A substation DC power supply fault monitoring and processing method and system

Abstract

This invention discloses a method and system for monitoring and handling DC power supply faults in substations. The method includes: acquiring a feature library of the operating status of the substation's DC power supply system; constructing topological connections, determining fault impact channels, and obtaining initial fault risk assessment results based on the feature library; acquiring coupling strength coefficients between devices to identify cascading fault modes and correcting the initial risk score to obtain a corrected fault risk assessment result; performing time-series backtracking analysis on the target devices, generating fault feature code sequences, extracting fault trigger data, determining a fault occurrence probability table based on the corrected risk score, and reconstructing the fault state evolution path; integrating the operating status feature library, corrected risk score, coupling strength coefficient, and fault state evolution path to perform gradient distribution analysis and locate key fault source devices; calculating the priority coefficient of the fault source devices, generating a fault handling sequence after risk compensation, and outputting a graded disposal plan to improve the accuracy of fault handling.

Description

Technical Field

[0001] This invention relates to the field of power system fault diagnosis technology, and in particular to a method and system for monitoring and handling DC power supply faults in substations. Background Technology

[0002] The DC power supply system of a substation provides uninterrupted power to relay protection devices, circuit breaker control circuits, signaling systems, and communication equipment. Its reliable operation is directly related to the safety and stability of the substation and even the entire power grid. Due to the intertwined effects of various factors such as aging charging modules, battery deterioration, bus abnormalities, and load fluctuations, DC power supply system faults exhibit characteristics of diverse triggering factors, intertwined impact paths, and widespread impact range, making fault identification and handling decisions difficult.

[0003] Existing DC power supply fault analysis technologies mainly rely on threshold alarms from single-point data and human experience-based judgment. They lack the ability to systematically integrate monitoring information scattered across different time points, making it difficult to extract correlation patterns between devices from massive operational records. Traditional technologies often only trigger responses after faults become apparent, lacking the ability to trace the fault formation process and distinguish which devices are the origin of the fault and which are affected. This results in weak targeting of handling measures and low resource allocation efficiency. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention discloses a method and system for monitoring and handling DC power supply faults in substations, which improves the accuracy of fault monitoring and handling.

[0005] To achieve the above objectives, this invention discloses a method for monitoring and handling DC power supply faults in substations, comprising:

[0006] The system acquires the runtime sequence data and physical equipment information of each target device in the DC power supply system of the substation, and identifies the early degradation characteristic data of each target device based on the runtime sequence data to construct an operating status characteristic library.

[0007] The topological connection relationship of the substation DC power supply system is obtained according to the operating status feature database, and the fault impact channel of the substation DC power supply system is determined according to the topological connection relationship and the operating status feature database.

[0008] The initial fault risk assessment result of each target device is determined based on the fault impact channel, and the coupling strength coefficient between different target devices is obtained based on the operating status feature library;

[0009] Based on the coupling strength coefficient, the cascading failure modes between different target devices are determined, and the initial failure risk assessment results are corrected according to the cascading failure modes to obtain the corrected failure risk assessment results for each target device.

[0010] Based on the operating status feature library, a time-series backtracking analysis is performed on each target device to determine the failure probability table for each target device. The failure probability table is then corrected based on the corrected failure risk assessment results to determine the failure state evolution path for each target device.

[0011] Based on the operating status feature library and the corrected fault risk assessment results, coupling strength coefficient and fault state evolution path of each target device, several fault source devices are determined from the multiple target devices;

[0012] The priority of each of the fault source devices is determined, and a fault handling sequence of several of the fault source devices is determined according to the priority, and the fault handling of the substation DC power supply system is completed based on the fault handling sequence.

[0013] This invention discloses a method for monitoring and handling DC power supply faults in substations. By acquiring operational sequence data and physical equipment information, it identifies early degradation characteristics, constructs an operational status feature library, and systematically integrates scattered monitoring information, solving the problem of insufficient integration of massive amounts of data and providing a comprehensive foundation for subsequent analysis. Based on the operational status feature library, it obtains topological connection relationships and determines fault impact channels. It uses physical connections of equipment to identify fault propagation paths, solving the problem of not being able to extract correlation patterns between equipment. Based on the fault impact channels, it determines the initial fault risk assessment result. Simultaneously, it obtains coupling strength coefficients from the operational status feature library to initially quantify risks and capture inter-equipment dependencies, laying the foundation for corrective assessment. Based on the coupling strength coefficients, it determines cascading fault modes and corrects the initial fault risk assessment result accordingly, considering the fault cascading effect and improving the accuracy of risk assessment. Based on the operational status feature library, it performs time-series backtracking analysis to determine a fault occurrence probability table. This table is then corrected in conjunction with the corrected fault risk assessment result to determine the fault state evolution path. It backtracks historical data to predict fault development, solving the problem of not being able to trace the fault formation process. By integrating the operational status feature library, correction assessment, coupling coefficient, and evolution path, the fault source equipment is identified, the source is accurately located, and the fault source and affected objects are distinguished.

[0014] Finally, the priority of the fault sources was determined, a fault handling sequence was formulated, fault handling was completed, resource allocation was optimized, and precise fault handling was achieved.

[0015] As a preferred example, the step of acquiring the runtime sequence data and physical equipment information of each target device in the substation DC power supply system, and identifying early degradation characteristic data of each target device based on the runtime sequence data to construct an operating status characteristic library, includes:

[0016] For any target device in the DC power supply system of the substation:

[0017] The real-time internal resistance, real-time discharge current, initial internal resistance, and real-time voltage and temperature of each individual battery cell in the target device are obtained.

[0018] The internal resistance growth rate of the target device is obtained based on the real-time internal resistance and the initial internal resistance.

[0019] The discharge capacity of the target device is obtained based on the real-time discharge current.

[0020] The average voltage of the target device is determined based on the real-time voltage of all the individual cells in the target device, so as to obtain the voltage dispersion of each individual cell based on the average voltage and the real-time voltage;

[0021] For any single battery cell, if the real-time temperature of the single battery cell is higher than a preset temperature threshold, the single battery cell is determined to be an abnormal battery.

[0022] The number of abnormal batteries is obtained, and the internal resistance growth rate, discharge capacity and voltage dispersion are marked according to the preset degradation degree threshold to obtain the early degradation characteristic data of the target device.

[0023] The real-time internal resistance, real-time discharge current, real-time voltage, and real-time temperature are collected according to a preset sampling frequency to obtain the runtime sequence data of each target device.

[0024] Obtain the device identifier, physical location, and port physical connection relationship of each target device to obtain the physical device information of each target device;

[0025] Based on the device identifier, the operating sequence data, physical device information and early degradation characteristic data corresponding to each target device are associated to obtain the operating status characteristic library of the substation DC power supply system.

[0026] The above-described scheme acquires real-time internal resistance, real-time discharge current, initial internal resistance, real-time voltage, and real-time temperature, providing comprehensive parameter coverage of the equipment's operating status and avoiding omission of key degradation indicators. It obtains the internal resistance growth rate based on real-time and initial internal resistance, quantifying the degree of equipment aging and directly reflecting internal degradation trends. It obtains the discharge capacity based on real-time discharge current, assessing the actual performance degradation of the equipment and supplementing degradation characteristics in energy output. It determines the average voltage based on the real-time voltage of all individual cells, and then combines this with the real-time voltage to obtain the voltage dispersion of each individual cell, identifying inconsistencies between cells and capturing potential fault points. For any individual cell, if the real-time temperature exceeds a preset temperature threshold, it is determined to be an abnormal cell. Timely detection of overheating risks prevents the spread of localized faults; acquisition of the number of abnormal batteries and labeling of internal resistance growth rate, discharge capacity, and voltage dispersion based on preset degradation thresholds, systematically marking early fault characteristics and improving feature recognition accuracy; collection of real-time data according to preset sampling frequency to form runtime sequence data, ensuring data continuity and feasibility of time-series analysis; acquisition of device identifiers, physical locations, and port physical connection relationships to provide device physical context and support topology association; finally, construction of an operating status feature library based on device identifier association with runtime sequence data, physical device information, and early degradation feature data, achieving multi-source data integration and providing a comprehensive and accurate foundation for subsequent fault monitoring.

[0027] As a preferred example, the step of obtaining the topological connection relationship of the substation DC power supply system according to the operating status feature database, and determining the fault impact channel of the substation DC power supply system according to the topological connection relationship and the operating status feature database, includes:

[0028] The topology of the DC power supply system of the substation is obtained based on the physical connection relationship of the ports, and the topology is decomposed into multiple charger power supply paths and multiple load power supply paths; wherein, the charger power supply path is the path from the charging module to the bus to the battery; and the load power supply path is the path from the bus to the load.

[0029] For any one of the multiple charger power supply paths and multiple load power supply paths, the voltage transmission sequence between different target devices on that path is obtained based on the runtime sequence data, so as to determine the location of abnormal voltage drop according to the voltage transmission sequence.

[0030] Based on the location of the abnormal voltage drop, multiple transmission interruption devices are determined on the path, and time-series fluctuation analysis is performed on the path based on the multiple transmission interruption devices to obtain the path convergence corresponding to the path.

[0031] Based on the path convergence, multiple fault-affected channels of the substation DC power supply system are determined from multiple charger power supply paths and multiple load power supply paths.

[0032] The above scheme obtains the topology connection relationship based on the physical connection relationship of the ports, ensuring the physical authenticity and reliability of the topology and providing an accurate basis for subsequent path decomposition. Secondly, the topology connection relationship is decomposed into multiple charger power supply paths and multiple load power supply paths. By distinguishing path types, the propagation characteristics of faults in different power supply directions are clarified, facilitating targeted analysis. Next, for each path, the voltage transmission sequence between different target devices on that path is obtained based on runtime timing data. The dynamic characteristics of timing data are used to capture voltage change sequences and identify the timing patterns of fault propagation. Finally, the location of abnormal voltage drops is determined based on the voltage transmission sequence, accurately locating the voltage. The sudden drop point reveals the location of potential fault sources. Then, based on the location of the abnormal voltage drop, multiple transmission interruption devices along the path are identified, identifying devices that may interrupt power supply and quantifying the scope of fault impact. Next, time-series fluctuation analysis is performed on the path based on these multiple transmission interruption devices. By analyzing the time-series fluctuations of the interruption devices, the stability and anti-interference capability of the path are evaluated. Subsequently, the path convergence degree is obtained, quantifying the fault convergence degree of the path and reflecting its vulnerability to faults. Finally, based on the path convergence degree, multiple fault impact channels are identified from multiple paths, prioritizing paths with high convergence degrees as key impact channels to improve identification accuracy and targeting.

[0033] As a preferred example, the step of determining multiple transmission interruption devices on the path based on the location of the abnormal voltage drop, and performing time-series fluctuation analysis on the path based on the multiple transmission interruption devices to obtain the path convergence, includes:

[0034] For any target device on the path, obtain the voltage timing data of the target device in the operating status feature library. When there is a voltage lower than a preset voltage threshold in the voltage timing data, determine that the target device is a transmission interruption device.

[0035] Obtain the number of target devices and the number of transmission interruption devices for all target devices on the path, and use the device tree ratio of the number of transmission interruption devices to the number of target devices as the path convergence of the path.

[0036] The above scheme identifies multiple transmission interruption devices along the path based on the location of abnormal voltage drops. This location-based approach pinpoints potential fault points and provides a starting point for subsequent analysis. Time-series fluctuation analysis is performed on the path based on multiple transmission interruption devices, utilizing data from multiple devices for overall evaluation and avoiding the limitations of single-point analysis. For any target device along the path, its voltage time-series data from the operational status feature library is obtained, ensuring the real-time nature and reliability of the data source. When the voltage time-series data shows a voltage below a preset threshold, the target device is identified as a transmission interruption device. This preset threshold provides an objective standard for accurate identification of interruption devices and reduces misjudgments. The number of target devices and the number of transmission interruption devices along the path are obtained, providing a basis for quantification of the overall device status. The ratio of the number of transmission interruption devices to the number of target devices can be used as the path convergence. This ratio directly quantifies the fault concentration of the path; a higher ratio indicates lower convergence and more severe fault impact, thus achieving a standardized evaluation of path convergence.

[0037] As a preferred example, the step of determining the initial fault risk assessment result for each target device based on the fault impact channel, and obtaining the coupling strength coefficient between different target devices based on the operating status feature library, includes:

[0038] Based on the fault impact channel and the operating status feature library of the substation DC power supply system, multiple affected devices and the connection relationships between different affected devices are determined from multiple target devices, so as to construct an affected device connection diagram based on the connection relationships;

[0039] For any affected device in any of the fault-affected channels, the affected device is used as the starting point for traversal of the connection graph of the affected devices to obtain the fault propagation path corresponding to the affected device.

[0040] For any fault propagation path, obtain the fault impact degree and hierarchical weight coefficient corresponding to each affected device on the fault propagation path, so as to obtain the path impact degree of the fault propagation path based on the fault impact degree and the hierarchical weight coefficient.

[0041] The influence of all the fault propagation paths is organized according to the propagation level to obtain the propagation path table of the substation DC power supply system.

[0042] Obtain the correlation index of each target device in the propagation path table, and determine the initial fault risk assessment result of each target device based on the correlation index and the path influence degree;

[0043] The operating status feature library is subjected to time-series feature analysis to obtain the time-series correlation between different target devices, and the coupling strength coefficient between different target devices is obtained based on the time-series correlation.

[0044] The above scheme identifies affected devices and their connections based on the fault impact channels and operational status feature database, constructing an affected device connection graph. This provides a structured foundation for subsequent path traversal, ensuring that risk assessment is based on the actual system topology rather than isolated data. Next, the affected devices are used as starting points to traverse the connection graph, generating fault propagation paths and simulating the fault's spread among devices to help identify potential cascading paths. Then, for each fault propagation path, the path impact degree is calculated by combining the fault's impact level and hierarchical weight coefficients. The hierarchical weight coefficients reflect the importance of the device in the system, making the impact quantification more closely match the actual risk distribution. Afterward, all path impact degrees are organized into a propagation path table according to the propagation hierarchy, forming a global view for easy analysis of the overall propagation pattern. Based on the propagation path table, the initial fault risk assessment results are determined using correlation indicators and path impact degrees. Correlation indicators capture the role of devices in propagation, comprehensively improving assessment accuracy. Finally, temporal feature analysis is performed on the operational status feature database to obtain the temporal correlation relationships between devices, and coupling strength coefficients are calculated accordingly. The temporal correlation relationships consider dynamic data changes, enhancing the predictive ability for cascading faults.

[0045] As a preferred example, the step of performing time-series feature analysis on the operating state feature library to obtain the time-series correlation between different target devices, and obtaining the coupling strength coefficient between different target devices based on the time-series correlation, includes:

[0046] Runtime sequence data of each target device is extracted from the running status feature library, and cross-correlation analysis is performed on different target devices based on the runtime sequence data to obtain the device timing correlation relationship between different target devices;

[0047] The load timing data of each target device is obtained based on the runtime timing data, so as to obtain the load fluctuation characteristics of each target device according to the device timing correlation and the load timing data.

[0048] A load fluctuation map is constructed based on all the load fluctuation features, and a similarity detection is performed on the load fluctuation features of different target devices in the load fluctuation map to generate a device similarity table between different target devices.

[0049] The correlation stability between different target devices is obtained based on the device similarity table, and a weak correlation correction factor between different target devices is determined based on the correlation stability.

[0050] The basic similarity between different target devices is determined according to the device similarity table, and the basic similarity is corrected according to the weak correlation correction factor to obtain the coupling strength coefficient between different target devices.

