Circuit breaker health state assessment method and system
By constructing a classification model based on the PSO-AdaBoost-RF algorithm and combining wavelet filtering technology with state variable extraction, efficient assessment of circuit breaker health status is achieved, improving the accuracy of fault identification and solving the problem of inaccurate circuit breaker health status judgment in existing technologies.
Patent Information
- Application Number
- CN202511043488.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-28
- Publication Date
- 2025-12-19
AI Technical Summary
The accuracy of existing technology in analyzing mechanical fault types and assessing the health status of circuit breakers is not high.
A classification model is constructed using the PSO-AdaBoost-RF algorithm. By collecting the mechanical characteristic curves of the circuit breaker in real time, the state variables are extracted, and wavelet filtering technology is used to process the data. The hyperparameters of the RF algorithm are optimized by combining the PSO algorithm and the weights of the weak classifiers are adjusted by the AdaBoost algorithm to form a strong classifier, thereby realizing the assessment of the health status of the circuit breaker.
It improved the accuracy of circuit breaker health status assessment, with a fault identification accuracy rate of 98.3%, providing an effective solution for intelligent fault diagnosis of high-voltage circuit breakers.
Smart Images
Figure CN121167501A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application relates to a circuit breaker health state evaluation method and system, and belongs to the field of circuit breaker state evaluation. BACKGROUND
[0002] A circuit breaker is a crucial device in a power system. With the use of the circuit breaker, different changes occur in the mechanical characteristics of the circuit breaker. Once the changes exceed the limit, the actual use of the circuit breaker is affected. Therefore, the mechanical characteristics of the used circuit breaker need to be tested to evaluate the health state of the circuit breaker.
[0003] However, when the mechanical characteristics of the circuit breaker are evaluated in the prior art, the monitored parameters are single, so that the accuracy of the mechanical fault type analysis of the circuit breaker and the health state judgment of the circuit breaker is not high. SUMMARY
[0004] The application aims to provide a circuit breaker health state evaluation method and system to solve the problem of low accuracy of circuit breaker health state evaluation.
[0005] To achieve the above-mentioned purpose, the application provides the following scheme.
[0006] The circuit breaker health state evaluation method comprises the following steps: 1) collecting the mechanical characteristic curve of the circuit breaker in action in real time; 2) obtaining the state quantity of the circuit breaker in action according to the mechanical characteristic curve; 3) inputting the state quantity into the trained classification model to obtain the corresponding circuit breaker state including the fault state and the normal state.
[0007] The construction process of the classification model comprises the following steps: using the PSO algorithm to determine the hyperparameters in the RF algorithm, the hyperparameters including the number of decision trees; adjusting the parameters of each decision tree through the training set, and taking the decision tree as a weak classifier of the AdaBoost algorithm and adjusting the weight of the weak classifier based on the AdaBoost algorithm during the training process; after the training is completed, the weak classifiers with the adjusted weights are summed to obtain a strong classifier of the AdaBoost algorithm, and the strong classifier is taken as the classification model.
[0008] Further, the PSO algorithm is also used to preliminarily determine the parameters of each decision tree in the RF algorithm, and then the parameters of each preliminarily determined decision tree are adjusted through the training set.
[0009] Further, the mechanical characteristic curve comprises a stroke characteristic curve and a split and close current characteristic curve.
[0010] Further, the state quantity comprises the opening distance, the overstroke, the just split point, the just close point, the split and close current, the split and close time, and the split and close speed of the circuit breaker.
[0011] Further, in step 2), the mechanical characteristic curve is wavelet filtered, and then the state quantity is obtained.
[0012] A circuit breaker health state evaluation system comprises a processor, and the processor executes a computer program to realize the steps of the following method: 1) collecting a mechanical characteristic curve in real time when the circuit breaker is operated; 2) obtaining a state quantity when the circuit breaker is operated according to the mechanical characteristic curve; 3) inputting the state quantity into a trained classification model to obtain a corresponding circuit breaker state including a fault state and a normal state;
[0013] The construction process of the classification model comprises the following steps: determining hyperparameters in the RF algorithm by using the PSO algorithm, the hyperparameters comprising the number of decision trees; adjusting the parameters of each decision tree through a training set, and taking the decision tree as a weak classifier of the AdaBoost algorithm during the training process and adjusting the weight of the weak classifier based on the AdaBoost algorithm; after the training is completed, the weak classifiers with the adjusted weights are summed to obtain a strong classifier of the AdaBoost algorithm, and the strong classifier is taken as the classification model.
[0014] Further, the PSO algorithm is also used to preliminarily determine the parameters of each decision tree in the RF algorithm, and then the parameters of each preliminarily determined decision tree are adjusted through the training set.
