Power grid safety joint control method for multi-parameter monitoring of large-scale power grid distribution cabinet

Abstract

The application discloses a power grid safety joint control method for multi-parameter monitoring of a large-scale power grid distribution cabinet, and aims to solve the problems of existing power distribution network distribution cabinet monitoring, such as lagging behind in abnormal identification, insufficient joint control closed loop, and low risk control precision. The method divides the power distribution cluster by constructing a global topology mapping model of the distribution cabinet and configuring an edge computing node, synchronously collects multi-dimensional operation parameters of the distribution cabinet, uploads the cloud after time-domain and frequency-domain joint feature extraction and initial screening, completes operation risk assessment and abnormal influence range determination through a multi-parameter coupling abnormality evaluation model, generates matched partition and hierarchical safety joint control instructions, and completes joint control effect closed loop verification after the instructions are executed, thereby improving the abnormal response speed and operation safety of the power distribution network.

Description

Technical Field

[0001] This application relates to the field of power grid distribution cabinet safety control technology, and in particular to a power grid safety control method for large-scale power grid distribution cabinet multi-parameter monitoring. Background Technology

[0002] Against the backdrop of new power system construction, large-scale power grids have integrated numerous distributed power sources and diversified controllable loads on their distribution network side, significantly increasing the dynamics, complexity, and coupling of power grid operation. As the core terminal node for power distribution and equipment management in the distribution network, the operating status of the switchgear directly determines the power supply reliability and overall security and stability of the large-scale power grid. With the continuous upgrading of power grid security management standards, accurate monitoring of the multi-dimensional operating status of switchgear, proactive identification of abnormal risks, and rapid coordinated control of power grid faults have become core requirements for the safe operation and management of large-scale power grids.

[0003] Currently, existing technologies have been researched and applied in the monitoring of distribution cabinet operation and the joint control of power grid safety: some basic solutions deploy sensing units in the distribution cabinet to collect basic electrical parameters such as voltage and current and trigger over-limit alarms, completing simple anomaly monitoring at the single cabinet level; some centralized control solutions build a cloud-based control platform to centrally analyze the monitoring data uploaded by the distribution cabinet and generate corresponding control commands to achieve global control of the power grid; and some optimized solutions introduce edge computing technology to deploy edge nodes on the distribution network side to achieve local preprocessing of monitoring data, which reduces the computing pressure on the cloud to a certain extent and improves the real-time response to anomalies.

[0004] However, the aforementioned existing technologies still have core shortcomings that make them difficult to adapt to the complex operating conditions of large-scale power grids: First, existing monitoring schemes mostly adopt a fixed-frequency full-parameter acquisition mode, without classifying and controlling the impact of parameters on power grid safety, resulting in high data redundancy, heavy processing pressure at the edge, and insufficient monitoring accuracy and response speed for key abnormal parameters; Second, existing anomaly identification schemes mostly use single threshold judgment or simple time-domain feature analysis, without realizing the joint extraction and coupling correlation analysis of time-domain and frequency-domain features of multi-dimensional operating parameters, resulting in problems such as delayed anomaly identification, high false alarm rate, and inability to accurately locate the root cause of anomalies and the scope of fault impact; Third, existing safety joint control schemes mostly adopt a one-size-fits-all centralized control mode, without realizing precise joint control by partition and level based on the anomaly risk level and impact scope, lacking a complete closed-loop verification mechanism for joint control effect, and are prone to control failure under extreme conditions such as communication interruption, failing to guarantee the continuous safe and stable operation of large-scale power grids. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing large-scale power grid distribution cabinet monitoring and safety joint control technologies, such as delayed anomaly identification, insufficient control accuracy, poor closed-loop control, and weak adaptability to extreme operating conditions. This application provides a power grid safety joint control method for multi-parameter monitoring of large-scale power grid distribution cabinets.

[0006] To achieve the above objectives, the power grid safety joint control method for multi-parameter monitoring of large-scale power grid distribution cabinets provided in this application adopts the following technical solution:

[0007] A method for integrated power grid security control based on multi-parameter monitoring of large-scale power grid distribution cabinets includes:

[0008] Based on the power distribution network architecture of a large-scale power grid, a global topology mapping model for power distribution cabinets is constructed. According to the electrical connection relationship, power supply area and load level of the power distribution cabinets, the global power distribution cabinets are divided into multiple power distribution clusters, and a corresponding edge computing node is configured for each power distribution cluster.

[0009] By deploying sensing units in each distribution cabinet, multi-dimensional operating parameters of the corresponding distribution cabinet are collected synchronously. These multi-dimensional operating parameters include electrical operating parameters, equipment status parameters, and insulation performance parameters.

[0010] Each edge computing node performs real-time preprocessing of the multi-dimensional operating parameters of the distribution cabinets within its distribution cluster, and uploads the parameters with joint time-frequency domain anomalies and the unique topology identifier information of the corresponding distribution cabinet in the global topology mapping model to the cloud management platform.

[0011] The cloud-based management platform will perform the following operations:

[0012] By using a pre-built and solidified multi-parameter coupled anomaly assessment model, the parameters of the received time-domain-frequency domain joint characteristic anomalies are input into the multi-parameter coupled anomaly assessment model. The model performs coupling correlation analysis on the multi-dimensional parameters and conducts operational risk assessment on the distribution cabinets corresponding to the parameters of the time-domain-frequency domain joint characteristic anomalies. The anomaly risk level of the corresponding distribution cabinet is obtained. Combined with the global topology mapping model, the power grid impact range associated with the anomaly of the corresponding distribution cabinet and the upstream and downstream related distribution cabinets and electrical equipment are determined.

[0013] Based on the abnormal risk level of the corresponding distribution cabinet, the scope of the power grid impact, and the load attributes of the distribution cabinets and electrical equipment associated with upstream and downstream, a zoned and graded safety joint control instruction matching the joint anomaly characteristics of the time domain and frequency domain is generated. The zoned and graded safety joint control instruction is divided into early warning prompt instruction, load adjustment instruction, fault isolation instruction and emergency power supply switching instruction according to the abnormal risk level from low to high.

[0014] The zoned and hierarchical security joint control command is sent to the edge computing node of the corresponding power distribution cluster, and the edge computing node performs the following operations:

[0015] The zoned and graded safety joint control instructions are broken down and sent to the execution unit of the target distribution cabinet to complete the instruction execution; the execution unit includes circuit breakers, load switches, and disconnect switches in the distribution cabinet;

[0016] After the command is executed, the target power distribution cabinet and its associated power distribution cabinets and electrical equipment are collected by the edge computing node. The data is then uploaded to the cloud management platform to complete the closed-loop verification of the joint control effect. If the closed-loop verification of the joint control effect fails, the multi-dimensional operating parameters collected after the command is executed are combined to dynamically correct the zoning and hierarchical safety joint control command and re-execute it until the closed-loop verification of the joint control effect passes.

[0017] Preferably, when constructing the global topology mapping model of the power distribution cabinet, a dynamic verification and update step is also included:

[0018] Real-time acquisition of the open / close status data of circuit breakers and disconnect switches in each distribution cabinet; real-time verification of the electrical connection relationship of the cabinet based on the conduction logic of electrical connection.

[0019] When the actual open / closed status of circuit breakers and disconnectors in the distribution cabinet is found to be inconsistent with the electrical connection relationship in the global topology mapping model, the global topology mapping model is dynamically updated to correct the electrical connection relationship of the corresponding distribution cabinet, the information of upstream and downstream related distribution cabinets and electrical equipment, and update the division boundary of each distribution cluster.

[0020] The updated global topology mapping model is synchronously distributed to all edge computing nodes to verify the consistency of topology data between edge computing nodes and the cloud management platform.

[0021] Preferably, when synchronously collecting multi-dimensional operating parameters of the corresponding power distribution cabinet, a safety priority hierarchical sampling step is also included:

[0022] Based on the impact weight of multi-dimensional operating parameters on power grid security and abnormal transient characteristics, multi-dimensional operating parameters are divided into first-level key parameters, second-level important parameters, and third-level routine parameters, and corresponding basic sampling frequencies are configured for parameters in different multi-dimensional operating parameters.

[0023] The feature change rate of each parameter in the multi-dimensional operation parameters is obtained in real time through edge computing nodes during the joint feature extraction process of time domain and frequency domain, and a trend warning threshold is preset between the normal threshold and the initial screening threshold of the parameter.

[0024] When the characteristic change rate of any parameter in the multi-dimensional operation parameters exceeds the trend warning threshold, the sampling frequency is increased according to the level of the parameter, and the parameter data collected after the sampling frequency is increased is subjected to time-series alignment and outlier removal.

[0025] The processed parameter data is marked with transmission priority according to its level and transmitted to the edge computing node in sequence according to the transmission priority for joint time-domain and frequency-domain feature extraction and anomaly screening. The first-level key parameters and parameters that trigger trend warning thresholds are cached locally.

[0026] Preferably, when performing time-domain-frequency domain joint feature extraction on electrical operating parameters, equipment status parameters, and insulation performance parameters, and completing the initial screening of anomalies through preset parameter thresholds, adaptive feature optimization is performed. Adaptive feature optimization includes:

[0027] Based on the historical abnormal fault events and corresponding parameters of the distribution cabinets in the distribution cluster, stored in the corresponding edge computing nodes during the historical operation of the distribution cabinets, the correlation between each feature dimension in each time-frequency domain joint feature and the abnormal state of the distribution cabinet is calculated. Adaptive weight coefficients are assigned to each feature dimension in the joint feature vector to increase the proportion of features with high correlation to abnormal state in anomaly identification.

[0028] The feature dimensions after assigning adaptive weight coefficients are reduced in dimensionality to remove redundant features. Redundant features are invalid feature dimensions that have low correlation with the abnormal state of the power distribution cabinet, have overlapping information, have no effective gain for anomaly identification, and increase the computational load of edge computing nodes.

