CN119167166A

CN119167166A - Circuit breaker fault detection optimization method and system based on convolutional neural network

Info

Publication number: CN119167166A
Application number: CN202411192109.XA
Authority: CN
Inventors: 刘春平; 张晓华; 王云会; 方建博; 郎永波; 谢震伟; 唐术公; 马超; 徐利广
Original assignee: Yanbian Power Supply Co of Jilin Electric Power Co Ltd
Current assignee: Yanbian Power Supply Co of Jilin Electric Power Co Ltd
Priority date: 2024-08-28
Filing date: 2024-08-28
Publication date: 2024-12-20

Abstract

本发明公开了一种基于卷积神经网络的断路器故障检测优化方法及系统，包括：获取断路器处于运行状态时产生的原始振动信号；对原始振动信号进行预处理，基于卷积神经网络对预处理后的原始振动信号进行特征提取，得到提取后的目标特征数据集；构建MTBO‑CNN模型，以CNN卷积神经网络作为基础模型，并引入MTBO算法自动搜索最佳参数配置；利用所述训练集对MTBO‑CNN模型进行训练，得到训练好的MTBO‑CNN模型；将提取到的目标特征数据输入至训练好的MTBO‑CNN模型，采用优化后的网络对断路器振动信号进行分类，得到断路器故障识别结果，能够解决传统CNN模型不稳定的问题，基于MTBO算法能够稳定地搜索到CNN的最优参数，提高系统的性能稳定性。The invention discloses a circuit breaker fault detection optimization method and system based on convolutional neural network, comprising: obtaining an original vibration signal generated when the circuit breaker is in an operating state; preprocessing the original vibration signal, extracting features of the preprocessed original vibration signal based on the convolutional neural network, and obtaining an extracted target feature data set; constructing a MTBO-CNN model, taking the CNN convolutional neural network as a basic model, and introducing the MTBO algorithm to automatically search for an optimal parameter configuration; training the MTBO-CNN model using the training set to obtain a trained MTBO-CNN model; inputting the extracted target feature data into the trained MTBO-CNN model, using the optimized network to classify the circuit breaker vibration signal, and obtaining a circuit breaker fault identification result, which can solve the problem of instability of the traditional CNN model, can stably search for the optimal parameters of the CNN based on the MTBO algorithm, and improve the performance stability of the system.

Description

Breaker fault detection optimization method and system based on convolutional neural network

Technical Field

The invention relates to the technical field of electrical equipment overhaul, in particular to a circuit breaker fault detection optimization method and system based on a convolutional neural network.

Background

High voltage circuit breakers are indispensable elements in the electrical mains wiring of the power system, playing a critical role in maintaining the stability, reliability and economy of the power system. The power grid switching device has the main functions of switching normal current to cope with power grid load change and ensure normal operation of a circuit, and switching off fault current to isolate a fault area and protect safe operation of a power system. When high voltage circuit breakers lose reliability or mechanical component damage occurs, two major problems are presented. On the one hand, instability of the power system or the power equipment may occur. On the other hand, relay protection can not be triggered accurately, and the capability of quick response is lost. Thus, the operation of the upper level protection can be initiated, the fault range is enlarged, the selectivity is lost, and the disconnection of the whole system can be caused when the operation is serious.

At present, fault diagnosis is carried out on a circuit breaker, which mainly comprises preprocessing of wavelet transformation, empirical mode decomposition and Hilbert transformation for taking an envelope curve method on diagnosis signals, and then fault classification is carried out by adopting algorithms such as a support vector machine, a random forest, a neural network and the like. In fault diagnosis, the presence of noise and outliers may negatively impact the accuracy of the model. High-dimensional sparse data can cause node splitting of the decision tree to be difficult, and more samples may be required for efficient training and modeling. Convolutional neural networks (Convolutional Neural Network, CNN) typically have better modeling capabilities when dealing with non-linearity problems than support vector machines. CNNs can learn complex nonlinear relationships by using nonlinear activation functions and multi-layer network structures to better accommodate modeling requirements of nonlinear problems. However, since the parameters of the neural network need to be determined manually, the selection of the super-parameters is critical. Once the selection is inaccurate, it is easy to get into a locally optimal solution. While the performance of CNNs is highly dependent on their parameter settings, efficient parameter optimization is required. Conventional CNN methods typically use randomly initialized parameters, which may result in poor model performance, require long training and manual parameter adjustment to achieve satisfactory performance, which increases system complexity and computational overhead, and current techniques rely on domain experts to select and adjust the CNN parameter settings, which limits their widespread use because not every operator has expertise in the relevant domain.

Disclosure of Invention

The invention aims to provide a circuit breaker fault detection optimization method, device and equipment based on a convolutional neural network, which aim to solve the problem of instability of a traditional CNN model, and can stably search optimal parameters of CNN based on MTBO algorithm so as to improve the performance stability of a system.

