CN106507398A - A Network Self-optimization Method Based on Continuous Learning - Google Patents

Abstract

The invention discloses a network self-optimization method based on continuous learning; it includes a continuous learning process and a network optimization process; the network self-optimization method provided by the invention can greatly reduce the input of manpower and material resources, save costs, shorten the optimization process, and improve optimization efficiency , and at the same time solve the defects of the above-mentioned invention that the optimization time is long and may not be the best optimization strategy; quickly discover problems in the network, shorten the duration of network failures, restore the normal working state of the network in time, and achieve the purpose of optimizing network performance .

Description

A Network Self-optimization Method Based on Continuous Learning

technical field

The invention belongs to the field of network optimization, in particular to a network self-optimization method based on continuous learning.

Background technique

With the large-scale growth of mobile devices and the rapid development of wireless services, users' requirements for wireless network access quality are also increasing. Therefore, if any user can obtain quality-guaranteed services at any time and any place, it is necessary to Continuously optimize the network.

Due to the explosive growth of users and services in wireless communication systems, network management and optimization often become very difficult. The first problem facing network operators is that there are thousands of network elements (such as base stations) in the network, and each of these network elements may have problems or failures, some of which may not be obvious in its initial stage. If you have to rely on manual discovery and troubleshooting, the workload will be huge or even impossible. Until now, network optimization still requires a large number of professional talents and sophisticated equipment, which greatly increases the cost of operators and reduces the profit margin. At present, the high operation and maintenance costs incurred by operators can hardly be compensated by the benefits brought about by the improvement of additional service performance, because the average revenue of each user is constantly declining.

For example, a method and device for network self-optimization with the application number of 201410371501.0; the method and device for network self-optimization provided by this invention firstly obtains a special optimization strategy, and then sequentially executes each sub-optimization included in the special optimization strategy Strategy, and evaluate the effect of each sub-optimization strategy that has been executed to determine whether the preset effect index is reached. When it is reached, the entire self-optimization process is completed, and the self-optimization process is ended, otherwise continue to execute the next step For the sub-optimization strategy, repeat the above steps. Compared with the manual optimization method in the prior art, the improved invention realizes network self-optimization through a special optimization strategy including a series of sub-optimization strategies, and evaluates the effect after each sub-optimization strategy is executed. The strategy-oriented self-optimization method of the evaluation mechanism, compared with the optimization process of manually finding problems, analyzing problems, and solving problems, has less manual labor, so it can reduce the input of corresponding manpower and material resources in the optimization process and save costs. The self-optimization method can be automatically executed after being triggered, and has a high degree of automation, which can shorten the optimization process and improve optimization efficiency. However, this invention simply executes the optimization strategies sequentially until the optimization effect improves. There are two defects: 1. Selecting the way to execute the optimization strategy sequentially will lead to long optimization time; 2. When the optimization effect is improved, the entire self-optimization process will end, which may cause the optimization strategy to be not the optimal strategy. .

Contents of the invention

The technical problem to be solved by the present invention is to propose a network self-optimization method based on continuous learning in order to solve the shortcomings and deficiencies in the prior art; it reduces the input of manpower and material resources, saves costs, shortens the optimization process, and improves the optimization efficiency. To solve the above-mentioned defects that the optimization time of the invention is long and may not be the best optimization strategy.

The present invention adopts the following technical solutions to solve the above-mentioned technical problems

A network self-optimization method based on continuous learning, including a continuous learning process and a network optimization process;

The continuous learning process specifically includes the following steps;

Step 1, prepare classified network optimization problems;

Step 2, extracting the feature information of the network optimization problem prepared in step 1, and training the extracted feature information into a prior model;

Step 3, using the prior model to classify subsequent network problems;

Step 4, if there is a new network optimization problem, use unsupervised learning to extract the feature information of the new network optimization problem;

Step 5, update and improve the prior model trained in step 2 through active learning according to the feature information obtained in step 4;

The network optimization process specifically includes the following steps:

Step 6, after the classification of known network optimization problems, obtain the optimization strategy belonging to this type of network optimization problems;

Step 7, implementing an optimization strategy for this type of network optimization problem;

Step 8, obtaining the optimization effect corresponding to the executed optimization strategy;

Step 9: Evaluate each obtained optimization effect according to a preset effect evaluation index;

Step 10, compare all evaluations, and select the best network optimization strategy.

As a further preferred solution of the continuous learning-based network self-optimization method of the present invention, in step 2, the feature information of the network optimization problem prepared in step 1 is extracted by using a deep learning method.