[0029] Based on the real-time load level of the distribution cluster and the seasonal environmental parameters of the distribution cabinet, combined with the feature dimensions of each feature in the joint feature vector obtained by time-frequency domain joint feature extraction, the normal threshold range corresponding to each parameter is dynamically adjusted, and the initial screening of anomalies is completed based on the adjusted normal threshold range.

[0030] Preferably, when completing the coupling correlation analysis of multi-dimensional parameters and determining the abnormal risk level of the corresponding power distribution cabinet, the method also includes an abnormal root cause localization step:

[0031] The multi-parameter coupled anomaly assessment model is used to calculate the coupling correlation degree between electrical operation parameters, equipment status parameters and insulation performance parameters, and to construct the anomaly parameter coupling correlation matrix for the corresponding distribution cabinet anomaly.

[0032] Based on the abnormal parameter coupling correlation matrix, the propagation path of the abnormal parameters is drawn to locate the initial distribution cabinet, initial circuit and initial fault point where the abnormality occurs.

[0033] By combining the type, trend and coupling correlation of abnormal parameters, the root cause type of the corresponding distribution cabinet is identified. The root cause types include equipment failure, abnormal load fluctuation, grid disturbance and external environmental interference.

[0034] Preferably, when determining the power grid impact range associated with an anomaly of a corresponding distribution cabinet and the upstream and downstream related distribution cabinets and electrical equipment by combining a global topology mapping model, dynamic simulation is performed, including:

[0035] Based on the identified root cause type and initial fault point, the fault propagation path and diffusion speed of the corresponding distribution cabinet are simulated using a global topology mapping model.

[0036] Based on the simulation results, the direct impact area, indirect impact area and safe area of ​​the power distribution cabinet anomaly are divided, and the affected power distribution cabinets, electrical equipment and upstream and downstream power nodes in each area are identified.

[0037] For each affected area, calculate the corresponding load transfer capacity, power supply margin, and grid stability after fault isolation to complete a quantitative assessment of the scope of the abnormal impact.

[0038] Preferably, when executing the partitioned and hierarchical security joint control instruction that matches the joint time-domain and frequency-domain feature anomalies, a multi-objective optimization generation step is performed, including:

[0039] With the optimization objectives of fastest abnormal fault isolation speed, smallest power outage range, shortest power outage duration for critical loads, and minimum power flow fluctuation in the power grid, a multi-objective optimization function for the regional and hierarchical safety joint control command is constructed.

[0040] Based on the quantitative assessment of the scope of abnormal impact and load transfer capability, a multi-objective optimization function is solved to generate multiple sets of alternative partitioned and hierarchical safety joint control instructions.

[0041] The candidate partitioned and hierarchical security joint control instruction sets are subjected to equipment interlock verification and power grid stability verification, and the optimal partitioned and hierarchical security joint control instruction set is selected.

[0042] Preferably, when the edge computing node breaks down and sends the partitioned and hierarchical security control commands to the execution unit of the target power distribution cabinet to complete the command execution, timing coordination is performed, including:

[0043] Edge computing nodes determine the action sequence and action logic of each target execution unit based on the global topology mapping model and the received partitioned and hierarchical security joint control instructions, and generate a timing execution plan table;

[0044] Based on the timing execution plan, control commands are issued to the corresponding execution units in the order of isolating the faulty circuit, switching the backup circuit, and adjusting the load.

[0045] During the execution of control commands, action feedback signals of each execution unit are collected in real time. Local backup protection is triggered for execution units that fail to complete actions in the correct sequence. The local backup protection is a protection logic based on pre-configured edge computing nodes that matches the partitioned and hierarchical security joint control commands. It directly issues tripping commands to the circuit breakers of the circuits corresponding to the execution units that fail to perform actions in the correct sequence, thereby completing the local fault isolation of the corresponding circuits.

[0046] Preferably, during the process of completing the closed-loop verification of the joint control effect, quantitative evaluation and model iteration actions are performed, including:

[0047] The cloud-based management and control platform pre-builds a quantitative evaluation system for joint control effectiveness, which quantitatively scores the execution effectiveness of regional and hierarchical safety joint control commands based on fault isolation time, power outage range, load loss, and power grid power flow fluctuation amplitude.

[0048] If the quantitative score is lower than the preset qualified threshold, the closed-loop verification of the joint control effect is deemed to have failed. Simultaneously, the full-process data of this abnormal event is extracted. The full-process data includes parameter data of characteristic anomalies, execution data of regional and hierarchical safety joint control instructions, and execution effect feedback data.

[0049] The extracted full-process data is used as incremental training samples for the multi-parameter coupled anomaly assessment model. The model is then trained online incrementally and iteratively to update the weight coefficients of the multi-parameter coupled anomaly assessment model.

[0050] Preferably, it also includes emergency joint control procedures for extreme working conditions:

[0051] Real-time monitoring of the communication link status between edge computing nodes and the cloud management platform, as well as the power grid operation status. When the duration of the communication link interruption between the edge computing node and the cloud management platform exceeds a preset threshold, the emergency joint control downgrade mode is triggered.

[0052] The cloud-based management platform delegates autonomous control permissions for power distribution clusters to edge computing nodes. The edge computing nodes, based on the global topology mapping model and the lightweight multi-parameter coupled anomaly assessment model pre-deployed on the edge computing nodes, complete the anomaly identification, operational risk assessment, and generation and execution of zoned and graded safety control commands within their respective power distribution clusters.

[0053] Compared with existing technologies, this invention provides a method for multi-parameter monitoring of large-scale power grid distribution cabinets for power grid safety joint control, which has the following beneficial effects:

[0054] 1. By employing multi-dimensional operational parameter safety priority hierarchical sampling, time-domain-frequency domain joint feature extraction, and adaptive feature optimization techniques, on the one hand, hierarchical differentiated control is achieved based on the impact weight of parameters on power grid safety, significantly reducing the redundancy of collected and processed data, alleviating the computational pressure on edge computing nodes, and ensuring the monitoring accuracy and response speed of key abnormal parameters; on the other hand, through multi-dimensional feature correlation weighting and dimensionality reduction optimization, combined with multi-parameter coupling correlation analysis, the false alarm rate of anomalies is effectively reduced, enabling advanced and accurate identification and root cause localization of distribution cabinet operational anomalies, providing a reliable decision-making basis for subsequent safety joint control.

[0055] 2. A precise safety joint control system based on the level of abnormal risk and the scope of impact was constructed. Through multi-objective optimization, joint control instructions adapted to abnormal scenarios were generated. Combined with a time-series collaborative execution mechanism, the orderly and reliable control actions were ensured. At the same time, a closed-loop verification mechanism for joint control effect and a dynamic correction mechanism for instructions were established. This system can isolate faults as quickly as possible, minimize the scope of power outages and load losses, ensure continuous power supply to important loads, and achieve continuous optimization of joint control strategies through closed-loop verification. This significantly improves the precision level and operational stability of safety management and control of large-scale distribution networks.

[0056] 3. Adopting a two-tiered management architecture that coordinates edge computing nodes and a cloud-based management platform, combined with a dynamic verification and update mechanism for the global topology mapping model of the distribution cabinet, it can adapt to the dynamic changes in the large-scale power grid distribution network architecture in real time, ensuring the consistency of topology data and management boundaries. At the same time, through an emergency joint control degradation mode for extreme operating conditions, autonomous management authority is delegated in extreme scenarios such as cloud communication interruption. Relying on the lightweight model of edge nodes, local autonomous joint control and cross-cluster collaboration of the distribution cluster are realized, overcoming the strong dependence of existing technologies on cloud communication, effectively avoiding management failure problems under extreme operating conditions, and comprehensively improving the power supply reliability and safety redundancy capabilities of large-scale power grids in all scenarios. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating the steps of the power grid safety joint control method for multi-parameter monitoring of large-scale power grid distribution cabinets according to an embodiment of this application.

[0058] Figure 2 This is a flowchart of the safety priority hierarchical sampling step in the power grid safety joint control method for multi-parameter monitoring of large-scale power grid distribution cabinets in this application embodiment.

[0059] Figure 3 This is a flowchart of the adaptive feature optimization steps in the power grid safety joint control method for multi-parameter monitoring of large-scale power grid distribution cabinets according to an embodiment of this application.

[0060] Figure 4This is a flowchart illustrating the multi-objective optimization generation step in the power grid safety joint control method for multi-parameter monitoring of large-scale power grid distribution cabinets according to embodiments of this application. Detailed Implementation

[0061] The following is in conjunction with the appendix Figure 1-4 This application will be described in further detail.

[0062] This application discloses a power grid safety joint control method for multi-parameter monitoring of large-scale power grid distribution cabinets. (Refer to...) Figure 1 A method for integrated power grid security control based on multi-parameter monitoring of large-scale power grid distribution cabinets, including:

[0063] S1. Based on the power distribution network architecture of a large-scale power grid, construct a global topology mapping model for the power distribution cabinets. According to the electrical connection relationship, power supply area and load level of the power distribution cabinets, the global power distribution cabinets are divided into multiple power distribution clusters, and a corresponding edge computing node is configured for each power distribution cluster.

[0064] Among them, a global topology mapping model of the distribution cabinet is constructed based on the distribution network design drawings and on-site electrical connection ledgers of the target large-scale power grid;

[0065] The global topology mapping model is a three-layer directed graph topology structure, including:

[0066] Node layer: Each distribution cabinet, circuit breaker / disconnector within the distribution cabinet, upstream and downstream transformers, and load equipment are considered as independent nodes. Each independent node is assigned a unique topology identifier. The topology identifier adopts a four-level coding format of power supply area code - distribution cluster number - distribution cabinet number - circuit number, which has global uniqueness.

[0067] Edge layer: The electrical connection relationship between independent nodes and the power cable parameters are used as the connection edges, and the electrical conduction logic and impedance parameters are marked.

