Intelligent diagnosis and autonomous treatment method and system for line loss of power distribution network

Abstract

The invention relates to the field of line loss treatment, in particular to a power distribution network line loss intelligent diagnosis and autonomous treatment method and system. The method comprises the following steps: collecting various data and carrying out restoration and normalization processing; checking the space consistency of topology, ledgers and archives through a block chain smart contract, and calculating a load space-time matrix; constructing a line loss gene map, and identifying an abnormal line; generating and executing a treatment strategy; and updating the gene map according to the treatment effect and carrying out closed-loop iteration. The problems that in the prior art, data processing is difficult, topology and ledger are inconsistent, and line loss diagnosis and treatment are not intelligent are solved, whole-process intelligence and automation from data collection to treatment are achieved, and the efficiency and accuracy of power distribution network line loss management are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of line loss management, specifically to a method and system for intelligent diagnosis and autonomous management of line losses in distribution networks. Background Technology

[0002] In current distribution network operations, line loss has consistently been a key factor affecting power system efficiency and economic benefits. With the continuous expansion of the power grid, the widespread integration of distributed power sources, and the increasing complexity of user electricity consumption behavior, traditional distribution network line loss management methods have gradually revealed numerous shortcomings. For example, in data processing, data from meter readings at the user acquisition system, distributed power source output data, and meteorological data are often incomplete, and the inconsistent data formats and dimensions of different data sources make them difficult to use directly for subsequent analysis, significantly reducing the accuracy of line loss calculations and failing to provide a reliable basis for mitigation efforts. Regarding topology and ledger management, there is a lack of effective spatial consistency verification methods between GIS topology, PMS ledgers, and marketing records, and the timestamps for power supply and sales are often difficult to align. This results in significant errors when calculating the load spatiotemporal matrix, failing to accurately reflect the actual operating status of the distribution network. Furthermore, existing technologies lack systematicity and intelligence in the diagnosis and management of line loss anomalies. They fail to form a complete closed loop encompassing data processing, topology verification, line loss analysis, anomaly diagnosis, and management strategy generation. This makes it difficult to quickly build a line loss management system covering the entire process. In particular, they lack the ability to classify line loss genes, identify abnormal lines, and autonomously control them based on multi-source data fusion and intelligent algorithms. Consequently, they cannot achieve full-chain automation and intelligence from data acquisition, spatiotemporal calibration, gene map construction to diagnosis and management. For example, the patent CN119695879A, "A method and system for reconstructing a distribution network considering dynamic line loss management," while improving dynamic line loss management in distribution networks to some extent, still needs improvement in the comprehensiveness and synergy of data processing, intelligent diagnosis, and autonomous management.

[0003] To address the core issues of inefficient multi-source data processing, lack of topology and ledger consistency verification, lack of intelligent classification system for line loss analysis, and fragmented generation of anomaly diagnosis and governance strategies in the aforementioned traditional methods, this application proposes a method and system for intelligent diagnosis and autonomous governance of distribution network line losses. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for intelligent diagnosis and autonomous management of distribution network line losses, in order to solve the problems of inefficient multi-source data processing, lack of topology and ledger consistency verification, lack of intelligent classification system for line loss analysis, and fragmented generation of anomaly diagnosis and management strategies in traditional methods.

[0005] To achieve the above objectives, the following technical solution is adopted.

[0006] A method for intelligent diagnosis and autonomous management of distribution network line losses, comprising: S1: Collect meter data from the data collection system, distributed power output data, and meteorological data; repair and normalize missing data; output a preprocessed dataset. S2: Based on the preprocessed dataset, verify the spatial consistency of GIS topology, PMS ledger, and marketing files using blockchain smart contracts; align power supply and sales timestamps and calculate the load spatiotemporal matrix to generate spatiotemporal calibration data. S3: Extract line feature vectors from the spatiotemporal calibration data; construct a line loss gene map containing Class I low-loss genes, Class II sensitive genes, and Class III high-loss genes through cluster analysis; output gene classification results. S4: Identify high-loss / negative-loss lines using gene classification results; perform diagnosis at the edge terminal: verify CT ratio errors and distributed power supply backfeed metering; correlate infrared temperature and current surges to locate contact resistance anomalies; generate equipment-level anomaly reports. S5: Generate governance strategies based on anomaly reports: control reactive power compensation and transformer voltage regulation at the edge layer; generate topology reconfiguration commands in the cloud and issue control signals. S6: Update the gene map based on the post-governance line loss rate and perform closed-loop iteration.