The invention provides a circuit breaker fault detection optimization method based on a convolutional neural network, which comprises the following steps:

Acquiring an original vibration signal generated when the circuit breaker is in an operating state;

preprocessing the original vibration signal, and extracting features of the preprocessed original vibration signal based on a convolutional neural network to obtain an extracted target feature data set;

Constructing MTBO-CNN model, taking CNN convolutional neural network as basic model, introducing MTBO algorithm to automatically search the optimal parameter configuration, and optimizing the super parameters of MTBO-CNN model;

Dividing a target characteristic data set into a training set and a testing set according to a preset proportion, training the MTBO-CNN model by using the training set, and testing the MTBO-CNN model by using the testing set to obtain a trained MTBO-CNN model;

Inputting the extracted target characteristic data into a trained MTBO-CNN model, and classifying the breaker vibration signals by adopting an optimized network to obtain a breaker fault identification result.

Preferably, the acquiring the original vibration signal generated when the circuit breaker is in the operation state further includes:

constructing a vibration signal acquisition system with a circuit breaker, arranging a vibration sensor on the circuit breaker, and starting a working operation mode to simulate different mechanical fault types of the circuit breaker;

The method comprises the steps that a signal acquisition device is activated based on the vibration signal acquisition system to capture an analog vibration signal in a mechanical state of the circuit breaker, the captured analog vibration signal is converted into a digital signal, acquisition and recording are carried out according to a preset sampling rate, and the acquired analog vibration signal is transmitted to a fault diagnosis system to carry out fault detection operation of the circuit breaker;

The vibration signal acquisition system comprises a circuit breaker, a signal acquisition device and a fault diagnosis system which are sequentially connected, the signal acquisition device is powered by an external power supply to ensure the normal operation of acquisition, and the simulated vibration signals comprise base screw looseness, iron core jamming and insufficient lubrication of the circuit breaker and related fault signals with different degrees which are simulated and generated by adjusting corresponding mechanical parts on the circuit breaker.

Preferably, in the construction MTBO-CNN model, a CNN convolutional neural network is adopted to sequentially perform signal processing and feature extraction according to the operation sequence of the neural network, so that multi-level information fusion can be efficiently realized, the CNN convolutional neural network comprises an input layer, a one-dimensional convolutional layer, a two-dimensional convolutional layer, a correction layer, a maximum value pooling layer and a full connection layer which are sequentially connected, an original vibration signal is received from a test source output as the input of the input layer, and is transmitted to the next network layer by layer, and a fault identification result is output.

Preferably, the introducing MTBO algorithm automatically searches the optimal parameter configuration, so as to optimize the super parameters of the MTBO-CNN model comprises:

Defining the parameter combination of the CNN convolutional neural network as theta= { alpha, k, n _k }, so as to obtain the optimal parameter combination theta, maximizing the objective function F (theta) on the verification set as a target, and setting the objective function as a weighted combination of the classification accuracy and the training time

F(θ)=Accuracy(θ)-λTraining Time(θ);

Wherein, alpha represents learning rate, k represents convolution kernel size, n _k represents convolution kernel number, accuracy (theta) is classification Accuracy under parameter combination theta, TRAINING TIME (theta) is training time under parameter combination theta, lambda is time penalty coefficient;

Initializing parameters in the super parameters based on a mountain climbing optimization algorithm, and defining mountain climbing teams, wherein each mountain climbing team comprises a plurality of mountain climbing teams, each mountain climbing team corresponds to an initial parameter combination theta _i＝{α_i,k_i,n_k,i, and i is a subgroup index;

And training the CNN model by taking each initial parameter combination thetai as a super parameter of the CNN model, calculating the corresponding Accuracy Accuracy (theta) on a verification set, recording training time TRAINING TIME (theta), calculating an objective function F (theta), and evaluating and optimizing the initial parameter combinations thetai according to a calculation result.

Preferably, the introducing MTBO algorithm automatically searches the optimal parameter configuration, so as to optimize the super-parameters of the MTBO-CNN model further comprises:

searching and updating each sub-group according to the current adaptability based on a search strategy corresponding to MTBO;

if a local search is performed, the gradient of each initial parameter combination is calculated:

Wherein the method comprises the steps of A gradient representing a current initial parameter combination; representing the deviation of the objective function to the learning rate alpha; A partial derivative of the convolution kernel size k for the objective function; The partial derivative of the number n _k of the convolution kernels is the target function;

the parameter combination after the local search is as follows:

Wherein eta is a step size parameter, thetai is a current initial parameter combination, and theta _i-local is a parameter combination after local search;

If the cross-population learning is performed, the parameters of the self are adjusted by information sharing, and the parameter combination with the best fitness at the t generation is set as θ, then:

θ_i-cross＝γ·θ*+(1-γ)·θ_i-local;

Wherein, gamma is information reference factor, theta is parameter combination with optimal adaptability in the t generation, and theta _i-cross is parameter combination after cross-group learning.

Synthesizing the parameter combinations subjected to the cross-population learning, and selecting an optimal parameter combination theta ^best:

θ^best＝argmaxF(θ_i-cross)

wherein argmax is expressed as an argument that makes the function take a maximum value, and F (θ _i-cross) is an objective function to be maximized;

Introducing random variation to maintain the diversity of the search, the updated new parameter combination θ (i, new) is expressed as follows:

θ_i,new＝θ_i-cross+∈·ramdn();

wherein, E is variation intensity, randn () is standard normal random noise;

and replacing the original thetai with the updated new parameter combination theta (i, new) to serve as a starting point of the next iteration.

judging whether the optimizing process meets the following termination conditions:

Determining whether the maximum number of iterations Tmax is reached and/or,

Judging whether the corresponding fitness value of the current optimal combination is improved or not according to the comparison between the current t-generation optimal fitness and the t-1 generation optimal fitness, namely:

Where, δ is the convergence threshold, The optimal result is the fitness value under the t generation optimal parameter combination theta best;

If the termination condition is met, after the optimization process is finished, determining a final optimal parameter combination configuration theta ^best and training a final model;

if the termination condition is not satisfied, the step of evaluating and optimizing the initial parameter combination thetai is returned to continue iteration.

