A circuit breaker state identification method and device based on agent analysis and closed-loop feedback

Abstract

The application discloses a circuit breaker state identification method and device based on an intelligent agent analysis and closed-loop feedback, and belongs to the field of circuit breaker state monitoring and artificial intelligence technology. The method takes an intelligent agent as the core, fuses and processes multi-source heterogeneous data such as a SCADA system, a fault recorder and a protection device, and constructs a dynamic characteristic matrix. A time sequence analysis and a deep learning model are called to decouple and identify an operation intention, and normal operation and hidden faults such as switch jump are distinguished. Circuit breaker state identification is carried out based on a knowledge graph knowledge base, and professional treatment suggestions are generated. The intelligent agent continuously collects feedback information, and a closed-loop mechanism of demand sensing, strategy planning, model evolution and verification deployment is used to realize autonomous evolution of the model. The application effectively improves the accuracy, robustness and self-adaptive capacity of circuit breaker state identification, and provides reliable technical support for safe operation of a power supply system.

Description

Technical Field

[0001] This invention relates to the fields of circuit breaker condition monitoring and artificial intelligence technology, and in particular to a method, device, storage medium and electronic device for circuit breaker condition identification based on agent analysis and closed-loop feedback. Background Technology

[0002] In railway power supply systems and power systems, circuit breakers, as critical switching devices, bear the important responsibility of controlling the opening and closing of circuits and providing fault protection. Accurate identification of the circuit breaker's status is a key link in fault location, emergency response, and safe system operation, directly affecting power supply reliability and power grid security and stability.

[0003] Traditional circuit breaker status identification methods mainly rely on single "remote / local" location signals or operation command records collected by SCADA (Supervisory Control and Data Acquisition) systems, combined with the experience of dispatchers for manual judgment. However, this type of method has obvious limitations: on the one hand, a single signal source is easily affected by factors such as electromagnetic interference, contact jitter, and communication errors, leading to misjudgments of status; on the other hand, for hidden faults such as "switching without warning," the lack of obvious corresponding operation commands makes it difficult for traditional methods to accurately identify them, easily creating monitoring blind spots.

[0004] In recent years, with the development of artificial intelligence technology, some studies have attempted to use machine learning or deep learning models for circuit breaker status identification. However, most existing technologies use static rules or offline training models, which have the following problems: First, diagnostic conclusions cannot drive model self-correction, and the conversion efficiency of human experience into system knowledge is low; second, the system has poor adaptability to changes in operating conditions and new fault modes, and the model performance will gradually decline as the power grid structure and operating mode evolve; finally, existing methods mostly analyze a single data source and fail to fully utilize the complementary information from multi-source heterogeneous data such as SCADA systems, fault recorders, and protection devices, and the identification accuracy and robustness need to be improved.

[0005] An agent, as an intelligent entity possessing autonomous perception, intent reasoning, and closed-loop learning capabilities, can simulate expert decision-making processes and perform logical inferences by integrating multi-dimensional information even when signals are incomplete or contradictory. Applying agent technology to the field of circuit breaker status identification is expected to solve problems such as monitoring blind spots, high false positive rates, and lack of adaptive capabilities inherent in traditional methods. Summary of the Invention

[0006] The technical problem to be solved by this invention is: to address the shortcomings of traditional circuit breaker status identification methods, such as difficulty in accurately identifying hidden faults such as "switching without authorization", susceptibility to interference leading to misjudgment, and lack of adaptive evolution capabilities, a circuit breaker status identification method based on agent analysis and closed-loop feedback is provided to improve the accuracy, robustness and adaptability of circuit breaker status identification.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: a circuit breaker state identification method based on agent analysis and closed-loop feedback, which takes an agent as the core execution unit, completes the fusion processing of multi-source heterogeneous data, constructs a dynamic feature matrix that integrates time correlation and feature dimension, calls time series analysis and deep learning models to decouple and identify operation intentions, distinguishes between normal operation and hidden faults, establishes a knowledge base for circuit breaker state identification, and realizes the autonomous evolution of the model through a closed-loop feedback mechanism.

