CN120528815A - Intelligent network prediction and dynamic optimization method based on machine learning - Google Patents

Abstract

The application provides an intelligent network prediction and dynamic optimization method based on machine learning, which relates to the technical field of network communication and optimization and comprises the steps of predicting the future network condition change trend through a machine learning algorithm based on historical data and detection results to obtain a prediction result; and constructing decision logic based on the field expertise and experience, generating a deployment proposal according to the prediction result and the decision logic, and selecting a deployment scheme according to the deployment proposal.

Description

Intelligent network prediction and dynamic optimization method based on machine learning

Technical Field

The application belongs to the technical field of network communication and optimization, and particularly relates to an intelligent network prediction and dynamic optimization method based on machine learning.

Background

With the rapid development of information technology, network communication has been advanced to every corner of social life, from early fixed telephone networks to the ubiquitous mobile internet today, and demands of people for network speed, stability and security are increasing. Particularly, the rise of the emerging fields of Internet of things (IoT), 5G communication technology, cloud computing and the like in recent years promotes the wide application of intelligent embedded equipment remote management systems, and the systems are required to process massive data transmission tasks, respond to diversified demands of users in real time and ensure the efficient operation of various applications and services. Currently, most intelligent embedded device remote management systems rely on static configuration and a fixed set of rules to manage and optimize network performance, for example, in conventional Wide Area Network (WAN) optimization solutions, a series of parameters such as bandwidth allocation ratio, qoS policy, and routing are typically preset, and it is expected that these settings will remain valid for a long period of time. In addition, some more advanced systems have started to introduce machine learning algorithms, and predict the flow change trend possibly occurring in the future through analysis of historical data so as to make corresponding adjustment measures in advance

Although the prior art improves the automation level of network management to a certain extent, the prior art still has defects when facing to dynamic and changeable practical application scenes, particularly has the defects of lack of foresight, the traditional method often carries out resource allocation based on fixed time periods or periodic modes, is difficult to accurately capture the sudden flow peak or abnormal activity signs possibly occurring in a short period, causes untimely coping and influences user experience, has poor adaptability, cannot flexibly respond to specific requirements of various applications due to insufficient consideration of differences of different service types and service levels, is easy to cause interference or interruption of key services particularly in the application scenes with strong instantaneity, has limited intelligentization degree, mainly focuses on post analysis and summarization rules, lacks active prediction capability in the future, cannot provide enough support for decision making, and has the complex mode identification and self-adaptive adjustment functions caused by lack of deep learning, has insufficient safety, is difficult to rapidly establish a complete and complete security-monitoring framework once the prior system is difficult to rapidly detect and rapidly-locate the dangerous situation, and the problem is difficult to rapidly detect the security-oriented and the security-related fault.

Disclosure of Invention

The application provides an intelligent network prediction and dynamic optimization method based on machine learning, which aims to solve the problems of unreasonable resource allocation, affected key business and untimely handling of emergency caused by lack of prospective prediction and flexible capability of adapting to dynamic network environment change in the prior art.

The technical scheme adopted by the application is as follows:

The embodiment of the application provides an intelligent network prediction and dynamic optimization method based on machine learning, which comprises the following steps:

Predicting a future network condition change trend through a machine learning algorithm based on the historical data and the detection result to obtain a prediction result;

constructing decision logic based on field expertise and experience;

Generating a deployment proposal according to the prediction result and the decision logic, and selecting a deployment proposal according to the deployment proposal.

According to an embodiment of the present application, the predicting the future network condition change trend by a machine learning algorithm based on the historical data and the detection result, specifically, the predicting result is:

acquiring the historical data and the detection result, and preprocessing;

extracting features of the preprocessed data;

based on the trained model, obtaining the prediction result according to the characteristics;

the detection result comprises network performance indexes, environment parameters and log records which are measured through preset time intervals;

and selecting the detection result of a similar time period from the historical data of the past year.

According to one embodiment of the application, the decision logic is constructed based on the field expertise and experience, specifically:

And determining key factors influencing network performance and early warning thresholds by combining experience of field experts and past cases, and setting specific rules to construct the decision logic.

According to one embodiment of the present application, the generating a deployment proposal according to the prediction result and the decision logic, and selecting a deployment scheme according to the deployment proposal specifically comprises:

Inputting the prediction result into the decision logic, and checking whether the prediction result meets the condition of one or more rules in the decision logic or not so as to obtain the deployment suggestion;

Supplementing the decision logic in combination with a trained machine learning model;

carrying out multi-factor comprehensive evaluation on the deployment suggestion, wherein the multi-factor comprehensive evaluation comprises cost benefit analysis, risk evaluation and implementation difficulty;

The deployment scheme is selected through the deployment proposal, and the deployment scheme comprises a direct deployment scheme and an indirect bridging scheme.

According to one embodiment of the application, the direct deployment scheme specifically optimizes the existing equipment configuration, including parameter adjustment and network topology optimization.

According to one embodiment of the application, the indirect bridging scheme is specifically:

adopting CPE network bridge equipment and dynamically optimizing a network link;

the dynamic optimization comprises automatically adjusting the transmitting power according to the change of the real-time environment;

Using dynamic frequency selection function to make access point identify and avoid occupied or interfered channel;

And dynamically adjusting the service quality setting according to the real-time service requirement.

According to one embodiment of the application, the service quality setting is dynamically adjusted according to real-time service requirements, specifically, bandwidth resources are allocated according to application types and user role factors by defining different service levels and service strategies.

An electronic device comprising a memory storing a computer program and a processor implementing the steps of the method when the processor executes the computer program.

A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of the method.

A computer program product comprising instructions which, when run on a device, cause the device to perform steps in implementing the method.

By adopting the technical scheme, the application has the following beneficial effects:

The application adopts a technical scheme based on historical data and detection results and predicting the change trend of the future network condition through a machine learning algorithm, combines the field expertise and experience to construct decision logic, realizes the remarkable optimization of the disconnection reconnection method of the intelligent embedded equipment remote management system, not only improves the accuracy and the foresight of network performance prediction, but also enhances the capability of the system for coping with sudden flow peaks or abnormal activity signs, ensures that effective response can be timely made even under the condition of rapid change of network conditions, avoids the occurrence of potential problems, and can dynamically adjust service quality setting according to real-time service demands, preferentially ensure that key services are not interfered, simultaneously maximize the utilization rate of the whole network resources, improve service quality and user experience, further, through the selection of direct deployment schemes and indirect bridging schemes, including parameter adjustment, network topology optimization, dynamic optimization by CPE bridge equipment and other measures, the system can realize high-efficiency flexible configuration without increasing excessive cost, ensure that the system can realize effective response in time even under the condition of rapid change of network conditions, avoid the occurrence of potential problems, and can dynamically adjust service quality setting according to real-time service demands, simultaneously maximize the utilization rate of the whole network resources, improve the service quality, improve the reliability, promote the reliability of the system, promote the continuous service performance of the system, improve the reliability, and estimate, improve the reliability, and improve the reliability of the system, and continuously, improve the reliability, and the reliability of the network, and the reliability of the performance of the system is improved, and the performance, provides scientific basis and technical guarantee for future network management. .

Drawings

The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the application and do not constitute a limitation on the application. In the drawings:

fig. 1 is a schematic flow chart of an intelligent network prediction and dynamic optimization method based on machine learning according to an embodiment of the present application;

Fig. 2 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Reference numerals:

810. processor 820, communication interface 830, memory 840, communication bus.

Detailed Description

In order to more clearly illustrate the general inventive concept, a detailed description is given below by way of example with reference to the accompanying drawings.

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application, but the present application may be practiced in other ways than those described herein, and therefore the scope of the present application is not limited to the specific embodiments disclosed below. It should be noted that, without conflict, embodiments of the present application and features in each embodiment may be combined with each other.

In the present application, unless expressly stated or limited otherwise, a first feature "up" or "down" a second feature may be the first and second features in direct contact, or the first and second features in indirect contact via an intervening medium. In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present application. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

As shown in fig. 1, an intelligent network prediction and dynamic optimization method based on machine learning includes:

based on the historical data and the detection result, predicting the future network condition change trend through a machine learning algorithm to obtain a prediction result.