[0051] The above scheme extracts runtime sequence data for each target device from the runtime status feature library, providing a temporal basis for device operating status and ensuring that the analysis is based on actual operating records. It also performs cross-correlation analysis on different target devices based on the runtime sequence data, capturing the temporal correlation between devices and thus obtaining the temporal relationship between them. This avoids the limitations of single-point data and systematically integrates device interaction patterns. Load time-series data for each target device is obtained from the runtime sequence data. Combined with the temporal relationship between devices and the load time-series data, the load fluctuation characteristics of each target device are obtained. By fusing load change data with the correlation, the fluctuation pattern of device load is identified, enhancing the comprehensiveness of feature extraction. A load fluctuation map is constructed based on all load fluctuation characteristics, structurally displaying the load fluctuation relationship between devices for subsequent analysis. Furthermore, similarity detection is performed on the load fluctuation characteristics of different target devices in the load fluctuation map, directly comparing device fluctuation patterns and generating a device similarity table between different target devices. This quantifies device similarity and lays the foundation for stability assessment. The stability of associations between different target devices is obtained based on the device similarity table to assess the reliability of the associations. A weak correlation correction factor can be determined based on the association stability, and a correction mechanism is introduced for unstable or weakly correlated associations to eliminate bias. The basic similarity between different target devices is determined based on the device similarity table to obtain initial similarity values. The basic similarity can be corrected based on the weak correlation correction factor, and the basic values ​​are adjusted by applying the correction factor to finally obtain the coupling strength coefficient between different target devices. This ensures that the coefficient accurately reflects the true coupling strength and supports subsequent fault mode identification.

[0052] As a preferred example, the step of determining the cascading failure modes among different target devices based on the coupling strength coefficient, and then correcting the initial failure risk assessment result based on the cascading failure modes to obtain a corrected failure risk assessment result for each target device, includes:

[0053] Based on the coupling strength coefficient, multiple strongly coupled device pairs in the substation DC power supply system are determined.

[0054] The strongly coupled device is used to trace the cascading propagation path of the fault in the DC power supply system of the substation, so as to identify multiple cascading fault modes corresponding to the DC power supply system of the substation based on the cascading propagation path.

[0055] For any one of the target devices, the number of times the target device appears in the multiple cascading failure modes is obtained, and the number of occurrences is used as the failure participation degree of the target device.

[0056] The initial fault risk assessment result is corrected based on the fault participation rate to obtain the corrected fault risk assessment result for each target device.

[0057] The above scheme identifies strongly coupled device pairs based on the coupling strength coefficient, directly identifying strong correlations between devices, focusing on high-risk device combinations, avoiding blind analysis of all devices, and ensuring efficient resource allocation. It tracks the cascading propagation path of faults in the system based on these strongly coupled device pairs, revealing the mechanism of fault propagation from local to global by simulating the fault diffusion process, laying the foundation for identifying cascading patterns. Based on the cascading propagation path, it identifies multiple cascading fault modes corresponding to the system, defines specific fault scenarios based on the propagation path, and transforms abstract propagation into an operational mode. For any target device, it obtains the number of times it appears in the cascading fault mode, statistically analyzes the frequency of the device in the identified mode, and quantifies its degree of participation in the cascading fault. The number of occurrences is used as the fault participation degree, and the statistical value is transformed into an indicator that directly reflects the role and weight of the device in the cascading fault. The initial fault risk assessment result is corrected based on the fault participation degree, integrating the participation degree into the initial assessment and dynamically adjusting the risk value to include cascading effects. The corrected fault risk assessment result for each target device is obtained, and the corrected assessment is output, providing a more reliable basis for subsequent fault location and handling.

[0058] As a preferred example, the step of performing time-series backtracking analysis on each target device based on the operating state feature library to determine a fault occurrence probability table for each target device, and then revising the fault occurrence probability table based on the revised fault risk assessment results to determine the fault state evolution path for each target device, includes:

[0059] For any one of the target devices, after discretizing and classifying the runtime sequence data of the target device, a fault feature code sequence of the target device is formed;

[0060] The fault signature sequence is subjected to time-series backtracking analysis to identify the period of cross-level sudden change, and the operation data of the period of cross-level sudden change is extracted as fault triggering data.

[0061] Based on the fault triggering data and the historical fault database of the target device, a fault correlation analysis is performed to obtain a fault occurrence probability table for the target device; wherein, the fault occurrence probability table records the fault type and fault occurrence probability corresponding to each fault triggering data;

[0062] Based on the corrected fault risk assessment results of the target equipment, the fault occurrence probability is weighted and corrected to obtain the corrected fault occurrence probability table.

[0063] From the revised fault occurrence probability table, extract the historical fault feature code sequence with increasing fault occurrence probability as the fault state evolution path of the target device.

[0064] The above scheme discretizes runtime sequence data into fault feature code sequences, transforming continuous data into structured hierarchical sequences to facilitate the identification of abnormal change points and provide an operational basis for subsequent analysis. Time-series backtracking analysis of the fault feature code sequences identifies periods of abrupt changes across levels and extracts runtime data as fault triggering data. By backtracking historical data, abrupt events are captured; these abrupt change periods represent potential fault trigger points, ensuring the effective utilization of critical data. Fault correlation analysis is performed based on fault triggering data and a historical fault database to obtain a fault occurrence probability table. By combining historical fault records with current triggering data, fault types and probabilities are quantified, and a probability table is established to reflect the likelihood of faults. The fault occurrence probabilities are weighted and corrected according to the corrected fault risk assessment results to obtain a corrected fault occurrence probability table. The probability weights are adjusted using the corrected risk assessment results (such as inter-equipment cascading effects) to enhance the reliability of the probability table and reflect dynamic risk factors. Historical fault feature code sequences with increasing fault occurrence probabilities are extracted from the corrected fault occurrence probability table as fault state evolution paths. By extracting increasing probability sequences, the fault evolution trend is visually displayed, providing a clear path for fault source localization.

[0065] As a preferred example, the step of determining several fault source devices from multiple target devices based on the operating state feature library and the modified fault risk assessment results, coupling strength coefficient, and fault state evolution path of each target device includes:

[0066] Based on the operating status feature library, the corrected fault risk assessment results of each target device and the fault status evolution path, the parameter gradient distribution analysis of multi-source data is performed to identify the set of target devices whose parameter gradient values ​​exceed the gradient threshold as high gradient clustering areas.

[0067] At the boundary of the high gradient cluster region, identify the region where the parameter gradient value decreases significantly, and generate a gradient reduction marker.

[0068] Based on the gradient descent marker, the parameter gradients of adjacent target devices are traced step by step along the direction of parameter gradient descent to obtain the low gradient propagation path.

[0069] Based on the corrected fault risk assessment results, a set of hidden fault candidate target devices is selected from the low-gradient transmission path;

[0070] Calculate the gradient correlation degree of each hidden fault candidate target device to the high gradient clustering area via the low gradient transmission path;

[0071] Multiple fault source devices are selected from the set of hidden fault candidate target devices based on the gradient correlation.

[0072] The above scheme performs parameter gradient distribution analysis on multi-source data based on the operational status feature library, the corrected fault risk assessment results, and the fault state evolution path. This comprehensively captures dynamic changes between devices, avoiding the bias caused by relying on single-point data, thus laying the foundation for subsequent steps. Identifying the set of target devices whose parameter gradient values ​​exceed the gradient threshold as high-gradient clusters helps focus on fault hotspots, but further differentiation between fault sources and affected points is needed. Identifying regions where parameter gradient values ​​significantly decrease at the boundaries of high-gradient clusters and generating gradient reduction markers effectively captures gradient abrupt change points, indicating potential fault propagation starting points and overcoming the shortcomings of traditional methods that ignore boundary changes. Based on the gradient reduction markers, the parameter gradients of adjacent target devices are traced step-by-step along the direction of parameter gradient reduction to obtain low-gradient propagation paths. By tracing low-gradient regions, hidden fault sources can be discovered, preventing the omission of critical equipment. Based on the corrected fault risk assessment results, a set of candidate target devices for hidden faults is selected from the low-gradient propagation paths. Combining this with the risk assessment, high-risk but low-gradient devices are prioritized, enhancing the targeting of hidden faults. The gradient correlation degree of each hidden fault candidate device to the high gradient clustering area via the low gradient propagation path is calculated to quantify the influence relationship between devices, ensuring that the selected devices are strongly correlated with the fault area and improving the identification accuracy. Based on the gradient correlation degree, multiple fault source devices are selected from the set of hidden fault candidate devices. The selection process of fault source is optimized through correlation degree screening, and finally, accurate fault location is achieved.

[0073] On the other hand, the present invention discloses a substation DC power supply fault monitoring and processing system, including an operation characteristic module, a fault channel module, a coupling analysis module, a risk assessment module, a fault evolution module, a fault location module, and a fault processing module.

[0074] The operation feature module is used to acquire the operation sequence data and physical equipment information of each target device in the DC power supply system of the substation, and to identify the early deterioration feature data of each target device based on the operation sequence data, so as to construct an operation status feature library.

[0075] The fault channel module is used to obtain the topology connection relationship of the substation DC power supply system according to the operating status feature library, so as to determine the fault impact channel of the substation DC power supply system according to the topology connection relationship and the operating status feature library.

[0076] The coupling analysis module is used to determine the initial fault risk assessment result of each target device according to the fault impact channel, and to obtain the coupling strength coefficient between different target devices according to the operating status feature library;

[0077] The risk assessment module is used to determine the cascading failure modes between different target devices based on the coupling strength coefficient, and to correct the initial failure risk assessment result based on the cascading failure modes to obtain the corrected failure risk assessment result for each target device.

[0078] The fault evolution module is used to perform time-series backtracking analysis on each target device according to the operating status feature library, determine the fault occurrence probability table of each target device, and revise the fault occurrence probability table according to the revised fault risk assessment result to determine the fault state evolution path of each target device.

[0079] The fault location module is used to determine several fault source devices from multiple target devices based on the operating status feature library and the modified fault risk assessment results, coupling strength coefficient and fault state evolution path of each target device.

[0080] The fault handling module is used to determine the priority of each fault source device, to determine a fault handling sequence of several fault source devices according to the priority, and to complete the fault handling of the substation DC power supply system based on the fault handling sequence.

[0081] This invention discloses a substation DC power supply fault monitoring and handling system. By acquiring operational sequence data and physical equipment information, it identifies early degradation characteristics, constructs an operational status feature library, and systematically integrates scattered monitoring information, solving the problem of lack of integration for massive amounts of data and providing a comprehensive foundation for subsequent analysis. Based on the operational status feature library, it obtains topological connection relationships and determines fault impact channels. It uses physical connections of equipment to identify fault propagation paths, solving the problem of not being able to extract correlation patterns between equipment. Based on the fault impact channels, it determines the initial fault risk assessment result. Simultaneously, it obtains coupling strength coefficients from the operational status feature library to initially quantify risks and capture inter-equipment dependencies, laying the foundation for corrective assessment. Based on the coupling strength coefficients, it determines cascading fault modes and corrects the initial fault risk assessment result accordingly, considering the fault cascading effect and improving the accuracy of risk assessment. Based on the operational status feature library, it performs time-series backtracking analysis to determine a fault occurrence probability table. This table is then corrected in conjunction with the corrected fault risk assessment result to determine the fault state evolution path. It backtracks historical data to predict fault development, solving the problem of not being able to trace the fault formation process. By integrating the operational status feature library, correction assessment, coupling coefficient, and evolution path, the fault source equipment is identified, the source is accurately located, and the fault source and affected objects are distinguished.

[0082] Finally, the priority of the fault sources was determined, a fault handling sequence was formulated, fault handling was completed, resource allocation was optimized, and precise fault handling was achieved. Attached Figure Description

[0083] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0084] Figure 1 This is a flowchart illustrating a method for monitoring and handling DC power supply faults in a substation, as disclosed in an embodiment of the present invention.

[0085] Figure 2 This is a schematic diagram of the structure of a substation DC power supply fault monitoring and processing system disclosed in an embodiment of the present invention. Detailed Implementation

[0086] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0087] See Figure 1 To address the technical problem of existing technologies being unable to accurately and quickly handle faults, this embodiment provides a method for monitoring and handling DC power supply faults in substations, mainly including:

[0088] Step 101: Obtain the runtime sequence data and physical equipment information of each target device in the DC power supply system of the substation, and identify the early degradation characteristic data of each target device based on the runtime sequence data to construct an operating status characteristic library.

[0089] Step 102: Obtain the topology connection relationship of the substation DC power supply system according to the operating status feature library, so as to determine the fault impact channel of the substation DC power supply system according to the topology connection relationship and the operating status feature library;

[0090] Step 103: Determine the initial fault risk assessment result of each target device according to the fault impact channel, and obtain the coupling strength coefficient between different target devices according to the operating status feature library;

[0091] Step 104: Determine the cascading failure modes between different target devices based on the coupling strength coefficient, and correct the initial failure risk assessment results based on the cascading failure modes to obtain the corrected failure risk assessment results for each target device;

[0092] Step 105: Perform time-series backtracking analysis on each target device according to the operating status feature library, determine the failure probability table for each target device, and revise the failure probability table according to the revised failure risk assessment results to determine the failure state evolution path of each target device.

[0093] Step 106: Based on the operating status feature library and the modified fault risk assessment results, coupling strength coefficient and fault state evolution path of each target device, determine several fault source devices from the multiple target devices;

[0094] Step 107: Determine the priority of each of the fault source devices, determine a fault handling sequence of several of the fault source devices according to the priority, and complete the fault handling of the substation DC power supply system based on the fault handling sequence.

[0095] In this embodiment, step 101 includes:

[0096] Step 1011: For any target device in the DC power supply system of the substation: obtain the real-time internal resistance, real-time discharge current, initial internal resistance of the target device, and the real-time voltage and real-time temperature of each individual cell in the target device.

[0097] Step 1012: Obtain the internal resistance growth rate of the target device based on the real-time internal resistance and the initial internal resistance.

[0098] Step 1013: Obtain the discharge capacity of the target device based on the real-time discharge current.

[0099] Step 1014: Determine the average voltage of the target device based on the real-time voltage of all the individual cells in the target device, so as to obtain the voltage dispersion of each individual cell based on the average voltage and the real-time voltage.

[0100] Step 1015: For any one of the individual cells, when the real-time temperature of the individual cell is higher than a preset temperature threshold, the individual cell is determined to be an abnormal cell.

[0101] Step 1016: Obtain the number of abnormal batteries, and mark the internal resistance growth rate, discharge capacity and voltage dispersion according to the preset degradation degree threshold to obtain the early degradation characteristic data of the target device.

[0102] Step 1017: Collect the real-time internal resistance, the real-time discharge current, the real-time voltage, and the real-time temperature according to the preset sampling frequency to obtain the running sequence data of each target device.

[0103] Step 1018: Obtain the device identifier, physical location and port physical connection relationship of each target device to obtain the physical device information of each target device.

[0104] Step 1019: Based on the device identifier, associate the runtime sequence data, physical device information and early degradation characteristic data corresponding to each target device to obtain the operating status characteristic library of the substation DC power supply system.

[0105] This embodiment takes the DC power supply system of a 220kV substation as an example. The system includes target equipment such as charging modules, battery banks, DC buses, and loads. There are two battery banks, each consisting of 110 individual batteries connected in series. Data acquisition devices are deployed at the various target equipment nodes of the substation's DC power supply system. Specifically, a voltage acquisition module is installed at the terminal of each individual battery in each battery bank target equipment to collect the real-time voltage of each individual battery. The sampling frequency is set to once every 30 seconds.