[0015] Further, the mechanical characteristic curve comprises a stroke characteristic curve and a closing and opening current characteristic curve.
[0016] Further, the state quantity comprises an opening distance, an overstroke, a just opening point, a just closing point, a closing and opening current, a closing and opening time and a closing and opening speed of the circuit breaker.
[0017] Further, in step 2), the mechanical characteristic curve is wavelet filtered, and then the state quantity is obtained.
[0018] The circuit breaker health state evaluation system has the following beneficial effects: the model used in the evaluation of the health state of the circuit breaker is jointly constructed based on the PSO-AdaBoost-RF algorithm, that is, the hyperparameters in the RF algorithm are determined by using the PSO algorithm, the hyperparameters comprising the number of decision trees; the parameters of each decision tree are adjusted through a training set, and the decision tree is taken as a weak classifier of the AdaBoost algorithm during the training process and the weight of the weak classifier is adjusted based on the AdaBoost algorithm; after the training is completed, the weak classifiers with the adjusted weights are summed to obtain a strong classifier of the AdaBoost algorithm, and the strong classifier is taken as the classification model. After the corresponding classification model is obtained, it is deployed online, then the mechanical characteristic curve in real time when the circuit breaker is operated is collected, and then the state quantity obtained according to the mechanical characteristic curve is input into the classification model, so that the evaluation of the health state of the circuit breaker can be realized.
[0019] The PSO-AdaBoost-RF algorithm adopts a random forest classifier as a basic model for fault diagnosis of the circuit breaker, and introduces a particle swarm optimization algorithm and an AdaBoost algorithm to optimize the random forest algorithm. The particle swarm optimization can reduce the dimension of data by selecting the most important features for fault diagnosis, the AdaBoost can enhance the sensitivity of the classifier to difficult samples and reduce redundant features, and the combination of the two can reduce the training time and improve the generalization ability of the algorithm model. The PSO-AdaBoost-RF algorithm can identify faults in a short time, and the identification accuracy of the fault is as high as 98.3%, which provides an effective solution for intelligent fault diagnosis of high-voltage circuit breakers. BRIEF DESCRIPTION OF DRAWINGS
[0020] Figure 1 is a circuit breaker state evaluation process of the present application;
[0021] Figure 2 is a classification model determination process of the present application. DETAILED DESCRIPTION
[0022] In order to make the purpose, technical scheme and advantages of the present application more clear and obvious, the present application will be further described in detail below with reference to the drawings and embodiments.
[0023] The idea of the present application is to collect the mechanical characteristic curve of the circuit breaker in action in real time, obtain the corresponding state quantity according to the collected mechanical characteristic curve, input the state quantity into the trained classifier, and output the corresponding circuit breaker state corresponding to the state quantity.
[0024] Method embodiment:
[0025] The present embodiment provides a circuit breaker health state evaluation method, as shown in Figure 1 , comprising the following steps: 1) collecting the mechanical characteristic curve of the circuit breaker in action in real time; 2) obtaining the state quantity of the circuit breaker in action according to the obtained mechanical characteristic curve; 3) inputting the state quantity into the trained classification model to obtain the corresponding circuit breaker state.
[0026] Specifically, the mechanical characteristic curve includes a stroke characteristic curve and a split and close current characteristic curve. The present embodiment takes the split case as an example to explain step 2):
[0027] From the stroke characteristic curve, the total stroke, rebound value and other state quantities of the circuit breaker in split can be directly observed.
[0028] In the split and close current characteristic curve, the starting point of the current is the time (t1) when the circuit breaker receives the split instruction.
[0029] The opening point is calculated by combining the travel-time curve and the opening and closing current curve. At the opening point, the two contacts are affected by the force of the arc, causing the opening to shake slightly. After the contacts are separated, the current drops suddenly. The shaking time before the sudden drop is the opening point of the circuit breaker, i.e., the time (t2) at which the arc contacts are separated.
[0030] The arc extinguishing time can be determined from the acceleration-time curve. During the opening process of the circuit breaker, the strong attraction of the arc to the contacts makes it difficult for the contacts to separate, and the acceleration is not very large. When the arc is extinguished, the attraction to the contacts disappears, and the acceleration increases instantaneously. According to the acceleration curve, the maximum acceleration point is determined by combining the first derivative and the second derivative, and the maximum point is the arc extinguishing time (t3).
[0031] High-order derivative formula:
[0032]
[0033] The opening time is the time at which the arc contacts in all phases are separated after the circuit breaker receives the opening instruction, i.e., the opening time (short opening time) = arc time + opening time.
[0034] Arc time = circuit breaker breaking time (t3-t1) - opening time (t2-t1).