[0008] Specifically, the technical solution of the present invention includes the following steps: The intelligent agent performs timestamp alignment and standardization on the multi-source heterogeneous data generated during the circuit breaker operation process, and extracts time-domain and frequency-domain features that reflect the equipment status. The intelligent agent calls time series data analysis algorithms to process strong time series data, captures the changing trends and patterns before and after the circuit breaker operation, constructs a dynamic feature matrix, and calls a deep learning model to determine the nature of the operation. A knowledge base is built from power system operation procedures, maintenance guidelines and expert experience. The intelligent agent identifies the status of circuit breakers based on the knowledge base and generates diagnostic results and operation suggestions. The intelligent agent continuously collects diagnostic feedback information, autonomously perceives evolutionary needs, updates model parameters, achieves model self-adaptation, and forms a closed-loop feedback.

[0009] In a preferred embodiment of the present invention, the sources of the multi-source heterogeneous data include a SCADA system, a fault recorder, and a protection device; wherein, the data from the SCADA system includes switch quantity change signals, opening and closing status, and remote / local position signals; the data from the fault recorder includes voltage and current transient waveform data; and the data from the protection device includes action messages and operation log text.

[0010] In a preferred embodiment of the present invention, the intelligent agent invokes a deep learning model to decouple and identify the operation intent, including: for the strong temporal correlation of circuit breaker operation, invoking a Transformer-based time series model to extract nonlinear discriminant features between normal voltage drops and voltage drops caused by fault tripping; and based on the nonlinear discriminant features, determining whether the nature of the circuit breaker operation is normal operation or abnormal action.

[0011] As a preferred embodiment of the present invention, the construction of a dynamic feature matrix that integrates time correlation and feature dimension includes: encapsulating multi-source, multi-dimensional, and time-varying feature information in a structured manner to form a two-dimensional matrix structure containing time dimension and feature dimension; and calling a machine learning classification algorithm to analyze the dynamic feature matrix and distinguish the operation type as normal operation and abnormal action.

[0012] In a preferred embodiment of the present invention, the knowledge base is a knowledge graph structure, with circuit breakers, protection devices, fault phenomena, and handling measures as core entities; the intelligent agent identifies the status of the circuit breaker based on the preset knowledge base, including: after diagnosing the specific fault nature, querying the knowledge graph to generate specialized handling suggestions that comply with power regulations.

[0013] As a preferred embodiment of the present invention, the generation of diagnostic results and operational suggestions includes: structurally combining the fault time, equipment name, fault type, confidence level and recommended measures to form diagnostic results and operational suggestions output.

[0014] As a preferred embodiment of the present invention, the intelligent agent autonomously perceives evolutionary needs, including: continuously monitoring the confidence distribution of its own diagnostic results, human feedback information, and changes in the distribution of newly added data; when it detects that the model performance index continuously declines beyond a preset threshold, the human correction rate exceeds a threshold, or the data distribution drifts significantly, it autonomously judges that the current model is not adaptable enough and triggers evolutionary needs.

[0015] As a preferred embodiment of the present invention, the step of updating model parameters through a closed-loop feedback mechanism to achieve adaptive model evolution includes: Strategy planning steps: Based on the new sample size and changes in data distribution, select the corresponding evolutionary strategy, which includes incremental learning strategy, transfer learning strategy, and feature weight calibration strategy; at the same time, evaluate the system's computing resources and formulate the execution time and resource allocation plan for the evolutionary task. Model evolution steps: construct a sample set, trigger training tasks, record key information in the evolution process, and form a traceable evolution log; Verification and deployment steps: Test and evaluate the updated model. If the new model performs better than the original model, deploy it; otherwise, automatically roll back to the original model.

[0016] As a preferred embodiment of the present invention, the selection of the corresponding evolutionary strategy includes: if the new sample size is small, selecting an incremental learning strategy to fine-tune the model parameters; if the new sample size is sufficient and the data distribution changes significantly, selecting a transfer learning strategy; if the weights of individual features are unbalanced, selecting a feature weight calibration strategy to adjust the weight coefficients in the dynamic feature matrix.