[0106] Current sensors are installed at the positive and negative busbars of the battery pack to collect the charging current and real-time discharge current during the discharge process. Temperature sensors are placed at key points on the surface of the battery pack and inside the battery cabinet to collect the real-time temperature of each individual battery cell and the ambient temperature of the battery cabinet. An internal resistance measurement module is used to periodically measure the real-time internal resistance of each individual battery cell in the target device. The initial internal resistance of each individual battery cell is its pre-stored reference measurement value before operation. Voltage and current monitoring devices are installed at the output terminal of each charging module in the target device to collect its output voltage and output current data.

[0107] The voltage, current, temperature, and internal resistance data collected synchronously by all the aforementioned sensors according to timestamps are categorized and organized according to device identification to form continuous and ordered runtime sequence data corresponding to each target device. For example, for a battery pack with 110 cells, its runtime sequence data at a sampling point includes 110 voltage values, 1 current value, multiple temperature values, and a periodic sequence of internal resistance values.

[0108] This embodiment focuses on the early degradation stage, which is the period when equipment performance begins to decline but has not yet triggered a serious failure. For example, the actual capacity decay of the battery pack has not yet exceeded 10% of its rated capacity. During this stage, the following key features are extracted from the runtime sequence data: First, for the target device, the battery pack, its actual internal resistance and initial internal resistance are obtained to calculate the internal resistance growth rate, directly reflecting the health degradation within the battery. For example, if the initial internal resistance of a single cell is 0.5 mΩ and the current real-time internal resistance is 0.68 mΩ, then its internal resistance growth rate is 36%. Second, for the target device, the battery pack, during a complete discharge process, the collected real-time discharge current is integrated over time to calculate its actual discharge capacity. For example, if the average discharge current is 100 A and the discharge lasts for 2 hours, the calculated discharge capacity is 200 Ah. Comparing this value with the rated capacity allows for the assessment of capacity retention. Furthermore, for the target battery pack, to assess its consistent degradation trend, the following steps are used to calculate voltage dispersion: A statistical time window (e.g., 24 hours) is set, and all voltage samples of each individual cell within this window are extracted. Then, the average voltage of the battery pack within this time window is calculated. Based on the deviation of each individual cell's voltage value from the average voltage, the voltage balance within the battery pack is quantified. Increased dispersion is a significant indicator of early consistent degradation. Finally, for the target battery pack, the real-time temperature of all its individual cells is analyzed. A cell is considered to have an abnormal temperature if its real-time temperature meets any of the following conditions: the cell temperature is more than 10°C higher than the ambient temperature, or the temperature difference between the cell and the average temperature of the same group of cells exceeds 5°C. The number of all cells with abnormal temperatures in the target device is counted to obtain the characteristic of the number of abnormal cells.

[0109] The calculated internal resistance growth rate, discharge capacity, voltage dispersion, and number of abnormal batteries are compared with preset degradation thresholds to complete early degradation labeling and form early degradation characteristic data. Examples of thresholds are as follows: if the voltage dispersion exceeds 0.1V, it is labeled as "voltage dispersion degradation"; if the internal resistance growth rate exceeds 30%, it is labeled as "internal resistance growth degradation"; if the discharge capacity is less than 90% of the rated capacity, it is labeled as "capacity decay degradation"; if the number of abnormal batteries exceeds 10% of the total number of batteries, it is labeled as "temperature consistency degradation". Simultaneously, the physical equipment information of each target device is acquired. This information includes: Equipment identifier: a unique identifier obtained from the data acquisition device or asset management system. For example, the equipment identifier for charging module 1 is "CHG-01", and the equipment identifier for battery group A is "BAT-A"; Physical location: the installation location is recorded using structured coding, such as "distribution room - cabinet 01 - floor 03". Port physical connection relationship: The electrical connection relationship of the device ports is recorded in a structured manner, in the format of: {Device Identifier: CHG-01, Output Port: P1, Connected to: Device Identifier: DC-BUS-I, Input Port: IN1}. This information fully describes the topology relationship of power input and output.

[0110] Finally, using the device identifier as the core key field, the runtime sequence data, physical device information, and early degradation characteristic data corresponding to each target device are correlated and integrated to construct a unified runtime status characteristic library. Logically, this library includes a raw time-series data table, a degradation characteristic data table, and a device basic information table, and allows for correlation queries through the device identifier, providing comprehensive data support for all subsequent analysis steps.

[0111] In this embodiment, step 102 includes:

[0112] Step 1021: Obtain the topology connection relationship of the substation DC power supply system based on the physical connection relationship of the ports, and decompose the topology connection relationship into multiple charger power supply paths and multiple load power supply paths; wherein, the charger power supply path is the path from the charging module to the bus to the battery; the load power supply path is the path from the bus to the load.

[0113] Step 1022: For any one of the multiple charger power supply paths and multiple load power supply paths, obtain the voltage transmission sequence between different target devices on that path based on the runtime sequence data, so as to determine the location of abnormal voltage drop according to the voltage transmission sequence.

[0114] Step 1023: Based on the location of the abnormal voltage drop, determine multiple transmission interruption devices on the path, and perform time-series fluctuation analysis on the path based on the multiple transmission interruption devices to obtain the path convergence degree corresponding to the path; wherein, for any target device on the path, obtain the voltage time-series data of the target device in the operating status feature library, and when there is a voltage lower than a preset voltage threshold in the voltage time-series data, determine that the target device is a transmission interruption device; obtain the number of target devices of all target devices on the path and the number of transmission interruption devices of all transmission interruption devices, and use the device tree ratio of the number of transmission interruption devices to the number of target devices as the path convergence degree of the path.

[0115] Step 1024: Determine multiple fault-affected channels of the substation DC power supply system from multiple charger power supply paths and multiple load power supply paths based on the path convergence.

[0116] In this embodiment, the physical connection relationship of each target device's ports is first extracted from the device basic information table in the operating status feature library. This information is recorded in a structured form, clearly indicating the upstream device identifier connected to the device's input port and the downstream device identifier connected to the device's output port. For example: {Device Identifier: CHG-01, Output Port: P1, Connected to: {Device Identifier: DC-BUS-I, Input Port: IN1}}, indicating that the output port P1 of the charging module CHG-01 is connected to the input port IN1 of the DC bus DC-BUS-I.

[0117] Next, based on the physical connection relationships of all devices' ports, the topology of the substation's DC power supply system is constructed. Preferably, a directed graph model is used for representation: the vertices of the graph represent each target device (including charging module nodes, battery bank nodes, DC bus nodes, and load nodes); the edges of the graph represent the electrical connections between devices, and the direction of the edges represents the direction of power transmission; each edge is labeled with its connection type (such as "charging power supply", "load power supply", "charging / discharging") and electrical parameters (rated voltage, rated current).

[0118] Taking the 220kV substation in this embodiment as an example, its DC power supply system includes 4 charging module nodes, 2 sets of battery bank nodes, 2 DC bus nodes, and 15 load nodes. The directed graph constructed based on the physical connection relationship contains 23 vertices and corresponding directed edges. Each edge is labeled with parameters such as rated voltage (e.g., the rated output voltage of a certain connection edge is 230V) and rated current (e.g., 50A). The topology connection relationship is stored as graph structure data, including node set, edge set, and node attributes and edge attributes.

[0119] Next, the topological connection relationship is decomposed into two types of functional sub-path sets, such as charger power supply path sets and load power supply path sets. The directed graph is traversed to identify all paths starting from any charging module node, passing through the DC bus node, and finally reaching the battery pack node. This embodiment identifies a total of 8 charger power supply paths. For example, a typical path is charging module CHG-01 → DC bus I → battery pack A, with node voltages at 232V, 230V, and 228V respectively, exhibiting a progressively decreasing characteristic. Another charger power supply path is charging module CHG-02 → DC bus II → battery pack B. This path serves as a backup power supply channel or identifies all paths starting from the DC bus node and reaching each load node. This embodiment identifies a total of 15 load power supply paths. For example, DC bus I → control load (power supply object is circuit breaker control circuit), DC bus I → power load (power supply object is electric mechanism), DC bus I → communication load (power supply object is communication equipment), etc.

[0120] In this embodiment, each power supply path records its starting node, ending node, and the sequence of intermediate nodes, and labels the path. For each charger power supply path and load power supply path extracted above, the voltage time series data of each node (target device) on the path is extracted from the original time series data table of the operating status feature library. The monitoring data includes the node voltage value and sampling timestamp to analyze the voltage transmission sequence in the power transmission process. Under normal operating conditions, the voltage decreases slightly step by step along the power transmission direction (from the starting node to the ending node), and the voltage of each node should be stable within the rated operating range. In this embodiment, the voltage threshold is set as: below 85% of the rated voltage or above 115% of the rated voltage. When the real-time voltage of a node is below 85% of the rated voltage or above 115% of the rated voltage, it is marked as a voltage anomaly point.

[0121] To accurately pinpoint the first abnormal location on the path affected by the fault, this embodiment employs the following determination logic: along the power transmission direction of the path (from the power source to the load), the first node identified as having a voltage anomaly is determined as the location of the voltage drop anomaly on the path. The target device corresponding to this location is the transmission interruption device on the path, and its device identification, voltage drop magnitude, and occurrence time are recorded.

[0122] For example, in a charger power supply path from charging module CHG-01 → DC bus I → battery pack A, at a specific moment, the charging module output voltage is 230V (normal), and the voltage of DC bus I suddenly drops to 195V (below 85% of the rated voltage of 230V), causing the voltage at battery pack A to drop to 190V. At this time, the node of DC bus I is marked as a voltage anomaly point, and it is the first anomaly point along the transmission direction. Therefore, DC bus I is determined to be the location of the voltage anomaly drop in this path, and the equipment in DC bus I is the transmission interruption device in this path.

[0123] For any target device on any power supply path, retrieve the device's voltage timing data from the operating status feature library. If the voltage timing data shows a voltage lower than a preset voltage threshold (85% of the rated voltage), the target device is determined to be a transmission interruption device. Here, the transmission interruption device is consistent with the device at the abnormal voltage drop location determined in step 1022, and may also include other devices on the path that meet the condition of voltage below the threshold.

[0124] For each power supply path, the following two parameters are calculated: Total number of target devices (N_path): the total number of target device nodes included in this path; Number of devices with transmission interruption (N_interrupt): the number of target devices on this path that are identified as transmission interruption devices. Then, the path convergence is calculated based on the ratio of the number of transmission interruption devices to the total number of target devices. Path convergence measures the degree of concentration of fault impact on this path; a higher convergence indicates that the path carries more faulty nodes and is the main carrier of fault propagation.

[0125] To improve the accuracy of fault channel identification, this embodiment further performs joint analysis and deep feature extraction on all transmission interruption devices on the power supply path to form an interruption location set. Specifically, for each transmission interruption device, its voltage drop curve is analyzed. The voltage drop critical point is defined as the moment when the voltage drop rate changes significantly. The location of the abrupt change in the drop rate is identified by taking the second derivative of the voltage curve. For example, during the voltage drop process of a charging module, the voltage is 230V at time T1, drops to 210V at time T2, drops to 195V at time T3, and drops to 185V at time T4. Then, times T2, T3, and T4 are adjacent drop critical points. The time interval between adjacent drop critical points reflects the stage characteristics of voltage drop. The drop interval features of the transmission interruption device are extracted, including: interval duration sequence: the time interval between adjacent critical points, such as 15 seconds, 22 seconds, 18 seconds, and 30 seconds; interval change rate: the ratio of the difference between adjacent intervals to the previous interval. Abnormal interval points in the interval sequence are marked based on the drop interval feature. When the duration of an interval exceeds twice the average interval, it is marked as an abnormal interval point.

[0126] The drop interval characteristics are arranged in chronological order to form an interval sequence. Periodic analysis is performed on this sequence to identify whether the intervals exhibit a periodic variation pattern. Specifically, the autocorrelation coefficient of the interval sequence is calculated; a coefficient value close to 1 indicates a strong periodicity in the intervals, which is used for autocorrelation analysis. A Fast Fourier Transform (FFT) is used to convert the time-domain interval sequence into a frequency-domain drop feature spectrum for spectral analysis. The horizontal axis of the drop feature spectrum represents frequency, and the vertical axis represents spectral amplitude. The peak frequency corresponds to the main periodic component of the interval. For example, the drop feature spectrum at a transmission interruption point shows a main peak at 0.05 Hz, corresponding to a main period of 20 seconds. The characteristic frequencies and corresponding amplitudes in the drop feature spectrum are recorded to form a spectral feature vector.

[0127] Based on the drop characteristic spectrum and the original interval sequence, the following non-uniformity parameters are extracted: Spectral distribution concentration: calculated based on the ratio of the amplitude of the main peak to the sum of the amplitudes of all frequency points, reflecting the strength of the interval regularity;

[0128] Interval nonuniformity: calculated based on the ratio of the interval standard deviation to the interval mean; mean interval rate of change: reflects the severity of interval fluctuations. For example, if the interval standard deviation of a transmission interruption point is 5 seconds and the interval mean is 20 seconds, then the interval nonuniformity is 0.25; its spectral concentration is 0.6, and the mean interval rate of change is 0.3. A nonuniformity parameter table is constructed, including fields such as transmission interruption point identifier, nonuniformity value, spectral concentration, and mean interval rate of change.

[0129] The non-uniformity parameter table was normalized, scaling the mean values ​​of non-uniformity, spectral concentration, and interval change rate to the range of 0-1. Using the K-means clustering algorithm, all transmission interruption points were divided into three categories based on parameters such as non-uniformity: regular drop segments (non-uniformity < 0.2 and spectral concentration > 0.6); semi-regular drop segments (non-uniformity between 0.2 and 0.4 and spectral concentration between 0.4 and 0.6); and irregular drop segments (non-uniformity > 0.4 and spectral concentration < 0.4).

[0130] In this embodiment, a total of 12 transmission interruption points participate in clustering. Regular drop segments contain 4 interruption points, semi-regular drop segments contain 5 interruption points, and irregular drop segments contain 3 interruption points. Transmission interruption points of the same category are grouped into one drop segment, and all drop segments together constitute an interruption location set. The interruption location set contains 3 segments, and each segment records a list of transmission interruption point identifiers.

[0131] For each power supply path, based on its path convergence at different time windows, a curve showing the change in convergence over time is extracted. The fluctuation characteristics of this curve within the monitoring period (e.g., 24 hours) are analyzed, and the difference between the maximum and minimum convergence values ​​is taken as the fluctuation amplitude. For example, if the convergence of a power supply path fluctuates from 0.2 to 0.6 within 24 hours, then its fluctuation amplitude is 0.4.

[0132] In this embodiment, the fluctuation amplitude threshold is set to 0.3. Power supply paths that simultaneously meet the following two conditions are selected as fault impact channels: the path convergence exceeds a preset convergence threshold (0.2 in this example); and the convergence timing fluctuation amplitude exceeds 0.3. Such paths reflect the dual concentration and dynamic diffusion characteristics of fault impact in both time and space dimensions, and are the most critical fault propagation trunk lines in the system.

[0133] Based on the above analysis, this embodiment identified three fault-affected channels from the 23 power supply paths (8 charger power supply paths + 15 load power supply paths): 1. Charging module power supply channel: involving the power transmission link from the charging module to the DC bus to the battery pack; 2. Battery pack discharge channel: involving the link where the battery pack discharges to the load via the DC bus; 3. Bus load power supply channel: involving the link where the DC bus distributes power to important loads such as control loads, power loads, and communication loads. The identification results of these fault-affected channels provide a precise analytical scope and critical path data foundation for the initial fault risk assessment in subsequent steps.