[0068] Attribute layer: Label each independent node with attribute information such as power supply area, load level, rated operating parameters, and equipment model.

[0069] The edge computing nodes are deployed in the power distribution room of the corresponding power distribution cluster. They establish industrial Ethernet + wireless dual-link communication with all sensing units and execution units of the power distribution cluster and are responsible for data acquisition, local preprocessing, instruction execution and feedback within their respective power distribution clusters.

[0070] S2. Through the sensing units deployed in each distribution cabinet, the multi-dimensional operating parameters of the corresponding distribution cabinet are collected synchronously. The multi-dimensional operating parameters include electrical operating parameters, equipment status parameters and insulation performance parameters.

[0071] The electrical operating parameters, equipment status parameters, and insulation performance parameters mentioned above are conventional power operation parameters, and will not be described in detail here.

[0072] S3. Real-time preprocessing of multi-dimensional operating parameters of distribution cabinets within the distribution cluster is performed by each edge computing node. Real-time preprocessing includes time-domain and frequency-domain joint feature extraction of electrical operating parameters, equipment status parameters and insulation performance parameters. Anomalies are initially screened by preset parameter thresholds to obtain the anomaly initial screening judgment boundary. Invalid and redundant data are removed. The parameters with time-domain and frequency-domain joint feature anomalies and the unique topology identifier information of the corresponding distribution cabinet in the global topology mapping model are uploaded to the cloud management platform.

[0073] In the process of joint time-frequency domain feature extraction, time-domain features and frequency-domain features are extracted for the synchronously acquired electrical operation parameters, equipment status parameters and insulation performance parameters, and then spliced ​​together to form a joint feature vector.

[0074] The time-domain features were extracted using a sliding window statistical method, with a sliding window size of 200ms and a step size of 50ms. The extracted time-domain features included mean, peak value, valley value, variance, rate of change, kurtosis, and skewness. The frequency-domain features were extracted using Fast Fourier Transform (FFT) + three-level wavelet packet decomposition, which decomposed the frequency into eight frequency bands. The extracted frequency-domain features included energy entropy, spectral kurtosis, and the proportion of each harmonic in each frequency band.

[0075] Finally, the time-domain features and frequency-domain features of each parameter are concatenated in sequence to form the time-domain-frequency-domain joint feature vector of that parameter.

[0076] It should be noted that the preset parameter thresholds are based on national standards such as the "Low-voltage Power Distribution Design Code" and the "Technical Specification for Relay Protection and Safety Automatic Devices". The preset parameter thresholds will not be described in detail here.

[0077] Furthermore, based on the rated operating parameters of the corresponding power distribution cabinet and the preset benchmark thresholds of historical normal operating data, it is divided into normal fluctuation range, early warning range, and abnormal range;

[0078] By comparing the extracted time-frequency domain joint features with preset parameter thresholds, anomaly screening is completed, and anomaly screening judgment boundary that distinguishes normal / abnormal features is obtained. Invalid redundant data includes outliers exceeding the sensor's range, constant invalid values ​​without fluctuations, and noise data with a signal-to-noise ratio below 10dB, which are removed using the 3σ criterion and median filtering method.

[0079] When uploading parameters with joint time-frequency domain anomalies and the unique topology identifiers of the corresponding distribution cabinets in the global topology mapping model to the cloud management platform, only the parameters determined to be joint time-frequency domain anomalies, along with the unique topology identifiers of the corresponding distribution cabinets and abnormal circuits, are uploaded to the cloud management platform. Normally operating parameter data is only encrypted and cached locally on the edge computing nodes, reducing the computational load and communication bandwidth usage of the cloud management platform.

[0080] S4. Through the pre-built and solidified multi-parameter coupled anomaly assessment model of the cloud management platform, the parameters of the received time-domain-frequency domain joint characteristic anomaly are input into the multi-parameter coupled anomaly assessment model to complete the coupling correlation analysis of multi-dimensional parameters and the operation risk assessment of the distribution cabinet corresponding to the parameters of the collected time-domain-frequency domain joint characteristic anomaly. The anomaly risk level of the corresponding distribution cabinet is obtained. Combined with the global topology mapping model, the power grid impact range associated with the anomaly of the corresponding distribution cabinet and the upstream and downstream related distribution cabinets and electrical equipment are determined.

[0081] The multi-parameter coupled anomaly assessment model adopts a deep learning architecture of bidirectional long short-term memory network (BiLSTM) + attention mechanism. In the pre-training stage of the multi-parameter coupled anomaly assessment model, supervised learning is completed using historical anomaly event data of the target distribution network and multi-dimensional operating parameter data. After training, it is fixed on the cloud management and control platform after simulation quantization and pruning optimization.

[0082] The multi-parameter coupled anomaly assessment model takes as input the parameters of time-domain-frequency domain joint feature anomalies uploaded by edge nodes, and outputs the coupling correlation degree of multi-dimensional parameters and the anomaly risk level of the distribution cabinet.

[0083] S5. Based on the abnormal risk level of the corresponding distribution cabinet, the scope of the power grid impact, and the load attributes of upstream and downstream related distribution cabinets and electrical equipment, the cloud-based management and control platform generates a zoned and graded safety joint control command that matches the joint anomaly characteristics of the time domain and frequency domain. The zoned and graded safety joint control command is divided into early warning prompts, load adjustment commands, fault isolation commands, and emergency power supply switching commands according to the abnormal risk level from low to high.

[0084] S6. The hierarchical safety joint control command is sent to the edge computing node of the corresponding power distribution cluster through the cloud management and control platform. The edge computing node breaks down the hierarchical safety joint control command and sends it to the execution unit of the target power distribution cabinet to complete the command execution. The execution unit includes circuit breakers, load switches and disconnect switches in the power distribution cabinet. The edge computing node collects multi-dimensional operating parameters of the target power distribution cabinet and its associated power distribution cabinets and electrical equipment after the command execution, and uploads them to the cloud management and control platform to complete the closed-loop verification of the joint control effect. If the closed-loop verification of the joint control effect fails, the hierarchical safety joint control command is dynamically corrected based on the multi-dimensional operating parameters collected after the command execution and re-executed until the closed-loop verification of the joint control effect passes.

[0085] The cloud-based management platform has pre-set standards for judging the effectiveness of joint control. These standards include: whether the fault point is effectively isolated, whether abnormal parameters have recovered to the normal fluctuation range, whether important loads have not been interrupted in power supply, and whether power grid flow fluctuations are within the allowable range of national standards. Within 100ms after the command is executed, the edge computing node synchronously collects multi-dimensional operating parameters of the target distribution cabinet and related distribution cabinets and electrical equipment, and uploads them to the cloud-based management platform. The cloud compares the collected post-execution data with the standards to complete the closed-loop verification of the joint control effectiveness.

[0086] Furthermore, the construction of the global topology mapping model for the power distribution cabinet also includes a dynamic verification and update step:

[0087] S11. Real-time acquisition of the open / close status data of circuit breakers and disconnectors in each distribution cabinet, and real-time verification of the electrical connection relationship of the cabinet based on the conduction logic of electrical connection.

[0088] By using the auxiliary contacts of the circuit breakers and disconnectors in each distribution cabinet, combined with the digital input (DI) acquisition module of the corresponding distribution cluster edge computing node, the open / close status data of all circuit breakers and disconnectors in the global topology mapping model are collected in real time.

[0089] Among them, the basic frequency of the switch opening and closing state data acquisition is matched with the basic sampling frequency of the multi-dimensional operating parameters. In this embodiment, it is preferred that both are set to 1Hz by default. When a change in the switch opening and closing state is detected, event-driven millisecond-level high-frequency acquisition is immediately triggered to ensure that no switch state change is missed.

[0090] Furthermore, the conduction logic of the electrical connection is as follows: when all circuit breakers and disconnectors connected in series in a single electrical circuit are in the closed state, the topology connection edge in the global topology mapping model corresponding to the electrical circuit is marked as connected;

[0091] At the same time, when any of the series-connected switching devices is in the open state, the topology connection edge corresponding to the electrical circuit is marked as disconnected;

[0092] Therefore, the actual conduction relationship corresponding to the switch opening and closing states collected in real time is compared with the electrical connection relationship pre-stored in the global topology mapping model loop by loop to complete the real-time verification of the electrical connection relationship of the entire network.

[0093] S12. When the actual open / closed state of the circuit breaker and disconnector in the distribution cabinet is found to be inconsistent with the electrical connection relationship of the global topology mapping model, the global topology mapping model is dynamically updated to correct the electrical connection relationship of the corresponding distribution cabinet, the information of upstream and downstream related distribution cabinets and electrical equipment, and update the division boundary of each distribution cluster.

[0094] In actual work, the following steps need to be performed:

[0095] S121. Correction of topology nodes and connection relationships: For the topology nodes of distribution cabinets and switchgear corresponding to circuits with inconsistent positioning states, correct the conduction attribute of the node and the conduction / disconnection status of the corresponding topology connection edge based on the actual open / closed state.

[0096] If the inconsistency in positioning status is caused by on-site circuit reconnection, addition / removal of distribution cabinets, or equipment replacement, the corresponding topology nodes and connection edges are added / deleted in the global topology mapping model simultaneously. The unique topology identifier information and attribute layer information of the topology node are updated, and the connection relationship of the distribution cabinets and electrical equipment associated with the upstream and downstream of the topology node is corrected simultaneously so that the updated topology model matches the actual electrical conditions on site.

[0097] S122, Distribution Cluster Boundary Update: Based on the corrected topology connection relationship, re-verify the rationality of the division of each distribution cluster;

[0098] If topology modification leads to a decrease in electrical connection coupling within the original distribution cluster or the emergence of strong electrical coupling relationships across clusters, the division boundaries of each distribution cluster will be updated synchronously according to the three classification principles of electrical connection relationship, power supply area, and load level. The distribution cluster affiliation of the corresponding distribution cabinets will be adjusted so that the division of distribution clusters always adapts to the current actual operating architecture of the power grid.

