Circuit breaker fault diagnosis method and system, and computer device

Abstract

The invention belongs to the technical field of power switch equipment monitoring, and particularly relates to a circuit breaker fault diagnosis method and system and a computer device. The method comprises the following steps: firstly, obtaining characteristic quantity data of the circuit breaker, wherein the characteristic quantity is a characteristic quantity which is obtained by carrying out LDA dimension reduction processing on an obtained multi-source characteristic signal reflecting the operation condition of the circuit breaker and is required for fault diagnosis of the circuit breaker; inputting the obtained characteristic quantity data into a fault diagnosis model to obtain a circuit breaker fault classification result; the fault diagnosis model is obtained by training a machine learning model by using a training set, and the training set is a data set containing multiple groups of historical characteristic quantity data of fault classification labels. According to the method, some redundant and interference diagnosis information can be removed through LDA dimension reduction processing, enough important features are reserved, and category information of original multi-source feature signals can be accurately represented, so that the efficiency and accuracy of a fault diagnosis model are relatively high.

Description

Technical Field

[0001] The present invention belongs to the technical field of power switch equipment monitoring, and in particular relates to a circuit breaker fault diagnosis method and system, and a computer device. Background Art

[0002] Circuit breakers play an important role in the power industry. With the rapid development and changes of power systems, higher requirements are placed on the quality and performance of circuit breakers. However, according to relevant statistical reports, about 25% to 30% of power grid accidents are related to circuit breaker failures each year, and this situation has not improved in recent years. The circuit breaker online monitoring system can obtain the operating status data of the circuit breaker in real time through the online monitoring device, analyze the data and judge the health status of the circuit breaker, detect faults in time, and issue early warnings and alarms.

[0003] At present, the online monitoring of circuit breaker fault diagnosis is usually based on coil current signals, contact travel signals or mechanical vibration signals, and the collected signals are combined with artificial intelligence algorithms for single-source diagnosis. However, single-source signal diagnosis cannot fully utilize data, nor is it sufficient to distinguish all types of faults. For example, a single coil current feature confuses a large number of faults caused by the opening spring and tightness in normal situations during the opening process; although the contact travel has a high diagnosis rate, it is also difficult to distinguish faults caused by the spring during the closing process.

[0004] In order to solve this problem, a multi-source monitoring system can be used to obtain the circuit breaker's electrical, mechanical, position and other information, and the multi-source signals can be combined with artificial intelligence algorithms to perform fault diagnosis, so as to reduce the errors and deviations that may exist in a single monitoring source and improve the reliability of the data. However, a large amount of multi-source data is also a major challenge to the computer's processing power. Multi-source data makes the training process of artificial intelligence algorithms cumbersome and takes a long time to train, which affects the efficiency of fault diagnosis. In addition, there may be some redundant signals in these multi-source data, which also affects the efficiency of fault diagnosis. Furthermore, there may be some interference signals in these multi-source data, which will affect the circuit breaker fault diagnosis and the accuracy of fault diagnosis. Summary of the invention

[0005] The object of the present invention is to provide a circuit breaker fault diagnosis method and system, and a computer device, so as to solve the problem that the existing multi-source signal fault diagnosis method has low diagnostic efficiency and low diagnostic accuracy.

[0006] In order to solve the above technical problems, the present invention provides a circuit breaker fault diagnosis method, comprising the following steps: Acquire characteristic quantity data of the circuit breaker, wherein the characteristic quantity is obtained by using LDA dimension reduction processing on the multi-source characteristic signals reflecting the operation status of the circuit breaker, and is the characteristic quantity required for circuit breaker fault diagnosis; The acquired characteristic quantity data is input into the fault diagnosis model to obtain the circuit breaker fault classification result; the fault diagnosis model is obtained by training the machine learning model using the training set, and the training set is a data set containing multiple groups of historical characteristic quantity data with fault classification labels.

[0007] Furthermore, the machine learning model is an improved CatBoost model obtained by improving the traditional CatBoost model, and the improvement includes that the sampling strategy in the model adopts a Bayesian sampling strategy.