Specifically, a large amount of historical network operation data is collected from a plurality of sources as training samples, the data comprise router logs, traffic statistics information, qoS monitoring system records and the like, and meanwhile, the real-time monitoring results of the current network environment, such as the latest key indexes of traffic patterns, delay time, packet loss rate and the like, are combined and integrated into a prediction model, and the original data need to undergo preprocessing steps, such as abnormal point removal, vacancy filling data, standardized numerical range and the like, so as to ensure the quality and applicability of the data; then selecting machine learning algorithm which can process complex nonlinear relation and has good generalization ability, such as random forest, support Vector Machine (SVM), gradient lifting decision tree (GBDT), long and short time memory network (LSTM), especially LSTM which is good for capturing long time dependence characteristic, showing prominence in processing time sequence data, selecting algorithm, determining optimum superparameter configuration, implementing by grid search, random search or Bayesian optimization method, implementing k-fold cross-validation strategy in whole training process to avoid deviation caused by accidental error, making prediction for network condition in a certain time period after model training is completed, inputting a batch of real-time data and necessary context information, outputting predicted various item performance and confidence interval, building a complete evaluation system for predicting accuracy, the commonly used measurement indexes comprise Mean Square Error (MSE), mean Absolute Error (MAE) and R square (R2), visual display can be carried out by drawing a comparison graph between an actual observed value and a predicted value, any predicted model is gradually disabled along with the time, so that a continuously improved feedback loop is established, a training data set is timely updated to ensure that the model is always in a latest state, the validity and algorithm applicability of selected features are regularly and re-checked to make corresponding adjustment when necessary, and finally, after the development of the predicted module is completed, the predicted module is seamlessly embedded into an existing network management platform, so that good compatibility between the two is ensured, special requirements in actual application scenes such as response speed, resource consumption and the like are considered, an accurate and reliable predicted result is taken as a support, a more intelligent early warning system and an automatic decision process can be further designed, such as automatic triggering of load balancing measures or early planning of a maintenance window period in advance to reduce the influence on normal business as much as possible before serious congestion is predicted.

For example, find particular application in a practical telecom operator network management scenario, e.g., a large telecom operator wants to enhance user experience by improving quality of service (QoS) of its 4G/5G wireless network, reduce customer complaints, and optimize network resource allocation. To achieve this goal, the operator decides to deploy a set of intelligent predictive systems, using machine learning algorithms to analyze past traffic patterns, device performance metrics, and real-time monitoring data, to identify potential network problems ahead of time and to take precautions.

First, the operator collects from its core network element equipment, base stations (eNodeB/gNodeB), probe systems and other monitoring tools detailed operational records over the past year, which cover not only basic information during daily operations, such as data throughput per hour, number of connections, signal strength, etc., but also status snapshots at the occurrence of special events, such as holiday peak hours or abnormally high load conditions during live events of important sports. Meanwhile, in order to ensure that the model can capture the latest network behavior characteristics, sensor nodes in the whole country are also arranged, and the current network state is continuously updated back to the central database.

Next, preprocessing work is performed on these massive and complex data, including cleaning up invalid entries, filling in missing values, converting non-numeric fields into a computable format, and rearranging all observation points according to time-series characteristics. Then, a series of parameters closely related to network performance, such as distribution density of User Equipment (UE), flow rate duty ratio of specific service type, delay difference of uplink and downlink, etc., are extracted by adopting advanced feature engineering technology, and core elements which can most represent network health condition are screened out by using a statistical method as a basis of subsequent modeling.

In the stage of constructing the predictive model, a long-short-term memory network (LSTM) is selected, which is a recurrent neural network architecture particularly suited to process and predict time-series data with long-term dependencies. LSTM can effectively remember past traffic patterns and infer future trends accordingly. In the training process, through adjusting super parameters such as learning rate, batch size, hidden layer number and the like and combining with a k-fold cross validation strategy, the optimal configuration capable of achieving the minimum Mean Square Error (MSE) on a test set is finally determined. In addition, an online learning mechanism is introduced to allow the model to automatically update weights along with the arrival of new data so as to maintain prediction accuracy in consideration of the dynamic property and unpredictability of the network environment.

After the intelligent prediction system is formally online, the intelligent prediction system is integrated into an existing Operation Support System (OSS) platform and becomes an integral part of daily maintenance work. Whenever it is desired to evaluate the network performance in an area within a few hours of the future, the system will quickly generate a detailed predictive report containing the predicted bandwidth requirements, the likely congestion hot spot locations, and the proposed mitigation strategies by simply entering the latest set of real-time data collected. For example, before predicting that the upcoming morning commute peak period may cause the overload of base stations near some commercial areas, the administrator may adjust the power setting between adjacent cells in advance or temporarily add additional spectrum resource allocation according to the pre-warning information, so as to avoid the service interruption phenomenon.

Furthermore, the system can help operators to plan long-term network expansion plans more accurately. By periodically analyzing the historical flow peaks in different seasons, holidays and major activities, the change rule of the public communication demands can be accurately mastered, and scientific basis is provided for infrastructure construction. Moreover, as the whole process is highly automated, the manual intervention requirement is reduced, and the working efficiency and the service response speed are remarkably improved. Most importantly, by virtue of the prospective and accuracy of the network condition, the internet surfing experience of the terminal user is greatly improved, the discontent emotion caused by network faults is reduced, and the growth of enterprise images and market share is promoted.

Furthermore, the predictive network planning and resource allocation can be performed by using the prediction result, the intelligent fault diagnosis system can be developed to identify and solve potential problems in advance, the service quality of key business can be ensured by optimizing the existing QoS strategy, and the anti-attack capability of the system can be enhanced by designing a self-adaptive network security protection mechanism; in addition, in order to improve the prediction precision and practicability, external factors such as weather information, social activity schedules and the like with more dimensions can be introduced, and model parameters can be dynamically adjusted by combining deep reinforcement learning, meanwhile, a customized solution is an indispensable part in consideration of special requirements possibly existing in different areas or industries, which means that the choice of a technical framework and a methodology of feature engineering needs to be flexibly adjusted according to the characteristics of different application scenes, furthermore, along with the large-scale deployment of the Internet of things equipment, the optimization of the transmission characteristics of massive small data packets is also important, which relates to the support of a low-power wide area network (LPWAN) protocol and the understanding of corresponding flow modes, and the ultra-low delay and high reliability requirements under the future 5G or even 6G communication standard, the exploration of a real-time prediction algorithm suitable for a high-speed mobile scene is also an important research direction, so that the problems of user privacy protection and data security are required to be emphasized while the functions are realized, the information of all participants are ensured not to be leaked by adopting advanced encryption technology and anonymization processing means, finally, a real-time response system is not required to be established for the real-time performance of a real-time response system to be estimated and a real-time response system is not to be predicted and maintained, these are key elements necessary to ensure the stable and reliable operation of the whole prediction system.

Based on domain expertise and experience, decision logic is constructed.