[0099] S13. The updated global topology mapping model is synchronously distributed to all edge computing nodes to complete the topology data consistency verification between edge computing nodes and the cloud management platform.

[0100] After completing the dynamic update of the global topology mapping model, the cloud management platform will fully encrypt the updated global topology mapping model data packet and synchronously distribute it to the edge computing nodes corresponding to all power distribution clusters through the power dedicated communication network in a point-to-point manner.

[0101] Furthermore, after receiving the update data packet, the edge computing node completes the decryption and integrity verification of the update data packet, replaces the historical version of the topology model stored locally, and sends the full node hash verification value of the locally updated topology model back to the cloud management platform.

[0102] Furthermore, the cloud-based management platform compares the hash verification values ​​returned by the edge computing nodes with the standard hash values ​​of the locally updated global topology mapping model node by node:

[0103] If the hash values ​​are completely identical, then the topology data consistency check of the edge computing node is deemed to have passed.

[0104] If the hash values ​​are inconsistent, the data packet retransmission mechanism is immediately triggered to resend the update data packet to the edge node until all edge computing nodes have completed the consistency verification.

[0105] This ensures that the topology data of the cloud management platform and all edge computing nodes are completely unified, providing accurate topology basis for subsequent determination of the scope of anomaly impact and issuance of zoned and graded security joint control commands, effectively solving the problems of command issuance errors and fault isolation failures caused by inconsistent topology data.

[0106] Furthermore, when simultaneously collecting multi-dimensional operating parameters of the corresponding power distribution cabinet, refer to Figure 2 It also includes a security priority tiered sampling step:

[0107] S21. Based on the impact weight of multi-dimensional operating parameters on power grid security and abnormal transient characteristics, multi-dimensional operating parameters are divided into first-level key parameters, second-level important parameters, and third-level routine parameters, and corresponding basic sampling frequencies are configured for parameters in different multi-dimensional operating parameters.

[0108] S22. Real-time acquisition of the feature change rate of each parameter in the multi-dimensional operation parameters during the joint feature extraction process in the time domain and frequency domain through edge computing nodes, and preset trend warning threshold between the normal threshold of the parameter and the abnormal initial screening threshold corresponding to the preset parameter threshold;

[0109] S23. When the characteristic change rate of any parameter in the multi-dimensional operation parameters exceeds the trend warning threshold, the sampling frequency is increased according to the level of the parameter, and the parameter data collected after the sampling frequency is increased is subjected to time-series alignment and outlier removal.

[0110] S24. The processed parameter data is marked with transmission priority according to its level, and transmitted to the edge computing node in sequence according to the transmission priority for time-domain-frequency domain joint feature extraction and anomaly screening. The first-level key parameters and parameters that trigger trend warning thresholds are cached locally.

[0111] Specifically, by combining the Analytic Hierarchy Process (AHP) with historical fault cause statistics, the impact weight ω of each dimension of operating parameters on power grid security is quantitatively calculated. The calculation formula is as follows:

[0112]

[0113] in,

[0114] : Parameter fault cause weight, with a value of 0-1, determined by statistically analyzing the proportion of historical events that caused distribution network faults due to abnormalities in this parameter;

[0115] Fault impact coefficient, with a value range of 0-1, is determined based on the fault impact range and load loss level corresponding to the parameter anomaly.

[0116] : The transient characteristic weight of the parameter, with a value range of 0-1, is determined based on the transient duration and rate of change of the parameter anomaly. The longer the transient duration and the faster the rate of change, the greater the weight. The larger the value;

[0117] Transient risk coefficient, with a value range of 0-1, is determined based on the probability of a fault caused by a transient anomaly in the parameter.

[0118] The safety priority classification and sampling frequency configuration of multi-dimensional operating parameters are shown in Table 1 below:

[0119] Level 1 key parameters ω≥0.7 Electrical operating parameters include three-phase current, three-phase voltage, and zero-sequence current; insulation performance parameters include partial discharge and insulation resistance. Abnormal transients last in the microsecond to millisecond range and change rapidly, making them prone to causing major faults such as short circuits and insulation breakdowns. Secondary important parameters 0.3 ≤ ω < 0.7 Electrical operating parameters include active power, reactive power, power factor, and harmonic content; equipment condition parameters include circuit breaker contact temperature, busbar temperature, and opening / closing displacement. Abnormal transients last for milliseconds to seconds and can easily lead to secondary failures such as equipment overheating and abnormal loads. Level 3 conventional parameters ω < 0.3 The equipment status parameters include the ambient temperature and humidity inside the cabinet and the vibration amplitude of the switching mechanism; the insulation performance parameters include the circuit leakage current. Slow abnormal change rates, lasting from seconds to minutes, are mostly indicative of equipment degradation trends and do not pose a risk of sudden failure.

[0120] Table 1

[0121] Furthermore, the definition methods for the normal threshold, the initial screening threshold for anomalies, and the trend warning threshold are the same as those for the preset parameter thresholds.

[0122] Normal threshold for parameter: The normal fluctuation range determined by the 3σ statistical criterion based on the historical normal operation data of this parameter, denoted as ;

[0123] Anomaly initial screening threshold: The anomaly initial screening judgment boundary corresponding to the preset parameter threshold is the critical value for parameter anomalies, denoted as . If the value exceeds this threshold, it will be directly judged as a feature anomaly;

[0124] Trend warning threshold: Located between the normal threshold and the initial screening threshold for anomalies, this threshold is used to detect early signs of a parameter moving towards an abnormal state. It is divided into upward trend warning thresholds. With the downward trend warning threshold The calculation formula is as follows:

[0125]

[0126]

[0127] In the above formula, k is the early warning coefficient, with a value range of 0.2-0.8. It can be adjusted according to the different safety priority levels of the parameters. For the first-level key parameter, k is 0.2-0.4 to achieve early detection of abnormal trends. For the third-level regular parameter, k is 0.6-0.8 to reduce invalid early warnings.

[0128] Simultaneously, the edge computing nodes acquire the feature change rate of the joint time-frequency domain features of each parameter in real time. The feature change rate is the magnitude of the change in feature value per unit time, calculated using the following formula:

[0129]

[0130] in, Let be the eigenvalue at the current time t. For the previous moment eigenvalues, The sliding time window is consistent with the sliding window for time-domain feature extraction. In this embodiment, it can be 200ms.

[0131] Furthermore, when any parameter is detected to meet any of the following trigger conditions, the sampling frequency of that parameter is immediately adaptively increased:

[0132] a. The characteristic rate of change v of the parameter exceeds the rate of change limit corresponding to the trend warning threshold;

[0133] b. The real-time value of the parameter enters the trend warning threshold range.

[0134] The sampling frequency enhances the security priority level of the matching parameter, specifically as follows:

[0135] First-level key parameter: The sampling frequency is increased from 10kHz to 50kHz to achieve high-precision capture of microsecond-level transient anomalies;

[0136] Secondary important parameter: The sampling frequency is increased from 1kHz to 10kHz to match its moderate transient characteristics;

[0137] Level 3 conventional parameters: The sampling frequency is increased from 1Hz to 1kHz, enhancing the monitoring accuracy of equipment degradation trends.

[0138] Furthermore, for the parameter data acquired after increasing the sampling frequency, the following standardized preprocessing operations are performed simultaneously to improve data quality, including:

[0139] Timing alignment: Based on the IEEE 1588 precision time synchronization protocol, each sampled data point is marked with a unique nanosecond-level timestamp. Linear interpolation is used to align the time axis of parameter data with different sampling frequencies and timing lengths, unify the time base of the data, and ensure the timing consistency of subsequent multi-parameter coupling analysis.

[0140] Outlier removal: The 3σ criterion combined with the Grubbs criterion is used to remove outliers in the sampled data. The positions of the removed outliers are filled with the median value of the adjacent data points to avoid outliers interfering with subsequent feature extraction and anomaly screening.

[0141] Outliers in the sampled data that were removed were: abnormal data points exceeding 3 times the standard deviation, and invalid data caused by sensor malfunction.

[0142] When performing step S24 above, the following is required:

[0143] For the preprocessed parameter data, a corresponding transmission priority is assigned according to the security priority level: Level 1 critical parameters are marked with the highest transmission priority, Level 2 important parameters are marked with medium transmission priority, and Level 3 routine parameters are marked with ordinary transmission priority. Edge computing nodes employ a strict priority (SP) scheduling algorithm to queue parameter data of different priorities: data in the highest priority queue is transmitted first to the feature extraction and anomaly screening module with no transmission delay; data in the medium priority queue is transmitted when the highest priority queue is idle; data in the ordinary transmission priority queue is transmitted when both the high and medium priority queues are idle, ensuring that anomaly features of critical parameters are processed first, improving anomaly response speed.

[0144] After the transmission scheduling is completed, the parameter data is transmitted to the processing module of the edge computing node in order of priority to perform time-domain-frequency domain joint feature extraction and anomaly screening.

[0145] Meanwhile, the full sample data of the first-level key parameters and the full data of all parameters that trigger the trend warning threshold are locally encrypted and cached using the AES-256 encryption algorithm, with a caching period of no less than 90 days; the cached data can be used for subsequent abnormal event tracing and incremental training of multi-parameter coupled abnormal evaluation models.

[0146] Furthermore, when performing time-domain-frequency domain joint feature extraction on electrical operating parameters, equipment status parameters, and insulation performance parameters, and completing the initial anomaly screening through preset parameter thresholds, adaptive feature optimization is performed, referring to... Figure 3 Adaptive feature optimization includes:

[0147] S31. Based on the historical abnormal fault events and corresponding parameters stored in the corresponding edge computing nodes during the historical operation of the distribution cabinets in the distribution cluster, calculate the correlation between each feature dimension in each time-frequency domain joint feature and the abnormal state of the distribution cabinet, assign adaptive weight coefficients to each feature dimension in the joint feature vector, and increase the proportion of features with high correlation to the abnormal state in abnormal identification.