As a further preferred solution of a network self-optimization method based on continuous learning in the present invention, the optimization strategies are respectively an optimization strategy for coverage problems and an optimization strategy for signal quality problems;

Among them, the coverage problem optimization strategy includes adjusting the antenna feed, adjusting the pilot power, adjusting the system coverage, and checking whether the adjacent station RxLev is normal;

Signal quality optimization strategies include PCI optimization, antenna feed adjustment, dominant coverage enhancement, and pilot power adjustment.

As a further preferred solution of the continuous learning-based network self-optimization method of the present invention, the effect evaluation indicators include coverage indicators, call establishment indicators, call maintenance indicators and delay indicators

As a further preferred solution of the continuous learning-based network self-optimization method of the present invention, the coverage indicators include RSRP, RSSI, RSRQ, and SINR.

As a further preferred solution of the continuous learning-based network self-optimization method of the present invention, the call establishment indicators include RRC connection establishment success rate, RRC connection establishment success rate, E-RAB establishment success rate, and wireless connection rate.

As a further preferred solution of the continuous learning-based network self-optimization method of the present invention, the call holding indicators include RRC connection abnormal call drop rate and E-RAB call drop rate.

As a further preferred solution of the continuous learning-based network self-optimization method of the present invention, the delay class indicators include UE transition delay from Idle state to Active state, Attach delay, user plane delay, and X2 switching services within the system Interruption delay, intra-system S1 switching service interruption delay, inter-system switching service interruption delay.

Compared with the prior art, the present invention adopts the above technical scheme and has the following technical effects:

The network self-optimization method provided by the present invention can greatly reduce the input cost of manpower and material resources, reduce the cost of network construction and maintenance; automatically and quickly find problems in the network, shorten the duration of network failure, and restore the normal work of the network in time State, to achieve the purpose of optimizing network performance.

Description of drawings

Fig. 1 is the continuous learning process of the present invention;

Fig. 2 is a schematic diagram of the prior model of the present invention;

Fig. 3 is a schematic diagram of the updated prior model of the present invention;

Fig. 4 is the network optimization process of the present invention;

Fig. 5 is the database provided by Example 1 of the present invention.

detailed description

Below in conjunction with accompanying drawing, technical scheme of the present invention is described in further detail:

A network self-optimization process based on continuous learning is as follows: including a continuous learning process and a network optimization process;

As shown in Figure 1, the process of continuous learning is as follows:

Step 1, prepare the classified network optimization problem; use the deep learning method to extract the characteristic information of the network optimization problem prepared in step 1.

Step 2, extracting the feature information of the network optimization problem prepared in step 1, and training the extracted feature information into a prior model;

Step 3, using the prior model to classify subsequent network problems;

Step 4, if there is a new network optimization problem, use unsupervised learning to extract the feature information of the new network optimization problem;

Step 5, use active learning to gradually update and improve the model according to the feature information obtained in step 4, so as to achieve the purpose of continuous learning.

The purpose of continuous learning is to refine the prior model and classify different network optimization problems. After classification, different optimization strategies can be used for different optimization problems. The essence of the prior model is the correspondence between feature information and problem categories, as shown in the figure:

When a new network problem occurs, it means the emergence of new feature information. Assuming that the prior model is shown in Figure 2, when a new network problem appears, its feature information is feature information 1, feature information 3, feature information 4 and new feature information 9, then the network problem will be classified into categories 1. At the same time, the prior model will be updated, and the feature information 9 will also be pointed to the problem category 1. The prior model in Figure 2 will be updated as shown in Figure 3.

After classifying the network optimization problem, start the self-optimization of the network, as shown in Figure 4, the specific process is as follows:

Step 6, after the classification of known network problems, obtain the optimization strategy belonging to this type of optimization problem;

Step 7, execute all optimization strategies of this type;

Step 8, obtaining the optimization effect corresponding to the executed optimization strategy;

Step 9, evaluating each optimization effect according to the preset effect evaluation index;

Step 10, comparing all evaluations, and selecting the best network optimization strategy. The optimization strategies described are respectively an optimization strategy for coverage problems and an optimization strategy for signal quality problems;

Among them, the coverage problem optimization strategy includes adjusting the antenna feed, adjusting the pilot power, adjusting the system coverage, and checking whether the adjacent station RxLev is normal;

Signal quality optimization strategies include PCI optimization, antenna feed adjustment, dominant coverage enhancement, and pilot power adjustment.

The effect evaluation indicators include coverage indicators, call establishment indicators, call maintenance indicators and delay indicators

The coverage indicators include RSRP, RSSI, RSRQ, and SINR.

The call establishment indicators include RRC connection establishment success rate, RRC connection establishment success rate, E-RAB establishment success rate, and wireless connection rate.

The call holding indicators include RRC connection abnormal call drop rate and E-RAB call drop rate.

The delay indicators include UE transition delay from Idle state to Active state, Attach delay, user plane delay, intra-system X2 handover service interruption delay, intra-system S1 handover service interruption delay, and inter-system handover service interruption delay.

