CN-121997064-A - Breaker fault monitoring system and method based on artificial intelligence

CN121997064ACN 121997064 ACN121997064 ACN 121997064ACN-121997064-A

Abstract

The invention discloses a breaker fault monitoring system and method based on artificial intelligence, and relates to the technical field of fault monitoring.A physical data of a breaker is utilized to calculate and generate different key physical models, waveform data of the different key physical models are output, and a simulation data set is formed; extracting abstract features from waveform data by using a signal processing algorithm, constructing a mixed feature space by using physical features and abstract features, training a breaker fault prediction model by using a neural network algorithm, forming a total loss function by combining prediction error loss with physical residual error loss, extracting training samples from a simulation data set, performing meta-learning on the fault prediction model, extracting real-time physical data of the breaker, mapping the real-time physical data of the breaker into the mixed feature space, calculating a similar distance between the real-time physical data and each prototype, judging real-time fault types, constructing a knowledge graph according to expert experience, and searching dimension actions and fault reasons of the real-time fault types in the knowledge graph.

Inventors

LV JIAWEI
YI GUANWEN
SHEN XIAOSONG
Lai Yingdong
LIN SEN
CHENG JUNQI
WANG LIUHUO
LI HAIXIAN
LI YE
He Bingze

Assignees

广东电网有限责任公司江门供电局

Dates

Publication Date: 20260508
Application Date: 20251231

Claims (10)