In particular, a set of rules or algorithm flows capable of guiding automatic or semi-automatic decision making are formulated through systematic methodology by combining deep knowledge in the fields of communication engineering, computer science, best practices of specific industries and the like with actual working experience, the decision logic covers deep understanding and analysis of network performance indexes (such as bandwidth utilization rate, delay, jitter, packet loss rate and the like), influences of various aspects such as business requirements, user behavior patterns, cost benefit evaluation and the like are considered, the intelligent support is provided for key tasks such as resource allocation, fault diagnosis, quality of service guarantee, safety strategy implementation and the like, in specific application, data from different channels including but not limited to historical operation records, real-time monitoring information, external environment parameters and the like are firstly required to be collected and arranged, accuracy and applicability are ensured through preprocessing steps such as data cleaning, feature extraction and the like, then a machine learning model is trained by utilizing the data so as to identify potential patterns and trends, further more accurate prediction results are generated, simultaneously, human experts can be combined with the influence factors of aspects such as service requirements, fault diagnosis and cost benefit evaluation and the like, the like can be set up in a series of dynamic environment conditions such as to be different from the prior to the prior art, the dynamic conditions can be automatically set up by the fuzzy logic, the dynamic conditions can be adjusted to have a high-level of the self-adaptive situation, the dynamic conditions can be set up to the optimal conditions, and the dynamic conditions can be automatically changed according to the dynamic conditions, the dynamic conditions of the dynamic conditions can be well-adapted to the dynamic conditions, and the dynamic conditions can be set up by the dynamic conditions of the decision logic, and the system is required to have the requirements of the dynamic conditions, and the requirements of the conditions of the requirements of the dynamic conditions can be automatically changed, and the conditions according to the conditions and the conditions can be self-adaptively change, such decision logic is often deployed as part of a microservice architecture to facilitate efficient collaboration across multiple geographic locations and organizational boundaries, notably where relevant standards and technical specifications must be strictly followed during construction to ensure interoperability and smooth integration between the various components, while also fully respecting the design of the user experience.

For example, in a data center environment of a large enterprise, network engineers can design a complex decision logic for a machine learning-driven traffic prediction model according to years of operation and maintenance experience and understanding of service flows, the logic not only considers the characteristics of data interaction among different departments in daily office time, but also makes special countermeasures especially for large-scale file transmission activities possibly occurring during final financial settlement, when the prediction algorithm indicates that the bandwidth requirements between financial departments and servers in a certain period of time are obviously increased, the decision logic can automatically trigger a series of operations, firstly inform an IT support team to check the states of related hardware facilities in advance to ensure that all devices are in an optimal working state, secondly adjust QoS policies to ensure that the data flows of the critical services are smooth, and properly limit the resource occupation of a non-emergency application such as a video conference system, furthermore, a redundant link is started to disperse the pressure of a backbone network, so as to prevent possible faults caused by sudden peak traffic, in addition, important data are backed up regularly and the effectiveness of a recovery scheme is tested, meanwhile, in order to further improve the efficiency of the policy, even though the policy is further provided by the additional policy, the policy is further provided by the DDS, the policy is further improved, the policy is further can be completely and the network is further protected by the remote protection and the policy is further provided by the policy, which is more than can be completely and is better in the mode, even if the policy is more than the policy is completely protected, a feedback loop mechanism is established to collect user satisfaction scores and technical index performances after each event processing as important reference bases for subsequent optimization decision logic, so that a virtuous circle is formed, and the stability and flexibility of enterprise network infrastructure are continuously improved.

Furthermore, according to the historical flow mode and the expected growth trend, the factors such as geographic distribution characteristics, user behavior habits and the like can be combined, and the simulation tool is utilized to simulate the performance under different schemes, so that the optimal infrastructure expansion path is selected or the existing equipment layout is adjusted to meet the increasing service demands; in fault management, expert systems can integrate solutions in past case libraries, automatically match the most similar situations when abnormal signals are monitored, give targeted repair suggestions and even directly execute preset operation instructions such as restarting services, switching standby lines and the like, simultaneously deeply explore problem sources by means of causal analysis methods to avoid similar events from happening again, dynamically prioritize the priority queues based on deep understanding of various service characteristics (video conferences with high practical requirements compared with file downloading with higher elasticity) for QoS policy optimization, ensure that important tasks obtain enough bandwidth support without affecting normal development of other conventional activities, quickly respond to maintain consistency and stability of the whole service level, establish a complete risk assessment framework by virtue of rich attack and defense experiences to cover aspects of scanning, intrusion detection, threat intelligence sharing and the like, immediately start emergency response pre-solutions once suspicious activities are found, including but not limited to limiting access, isolating infected nodes, updating potential signature and the like, effectively diffusing, completing the existing policy and the like, complete network system (6) and the like, complete the requirements of the existing logic system (6) and the like along with the development of the prior art, in order to ensure that the constructed decision logic not only follows the steps of the time but also is not robust and reliable, a closed loop flow with continuous learning and iterative improvement must be established, on one hand, the latest research results and technical trends in academia are closely focused, on the other hand, the actual problems and improvement suggestions fed back by first-line operation staff are actively collected, the actual problems and improvement suggestions are regularly organized to communicate with each other across department seminars, the best practice cases are jointly discussed, so that the evolution and upgrading of the whole decision system are promoted, and finally, but also importantly, the artificial design concept is always followed in the whole process, the experience feeling of an end user is fully considered, whether the operation interface is simplified or the transparency is enhanced, and the user is focused on the more friendly, efficient and safe network environment.

Generating a deployment proposal according to the prediction result and the decision logic, and selecting a deployment proposal according to the deployment proposal.

Specifically, upon receiving future network condition prediction data output by the machine learning model, the information is input into a pre-designed decision support system that fuses multiple expertise from network engineering, communication technology, information system management, and network security, and is able to deeply interpret the prediction results and formulate a series of targeted and prospective deployment suggestions accordingly, e.g., if the prediction shows that a particular area will face significant traffic growth over a future period of time, the decision logic may suggest to increase bandwidth resources of the area in advance or optimize existing routing configurations to ensure good quality of service even under high load conditions, while, for possible failure points or security risks, the system may propose preventive measures based on best practices in a historical case library, such as enhancing monitoring efforts, deploying additional security protection layers, or preparing emergency response plans.

Further, for different types of traffic demands and Service Level Agreements (SLAs), the decision logic may be further refined to more specific guidelines to help determine which applications should share higher priority processing rights and how to achieve this with QoS policies, e.g., for video conferencing or online gaming services with extremely high real-time requirements, the system may suggest to allocate dedicated transmission channels and enable high-level error correction mechanisms to minimize delay and packet loss, while for non-critical tasks, the constraints may be relaxed appropriately to allow them to share the remaining bandwidth without affecting overall performance, thus not only ensuring that important traffic is not disturbed, but also improving the utilization of the overall network resources.

Furthermore, given the complexity and diversity in the actual operating environment, the generated deployment recommendations are often not single fixed, but rather include a variety of alternatives for reference by the decision maker. Each solution has unique advantages and disadvantages and application scenes, so that comprehensive trade-offs are required by combining currently available technical means, budget limitation, human resources and other factors, for example, in the process of deciding whether to upgrade the existing infrastructure by adopting the latest 5G technology, cost benefit ratio, compatibility problem and long-term maintenance cost are also required to be considered in addition to evaluating the performance improvement brought by the latest 5G technology, and when the third party cloud service platform is introduced, the contents of service level protocol, data privacy protection policy and other aspects are also required to be carefully examined in addition to the functional characteristics provided by the third party cloud service platform, so that all choices are ensured to meet the strategic targets and compliance requirements of the organization.

Last but not least, in order for the final selected deployment scenario to achieve the desired effect in practice, a complete set of implementation planning and tracking evaluation mechanisms needs to be established. The method comprises the steps of defining task division, time nodes and responsibility main bodies of each stage, ensuring orderly development of each work, setting up Key Performance Indicators (KPIs) for measuring actual performance of a new scheme, periodically collecting user feedback opinions for analysis and summarization so as to discover problems in time and make corresponding adjustment, and in addition, keeping flexible coping attitudes in view of quick change characteristics of network environment and technical development, preparing to update original deployment strategies according to the latest condition at any time, and continuously optimizing network performance and service quality.

For example, assume that a nationwide network infrastructure is managing a large enterprise having multiple branches distributed across the country, each of which relies on a stable and efficient network connection for everyday operations such as video conferencing, data synchronization, remote collaboration, etc. Based on historical data and real-time monitoring results, the machine learning model predicts that in the coming month, due to the upcoming annual stakeholder meeting and the consequent massive remote participation demands, bandwidth utilization between headquarters and subdivisions will increase significantly, especially short but severe flow peaks may occur during the meeting.