[0035] The circuit breaker breaking time is the time at which the arc contacts in all phases are separated after the circuit breaker receives the opening instruction. Before the circuit breaker receives the opening instruction, the motor needs to store enough energy to separate the moving and stationary contacts. The opening time is the time at which the arc contacts in all phases are separated instantaneously after the circuit breaker receives the opening instruction (t1). The opening point of the circuit breaker is the time at which the arc contacts are separated (t2).
[0036] The opening speed v, i.e., the opening speed, is the instantaneous speed of the moving contact in the pole column when it just separates from the stationary contact after the opening instruction is issued. In some cases, the opening speed can also be defined as the average speed within 6 millimeters (mm) after the opening. The time at which 6 mm has passed after the opening is t4, and v = 6 / (t2-t4), t2-t4: the time at which the opening point is located - the time at which 6 mm has passed after the opening.
[0037] The travel curve and the opening and closing current curve can be combined to observe the distance (x) of the moving contact from the opening point to the time at which the arc contacts are separated after the opening instruction is received.
[0038] Further, the collected travel characteristic curve and the opening and closing characteristic curve are wavelet filtered, and the opening distance, overtravel, just opening point, just closing point, opening and closing current, opening and closing time, opening and closing speed and other state quantities during the action of the circuit breaker are extracted. After the wavelet filtering, the curve retains the time information of the abrupt signal, and the time information reflects the repetition frequency of the abrupt signal and their change law, and contains the information of the state quantity.
[0039] By Hibbert transformation, suppose that the wavelet coefficient d(t) of a certain space, the Hibert transformation of d(t) is:
[0040]
[0041] The envelope signal of the wavelet coefficient d(t) is:
[0042]
[0043] The state quantity extraction by the wavelet correlation filtering envelope analysis method mainly has two stages: the wavelet correlation filtering denoising idea is adopted, the wavelet transformation coefficients of the signal on adjacent scales are directly multiplied, and the wavelet decomposition coefficients of the abrupt signal are multiplied to be enhanced. Hibert envelope detailed spectrum analysis is performed on the wavelet coefficients d after the filtering processing, and the time in the state quantity is obtained.
[0044] Further, in step 3), the classification model is constructed by the PSO-AdaBoost-RF algorithm, as shown in Figure 2 The optimal hyperparameter setting of the random forest is searched in the solution space in parallel by the PSO algorithm, and the hyperparameter setting includes the number of decision trees, the maximum depth and the like. The optimized PSO-RF is embedded into the AdaBoost algorithm framework as a base classifier (weak classifier).
[0045] Firstly, the training set samples are divided into a corresponding number of subsets according to the number of decision trees, and the same initial sample weight is assigned to the subsets. In the process of iterative training, the fault samples classified incorrectly by the weak classifier in each round are subjected to exponential weight increment, and the sample weight corresponding to the correctly classified samples is reduced, so that the subsequent weak classifier focuses on the difficult-to-classify samples near the decision boundary, and in each round of training, the classifier weight corresponding to the weak classifier with low error rate is larger. After the training is completed, the weighted sum of all weak classifiers and the weights corresponding to the weak classifiers is performed to form a final strong classifier, so that the strong classifier is used as a trained classification model to perform state classification.
[0046] Specifically, due to the differences in feature data of various fault types, in order to mine the feature data of different faults and make full use of all the information that can be collected, the coil characteristic current data and the contact stroke characteristic parameters collected in real time are taken as the input signals of the PSO-AdaBoost-RF algorithm (as shown in Table 1), and the fault states such as operating voltage offset, tripping coil short circuit, oil buffer fault and the normal operation state (as shown in Table 2) are taken as the final output results of the diagnostic model as an example for illustration:
[0047] Table 1 Signal input quantity
[0048]
[0049] Table 2 Circuit breaker state
[0050]
[0051] Exemplarily, when the coil characteristic current is I and the contact stroke characteristic value is I, it is a tripping normal operation state; when the coil characteristic current is III and the contact stroke characteristic value is II, it is a tripping coil turn-to-turn short circuit.