[0148] This includes:

[0149] S311. Dedicated Dataset Construction: A dedicated dataset is constructed based on the historical data stored locally on the corresponding edge computing nodes, and is divided into two categories:

[0150] Anomaly sample set: Contains full data of anomaly features corresponding to all types of fault events within a certain period in the power distribution cluster;

[0151] Normal sample set: contains baseline characteristic data of normal operating conditions under different loads within the cluster;

[0152] S312, Dual-dimensional comprehensive correlation calculation: For each feature dimension of the time-frequency domain joint feature vector, simultaneously calculate the linear and non-linear correlation with the abnormal state;

[0153] The linear correlation is calculated using the Pearson correlation coefficient, a standard method in the relevant field, to capture the linear correspondence between features and anomalies. The nonlinear correlation is calculated using the maximum mutual information coefficient (MIC), a standard method in the relevant field, to capture the nonlinear coupling anomalies unique to the distribution cabinet parameters. Finally, a proprietary fusion formula is used to obtain the comprehensive correlation between each feature dimension and the anomaly state. The formula is:

[0154]

[0155] in,

[0156] Let Pearson correlation coefficient be the value of the i-th feature dimension.

[0157] The maximum mutual information coefficient for the i-th feature dimension.

[0158] These are the weighting coefficients, and ,

[0159] In this embodiment, it is preferred To adapt to scenarios where the nonlinear characteristics of power grid parameters are prominent;

[0160] S313, Adaptive Weight Allocation of Abnormal Contribution: Based on the Calculated Comprehensive Relevance An adaptive weight coefficient associated with the contribution of anomaly detection is assigned to each feature dimension. The formula is:

[0161]

[0162] Where M is the total number of feature dimensions of the joint feature vector.

[0163] S32. Perform dimensionality reduction on the feature dimensions after assigning adaptive weight coefficients and remove redundant features. Redundant features are invalid feature dimensions that have low correlation with the abnormal state of the power distribution cabinet, have overlapping information, have no effective gain for abnormal identification, and increase the computing load of edge computing nodes.

[0164] include:

[0165] S321. Pre-screening and removing redundant features with low relevance: Based on the comprehensive relevance, the first round of redundancy removal is carried out to screen feature dimensions with a comprehensive relevance of less than 0.1 and no effective gain for anomaly identification, thereby reducing the computational workload of subsequent dimensionality reduction processing.

[0166] S322. Principal Component Analysis (PCA) is used to reduce the dimensionality of the pre-screened features, eliminating redundant features with high information overlap and extremely low contribution rate, and finally obtaining the optimized low-dimensional feature vector.

[0167] S323. Specific judgment rules for redundant features, including: invalid features with low correlation to abnormal state of distribution cabinet, duplicate features with overlapping information, and interference features that do not effectively improve anomaly identification and increase the computational load of edge computing nodes.

[0168] S33. Based on the real-time load level of the distribution cluster and the seasonal environmental parameters of the distribution cabinet, and combined with the feature dimensions of the joint feature vector obtained by time-frequency domain joint feature extraction, dynamically adjust the normal threshold range corresponding to each parameter, and complete the initial screening of anomalies based on the adjusted normal threshold range.

[0169] In actual implementation, the operating parameters of the power distribution cluster to which the edge node belongs are obtained in real time. These operating parameters include:

[0170] Real-time load level L is the ratio of the real-time value of the total active power of the distribution cluster to its rated capacity.

[0171] The ambient temperature T, i.e. the measured temperature of the environment where the distribution cabinet is located, provides the operating condition basis for adjusting the normal threshold range.

[0172] Furthermore, based on the different operating condition sensitivities of electrical operating parameters, equipment status parameters, and insulation performance parameters, differentiated influence coefficients are set, and load adjustment coefficients are calculated. With temperature adjustment coefficient The formula is:

[0173]

[0174]

[0175] in,

[0176] The baseline load level;

[0177] The reference ambient temperature;

[0178] This is the load influence factor;

[0179] The temperature influence coefficient is set differently according to the parameter type: higher values ​​are taken for electrical parameters α that are sensitive to load changes, and higher values ​​are taken for equipment status and insulation parameters that are sensitive to temperature.

[0180] Furthermore, based on the adjustment coefficient During the normal fluctuation period of the parameter [ Real-time dynamic updates are performed to obtain a real-time normal threshold range adapted to the current operating conditions. The updated formula is:

[0181] × ×

[0182]

[0183] Based on the adjusted real-time normal threshold range, and combined with the weighted and optimized feature vector, the initial screening of anomalies is completed.

[0184] Furthermore, in completing the coupling correlation analysis of multi-dimensional parameters and determining the abnormal risk level of the corresponding distribution cabinet, the process also includes an abnormal root cause localization step:

[0185] S41. Using a multi-parameter coupled anomaly assessment model, calculate the coupling correlation degree between electrical operation parameters, equipment status parameters, and insulation performance parameters, and construct the anomaly parameter coupling correlation matrix for the corresponding distribution cabinet anomaly.

[0186] S42. Based on the abnormal parameter coupling correlation matrix, draw the propagation path of the abnormal parameter and locate the initial distribution cabinet, initial circuit and initial fault point where the abnormality occurred.

[0187] S43. Based on the type, trend and coupling correlation of abnormal parameters, identify the root cause type of the corresponding distribution cabinet. The root cause types include equipment failure, abnormal load fluctuation, grid disturbance and external environmental interference.

[0188] Specifically, the multi-parameter coupling anomaly assessment model calculates the coupling correlation degree between three types of parameters—electrical, state, and insulation—using a grey relational analysis algorithm to address the uploaded anomaly parameters, constructs an N×N-dimensional anomaly parameter coupling correlation matrix, and quantifies the causal relationship between the anomaly parameters.

[0189] Furthermore, based on the coupling correlation matrix, with a strong coupling correlation degree ≥ 0.7 as the threshold, a directed acyclic graph of anomaly propagation is constructed in combination with the anomaly occurrence sequence. The node with an in-degree of 0 is the anomaly starting point, which accurately locates the initial distribution cabinet, the initial circuit and the initial fault point, while clearly distinguishing between primary anomalies and secondary anomalies.

[0190] Correspondingly, by combining the types of abnormal parameters, their changing trends, and their coupling correlation, multi-dimensional identification rules are established to accurately identify four types of root causes: equipment failure, abnormal load fluctuations, grid-side disturbances, and external environmental interference. These rules are then simultaneously output to risk assessment and regional safety control instructions.

[0191] On the other hand, when combining the global topology mapping model to determine the power grid impact range associated with the anomaly of the corresponding distribution cabinet and the upstream and downstream related distribution cabinets and electrical equipment, dynamic simulation is performed, including:

[0192] S4a. Based on the identified root cause type and initial fault point, simulate the fault propagation path and spread speed of the corresponding distribution cabinet anomaly using the global topology mapping model.

[0193] S4b. Based on the simulation results, delineate the direct impact area, indirect impact area, and safety area of ​​the power distribution cabinet anomaly, and determine the affected power distribution cabinets, electrical equipment, and upstream and downstream power nodes in each area.

[0194] S4c: For each affected area, calculate the corresponding load transfer capacity, power supply margin, and grid stability after fault isolation to complete the quantitative assessment of the scope of the abnormal impact.

[0195] In actual execution, the initial fault point is used as the simulation starting node. The electrical parameters, upstream and downstream connections, and distribution cluster attributes of the initial fault point are retrieved from the global topology mapping model. Simultaneously, based on the anomaly root cause type, differentiated propagation models and simulation parameters are matched, including:

[0196] For equipment faults such as short circuits and insulation breakdowns, an electromagnetic transient propagation model is matched, with a simulation step size of 1ms, focusing on simulating the propagation path of the fault current.

[0197] To address abnormal load fluctuations and grid-side disturbances, an electromechanical transient propagation model is matched, with a simulation step size of 10ms, focusing on simulating the propagation paths of power and voltage fluctuations.

[0198] To address external environmental interference, a slow characteristic degradation propagation model was matched, with a simulation step size of 1 second, focusing on simulating the degradation diffusion path of insulation and temperature parameters.

[0199] Furthermore, based on the matching propagation model and combined with the electrical conduction logic of the topology nodes, the full-time propagation process of a fault from the initial node along the electrical connections is simulated through distribution network time-domain power flow calculation. The affected nodes and fault parameter amplitudes at each simulation moment are output, and the propagation velocity of the fault in different topology branches is calculated using the following formula:

[0200]

[0201] in,

[0202] The propagation speed of the fault in the branch from node m to node n in the topology;

[0203] This refers to the change in branch power.

[0204] For simulation compensation;

[0205] This represents the branch impedance value.

[0206] Furthermore, when the amplitude of the simulated fault parameter decays to within the normal threshold range, or propagates to the electrical isolation node, the simulation stops, and the complete fault propagation path, fault arrival time at each node, and propagation speed data are output.

[0207] Furthermore, the distribution cabinet where the initial fault point is located, the initial circuit where the fault occurred, and the nodes where the fault parameters exceed the initial screening threshold during the simulation are designated as the directly affected areas. The distribution cabinets, circuit breakers / disconnect switches and other electrical equipment, as well as the directly related load equipment, that are affected within the directly affected areas are identified, providing a target range for subsequent fault isolation commands.

[0208] Furthermore, nodes that are electrically connected to the directly affected area, whose fault parameters enter the trend warning threshold range during the simulation but do not exceed the initial screening threshold for anomalies are designated as indirectly affected areas. The affected distribution cabinets, upstream and downstream power nodes, and interconnection switchgear within the indirectly affected areas are identified to provide the target range for subsequent load regulation and emergency power supply switching instructions.

[0209] Furthermore, nodes that are not electrically connected to the fault propagation path and whose parameters remain within the normal threshold range throughout the simulation are designated as safe zones, and it is clear that no control actions need to be performed in safe zones.