1. The circuit breaker fault monitoring method based on artificial intelligence is characterized by comprising the following steps of: S100, calculating and generating different key physical models by utilizing physical data of the circuit breaker, collecting physical data of the health state and fault state of the circuit breaker in history, inputting the physical data into the key physical models for simulation, and outputting waveform data of the different key physical models to form a simulation data set; S200, extracting physical characteristics related to a key physical model in the working process of the circuit breaker, extracting abstract characteristics from waveform data by using a signal processing algorithm, and constructing a mixed characteristic space by using the physical characteristics and the abstract characteristics; S300, training a breaker fault prediction model by using a neural network algorithm, additionally designing physical residual error loss in prediction error loss constructed during training of the neural network algorithm, and forming a total loss function by combining the prediction error loss with the physical residual error loss; S400, extracting feature vectors in a mixed feature space and calculating feature vector mean values to form a prototype of the fault type aiming at each fault type, extracting real-time physical data of the circuit breaker and mapping the real-time physical data to the mixed feature space to calculate a similar distance with each prototype when the fault prediction model monitors the fault of the circuit breaker, and judging the real-time fault type; S500, constructing a knowledge graph according to expert experience, and searching dimension actions and fault reasons of the real-time fault type in the knowledge graph.
2. The method for monitoring circuit breaker faults based on artificial intelligence according to claim 1, wherein the specific steps of forming the simulation data set in S100 are as follows: S101, the key physical model comprises an electromagnetic operating mechanism model, a mechanical motion model and a contact electric contact model, wherein the electromagnetic operating mechanism model is specifically constructed by collecting coil current, coil voltage, electromagnetic force and movable iron core displacement in the operation of a circuit breaker, and constructing the influence relationship of the coil current, the electromagnetic force and the movable iron core displacement on the coil voltage according to the physical principle in the operation of the circuit breaker, wherein the formula is as follows: ; In the formula, U represents coil voltage, L represents coil inductance, i represents coil current, R represents coil direct current resistance, K e represents back electromotive force constant, x represents movable iron core displacement, and t represents time; S102, analyzing the force born by the breaker during operation, collecting the spring counter force, friction force, damping force and total mass during driving the iron core to move, and calculating the resultant force of acceleration generated by the total mass of the breaker, wherein the formula is as follows: ; In the formula, m represents the total mass, F spring represents the spring reaction force, F friction represents the friction force, F damping represents the damping force, The second derivative of the displacement of the movable iron core with respect to time is represented, and a resultant force calculation formula is used as a mechanical motion model; s103, a professional actively builds a contact electric contact model according to the relation between contact resistance and contact pressure, the relation between materials and the relation between cauterization degree; Physical data of the circuit breaker in the history and different fault states are collected and input into three models to obtain waveform data to form a simulation data set.
3. The method for monitoring faults of circuit breakers based on artificial intelligence according to claim 2, wherein the specific steps of constructing a hybrid feature space by utilizing physical features and abstract features in the step S200 are as follows: S201, collecting all working data related to the work of the circuit breaker, calculating correlation coefficients of each type of working data and each type of physical data in a key physical model when the circuit breaker breaks down, and selecting the working data with the correlation coefficient not being 0 as physical characteristics related to the key physical model of the circuit breaker; s202, extracting frequency domain features from waveform data in a simulation data set by using a wavelet transformation algorithm, taking the frequency domain features as abstract features of the circuit breaker, and forming a mixed feature space by using the abstract features and physical features.
4. The method for monitoring circuit breaker failure based on artificial intelligence according to claim 3, wherein the specific steps of constructing the total loss function by combining the prediction error loss and the physical residual loss in S300 are as follows: S301, training a breaker fault prediction model by using a neural network algorithm, wherein a prediction error loss constructed during the training of the neural network algorithm is L date , the prediction error loss is used for measuring an error between a prediction result output by the fault prediction model and a true value, and a physical residual error loss is additionally designed in the prediction error loss constructed during the training of the neural network algorithm, wherein the formula is as follows: ; In the formula, L physics represents physical residual error loss, physics law represents different key physical models, f θ (s) represents different physical predicted values of a fault prediction model when circuit breaker fault prediction is carried out, and y physics represents a physical reality value of the physical predicted value corresponding to the observed physical reality value; S302, combining the prediction error loss and the physical residual error loss to form a total loss function, wherein the formula is as follows: ; in the formula, L total represents total loss, and lambda represents super-parameters; S303, extracting different states of the circuit breaker in the simulation data set, wherein each state provides k samples to form a support set, extracting n samples as a query set, and forming a task by the support set and the query set; learning a support set and a query set by adopting model independent element learning, setting an inner loop and an outer loop, extracting a task in the inner loop, and using initial parameters of a fault prediction model, wherein the initial parameters represent initial values of different model parameters in the fault prediction model; The method comprises the steps of using an adapted fault prediction model for tasks to forward spread in a query set in an outer loop, calculating total loss, summarizing the query total loss of all tasks to form element loss, calculating the gradient of the element loss relative to initial parameters, subtracting the gradient from the initial parameters to obtain optimized parameters, optimizing the initial parameters of the fault prediction model in the outer loop in element learning, and executing the adaptation task by taking the optimized parameters as the initial parameters by the inner loop to obtain the adaptation parameters.
5. The method for monitoring circuit breaker faults based on artificial intelligence as claimed in claim 4, wherein the specific steps of judging the type of the real-time faults in the step S400 are as follows: s401, converting the characteristics of each breaker fault type into characteristic vectors in a mixed characteristic space, calculating the average value of each characteristic vector for each fault type, and using the average value of each characteristic vector as a prototype p c of the corresponding fault type; S402, when a breaker fault is detected, mapping breaker data output in a fault prediction model during the fault into a mixed feature space, extracting feature vectors of real-time faults, and calculating the similar distance between the real-time faults and each prototype of the breaker, wherein the formula is as follows: ; In the formula, sim represents the similarity distance between the real-time fault and each prototype of the circuit breaker, p (s test ) represents the characteristic vector of the real-time fault, and the fault type corresponding to the prototype with the minimum similarity distance is selected as the real-time fault type.
6. The method for monitoring faults of circuit breakers based on artificial intelligence according to claim 5, wherein the specific steps of searching the dimension action and the fault cause of the real-time fault type in the knowledge graph in S500 are as follows: S501, an expert actively inputs maintenance actions, fault reasons and association relations when different fault types exist, the fault types, the fault reasons, the fault characteristics and the maintenance actions are taken as nodes, the association relations are taken as edges, and a knowledge graph is constructed according to expert experience; s502, after judging the fault type of the real-time fault, positioning the corresponding fault type in the knowledge graph, extracting the nodes related to the real-time fault type according to the knowledge graph, extracting the maintenance action and the fault reason of the real-time fault pair, and pushing the maintenance action and the fault reason to maintenance personnel.
7. The breaker fault monitoring system based on artificial intelligence is characterized by comprising a data acquisition module, a feature extraction module, a model training module, a fault type searching module and a fault reasoning module; The data acquisition module is used for calculating and generating different key physical models by utilizing the physical data of the circuit breaker and acquiring the physical data of the health state and the fault state of the circuit breaker in the history; the characteristic extraction module is used for extracting physical characteristics related to a key physical model in the working process of the circuit breaker, extracting abstract characteristics from waveform data by using a signal processing algorithm, and constructing a mixed characteristic space by using the physical characteristics and the abstract characteristics; the model training module is used for training a breaker fault prediction model by utilizing a neural network algorithm, extracting training samples from a simulation data set and performing meta-learning on the fault prediction model; The fault type searching module is used for extracting real-time physical data of the circuit breaker and mapping the real-time physical data to a mixed feature space to calculate a similar distance with each prototype when the fault prediction model monitors the fault of the circuit breaker, and judging the real-time fault type; The fault reasoning module is used for constructing a knowledge graph according to expert experience, and searching dimension actions and fault reasons of the real-time fault type in the knowledge graph.
8. The breaker failure monitoring system of claim 7, wherein the data acquisition module comprises a key physical model unit and a simulation unit; The key physical model unit is used for calculating and generating different key physical models by utilizing physical data of the circuit breaker, wherein the key physical models comprise an electromagnetic operating mechanism model, a mechanical motion model and a contact electric contact model; The simulation unit is used for acquiring physical data of the circuit breaker in the history and in different fault states, and inputting the physical data into the three models to obtain waveform data to form a simulation data set.
9. The circuit breaker fault monitoring system based on artificial intelligence of claim 7 wherein the feature extraction module comprises a physical feature unit and an abstract feature unit; The physical characteristic unit is used for calculating the correlation coefficient of each type of working data and each type of physical data in the key physical model when the breaker breaks down, and selecting the working data with the correlation coefficient not being 0 as the physical characteristic related to the key physical model of the breaker; The abstract feature unit is used for extracting frequency domain features from waveform data in the simulation data set by utilizing a wavelet transformation algorithm, and taking the frequency domain features as abstract features of the circuit breaker.
10. The breaker failure monitoring system based on artificial intelligence of claim 7, wherein the model training module comprises a total loss function construction unit and a meta learning unit; The total loss function construction unit is used for combining the prediction error loss and the physical residual error loss to form a total loss function; the meta learning unit is used for extracting training samples from the simulation data set and performing meta learning on the fault prediction model.