[0017] The present invention also provides a circuit breaker state identification device based on agent analysis and closed-loop feedback, comprising: The data fusion processing module is used to fuse multi-source heterogeneous data generated during the operation of the circuit breaker through an intelligent agent. The fusion processing includes timestamp alignment and standardization processing, and extracts time-domain and frequency-domain features that reflect the equipment status. The operation intent identification module is used to process strong time-series data by calling the time-series analysis algorithm through the intelligent agent, capture the change trend before and after the circuit breaker operation, construct a dynamic feature matrix that integrates time correlation and feature dimension, and call a deep learning model to decouple and identify the operation intent, and distinguish between normal operation and hidden fault. The status diagnosis module is used to identify the status of the circuit breaker through the intelligent agent based on a preset knowledge base, and generate diagnostic results and operation suggestions. The knowledge base includes power system operation procedures, maintenance guidelines and expert experience. The closed-loop feedback evolution module is used to continuously collect diagnostic result feedback information through the intelligent agent, autonomously perceive evolutionary needs, update model parameters through a closed-loop feedback mechanism, and realize adaptive evolution of the model.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: First, by integrating multi-source heterogeneous data through intelligent agents, and making full use of complementary information from different data sources such as SCADA systems, fault recorders, and protection devices, the misjudgment rate caused by interference from a single signal source can be effectively reduced. Second, by combining time series analysis with deep learning, we can capture subtle differences between normal operation and hidden faults such as "switching switches unexpectedly," thereby improving the ability to identify hidden faults. Third, knowledge bases based on knowledge graphs provide professional procedural guidelines and treatment suggestions for diagnosis, improving the practicality and professionalism of diagnostic results; Fourth, the closed-loop feedback mechanism enables the model to evolve autonomously, adapt to changes in operating conditions and new fault modes, maintain identification accuracy, and provide reliable technical support for the safe operation of the power supply system. Attached Figure Description

[0019] Figure 1 A schematic diagram of the overall process of the circuit breaker state identification method based on agent analysis and closed-loop feedback provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the model evolution process in the closed-loop feedback mechanism provided in an embodiment of the present invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0021] This invention provides a circuit breaker state identification method based on agent analysis and closed-loop feedback. Its core idea is to use an agent as the core execution unit and achieve accurate, robust, and adaptive identification of the circuit breaker state through four key links: multi-source heterogeneous data fusion, intent recognition combining time series analysis and deep learning, knowledge graph-assisted diagnosis, and closed-loop feedback autonomous evolution.

[0022] Please see Figure 1 The method of this invention generally includes the following four steps: Step S101: Multi-source heterogeneous data processing. For multi-source heterogeneous data generated during circuit breaker operation, the agent performs timestamp alignment and standardization on data from different sources, of different types, and with different sampling frequencies, and extracts time-domain and frequency-domain features that can reflect the equipment status.

[0023] Specifically, the sources of the multi-source heterogeneous data include, but are not limited to, SCADA systems, fault recorders, and protection devices. The data recorded by the SCADA system includes switching quantity change signals, such as opening and closing status, and remote / local location; the data recorded by the fault recorder includes transient waveform data, such as voltage and current transient records; and the data provided by the protection device includes action messages and operation log text.

[0024] Timestamp alignment refers to unifying data from different sampling frequencies and clock sources to the same time base, ensuring data consistency across time. Standardization processes include data format unification, unit normalization, and outlier removal, providing a high-quality data foundation for subsequent feature extraction and model analysis.

[0025] In the feature extraction stage, time-domain features include, but are not limited to, amplitude, rate of change, duration, and number of zero crossings; frequency-domain features include, but are not limited to, fundamental amplitude, harmonic content, and spectral distribution. These features can reflect the operating status of the circuit breaker from different dimensions.

[0026] Step S102: Time series analysis and deep learning to determine the nature of the operation. The agent calls a time series data analysis algorithm to process the strong time series data generated by SCADA, capture the changing trends and patterns before and after the circuit breaker operation, construct a dynamic feature matrix, and the agent calls a deep learning model to determine the nature of the operation.

[0027] To address the strong temporal correlations inherent in circuit breaker operations, this embodiment employs a Transformer-based time series model. Utilizing its self-attention mechanism, the agent captures long-distance temporal dependencies, acquiring nonlinear discriminative features between normal voltage dips and voltage dips caused by fault-induced tripping. Compared to traditional threshold- or rule-based methods, the Transformer-based time series model can learn more complex nonlinear patterns, improving the ability to identify hidden faults.

[0028] The construction of the dynamic feature matrix involves structurally encapsulating multi-source, multi-dimensional, and time-varying feature information into a two-dimensional matrix structure containing both time and feature dimensions. This matrix can fully characterize the state evolution process of the circuit breaker before and after operation. The agent uses machine learning classification algorithms (such as support vector machines, random forests, or neural network classifiers) to analyze the dynamic feature matrix and distinguish between normal operations and abnormal actions.