[0210] After performing the above actions, when calculating the corresponding load transfer capacity, power supply margin, and grid stability after fault isolation for each affected area, and completing the quantitative assessment of the scope of the abnormal impact, the following multi-dimensional quantitative assessment of the scope of the abnormal impact is performed:

[0211] S4c1, Load Transfer Capacity Calculation: For the directly affected area and the indirectly affected area, calculate the total amount of non-critical loads that can be transferred in the directly affected area and the maximum carrying capacity of the backup power supply circuit. The ratio of the total amount of non-critical loads to the backup power supply circuit is the load transfer capacity.

[0212] S4c2, Power Margin Calculation: For upstream and downstream power nodes in the affected area, calculate the difference between the current load rate and the rated capacity of the power node, which is the power margin, representing the backup power supply capacity of the power supply after fault isolation.

[0213] S4c3, Calculation of power grid stability after fault isolation: Based on the simulated fault propagation results, the voltage deviation rate, power flow overload rate, and frequency fluctuation range of the system after fault isolation are calculated through static security analysis of the distribution network. When all three indicators meet the national standard requirements, the power grid is judged to be stable, and the stability margin value is output.

[0214] After performing a multi-dimensional quantitative assessment of the scope of anomaly impact, a quantitative assessment report of the scope of anomaly impact is output.

[0215] Furthermore, when executing the partitioned and hierarchical security joint control command that matches the joint time-domain and frequency-domain feature anomalies, a multi-objective optimization generation step is performed, referring to... Figure 4 ,include:

[0216] S51. With the optimization objectives of the fastest abnormal fault isolation speed, the smallest power outage range, the shortest power outage duration of important loads, and the minimum power flow fluctuation of the power grid, a multi-objective optimization function for the partitioned and hierarchical safety joint control command is constructed.

[0217] S511. Sub-objective function standardization: The following four optimization objectives are transformed into sub-objective functions that require minimum solutions. All sub-objective functions are normalized to eliminate dimensional differences, wherein:

[0218] Objective 1: To achieve the fastest speed in isolating abnormal faults, corresponding to the sub-objective function. The total time from the occurrence of a fault to complete isolation is used as the optimization variable; the shorter the total time, the smaller the function value.

[0219] Objective 2: Minimize the power outage area, corresponding to the sub-objective function. The proportion of the total load affected by the power outage to the total load of the abnormally affected area is used as the optimization variable. The smaller the proportion, the smaller the function value.

[0220] Objective 3: Minimize the duration of power outages for critical loads, corresponding to the sub-objective function. The cumulative power outage duration of primary and secondary critical loads is used as the optimization variable; the shorter the duration, the smaller the function value.

[0221] Objective 4: Minimize power flow fluctuations in the power grid, corresponding to the sub-objective function. The mean square values ​​of the grid node voltage deviation rate and branch power flow overload rate after the command is executed are used as optimization variables. The smaller the mean square value, the smaller the function value.

[0222] S512. Integration of Multi-Objective Optimization Functions: Based on all the above sub-objective functions, construct a multi-objective optimization function with weighted constraints, the formula of which is:

[0223]

[0224] in,

[0225] x is the set of optimization variables, which includes the list of devices to be acted upon, the timing of the actions, and the control parameters;

[0226] , These are the weighting coefficients for each sub-objective, with a total weight of 1, which can be dynamically adjusted based on the output anomaly risk level.

[0227] For example, in high-risk / major-risk scenarios, improving the speed of fault isolation corresponds to Weighting; in low-to-medium risk scenarios, increase the scope of power outages and ensure the protection of critical loads. , Weights.

[0228] S52. Based on the quantitative assessment of the scope of abnormal impact and load transfer capability, solve the multi-objective optimization function to generate multiple sets of alternative partitioned and hierarchical safety joint control instructions.

[0229] In the multi-objective optimization generation step, the abstract power grid security management requirements are transformed into quantifiable and solvable standardized alphabetic functions, while rigid constraints are set. The specific construction process is as follows:

[0230] Among them, the output quantitatively evaluated anomaly impact range, load transfer capacity, power supply margin, and grid stability margin are used as core boundaries, and the anomaly risk level and time-frequency joint characteristic anomaly type are used as auxiliary inputs. The non-dominated sorting genetic algorithm (NSGA-II), which is commonly used in the field of power system multi-objective optimization, is used to solve the multi-objective optimization function.

[0231] The zoned and graded safety joint control instructions include early warning prompts, load adjustment, fault isolation, and emergency power supply switching actions that match the abnormal risk level. Each action is clearly marked with the executing entity, target equipment, action sequence, and control parameters, ensuring that the instructions can be directly broken down and sent to edge computing nodes for execution.

[0232] S53. Perform equipment interlock verification and power grid stability verification on the candidate partitioned and hierarchical security joint control instruction sets, and select the optimal partitioned and hierarchical security joint control instruction set.

[0233] The specific execution process is as follows:

[0234] S531, Equipment Interlock Verification: For each set of candidate instructions, based on the electrical connection relationships of the global topology mapping model, verify the compliance of the equipment action logic, including:

[0235] Verify that the operating sequence of circuit breakers, disconnectors, and load switches meets the five electrical safety requirements.

[0236] Verify whether there are interlocking conflicts in the operation of devices on the same circuit;

[0237] Verify whether there are timing inconsistencies in the action logic of upstream and downstream devices. Remove candidate instructions that fail verification and retain compliant instructions.

[0238] Power grid stability verification: For candidate instructions that have passed the equipment interlock verification, the power grid operation status after the instruction is executed is simulated through power flow simulation of the distribution network. The node voltage deviation, branch power flow overload rate, and system frequency fluctuation are verified to meet the requirements. At the same time, the power supply margin of the power grid after fault isolation is verified to meet the safety requirements. Candidate instructions that show the risk of power grid over-limit and instability after simulation are eliminated, and the set of compliant candidate instructions that have passed the dual verification is retained.

[0239] Optimal instruction set selection: For compliant candidate instructions that pass the double verification, a fuzzy comprehensive evaluation method is adopted. Combined with the weight setting of the multi-objective optimization function, the four sub-objectives of each group of instructions are comprehensively scored, and the group with the highest comprehensive score is selected as the final output optimal partition-level security joint control instruction set.

[0240] Furthermore, when the edge computing node breaks down and distributes the zoned and hierarchical security control commands to the execution unit of the target power distribution cabinet to complete the command execution, timing coordination is performed, including:

[0241] S61. The edge computing node determines the action sequence and action logic of each target execution unit based on the global topology mapping model and the received partitioned and hierarchical security joint control instructions, and generates a timing execution plan table.

[0242] Specifically, after receiving the partitioned and hierarchical security joint control instructions from the cloud management platform, the edge computing node first completes the standardized decomposition of the instructions: the global joint control instructions issued by the cloud are decomposed into atomic control actions corresponding to each target execution unit. Each atomic action corresponds to a unique target device, where the target device is the circuit breaker, load switch, or disconnect switch in the power distribution cabinet, and the unique topology identifier information of the target device is matched in the global topology mapping model to ensure that the action target is accurate and without deviation.

[0243] Furthermore, based on the electrical connection relationships and electrical safety rules of the global topology mapping model, combined with the core control objectives of this joint control command, which include fault isolation, load regulation, and emergency power supply, the action logic, pre-trigger conditions, and equipment interlocking constraints of each atomic action are determined. At the same time, a precise execution timestamp based on the IEEE1588 precision time synchronization protocol is assigned to each action, and finally a standardized time sequence execution plan table is generated.

[0244] Correspondingly, the core content of the time-series execution plan table includes:

[0245] ① A unique topology identifier for the target execution unit;

[0246] ② Action type, such as tripping, closing, and load regulation;

[0247] ③ Accurately execute timestamps;

[0248] ④ Pre-triggered conditions, such as successful feedback of the preceding action;

[0249] ⑤ Equipment interlock constraint rules;

[0250] ⑥ Action feedback timeout threshold, which is preferably set to 100ms in this embodiment;

[0251] Furthermore, all timing plans meet the core principles of no electrical conflicts, no interlock violations, and priority isolation of faults.

[0252] S62. Based on the timing execution plan, control commands are issued to the corresponding execution units in the order of isolating the faulty circuit first, then switching the backup circuit, and then adjusting the load.

[0253] Specifically, edge computing nodes strictly follow the timing execution schedule, with electrical safety as the core principle, and issue control commands with precise timestamps to the corresponding execution units in a fixed priority order. The specific execution logic is as follows:

[0254] First priority: Fault circuit isolation command issuance: Prioritize issuing trip control commands to the circuit where the initial fault point is located, the fault isolation circuit breaker and load switch in the directly affected area; only after receiving a successful feedback signal from the execution unit of the circuit and confirming that the fault point has been effectively electrically isolated can the next priority command be executed, so as to prevent the fault from spreading from the source.

[0255] Second priority: Issuance of backup power supply circuit switching command: Issuance of closing control command to the power supply interconnection switch and backup power supply incoming switch marked in the indirectly affected area and the global topology mapping model, executes uninterrupted timing control of fault circuit opening and blocking confirmation → backup circuit closing, and completes seamless switching of backup power supply; throughout the switching process, the risk of loop closing is verified to ensure uninterrupted power supply to important loads.

[0256] Third priority: Load regulation command issuance: Issue load peak shaving and staggered peak control commands to the regulating switches corresponding to non-critical loads to optimize the power flow distribution of the power grid and reduce the operating pressure of the power grid after fault recovery.

[0257] S63. During the execution of control commands, the action feedback signals of each execution unit are collected in real time, and local backup protection is triggered for execution units that fail to complete the action in the correct sequence. The local backup protection is a protection logic based on pre-configured edge computing nodes that matches the partitioned and hierarchical security joint control commands. It directly issues tripping commands to the circuit breakers of the circuits corresponding to the execution units that fail to perform the action in the correct sequence, thereby completing the local fault isolation of the corresponding circuits.