In combination with this prediction, decision logic built using domain expertise comes into play. Firstly, taking into account the upcoming significant activities and their requirements for network performance, it is decided to take action in advance, ensuring that all critical traffic can be fully guaranteed. In order to cope with possible high load situations, temporary leasing of additional cloud resources can be considered to share the pressure of a local server, meanwhile, an existing Content Delivery Network (CDN) is optimized, so that multimedia files can be transmitted to remote users more quickly, for critical links with frequent congestion phenomenon, the robustness of the critical links can be improved by adjusting routing strategies or introducing new redundant paths, service interruption caused by single-point faults is avoided, in addition, differentiated QoS policies are implemented for different types of traffic, the quality of voice calls and video streaming media is preferably guaranteed not to be influenced, and for other non-real-time applications, the limit is properly relaxed to balance the overall bandwidth allocation, and in the network security level, boundary protection measures must be enhanced, such as updating firewall rules, deploying Intrusion Detection Systems (IDSs) and enhancing monitoring strength, so that any suspicious behaviors can be found and processed in time, in view of the targets of hacking.

The specific deployment proposal is then generated based on the analysis described above. For the main communication lines from headquarters to each subsection, it is suggested to complete all necessary capacity expansion preparation work at least one week before the peak period and conduct pressure test to verify the validity of the new configuration, for unplanned emergency situations, a set of detailed emergency plans including but not limited to emergency contact list, alternative communication means, technical support hotline and other information are prepared so as to quickly respond to possible problems, in consideration of relatively poor network conditions in partial remote areas, it is suggested to send technicians to these areas in advance to carry portable satellite communication equipment as a standby plan to ensure that basic communication capability can be maintained even in extreme cases, in addition, a comprehensive security inspection should be organized to check whether the existing defense system has loopholes or weak links and adjust related settings accordingly, such as enabling stronger identity verification mechanisms, encryption sensitive data transmission processes and the like, finally, special people are arranged to monitor the network state in real time and prepare flexible adjustment strategies according to actual situations at any time during the whole activity period to ensure optimal service quality.

The final step is to select the most appropriate solution based on the deployment recommendation. In this case, the final decision is made by combining a number of factors, including cost effectiveness, technical feasibility, risk assessment, etc. For example, while temporary renting cloud resources does effectively alleviate traffic pressure in the short term, if this is not economical in the long term, it is desirable to trade off whether it is worth investing in purchasing more permanent hardware facilities, and likewise, while introducing advanced security techniques and equipment to help promote the level of protection, it is also desirable to consider whether they introduce additional operational complexity or compatibility issues, and when cross-regional coordination is involved, it is also desirable to take special attention to the differences in legal regulations in different regions to ensure that all actions are performed with legal compliance.

Furthermore, the method can be combined with real-time network traffic analysis and user behavior pattern recognition technology to more accurately capture the possible sudden traffic peak or abnormal activity signs in a short period, thereby early warning and dynamically adjusting resource allocation, and can also introduce a deep reinforcement learning algorithm to simulate and train the optimal response strategies under different scenes, so that the self-optimization capacity of the system in complex and changeable environments is gradually improved through continuous trial and error accumulation experience, in addition, in order to cope with the increasing data security challenges, advanced Encryption Standard (AES) and other advanced network security protocols can be integrated on the basis of the prior art, and the confidentiality and the integrity of all transmitted data are ensured; meanwhile, a comprehensive log record and audit trail mechanism is established, a detailed and reliable basis is provided for subsequent problem investigation, and in consideration of the development trend of edge calculation, the possibility of migrating part of core processing tasks to be executed on a miniaturized and low-power-consumption computing node close to a data source position is actively explored, so that delay time is shortened, response speed is improved, user experience is improved, meanwhile, a set of efficient load balancing solution is designed according to the requirement of cross-region cooperation in a large-scale distributed system, flexible flow scheduling is realized by utilizing a Software Defined Network (SDN) controller, and optimal path selection is realized by matching with intelligent DNS analysis service.

In some embodiments of the present application, the predicting the future network status change trend by a machine learning algorithm based on the historical data and the detection result to obtain a predicted result specifically includes:

acquiring the historical data and the detection result, and preprocessing;

extracting features of the preprocessed data;

based on the trained model, obtaining the prediction result according to the characteristics;

the detection result comprises network performance indexes, environment parameters and log records which are measured through preset time intervals;

and selecting the detection result of a similar time period from the historical data of the past year.

Specifically, first, the history data and the detection result are acquired, and preprocessing is performed. The historical data refers to the network performance index, the environmental parameter and the log record in the similar time period in the past year, and the detection result refers to the current network performance index, the environmental parameter and the log record measured by the set time interval (such as every hour or every day). To ensure that these data can be effectively used for subsequent analysis, they need to be subjected to a series of preprocessing operations, such as removing outliers, filling in missing data points, normalizing the range of values, and possibly even data conversion, in order to convert non-numeric data (e.g., state codes) into a form that can be understood by the model. In addition, preprocessing involves data cleansing, i.e., clearing duplicate or erroneous data entries, to ensure the quality of the data input to the machine learning model.

The feature extraction is then performed on the preprocessed data. This step aims at mining the most useful information for prediction from the raw data. Feature extraction may be accomplished in a variety of ways, such as calculating statistical features of average throughput, peak bandwidth utilization, number of active users, etc., or more complex features, such as rate of change of traffic patterns, traffic duty cycle of a particular traffic type, etc. In some cases, automated feature selection methods, such as Recursive Feature Elimination (RFE), principal Component Analysis (PCA), may also be used to screen out core elements that are most representative of network health. This process is critical to simplifying the model structure and improving the prediction accuracy, as it can help exclude irrelevant or redundant information, reducing the risk of overfitting.

And then, based on the trained model, obtaining the prediction result according to the characteristics. This means that future network conditions are predicted using the machine learning model that has been trained and validated, with the extracted features as input variables. The choice of machine learning model depends on the data characteristics and problem requirements, common models include random forests, support Vector Machines (SVMs), gradient-lifting decision trees (GBDT), and particularly long-term memory networks (LSTM) that are good at capturing long-term dependencies. After the model is selected, the optimal hyper-parameter configuration needs to be determined, and the optimal hyper-parameter configuration is usually realized by adopting methods such as grid search, random search or Bayesian optimization. The k-fold cross-validation strategy is implemented throughout the training process to avoid bias from occasional errors. Finally, the trained model can make predictions of network conditions over a future period of time, outputting predicted performance index values and confidence intervals thereof.

Finally, in order to measure the prediction accuracy, it is important to build a complete evaluation system. Commonly used metrics include Mean Square Error (MSE), mean Absolute Error (MAE), R square (R2), and can also be visually demonstrated by plotting a comparison graph between actual observed values and predicted values. Any predictive model will gradually lose its effectiveness over time, so a continuously improved feedback loop is established to update the training data set in time, ensuring that the model is always up-to-date. The validity of the selected features and the applicability of the algorithm are reviewed periodically, with corresponding adjustments made as necessary. After the development of the prediction module is completed, the prediction module is seamlessly embedded into the existing network management platform, so that good compatibility between the prediction module and the existing network management platform is ensured, special requirements in practical application scenes such as response speed, resource consumption and the like are considered, accurate and reliable prediction results are used as supports, and a more intelligent early warning system and an automatic decision flow can be further designed, such as automatic triggering of load balancing measures or early planning of maintenance window periods before serious congestion is predicted, so that influence on normal business is reduced as much as possible.

In some embodiments of the present application, the decision logic is constructed based on domain expertise and experience, specifically:

And determining key factors influencing network performance and early warning thresholds by combining experience of field experts and past cases, and setting specific rules to construct the decision logic.

Specifically, firstly, combining experience of a field expert and past cases, determining key factors and early warning thresholds which influence network performance. This step requires collecting and analyzing a large amount of historical data including, but not limited to, network traffic patterns, device performance metrics, user behavior habits, changes in business needs, seasonal fluctuations, special events (such as holiday peaks), etc., and identifying those factors that have a significant impact on network stability and quality of service. For example, in a telecom operator scenario, factors such as load balancing between base stations, spectrum resource allocation, mobile terminal connection quality, etc. may need to be considered, and in an enterprise data center environment, aspects such as response time of internal applications, synchronization efficiency between servers, effectiveness of security protection measures, etc. are more concerned. And setting reasonable early warning thresholds according to the historical performances of the key factors and the correlation of the key factors with network faults or performance degradation, and triggering corresponding alarm mechanisms when the monitored data exceeds the thresholds.