The invention also provides a circuit breaker fault detection optimization system based on the convolutional neural network, which comprises:

the data acquisition module is used for acquiring an original vibration signal generated when the circuit breaker is in an operating state;

The preprocessing module is used for preprocessing the original vibration signals, and extracting the characteristics of the preprocessed original vibration signals based on a convolutional neural network to obtain an extracted target characteristic data set;

MTBO-CNN model, which is used to construct MTBO-CNN model, takes CNN convolutional neural network as basic model, and introduces MTBO algorithm to automatically search the optimal parameter configuration, so as to optimize the super parameter of MTBO-CNN model;

the model training module is used for dividing the target characteristic data set into a training set and a testing set according to a preset proportion, training the MTBO-CNN model by using the training set, and testing the MTBO-CNN model by using the testing set to obtain a trained MTBO-CNN model;

The fault detection module is used for inputting the extracted target characteristic data into a trained MTBO-CNN model, and classifying the vibration signals of the circuit breaker by adopting an optimized network to obtain a fault identification result of the circuit breaker.

The invention also provides an electronic device, comprising:

a memory for storing a processing program;

And the processor is used for realizing the breaker fault detection optimization method based on the convolutional neural network when executing the processing program.

The invention also provides a readable storage medium, wherein a processing program is stored on the readable storage medium, and the processing program realizes the breaker fault detection optimization method based on the convolutional neural network when being executed by a processor.

Aiming at the prior art, the invention has the following beneficial effects:

The invention provides a fault detection optimization method of a circuit breaker based on a convolutional neural network, which is characterized in that optimal parameter configuration of the CNN is automatically searched through MTBO algorithm, the requirement of manual parameter adjustment is remarkably reduced, the efficiency of model adjustment is improved, the application of MTBO algorithm ensures that the optimal solution is quickly and effectively found in a parameter space, the inefficiency and blindness which possibly occur in the traditional trial-and-error method are avoided, a MTBO-CNN model is adopted, a large amount of vibration signal data are learned, key features are automatically extracted, effective pattern recognition is carried out, therefore, various fault types of the circuit breaker can be accurately identified, compared with the traditional fault detection method, the method has higher accuracy and lower false alarm rate, and because MTBO algorithm considers the performance of the model on a verification set in the parameter optimization process, the obtained MTBO-CNN model has good performance and good generalization capability on a training set, can effectively respond to the newly-occurring fault types, the high-efficiency parameter optimization is greatly reduced, and the cost of professional personnel is greatly reduced. Meanwhile, the rapid and accurate fault diagnosis reduces equipment downtime, improves the overall efficiency of the system, improves the accuracy of fault detection of the circuit breaker, reduces manual intervention through automatic parameter adjustment, and improves the efficiency and reliability of fault detection.

The invention provides a mountain climbing optimization algorithm for optimizing a convolutional neural network (MTBO-CNN) by taking a breaker vibration signal as a research object. The algorithm aims at finding the global optimal solution by simulating the behavior of the mountain climbing team. Specifically, MTBO optimizes the performance of the model by adjusting the hyper-parameters of the CNN, such as convolution kernel size, number of convolution layers, activation functions, etc. Experiments prove that MTBO-CNN is obviously improved in fault diagnosis of the vibration signal of the circuit breaker. Compared with the traditional method for manually determining parameters, MTBO-CNN can automatically search the optimal super-parameter combination, so that the performance and generalization capability of the model are improved. In addition, MTBO-CNN has better robustness, and can obtain good effects on different datasets and problems.

Drawings

Fig. 1-2 are schematic diagrams illustrating steps of a circuit breaker fault detection optimization method based on convolutional neural network according to an embodiment of the present invention;

FIG. 3 is a schematic block diagram of a vibration signal acquisition system according to an embodiment of the present invention;

FIG. 4 is a diagram illustrating an exemplary signal acquisition device according to an embodiment of the present invention;

FIG. 5 is a diagram illustrating steps of an optimization algorithm for MTBO-CNN in accordance with one embodiment of the present invention;

FIG. 6 is a diagram of an exemplary neural network according to an embodiment of the present invention;

FIG. 7 is a diagram of an example of a convolutional layer in an embodiment of the invention;

FIG. 8 is a diagram illustrating an example convolution operation in a one-dimensional convolution layer according to one embodiment of the present disclosure;

FIG. 9 is a diagram illustrating an example convolution operation in a two-dimensional convolution layer according to one embodiment of the present disclosure;

FIG. 10 is a diagram of the original vibration signal normal state signal according to an embodiment of the present invention;

FIG. 11 is a diagram of an original vibration signal stuck state signal according to an embodiment of the present invention;

FIG. 12 is a diagram of an original vibration signal screw loosening status signal according to an embodiment of the present invention;

FIG. 13 is a diagram of an initial vibration signal lubrication deficiency status signal in an embodiment of the present invention;

FIG. 14 is a graph showing the iterative variation of the error in accordance with one embodiment of the present invention;

FIG. 15 is a diagram illustrating a sample analysis of the diagnosis results by the confusion matrix according to an embodiment of the present invention.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

The term "including" and variations thereof as used herein are intended to be open-ended, i.e., including, but not limited to. The term "based on" is based at least in part on. The term "one embodiment" means "at least one embodiment," another embodiment "means" at least one additional embodiment, "and" some embodiments "means" at least some embodiments. Related definitions of other terms will be given in the description below.