[0007] Optionally, step S2 specifically includes, Step S2.1: Retrieve the GIS topology node coordinate information, PMS ledger records of device parameters, and marketing system user profile linkage relationships from the preprocessed dataset, and perform spatial consistency verification through a blockchain smart contract: Traverse the connection relationship between each user and the transformer in the marketing user profile. When the coordinate position of the user in the GIS system exceeds the power supply radius of the corresponding transformer, mark it as a spatially inconsistent node. Triggering a blockchain smart contract to attach spatially inconsistent nodes to the nearest transformer in the GIS topology generates a block storage record containing the original attachment relationship, the new attachment relationship, and a timestamp, and outputs spatial calibration data; Step S2.2: Align the timestamps of the power supply from the data acquisition system with the power sales from the marketing system: The time difference between the meter reading time of the data collection system and the meter reading time of the marketing system is identified. When the time difference exceeds a preset threshold, the cubic spline interpolation algorithm is used to unify the electricity data of the two systems to the same time series to obtain time-aligned data. Step S2.3: Based on the transformer impedance parameters in the spatial calibration data and the nodal power timing data in the time-aligned data, calculate the load spatiotemporal matrix: Based on the PMS ledger, obtain the equivalent impedance from each node to the power supply point and calculate the inverse impedance weight. Multiply the power value of each node by the corresponding inverse impedance weight, and then sum the weighted values ​​of all nodes to generate a load spatiotemporal matrix as spatiotemporal calibration data.

[0008] Optionally, step S3 includes: Step S3.1: Extracting line feature quantities based on spatiotemporal calibration data: Obtain line length data from GIS topology. Calculate the average load rate from power timing data Daily average power factor and three-phase imbalance are extracted from electrical quantity data. Calculate power penetration rate from distributed power monitoring data to form a set of feature vectors; Step S3.2: Perform cluster analysis on the feature vector set: Density clustering algorithm is used to divide the line clusters. Set a first threshold and a second threshold based on historical line loss rates. Clusters with a center line loss rate below the first threshold are labeled as Class I low-loss gene lines. Cluster centerline loss rates between the first and second thresholds are labeled as Class II sensitive gene circuits. The cluster center line loss rate is marked as Class III high-loss gene lines, and the gene classification label set is output. Step S3.3: Map the gene classification tag set to the GIS geographic information system: The class I gene circuits are identified as the first visual code. The class II gene circuits are identified as the second visual code. The Class III gene circuits are identified as the third visualization code, and a heat map of line loss genes is generated.

[0009] Optionally, step S4 includes: Step S4.1: Identifying abnormal circuits based on gene classification results: Type III high-loss gene circuits are marked as high-loss abnormal circuits. Mark lines with a line loss rate below zero as negative loss abnormal lines and output a set of abnormal line identifiers; Step S4.2: Perform edge diagnosis on the negative loss abnormal line: Access the CT multiplier parameters recorded in the marketing system. Compare the calculated values ​​of the CT secondary current collected from the edge terminal of the line to be diagnosed. When the deviation exceeds the preset current error threshold, a CT ratio correction command is generated; Synchronous detection of the voltage phase angle at the grid connection point of distributed power sources. When the phase angle is greater than 90 degrees and the inverter is in grid-connected state, the back-feeding power metering supplementary recording is triggered. Step S4.3: Perform edge diagnosis on high-loss abnormal lines: Obtain the timing data of the connector temperature collected by the infrared temperature measurement module of this line. The load current time-series data collected by the associated edge terminal When the ratio of the rate of temperature change to the rate of current change exceeds the preset temperature rise threshold, locate the coordinates of the contact resistance fault point. Step S4.4: Integrate the CT ratio correction command, the reverse power metering supplementary recording command, and the fault point coordinates to generate an equipment-level anomaly report.

[0010] Optionally, step S5 includes: Step S5.1: When the equipment-level anomaly report contains power factor exceeding the limit information, generate a parallel capacitor switching control signal and output it to the edge layer reactive power compensation device. Step S5.2: When the equipment-level anomaly report contains voltage fluctuation limit information, generate an on-load tap changer adjustment command and output it to the transformer control terminal; Step S5.3: When the equipment-level anomaly report contains the identifier of a Class III high-loss gene line, a load transfer tie switch operation sequence is generated based on a GIS topology search for nearby Class I low-loss gene lines. Step S5.4: When the equipment-level anomaly report contains the coordinates of the contact resistance fault point, generate an equipment maintenance work order and assign it to the mobile terminal; Step S5.5: When the equipment-level anomaly report contains a CT ratio error identifier, update the marketing system's metering parameter configuration; Step S5.6: Integrate control signals, adjustment commands, operation sequences, maintenance work orders, and parameter configuration commands to form a set of autonomous governance control signals.