The network self-optimization method provided by the present invention first uses a continuous learning method to train a priori model that is continuously updated and optimized, then classifies unclassified network optimization problems, and then selects an optimization strategy belonging to this category to optimize the network. Finally, the preset evaluation index is used to evaluate the effect and select the best optimization strategy.

Specific examples are as follows:

Example 1, when an unknown problem occurs in the network, and the problem is a new network problem, the processing steps are as follows:

Step 1: first extract feature information;

Step 2: Then use the prior model to classify the problem;

Step 3: Then use the extracted feature information to update the prior model;

Step 4: Suppose the problem is classified as a downlink signal quality problem (network problem category 1), there are three optimization strategies for this type of problem;

Step 5: Execute all three optimization strategies of this type to obtain the corresponding optimization effect;

Step 6: Evaluate the three optimization effects according to the preset effect evaluation indicators;

Step 7: Compare the three evaluations and choose the best network optimization strategy;

The database involved in this example is shown in Figure 5:

Example 2, when a problem occurs in the network and the problem has been classified, the processing steps are as follows:

Step 1: first extract feature information;

Step 2: Then use the prior model to classify the problem;

Step 3: Suppose the problem is classified as a downlink signal quality problem (network problem category 1), there are three optimization strategies for this type of problem;

Step 4: Execute all three optimization strategies of this type to obtain the corresponding optimization effect;

Step 5: Evaluate the three optimization effects according to the preset effect evaluation indicators;

Step 6: Compare the three evaluations and choose the best network optimization strategy.

The key points in continuous learning are: preparing a series of classified network optimization problems and using deep learning methods or models to extract the feature information and train it into a priori model. These two operations do not need to be completed by the user. After the network operator trains a prior model, the model will be continuously updated and improved through continuous learning.

The key to implementing the optimization strategy is not to execute in order until the preset evaluation index is met, but to execute all the optimization strategies under this category, and select the strategy with the best evaluation effect to apply to the network.

Claims (8)

1. A network self-optimization method based on continuous learning, characterized in that: comprising a continuous learning process and a network optimization process; The continuous learning process specifically includes the following steps; Step 1, prepare classified network optimization problems; Step 2, extracting the feature information of the network optimization problem prepared in step 1, and training the extracted feature information into a prior model; Step 3, using the prior model to classify subsequent network problems; Step 4, if there is a new network optimization problem, use unsupervised learning to extract the feature information of the new network optimization problem; Step 5, update and improve the prior model trained in step 2 through active learning according to the feature information obtained in step 4; The network optimization process specifically includes the following steps: Step 6, after the classification of known network optimization problems, obtain the optimization strategy belonging to this type of network optimization problems; Step 7, implementing an optimization strategy for this type of network optimization problem; Step 8, obtaining the optimization effect corresponding to the executed optimization strategy; Step 9: Evaluate each obtained optimization effect according to a preset effect evaluation index; Step 10, compare all evaluations, and select the best network optimization strategy. 2. A network self-optimization method based on continuous learning according to claim 1, characterized in that: in step 2, the feature information of the network optimization problem prepared in step 1 is extracted by using a deep learning method. 3. a kind of network self-optimization method based on continuous learning according to claim 1, is characterized in that: in step 6, described optimization strategy is respectively the optimization strategy of coverage class problem and the optimization strategy of signal quality class problem ; Among them, the coverage problem optimization strategy includes adjusting the antenna feed, adjusting the pilot power, adjusting the system coverage, and checking whether the adjacent station RxLev is normal; Signal quality optimization strategies include PCI optimization, antenna feed adjustment, dominant coverage enhancement, and pilot power adjustment. 4. a kind of network self-optimization method based on continuous learning according to claim 1, is characterized in that: in step 9, described effect evaluation index comprises coverage class index, call establishment class index, call hold class index and time extended index. 5. A kind of network self-optimization method based on continuous learning according to claim 4, characterized in that: said coverage index comprises RSRP, RSSI, RSRQ, SINR. 6. a kind of network self-optimization method based on continuous learning according to claim 4, is characterized in that: described call establishment class index comprises RRC connection establishment success rate, RRC connection establishment success rate, E-RAB establishment success rate, Wireless connection rate. 7. A kind of network self-optimization method based on continuous learning according to claim 4, characterized in that: said call holding class index comprises RRC connection abnormal call drop rate, E-RAB call drop rate. 8. A network self-optimization method based on continuous learning according to claim 4, characterized in that: said delay class indicators include UE transition delay from Idle state to Active state, Attach delay, user plane delay , Intra-system X2 switching service interruption delay, intra-system S1 switching service interruption delay, and inter-system switching service interruption delay.