It should be noted that the terms "first," "second," and the like in this disclosure are merely used to distinguish between different devices, modules, or units and are not intended to limit the order or interdependence of functions performed by such devices, modules, or units.

It should be noted that references to "one" or "a plurality" in this disclosure are intended to be illustrative rather than limiting, and those of ordinary skill in the art will appreciate that "one or more" is intended to be interpreted as "one or more" unless the context clearly indicates otherwise.

Example 1

As shown in fig. 1, the invention further provides a circuit breaker fault detection optimization method based on the convolutional neural network, which comprises the following steps:

S1, acquiring an original vibration signal generated when a circuit breaker is in an operating state;

S2, preprocessing the original vibration signal, and carrying out feature extraction on the preprocessed original vibration signal based on a convolutional neural network to obtain an extracted target feature data set;

s3, constructing MTBO-CNN model, taking CNN convolutional neural network as basic model, introducing MTBO algorithm to automatically search the optimal parameter configuration, and optimizing the super parameters of MTBO-CNN model;

S4, dividing a target characteristic data set into a training set and a testing set according to a preset proportion, training the MTBO-CNN model by using the training set, and testing the MTBO-CNN model by using the testing set to obtain a trained MTBO-CNN model;

S5, inputting the extracted target characteristic data into a trained MTBO-CNN model, and classifying the vibration signals of the circuit breaker by adopting an optimized network to obtain a fault recognition result of the circuit breaker.

In the embodiment, firstly, vibration signals are collected when the circuit breaker works, and the collected signals are subjected to necessary preprocessing steps, such as denoising, normalization, feature extraction and the like, so as to ensure the reliability and the quality of data. The processed vibration signal data set is divided into a training set and a testing set according to the proportion of 8:2. The training set is used for training MTBO-CNN network models to help the models learn complex relations between vibration signals and specific fault types, and the testing set is used for evaluating generalization capability and recognition accuracy of the models. These preprocessed data are then input into a deep learning architecture that combines MTBO (Mountain Team-based Optimization) optimization algorithms with Convolutional Neural Networks (CNNs). The MTBO algorithm is used to automatically adjust the hyper-parameters of the CNN to obtain better model performance. Through multiple iterations and cross-subgroup learning, the MTBO-CNN model can effectively improve the accuracy and efficiency of fault identification, and the overall flow is shown in FIG. 2.

In this embodiment, the acquiring the original vibration signal generated when the circuit breaker is in the operating state in step S1 further includes:

In order to effectively detect faults of the circuit breaker, the embodiment designs and builds a high-precision vibration signal acquisition system, combines advanced algorithm optimization, improves accuracy and efficiency of fault identification, and specifically acquires and identifies the process as shown in fig. 3, step 1, a signal acquisition device supplies power. And 2, configuring a vibration sensor, namely attaching the vibration sensor to the upper part of an arc extinguishing chamber of the circuit breaker. And 3, vibration signal acquisition, namely activating a signal acquisition device to start to acquire vibration data, converting an analog vibration signal into a digital signal by an analog-to-digital converter (ADC) built in the acquisition device, and recording the digital signal at a preset sampling rate such as 25.6 kHz. And 4, inputting data into a fault diagnosis system, namely transmitting the acquired vibration signal data to the fault diagnosis system in a wired mode. And 5, analyzing and identifying fault data, namely firstly filtering, reducing noise and enhancing signals by a preprocessing module to highlight fault characteristics, and then carrying out state identification by using the fault diagnosis algorithm provided by the embodiment. And 6, displaying the identification result, namely displaying the analysis result through a user interface, wherein a user can directly observe the state information of the circuit breaker.

The specific construction of the vibration signal acquisition system in this embodiment is as follows:

A vibration signal acquisition system is built by adopting PCB Piezotronics accelerometer A356A 04 piezoelectric acceleration sensor and a NI 9234 data acquisition card manufactured by National Instruments (NI) company.

(1) The sensor is selected by adopting a CT1005LC type acceleration sensor because the vibration signal of the circuit breaker has the characteristics of short effective time and large vibration impact. The sensor is fixed above the arc extinction of the circuit breaker by the magnet adsorption, so that the signal repeatability is ensured and the installation is convenient. By virtue of its high performance SMR sensing element and sophisticated internal amplification circuitry, the sensor has high resolution and high sensitivity. (2) And signal acquisition, namely reading ADC data by the sensor through a parallel interface, wherein the sampling frequency is up to 1MHz. The acquired data is downsampled to 32kHz by the FPGA module, making it more suitable for subsequent signal processing and analysis, as shown in fig. 4. (3) And the acquisition setting is that the coordinate zero point of the sampling signal is set to be when the circuit breaker is about to act, so as to ensure that the data acquired each time are kept consistent. Vibration signals are captured under four mechanical states, including faults such as loosening of base screws, jamming of iron cores, insufficient lubrication and the like, and faults with different degrees are simulated by adjusting conditions of related mechanical components.