[0011] Optionally, step S6 includes: Step S6.1: Collect power supply and sales data after the treatment, and calculate the real-time line loss rate; Step S6.2: Update the line feature vector based on the real-time line loss rate value; Step S6.3: Input the updated feature vector into the cluster analysis model to regenerate a new version of the line loss gene map; Step S6.4: Feed back the new version of the line damage gene map to step S4 for a new round of abnormal diagnosis.

[0012] Optionally, locating the contact resistance fault point in step S4.3 includes: Construct a user electricity consumption behavior game matrix, including strategies for normal electricity consumption and strategies for electricity theft; calculate the regret values ​​of the costs of misjudgment and missed judgment; when the minimum and maximum regret values ​​exceed a preset safety threshold, initiate joint verification using multiple evidence chains. Compare the similarity of historical electricity consumption patterns. Verify the seal status image of the meter. A comprehensive judgment is made by combining the current phase offset; finally, a high-confidence user identification for suspected electricity theft is output and written into the device-level anomaly report.

[0013] Optionally, the diagnostic model running on the edge terminal is deployed through the following steps: Train a deep neural network model for contact resistance diagnosis on a cloud platform; The knowledge distillation technique is used to compress deep neural network models into lightweight models; The compressed lightweight model is converted into a format executable by the edge computing engine; Deploy it to the edge terminal of the power distribution IoT and set up a periodic parameter synchronization mechanism.

[0014] A power distribution network line loss intelligent diagnosis and autonomous management system includes, Data acquisition layer: includes smart meters, distributed power monitoring units, and infrared thermometers, used to collect raw monitoring data; The edge processing layer includes, The blockchain verification module connects to the data acquisition layer at its input end and performs spatial consistency verification and timestamp alignment. The diagnostic analysis module connects to the blockchain verification module at its input end and outputs anomaly reports based on the line loss gene map; cloud-based decision-making layer: The strategy generation module, whose input end is connected to the edge processing layer, generates reactive power compensation control, transformer voltage regulation, and topology reconfiguration instructions. The dynamic evaluation module connects to the strategy generation module at its input end, updates the line loss gene map, and feeds it back to the edge processing layer. Execution feedback layer: including reactive power compensation devices, transformer tap changers, and interconnection switches, receives and executes instructions from the cloud decision layer.

[0015] Optionally, the edge processing layer includes: The data cleaning unit uses a long short-term memory network to predict and complete missing data. The blockchain consensus unit deploys smart contracts that automatically correct ledgers; The gene map storage unit stores the cluster center vectors of type I, type II, and type III line-loss genes; The edge diagnostic unit takes current timing data and temperature gradient data as input and outputs the coordinates of the contact resistance fault point.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The intelligent diagnosis and autonomous governance method for distribution network line losses proposed in this application has significant advantages. By collecting various types of data and performing repair and normalization processing, it utilizes blockchain smart contracts to verify the spatial consistency of topology, ledgers, and archives, calculates the load spatiotemporal matrix, constructs a line loss gene map, identifies abnormal lines, generates governance strategies, and finally iterates and updates the gene map in a closed loop. This effectively solves the problems of data processing difficulties, inconsistencies between topology and ledgers, and lack of intelligence in line loss diagnosis and governance in existing technologies, realizing full-process intelligence and automation from data collection to governance. The refinement of step S2 enables precise spatial consistency verification, timestamp alignment, and accurate calculation of the load spatiotemporal matrix, providing a reliable data foundation for subsequent analysis. The process of line feature vector extraction and cluster analysis is described in detail, and the generated line loss gene heat map helps to intuitively understand the line loss status. The edge diagnosis process for abnormal lines is specifically explained, enabling rapid location of equipment-level anomalies. Corresponding governance strategies are generated for different anomalies, forming an autonomous governance control signal set to achieve precise governance. The closed-loop iterative mechanism ensures continuous optimization of the method. Further improvements to the contact resistance fault location method have enhanced the accuracy of identifying suspected electricity theft users and ensured the efficient deployment of the edge terminal diagnostic model. The system architecture, from the data acquisition layer to the execution feedback layer, ensures smooth implementation of the entire method through collaborative work among all layers. The refinement of the edge processing layer further improves the system's processing capacity and reliability. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of an embodiment of a method for intelligent diagnosis and autonomous management of distribution network line losses according to the present invention.

[0018] Figure 2 This is a schematic diagram of a module according to an embodiment of a distribution network line loss intelligent diagnosis and autonomous management system based on the present invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0020] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0021] Example 1 like Figure 1As shown, this invention relates to a method and system for intelligent diagnosis and autonomous management of distribution network line losses, aiming to solve problems such as inefficient multi-source data processing, lack of topology and ledger consistency verification, lack of intelligent classification system for line loss analysis, and fragmented generation of anomaly diagnosis and management strategies in traditional distribution network line loss management through intelligent means.