[0008] Furthermore, the improvement also includes that the Bayesian bagging control strength in the model is obtained by optimizing through a genetic algorithm.

[0009] Furthermore, the improvement also includes that the growth strategy in the model adopts a depth-first growth strategy.

[0010] Furthermore, the improvement also includes that at least one parameter among the learning rate, decision tree depth and regularization parameter in the model is obtained by optimizing through a genetic algorithm.

[0011] Furthermore, the multi-source characteristic signal includes a coil current signal, a contact travel signal and an auxiliary switch signal.

[0012] Furthermore, the acquired feature quantity data is data that has been subjected to denoising and normalization processing.

[0013] In order to solve the above technical problem, the present invention further provides a computer device, including a processor, wherein the processor is used to execute a computer program to implement the steps of the circuit breaker fault diagnosis method introduced above.

[0014] In order to solve the above technical problems, the present invention also provides a circuit breaker fault diagnosis system, including a data acquisition device, a data transmission device and a data processing device. The data acquisition device is used to collect characteristic quantity data of the circuit breaker and send it to the data processing device through the data transmission device. The data processing device includes a memory, a processor and a computer program stored in the memory. The processor is used to execute the computer program to implement the steps of the circuit breaker fault diagnosis method introduced above.

[0015] Furthermore, each software function module corresponding to the computer program is encapsulated using container technology.

[0016] Furthermore, the data received by the data processing device is stored in the memory using a persistence technology.

[0017] The present invention is an improved invention creation, and its beneficial effect is: when the present invention uses the fault diagnosis model obtained by training the machine learning model for fault diagnosis, the feature quantity used as the input is the feature quantity required for circuit breaker fault diagnosis obtained after LDA dimensionality reduction processing of the multi-source feature signal reflecting the operation status of the circuit breaker. The dimensionality reduction processing method reduces the complexity of the model, improves the training speed, and thus improves the efficiency of fault diagnosis; moreover, some redundant information can be removed through LDA dimensionality reduction processing, improves the efficiency of fault diagnosis, and can remove some information that interferes with diagnosis, retains sufficiently important features, and can accurately represent the category information of the original multi-source feature signal, so that the accuracy of the fault diagnosis model is relatively high. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a structural block diagram of a circuit breaker fault diagnosis system of the present invention; Figure 2 is a flow chart of a circuit breaker fault diagnosis method of the present invention; Figure 3a It is a relationship diagram of the sampling strategy, growth strategy and diagnostic error rate of the CatBoost model for coil current of the present invention; Figure 3b It is a relationship diagram of the sampling strategy, growth strategy and diagnostic error rate of the CatBoost model for contact travel of the present invention; Figure 4a It is a relationship diagram of the sampling strategy, growth strategy and training time of the CatBoost model for coil current of the present invention; Figure 4b It is a relationship diagram of the sampling strategy, growth strategy and training time of the CatBoost model for contact travel of the present invention; Figure 5a It is a schematic diagram of the loss value and accuracy rate of the CatBoost model using the Bayesian sampling strategy and the depth-first growth strategy for coil current during the iteration process of the present invention; Figure 5b It is a schematic diagram of the loss value and accuracy rate of the CatBoost model using the Bayesian sampling strategy and the depth-first growth strategy for the contact stroke according to the present invention during the iteration process; Figure 6a It is a tuning curve diagram of the CatBoost model using GA for parameter tuning of coil current of the present invention; Figure 6b It is a tuning curve diagram of the CatBoost model using GA for parameter tuning for contact stroke of the present invention; Figure 7 It is a comparison chart of the fault diagnosis error rates of the method of the present invention and several other machine learning algorithms; Figure 8a It is a GA-CatBoost confusion matrix diagram for coil current of the present invention; Figure 8b It is a GA-CatBoost confusion matrix diagram for contact travel of the present invention; Fig. 9 It is a schematic diagram of the fault accuracy rate after LDA dimensionality reduction processing of the present invention; Fig.10 It is a T-SNE cluster diagram of the feature quantity after LDA dimensionality reduction processing of the present invention. DETAILED DESCRIPTION