In the embodiment, the MTBO-CNN-based optimization algorithm optimizes three key parameters of the convolutional neural network, namely the learning rate, the convolutional kernel size and the convolutional kernel number, by using the highest test set precision as an objective function, and can obtain better performance and more accurate results in the network training process. In this optimization process, the mountain climbing algorithm is different from the conventional optimization algorithm. It simulates the process of a mountain climbing team climbing a mountain, by continually exploring different points and directions in the parameter space, to find better parameter combinations. Specifically, this algorithm first randomly initializes a set of parameter combinations as a starting point and calculates the corresponding test set precision. It then searches for nearby parameter points with minor changes and adjustments, and at each step evaluates the improvement in test set accuracy. If a better combination of parameters is found, the mountain climbing algorithm updates the parameters and explores the surrounding neighborhood further. This process continues until a locally optimal solution is found or a preset termination condition is reached. The algorithm flow chart is shown in fig. 5.

In the CNN identification network in this embodiment, the fault diagnosis system adopts a Convolutional Neural Network (CNN), and includes six key layers, which are an input layer, a one-dimensional convolutional layer, a two-dimensional convolutional layer, a correction layer, a maximum value pooling layer, and a full connection layer. The layers sequentially perform signal processing and feature extraction according to the operation sequence of the neural network, so that multi-level information fusion can be efficiently realized, and the accuracy of fault diagnosis is improved. A schematic of the structure is shown in fig. 6. In the construction MTBO-CNN model, a CNN convolutional neural network is adopted to sequentially perform signal processing and feature extraction according to the operation sequence of the neural network, multi-level information fusion can be efficiently realized, the CNN convolutional neural network comprises an input layer, a one-dimensional convolutional layer, a two-dimensional convolutional layer, a correction layer, a maximum value pooling layer and a full connection layer which are sequentially connected, an original vibration signal is received from test source output as input of the input layer, and is transmitted to the next network layer by layer, and a fault identification result is output.

(1) Input stage

In the input phase, the system first receives raw data of the vibration signal from the test source output and performs preprocessing. The data are arranged in time sequence, and one data sample is input every 16 periods. The process ensures the order of the input data and lays a foundation for subsequent convolution processing. Once the data is collected, it is gradually passed on to the next network layer for further processing. Wherein the convolution layer is composed of a plurality of convolution cores, each of which is composed of two multipliers and an adder, the structure of which is shown in fig. 7.

(2) One-dimensional convolution layer

The one-dimensional convolution layer is made up of 16 parallel convolution cores. After this layer of processing, each input data item can generate 16 different output data simultaneously, and the convolution operation is shown in fig. 8. The design enhances the density and effectiveness of feature extraction and improves the quality of the preliminary analysis stage.

(3) Two-dimensional convolution layer

The two-dimensional convolution layer convolves 16 data (after correction) output by the one-dimensional convolution layer, and because the 16 data are calculated through 16 convolution cores respectively, in order to save resources, 16 groups of convolution parameters are circularly input into the 16 convolution cores according to the period, and one-time two-dimensional convolution calculation is completed by using 16 clock periods, as shown in fig. 9.

(4) Correction layer

A correction operation is performed on the output of the convolutional layer, with the negative value set to 0, to ensure that all output data is non-negative, enhancing the interpretability of the model and subsequent processing stability.

(5) Maximum pooling layer

A pooling operation is applied to the output results of the two-dimensional convolution layers, extracting a maximum value from each convolution output channel. Finally, the maximum of the 16 channels is integrated as the pooling layer output.

(6) Full connection layer

And arranging the 16 values processed by the pooling layer according to the channel sequence, and carrying out convolution calculation with 4 groups of preset coefficients. And adding offset value for adjustment to obtain the weight value of each possible result. And selecting the final classification result with the largest weight value, namely judging the fault type calculated at the time.

The original convolutional neural network usually requires tedious manual parameter adjustment, including learning rate, hierarchy, and selection of activation functions. This process not only takes a lot of time, but also requires a lot of expertise to ensure the rationality of the parameters and the validity of the model. In order to simplify the process and improve the performance of the model, a MTBO optimized CNN parameter process is adopted in the embodiment, and MTBO (Mountain Team-based Optimization) algorithm is introduced to automatically optimize the parameter configuration of CNN, and the specific steps are as follows:

1. Definition problem

Setting the parameter combination of the CNN convolutional neural network as theta= { alpha, k, n _k }, so as to obtain the optimal parameter combination theta, maximizing an objective function F (theta) on a verification set as a target, and setting the objective function as a weighted combination of classification accuracy and training time:

F(θ)=Accuracy(θ)-λTrainming Time(θ);

2. Initialization of

The mountain-climbing team optimization algorithm is based on initializing each parameter in the super parameters, and the scope is defined as alpha epsilon [ alpha _min,α_max],k∈{3,5,7},n_k∈[n_k,min,n_k,max ], wherein alpha min and alpha max are respectively the minimum value and the maximum value preset by the parameter alpha, and n _k,min,n_k,max is respectively the minimum value and the maximum value preset by the parameter n _k.