Then, specific rules are set to construct the decision logic. This process involves creating a series of if-then rules or using fuzzy logic to deal with problem domains with unclear boundaries to simulate the expert's thinking process in the face of complex situations. For example, if it is detected that the bandwidth utilization of a specific area is continuously higher than 80% within a certain period of time and the delay time exceeds a preset safety range, it is determined that the area is about to be at potential congestion risk, a load sharing plan is immediately started, an additional transmission channel is added or a QoS policy is adjusted, and it is ensured that the critical service is not affected. In addition, the machine learning model can be utilized to assist in generating rules, and an effective decision path is automatically extracted through learning of a large number of real cases and is self-optimized along with the continuous accumulation of new data.

Further, in consideration of the dynamic change characteristics of the network conditions, the built decision logic also needs to have self-adaptive capability, i.e. to automatically adjust the parameter settings of the decision logic according to new data input or environmental changes, so that the optimal working state is always maintained. For example, in the face of sudden high-traffic impact, the system should be able to quickly identify abnormal patterns and react immediately, such as automatically expanding cloud resources, enabling backup links, enhancing monitoring forces, etc., while for long-term trend changes, such as increase in user base or application of new technologies, infrastructure upgrade schemes need to be planned in advance to ensure service capability in a future period of time.

For large distributed systems, such decision logic is often deployed as part of a microservice architecture to facilitate efficient collaboration across multiple geographic locations and organizational boundaries. This means that not only the interoperability and smooth integration between the components are realized, but also the design of the user experience should be fully emphasized, the operation interface is simplified, the transparency is enhanced, and the user is enabled to feel a more friendly, efficient and safe network environment. Meanwhile, in order to ensure that decision logic is not only immediately following the step of the time, but also is not lost in robustness and reliability, a closed loop flow of continuous learning and iterative improvement must be established, wherein on one hand, the latest research results and technical trends in academia are closely concerned, on the other hand, the actual problems and improvement suggestions fed back by first-line operation and maintenance personnel are actively collected, and the best practice cases are studied together by periodic organization and cross-department seminar exchange, so that the evolution and the upgrading of the whole decision system are promoted.

In some embodiments of the present application, the generating a deployment proposal according to the prediction result and the decision logic, and selecting a deployment scheme according to the deployment proposal specifically includes:

Inputting the prediction result into the decision logic, and checking whether the prediction result meets the condition of one or more rules in the decision logic or not so as to obtain the deployment suggestion;

Supplementing the decision logic in combination with a trained machine learning model;

carrying out multi-factor comprehensive evaluation on the deployment suggestion, wherein the multi-factor comprehensive evaluation comprises cost benefit analysis, risk evaluation and implementation difficulty;

The deployment scheme is selected through the deployment proposal, and the deployment scheme comprises a direct deployment scheme and an indirect bridging scheme.

Specifically, first, the prediction result is input into the decision logic, and whether the prediction result meets the condition of one or more rules in the decision logic is checked to obtain the deployment suggestion. This means that predictive data (e.g., bandwidth requirements, possible congestion hotspot locations, predicted service outage times, etc.) for future network conditions using machine learning models is imported into a pre-designed decision support system. The system integrates professional experiences from multiple fields such as network engineering, communication technology, information system management, network security and the like, can deeply read prediction results, and can be used for preparing a series of highly targeted deployment suggestions. For example, if a prediction shows that a particular region will be subject to significant traffic growth over a period of time in the future, then decision logic may suggest to increase bandwidth resources in the region in advance or optimize existing routing configurations to ensure good quality of service is maintained even under high load conditions, while for possible failure points or security risks, the system may present precautions based on best practices in the historical case base, such as enhancing monitoring efforts, deploying additional security guards, or preparing emergency response plans.

Second, the decision logic is supplemented in conjunction with a trained machine learning model. This step emphasizes the ability to dynamically update and improve decision logic. By continuously collecting new data and retraining the machine learning model, the latest trend of change and technology development dynamics in the network environment can be captured, thereby continuously adjusting and perfecting the rule set in the decision logic. For example, when the new application type or user behavior mode is faced, the corresponding threshold setting and early warning mechanism can be updated by analyzing the influence of the new application type or user behavior mode on the network performance, and when a more advanced algorithm or tool is introduced, how to integrate the new application type or user behavior mode into the existing decision frame is considered, so that the overall intelligent level is improved. In addition, advanced technologies such as reinforcement learning and the like can be utilized to enable the system to automatically explore the optimal solution in the unknown field, so that the adaptability and the flexibility of the system are further enhanced.

And then, carrying out multi-factor comprehensive evaluation on the deployment proposal, wherein the multi-factor comprehensive evaluation comprises cost benefit analysis, risk evaluation and implementation difficulty. This means that each deployment proposal generated by the decision logic will be fully and carefully reviewed, ensuring that the final selected solution will not only meet the current specifications, but also be economically viable, potentially risky, and complex in practice. For example, in deciding whether to upgrade an existing infrastructure with the latest 5G technology, in addition to evaluating the performance improvement that it brings, cost-effectiveness, compatibility issues, and long-term maintenance costs must be considered, and likewise, when it comes to introducing a third-party cloud service platform, it is necessary to carefully examine the contents of service level agreements, data privacy protection policies, etc. in addition to the functional characteristics that it provides, to ensure that all choices meet the strategic goals and compliance requirements of the organization. By comprehensively considering the key factors, the manager can be helped to better balance the advantages and disadvantages and make a more intelligent choice.

Finally, the deployment scheme is selected through the deployment proposal, and the deployment scheme comprises a direct deployment scheme and an indirect bridging scheme. The two solutions mentioned here represent different implementation paths and technical means. While the direct deployment scheme generally refers to the optimization adjustment of the existing device configuration, such as parameter modification, network topology optimization, etc., the manner tends to be quick and easy to manage, the indirect bridging scheme more involves introducing external resources or new technologies, such as adopting CPE bridge devices, and dynamically optimizing network links, including automatically adjusting transmit power according to the change of real-time environment, avoiding occupied or interfered channels with dynamic frequency selection function, dynamically adjusting quality of service settings according to real-time service requirements, etc. Each solution has its own unique advantages and disadvantages and application scenario, so that the most appropriate choice needs to be made in combination with specific business requirements, state of the art and development planning.

In some embodiments of the application, the direct deployment scheme specifically optimizes existing device configurations, including parameter adjustment and network topology optimization.

In particular, parameter tuning refers to fine tuning of various settings in the network devices and software to ensure that they are operating in an optimal state. This may involve multiple levels:

bandwidth allocation, namely reallocating bandwidth resources according to service requirements, preferentially ensuring that key applications (such as video conferences and online transactions) obtain enough bandwidth support, and simultaneously reasonably planning traffic of non-real-time applications to avoid excessive occupation of resources.

QoS (quality of service) policy, defining different service levels and service policies, and distributing bandwidth resources according to different application types, user roles and other factors. For example, for services with high real-time requirements, a high-level error correction mechanism is enabled, while for tasks such as file downloading with higher elasticity, the limiting conditions can be properly relaxed.

Routing protocol optimization, namely checking and optimizing routing algorithms and protocol configurations, such as OSPF, BGP and the like, so as to reduce data packet transmission delay and packet loss rate. By dynamically adjusting the routing weights or introducing new path selection rules, it is ensured that the data stream can be transmitted along the shortest and most stable path.

Security setup, namely strengthening boundary protection measures, updating firewall rules and Intrusion Detection System (IDS) configuration, and strengthening encryption protection for sensitive data transmission processes, and preventing potential security threats.

Hardware configuration, namely checking the CPU utilization rate, memory use condition and other performance indexes of core equipment such as a server, a switch, a router and the like, and upgrading or replacing old equipment if necessary to maintain a high-efficiency stable running state.