[0019] The main scheme of the present invention is: obtaining characteristic quantity data of the circuit breaker, the characteristic quantity is the characteristic quantity required for circuit breaker fault diagnosis obtained by LDA dimensionality reduction processing of the multi-source characteristic signals reflecting the operation status of the circuit breaker; inputting the obtained characteristic quantity data into the fault diagnosis model to obtain the circuit breaker fault classification result; the fault diagnosis model is obtained by training the machine learning model with the training set, and the training set is a data set containing multiple groups of historical characteristic quantity data with fault classification labels. Through LDA dimensionality reduction processing, while improving the processing speed, because LDA dimensionality reduction processing can remove redundant and interfering information, only retaining sufficiently important information, the diagnostic accuracy of the fault diagnosis model is improved.

[0020] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0021] Circuit breaker fault diagnosis system embodiment: A circuit breaker fault diagnosis system of the present invention has a structure diagram as shown in FIG. Figure 1 As shown, it includes a data acquisition device, a data transmission device and a data processing device.

[0022] The data acquisition device is used to collect multi-source characteristic signals reflecting the operation status of the circuit breaker and capture the action data of the circuit breaker. Specifically, it may include a current sensor, an auxiliary switch signal acquisition module and a travel sensor, which are used to collect the coil current signal, auxiliary switch signal and contact travel signal of the circuit breaker respectively.

[0023] The data transmission device is used to perform signal conditioning and transformation on the data collected by the data acquisition device and then output it to the data processing device. Specifically, the three-way processing of the coil current signal, the auxiliary switch signal and the contact stroke signal are as follows: firstly, the coil current signal is amplified and filtered by the amplification and filtering circuit, and then the analog-to-digital conversion is performed by the analog-to-digital conversion unit (built-in ADC) and transmitted to the central processing unit, and then the central processing unit transmits it to the main control unit through the TTL to differential 485 converter and the EMC filter (electromagnetic compatibility filter); firstly, the auxiliary switch signal is transmitted to the central processing unit through the optical coupling isolation, and then the central processing unit transmits it to the main control unit through the TTL to differential 485 converter and the EMC filter; firstly, the contact stroke signal is filtered by the EMC filter, and then transmitted to the central processing unit after filtering, and the central processing unit transmits it to the main control unit through the TTL to differential 485 converter and the EMC filter. Finally, the main control unit transmits the received three-way signals to the data processing device.

[0024] The data processing device is a host computer, which includes a memory, a processor and a computer program stored in the memory. The processor executes the computer program to implement the steps of a circuit breaker fault diagnosis method of the present invention. In addition, the data processing device introduces Docker and Docker Compose technology to encapsulate the software function modules corresponding to the computer program, realize the unified management, deployment, maintenance, uninstallation and expansion of each program, and completes the unified configuration of each image and the restriction of resources such as CPU and memory based on docker-compose.yml. The host computer uses persistence technology to store the received data in the memory to prepare for subsequent fault diagnosis. In addition, the data processing device also includes a display interface to meet the visual interface requirements of online monitoring, which can display various aspects such as fault diagnosis results and real-time monitoring data. Specifically, the application framework Electron is combined with the front-end framework Vue and the drawing tool Echarts as a visualization solution to meet the cross-platform and multi-type data requirements of the circuit breaker online monitoring system. In addition, the host computer introduced a design scheme of multi-level routing combined with sandwich layout and Flex layout; adopted a data access module based on two-layer encapsulation to improve the reuse rate of query code; adopted IPC communication to enable the rendering process to directly access the database; and introduced YAML files to meet the requirements of configuration files, and realized the memory function of custom settings through localStorage.