[0022] In the intelligent diagnosis and autonomous management methods for distribution network line losses, it is first necessary to collect data from multiple data sources, including meter data from the power consumption system, distributed generation output data, and meteorological data. These data sources are extensive and cover all aspects of distribution network operation. However, the actual collected data may have problems such as missing data, anomalies, or inconsistent formats, therefore, preprocessing of this data is necessary.

[0023] Specifically, for missing data, data repair techniques can be employed, such as predictive completion based on machine learning methods like Long Short-Term Memory (LSTM) networks. LSTM is a neural network capable of processing time-series data, suitable for handling time-dependent missing data problems. By training an LSTM model, the values ​​of missing data points can be predicted based on existing data sequences, thus achieving data repair. Simultaneously, to eliminate differences in units and numerical ranges between different data sources, data normalization is necessary to unify the data to a standard range, facilitating subsequent analysis and processing. Normalization typically employs the min-max normalization method, scaling the data to the [0,1] interval, as shown in the following formula: in, For the original data points, and These are the minimum and maximum values ​​of the data column, respectively. This process outputs a preprocessed dataset, providing a high-quality data foundation for subsequent steps.

[0024] Based on the preprocessed dataset, spatial consistency verification of GIS topology, PMS ledgers, and marketing files is performed using blockchain smart contracts. The introduction of blockchain technology ensures the immutability and traceability of data, thereby improving data credibility.

[0025] In practice, the system retrieves GIS topology node coordinates from the preprocessed dataset, device parameters from the PMS ledger, and user profile associations from the marketing system. Through a blockchain smart contract, it iterates through the associations of each user with a transformer in the marketing user profile, checking whether the user's coordinates in the GIS system exceed the power supply radius of the corresponding transformer. If they do, the node is identified as spatially inconsistent, and the blockchain smart contract is triggered to associate the node with the nearest transformer in the GIS topology. Simultaneously, a block record containing the original association, the new association, and a timestamp is generated, and spatial calibration data is output.

[0026] In addition, it is necessary to align the timestamps of the electricity consumption data from the meter reading system and the electricity sales data from the marketing system. By identifying the time difference between the meter reading times of the two systems, when the time difference exceeds a preset threshold, a cubic spline interpolation algorithm is used to unify the electricity data to the same time series, thus obtaining time-aligned data. Cubic spline interpolation is a commonly used interpolation method that can generate smooth curves and is suitable for handling time difference issues in time series data. The specific steps are as follows: Determine the time difference .

[0027] if Exceeding the preset threshold Then, interpolation processing is performed on the electricity data of the data collection system and the marketing system.

[0028] Using the cubic spline interpolation formula: in, and The time of the known data points. The coefficients are calculated using known data points.

[0029] Based on this, the load spatiotemporal matrix is ​​calculated using transformer impedance parameters from the spatial calibration data and node power timing from the time-aligned data. The equivalent impedance from each node to the power source is obtained from the PMS ledger, and the inverse impedance weights are calculated. Then, the power value of each node is multiplied by its corresponding inverse impedance weight, and the weighted sum is applied to all nodes to generate the final load spatiotemporal matrix, which serves as the spatiotemporal calibration data. This process ensures the accuracy of the data in both spatial and temporal dimensions, providing reliable data support for subsequent line loss analysis.

[0030] Based on spatiotemporal calibration data, line feature vectors are extracted, and then a line loss gene map is constructed through cluster analysis. Specifically, line length data is obtained from GIS topology, average load rate is calculated from power time-series data, daily average power factor and three-phase imbalance are extracted from electrical quantity data, and power penetration rate is calculated from distributed power source monitoring data, forming a set of feature vectors. These feature vectors can comprehensively reflect the line's operating status and line loss characteristics.

[0031] Next, cluster analysis is performed on the feature vector set. Density clustering is used to divide the lines into clusters, and a first threshold and a second threshold are set based on historical line loss rates. Clusters with line loss rates below the first threshold are labeled as Class I low-loss gene lines; those with line loss rates between the first and second thresholds are labeled as Class II sensitive gene lines; and those with line loss rates above the second threshold are labeled as Class III high-loss gene lines. A set of gene classification labels is output. Density clustering algorithms (such as DBSCAN) can cluster data points based on their density distribution and are suitable for handling clusters with different shapes and sizes.

[0032] Finally, the gene classification tag set was mapped to a GIS geographic information system, with Class I gene lines identified by the first visual code, Class II gene lines by the second visual code, and Class III gene lines by the third visual code, generating a heat map of line loss genes. This method provides a clear visual representation of the line loss status of different lines, offering a scientific basis for subsequent anomaly diagnosis and remediation strategy development.