Defining mountain climbing teams, wherein each mountain climbing team comprises a plurality of mountain climbing teams, each mountain climbing team corresponds to an initial parameter combination theta _i＝{α_i,k_i,n_k,i, and i is a subgroup index;

3. Parameter assessment

For each initial parameter combination θi is used as a hyper-parameter of the CNN model:

1) Model construction and training, namely constructing a CNN model and training by using training data. 2) And (3) calculating the corresponding Accuracy Accuracy (theta) on the verification set, recording the training time TRAINING TIME (theta), calculating the objective function F (theta), and evaluating and optimizing the initial parameter combination thetai according to the calculation result.

4. Updating and searching

(1) Local search:

the gradient for each initial parameter combination is calculated:

the parameter combination after the local search is as follows:

(2) Group-crossing learning:

Information sharing is carried out among the sub-groups, and the parameters of the sub-groups are adjusted according to the optimal parameter combination of other teams. The parameters of the information sharing adjustment are set to be θ, which is the optimal parameter combination for the t generation of fitness:

θ_i-cross＝γ·θ^*+(1-γ)·θ_i-local;

5. Fusion and optimization

(1) Global fusion:

θ^best＝argmaxF(θ_icross);

(2) Diversity retention:

Introducing random variation to maintain diversity of search and avoid early convergence, the new parameter combination θ (i, new) after update is expressed as follows:

θ_i,new＝θ_i-cross+∈·randn();

wherein, E is variation intensity, randn () is standard normal random noise;

(3) Parameter updating:

And replacing the original thetai with the updated new parameter combination theta (i, new) to serve as a starting point thetat _i←θ_i,new of the next iteration.

6. Termination condition

(1) Judging whether the maximum iteration number Tmax is reached;

(2) Judging whether the corresponding fitness value of the current optimal combination is improved or not according to the comparison between the current t-generation optimal fitness and the t-1 generation optimal fitness, namely:

Where, δ is the convergence threshold, The optimal result is the fitness value under the t generation optimal parameter combination theta ^best;

7. Iteration and final model

if the termination condition is not satisfied, returning to the step of evaluating and optimizing the initial parameter combination thetai, namely, continuing iteration in step 3.

Conventional fault diagnosis models often use manual parametric approaches to optimize the model, which is time consuming and requires expertise. The MTBO-CNN model introduces MTBO (Mountaineering Team-Based Optimization) algorithm, and utilizes an optimization algorithm to automatically search the optimal parameter configuration without manually adjusting parameters, thereby greatly improving the parameter optimization efficiency and accuracy. The MTBO-CNN model adopts a Convolutional Neural Network (CNN) as a basic model, and has strong feature extraction capability. CNNs are able to automatically learn and extract important features from raw data, playing a key role in fault identification and diagnosis. The MTBO-CNN model considers the influence of uncertainty and noise in the parameter optimization process, and improves the robustness and stability of the model. In addition, by optimizing CNN parameters, the MTBO-CNN model can also improve the interpretability of the model, so that engineers can better understand fault diagnosis results and judge fault reasons more accurately. Compared with the CNN model or other traditional methods for fault diagnosis, the MTBO-CNN model has advantages in parameter optimization and feature extraction, so that the accuracy and performance of fault diagnosis can be improved. By automatically optimizing the CNN parameters, a better effect is achieved in the fault diagnosis task. The MTBO-CNN model fully considers the mobility of the model in the training process, and the model can be better adapted to different breaker fault diagnosis tasks by pre-training on a large-scale data set and then fine-tuning on a target task. Therefore, the requirement for a large amount of labeling data can be reduced, and the training efficiency and generalization capability of the model are improved.

Specific examples of the fault detection flow executed based on the above-described optimization step are as follows:

(1) And (3) data acquisition, namely acquiring vibration signals by the sensor under different fault scenes to form a sample data set containing various fault degrees, wherein the original vibration signals in four different states are shown in figures 10-13.

(2) CNN is optimized by MTBO, the learning rate, the convolution kernel size and the number of convolution kernels of the convolution neural network are optimized according to a mountain climbing team optimization algorithm, and the optimized parameters are shown in the following table 1.

TABLE 1cnn optimization parameter results

(3) Training and evaluating the model, namely training the collected data by using optimized parameter configuration, quantitatively evaluating the classification performance of the model under different fault scenes, and ensuring that the network can accurately identify various mechanical faults of the circuit breaker. The error iteration change curve is drawn by taking the test error as a radius label and the iteration number as an angle label as shown in fig. 14. Analysis of the graph shows that after 6 rounds of training, the training error is reduced to 0.0008 and the training becomes stable.

In order to further examine the capacity of MTBO-CNN model to identify faults of the circuit breaker, a confusion matrix is introduced to quantitatively analyze diagnosis results. The confusion matrix reflects in detail the number of erroneous decisions and the type of erroneous decisions for different types of faults of the rolling bearing, as shown in fig. 15. In fig. 15, the abscissa represents the predicted category of different faults, the ordinate represents the real label of different faults, and the number on the main diagonal of the matrix represents the number of samples for correctly classifying each type of faults, and the result shows that 100% of identification accuracy can be realized for all the four types of circuit breakers in different states.