[0258] Specifically, throughout the entire process of controlling command execution, the edge computing node collects the action feedback signals of each execution unit in real time through the auxiliary contact of the execution unit, and compares the actual action time, action status with the timing execution plan in real time: if the execution unit completes the corresponding action within the preset timeout threshold and the action status is completely consistent with the command requirements, the action is judged to be executed successfully; if the execution unit fails to provide an action completion signal within the timeout threshold or the action status does not match the command requirements, the action is judged to be executed unsuccessfully, and local backup protection is immediately triggered.

[0259] The implementation logic of the local backup protection is as follows: while receiving the cloud-based partitioned and hierarchical security joint control command, the edge computing node simultaneously receives and embeds the loop-level fault isolation protection logic that is completely matched with the current joint control command in the local programmable logic controller (PLC).

[0260] After the backup protection is triggered, the edge computing node bypasses the communication link of the original execution unit and directly issues a trip command to the upstream incoming circuit breaker of the corresponding circuit of the execution unit that has not completed the action in the correct sequence through hard wiring. This completes the local forced isolation of the corresponding fault circuit and completely cuts off the fault propagation path.

[0261] Meanwhile, edge computing nodes will report abnormal action execution information and backup protection actions to the cloud management platform in real time, providing complete data support for the closed-loop verification of subsequent joint control effects.

[0262] Furthermore, during the process of completing the closed-loop verification of the joint control effect, quantitative evaluation and model iteration actions are performed, including:

[0263] S6a and the cloud-based management and control platform pre-build a quantitative evaluation system for joint control effects, which quantitatively scores the execution effect of the regional and hierarchical safety joint control commands based on fault isolation time, power outage range, load loss, and power grid power flow fluctuation amplitude.

[0264] The cloud-based management platform pre-builds a standardized quantitative evaluation system for joint control effectiveness, which includes four core evaluation dimensions to eliminate differences in measurement units, as detailed below:

[0265] Fault Isolation Time Dimension: This corresponds to the optimization objective of fastest fault isolation speed. The quantifiable metric is the total time from the occurrence of the anomaly to complete electrical isolation of the faulty circuit. The shorter the time, the higher the score. The normalized score is denoted as... ;

[0266] Power Outage Scope Dimension: This dimension corresponds to the area affected by the anomaly and aims to minimize the power outage scope. The quantifiable metric is the ratio of the actual total power outage load to the total load of the affected area; a lower ratio results in a higher score. The normalized score is denoted as [missing metric]. ;

[0267] Load loss dimension: This corresponds to the optimization objective of minimizing the power outage duration of critical loads and the corresponding load transfer capacity. The quantifiable metric is the cumulative power loss caused by this abnormal event. A lower loss result results in a higher score. The normalized score is denoted as... ;

[0268] Power flow fluctuation amplitude dimension: This corresponds to the optimization objective of minimizing power flow fluctuation. The quantified indicators are the root mean square values ​​of the grid node voltage deviation rate and branch power flow overload rate before and after command execution. A smaller fluctuation amplitude results in a higher score. The normalized score is denoted as... ;

[0269] Based on the sub-scoring of the above dimensions, a comprehensive quantitative scoring formula is constructed:

[0270]

[0271] S represents the comprehensive quantitative score of the joint control effect, with a maximum score of 100 points;

[0272] These are the weighting coefficients for each dimension, with a total weight of 1, which are dynamically adjusted based on the output anomaly risk level.

[0273] S6b. If the quantitative score is lower than the preset qualified threshold, the closed-loop verification of the joint control effect is deemed to have failed. Simultaneously, the full-process data of this abnormal event is extracted. The full-process data includes parameter data of characteristic anomalies, execution data of regional and hierarchical safety joint control instructions, and execution effect feedback data.

[0274] When the comprehensive quantitative score S is lower than the preset qualified threshold, the closed-loop verification of the joint control effect is deemed to have failed, and the extraction and encrypted storage of the entire process data of this abnormal event are triggered simultaneously, including:

[0275] Parametric data of feature anomalies: Full parameter data of joint time-frequency domain feature anomalies uploaded by edge computing nodes, including parameter time-series data, joint feature vectors, and unique topology identifier information of the corresponding power distribution cabinet;

[0276] Partitioned and hierarchical security joint control command execution data: including the optimal joint control command set generated in the cloud, atomic control actions after edge node decomposition, timing execution plan table, and action timing and feedback data of each execution unit;

[0277] Execution effect feedback data includes multi-dimensional running parameters collected by edge nodes after command execution, detailed quantitative scoring of joint control effect, analysis data of reasons for verification failure, and full execution data of dynamically corrected joint control commands.

[0278] S6c. Use the extracted full-process data as incremental training samples for the multi-parameter coupled anomaly assessment model, perform online incremental iterative training on the multi-parameter coupled anomaly assessment model, and update the weight coefficients of the multi-parameter coupled anomaly assessment model.

[0279] The multi-parameter coupled anomaly assessment model is a pre-built deep learning model embedded in the cloud management platform. The specific iterative training process is as follows:

[0280] Incremental sample preprocessing: Data cleaning, time-series alignment, and feature standardization are performed on the extracted full-process data. The abnormal risk level, root cause type, and joint control execution effect label of the sample are labeled. The training set and validation set are divided in an 8:2 ratio.

[0281] Online incremental iterative training: Using transfer learning + incremental learning, iterative training is performed on the pre-trained model architecture based on incremental training samples; during the training process, the weights of the model's bottom feature extraction layer are frozen, and only the weight coefficients of the top coupling correlation analysis and risk assessment layers are updated to adapt to newly added abnormal working condition data.

[0282] Model Validation and Consolidation: After training, the accuracy of the iterated model is validated using a validation set. When the accuracy of the model in identifying abnormal risk levels and the accuracy of root cause localization are not lower than the accuracy before iteration, the iteration is deemed valid, and the model with updated weight coefficients is consolidated into the cloud management platform to replace the original model.

[0283] Edge model synchronization: The iterated model is lightly pruned and quantized, and then synchronously updated to all edge computing nodes to ensure the consistency of model accuracy between the cloud and the edge.

[0284] It also includes emergency joint control procedures for extreme working conditions:

[0285] Real-time monitoring of the communication link status between edge computing nodes and the cloud management platform, as well as the power grid operation status. When the duration of the communication link interruption between the edge computing node and the cloud management platform exceeds a preset threshold, the emergency joint control downgrade mode is triggered.

[0286] The cloud-based management platform and edge computing nodes use a heartbeat mechanism to monitor the connection status, transmission latency, and packet loss rate of the dedicated power communication link in real time, while simultaneously monitoring the real-time operating conditions and abnormal risk levels of the power grid. In this embodiment, the preset communication interruption duration threshold is 30 seconds, which can be flexibly adjusted according to the distribution network management level. When the communication link interruption duration between a single or multiple edge computing nodes and the cloud exceeds the preset threshold, the emergency joint control downgrade mode is immediately triggered for the corresponding distribution cluster.

[0287] The cloud-based management platform delegates autonomous control permissions for power distribution clusters to edge computing nodes. The edge computing nodes, based on the global topology mapping model and the lightweight multi-parameter coupled anomaly assessment model pre-deployed on the edge computing nodes, complete the anomaly identification, operational risk assessment, and generation and execution of zoned and graded safety control commands within their respective power distribution clusters.

[0288] During normal operation of the communication link, the cloud-based management and control platform has pre-configured autonomous control permissions for the power distribution cluster to all edge computing nodes and synchronously completed authorization verification. Once the degradation mode is triggered, the permissions automatically take effect without the need for secondary interaction after communication interruption. Based on the dynamically updated global topology mapping model and the lightweight multi-parameter coupling anomaly evaluation model pre-deployed locally after iterative synchronization, the edge computing nodes complete the closed-loop management of the entire process within their respective power distribution clusters.

[0289] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for multi-parameter monitoring of large-scale power grid distribution cabinets for integrated power grid safety control, characterized in that, include: Based on the power distribution network architecture of a large-scale power grid, a global topology mapping model for power distribution cabinets is constructed. According to the electrical connection relationship, power supply area and load level of the power distribution cabinets, the global power distribution cabinets are divided into multiple power distribution clusters, and a corresponding edge computing node is configured for each power distribution cluster. By deploying sensing units in each distribution cabinet, multi-dimensional operating parameters of the corresponding distribution cabinet are collected synchronously. These multi-dimensional operating parameters include electrical operating parameters, equipment status parameters, and insulation performance parameters. Each edge computing node performs real-time preprocessing of the multi-dimensional operating parameters of the distribution cabinets within its distribution cluster, and uploads the parameters with joint time-frequency domain anomalies and the unique topology identifier information of the corresponding distribution cabinet in the global topology mapping model to the cloud management platform. The cloud-based management platform will perform the following operations: By using a pre-built and solidified multi-parameter coupled anomaly assessment model, the parameters of the received time-domain-frequency domain joint characteristic anomalies are input into the multi-parameter coupled anomaly assessment model. The model performs coupling correlation analysis on the multi-dimensional parameters and conducts operational risk assessment on the distribution cabinets corresponding to the parameters of the time-domain-frequency domain joint characteristic anomalies. The anomaly risk level of the corresponding distribution cabinet is obtained. Combined with the global topology mapping model, the power grid impact range associated with the anomaly of the corresponding distribution cabinet, as well as the upstream and downstream related distribution cabinets and electrical equipment, are determined, and dynamic simulation is performed. Based on the abnormal risk level of the corresponding distribution cabinet, the scope of the power grid impact, and the load attributes of the distribution cabinets and electrical equipment associated with upstream and downstream, a zoned and graded safety joint control instruction matching the joint anomaly characteristics of the time domain and frequency domain is generated. The zoned and graded safety joint control instruction is divided into early warning prompt instruction, load adjustment instruction, fault isolation instruction and emergency power supply switching instruction according to the abnormal risk level from low to high. The zoned and hierarchical security joint control command is sent to the edge computing node of the corresponding power distribution cluster, and the edge computing node performs the following operations: The zoned and graded safety joint control instructions are broken down and sent to the execution unit of the target distribution cabinet to complete the instruction execution; the execution unit includes circuit breakers, load switches, and disconnect switches in the distribution cabinet; After the command is executed, the target power distribution cabinet and its associated power distribution cabinets and electrical equipment are collected by the edge computing node. The data is then uploaded to the cloud management platform to complete the closed-loop verification of the joint control effect. If the closed-loop verification of the joint control effect fails, the multi-dimensional operating parameters collected after the command is executed are combined to dynamically correct the zoning and hierarchical safety joint control command and re-execute it until the closed-loop verification of the joint control effect passes.