Description

Breaker fault monitoring system and method based on artificial intelligence Technical Field The invention relates to the technical field of fault monitoring, in particular to a circuit breaker fault monitoring system and method based on artificial intelligence. Background Electric power systems are vital in modern society, supporting various aspects of industrial production, commercial operations, and residential life. The high-voltage circuit breaker is used as a key device in the power system, plays an important role in controlling and protecting a circuit, can switch on and off load current under normal conditions, and can rapidly switch off short-circuit current under fault conditions, so that other devices in the power system are protected from being damaged, and the fault range is prevented from being expanded. With the continuous expansion of the power system, the voltage class is continuously improved, and the number of high-voltage circuit breakers is also increased. The traditional fault diagnosis method mainly relies on manual inspection and experience judgment, has the problems of low efficiency, poor accuracy, insufficient real-time performance and the like, and is difficult to meet the requirements of a large-scale high-voltage-class power system on fault diagnosis of a high-voltage circuit breaker. The intelligent fault diagnosis technology integrates the knowledge and the method of multiple disciplines such as a sensor technology, a signal processing technology, a data analysis technology, an artificial intelligence technology and the like, can monitor and analyze the running state of the high-voltage circuit breaker in real time, timely and accurately diagnose the fault type and the fault position, give out corresponding fault treatment suggestions, realize the state maintenance of the high-voltage circuit breaker, avoid blindness and excessive maintenance of the traditional regular maintenance, effectively improve the running reliability of equipment, reduce the running maintenance cost and improve the overall running efficiency of a power system. However, in the conventional breaker fault monitoring, when a brand new breaker or a new fault mode does not have historical data accumulation, the conventional data driving model cannot be trained or has extremely poor diagnosis effect, and a model trained on one working condition or one type of breaker is rapidly reduced in another slightly different working condition or model, and has no universality. And when the pure data driven deep learning model gives a diagnosis result, the pure data driven deep learning model lacks of physical mechanism support and logic interpretation and is difficult to accept and trust by field engineers and field experts. For special faults with very low occurrence probability, the traditional model often ignores the special faults as noise due to the rare samples, so that the special faults are not reported. Disclosure of Invention The invention aims to provide a breaker fault monitoring system and method based on artificial intelligence, which are used for solving the problems in the prior art. In order to achieve the above purpose, the present invention provides the following technical solutions: A circuit breaker fault monitoring method based on artificial intelligence, the method comprising the steps of: S100, calculating and generating different key physical models by utilizing physical data of the circuit breaker, collecting physical data of the health state and fault state of the circuit breaker in history, inputting the physical data into the key physical models for simulation, and outputting waveform data of the different key physical models to form a simulation data set; further, the specific steps of forming the simulation data set are as follows: S101, the key physical model comprises an electromagnetic operating mechanism model, a mechanical motion model and a contact electric contact model, wherein the electromagnetic operating mechanism model is specifically constructed by collecting coil current, coil voltage, electromagnetic force and movable iron core displacement in the operation of a circuit breaker, and constructing the influence relationship of the coil current, the electromagnetic force and the movable iron core displacement on the coil voltage according to the physical principle in the operation of the circuit breaker, wherein the formula is as follows: ; In the formula, U represents coil voltage, L represents coil inductance, i represents coil current, R represents coil direct current resistance, K e represents back electromotive force constant, x represents movable iron core displacement, and t represents time; S102, analyzing the force born by the breaker during operation, collecting the spring counter force, friction force, damping force and total mass during driving the iron core to move, and calculating the resultant force of acceleration generated by the total mass of the breaker, whe