Example two

Based on the same conception, the invention provides a breaker fault detection optimization system based on a convolutional neural network, which comprises the following steps:

It should be noted that, the division of each module in the embodiment of the apparatus/system is merely a division of a logic function, and may be fully or partially integrated into a physical entity or may be physically separated. The modules can be realized in a form of calling the processing element through software, can be realized in a form of hardware, can also be realized in a form of calling the processing element through part of units, and can be realized in a form of hardware.

The implementation principle of the above modules has been described in the foregoing embodiments, so that a repetition of the description is omitted here.

Example III

Based on the same conception, in some embodiments of the application, an electronic device is also provided. The electronic equipment comprises a memory and a processor, wherein the memory is used for storing a processing program, and the processor executes the processing program according to the instruction. The processor executes the processing program, so that the breaker fault detection optimization method based on the convolutional neural network in the previous embodiment is realized.

In some embodiments of the present application, a readable storage medium is also provided, which may be a non-volatile readable storage medium or a volatile readable storage medium. The readable storage medium has instructions stored therein that, when executed on a computer, cause an electronic device comprising such a readable storage medium to perform the aforementioned convolutional neural network-based breaker failure detection optimization method.

It will be appreciated that for the aforementioned convolutional neural network based breaker failure detection optimization method, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in essence or a part contributing to the prior art or all or part of the technical solution in the form of a software product stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of 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 (Randomaccessmemory, RAM), a magnetic disk or an optical disk, etc. which can store the program code.

The computer readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, with readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A readable storage medium may also be any readable medium that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Program code for carrying out the disclosed aspects of the present application may be written in any combination of one or more programming languages, including an object oriented programming language such as Python, c++ or the like and conventional procedural programming languages, such as the C-language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device, partly on a remote computing device, or entirely on the remote computing device or server. In the case of remote computing devices, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., connected via the Internet using an Internet service provider).

It should be noted that the above-mentioned embodiments are merely for illustrating the technical solution of the present invention, and not for limiting the same, and although the present invention has been described in detail with reference to the above-mentioned embodiments, it should be understood by those skilled in the art that the technical solution described in the above-mentioned embodiments may be modified or some technical features may be equivalently replaced, and these modifications or substitutions do not make the essence of the corresponding technical solution deviate from the spirit and scope of the technical solution of the embodiments of the present invention.

Claims

1. A circuit breaker fault detection optimization method based on convolutional neural network, characterized by comprising:

Acquire the original vibration signal generated when the circuit breaker is in operation;

Construct the MTBO-CNN model, using the CNN convolutional neural network as the basic model, and introduce the MTBO algorithm to automatically search for the best parameter configuration, thereby optimizing the hyperparameters of the MTBO-CNN model;

Dividing the target feature data set into a training set and a test set according to a preset ratio, and using the training set to train the MTBO-CNN model, and using the test set to test the MTBO-CNN model, to obtain a trained MTBO-CNN model;

The extracted target feature data is input into the trained MTBO-CNN model, and the optimized network is used to classify the circuit breaker vibration signal to obtain the circuit breaker fault identification result.

2. The circuit breaker fault detection optimization method based on convolutional neural network according to claim 1, characterized in that the step of obtaining the original vibration signal generated when the circuit breaker is in operation further comprises:

Construct a vibration signal acquisition system with a circuit breaker, set a vibration sensor on the circuit breaker, and start a working operation mode to simulate different mechanical fault types of the circuit breaker;

Based on the vibration signal acquisition system, the signal acquisition device is activated to capture the analog vibration signal of the circuit breaker in the mechanical state, and the captured analog vibration signal is converted into a digital signal, which is collected and recorded according to a preset sampling rate, and the collected analog vibration signal is transmitted to the fault diagnosis system to perform circuit breaker fault detection operation;

Among them, the vibration signal acquisition system includes a circuit breaker, a signal acquisition device and a fault diagnosis system connected in sequence, and the signal acquisition device is powered by an external power supply to ensure the normal operation of the acquisition work; the simulated vibration signal includes loose screws on the base of the circuit breaker, stuck iron core, insufficient lubrication and related fault signals of different degrees simulated by adjusting the corresponding mechanical parts on the circuit breaker.

3. The circuit breaker fault detection optimization method based on convolutional neural network according to claim 1 is characterized in that, in the construction of the MTBO-CNN model, a CNN convolutional neural network is used to perform signal processing and feature extraction in sequence according to the operation sequence of the neural network, and multi-level information fusion can be efficiently realized. The CNN convolutional neural network includes an input layer, a one-dimensional convolutional layer, a two-dimensional convolutional layer, a correction layer, a maximum pooling layer and a fully connected layer connected in sequence, and receives the original vibration signal from the test source output as the input of the input layer, which is passed to the next network layer layer by layer to output the fault identification result.