[0029] Step S103: Construct a knowledge base and perform circuit breaker status identification. A knowledge base is constructed using power system operation procedures, maintenance guidelines, and the experience of senior experts. An intelligent agent is then used to identify the circuit breaker status, generating diagnostic results and operational suggestions.

[0030] In this embodiment, the knowledge base adopts a knowledge graph structure, with circuit breakers, protection devices, fault phenomena, and handling measures as core entities, which are connected through semantic relationships. When the intelligent agent diagnoses the specific nature of the fault, it retrieves the corresponding handling measures and maintenance suggestions by querying the knowledge graph, enabling it to generate professional handling suggestions that comply with power regulations.

[0031] The diagnostic results and operational recommendations are output in a structured format, including fields such as fault time, equipment name, fault type, confidence level, and recommended measures, which facilitates quick understanding and execution by dispatchers.

[0032] Step S104: Autonomous perception, model evolution, and closed-loop feedback. The agent continuously collects diagnostic result feedback information, autonomously perceives evolutionary needs, updates model parameters, achieves model adaptation, and forms closed-loop feedback.

[0033] Please see Figure 2 The closed-loop feedback mechanism based on the intelligent agent framework includes four key stages: demand perception, strategy planning, model evolution, and verification and deployment.

[0034] In the demand perception phase, the agent continuously accesses system operation data to monitor model performance in real time, including performance metrics such as the model's confidence distribution on recent data, the trend of confusion matrix changes, and the accuracy of identifying jump vulnerabilities. When the accuracy continuously declines beyond a preset threshold, an evolutionary warning is triggered. Simultaneously, the agent listens to feedback signals from the scheduler regarding the adoption rate of diagnostic results, correction records, and the number of rejections, performing feedback stream analysis. Furthermore, the agent uses data drift detection to compare the current data distribution with the training set distribution, autonomously identifying changes in the model's applicable boundaries.

[0035] During the strategy planning phase, the agent selects the strategy type based on the new samples. If the number of new samples is small, the agent chooses an incremental learning strategy to fine-tune the model parameters; if the number of new samples is sufficient and the data distribution has changed significantly, the agent chooses a transfer learning strategy; if the weights of individual features are unbalanced, the agent chooses a feature weight calibration strategy to adjust the weight coefficients in the dynamic feature matrix. Simultaneously, the agent assesses the current system's computing resources and autonomously determines the execution time and resource allocation scheme for the evolution task to avoid impacting normal diagnostic operations.

[0036] During the model evolution phase, the agent invokes encapsulated tools to construct sample sets, trigger training tasks, and record intermediate states. The agent calls the machine learning training platform via API, submits training tasks, monitors task status, and generates a traceable evolution log containing key information for auditing and troubleshooting.

[0037] During the verification and deployment phase, the agent tests and evaluates the updated model, collecting metrics such as diagnostic accuracy and confidence distribution. If the new model outperforms the original model, it is deployed; if the performance is poor or anomalies occur, the agent automatically reverts to the original model to ensure system stability and reliability.

[0038] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A circuit breaker state identification method based on agent analysis and closed-loop feedback, characterized in that, Includes the following steps: The intelligent agent performs fusion processing on multi-source heterogeneous data generated during the operation of the circuit breaker. The fusion processing includes timestamp alignment and standardization processing, and extracts time-domain and frequency-domain features that reflect the equipment status. The intelligent agent calls a time series analysis algorithm to process strong time series data, capture the changing trends before and after the circuit breaker operation, construct a dynamic feature matrix that integrates time correlation and feature dimensions, and calls a deep learning model to decouple and identify the operation intention, distinguishing between normal operation and hidden fault. The intelligent agent identifies the status of the circuit breaker based on a preset knowledge base, and generates diagnostic results and operation suggestions. The knowledge base includes power system operation procedures, maintenance guidelines and expert experience. The intelligent agent continuously collects diagnostic feedback information, autonomously perceives evolutionary needs, and updates model parameters through a closed-loop feedback mechanism to achieve adaptive model evolution.

2. The circuit breaker state identification method based on agent analysis and closed-loop feedback according to claim 1, characterized in that, The sources of the multi-source heterogeneous data include SCADA systems, fault recorders, and protection devices; wherein, the data from the SCADA system includes switch quantity change signals, opening and closing status, and remote / local position signals; the data from the fault recorder includes voltage and current transient waveform data; and the data from the protection device includes action messages and operation log text.