[0134] In this embodiment, step 103 includes:

[0135] Step 1031: Based on the fault impact channel and the operating status feature library of the substation DC power supply system, determine multiple affected devices and the connection relationships between different affected devices from multiple target devices, so as to construct an affected device connection diagram based on the connection relationships.

[0136] Step 1032: For any affected device in any of the fault-affected channels, use the affected device as the starting point to traverse the connection graph of the affected devices to obtain the fault transmission path corresponding to the affected device.

[0137] Step 1033: For any of the fault propagation paths, obtain the fault impact degree and hierarchical weight coefficient corresponding to each affected device on the fault propagation path, so as to obtain the path impact degree of the fault propagation path based on the fault impact degree and the hierarchical weight coefficient.

[0138] Step 1034: Organize the influence of all the fault propagation paths according to the propagation level to obtain the propagation path table of the substation DC power supply system.

[0139] Step 1035: Obtain the correlation index of each target device in the propagation path table, so as to determine the initial fault risk assessment result of each target device based on the correlation index and the path influence degree.

[0140] Step 1036: Perform time-series feature analysis on the operating status feature library to obtain the time-series correlation between different target devices, and obtain the coupling strength coefficient between different target devices based on the time-series correlation; wherein, extract the operating time-series data of each target device from the operating status feature library, and perform cross-correlation analysis on different target devices based on the operating time-series data to obtain the device time-series correlation between different target devices; obtain the load time-series data of each target device based on the operating time-series data, and obtain the load fluctuation characteristics of each target device based on the device time-series correlation and the load time-series data; construct a load fluctuation map based on all the load fluctuation characteristics, and perform similarity detection on the load fluctuation characteristics of different target devices in the load fluctuation map to generate a device similarity table between different target devices; obtain the correlation stability between different target devices based on the device similarity table, and determine the weak correlation correction factor between different target devices based on the correlation stability; determine the basic similarity between different target devices based on the device similarity table, and correct the basic similarity based on the weak correlation correction factor to obtain the coupling strength coefficient between different target devices.

[0141] In this embodiment, based on the fault impact channels and the operating status feature library determined in the aforementioned steps, all target device nodes and their electrical connections involved in each fault impact channel are extracted. The extracted device nodes include fault source nodes, intermediate transmission nodes, and affected nodes. Taking a charging module power supply channel as an example, the device nodes involved in this channel include: the fault source node is the charging module CHG-01, the intermediate transmission node is DC bus section I, and the affected nodes are battery pack A and control loads, power loads, etc.

[0142] The electrical connections between the aforementioned device nodes are extracted, including series connections, parallel connections, and branch connections. For example, some charging modules are connected in parallel to the same DC bus, with two parallel paths; the bus is connected in series to each load node, with the impedance value accumulating step by step along the series path; the DC bus branches off into multiple branches to supply power to different loads, forming branch connections. Using all extracted device nodes as vertices and the electrical connections between devices as edges, an affected device connection graph is constructed. In this embodiment, the graph contains 23 vertex nodes and 35 connection edges. Each edge is labeled with its connection type and electrical parameters, including impedance values ​​for series connections, current shunting coefficients for parallel connections, and branch direction and load type for branch connections.

[0143] A connectivity check is performed on the constructed affected device connection graph to identify isolated nodes and broken paths, ensuring the integrity of the device connection graph structure and providing an accurate basis for subsequent path traversal. For any affected device in any fault-affected channel, the affected device connection graph is traversed from that device as the starting point. Preferably, this embodiment uses a depth-first search algorithm, starting from the traversal starting point and searching for all paths to other affected device nodes in the channel. To avoid excessive computation due to searching excessively long paths, the maximum depth of the path search is set to 5 layers. For example, taking the charging module CHG-01 as the traversal starting point, the first layer searches to DC bus I segment, the second layer searches to battery pack A, control load, power load, and other nodes, and the third and subsequent layers extend downstream sequentially. The fault propagation paths obtained through the search include, but are not limited to: Charging module CHG-01 → DC bus I → Battery pack A; Charging module CHG-01 → DC bus I → Control load; Charging module CHG-01 → DC bus I → Power load. For nodes connected in parallel, all parallel branches are searched simultaneously, and paths passing through different parallel branches are recorded as independent fault propagation paths. The node sequence and path length (i.e., the number of nodes in the node sequence) of each fault propagation path are recorded. For example, the path Charging module CHG-01 → DC bus I → Battery pack A has a path length of 2, and the node sequence is [Charging module CHG-01, DC bus I, Battery pack A].

[0144] All fault propagation paths obtained from the search are filtered out, removing excessively long paths exceeding 5 hops and reverse paths that do not conform to the direction of power transmission, ensuring the relevance and effectiveness of subsequent analysis. For each filtered fault propagation path, it is divided into multiple propagation levels according to the direction of power transmission: Level 1: Nodes directly connected to the fault source node (or traversal starting point); Level 2: Nodes connected to Level 1 nodes; Level 3: Nodes connected to Level 2 nodes; and so on, forming a multi-level structure.

[0145] For each affected equipment node at each level, the degree of fault impact is quantified. The degree of fault impact is quantified by the magnitude of voltage or current change at that node during the fault period and the normal period. The impact magnitude of each level of node is obtained by comparing the monitored value during the fault period with the normal baseline value. The calculation formula is: Node impact magnitude = (Baseline value - Monitored value) / Baseline value × 100%; where the baseline value is the standard voltage or current value of the node under normal operating conditions, and the monitored value is the real-time value collected during the fault period. For example, if the normal voltage of a bus node is 230V and the monitored voltage during the fault period is 195V, then the impact magnitude of that node is 15.2%. After calculating the impact magnitude of nodes at each level, the nonlinear jump characteristics of the impact between levels are identified. The judgment rule is: when the ratio of the impact magnitude of the second-to-last level node to the impact magnitude of the previous level node deviates from 1.0 by more than 20%, that position is marked as a nonlinear jump point. Subsequently, the formula for calculating the path impact of the fault propagation path is: Path Impact = Σ (Influence Amplitude of Each Level Node × Corresponding Level Weight Coefficient); where the level weight coefficient adopts an exponential decay allocation strategy, reflecting the physical law that the fault impact decreases as the propagation distance increases. The formula for calculating the weight coefficient of the nth level is: W_n = W1 × α^(n-1)

[0146] In this embodiment, the first-level weight coefficient W1 is set to 1.0, and the attenuation factor α is set to 0.8. Therefore: the first-level weight coefficient is 1.0; the second-level weight coefficient is 0.8; the third-level weight coefficient is 0.64; and so on for subsequent levels.

[0147] For example, if the influence of a fault propagation path's first-level node is 15%, the second level is 12%, and the third level is 8%, with corresponding weighting coefficients of 1.0, 0.8, and 0.64 respectively, then its path influence is: 15% × 1.0 + 12% × 0.8 + 8% × 0.64 = 15% + 9.6% + 5.12% = 29.72%. Record the path identifier, path start point, path end point, number of propagation levels, list of nodes at each level, path influence value, and nonlinear transition point location for each fault propagation path. Structure all fault propagation paths and their corresponding path influence values ​​according to propagation levels to construct a propagation path table. The propagation path table is stored in relational data table format, with each row corresponding to a fault propagation path. Column fields include: path identifier (e.g., P1, P2, etc.), path start point device identifier, path end point device identifier, number of propagation levels, list of nodes at each level, path influence (value), and nonlinear transition point location. For example, a record might be described as "Path P1, Start point = Charging module CHG-01, End point = Battery pack A, Number of levels = 2, Node list = [Charging module CHG-01, DC bus I segment, Battery pack A], Influence = 29.72%, Jump point = DC bus I segment".

[0148] In this embodiment, a total of 12 fault propagation path records were generated, forming a complete propagation path table. The table was sorted in descending order of path impact, with paths of higher impact listed first. Paths with an impact exceeding 25% were defined as critical propagation paths. This embodiment identified 4 critical propagation paths, which are the main channels through which the fault impact occurs. Simultaneously, the number of propagation levels for each path was marked; more levels indicate a longer chain of fault impact propagation.

[0149] The propagation path table is used to perform correlation detection, generating an initial fault risk assessment result for each target device. First, the frequency of occurrence of each target device node in different propagation paths is extracted from the propagation path table. Frequency reflects the importance of a node in fault propagation; a higher frequency indicates a higher level of participation in the fault propagation network. For example, the DC bus node appears 11 times in 12 propagation paths, having the highest frequency. Second, the node correlation degree of each target device node is calculated using the formula: Node Correlation Degree = Node Occurrence Frequency / Total Number of Propagation Paths. For example, the node correlation degree of the DC bus node is 11 / 12 ≈ 0.92, indicating a high correlation between this node and fault propagation.

[0150] Furthermore, path correlation reflects the overall importance of a propagation path. The formula for calculating the path correlation of each propagation path is: Path Correlation = Σ (Correlation of each node traversed by the path) / Number of nodes on the path; for example, if a path passes through a charging module (node ​​correlation 0.75), a DC bus (node ​​correlation 0.92), and a battery pack (node ​​correlation 0.58), then its path correlation is (0.75 + 0.92 + 0.58) / 3 = 0.75. Subsequently, the formula for calculating the overall failure risk value of the system is: Overall Failure Risk Value = Σ (Path Influence × Path Correlation); in this embodiment, the calculated overall failure risk value is 68 points, corresponding to a medium risk level. Finally, by combining the node correlation of each target device with the overall failure risk value of the system, an initial failure risk assessment result (i.e., initial failure risk score) is generated for each target device. For example: Charging module CHG-01: 75 points; DC bus section I: 82 points; Battery pack A: 60 points; The initial fault risk score is divided into risk levels according to preset thresholds: above 80 points is high risk, 60-80 points is medium risk, and below 60 points is low risk. This step involves in-depth mining of time-series data in the operating status feature library to analyze the dynamic correlation between devices and calculate the coupling strength coefficient, which characterizes the mutual influence of devices.

[0151] Runtime timing data for each target device is extracted from the operational status feature library. This timing data includes the device's voltage timing, current timing, temperature timing, and internal resistance timing. Taking the charging module CHG-01 as an example, its output voltage timing data over 24 hours is extracted from the operational status feature library, with a sampling interval of 30 seconds, resulting in a timing sequence containing 2880 sampling points. Timing feature analysis is performed on the timing data of each device to identify the data's variation over time. The temporal correlation between the timing data of different devices is analyzed to identify the timing response relationship between devices. For example, when the output current of the charging module increases, the battery terminal voltage rises accordingly, indicating a timing correlation between the two. The timing response relationship between devices is obtained using cross-correlation analysis. For the timing data sequences x(t) and y(t) of devices A and B, their cross-correlation function is defined as: R_xy(τ) = Σ x(t) × y(t+τ); where τ is the time lag, and in this embodiment, the value range is set to -30 minutes to +30 minutes. The peak position τ_max of the cross-correlation function reflects the response delay time of device B relative to device A, and the magnitude of the peak reflects the strength of the response correlation. For example, in the cross-correlation analysis of charging module CHG-01 and battery pack A, the peak appears at τ=2 minutes, indicating that the response of battery pack A lags behind charging module CHG-01 by about 2 minutes, and there is a strong time-series correlation between the two. Cross-correlation analysis is performed on the time-series data of all devices in the operating status feature library. The cross-correlation coefficient reflects the degree of correlation between the time-series data of two devices at different time lags; a high cross-correlation coefficient indicates a strong time-series correlation. Device time-series correlation relationships are established, and the time-series correlation strength and response delay time between each pair of devices are recorded.

[0152] Load fluctuation analysis is performed based on the device timing relationships to construct a load fluctuation map. First, the load timing data of each device is extracted from the device timing relationships. The load timing data reflects the load status of the device at different times: the load of the charging module is its output current; the load of the battery pack is its charging and discharging current; the load of the DC bus is its total output current; and the load of the load device is its operating current.

[0153] Load fluctuation characteristics are analyzed based on the rate of change of load values ​​over time. Load fluctuation rate reflects the speed of load changes. For example, if the output current of a charging module increases from 20A to 35A during a certain period, the load fluctuation rate is 15A / hour; during the same period, the charging current of the battery pack increases from 5A to 18A, and the load fluctuation rate is 13A / hour. The time-series correlation of the devices shows that the load fluctuation rates of the two are close, indicating that the load changes are synchronous.

[0154] The peak times and amplitudes of load fluctuations for each device are identified, revealing a temporal correlation between the load fluctuations of different devices. The load fluctuation curves of different devices are then displayed on the same coordinate system to construct a load fluctuation map. This map uses time as the horizontal axis and load fluctuation rate as the vertical axis, showing the load fluctuation curves of different devices within the same load fluctuation map. For example, the peak times of the fluctuation curves of a charging module and a battery pack in the load fluctuation map are basically the same, with only a time offset of a few minutes, and the curve shapes show similar fluctuation patterns.

[0155] Similarity detection is performed on the load fluctuation curves of each device in the load fluctuation map to obtain a device similarity table. The load fluctuation curves of each device in the load fluctuation map are extracted, and after normalization to eliminate the influence of dimensions, a dynamic time warping algorithm is used to calculate the similarity between the curves of different devices. The dynamic time warping algorithm can identify the alignment relationship of time series on the time axis in the load fluctuation map, and can accurately calculate the similarity even if there is a time offset between two curves.

[0156] For example, the load fluctuation curves of charging module CHG-01 and charging module CHG-02 have a similarity of 0.88, indicating that their load fluctuation patterns are highly similar; the load fluctuation curves of charging module CHG-01 and battery pack A have a similarity of 0.62, indicating that they have a moderate degree of similarity. The similarity calculation results for all device pairs are compiled into a device similarity table. This table is a symmetric matrix with device identifiers as row and column indices, and the matrix elements are the similarity values ​​of the corresponding device pairs. Based on similarity thresholds, device pairs are classified into relevance levels: strongly related device pairs: similarity > 0.7; moderately related device pairs: similarity between 0.4 and 0.7; weakly related device pairs: similarity < 0.4.

[0157] To assess the stability of the association between devices, weak correlation identification is performed based on the device similarity table, and a weak correlation correction factor is obtained. The similarity time-series data (i.e., the sequence of similarity changes over time) for each device pair is extracted from the device similarity table. The time-series change in similarity reflects the evolution of the correlation between devices over time. For example, the similarity of a device pair in the device similarity table is 0.65 in the morning, drops to 0.52 at noon, and rises back to 0.68 in the evening, showing a trend of first decreasing and then increasing. The similarity slope is determined based on the ratio of the change in similarity to the time interval; the similarity slope reflects the speed of correlation change. For example, a negative similarity slope for a device pair from morning to noon indicates a weakening correlation; a positive similarity slope from noon to evening indicates a strengthening correlation. Statistical analysis identifies the distribution characteristics of the similarity slope in the device similarity table; the mean and standard deviation of the slope are used to determine the abrupt change identification threshold. In this embodiment, the mean similarity slope is close to 0 and the standard deviation is 0.008 / hour within the 24-hour monitoring period, reflecting the degree of slope fluctuation. The criterion for determining a slope mutation point is: the similarity slope at a certain moment exceeds the mean plus twice the standard deviation. A device is marked as a slope mutation point when the similarity slope exceeds the mutation threshold at a specific moment. All marked slope mutation points and their mutation times are displayed in a coordinate system to construct a slope mutation map. This map uses time as the horizontal axis and similarity slope as the vertical axis. Mutation points are displayed with special markers in the slope mutation map, intuitively showing the dynamic change process of the correlation between devices. The mutation amplitude parameter of each mutation point is extracted from the slope mutation map. The mutation amplitude is a key parameter for identifying abnormal mutations. For example, if the similarity slope of a mutation point changes abruptly from a low level to a high level in a short period of time, the mutation amplitude is large. The criterion for determining an abnormal mutation point is: the mutation amplitude exceeds 1.5 times the average mutation amplitude. Such mutation points reflect drastic fluctuations in the correlation between devices. In slope mutation maps, the persistence characteristics of mutation points are analyzed, with persistence duration defined as the time difference from the onset of the mutation to its recovery. For example, a mutation point might persist for 45 minutes, during which the similarity slope remains high. Mutation points with longer persistence durations indicate that changes in the correlation between devices are not transient perturbations, but rather persistent changes in association patterns.