2. The power grid safety joint control method for multi-parameter monitoring of large-scale power grid distribution cabinets according to claim 1, characterized in that, The process of constructing the global topology mapping model of the power distribution cabinet also includes a dynamic verification and update step: Real-time acquisition of the open / close status data of circuit breakers and disconnect switches in each distribution cabinet; real-time verification of the electrical connection relationship of the cabinet based on the conduction logic of electrical connection. When the actual open / closed status of circuit breakers and disconnectors in the distribution cabinet is found to be inconsistent with the electrical connection relationship in the global topology mapping model, the global topology mapping model is dynamically updated to correct the electrical connection relationship of the corresponding distribution cabinet, the information of upstream and downstream related distribution cabinets and electrical equipment, and update the division boundary of each distribution cluster. The updated global topology mapping model is synchronously distributed to all edge computing nodes to verify the consistency of topology data between edge computing nodes and the cloud management platform.

3. The power grid safety joint control method for multi-parameter monitoring of large-scale power grid distribution cabinets according to claim 1, characterized in that, When synchronously collecting multi-dimensional operating parameters of the corresponding power distribution cabinet, a safety priority hierarchical sampling step is also included: Based on the impact weight of multi-dimensional operating parameters on power grid security and abnormal transient characteristics, multi-dimensional operating parameters are divided into first-level key parameters, second-level important parameters, and third-level routine parameters, and corresponding basic sampling frequencies are configured for parameters in different multi-dimensional operating parameters. The feature change rate of each parameter in the multi-dimensional operation parameters is obtained in real time through edge computing nodes during the joint feature extraction process of time domain and frequency domain, and a trend warning threshold is preset between the normal threshold and the initial screening threshold of the parameter. When the characteristic change rate of any parameter in the multi-dimensional operation parameters exceeds the trend warning threshold, the sampling frequency is increased according to the level of the parameter, and the parameter data collected after the sampling frequency is increased is subjected to time-series alignment and outlier removal. The processed parameter data is marked with transmission priority according to its level and transmitted to the edge computing node in sequence according to the transmission priority for joint time-domain and frequency-domain feature extraction and anomaly screening. The first-level key parameters and parameters that trigger trend warning thresholds are cached locally.

4. The power grid safety joint control method for multi-parameter monitoring of large-scale power grid distribution cabinets according to claim 3, characterized in that, When performing time-domain and frequency-domain joint feature extraction on electrical operating parameters, equipment status parameters, and insulation performance parameters, and completing the initial anomaly screening through preset parameter thresholds, adaptive feature optimization is then performed. Adaptive feature optimization includes: Based on the historical abnormal fault events and corresponding parameters stored in the corresponding edge computing nodes during the historical operation of the distribution cabinets in the distribution cluster, the correlation between each feature dimension in each time-frequency domain joint feature and the abnormal state of the distribution cabinet is calculated. Adaptive weight coefficients are assigned to each feature dimension in the joint feature vector to increase the proportion of features with high correlation to abnormal state in anomaly identification. The feature dimensions after assigning adaptive weight coefficients are reduced in dimensionality to remove redundant features. Redundant features are invalid feature dimensions that have low correlation with the abnormal state of the power distribution cabinet, have overlapping information, have no effective gain for anomaly identification, and increase the computational load of edge computing nodes. Based on the real-time load level of the distribution cluster and the seasonal environmental parameters of the distribution cabinet, combined with the feature dimensions of each feature in the joint feature vector obtained by time-frequency domain joint feature extraction, the normal threshold range corresponding to each parameter is dynamically adjusted, and the initial screening of anomalies is completed based on the adjusted normal threshold range.

5. The method for multi-parameter monitoring of large-scale power grid distribution cabinets for power grid safety joint control according to claim 1, characterized in that, When completing the coupling correlation analysis of multi-dimensional parameters and determining the abnormal risk level of the corresponding power distribution cabinet, the step of abnormal root cause localization is also included: The multi-parameter coupled anomaly assessment model is used to calculate the coupling correlation degree between electrical operation parameters, equipment status parameters and insulation performance parameters, and to construct the anomaly parameter coupling correlation matrix for the corresponding distribution cabinet anomaly. Based on the abnormal parameter coupling correlation matrix, the propagation path of the abnormal parameters is drawn to locate the initial distribution cabinet, initial circuit and initial fault point where the abnormality occurs. By combining the type, trend and coupling correlation of abnormal parameters, the root cause type of the corresponding distribution cabinet is identified. The root cause types include equipment failure, abnormal load fluctuation, grid disturbance and external environmental interference.

6. The power grid safety joint control method for multi-parameter monitoring of large-scale power grid distribution cabinets according to claim 5, characterized in that, When determining the power grid impact range associated with an anomaly of a corresponding distribution cabinet and its upstream and downstream related distribution cabinets and electrical equipment by combining a global topology mapping model, dynamic simulation is performed, including: Based on the identified root cause type and initial fault point, the fault propagation path and diffusion speed of the corresponding distribution cabinet are simulated using a global topology mapping model. Based on the simulation results, the direct impact area, indirect impact area and safe area of ​​the power distribution cabinet anomaly are divided, and the affected power distribution cabinets, electrical equipment and upstream and downstream power nodes in each area are identified. For each affected area, calculate the corresponding load transfer capacity, power supply margin, and grid stability after fault isolation to complete a quantitative assessment of the scope of the abnormal impact.

7. The method for multi-parameter monitoring of large-scale power grid distribution cabinets for power grid safety joint control according to claim 6, characterized in that, When executing the partitioned and hierarchical security joint control command that generates a joint feature anomaly matching the time-domain and frequency-domain, a multi-objective optimization generation step is performed, including: With the optimization objectives of fastest abnormal fault isolation speed, smallest power outage range, shortest power outage duration for critical loads, and minimum power flow fluctuation in the power grid, a multi-objective optimization function for the regional and hierarchical safety joint control command is constructed. Based on the quantitative assessment of the scope of abnormal impact and load transfer capability, a multi-objective optimization function is solved to generate multiple sets of alternative partitioned and hierarchical safety joint control instructions. The candidate partitioned and hierarchical security joint control instruction sets are subjected to equipment interlock verification and power grid stability verification, and the optimal partitioned and hierarchical security joint control instruction set is selected.

8. The method for multi-parameter monitoring of large-scale power grid distribution cabinets for power grid safety joint control according to claim 1, characterized in that, When the edge computing node breaks down and distributes the zoned and hierarchical security control commands to the execution unit of the target power distribution cabinet to complete the command execution, timing coordination is performed, including: Edge computing nodes determine the action sequence and action logic of each target execution unit based on the global topology mapping model and the received partitioned and hierarchical security joint control instructions, and generate a timing execution plan table; Based on the timing execution plan, control commands are issued to the corresponding execution units in the order of isolating the faulty circuit, switching the backup circuit, and adjusting the load. During the execution of control commands, action feedback signals of each execution unit are collected in real time. Local backup protection is triggered for execution units that fail to complete actions in the correct sequence. The local backup protection is a protection logic based on pre-configured edge computing nodes that matches the partitioned and hierarchical security joint control commands. It directly issues tripping commands to the circuit breakers of the circuits corresponding to the execution units that fail to perform actions in the correct sequence, thereby completing the local fault isolation of the corresponding circuits.

9. The method for multi-parameter monitoring of large-scale power grid distribution cabinets for power grid safety joint control according to claim 2, characterized in that, During the process of completing the closed-loop verification of the joint control effect, quantitative evaluation and model iteration actions are performed, including: The cloud-based management and control platform pre-builds a quantitative evaluation system for joint control effectiveness, which quantitatively scores the execution effectiveness of regional and hierarchical safety joint control commands based on fault isolation time, power outage range, load loss, and power grid power flow fluctuation amplitude. If the quantitative score is lower than the preset qualified threshold, the closed-loop verification of the joint control effect is deemed to have failed. Simultaneously, the full-process data of this abnormal event is extracted. The full-process data includes parameter data of characteristic anomalies, execution data of regional and hierarchical safety joint control instructions, and execution effect feedback data. The extracted full-process data is used as incremental training samples for the multi-parameter coupled anomaly assessment model. The model is then trained online incrementally and iteratively to update the weight coefficients of the multi-parameter coupled anomaly assessment model.

10. The power grid safety joint control method for multi-parameter monitoring of large-scale power grid distribution cabinets according to claim 1, characterized in that, It also includes emergency joint control procedures for extreme working conditions: Real-time monitoring of the communication link status between edge computing nodes and the cloud management platform, as well as the power grid operation status. When the duration of the communication link interruption between the edge computing node and the cloud management platform exceeds a preset threshold, the emergency joint control downgrade mode is triggered. The cloud-based management platform delegates autonomous control permissions for power distribution clusters to edge computing nodes. The edge computing nodes, based on the global topology mapping model and the lightweight multi-parameter coupled anomaly assessment model pre-deployed on the edge computing nodes, complete the anomaly identification, operational risk assessment, and generation and execution of zoned and graded safety control commands within their respective power distribution clusters.

Images