3. The circuit breaker state identification method based on agent analysis and closed-loop feedback according to claim 1, characterized in that, The intelligent agent invokes a deep learning model to decouple and identify the operation intent, including: for the strong temporal correlation of circuit breaker operation, invoking a Transformer-based time series model to extract nonlinear discriminant features between normal voltage drops and voltage drops caused by fault tripping; based on the nonlinear discriminant features, determining whether the nature of the circuit breaker operation is normal operation or abnormal action.

4. The circuit breaker state identification method based on agent analysis and closed-loop feedback according to claim 1, characterized in that, The construction of the dynamic feature matrix that integrates time correlation and feature dimension includes: encapsulating multi-source, multi-dimensional, and time-varying feature information in a structured manner to form a two-dimensional matrix structure containing time dimension and feature dimension; calling a machine learning classification algorithm to analyze the dynamic feature matrix and distinguish the operation type as normal operation and abnormal action.

5. The circuit breaker state identification method based on agent analysis and closed-loop feedback according to claim 1, characterized in that, The knowledge base is a knowledge graph structure, with circuit breakers, protection devices, fault phenomena, and handling measures as core entities. The intelligent agent identifies the status of the circuit breaker based on the preset knowledge base, including: after diagnosing the specific fault nature, querying the knowledge graph to generate professional handling suggestions that comply with power regulations.

6. The circuit breaker state identification method based on agent analysis and closed-loop feedback according to claim 1, characterized in that, The process of generating diagnostic results and operational recommendations includes: structurally combining the fault time, equipment name, fault type, confidence level, and recommended measures to form diagnostic results and operational recommendations output.

7. The circuit breaker state identification method based on agent analysis and closed-loop feedback according to claim 1, characterized in that, The intelligent agent autonomously perceives evolutionary needs, including: continuously monitoring the confidence distribution of its own diagnostic results, human feedback information, and changes in the distribution of newly added data; when it detects that the model performance index continuously declines beyond a preset threshold, the human correction rate exceeds a threshold, or the data distribution drifts significantly, it autonomously judges that the current model is not adaptable enough and triggers evolutionary needs.

8. The circuit breaker state identification method based on agent analysis and closed-loop feedback according to claim 7, characterized in that, The method of updating model parameters through a closed-loop feedback mechanism to achieve adaptive model evolution includes: Strategy planning steps: Based on the new sample size and changes in data distribution, select the corresponding evolutionary strategy, which includes incremental learning strategy, transfer learning strategy, and feature weight calibration strategy; at the same time, evaluate the system's computing resources and formulate the execution time and resource allocation plan for the evolutionary task. Model evolution steps: construct a sample set, trigger training tasks, record key information in the evolution process, and form a traceable evolution log; Verification and deployment steps: Test and evaluate the updated model. If the new model performs better than the original model, deploy it; otherwise, automatically roll back to the original model.

9. The circuit breaker state identification method based on agent analysis and closed-loop feedback according to claim 8, characterized in that, The selection of the corresponding evolutionary strategy includes: if the new sample size is small, selecting the incremental learning strategy to fine-tune the model parameters; if the new sample size is sufficient and the data distribution changes significantly, selecting the transfer learning strategy; if the weights of individual features are unbalanced, selecting the feature weight calibration strategy to adjust the weight coefficients in the dynamic feature matrix.

10. A circuit breaker state identification device based on agent analysis and closed-loop feedback, characterized in that, include: The data fusion processing module is used to fuse multi-source heterogeneous data generated during the operation of the circuit breaker through an intelligent agent. The fusion processing includes timestamp alignment and standardization processing, and extracts time-domain and frequency-domain features that reflect the equipment status. The operation intent identification module is used to process strong time-series data by calling the time-series analysis algorithm through the intelligent agent, capture the change trend before and after the circuit breaker operation, construct a dynamic feature matrix that integrates time correlation and feature dimension, and call a deep learning model to decouple and identify the operation intent, and distinguish between normal operation and hidden fault. The status diagnosis module is used to identify the status of the circuit breaker through the intelligent agent based on a preset knowledge base, and generate diagnostic results and operation suggestions. The knowledge base includes power system operation procedures, maintenance guidelines and expert experience. The closed-loop feedback evolution module is used to continuously collect diagnostic result feedback information through the intelligent agent, autonomously perceive evolutionary needs, update model parameters through a closed-loop feedback mechanism, and realize adaptive evolution of the model.

Images