[0158] Based on the direction of the mutation, mutation points are divided into two categories: positive mutations, where similarity increases rapidly, usually corresponding to multiple devices responding simultaneously to a certain load change; and negative mutations, where similarity decreases rapidly, usually corresponding to the differentiation of response patterns among devices. All anomalous mutation points are collected to generate a mutation marker set. This set records the labeling information of anomalous mutation points, including the mutation magnitude and duration, serving as the data foundation for subsequent magnitude feature extraction.

[0159] The mutation amplitude values ​​of each mutation marker are extracted from the mutation marker set. The absolute values ​​of the mutation amplitude values ​​are used to eliminate the influence of the mutation direction. For example, the absolute value of the mutation amplitude of one mutation marker is 0.017 / hour, and the absolute value of the mutation amplitude of another mutation marker is 0.021 / hour. Multiple mutation amplitude values ​​of the same device pair in the mutation marker set are statistically analyzed, and these multiple amplitude values ​​form the amplitude sequence of the device pair. For example, the amplitude sequence of a device pair is [0.017, 0.012, 0.019, 0.015], and the sequence length reflects the frequency of anomalous mutations in the device pair. The mean and maximum value of the amplitude sequence are calculated through statistical analysis. For example, the mean amplitude of a device pair is 0.016 / hour, and the maximum amplitude is 0.019 / hour. The mean amplitude reflects the average level of correlation fluctuation of the device pair, and the maximum amplitude reflects the extreme case of correlation fluctuation. Device pairs with larger mean amplitudes have poorer correlation stability and need to be adjusted using a correction factor. An amplitude feature vector is constructed based on the mean amplitude, and this vector includes the mean amplitude dimension.

[0160] The amplitude feature vector is segmented and mapped according to amplitude intervals to obtain the weak correlation correction factor. The mean amplitude parameter is extracted from the amplitude feature vector, and the segmentation criteria divide the mean amplitude into three intervals:

[0161] Low amplitude range: mean amplitude < 0.01 / hour, corresponding to a stable correlation; Medium amplitude range: mean amplitude between 0.01 and 0.02 / hour, corresponding to a moderately volatile correlation; High amplitude range: mean amplitude > 0.02 / hour, corresponding to a highly volatile correlation. The range division is based on historical statistical data, with each range corresponding to a stable, moderately volatile, or highly volatile correlation state, respectively. For example, the mean amplitude of the amplitude feature vector of a certain device pair is 0.016 / hour, belonging to the medium amplitude range. The weak correlation correction factor is determined for different amplitude ranges using a piecewise mapping function: F_corr = 1.0 - 0.1 × k; where k is the interval coefficient: k=1 for low amplitude range, k=3 for medium amplitude range, and k=5 for high amplitude range. Correspondingly: low amplitude range: F_corr = 0.9; medium amplitude range: F_corr = 0.7; high amplitude range: F_corr = 0.5; the smaller the value of the weak correlation correction factor, the worse the stability of the correlation between devices, and the greater the need for correction of the coupling strength.

[0162] The coupling strength coefficient is determined segmentally based on the distribution range of the weak correlation correction factor. First, the distribution characteristics of the weak correlation correction factor for each device pair are statistically analyzed. In this embodiment, the weak correlation correction factor exhibits three concentrated distribution ranges between 0.5 and 0.9, corresponding to low stability, medium stability, and high stability segments, respectively. The original similarity data for each device pair is extracted from the device similarity table as its basic similarity. The original similarity reflects the basic degree of association between devices, while the weak correlation correction factor reflects the stability of the association between devices. The original similarity is segmented and corrected according to the distribution range of the weak correlation correction factor, and the coupling strength coefficient is calculated: Coupling strength coefficient = Basic similarity × Weak correlation correction factor; for example, if the original similarity of a device pair is 0.65 and its weak correlation correction factor is 0.7 (medium amplitude range), then the coupling strength coefficient is 0.65 × 0.7 = 0.455. The coupling levels of equipment pairs are categorized based on their coupling strength coefficients: Strong coupling: coupling strength coefficient > 0.6; Medium coupling: coupling strength coefficient between 0.3 and 0.6; Weak coupling: coupling strength coefficient < 0.3. For example, a pair of equipment pairs with a coupling strength coefficient of 0.46 belongs to the medium coupling level. The coupling strength coefficients of all equipment pairs are then organized into a coupling strength coefficient matrix. This matrix is ​​symmetric, with diagonal elements representing 1.0 (the coupling strength of the equipment itself) and off-diagonal elements representing the coupling strength coefficients between the corresponding equipment pairs. This step completes the initial fault risk assessment results for each target device and the quantitative calculation of the coupling strength coefficients between different target devices. These two results will serve as the core inputs for cascading failure mode identification and dynamic correction of risk scores in subsequent steps.

[0163] In this embodiment, step 104 includes:

[0164] Step 1041: Determine multiple strongly coupled device pairs of the substation DC power supply system based on the coupling strength coefficient.

[0165] Step 1042: Based on the strong coupling device, trace the cascading propagation path of the fault in the substation DC power supply system to identify multiple cascading fault modes corresponding to the substation DC power supply system based on the cascading propagation path.

[0166] Step 1043: For any one of the target devices, obtain the number of times the target device occurs in the multiple cascading failure modes, and use the number of occurrences as the failure participation degree of the target device.

[0167] Step 1044: Correct the initial fault risk assessment result based on the fault participation rate to obtain the corrected fault risk assessment result for each target device.

[0168] This embodiment performs cascading failure mode identification and risk score correction. Based on the acquired inter-device coupling strength coefficients, it aims to identify strongly coupled device pairs, trace the cascading propagation path of faults in the system, and thus identify cascading failure modes. Furthermore, based on the degree of participation of each device in the cascading failure mode, the initial fault risk assessment result is dynamically weighted and corrected to obtain a corrected fault risk assessment result that better reflects the cascading risk of the system. Specifically, based on the coupling strength coefficient matrix generated in the above steps, all device pairs that meet the strong coupling condition are identified. Preferably, the strong coupling threshold is set to 0.6. The off-diagonal elements of the coupling strength coefficient matrix are traversed, and device pairs with a coupling strength coefficient greater than 0.6 are marked as strongly coupled device pairs. For example, in a 220kV substation DC power supply system: the coupling strength coefficient between charging module CHG-01 and DC bus section I is 0.72, which is greater than 0.6, and is marked as a strongly coupled device pair; the coupling strength coefficient between charging module CHG-02 and DC bus section I is 0.68, which is greater than 0.6, and is also marked as a strongly coupled device pair.

[0169] The coupling strength coefficient between DC bus I and battery group A is 0.55, which is less than 0.6, and therefore does not belong to a strongly coupled device pair. All strongly coupled device pairs constitute the basic set of associated edges for subsequent cascade propagation path analysis.

[0170] Next, based on the identified strongly coupled device pairs, the propagation process of the fault along the strong coupling relationship in the system is simulated, tracing the cascading propagation path from any initial faulty device, through the strongly coupled device pairs, to successively affect other devices. The specific tracing method is as follows: Take a certain device as the initial faulty device (e.g., charging module CHG-01); traverse all downstream devices (e.g., DC bus I segment) that form a strongly coupled device pair with this device, and propagate the fault effect to this downstream device; take this downstream device as the new starting point, and continue to traverse other devices that form a strongly coupled device pair with it (e.g., charging module CHG-02, battery pack A, etc.); repeat the above process until no new strongly coupled downstream devices can be found or the preset propagation depth limit is reached.

[0171] Furthermore, record each complete propagation chain from the initial faulty device to the ultimately affected device. Each complete cascading propagation path is identified as a cascading failure mode. A cascading failure mode describes the complete path of a fault propagating through strong coupling relationships in a directed sequence form, and can be supplemented with triggering conditions and hazard assessments. Example: Taking charging module CHG-01 as the initial faulty device, the following cascading failure mode is identified: Charging module CHG-01 fault → DC bus I section voltage drop → Charging module CHG-02 overload → Battery pack A over-discharge; the triggering condition for this cascading failure mode can be defined as:

[0172] The output voltage of charging module CHG-01 is lower than 85% of its rated voltage (i.e., 195V); and the coupling strength coefficient between charging module CHG-01 and DC bus section I is greater than 0.6. Hazard assessment: This embodiment comprehensively considers the number of devices affected by the cascading failure mode and the severity of the impact. For example, the above-mentioned cascading failure mode affects four critical devices (charging module CHG-01, DC bus section I, charging module CHG-02, and battery pack A), and may lead to serious consequences such as DC bus voltage loss, backup power depletion, and power outage of important loads. Therefore, its hazard level is assessed as high-risk.

[0173] The above tracing process is performed on all possible initially faulty devices to obtain a set of all cascading failure modes existing in the system at the current operating stage. This embodiment identifies a total of 8 typical cascading failure modes.

[0174] For each target device, the total number of times it appears in all the identified cascading failure modes is counted, and this number is defined as the failure participation degree of that target device. Failure participation degree quantitatively reflects the activity level and pivotal position of the device in the system's cascading failure risk network. Example: Charging module CHG-01 appears 5 times in 8 cascading failure modes, so its failure participation degree is 5; DC bus I appears 7 times in 8 cascading failure modes, so its failure participation degree is 7; battery pack A appears 3 times in 8 cascading failure modes, so its failure participation degree is 3. To further eliminate the dimensional impact of differences in system size and the total number of cascading failure modes, this embodiment also calculates a cascading failure participation factor for subsequent risk correction: Cascading failure participation factor = Device failure participation degree / Total number of cascading failure modes; for example, the cascading failure participation factor of charging module CHG-01 = 5 / 8 = 0.625.

[0175] Finally, based on the cascading failure participation factor for each target device, the initial failure risk assessment result (initial risk score) generated in the above steps is weighted and corrected to obtain a corrected failure risk assessment result that reflects the potential impact of the device in a cascading failure. This embodiment uses the following weighted correction formula: Corrected Risk Score = Initial Risk Score × (1 + Cascading Failure Participation Factor × Weighting Coefficient); where the weighting coefficient is a configurable parameter, set to 0.5 in this embodiment based on engineering experience. This coefficient is used to adjust the enhancement of the risk score by the cascading failure participation factor. Example: The initial risk score of charging module CHG-01 is 75 points, and its cascading failure participation factor is 0.625. Then: Corrected Risk Score = 75 × (1 + 0.625 × 0.5) = 75 × 1.3125 = 98.4375 points; after normalization, the score is limited to the range of 0-100 points, rounded to 98 points, with a correction of approximately 30.7%. Similarly: DC bus section I: initial risk score 82 points, cascading failure participation factor 0.875, corrected risk score = 82 × (1 + 0.875 × 0.5) = 82 × 1.4375 ≈ 118 points, normalized to 100 points (upper limit); Battery pack A: initial risk score 60 points, cascading failure participation factor 0.375, corrected risk score = 60 × (1 + 0.375 × 0.5) = 60 × 1.1875 = 71.25 points, normalized to 71 points. The corrected risk score significantly increases the priority weight of high cascading risk equipment, effectively overcoming the deficiency of traditional static risk assessment in insufficient consideration of cascading effects.

[0176] Preferably, to support subsequent fault tracing and handling decisions, this embodiment associates and stores the corrected fault risk assessment results, coupling strength coefficient, cascading fault participation factor, and key monitoring parameter list for each target device to construct an extended monitoring parameter library. For example, the record of the high-risk device charging module CHG-01 in the extended monitoring parameter library includes: device identifier: CHG-01; corrected risk score: 98 points; coupling strength coefficient: 0.72 with DC bus I segment, 0.55 with charging module CHG-02; cascading fault participation factor: 0.625; key monitoring parameter list: output voltage, output current, operating temperature. The above parameters will be given a higher monitoring frequency and alarm sensitivity to achieve key monitoring of high-risk, high-cascading-impact devices.

[0177] In this embodiment, step 105 includes:

[0178] Step 1051: For any one of the target devices, after discretizing and classifying the runtime sequence data of the target device, a fault feature code sequence of the target device is formed.

[0179] Step 1052: Perform time-series backtracking analysis on the fault feature code sequence to identify the period of cross-level sudden change, and extract the operation data of the period of cross-level sudden change as fault trigger data.

[0180] Step 1053: Perform fault correlation analysis based on the fault triggering data and the historical fault database of the target device to obtain a fault occurrence probability table of the target device; wherein, the fault occurrence probability table records the fault type and fault occurrence probability corresponding to each fault triggering data.

[0181] Step 1054: Based on the corrected fault risk assessment results of the target equipment, the fault occurrence probability is weighted and corrected to obtain the corrected fault occurrence probability table.

[0182] Step 1055: Extract the historical fault feature code sequence with increasing fault occurrence probability from the corrected fault occurrence probability table as the fault state evolution path of the target equipment.

[0183] This embodiment, based on the runtime sequence data of each target device, constructs a fault state evolution path characterizing the device's own state evolution law through discretization coding, time-series backtracking analysis, fault correlation statistics, and risk weighting correction, providing a quantitative basis for subsequent fault source location and degradation feature extraction. Specifically, for each target device, its monitoring parameter data is extracted from the extended monitoring parameter library. In this embodiment, the monitoring parameters include voltage, current, temperature, internal resistance, and corrected fault risk assessment results. The above continuous monitoring parameters are discretized into levels, converting continuous values ​​into discrete level labels. The classification rules are as follows: Voltage parameters: divided into 5 levels according to the percentage of rated voltage, denoted as L1 to L5; Current parameters: divided into 5 levels according to the percentage of rated current, denoted as L1 to L5; Temperature parameters: divided into 5 levels according to the temperature rise range, denoted as L1 to L5; Internal resistance parameters: divided into 5 levels according to the internal resistance growth rate, denoted as R1 to R5; Risk score: divided into 5 levels according to the score range, denoted as S1 to S5. Taking the CHG-01 charging module as an example, its parameter discretization results at a certain moment are: voltage level L2, current level L3, temperature level L2, internal resistance level R2, and risk score level S3.

[0184] Multiple parameter levels of the target device at the same time are combined and encoded in a predetermined order to form a fault feature code for that time. This embodiment uses the following string encoding format: [Device Identifier] - [Parameter Level Combination] - [Timestamp]. The parameter level combinations are arranged in the order of voltage level + current level + temperature level + internal resistance level + risk level. For example, the fault feature code for charging module CHG-01 at time T1 is: CHG01-L2L3L2R2S3-T1. The fault feature code fully characterizes the overall operating status of the device at a specific time, with different level combinations corresponding to different fault risk modes. For example, level combination L1L4L4R3S4 typically corresponds to an overload fault with severely insufficient power supply; level combination L5L2L4R2S3 typically corresponds to an overvoltage fault with an abnormal charging circuit; and level combination L3L3L2R1S2 typically corresponds to the normal operating status of the device.