[0052] The PSO is responsible for optimizing the hyperparameters of the random forest base classifier, the population size determines the coverage ability and the calculation complexity of the search space, if the population size is too small, the algorithm is easy to fall into local optimum, and if the population size is too large, the marginal benefit decreases with the increasing cost; the maximum iteration number is the termination condition for controlling the optimization process, and the specific parameter setting is shown in Table 3:
[0053] Table 3 PSO parameter setting
[0054] Parameter Setting / initial value Population size 40 Maximum number of iterations 170 adaptive parameter (w max) ]] 0.9 adaptive parameter (w min) ]] 0.4 learning factor c1 1.0 (initial) 2.0 (final) learning factor c2 2.5 (initial) 1.5 (final) Probability of perturbation 5％
[0055] The parameter design of the random forest includes the tree structure parameter and the feature selection parameter, and the specific parameter setting is shown in Table 4:
[0056] Table 4 Parameter setting of random forest decision tree
[0057] Parameter Setting / initial value Maximum depth 6 Minimum sample size for node split 3 Minimum sample size for leaf node 2 Maximum number of features auto
[0058] In the decision tree structure parameter, the maximum depth controls the complexity of a single decision tree, if the value is too small, the model may be under-fitted and unable to capture the fault features, and if the value is too large, the risk of over-fitting is high and the calculation cost increases; the minimum sample number for node splitting is to prevent over-fitting and improve the generalization ability; the minimum sample number for leaf node can limit the minimum sample number of the leaf node to prevent small sample noise. In the feature selection parameter, there are fewer parameters to be set, only the maximum number of features selected by each tree needs to be designed. In this embodiment, the maximum number of features is set to auto in the first training, and the parameter is updated according to the training result after the first round of training.
[0059] System implementation:
[0060] The present embodiment provides a circuit breaker health state evaluation system, and a computer program executed by a processor in the system is designed based on the method design as introduced in the method embodiment. Since the introduction of the method is clear enough, it will not be described here.
[0061] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application rather than limit them, and although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and any modification or equivalent replacement without departing from the spirit and scope of the present application should be covered in the protection scope of the claims of the present application.
Claims
1. A method for assessing the health status of a circuit breaker, characterized in that, The process includes the following steps: 1) Real-time acquisition of the mechanical characteristic curve of the circuit breaker during operation; 2) Obtaining the state variables of the circuit breaker during operation based on the mechanical characteristic curve; 3) Inputting the state variables into a trained classification model to obtain the corresponding circuit breaker states, including fault states and normal states. The process of constructing the classification model includes: using the PSO algorithm to determine the hyperparameters in the RF algorithm, the hyperparameters including the number of decision trees; adjusting the parameters of each decision tree through the training set, and using the decision trees as weak classifiers of the AdaBoost algorithm during training and adjusting the weights of the weak classifiers based on the AdaBoost algorithm; After training, the weighted weak classifiers are weighted and summed to obtain the strong classifier of the AdaBoost algorithm, and the strong classifier is used as the classification model.
2. The circuit breaker health status assessment method according to claim 1, characterized in that, The PSO algorithm is also used to initially determine the parameters of each decision tree in the RF algorithm, and then the parameters of each decision tree are adjusted using the training set.
3. The circuit breaker health status assessment method according to claim 1, characterized in that, The mechanical characteristic curves include the stroke characteristic curve and the opening and closing current characteristic curve.
4. The circuit breaker health status assessment method according to claim 1, characterized in that, The state quantities include the circuit breaker's opening distance, overtravel, initial opening point, initial closing point, opening and closing current, opening and closing time, and opening and closing speed.
5. The circuit breaker health status assessment method according to claim 1, characterized in that, In step 2), the mechanical characteristic curve is subjected to wavelet filtering to obtain the state quantity.
6. A circuit breaker health status assessment system, comprising a processor, characterized in that, The processor executes a computer program to implement the following steps: 1) real-time acquisition of the mechanical characteristic curve of the circuit breaker during operation; 2) obtaining the state quantity of the circuit breaker during operation based on the mechanical characteristic curve; 3) inputting the state quantity into a trained classification model to obtain the corresponding circuit breaker state including fault state and normal state. The process of constructing the classification model includes: using the PSO algorithm to determine the hyperparameters in the RF algorithm, the hyperparameters including the number of decision trees; adjusting the parameters of each decision tree through the training set, and using the decision trees as weak classifiers of the AdaBoost algorithm during training and adjusting the weights of the weak classifiers based on the AdaBoost algorithm; After training, the weighted weak classifiers are weighted and summed to obtain the strong classifier of the AdaBoost algorithm, and the strong classifier is used as the classification model.
7. The circuit breaker health status assessment system according to claim 6, characterized in that, The PSO algorithm is also used to initially determine the parameters of each decision tree in the RF algorithm, and then the parameters of each decision tree are adjusted using the training set.
8. The circuit breaker health status assessment system according to claim 6, characterized in that, The mechanical characteristic curves include the stroke characteristic curve and the opening and closing current characteristic curve.
9. The circuit breaker health status assessment system according to claim 6, characterized in that, The state quantities include the circuit breaker's opening distance, overtravel, initial opening point, initial closing point, opening and closing current, opening and closing time, and opening and closing speed.
10. The circuit breaker health status assessment system according to claim 6, characterized in that, In step 2), the mechanical characteristic curve is subjected to wavelet filtering to obtain the state quantity.