4. The circuit breaker fault detection optimization method based on convolutional neural network according to claim 1 is characterized in that the introduction of MTBO algorithm to automatically search for the optimal parameter configuration, thereby optimizing the hyperparameters of the MTBO-CNN model comprises:

Define the parameter combination of CNN convolutional neural network as θ = {α, k, _nk } to obtain the best parameter combination θ*, so that the objective function F(θ) on the validation set is maximized as the goal, and the objective function is set as a weighted combination of classification accuracy and training time

F(θ)=Accuracy(θ)-λTraining Time(θ);

Where α represents the learning rate, k represents the convolution kernel size, and n _k represents the number of convolution kernels; Accuracy(θ) is the classification accuracy under the parameter combination θ, Training Time(θ) is the training time under the parameter combination θ, and λ is the time penalty coefficient;

Initialize each parameter in the hyperparameter based on the climbing team optimization algorithm, define a climbing team, each climbing team includes multiple climbers, each climber corresponds to an initial parameter combination θ _i ={α _i ,k _i , _nk,i }, where i is the subgroup index;

Take each initial parameter combination θi as the hyperparameter of the CNN model, train the CNN model, calculate the corresponding accuracy Accuracy(θ) on the validation set, record the training time Training Time(θ), calculate the objective function F(θ), and evaluate and optimize the initial parameter combination θi according to the calculation results.

5. The circuit breaker fault detection optimization method based on convolutional neural network according to claim 1 is characterized in that the introduction of MTBO algorithm to automatically search for the optimal parameter configuration, thereby optimizing the hyperparameters of the MTBO-CNN model further comprises:

Based on the search strategy corresponding to MTBO, each sub-population is searched and updated according to the current fitness;

If a local search is performed, the gradient is calculated for each initial parameter combination:

in Represents the gradient of the current initial parameter combination; Represents the partial derivative of the objective function with respect to the learning rate α; is the partial derivative of the objective function with respect to the convolution kernel size k; is the partial derivative of the objective function with respect to the number of convolution kernels n _k ;

The parameter combination after local search is:

Among them, η is the step size parameter; θi is the current initial parameter combination, and θ _i-local is the parameter combination after local search;

If cross-group learning is performed, the parameters are adjusted through information sharing. Assuming that the best parameter combination for fitness in the tth generation is θ*, then:

θ _i-cross =γ·θ ^* +(1-γ)·θ _i-local ;

Among them, γ is the information reference factor; θ* is the parameter combination with the best fitness in the tth generation; θ _i-cross is the parameter combination after cross-group learning.

6. The circuit breaker fault detection optimization method based on convolutional neural network according to claim 1 is characterized in that the introduction of MTBO algorithm to automatically search for the optimal parameter configuration, thereby optimizing the hyperparameters of the MTBO-CNN model further comprises:

Comprehensively combine the parameter combinations after cross-group learning and select the optimal parameter combination θ ^best :

θ ^best =argmaxF(θ _i-cross );

Among them, argmax represents the independent variable that makes the function reach the maximum value, and F(θ _i-cross ) is the objective function to be maximized;

Introducing random mutations to maintain the diversity of the search, the updated new parameter combination θ(i,new) is expressed as follows:

θ _i,new =θ _i-cross +∈·randn();

Where, ∈ is the variation intensity; randn() is the standard normal random noise;

The updated new parameter combination θ(i,new) replaces the original θi as the starting point for the next round of iteration.

7. The circuit breaker fault detection optimization method based on convolutional neural network according to claim 6 is characterized in that the introduction of MTBO algorithm to automatically search for the optimal parameter configuration, thereby optimizing the hyperparameters of the MTBO-CNN model further comprises:

Determine whether the above optimization process meets the following termination conditions:

Determine whether the maximum number of iterations Tmax has been reached; and/or,

According to the comparison between the best fitness of the current t generation and the best fitness of the t-1 generation, it is judged whether the corresponding fitness value of the current optimal combination has been improved, that is:

In the formula, δ is the convergence threshold, is the optimization result under the t-th generation optimal parameter combination θ ^best, i.e., the fitness value;

If the above termination conditions are met, the optimization process is terminated, the final optimal parameter combination configuration θ ^best is determined, and the final model is trained;

If the above termination conditions are not met, the process returns to the step of evaluating and optimizing the initial parameter combination θi and continues the iteration.

8. A circuit breaker fault detection optimization system based on convolutional neural network, characterized by comprising:

A data acquisition module, used for acquiring an original vibration signal generated when the circuit breaker is in operation;

A preprocessing module, used to preprocess the original vibration signal, and extract features from the preprocessed original vibration signal based on a convolutional neural network to obtain an extracted target feature data set;

MTBO-CNN model, used to build the MTBO-CNN model, using the CNN convolutional neural network as the basic model, and introducing the MTBO algorithm to automatically search for the best parameter configuration, thereby optimizing the hyperparameters of the MTBO-CNN model;

A model training module, used for dividing the target feature data set into a training set and a test set according to a preset ratio, and using the training set to train the MTBO-CNN model, and using the test set to test the MTBO-CNN model, to obtain a trained MTBO-CNN model;

The fault detection module is used to input the extracted target feature data into the trained MTBO-CNN model, and use the optimized network to classify the circuit breaker vibration signal to obtain the circuit breaker fault identification result.

9. An electronic device, comprising:

A memory, the memory being used to store a processing program;

A processor, wherein when executing the processing program, the processor implements the circuit breaker fault detection optimization method based on convolutional neural network as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that a processing program is stored on the readable storage medium, and when the processing program is executed by a processor, the circuit breaker fault detection optimization method based on convolutional neural network according to any one of claims 1 to 7 is implemented.