[0185] All fault signature codes for each target device in a continuous time series are arranged chronologically to form a fault signature code sequence for that device. For example, the fault signature code sequence for charging module CHG-01 over 6 hours is [CHG01-L3L3L2R1S2-T0, CHG01-L2L3L2R1S2-T1, CHG01-L2L4L3R2S3-T2, CHG01-L1L4L4R3S4-T3]. This sequence shows that the voltage level decreases from L3 to L1, the current level increases from L3 to L4, the temperature level increases from L2 to L4, the internal resistance level increases from R1 to R3, and the risk level increases from S2 to S4, exhibiting a typical trend of accelerated degradation.

[0186] A time-series backtracking analysis is performed on the fault signature sequence of each target device, tracing back from the moment the fault occurred to trace the change trajectory of parameter levels and identify the key state change points that triggered the fault.

[0187] Each frame of the fault signature code sequence is decoded by string segmentation. The parsing rules are as follows: split the signature code string using "-" as the delimiter; extract the intermediate field to obtain the parameter level combination field; and sequentially segment according to a fixed character length (2 characters per segment) to extract the level label of each parameter dimension. For example, the signature code CHG01-L2L3L2R2S3-T1 is parsed to obtain: Voltage level: L2; Current level: L3; Temperature level: L2

[0188] Internal resistance level: R2; Risk level: S3; Perform the above parsing on all feature codes in the sequence to extract the level sequence of each parameter dimension changing over time. Taking the charging module CHG-01 as an example: Voltage level sequence: [L3, L2, L2, L1]; Current level sequence: [L3, L3, L4, L4]; Temperature level sequence: [L2, L2, L3, L4]; Internal resistance level sequence: [R1, R1, R2, R3]; Risk level sequence: [S2, S2, S3, S4].

[0189] The parameter change sequence visually reflects the evolution trajectory of each operating parameter over time: a continuous decrease in voltage level indicates a gradual weakening of power supply capacity; an increase in current level indicates increased load demand or equipment malfunction; an increase in temperature level indicates insufficient heat dissipation or internal faults; an increase in internal resistance level indicates worsening battery degradation; and an increase in risk level indicates an accumulation of overall fault risk. The rate of change of parameter levels in the parameter change sequence reflects the rate of accumulation of fault risk. For example, a voltage jump of two levels (L3→L1) within 2 hours indicates a rapid rate of degradation.

[0190] The time window is set to 2 hours. The parameter change sequence within the time window is extracted by working backward from the time when the equipment failure occurred (i.e., the time corresponding to the feature code representing the failure state in the fault feature code sequence).

[0191] Taking the charging module CHG-01 as an example, it malfunctions at time T3 (signature code CHG01-L1L4L4R3S4-T3). Counting back 2 hours, the time window covers times T1 to T3. Arranging the parameter change sequence within the time window in reverse chronological order generates the following reverse parameter sequences: Voltage level sequence in reverse order: [L1, L2, L2]; Current level sequence in reverse order: [L4, L4, L3]; Temperature level sequence in reverse order: [L4, L3, L2]; Internal resistance level sequence in reverse order: [R3, R2, R1]; Risk level sequence in reverse order: [S4, S3, S2].

[0192] Using backward time as the vertical axis and parameter levels as the horizontal axis, the aforementioned reverse parameter sequence is visualized, constructing a reverse tracing channel. This channel displays the backtracking path of each operating parameter from the moment of the fault to historical moments, revealing the state evolution process before the fault occurred. For example, a reverse tracing channel shows that: the voltage level experienced a decrease from L2 to L1 2 hours before the fault; the current level suddenly increased from L3 to L4 1 hour before the fault; the temperature level increased from L2 to L3 1.5 hours before the fault, and further increased to L4 0.5 hours later; the internal resistance level increased from R1 to R2 1 hour before the fault, and increased to R3 0.5 hours before the fault; the risk level increased from S2 to S3 1.5 hours before the fault, and increased to S4 0.5 hours before the fault.

[0193] In the reverse tracing channel, the time period for a sudden change in parameter level across levels is defined as: the time period during which the parameter level changes across levels between adjacent moments (i.e., the level difference is ≥2). For example: the voltage level suddenly drops from L2 to L1 (a level jump, which is also considered a level jump in this embodiment); the current level suddenly increases from L3 to L4 (a level jump); the temperature level suddenly increases from L2 to L3 (a level jump); the internal resistance level suddenly increases from R1 to R2 (a level jump); the risk level suddenly increases from S2 to S3 (a level jump).

[0194] The reverse tracing channel identifies cross-level abrupt change periods for all parameter dimensions. Taking the CHG-01 charging module as an example, five parameter abrupt change periods were identified: voltage drops from L2 to L1 at a certain moment; current rises from L3 to L4 at a certain moment; temperature rises from L2 to L3 at a certain moment; internal resistance rises from R1 to R2 at a certain moment; and risk rises from S2 to S3 at a certain moment. The pre-abrupt change state is extracted as fault trigger data. The pre-abrupt change state is defined as the parameter state at the moment preceding the parameter abrupt change period, including the level combination of all monitored parameters at that moment. For example, the pre-abrupt change state combination for the CHG-01 charging module is L2L3L2R1S2 (voltage L2, current L3, temperature L2, internal resistance R1, risk S2), and the duration of this state is statistically over 1 hour.

[0195] The fault triggering data of each device is compiled into a structured fault triggering clue record. Each clue includes: State combination: the multi-parameter level combination at the moment before the change; Duration: the stable maintenance duration of the state combination; Change parameter: the specific parameter dimension and change direction of the cross-level change; Change time: the time point when the change occurs. For example, a fault triggering clue is "A voltage drop triggers a fault after state L2L3L2R1S2 lasts for 1.2 hours".

[0196] Based on the fault triggering data extracted in the above steps and the historical fault database of the target device, fault correlation statistical analysis is performed to generate a fault occurrence probability table. From all historical fault cases, the pre-mutation state combinations corresponding to each fault event are extracted, and the total number of occurrences (N_total) of each state combination and the number of times that state ultimately leads to a fault (N_fault) are counted. Then, the fault occurrence probability corresponding to each state combination is calculated based on the ratio of the occurrence count to the total count. For example, if a pre-mutation state L2L3L2R1S2 appears 50 times in the historical database, and 12 of these occurrences ultimately lead to a fault, then its fault occurrence probability is 12 / 50 = 0.24.

[0197] Probability calculations are performed on all state combinations, and they are sorted from highest to lowest probability: High-risk state: probability of failure > 0.5; Medium-risk state: probability of failure between 0.2 and 0.5; Low-risk state: probability of failure < 0.2. This embodiment sets: a state with a probability exceeding 0.5 triggers an early warning; a state with a probability exceeding 0.7 triggers an emergency response.

[0198] The above statistical and calculation results are compiled into a fault occurrence probability table. This table is stored in structured data form and contains the following fields: Status code: a unique code for the combination of multiple parameter levels; Status description: the specific level of each parameter dimension; Occurrence count (N_total); Fault count (N_fault); Fault occurrence probability (P_fault); Risk level (high / medium / low).

[0199] For example, the failure probability table for the CHG-01 charging module contains the following records:

[0200]

[0201] Based on the corrected fault risk assessment results for each target device, the fault occurrence probabilities in its fault occurrence probability table are weighted and corrected to obtain a corrected fault occurrence probability that better reflects the current health status and risk level of the device. This embodiment uses the following weighted correction formula: P_corrected = P_fault × (1 + α × R_norm); where: P_fault is the original fault occurrence probability; R_norm is the normalized value (0-1 range) of the corrected fault risk assessment result, obtained by dividing the corrected risk score by 100;

[0202] α is the risk weighting coefficient, which is set to 0.3 in this embodiment based on engineering experience. Example: The corrected risk score for charging module CHG-01 is 98 points, and the normalized value R_norm = 0.98. For state L2L4L3R2S3, the original fault probability P_fault = 0.52, then: P_corrected = 0.52 × (1 + 0.3 × 0.98) = 0.52 × 1.294 = 0.673; the corrected probability increases to 0.673, and the risk level is upgraded from "medium risk" to "high risk". The above weighted correction is performed on all state combinations to generate a corrected fault occurrence probability table, which will replace the original probability table for subsequent state evolution path analysis.

[0203] From the revised failure probability table, extract the historical failure feature code sequence with increasing failure probability as the failure state evolution path of the target equipment. Traverse all state combinations in the failure probability table to identify state sequences with temporal continuity and monotonically increasing probability values. Such sequences reflect the typical evolution trajectory of the equipment from a low-risk state to a high-risk state, until final failure.

[0204] For example, a typical fault state evolution path for the charging module CHG-01 is: L3L3L2R1S2 (P=0.08) → L2L3L2R1S2 (P=0.24) → L2L4L3R2S3 (P=0.52) → L1L4L4R3S4 (P=0.68). The probability of failure in each state along the path is 0.08, 0.24, 0.52, and 0.68 respectively, showing a clear increasing trend, indicating a gradual accumulation of fault risk. On the identified fault state evolution path, state transition points with a probability jump exceeding 0.2 are marked as critical risk nodes. These nodes are turning points where the equipment state undergoes a qualitative change and the risk increases sharply.

[0205] For example, in the above path, the jump from L2L4L3R2S3 (P=0.52) to L1L4L4R3S4 (P=0.68) is 0.16, which does not reach the 0.2 threshold; while the jump from L2L3L2R1S2 (P=0.24) to L2L4L3R2S3 (P=0.52) is 0.28, which exceeds 0.2. Therefore, L2L4L3R2S3 is marked as a critical risk node. The identified fault state evolution path is output in the form of a state sequence, with the following information appended: Path identifier: unique identifier of the evolution path; Equipment identifier: target equipment; State sequence: state combinations arranged in chronological order; Probability sequence: the corrected probability of fault occurrence for each state; Critical risk node: the location of the marked critical node in the path; Typical fault mode: the fault type corresponding to the path (e.g., overload fault, overvoltage fault, etc.). For example, the fault state evolution path output for charging module CHG-01 is as follows: Path ID: PTH-CHG01-001, Device: CHG-01, State sequence: [L3L3L2R1S2] → [L2L3L2R1S2] → [L2L4L3R2S3] → [L1L4L4R3S4], Probability sequence: [0.08] → [0.24] → [0.52] → [0.68], Key risk node: [L2L4L3R2S3] and Typical fault mode: Charging module overload fault.

[0206] Preferably, based on the above-mentioned fault state evolution path, the change patterns of key parameters in the path are extracted as degradation features to construct a multi-dimensional fault feature table. The degradation features extracted in this embodiment include: single-parameter degradation features: voltage continuous decline rate: determined by the voltage level change gradient along the path; current rise duration: the duration during which the current level rises to a high level without recovering; temperature acceleration point: the critical state where the temperature begins to accelerate in the path; internal resistance growth rate: the step change rate of the internal resistance level along the path; multi-parameter coupled degradation features: voltage drop accompanied by current rise: reflecting excessive load or insufficient power supply capacity; temperature rise accompanied by internal resistance growth: reflecting intensified internal battery degradation; risk score jump accompanied by parameter mutation: reflecting the critical transition of the overall risk level. The multi-dimensional fault feature table is stored in a structured form and includes the following fields: path identifier: associated with the corresponding fault state evolution path; degradation feature type: single-parameter / multi-parameter coupled; feature name: such as "voltage continuous decline rate"; feature value: quantitative indicator (such as 2 levels / hour); hazard level: high / medium / low. This table provides core feature inputs for subsequent multi-source coupled fault localization and key fault source identification.

[0207] In this embodiment, step 106 includes:

[0208] Step 1061: Based on the operating status feature library, the corrected fault risk assessment results of each target device, and the fault status evolution path, perform parameter gradient distribution analysis on multi-source data to identify the set of target devices whose parameter gradient values ​​exceed the gradient threshold, as high gradient clustering areas.

[0209] Step 1062: At the boundary of the high gradient cluster region, identify the region where the parameter gradient value decreases significantly, and generate a gradient reduction marker;

[0210] Step 1063: Based on the gradient descent marker, trace the parameter gradients of adjacent target devices step by step along the direction of parameter gradient descent to obtain the low gradient propagation path;

[0211] Step 1064: Based on the corrected fault risk assessment results, select a set of hidden fault candidate target devices from the low-gradient transmission path;

[0212] Step 1065: Calculate the gradient correlation degree of each hidden fault candidate target device to the high gradient clustering area via the low gradient transmission path;

[0213] Step 1066: Select multiple fault source devices from the set of hidden fault candidate target devices based on the gradient correlation degree.

[0214] This embodiment aims to accurately locate the key fault source device in the system that has a source triggering effect, based on the operating status feature library, the corrected fault risk assessment results of each target device, and the fault state evolution path, through parameter gradient distribution analysis, boundary identification, reverse tracing, and gradient correlation matching of multi-source data. Specifically, based on the degradation characteristics represented by the operating status feature library, the corrected fault risk assessment results of each target device (from step 1044), and the fault state evolution path (from step 1055), parameter gradient distribution analysis of multi-source data is performed to identify the set of target devices whose parameter gradient values ​​exceed the gradient threshold, as high gradient clustering areas.

[0215] In this embodiment, the parameter gradient reflects the rate of change of parameter values ​​in space or time, and the high gradient region corresponds to the location where the parameter changes rapidly. The parameter gradient includes the following three categories: (1) Spatial gradient of risk score: determined by the difference between the corrected fault risk assessment results between adjacent target devices. For example, the risk gradient between charging module CHG-01 and adjacent charging module CHG-02 is calculated as: |risk score_CHG-01 - risk score_CHG-02| = |98 - 72| = 26, which is a large gradient value; (2) Temporal gradient of operating parameters: extracted from the extended monitoring parameter library, the time series data of parameters such as voltage and current are extracted and determined by the difference between parameter values ​​at adjacent times. For example, the voltage of DC bus I section continues to decrease over a period of time, and its voltage time gradient is calculated as: |voltage_t2 - voltage_t1| / Δt, which is a large value; (3) Intensity gradient of degradation features: extracted from the fault state evolution path, degradation features (such as voltage drop rate, current rise rate, etc.) are extracted and determined by the rate of change of degradation rate. For example, if the voltage drop rate of a device accelerates, its degradation intensity gradient increases significantly.

[0216] In this embodiment, the gradient threshold is determined based on the mean and standard deviation of historical statistical data. For each type of gradient parameter, the threshold is set as: Gradient threshold = Historical mean + 2 × Historical standard deviation. Gradients are calculated for all target devices in the system across all dimensions, and the gradient values ​​are normalized and fused to obtain a comprehensive gradient value. Target devices whose comprehensive gradient values ​​exceed the gradient threshold are marked as high-gradient devices. All adjacent high-gradient devices are aggregated into connected components to form a high-gradient cluster region. The high-gradient cluster region marks the core area affected by the fault, where the fault coupling strength between devices is highest.

[0217] Example: This embodiment identifies a high gradient cluster region containing the following three device nodes: charging module CHG-01 (risk gradient 26, voltage-time gradient 0.5V / 10min, degradation intensity gradient 0.3 level / hour); DC bus section I (risk gradient 18, voltage-time gradient 0.8V / 10min, degradation intensity gradient 0.4 level / hour); and battery pack A (risk gradient 15, voltage-time gradient 0.3V / 10min, degradation intensity gradient 0.2 level / hour). The gradient values ​​of the above devices are significantly higher than those of other devices in the system, constituting a high gradient cluster region G1.