[0025] The specific working mode of the system is as follows: after the data acquisition device is powered on, a series of initialization settings are carried out, including device address, signal trigger threshold, and RS485 serial port information. After the initialization settings are completed, data is continuously queried based on the timer interrupt; for signal acquisition, a multiple threshold trigger mechanism is adopted to avoid false triggering caused by signal interference, that is, the sampling program only works when the same signal meets the threshold for 5 consecutive times, and when it is not triggered, the data acquisition program still stores data to record the complete data, and the queried data is stored in the buffer area; when the circuit breaker signal is triggered multiple times, the time is recorded and 300 data are taken from the processor to ensure that the data before the action threshold is also recorded; after acquisition, wait for the host computer to poll to realize real-time collection and processing.

[0026] Combine the following Figure 2 A circuit breaker fault diagnosis method of the present invention is described in detail with reference to a flowchart.

[0027] Step 1: Obtain multi-source characteristic signals reflecting the operation status of the circuit breaker, and use LDA (Linear Discriminant Analysis) to reduce their dimensionality to obtain the characteristic quantities required for circuit breaker fault diagnosis. Specifically: The multi-source feature signals specifically include the pre-processed coil current signal, contact travel signal and auxiliary switch signal; these signals are first normalized and standardized, and then all the processed signals are spliced, and the spliced ​​feature quantities and their labels are subjected to LDA analysis to obtain the feature quantities after dimensionality reduction. In the LDA dimensionality reduction process, as the dimension decreases, the accuracy is first improved to a certain extent. At this time, some redundant and interfering diagnosis information is removed, and the retained features are still important enough to accurately represent the category information of the original data, thereby making the accuracy rate higher.

[0028] Step 2: Obtain historical circuit breaker fault data, including fault types (as labels) and corresponding feature data, and divide these data into training sets and test sets.

[0029] In this process, the fault type string needs to be converted into a numerical label; and to prevent overfitting, 30% of the training set needs to be extracted as a validation set.

[0030] Step three: Build a machine learning model, and use the training set and test set obtained in step two to train and test the built machine learning model to obtain a fault diagnosis model.

[0031] In this embodiment, a CatBoost model is specifically selected from the machine learning model. CatBoost is a decision tree algorithm based on gradient boosting, which can effectively process category features and has strong generalization ability.

[0032] The CatBoost model has hyperparameters bootstrap_type and grow_policy. Bootstrap_type is a sampling strategy that controls the way samples are sampled and affects the bias and variance of the model; grow_policy is a growth strategy that controls the way decision trees are generated and affects the complexity and efficiency of the model. Figure 3a and Figure 3b As shown in the figure, there are respectively the sampling strategy, growth strategy and diagnostic error rate relationship diagrams of the CatBoost model for coil current and contact stroke (that is, the inputs of the CatBoost model are coil current and contact stroke, respectively). The sampling strategies include Bayesian sampling, Bernoulli sampling and Poisson sampling, and the growth strategies include symmetric tree-based, depth-first-based and loss-guided growth strategies. It can be seen from these two figures that the overall distribution of coil current and contact stroke are close. In the coil current model, the model error rate based on Bernoulli sampling and loss-guided growth is the lowest, and in the contact stroke model, the model error rate based on Bayesian sampling and loss-guided growth and depth-first generation is also the lowest. As shown in the figure Figure 4a and Figure 4b As shown in the figure, there are respectively the sampling strategy, growth strategy and training time relationship diagrams of the CatBoost model for coil current and contact stroke (that is, the inputs of the CatBoost model are coil current and contact stroke, respectively). From these two figures, it can be seen that the training time of the models based on Bayesian sampling and depth-first, as well as Bernoulli sampling and depth-first in the coil current model are short, and the training time of the model based on Bayesian sampling and depth-first in the contact stroke model is the shortest. Therefore, considering the training time and fault diagnosis error rate, the Bayesian sampling strategy and the depth-first growth strategy are adopted. Based on the Bayesian sampling strategy and depth-first growth strategy adopted by the CatBoost model, its fault training process is as follows Figure 5a and Figure 5b As shown, by observing the training curve, it can be found that the diagnosis rates of coil current and contact stroke are better.