[0033] High-loss / negative-loss lines are identified using gene classification results, and diagnosis is performed at the edge terminal. Lines with a Class III high-loss gene are marked as high-loss abnormal lines, and lines with a loss rate below zero are marked as negative-loss abnormal lines, outputting a set of abnormal line identifiers. For negative-loss abnormal lines, the CT magnification parameter recorded in the marketing system is called and compared with the CT secondary current calculated by the edge terminal of the line to be diagnosed. When the deviation exceeds a preset current error threshold, a CT magnification correction command is generated. Simultaneously, the voltage phase angle of the distributed power grid connection point is detected. When the phase angle is greater than 90 degrees and the inverter is in grid-connected status, reverse power metering is triggered for supplementary recording.

[0034] For high-loss abnormal lines, acquire the joint temperature time-series data collected by the infrared temperature measurement module of the line, and correlate it with the load current time-series data collected by the edge terminal. When the ratio of the temperature change rate to the current change rate exceeds the preset temperature rise threshold, locate the coordinates of the contact resistance fault point. The specific steps are as follows: Acquire junction temperature timing data and load current timing data .

[0035] Calculate the rate of temperature change and rate of change of current .

[0036] Calculate the ratio .

[0037] if Exceeding the preset temperature rise threshold Then the coordinates of the contact resistance fault point can be located.

[0038] Finally, by integrating CT ratio correction instructions, reverse power metering supplementary recording instructions, and fault point coordinates, an equipment-level anomaly report is generated. Based on the anomaly report generation and management strategy, when the equipment-level anomaly report contains power factor exceeding limits, a parallel capacitor switching control signal is generated and output to the edge-layer reactive power compensation device; when it contains voltage fluctuation exceeding limits, an on-load tap changer adjustment instruction is generated and output to the transformer control terminal; when it contains Class III high-loss line identifiers, a load transfer tie switch operation sequence is generated based on GIS topology search for adjacent Class I low-loss lines; when it contains contact resistance fault point coordinates, an equipment maintenance work order is generated and allocated to the mobile terminal; when it contains CT ratio error identifiers, the marketing system metering parameter configuration is updated. Finally, the control signals, adjustment instructions, operation sequences, maintenance work orders, and parameter configuration instructions are integrated to form an autonomous management control signal set. This series of operations enables accurate diagnosis and targeted management of abnormal lines, improving the operating efficiency and reliability of the distribution network.

[0039] To ensure continuous system optimization and adaptability, the gene map is updated based on the post-treatment line loss rate and iterated in a closed loop. Specific steps include collecting power supply and sales data after treatment, calculating real-time line loss rate values; updating the line feature vector based on the real-time line loss rate values; inputting the updated feature vector into a clustering analysis model to regenerate a new version of the line loss gene map; and feeding the new line loss gene map back to the anomaly diagnosis stage for a new round of anomaly diagnosis. Through this closed-loop iterative mechanism, the system can continuously learn and adapt to changes in the distribution network's operating status, optimize line loss diagnosis and treatment strategies, and achieve long-term stable and efficient operation.

[0040] When performing edge diagnosis on high-loss abnormal lines, the method for locating contact resistance fault points is further optimized. A user electricity consumption behavior game matrix is ​​constructed, including strategies for normal electricity consumption and electricity theft, and the regret values ​​of misjudgment cost and missed judgment cost are calculated. When the minimum and maximum regret values ​​exceed preset safety thresholds, multi-evidence chain joint verification is initiated, including comparing the similarity of historical electricity consumption patterns, verifying meter seal status images, and combining current phase offset for comprehensive judgment. Finally, a high-confidence suspected electricity theft user identifier is output and written into the equipment-level anomaly report. This optimization process improves the accuracy of identifying suspected electricity theft users and provides stronger protection for the safe operation of the distribution network.

[0041] To ensure the efficient operation of the diagnostic model on edge terminals, the following steps are employed for deployment. First, a deep neural network model for contact resistance diagnosis is trained on a cloud platform, utilizing extensive historical data and advanced training algorithms to enable accurate diagnosis of contact resistance faults. Next, knowledge distillation technology is used to compress the deep neural network model into a lightweight model, reducing computational complexity and storage requirements, making it more suitable for edge terminal operation. Then, the compressed lightweight model is converted into an executable format for the edge computing engine, ensuring smooth loading and execution on the edge terminal. Finally, the model is deployed to the power distribution IoT edge terminal, with a periodic parameter synchronization mechanism implemented to ensure the model receives timely updates to the latest data and parameters, maintaining the stability and accuracy of its diagnostic performance. This deployment method enables the diagnostic model to operate efficiently on edge terminals, improving the system's real-time performance and response speed.