Network topology optimization refers to improving the design and layout of the overall network structure to improve data transmission efficiency and system robustness. The main measures include, but are not limited to:

and in the redundancy design, a plurality of data transmission paths which are mutually backed up are constructed, and when the main link fails, the main link is automatically switched to a standby line, so that the communication is ensured not to be interrupted. In addition, a ring topology or other architecture with self-healing capability can be adopted, so that the fault tolerance and reliability of the network are further enhanced.

And load balancing, namely implementing an effective load sharing strategy to ensure that the flow distribution among different nodes is more uniform, and avoiding the condition that certain areas are overloaded and other places are idle. Flexible traffic scheduling may be implemented with a Software Defined Network (SDN) controller that cooperates with intelligent DNS resolution services to select an optimal path.

Edge computing, namely exploring the possibility of migrating part of core processing tasks to be executed on a miniaturized and low-power-consumption computing node close to a data source position, so as to reduce delay time and improve response speed, and further improve user experience.

The wireless coverage is enhanced, namely, for a scene depending on wireless connection, such as a mobile office area in Wi-Fi coverage, the signal strength and stability can be enlarged by increasing the number of access points, optimizing the antenna angle and the like, so that good connection quality in all areas is ensured.

Physical cabling improvements such as inspecting existing cabling schemes, identifying potential bottlenecks and addressing, such as replacing aged copper cables with optical fibers, or re-planning wiring patterns inside the machine room, ensure stable power supply and minimized electromagnetic interference.

And the data center interconnection is realized by designing a reasonable interconnection scheme according to the high-efficiency cooperation requirement among the distributed data centers, and the connection among all places is enhanced by utilizing a high-speed backbone network or a special line, so that the data synchronization is ensured to be rapid and accurate, and meanwhile, the cross-region access delay is reduced.

Implementation and maintenance after the optimization measures are completed, a complete implementation plan and tracking evaluation mechanism are also required to be established so as to ensure that the new scheme can achieve the expected effect in practice. The method comprises the steps of defining task division, time nodes and responsibility main bodies of each stage, ensuring orderly development of each work, setting up Key Performance Indicators (KPIs) for measuring actual performance of a new scheme, and periodically collecting user feedback opinions for analysis and summarization so as to discover problems in time and make corresponding adjustment. In addition, in view of the rapid change characteristics of network environment and technology development, the flexible coping attitude should be kept, the original deployment strategy is ready to be updated according to the latest situation at any time, and the network performance and the service quality are continuously optimized.

In some embodiments of the present application, the indirect bridging scheme is specifically:

adopting CPE network bridge equipment and dynamically optimizing a network link;

the dynamic optimization comprises automatically adjusting the transmitting power according to the change of the real-time environment;

Using dynamic frequency selection function to make access point identify and avoid occupied or interfered channel;

And dynamically adjusting the service quality setting according to the real-time service requirement.

In particular, CPE bridge devices are employed

CPE bridge devices are key components connecting a customer premise with a service provider network, which can bridge between wireless and wired networks to seamlessly integrate data streams from different networks. Such devices typically have powerful processing capabilities and flexible configuration options that can support a variety of communication protocols and interface standards, allowing for quick deployment and ease of management.

The CPE equipment can expand the effective coverage of Wi-Fi or other wireless technologies, especially in remote areas or places with weak signals, and the defects of the original network are overcome by adding additional access points.

And the transmission speed is improved, namely, the CPE equipment can obviously improve the data transmission speed under the condition of not sacrificing stability by utilizing the latest modulation-demodulation technology and frequency spectrum efficiency algorithm, and the requirements of high-bandwidth applications such as high-definition video streaming, large-scale file downloading and the like are met.

The system has the advantages of simplifying installation and maintenance, ensuring that many modern CPE equipment has compact design and easy arrangement, supporting plug and play functions, reducing the need of on-premise services of professional technicians, and simultaneously providing remote monitoring and fault removal functions, thereby facilitating centralized management and solving the problems.

Dynamically optimizing network links

Dynamic optimization refers to automatically adjusting network parameters according to changes in real-time environment to maintain optimal performance. This mainly includes the following aspects:

Automatic adjustment of transmit power

Energy saving and anti-interference, namely, when other wireless devices exist in the surrounding environment, the CPE device can intelligently reduce the self-transmitting power, reduce unnecessary energy consumption and avoid mutual interference with other devices. Conversely, if the signal strength is insufficient or the quality is degraded, the transmission power is appropriately increased, and a stable connection is ensured.

Adaptive environmental changes-wireless propagation conditions may change (e.g., building block, weather effects) over time and place. At this point, the CPE device can perceive these changes in real time and respond, maintaining optimal communication quality.

Using Dynamic Frequency Selection (DFS)

Channel selection optimization-through built-in dynamic frequency selection function, the CPE device can scan the available frequency band, and identify and avoid the channel which is already occupied or severely interfered. This means that relatively idle frequency bands can be found for data transmission even in a crowded spectrum environment, effectively reducing the risk of congestion.

Regulatory compliance the DFS mechanism also allows devices to comply with national legal regulations for use in particular frequency bands, such as in the 5GHz band, where certain sub-bands are limited to outdoor use or require rapid switching to other unoccupied frequency bands upon detection of radar signals.

Dynamically adjusting quality of service (QoS) settings

Resources are allocated on demand-CPE devices can dynamically allocate bandwidth resources according to predefined service levels and service policies, depending on different types of traffic demands. For example, for video conferencing or online gaming services where real-time requirements are extremely high, the system may allocate dedicated transmission channels and enable high-level error correction mechanisms to minimize delay and packet loss, while for non-critical tasks, constraints may be relaxed appropriately to allow them to share the remaining bandwidth without affecting overall performance.

The user experience is prioritized, and the refined management based on the application type and the user role is helpful to ensure that important services are not interfered, and the utilization rate of the whole network resource is improved. Particularly in the face of sudden peak traffic surges, can respond quickly, maintaining consistent and stable service levels.

In some embodiments of the present application, the service quality setting is dynamically adjusted according to real-time service requirements, specifically, bandwidth resources are allocated according to application types and user role factors by defining different service levels and service policies.

In particular, different service levels are defined

The service levels (SERVICE LEVEL) refer to different levels classified according to importance and urgency of the service, each level corresponding to a specific quality of service standard. Common service levels include, but are not limited to:

Platinum level-is suitable for applications that are extremely sensitive to delay, such as financial transaction systems, real-time video conferencing, etc. Such services require the highest bandwidth guarantee, the lowest delay time and the smallest packet loss rate.

Gold-level-applications that require higher stability and response speed but can accept a certain degree of delay, such as online games, remote educational platforms, etc.

Silver-level, which covers activities such as general internet browsing, mail sending and receiving, and the like, the requirements of the activities on bandwidth are relatively low, and flexible allocation can be performed on the premise of not affecting user experience.

Copper level-for applications that are not real-time or that can tolerate large delays, such as file downloads, bulk data processing, etc., these tasks can be performed during network idle periods to fully utilize the remaining bandwidth resources.

Formulating service policies

Service Policy (Service Policy) refers to a set of rules that directs how the above Service levels are applied in different situations, taking into account factors such as time, geographic location, device type, and specific business scenarios. The key point of formulating an effective service strategy is to accurately capture and reflect the change trend of the actual service demand, thereby realizing intelligent resource management. The specific measures include:

time-based policies-adjusting bandwidth allocation according to traffic patterns for different time periods. For example, the smooth operation of the office system inside the enterprise is mainly ensured in daytime during workdays, and more resources can be inclined to home entertainment applications such as high-definition video streaming media playing during nighttime or weekends.

Location-based policies-personalized service policies are tailored for different areas taking into account individual network condition differences. For example, adding redundant links or optimizing antenna configuration where signal coverage is poor ensures a high quality service experience even in remote areas.

And identifying characteristics of the terminal equipment based on the equipment type policy, and providing differentiated services according to the characteristics. For mobile devices, battery endurance and connection stability may be more important, while for stationary workstations, high throughput data transmission may be more important.