[0033] Moreover, in order to avoid falling into the local optimum, this embodiment also uses a genetic algorithm to further tune the parameters of the CatBoost model. The parameters here are specifically the Bayesian bagging control strength. Of course, in order to further improve the effect, a genetic algorithm can also be used to optimize at least one of the learning rate, decision tree depth and regularization parameter. The following takes the optimization of these four parameters as an example to illustrate the entire optimization process. The first step is to initialize the population. Set the learning rate, decision tree depth, regularization parameter and Bayesian bagging control strength value ranges to [0,1], [1,10], [0,10] and [0,10] respectively, and randomly generate an initial parameter combination as the individual of the population. The second step is to evaluate the fitness. Use the CatBoost model to train each individual, calculate its multi-classification performance index, as the individual's fitness; the fitness function converts the loss function value into a value between 0 and 1, so that the higher the fitness of the individual, the smaller the corresponding loss function and the better the performance. The third step is to perform selection and crossover operations. According to the value of the fitness function, select individuals with high fitness for further reproduction; randomly select two individuals, perform crossover operation, and generate two new individuals. The fourth step is to perform mutation operation. Perform random mutation operation on the new individuals to generate new individuals. The fifth step is to add the newly generated individuals to the population and update the population. Finally, determine whether the termination condition is reached: when a certain number of iterations is reached or the fitness reaches a certain threshold, stop the iteration and output the optimal parameter combination. Based on the above process, the learning rate, decision tree depth, regularization parameter and Bayesian bagging control strength are tuned to obtain the tuning curve as shown below Figure 6a and Figure 6b Under the optimal parameter model, the model based on coil current achieved an accuracy of 81.47%, an increase of 5% compared to the non-optimized model, while the model based on contact stroke achieved an accuracy of 90.35%, an increase of 2.7% compared to the non-optimized model.

[0034] In the process of training the CatBoost model, the model parameters can be optimized by cross-validation. The specific training process is: first set the weight of each sample to an equal value, and randomly initialize the first weak classifier; calculate the value of the loss function and the residual of each sample, and update the weight of each sample according to the gradient of the loss function; CatBoost will give higher weights to samples with larger residuals so that they can be corrected in further iterations; train a new weak classifier based on the newly calculated sample weights. Then, add the new weak classifier to the model and update the model's predicted value for each sample. According to the preset stop condition, determine whether it is necessary to continue iterating until the preset stop condition is reached.

[0035] Step 4: Acquire characteristic quantity data of the circuit breaker, perform preprocessing, and input the preprocessed data into the fault diagnosis model obtained in step 3 to obtain the circuit breaker fault classification result.

[0036] The preprocessing here can be denoising and normalization preprocessing to ensure the quality and consistency of the data. In addition, an alarm signal is sent when a fault is determined to occur, so that maintenance personnel can respond and handle the fault in a timely manner.

[0037] At this point, the entire method process is completed.

[0038] in addition, Figure 7 A comparison chart of the fault diagnosis error rates of the method of the present invention and several other machine learning algorithms is shown. It can be seen from the figure that the diagnosis rate of SVC based on the optimal kernel function is very low, and KNN has a good diagnosis effect only for contact stroke. Among them, the random forest based on the optimal number of trees and the fault diagnosis model based on GA-CatBoost of the present invention have the lowest error rates, and GA-CatBoost is even better. The model based on coil current has an accuracy rate of 81.47%, and the model based on contact stroke has an accuracy rate of 90.35%.

[0039] Figure 8a and Figure 8b is the GA-CatBoost confusion matrix diagram of the present invention, Fig. 9 is a schematic diagram of the fault accuracy rate after LDA dimensionality reduction processing of the present invention, Fig.10 This is a T-SNE cluster diagram of the feature quantities after LDA dimensionality reduction processing of the present invention. It can be seen that the feature quantities after dimensionality reduction are clustered relatively well.