[0218] At the boundary of the high-gradient clustering area, regions where the parameter gradient values ​​significantly decrease are identified, and gradient reduction markers are generated. The boundary of the high-gradient clustering area is defined as the intersection between devices within the clustering area and devices outside the clustering area. At this intersection, the gradient value decreases significantly, reflecting the attenuation boundary as the fault impact propagates from the core area to the periphery. For each boundary device in the high-gradient clustering area, the gradient reduction magnitude between it and adjacent devices outside the clustering area is calculated: Gradient reduction magnitude = Boundary device's comprehensive gradient value - Adjacent device's comprehensive gradient value; In this embodiment, the gradient reduction threshold is set to 0.5 (normalized). When the gradient reduction magnitude exceeds this threshold, the boundary location is marked as a gradient reduction region, and a corresponding gradient reduction marker is generated. The gradient reduction marker records the following information: Boundary location: device identifier within the clustering area and device identifier outside the clustering area; Gradient reduction magnitude: specific value; Boundary type: "Charging module-charging module boundary", "bus-load boundary", etc.; Example: The boundary of the high-gradient clustering area G1 is located between charging module CHG-01 and charging module CHG-03. The calculated gradient values ​​are: CHG-01: 0.85 (normalized); CHG-03: 0.32 (normalized); gradient descent magnitude: 0.85 - 0.32 = 0.53 > 0.5; a gradient descent marker is generated: "Descent marker LM01: Boundary position = between CHG-01 and CHG-03, gradient descent magnitude = 0.53, boundary type = charging module - charging module".

[0219] In this embodiment, three gradient reduction markers were identified in the high gradient aggregation region G1, located at the boundaries with the charging module CHG-03, the load circuit 2, and the spare bus SP-BUS, respectively. The distribution characteristics of these reduction markers reflect the location of the fault propagation blockage and the attenuation boundary contour.

[0220] Based on the gradient descent markers, the parameter gradients of adjacent target devices are traced step by step along the direction of gradient descent to obtain the low gradient propagation path. In this embodiment, the gradient direction is defined as the direction in which the gradient value increases. Therefore, tracing in the opposite direction of the gradient direction (i.e., the gradient decreasing direction) can find the source of the gradient value. Starting from each gradient descent marker position, with the device outside the cluster area in the marker as the starting point of tracing, the adjacent devices are visited step by step along the gradient decreasing direction: the comprehensive gradient value of the current device is recorded; adjacent devices that have an electrical connection with the current device and are located further away from the high gradient cluster area are found; the comprehensive gradient values ​​of adjacent devices are compared, and the device with the lower gradient value is selected as the next hop; the above steps are repeated until no adjacent device with a lower gradient value can be found (i.e., a local gradient minimum is reached) or the tracing depth exceeds a preset threshold (set to 5 layers in this embodiment).

[0221] The device nodes traversed during the tracing process and their corresponding comprehensive gradient values ​​are recorded in the access order to form a low-gradient propagation path. Example: Starting from the gradient reduction marker LM01 (boundary between CHG-01 and CHG-03), with charging module CHG-03 as the tracing starting point, a step-by-step tracing is performed: 1st hop: CHG-03, comprehensive gradient value 0.32; 2nd hop: tracing to the input circuit breaker CB-03 of CHG-03, comprehensive gradient value 0.18; 3rd hop: tracing to the upstream AC power supply AC-IN of CB-03, comprehensive gradient value 0.05; 4th hop: no adjacent device with a lower gradient value can be found, tracing terminates; resulting in the low-gradient propagation path: Low-gradient propagation path LTP01: AC power supply AC-IN (0.05) → input circuit breaker CB-03 (0.18) → charging module CHG-03 (0.32). Although the devices on this path have a low risk gradient, they may be hidden triggers for faults. This embodiment identifies a total of 5 low gradient propagation paths, each corresponding to a different gradient descent marker.

[0222] Based on the corrected fault risk assessment results, a set of hidden fault candidate target devices is selected from the low-gradient propagation path. Hidden fault candidate target devices are defined as: target devices located on the low-gradient propagation path that exhibit abnormal symptoms but have relatively low corrected fault risk assessment results. These devices have the following characteristics: low gradient value: located in a low-gradient region, making them less likely to be detected by traditional threshold alarm methods; low risk score: corrected fault risk assessment results are below the system average level or high-risk threshold; abnormal symptoms: deterioration features (such as abnormal internal resistance growth rate, excessive voltage dispersion, abnormal temperature fluctuations, etc.) can be extracted from the operating state feature library or fault state evolution path.

[0223] This embodiment uses the following filtering rules: traverse all device nodes on each low gradient propagation path; for each device, obtain its fault correction risk assessment result (risk score); if the device's risk score is lower than the system average risk score (65 points in this embodiment) and lower than the high risk threshold (80 points).

[0224] Simultaneously, the device exhibits at least one early degradation characteristic in the operational status feature library (e.g., internal resistance growth rate > 30%, voltage dispersion > 0.1V, number of abnormal batteries > 10%); devices meeting these conditions are marked as candidate target devices for hidden faults. Example: The low gradient propagation path LTP01 contains three devices:

[0225] AC power supply AC-IN: Risk score 35, no monitored degradation characteristics → Not met; Input circuit breaker CB-03: Risk score 42, exhibits abnormally increased contact resistance degradation characteristics → Hidden fault candidate; Charging module CHG-03: Risk score 58, exhibits output voltage fluctuation degradation characteristics → Hidden fault candidate; All selected hidden fault candidate target devices are gathered to form a hidden fault candidate target device set. In this embodiment, a total of 4 hidden fault candidate target devices were selected, including: input circuit breaker CB-03, charging module CHG-03, temperature sensor TS-07, and a single cell CELL-B23 of battery pack B.

[0226] Calculate the gradient correlation degree of each of the concealed fault candidate devices to the high gradient clustering area via the low gradient transmission path. The gradient correlation degree quantitatively reflects the degree of influence of the concealed fault candidate device on the high gradient clustering area. The higher the correlation degree, the greater the contribution of the concealed fault candidate to the fault state of the high gradient clustering area, and the more likely it is a potential key fault source. In this embodiment, the gradient correlation degree is determined by the cumulative gradient change on the gradient transmission path. The calculation formula is as follows: Gradient correlation degree = Σ (gradient descent amplitude of each hop on the path) × path length attenuation factor; where: gradient descent amplitude: the difference in the comprehensive gradient value between two adjacent hop devices; path length attenuation factor: reflects the law of attenuation of influence as the transmission distance increases, and the calculation formula is λ^(k-1), where λ is the attenuation coefficient (0.8 in this embodiment), and k is the hop number. Example: Calculate the gradient correlation degree of concealed fault candidate CB-03 to the high gradient clustering area G1. Low gradient propagation path LTP01 (reverse tracing direction): AC-IN (0.05) → CB-03 (0.18) → CHG-03 (0.32) → G1 boundary; gradient descent magnitude (along the positive influence direction): CB-03 → CHG-03: 0.32 - 0.18 = 0.14 and CHG-03 → G1 boundary: 0.85 - 0.32 = 0.53; path length decay factor:

[0227] First hop (CB-03→CHG-03): λ^(0) = 1.0; Second hop (CHG-03→G1 boundary): λ^(1) = 0.8; Gradient correlation = 0.14 × 1.0 + 0.53 × 0.8 = 0.14 + 0.424 = 0.564. Similarly, the gradient correlation of another hidden fault candidate TS-07 is calculated. Its gradient accumulation on the path is small, and the correlation is 0.21.

[0228] Based on the gradient correlation, multiple fault source devices are selected from the set of hidden fault candidate target devices. In this embodiment, the gradient correlation threshold is set to 0.5. Hidden fault candidate target devices with a gradient correlation exceeding 0.5 are marked as suspected critical fault sources. For each suspected critical fault source, its degradation characteristics in the operating status feature library are verified. The verification rules are as follows: if a suspected critical fault source has two or more independent degradation characteristics, and the degradation degree shows a continuous deterioration trend, it is confirmed as a critical fault source;

[0229] If the degradation characteristic is singular or shows no obvious deterioration trend, it is marked as a device to be observed and is not included in the fault source device set. Hidden fault candidate CB-03: Gradient correlation degree 0.564 > 0.5, marked as a suspected critical fault source. Verification showed that this device exhibits two degradation characteristics: abnormally increased contact resistance (current value is 45% higher than the initial value) and abnormal contact temperature (15°C higher than the ambient temperature), which have shown a continuous deterioration trend over the past 3 months. Confirmed as a critical fault source. Hidden fault candidate CHG-03: Gradient correlation degree 0.432 < 0.5, not reaching the threshold, not included in the fault source device set. Hidden fault candidate TS-07: Gradient correlation degree 0.21 < 0.5, not reaching the threshold, not included in the fault source device set. Hidden fault candidate CELL-B23: Gradient correlation degree 0.58 > 0.5, marked as a suspected critical fault source. Verification revealed that the single battery cell exhibited two degradation characteristics: excessive voltage dispersion (0.15V) and abnormal internal resistance growth rate (42%), with capacity degradation reaching 8%. It was confirmed as a critical fault source. This embodiment ultimately identified two critical fault source devices: Input circuit breaker CB-03: located upstream of charging module CHG-03, with a low risk score (42 points) but high gradient correlation (0.564), exhibiting abnormally increased contact resistance and abnormal contact temperature degradation; Battery cell CELL-B23: located in battery pack B, with a medium-low risk score (58 points) but high gradient correlation (0.58), exhibiting excessive voltage dispersion and abnormal internal resistance growth rate degradation. These critical fault sources share the following characteristics: low risk score: underestimated in traditional scoring systems; high gradient correlation: significantly impacting high gradient clustering areas.

[0230] Significant deterioration characteristics: Early signs of deterioration are detectable.

[0231] In this embodiment, step 107 includes: determining a handling priority coefficient based on the critical fault source. The handling priority coefficient quantifies the urgency and effectiveness of handling different critical fault sources; a higher priority coefficient indicates that the fault source should be handled first. The calculation of the handling priority coefficient comprehensively considers three factors: the impact range of the critical fault source, the probability of fault occurrence, and the difficulty of handling. The impact range is determined by the number of devices associated with the critical fault source; a critical fault source affects multiple device nodes. The probability of fault occurrence is extracted from the probability table; a critical fault source has a high probability of triggering a fault. The difficulty of handling is determined by the maintenance complexity and the impact range of the power outage; handling a critical fault source requires a certain amount of time and specialized tools. The handling priority coefficient P_priority = w1 × N_affected + w2 × P_fault + w3 × (1 / D_repair), where w1 is the impact range weight, N_affected is the number of affected devices, w2 is the probability weight, P_fault is the fault probability, w3 is the difficulty weight, and D_repair is the handling difficulty. Critical fault sources with high priority coefficients should be given priority in handling resources.

[0232] In some embodiments, establishing a fault handling sequence using the handling priority coefficient includes: performing a hierarchical mapping transformation on the handling priority coefficient to generate a top-level mapping set; identifying nodes with excessive risk weight deviation from the top-level mapping set to generate deviation identifiers; extracting suppressed risk features based on the deviation identifiers to obtain a risk compensation list; and reordering the risk compensation list and the top-level mapping set according to risk accumulation to form a fault handling sequence.

[0233] The priority coefficients are transformed into a hierarchical mapping top-level set. This hierarchical mapping converts continuous priority coefficients into discrete priority levels, facilitating resource allocation decisions. Priority levels are divided into high, medium, and low priorities. High priority corresponds to a coefficient greater than 0.7, medium priority to a coefficient between 0.5 and 0.7, and low priority to a coefficient less than 0.5. For example, a critical fault source—the input circuit breaker—has a priority coefficient of 0.85, mapped to a high priority level. A critical fault source—the charging module—has a priority coefficient of 0.72, also mapped to a high priority level. A critical fault source—the temperature sensor—has a priority coefficient of 0.58, mapped to a medium priority level. Fault sources within the same priority level are sorted according to their original priority coefficient values. The top-level mapping set is extracted, containing all high-priority critical fault sources. A particular top-level mapping set contains both the input circuit breaker and the charging module. Fault sources in the top-level mapping set have the highest urgency and should be addressed immediately.

[0234] The system identifies nodes with excessive risk weight deviation from the top-level mapping and generates deviation markers. Risk weight reflects the contribution of a fault source to the overall fault risk, calculated as the ratio of the fault source's risk score to the sum of all fault source risk scores. A certain input circuit breaker has a low risk score, a high sum of all fault source risk scores, and therefore a low risk weight. A certain charging module has a high risk score and a high risk weight. Risk weight deviation measures the degree of matching between the risk weight and the handling priority coefficient, determined by the absolute value of the difference between the risk weight and the normalized priority coefficient. The normalized priority coefficient is determined by normalizing the handling priority coefficients of each fault source to a sum of 1. A certain input circuit breaker has a high normalized handling priority coefficient, a low risk weight, and a large deviation. A large deviation indicates a mismatch between the risk weight and the handling priority, potentially suggesting that the risk is underestimated or overestimated. Nodes with excessive deviation are identified, and deviation markers are generated for fault sources with deviations exceeding a threshold. For example, an input circuit breaker's deviation exceeds the threshold, thus generating a deviation marker. The risk weight, handling priority coefficient, and deviation degree value of the deviation identifier record node are generated based on the top-level mapping set.

[0235] A risk compensation list is obtained by extracting suppressed risk features based on deviation markers. Nodes with large deviations typically correspond to hidden fault sources with low risk scores but wide impact ranges; their risk features are suppressed in traditional scoring systems. Nodes with underestimated risks are identified from the deviation markers. For example, an input circuit breaker has a low risk score, but as a power input link, it affects all downstream devices, and its risk impact is underestimated by traditional scoring. Suppressed risk features are extracted; these are risk factors that the deviation marker node possesses but are not fully reflected in the risk score. The suppressed risk features of a certain input circuit breaker include critical path location and single-point failure risk. Critical path location indicates that the device is located on the inevitable path of fault propagation, and single-point failure risk indicates that a failure of this device will lead to a complete power outage. Risk compensation values ​​quantify the correction amount of suppressed risk features to the risk score; a certain input circuit breaker has a high risk compensation value. A risk compensation list is constructed, which organizes the suppressed risk features and compensation values ​​of all deviation marker nodes. A risk compensation list contains multiple nodes and their corresponding compensation values. The risk compensation mechanism ensures that hidden but critical fault sources are not ignored due to low risk scores. The compensated risk score more accurately reflects the true risk level of the fault source.

[0236] The risk compensation list and the top-level mapping set are reordered into a fault handling sequence based on risk accumulation. Risk accumulation combines the original risk score and the risk compensation value, with risk accumulation R_cumulative = R_original + R_compensation, where R_original is the original risk score and R_compensation is the risk compensation value. A certain input circuit breaker has a high risk accumulation, as does a certain charging module. Fault sources in the top-level mapping set are sorted in descending order of risk accumulation, with higher accumulation sources appearing earlier in the handling sequence. Even if a node in the risk compensation list has a low original risk score, it may still appear at the top of the handling sequence after compensation. A possible fault handling sequence is [charging module, input circuit breaker, DC bus maintenance, battery equalization charging]. The fault handling sequence considers the handling dependencies between fault sources. Although an input circuit breaker has a slightly lower risk accumulation, its handling can block the root cause of the charging module fault, therefore it is adjusted to be handled before the charging module. The final fault handling sequence is [input circuit breaker, charging module, DC bus maintenance, battery equalization charging], and the sequence length reflects the complexity of the fault handling.