Based on the strategy of application type, the behavior characteristics and technical requirements of various applications are deeply understood, and the most suitable network environment is created for the body measurement of the application. For example, voice clarity and low latency characteristics should be guaranteed preferentially for voice telephony applications, while sufficient memory and computational power support should be guaranteed for large data analysis tasks.

User role factors

The Role of a User (User Role) reflects the status of the User in the network and what the User is working on, which directly affects the level of their demand for network resources. Therefore, when setting the service level and the service policy, the influence of the user role must be fully considered, so as to ensure reasonable and targeted resource allocation. The main considerations are as follows:

Management layer-typically involving important decision support systems and high-level conference communication, such users often require the highest level of service assurance to ensure timely and accurate information delivery.

Ordinary staff relies on various applications and services inside the enterprise in daily work, and although the requirements on bandwidth are not as strict as those of a management layer, stable network connection is also required to improve the working efficiency.

Visitor/temporary users have limited access rights, mainly for basic internet browsing and email, and thus can be provided with the necessary network access through lower-level services.

Special groups of users, such as development teams, technical support personnel, etc., may require special network configurations or additional security measures to meet specific job requirements.

Dynamic adjustment mechanism

In order for the service level and service policy described above to be able to accommodate changing business requirements, a set of efficient dynamic adjustment mechanisms must be established. This usually involves several links:

And (3) monitoring and analyzing in real time, namely continuously tracking indexes such as network flow, equipment state and application performance, and utilizing a big data analysis technology to mine potential problems and predict future trend. In this way, upcoming bottlenecks or risk points can be identified in advance, providing basis for subsequent adjustment.

And the automatic decision engine is combined with a machine learning algorithm to construct an intelligent decision system, which can automatically analyze the current service condition and make optimal selection. For example, when detecting that a certain department is performing large-scale data migration, the decision engine automatically increases the bandwidth priority of the department, so as to ensure that the task is completed smoothly.

Feedback loop optimization, namely establishing a closed loop flow from practice to theory to improvement, and periodically collecting user feedback opinions and technical index performances as important references of a follow-up optimization service strategy. Thus, not only can the service level be continuously improved, but also the satisfaction and the loyalty of the user can be enhanced.

An embodiment of a second aspect of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, where the processor implements the disconnection reconnection method of the intelligent embedded device remote management system in any of the embodiments of the first aspect when the processor executes the program.

Fig. 2 illustrates a physical schematic diagram of an electronic device, as shown in fig. 2, which may include a processor 810, a communication interface (Communications Interface) 820, a memory 830, and a communication bus 840, where the processor 810, the communication interface 820, and the memory 830 perform communication with each other through the communication bus 840. The processor 810 may invoke logic instructions in the memory 830 to perform the intelligent network prediction and dynamic optimization method based on machine learning in any of the embodiments of the first aspect described above, the method comprising:

Predicting a future network condition change trend through a machine learning algorithm based on the historical data and the detection result to obtain a prediction result;

constructing decision logic based on field expertise and experience;

Generating a deployment proposal according to the prediction result and the decision logic, and selecting a deployment proposal according to the deployment proposal.

Further, the logic instructions in the memory 830 described above may be implemented in the form of software functional units and may be stored in a computer-readable storage medium when sold or used as a stand-alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. The storage medium includes a U disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, an optical disk, or other various media capable of storing program codes.

In another aspect, the present invention also provides a computer program product, where the computer program product includes a computer program, where the computer program can be stored on a non-transitory computer readable storage medium, where the computer program when executed by a processor can perform the intelligent network prediction and dynamic optimization method based on machine learning provided by the above methods, and the method includes:

Performing heartbeat detection on the client and the equipment according to a preset period, and checking the connection state of the TCP to judge whether a disconnection condition exists;

dividing the disconnection condition into WiFi signal problem, IP address change and equipment restarting or network configuration change;

selecting different reconnection modes according to different disconnection conditions, wherein the reconnection modes comprise a quick reconnection strategy, an exponential backoff algorithm and connection by using a new IP address;

The quick reconnection strategy is used for solving the WiFi signal problem, the exponential backoff algorithm is used for solving the equipment restart or network configuration change, and the new IP address is used for connection to solve the IP address change.

In yet another aspect, the present invention further provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, is implemented to perform the machine learning based intelligent network prediction and dynamic optimization method provided by the above methods, the method comprising:

Performing heartbeat detection on the client and the equipment according to a preset period, and checking the connection state of the TCP to judge whether a disconnection condition exists;

dividing the disconnection condition into WiFi signal problem, IP address change and equipment restarting or network configuration change;

selecting different reconnection modes according to different disconnection conditions, wherein the reconnection modes comprise a quick reconnection strategy, an exponential backoff algorithm and connection by using a new IP address;

The quick reconnection strategy is used for solving the WiFi signal problem, the exponential backoff algorithm is used for solving the equipment restart or network configuration change, and the new IP address is used for connection to solve the IP address change.

Example 2

In a large telecommunications carrier scenario, the carrier decides to deploy a set of intelligent predictive systems in order to improve the quality of service (QoS) of its 4G/5G wireless network, enhance the user experience and optimize the network resource allocation. The system analyzes the past flow mode, the equipment performance index and the real-time monitoring data by utilizing a machine learning algorithm, so that potential network problems are identified in advance and preventive measures are taken.

Data collection and preprocessing

Data source and collection:

First, the operator has collected detailed operational records over the past year from its core network element equipment, base stations (eNodeB/gNodeB), probe systems, and other monitoring tools. These records not only cover basic information during daily operations such as data throughput per hour, number of connections, signal strength, etc., but also status snapshots at the time of occurrence of special events such as holiday peak hours or abnormally high load conditions during live events of important sports. Meanwhile, in order to ensure that the model can capture the latest network behavior characteristics, sensor nodes in the whole country are also arranged, and the current network state is continuously updated back to the central database.

Data preprocessing:

Next, preprocessing work is performed on these massive and complex data, including:

cleaning up invalid entries, removing incomplete or erroneous data records.

Filling the missing value, namely filling the blank part in the data by using a statistical method so as to maintain the data integrity.

Converting non-numeric fields-converting non-numeric data (e.g., state codes) into a form understandable by the model.

And (3) standardizing the numerical range, namely adjusting the data scale and ensuring the comparability among different variables.

And rearranging all observation points according to the time sequence characteristics, namely ensuring that the data are arranged in time sequence, and facilitating the subsequent time sequence analysis.

Then, a series of parameters closely related to network performance, such as distribution density of User Equipment (UE), flow rate duty ratio of specific service type, delay difference of uplink and downlink, etc., are extracted by adopting advanced feature engineering technology, and core elements which can most represent network health condition are screened out by using a statistical method as a basis of subsequent modeling.

Constructing a predictive model

Selecting a model architecture:

In the stage of constructing the predictive model, a long-short-term memory network (LSTM) is selected, which is a recurrent neural network architecture particularly suited to process and predict time-series data with long-term dependencies. LSTM can effectively remember past traffic patterns and infer future trends accordingly.

Model training and optimizing:

In the training process, the optimal configuration capable of achieving the minimum Mean Square Error (MSE) on the test set is finally determined by adjusting the following super parameters and combining with a k-fold cross validation strategy:

Learning rate, which is to control the speed of weight updating in each iteration.

Batch size-the number of samples used per training is defined.

The number of hidden layers increases the complexity and expressive power of the model. In addition, an online learning mechanism is introduced to allow the model to automatically update weights along with the arrival of new data so as to maintain prediction accuracy in consideration of the dynamic property and unpredictability of the network environment.

Deployment and application

System integration and daily maintenance:

After the intelligent prediction system is formally online, the intelligent prediction system is seamlessly integrated into an existing Operation Support System (OSS) platform and becomes an integral part of daily maintenance work. Whenever it is desired to evaluate the network performance in an area within a few hours of the future, the system will quickly generate a detailed predictive report containing the predicted bandwidth requirements, the likely congestion hot spot locations, and the proposed mitigation strategies by simply entering the latest set of real-time data collected. For example, before predicting that the upcoming morning commute peak period may cause the overload of base stations near some commercial areas, the administrator may adjust the power setting between adjacent cells in advance or temporarily add additional spectrum resource allocation according to the pre-warning information, so as to avoid the service interruption phenomenon.