[0042] Example 2 like Figure 2 As shown, this invention also provides an intelligent diagnostic and autonomous management system for distribution network line losses. This system includes a data acquisition layer, an edge processing layer, a cloud decision-making layer, and an execution feedback layer. The data acquisition layer consists of smart meters, distributed power monitoring units, infrared thermometers, etc., and is responsible for collecting raw monitoring data. The edge processing layer includes a blockchain verification module and a diagnostic analysis module. The blockchain verification module performs spatial consistency verification and timestamp alignment, while the diagnostic analysis module outputs anomaly reports based on the line loss gene map. The cloud decision-making layer includes a strategy generation module and a dynamic evaluation module. The strategy generation module generates reactive power compensation control, transformer voltage regulation, and topology reconfiguration instructions based on the anomaly reports, while the dynamic evaluation module updates the line loss gene map and feeds it back to the edge processing layer. The execution feedback layer includes reactive power compensation devices, transformer tap changers, tie switches, etc., and receives and executes the instructions from the cloud decision-making layer. In addition, the edge processing layer includes a data cleaning unit that uses a long short-term memory network to predict and complete missing data; a blockchain consensus unit that deploys smart contracts to automatically correct the ledger; a gene map storage unit that stores the cluster center vectors of type I, II, and III line loss genes; and an edge diagnostic unit that takes current time-series data and temperature gradient data as input and outputs the coordinates of contact resistance fault points. Through the collaborative work between these layers, the entire system achieves intelligent and automated processes from data acquisition to management, effectively improving the efficiency and effectiveness of distribution network line loss management.

[0043] The intelligent diagnosis and autonomous management system for distribution network line losses of the present invention consists of multiple layers, which work together to realize the intelligent and automated process from data collection to management.

[0044] Data Acquisition Layer: This layer includes devices such as smart meters, distributed power supply monitoring units, and infrared thermometers, responsible for collecting raw monitoring data. These devices can acquire real-time operating status information of the power distribution network, providing basic data for subsequent data processing and analysis.

[0045] Edge Processing Layer: This layer includes a blockchain verification module, a diagnostic analysis module, a data cleaning unit, a blockchain consensus unit, a gene map storage unit, and an edge diagnostic unit. Among them: Blockchain verification module: Performs spatial consistency verification and timestamp alignment to ensure data accuracy and reliability.

[0046] Diagnostic Analysis Module: Outputs abnormal reports based on line damage gene maps, providing a basis for subsequent treatment strategies.

[0047] Data cleaning unit: Uses long short-term memory networks to predict and complete missing data, improving data integrity and usability.

[0048] Blockchain consensus unit: Deploys smart contracts that automatically correct ledgers to ensure data consistency and credibility.

[0049] Gene map storage unit: Stores the cluster center vectors of Class I, Class II, and Class III line loss genes, providing a reference for line loss analysis.

[0050] Edge diagnostic unit: Input current timing data and temperature gradient data, output the coordinates of the contact resistance fault point to achieve rapid fault location.

[0051] Cloud-based decision-making layer: This layer includes a strategy generation module and a dynamic evaluation module. Specifically: Strategy generation module: Generates reactive power compensation control, transformer voltage regulation and topology reconfiguration instructions based on anomaly reports, providing strategy support for the optimized operation of the distribution network.

[0052] Dynamic evaluation module: responsible for updating the gene map of line loss and feeding it back to the edge processing layer, realizing closed-loop iteration and continuous optimization of the system.

[0053] Execution Feedback Layer: This layer includes reactive power compensation devices, transformer tap changers, tie switches, and other equipment, responsible for receiving and executing instructions from the cloud-based decision-making layer. These devices can adjust the operating status of the distribution network in real time based on the generated control signals, achieving autonomous management of line losses.

[0054] In the diagnosis of high-loss abnormal lines, accurate location of the contact resistance fault point is one of the key steps. To improve the accuracy of location, this invention proposes a method based on joint verification of multiple evidence chains. The specific steps are as follows: Construct a user electricity consumption behavior game matrix: This matrix includes strategies for normal electricity consumption and strategies for electricity theft. By analyzing users' electricity consumption patterns, calculate the costs and benefits under different strategies, thereby assessing the rationality of user behavior.

[0055] Calculating the regret value of misjudgment and omission costs: Through game theory matrix analysis, the costs of misjudgment and omission are calculated under different decisions. The cost of misjudgment refers to the cost of misjudging a legitimate user as an electricity thief, while the cost of omission refers to the cost of misjudging an electricity thief as a legitimate user. By calculating the regret value, the reliability of the decision can be assessed.