[0040] Computer device embodiment: A computer device embodiment of the present invention includes a memory, a processor, an internal bus and a computer program stored in the memory, and the processor and the memory communicate and exchange data with each other through the internal bus. The processor executes the computer program to implement the steps of a circuit breaker fault diagnosis method introduced in the circuit breaker fault diagnosis system embodiment of the present invention. Among them, the processor can be a processing device such as a microprocessor MCU, a programmable logic device FPGA, etc.; the memory can be various memories that use electrical energy to store information, such as RAM, ROM, etc., or it can be a memory that uses other methods.

[0041] Circuit breaker fault diagnosis method embodiment: The circuit breaker fault diagnosis method of the present invention, the principle, specific implementation process and other contents of the method have been introduced in detail in the circuit breaker fault diagnosis system embodiment, and will not be repeated in this embodiment.

[0042] In summary, the present invention has the following characteristics: the present invention performs feature dimensionality reduction processing through the LDA algorithm, which can effectively screen out features that have an important impact on fault diagnosis, thereby improving the diagnostic accuracy and efficiency; the present invention performs parameter optimization through the GA algorithm, which can improve the accuracy of the fault diagnosis model, has strong generalization ability, can provide reliable fault diagnosis results in a complex operating environment, and has high application value in circuit breaker fault diagnosis.

[0043] Specific implementation methods are given above, but the present invention is not limited to the described implementation methods. The basic idea of ​​the present invention lies in the above basic scheme. For ordinary technicians in this field, it does not take creative work to design various deformed models, formulas, and parameters according to the teachings of the present invention. Changes, modifications, substitutions, and variations of the implementation methods without departing from the principles and spirit of the present invention still fall within the scope of protection of the present invention.

Claims

1. A circuit breaker fault diagnosis method, characterized in that: The steps include: Acquire characteristic quantity data of the circuit breaker, wherein the characteristic quantity is obtained by using LDA dimension reduction processing on the multi-source characteristic signals reflecting the operation status of the circuit breaker, and is the characteristic quantity required for circuit breaker fault diagnosis; The acquired characteristic quantity data is input into the fault diagnosis model to obtain the circuit breaker fault classification result; The fault diagnosis model is obtained by training a machine learning model using a training set, and the training set is a data set containing multiple groups of historical feature quantity data with fault classification labels.

2. The circuit breaker fault diagnosis method according to claim 1, characterized in that: The machine learning model is an improved CatBoost model obtained by improving the traditional CatBoost model, and the improvement includes adopting a Bayesian sampling strategy in the sampling strategy of the model.

3. The circuit breaker fault diagnosis method according to claim 2, characterized in that: The improvement also includes that the Bayesian bagging control strength in the model is obtained by optimizing through a genetic algorithm.

4. The circuit breaker fault diagnosis method according to claim 2, characterized in that: The improvement also includes that the growth strategy in the model adopts a depth-first growth strategy.

5. The circuit breaker fault diagnosis method according to claim 2, characterized in that: The improvement also includes that at least one parameter among the learning rate, decision tree depth and regularization parameter in the model is obtained by optimizing through a genetic algorithm.

6. The circuit breaker fault diagnosis method according to any one of claims 1 to 5, characterized in that: The multi-source characteristic signals include coil current signal, contact travel signal and auxiliary switch signal.

7. The circuit breaker fault diagnosis method according to any one of claims 1 to 5, characterized in that: The acquired feature data is the data that has been denoised and normalized.

8. A computer device comprising a processor, characterized in that: The processor is used to execute a computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A circuit breaker fault diagnosis system, comprising a data acquisition device, a data transmission device and a data processing device, wherein the data acquisition device is used to collect characteristic quantity data of the circuit breaker and send it to the data processing device through the data transmission device, and the data processing device comprises a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. The circuit breaker fault diagnosis system according to claim 9, characterized in that: Each software functional module corresponding to the computer program is encapsulated using container technology.

11. The circuit breaker fault diagnosis system according to claim 9 or 10, characterized in that: The data received by the data processing device is stored in the memory using a persistence technology.

Images