Long-term planning and extension:

Furthermore, the system can help operators to plan long-term network expansion plans more accurately. By periodically analyzing the historical flow peaks in different seasons, holidays and major activities, the change rule of the public communication demands can be accurately mastered, and scientific basis is provided for infrastructure construction. Moreover, as the whole process is highly automated, the manual intervention requirement is reduced, and the working efficiency and the service response speed are remarkably improved. Most importantly, by virtue of the prospective and accuracy of the network condition, the internet surfing experience of the terminal user is greatly improved, the discontent emotion caused by network faults is reduced, and the growth of enterprise images and market share is promoted.

Dynamic adjustment of quality of service

Defining a service level and a service policy:

Further, the quality of service setting is dynamically adjusted according to real-time traffic demand. Specifically, bandwidth resources are allocated according to factors such as different application types, user roles and the like by defining different service levels and service policies. For example, for video conference or online game service with extremely high real-time requirement, the system allocates dedicated transmission channel and starts high-level error correction mechanism to reduce delay and packet loss phenomena to the greatest extent, while for non-critical tasks, the limiting conditions are properly relaxed to allow them to share the residual bandwidth without affecting the overall performance, thus not only ensuring that important services are not interfered, but also improving the utilization rate of the whole network resources.

Implementation details:

platinum level is suitable for applications extremely sensitive to delay, such as financial transaction systems, real-time video conferences and the like, and the services require the highest bandwidth guarantee, the lowest delay time and the smallest packet loss rate.

Gold-level-applications that require higher stability and response speed but can accept a certain degree of delay, such as online games, remote educational platforms, etc.

Silver-level, which covers activities such as general internet browsing, mail sending and receiving, and the like, the requirements of the activities on bandwidth are relatively low, and flexible allocation can be performed on the premise of not affecting user experience.

Copper level-for applications that are not real-time or that can tolerate large delays, such as file downloads, bulk data processing, etc., these tasks can be performed during network idle periods to fully utilize the remaining bandwidth resources.

Application of indirect bridging scheme

Application of CPE bridge equipment:

when the problem of insufficient network coverage in remote areas or specific occasions is faced, CPE bridge equipment is adopted, and network links are dynamically optimized. This includes automatically adjusting transmit power based on real-time environmental changes, avoiding occupied or interfered channels with dynamic frequency selection functions, and dynamically adjusting quality of service settings based on real-time traffic demands. For example, at schools in a rural area, network usage is not large at ordinary times, but network demand is suddenly increased during holiday student return. By means of the dynamic optimization function of the CPE device, the system is able to automatically recognize this variation and adjust the relevant parameters, ensuring that the basic quality of service is maintained even in case of poor network conditions.

Dynamic optimization measures:

When detecting that other wireless devices exist in the surrounding environment, the CPE device intelligently reduces the self-transmitting power, reduces unnecessary energy consumption and avoids mutual interference with other devices; conversely, if the signal strength is insufficient or the quality is degraded, the transmission power is appropriately increased, and a stable connection is ensured.

With Dynamic Frequency Selection (DFS), CPE devices can scan the available frequency bands, identify and avoid channels that have been occupied or severely interfered with, by means of built-in DFS functionality, thereby effectively reducing the risk of congestion.

Based on the dynamic adjustment of the application type, the CPE device can dynamically allocate bandwidth resources according to predefined service levels and service policies according to different types of traffic demands. For example, for video conferencing or online gaming services where real-time requirements are extremely high, the system may allocate dedicated transmission channels and enable high-level error correction mechanisms to minimize delay and packet loss, while for non-critical tasks, constraints may be relaxed appropriately to allow them to share the remaining bandwidth without affecting overall performance.

Continuous improvement and feedback loop

Performance monitoring and assessment:

finally, to ensure stable and reliable operation of the system, a sophisticated performance monitoring system is built for tracking the evaluation model performance and a fast response team is set up to handle unforeseen circumstances. User feedback and technical index performance are collected regularly and used as important reference bases for subsequent optimization decision logic, a virtuous circle is formed, and stability and flexibility of enterprise network infrastructure are improved continuously.

Technology update and evolution upgrade:

In addition, with the development of new technology and the change of application scenes, the latest research results and technical trends in academia are continuously focused, the actual problems and improvement suggestions fed back by first-line operation and maintenance personnel are actively collected, the best practice cases are discussed together by periodic organization across department seminars, and the evolution and the upgrading of the whole decision-making system are promoted. For example, when the 5G technology is gradually popularized, the system can adapt to new communication standards in time to ensure that the technology is always at the front, and when the edge calculation becomes trend, the possibility of migrating part of core processing tasks to be executed on a miniaturized low-power-consumption computing node close to a data source position is explored, so that delay time is reduced, response speed is improved, and user experience is improved.

The application can be realized by adopting or referring to the prior art at the places which are not described in the application.

In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments.

The foregoing is merely exemplary of the present application and is not intended to limit the present application. Various modifications and variations of the present application will be apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. which come within the spirit and principles of the application are to be included in the scope of the claims of the present application.

Claims (10)

1. An intelligent network prediction and dynamic optimization method based on machine learning is characterized by comprising the following steps:

Predicting a future network condition change trend through a machine learning algorithm based on the historical data and the detection result to obtain a prediction result;

constructing decision logic based on field expertise and experience;

Generating a deployment proposal according to the prediction result and the decision logic, and selecting a deployment proposal according to the deployment proposal.

2. The method according to claim 1, wherein the predicting future network condition change trend by a machine learning algorithm based on the historical data and the detection result is performed to obtain a prediction result, specifically:

acquiring the historical data and the detection result, and preprocessing;

extracting features of the preprocessed data;

based on the trained model, obtaining the prediction result according to the characteristics;

the detection result comprises network performance indexes, environment parameters and log records which are measured through preset time intervals;

and selecting the detection result of a similar time period from the historical data of the past year.

3. The method according to claim 1, wherein the building of decision logic is based on domain expertise and experience, in particular:

And determining key factors influencing network performance and early warning thresholds by combining experience of field experts and past cases, and setting specific rules to construct the decision logic.

4. The method according to claim 1, wherein the generating deployment suggestions from the prediction results and the decision logic, selecting a deployment scenario according to the deployment suggestions, in particular:

Inputting the prediction result into the decision logic, and checking whether the prediction result meets the condition of one or more rules in the decision logic or not so as to obtain the deployment suggestion;

Supplementing the decision logic in combination with a trained machine learning model;

carrying out multi-factor comprehensive evaluation on the deployment suggestion, wherein the multi-factor comprehensive evaluation comprises cost benefit analysis, risk evaluation and implementation difficulty;

The deployment scheme is selected through the deployment proposal, and the deployment scheme comprises a direct deployment scheme and an indirect bridging scheme.

5. The method of claim 4, wherein the direct deployment scenario specifically optimizes existing device configuration, including parameter adjustment and network topology optimization.

6. The method according to claim 4, characterized in that the indirect bridging scheme is in particular:

adopting CPE network bridge equipment and dynamically optimizing a network link;

the dynamic optimization comprises automatically adjusting the transmitting power according to the change of the real-time environment;

Using dynamic frequency selection function to make access point identify and avoid occupied or interfered channel;

And dynamically adjusting the service quality setting according to the real-time service requirement.

7. The method of claim 6, wherein the dynamic adjustment of the qos setting is based on real-time traffic demand, specifically by defining different service levels and service policies, bandwidth resources are allocated based on application type and user role factors.

8. An electronic device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any one of claims 1 to 7 when the computer program is executed.

9. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 7.

10. A computer program product comprising instructions which, when run on a device, cause the device to perform the steps of implementing the method of any of claims 1 to 7.