[0056] Joint verification of multiple evidence chains: When the minimum and maximum regret values ​​exceed a preset safety threshold, joint verification of multiple evidence chains is initiated. Specific steps include: Compare historical electricity usage patterns: By analyzing the user's historical electricity usage data, calculate the similarity between the current electricity usage pattern and the historical normal pattern. If the similarity is lower than a preset threshold, the user may be suspected of abnormal behavior.

[0057] Verify meter seal status image: Use image recognition technology to check the status of the meter seal. If the seal status is abnormal, further suspicion is raised that the user may be stealing electricity.

[0058] A comprehensive judgment is made by combining the current phase offset: By analyzing the phase offset of the current, it is determined whether there is abnormal current flow. If the current phase offset exceeds a preset threshold, the user is suspected of electricity theft.

[0059] Output high-confidence suspected electricity theft user identifier: Based on the verification results of the above multi-evidence chain, output a high-confidence suspected electricity theft user identifier and write this identifier into the device-level anomaly report. This process not only improves the accuracy of identifying suspected electricity theft users, but also provides a strong basis for subsequent governance measures.

[0060] To ensure that diagnostic models can run efficiently on edge terminals, this invention proposes a model deployment method based on knowledge distillation. The specific steps are as follows: On a cloud platform, a deep neural network model for contact resistance diagnosis is trained using a large amount of historical data. This model is able to learn complex patterns and relationships, thereby accurately diagnosing contact resistance faults.

[0061] Knowledge distillation is used to compress deep neural network models into lightweight models. Knowledge distillation is a model compression technique that reduces the computational complexity and storage requirements of large models by transferring knowledge from large models to smaller models, while maintaining high diagnostic accuracy.

[0062] The compressed, lightweight model is converted into a format executable by the edge computing engine. This process ensures that the model can be successfully loaded and executed on edge terminals, improving the system's real-time performance and response speed.

[0063] After deploying the model at the edge terminal, a periodic parameter synchronization mechanism is set up. By periodically obtaining the latest model parameters and data updates from the cloud, it is ensured that the model on the edge terminal can adapt to changes in the operating status of the distribution network in a timely manner, maintaining the stability and accuracy of its diagnostic performance.

[0064] To ensure continuous optimization and adaptability of the system, this invention proposes a closed-loop iterative update mechanism. The specific steps are as follows: Data collection after governance: After implementing the governance strategy, power supply and sales data are collected again, and real-time line loss rate values ​​are calculated. This process can reflect the effectiveness of the governance measures in real time.

[0065] Update the line feature vector: Based on the real-time line loss rate, update the line feature vector. This process ensures that the feature vector reflects the latest operating status, providing accurate data support for subsequent analysis.

[0066] Regenerating the line loss gene map: The updated feature vectors are input into the clustering analysis model to regenerate a new version of the line loss gene map. This process can reflect changes in the operating status of the distribution network in a timely manner, providing the latest reference for anomaly diagnosis.

[0067] Feedback to the anomaly diagnosis stage: The new line loss gene map is fed back to the anomaly diagnosis stage for a new round of anomaly diagnosis. This closed-loop iterative mechanism ensures that the system can continuously learn and adapt to changes in the operating status of the distribution network, optimize line loss diagnosis and mitigation strategies, and achieve long-term stable and efficient operation.

[0068] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or its equivalents are included in this invention.

Claims

1. A power distribution network line loss intelligent diagnosis and autonomous management method, characterized in that, Comprising, S1: Collecting data from the metering system, distributed power output data and meteorological data, repairing missing data and normalizing, and outputting the preprocessed data set; S2: Based on the preprocessed data set, the spatial consistency of GIS topology, PMS account and marketing archives is verified through a blockchain smart contract, the time stamps of power supply and sales are aligned, and the load space-time matrix is calculated to generate space-time calibration data; S3: According to the space-time calibration data, the line feature vector is extracted, the line loss gene map containing type I low loss gene, type II sensitive gene and type III high loss gene is constructed through cluster analysis, and the gene classification result is output; S4: Using the gene classification result to identify high loss / negative loss lines, performing diagnosis at the edge terminal: verifying CT ratio error and distributed power reverse sending metering, associating infrared temperature and current mutation to locate contact resistance abnormalities, and generating device-level abnormality report; S5: Based on the abnormality report, generate governance strategy: edge layer control reactive power compensation and transformer voltage regulation, cloud end generates topology reconstruction instruction, and issues control signal; S6: According to the line loss rate after governance, update the gene map and close loop iteration.