[0237] Based on the fault handling sequence, a tiered handling plan is output. The tiered handling plan divides the fault handling sequence into three levels: emergency handling, planned handling, and preventative handling. Emergency handling targets the highest priority fault sources with high risk accumulation in the fault handling sequence. A typical emergency handling plan includes two tasks: replacing the input circuit breaker and overhauling the charging module's output rectifier module. Emergency handling requires completion within a short timeframe and necessitates the deployment of emergency repair teams and spare parts. Planned handling targets medium priority fault sources with moderate risk accumulation in the fault handling sequence. A typical planned handling plan includes two tasks: tightening DC bus joints and insulation testing. Planned handling is scheduled for execution within the planned power outage window. Preventative handling targets lower priority fault sources with lower risk accumulation in the fault handling sequence. A typical preventative handling plan includes two tasks: equalizing battery charging and calibrating temperature sensors. Preventative handling is scheduled for execution within the routine maintenance cycle and does not occupy the power outage time window. The tiered handling plan clearly defines the handling time limits, resource requirements, and responsible personnel for each level. A typical emergency tiered handling plan requires the team leader to arrive at the site promptly upon receiving instructions. The handling plan includes detailed operating procedures and safety measures. For example, a replacement plan for a specific input circuit breaker includes five steps: power outage, circuit breaker removal, new circuit breaker installation, electrical testing, and power restoration. The tiered handling plan forms a complete handling guidance document, covering all handling tasks at the emergency, planned, and preventative levels. Each task is accompanied by time limits, resource lists, operating procedures, and safety measures, providing a standardized action guide for handling DC power supply faults in substations.

[0238] On the other hand, refer to Figure 2This embodiment also discloses a substation DC power supply fault monitoring and processing system, including an operation characteristic module 201, a fault channel module 202, a coupling analysis module 203, a risk assessment module 204, a fault evolution module 205, a fault location module 206, and a fault processing module 207.

[0239] The operation feature module 201 is used to acquire the operation sequence data and physical equipment information of each target device in the DC power supply system of the substation, and to identify the early deterioration feature data of each target device based on the operation sequence data, so as to construct an operation status feature library.

[0240] The fault channel module 202 is used to obtain the topological connection relationship of the substation DC power supply system according to the operating status feature library, so as to determine the fault impact channel of the substation DC power supply system according to the topological connection relationship and the operating status feature library.

[0241] The coupling analysis module 203 is used to determine the initial fault risk assessment result of each target device according to the fault impact channel, and to obtain the coupling strength coefficient between different target devices according to the operating status feature library.

[0242] The risk assessment module 204 is used to determine the cascading failure modes between different target devices based on the coupling strength coefficient, and to correct the initial failure risk assessment result based on the cascading failure modes to obtain the corrected failure risk assessment result for each target device.

[0243] The fault evolution module 205 is used to perform time-series backtracking analysis on each target device according to the operating status feature library, determine the fault occurrence probability table of each target device, and revise the fault occurrence probability table according to the revised fault risk assessment result to determine the fault state evolution path of each target device.

[0244] The fault location module 206 is used to determine several fault source devices from multiple target devices based on the operating status feature library and the modified fault risk assessment results, coupling strength coefficient and fault state evolution path of each target device.

[0245] The fault handling module 207 is used to determine the priority of each fault source device, to determine a fault handling sequence of several fault source devices according to the priority, and to complete the fault handling of the substation DC power supply system based on the fault handling sequence.

[0246] This embodiment discloses a method and system for monitoring and handling DC power supply faults in substations. By acquiring runtime sequence data and physical equipment information, it identifies early degradation characteristics, constructs an operational status feature library, and systematically integrates scattered monitoring information, solving the problem of insufficient integration of massive amounts of data and providing a comprehensive foundation for subsequent analysis. Based on the operational status feature library, it obtains topological connection relationships and determines fault impact channels. It uses physical connections of equipment to identify fault propagation paths, solving the problem of being unable to extract correlation patterns between equipment. Based on the fault impact channels, it determines the initial fault risk assessment result. Simultaneously, it obtains coupling strength coefficients from the operational status feature library to initially quantify risks and capture inter-equipment dependencies, laying the foundation for corrective assessment. Based on the coupling strength coefficients, it determines cascading fault modes and corrects the initial fault risk assessment result accordingly, considering the cascading effect of faults and improving the accuracy of risk assessment. Based on the operational status feature library, it performs time-series backtracking analysis to determine a fault occurrence probability table. This table is then corrected in conjunction with the corrected fault risk assessment result to determine the fault state evolution path. It backtracks historical data to predict fault development, solving the problem of being unable to trace the fault formation process. By comprehensively analyzing the operational status feature database, corrective assessments, coupling coefficients, and evolution paths, the fault source equipment was identified, the source was precisely located, and the fault source and affected objects were distinguished. Finally, the priority of the fault source was determined, a fault handling sequence was formulated, fault handling was completed, resource allocation was optimized, and precise fault handling was achieved.

[0247] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that 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 for those skilled in the art.

Claims

1. A substation DC power supply fault monitoring and handling method, characterized by, include: The system acquires the runtime sequence data and physical equipment information of each target device in the DC power supply system of the substation, and identifies the early degradation characteristic data of each target device based on the runtime sequence data to construct an operating status characteristic library. The topological connection relationship of the substation DC power supply system is obtained according to the operating status feature database, and the fault impact channel of the substation DC power supply system is determined according to the topological connection relationship and the operating status feature database. The initial fault risk assessment result of each target device is determined based on the fault impact channel, and the coupling strength coefficient between different target devices is obtained based on the operating status feature library; Determining the cascading failure modes among different target devices based on the coupling strength coefficient includes: identifying multiple strongly coupled device pairs in the substation DC power supply system based on the coupling strength coefficient; tracing the cascading propagation path of the fault in the substation DC power supply system based on the strongly coupled device pairs, to identify multiple cascading failure modes corresponding to the substation DC power supply system based on the cascading propagation path; for any target device, obtaining the number of times the target device appears in the multiple cascading failure modes, and using the number of appearances as the fault participation degree of the target device; and correcting the initial fault risk assessment result based on the fault participation degree to obtain a corrected fault risk assessment result for each target device. Based on the operating status feature library, a time-series backtracking analysis is performed on each target device to determine the failure probability table for each target device. The failure probability table is then corrected based on the corrected failure risk assessment results to determine the failure state evolution path for each target device. Based on the operating status feature library and the corrected fault risk assessment results, coupling strength coefficient and fault state evolution path of each target device, several fault source devices are determined from the multiple target devices; The priority of each of the fault source devices is determined, and a fault handling sequence of several of the fault source devices is determined according to the priority, and the fault handling of the substation DC power supply system is completed based on the fault handling sequence.

2. The method for monitoring and handling DC power supply faults in a substation according to claim 1, characterized in that, The process of acquiring runtime sequence data and physical device information for each target device in the substation DC power supply system, and identifying early degradation characteristic data for each target device based on the runtime sequence data to construct an operational status characteristic library, includes: For any target device in the DC power supply system of the substation: The real-time internal resistance, real-time discharge current, initial internal resistance, and real-time voltage and temperature of each individual battery cell in the target device are obtained. The internal resistance growth rate of the target device is obtained based on the real-time internal resistance and the initial internal resistance. The discharge capacity of the target device is obtained based on the real-time discharge current. The average voltage of the target device is determined based on the real-time voltage of all the individual cells in the target device, so as to obtain the voltage dispersion of each individual cell based on the average voltage and the real-time voltage; For any single battery cell, if the real-time temperature of the single battery cell is higher than a preset temperature threshold, the single battery cell is determined to be an abnormal battery. The number of abnormal batteries is obtained, and the internal resistance growth rate, discharge capacity and voltage dispersion are marked according to the preset degradation degree threshold to obtain the early degradation characteristic data of the target device. The real-time internal resistance, real-time discharge current, real-time voltage, and real-time temperature are collected according to a preset sampling frequency to obtain the runtime sequence data of each target device. Obtain the device identifier, physical location, and port physical connection relationship of each target device to obtain the physical device information of each target device; Based on the device identifier, the operating sequence data, physical device information and early degradation characteristic data corresponding to each target device are associated to obtain the operating status characteristic library of the substation DC power supply system.

3. The method for monitoring and handling DC power supply faults in a substation according to claim 2, characterized in that, The step of obtaining the topology connection relationship of the substation DC power supply system according to the operating status feature database, and determining the fault impact channel of the substation DC power supply system according to the topology connection relationship and the operating status feature database, includes: The topology of the DC power supply system of the substation is obtained based on the physical connection relationship of the ports, and the topology is decomposed into multiple charger power supply paths and multiple load power supply paths; wherein, the charger power supply path is the path from the charging module to the bus to the battery; and the load power supply path is the path from the bus to the load. For any one of the multiple charger power supply paths and multiple load power supply paths, the voltage transmission sequence between different target devices on that path is obtained based on the runtime sequence data, so as to determine the location of abnormal voltage drop according to the voltage transmission sequence. Based on the location of the abnormal voltage drop, multiple transmission interruption devices on the path are identified. A time-series fluctuation analysis is performed on the path based on these multiple transmission interruption devices to obtain the path convergence. Specifically, for any target device on the path, voltage time-series data of the target device in the operating status feature library is obtained. When a voltage in the voltage time-series data is lower than a preset voltage threshold, the target device is determined to be a transmission interruption device. The total number of target devices and the total number of transmission interruption devices on the path are obtained, and the ratio of the number of transmission interruption devices to the number of target devices is used as the path convergence. Based on the path convergence, multiple fault-affected channels of the substation DC power supply system are determined from multiple charger power supply paths and multiple load power supply paths.

4. The method for monitoring and handling DC power supply faults in a substation according to claim 3, characterized in that, The step of determining the initial fault risk assessment result for each target device based on the fault impact channel, and obtaining the coupling strength coefficient between different target devices based on the operating status feature library, includes: Based on the fault impact channel and the operating status feature library of the substation DC power supply system, multiple affected devices and the connection relationships between different affected devices are determined from multiple target devices, so as to construct an affected device connection diagram based on the connection relationships; For any affected device in any of the fault-affected channels, the affected device is used as the starting point for traversal of the connection graph of the affected devices to obtain the fault propagation path corresponding to the affected device. For any fault propagation path, obtain the fault impact degree and hierarchical weight coefficient corresponding to each affected device on the fault propagation path, so as to obtain the path impact degree of the fault propagation path based on the fault impact degree and the hierarchical weight coefficient. The influence of all the fault propagation paths is organized according to the propagation level to obtain the propagation path table of the substation DC power supply system. Obtain the correlation index of each target device in the propagation path table, and determine the initial fault risk assessment result of each target device based on the correlation index and the path influence degree; The operating status feature library is subjected to time-series feature analysis to obtain the time-series correlation between different target devices, and the coupling strength coefficient between different target devices is obtained based on the time-series correlation.

5. The method for monitoring and handling DC power supply faults in a substation according to claim 4, characterized in that, The step of performing time-series feature analysis on the operating state feature library to obtain the time-series correlation relationship between different target devices, and obtaining the coupling strength coefficient between different target devices based on the time-series correlation relationship, includes: Runtime sequence data of each target device is extracted from the running status feature library, and cross-correlation analysis is performed on different target devices based on the runtime sequence data to obtain the device timing correlation relationship between different target devices; The load timing data of each target device is obtained based on the runtime timing data, so as to obtain the load fluctuation characteristics of each target device according to the device timing correlation and the load timing data. A load fluctuation map is constructed based on all the load fluctuation features, and a similarity detection is performed on the load fluctuation features of different target devices in the load fluctuation map to generate a device similarity table between different target devices. The correlation stability between different target devices is obtained based on the device similarity table, and a weak correlation correction factor between different target devices is determined based on the correlation stability. The basic similarity between different target devices is determined according to the device similarity table, and the basic similarity is corrected according to the weak correlation correction factor to obtain the coupling strength coefficient between different target devices.

6. The method for monitoring and handling DC power supply faults in a substation according to claim 1, characterized in that, The step of performing time-series backtracking analysis on each target device based on the operational status feature library to determine a fault occurrence probability table for each target device, and then revising the fault occurrence probability table based on the revised fault risk assessment results to determine the fault state evolution path for each target device, includes: For any one of the target devices, after discretizing and classifying the runtime sequence data of the target device, a fault feature code sequence of the target device is formed; The fault signature sequence is subjected to time-series backtracking analysis to identify the period of cross-level sudden change, and the operation data of the period of cross-level sudden change is extracted as fault triggering data. Based on the fault triggering data and the historical fault database of the target device, a fault correlation analysis is performed to obtain a fault occurrence probability table for the target device; wherein, the fault occurrence probability table records the fault type and fault occurrence probability corresponding to each fault triggering data; Based on the corrected fault risk assessment results of the target equipment, the fault occurrence probability is weighted and corrected to obtain the corrected fault occurrence probability table. From the revised fault occurrence probability table, extract the historical fault feature code sequence with increasing fault occurrence probability as the fault state evolution path of the target device.

7. The method for monitoring and handling DC power supply faults in a substation according to claim 6, characterized in that, The step of determining several fault source devices from multiple target devices based on the operational status feature library and the modified fault risk assessment results, coupling strength coefficient, and fault state evolution path of each target device includes: Based on the operating status feature library, the corrected fault risk assessment results of each target device and the fault status evolution path, the parameter gradient distribution analysis of multi-source data is performed to identify the set of target devices whose parameter gradient values ​​exceed the gradient threshold as high gradient clustering areas. At the boundary of the high gradient cluster region, identify the region where the parameter gradient value decreases significantly, and generate a gradient reduction marker. Based on the gradient descent marker, the parameter gradients of adjacent target devices are traced step by step along the direction of parameter gradient descent to obtain the low gradient propagation path. Based on the corrected fault risk assessment results, a set of hidden fault candidate target devices is selected from the low-gradient transmission path; Calculate the gradient correlation degree of each hidden fault candidate target device to the high gradient clustering area via the low gradient transmission path; Multiple fault source devices are selected from the set of hidden fault candidate target devices based on the gradient correlation.

8. A substation DC power supply fault monitoring and handling system, characterized in that, It includes a runtime characteristic module, a fault channel module, a coupling analysis module, a risk assessment module, a fault evolution module, a fault location module, and a fault handling module; The operation feature module is used to acquire the operation sequence data and physical equipment information of each target device in the DC power supply system of the substation, and to identify the early deterioration feature data of each target device based on the operation sequence data, so as to construct an operation status feature library. The fault channel module is used to obtain the topology connection relationship of the substation DC power supply system according to the operating status feature library, so as to determine the fault impact channel of the substation DC power supply system according to the topology connection relationship and the operating status feature library. The coupling analysis module is used to determine the initial fault risk assessment result of each target device according to the fault impact channel, and to obtain the coupling strength coefficient between different target devices according to the operating status feature library; The risk assessment module is used to determine the cascading failure modes between different target devices based on the coupling strength coefficient. This includes: determining multiple strongly coupled device pairs in the substation DC power supply system based on the coupling strength coefficient; tracing the cascading propagation path of a fault in the substation DC power supply system based on the strongly coupled device pairs, to identify multiple cascading failure modes corresponding to the substation DC power supply system based on the cascading propagation path; for any target device, obtaining the number of times the target device appears in the multiple cascading failure modes, using the number of appearances as the fault participation degree of the target device; and correcting the initial fault risk assessment result based on the fault participation degree to obtain a corrected fault risk assessment result for each target device. The fault evolution module is used to perform time-series backtracking analysis on each target device according to the operating status feature library, determine the fault occurrence probability table of each target device, and revise the fault occurrence probability table according to the revised fault risk assessment result to determine the fault state evolution path of each target device. The fault location module is used to determine several fault source devices from multiple target devices based on the operating status feature library and the modified fault risk assessment results, coupling strength coefficient and fault state evolution path of each target device. The fault handling module is used to determine the priority of each fault source device, to determine a fault handling sequence of several fault source devices according to the priority, and to complete the fault handling of the substation DC power supply system based on the fault handling sequence.

Images