2. The power distribution network line loss intelligent diagnosis and autonomous management method according to claim 1, characterized in that, Step S2 specifically includes, Step S2.1: Call the GIS topology node coordinate information in the preprocessed data set, the device parameters of the PMS account record, and the user profile hanging relationship in the marketing system, and execute spatial consistency verification through a blockchain smart contract: Iterate through the hanging relationship of each user and transformer in the marketing user profile, and when the user coordinate position in the GIS system exceeds the power supply radius range of the corresponding transformer, mark it as a spatial inconsistency node; Trigger the blockchain smart contract to hang the spatial inconsistency node to the nearest transformer in the GIS topology, generate a block storage record containing the original hanging relationship, the new hanging relationship and the timestamp, and output the spatial calibration data; Step S2.2: Align the time stamps of power supply and marketing system power supply: Identify the time difference between the meter reading time of the metering system and the meter reading time of the marketing system, and when the time difference exceeds the preset threshold, use the cubic spline interpolation algorithm to unify the power data of the two systems to the same time sequence, and obtain the time alignment data; Step S2.3: Based on the transformer impedance parameters in the spatial calibration data and the node power time sequence in the time alignment data, calculate the load space-time matrix: According to the PMS account, obtain the equivalent impedance of each node to the power point, and calculate the impedance inverse weight; Multiply the power value of each node by the corresponding impedance inverse weight, and then sum all the nodes to generate the load space-time matrix as the space-time calibration data.

3. The power distribution network line loss intelligent diagnosis and autonomous management method according to claim 1, characterized in that, The step S3 includes: Step S3.1: Extracting line feature vector based on space-time calibration data: Obtain line length data from GIS topology, Calculate the average load rate from power time sequence data, Extract daily average power factor and three-phase unbalance degree from electrical quantity data, Calculate the power penetration rate from distributed power monitoring data, and form a feature vector set; Step S3.2: Cluster analysis on feature vector set: Use density clustering algorithm to divide line clusters, Set the first threshold and the second threshold according to the historical line loss rate, marking the cluster center line loss rate below the first threshold as a first-class low-loss gene line, marking the cluster center line loss rate between the first threshold and the second threshold as a second-class sensitive gene line, marking the cluster center line loss rate higher than the second threshold as a third-class high-loss gene line, and outputting a gene classification label set; Step S3.3: mapping the gene classification label set to a GIS geographic information system: identifying the first-class gene line as a first visual coding, identifying the second-class gene line as a second visual coding, identifying the third-class gene line as a third visual coding, and generating a line loss gene heat map.

4. The power distribution network line loss intelligent diagnosis and autonomous management method of claim 1, wherein, The step S4 includes: Step S4.1: identifying an abnormal line according to the gene classification result: marking the third-class high-loss gene line as a high-loss abnormal line, marking the line loss rate below zero as a negative-loss abnormal line, and outputting an abnormal line identification set; Step S4.2: performing edge diagnosis on the negative-loss abnormal line: calling a CT multiplier parameter recorded by a marketing system, comparing a CT secondary current calculation value collected by an edge terminal of the line to be diagnosed, generating a CT multiplier correction instruction when the deviation exceeds a preset current error threshold; synchronously detecting a voltage phase angle of a distributed power supply grid connection point, triggering reverse power metering supplementary recording when the phase angle is greater than 90 degrees and the inverter state is grid-connected; Step S4.3: performing edge diagnosis on the high-loss abnormal line: obtaining joint temperature time series data collected by an infrared temperature measurement module of the line, associating load current time series data collected by the edge terminal, locating a contact resistance fault point coordinate when the ratio of the temperature change rate to the current change rate exceeds a preset temperature rise threshold; Step S4.4: integrating the CT multiplier correction instruction, the reverse power metering supplementary recording instruction, and the fault point coordinate to generate a device-level abnormal report.

5. The power distribution network line loss intelligent diagnosis and autonomous management method according to claim 1, characterized in that, The step S5 includes: Step S5.1: when the device-level abnormal report contains power factor out-of-limit information, generating a shunt capacitor switching control signal and outputting it to an edge layer reactive power compensation device; Step S5.2: when the device-level abnormal report contains voltage fluctuation out-of-limit information, generating an on-load tap changer adjustment instruction and outputting it to a transformer control terminal; Step S5.3: when the device-level abnormal report contains a third-class high-loss gene line identification, searching for adjacent first-class low-loss gene lines based on GIS topology to generate a load transfer tie switch operation sequence; Step S5.4: when the device-level abnormal report contains a contact resistance fault point coordinate, generating a device maintenance work order and distributing it to a mobile terminal; Step S5.5: when the device-level abnormal report contains a CT multiplier error identification, updating a marketing system metering parameter configuration; Step S5.6: integrating the control signals, the adjustment instructions, the operation sequence, the maintenance work order, and the parameter configuration instructions to form an autonomous governance control signal set.

6. The power distribution network line loss intelligent diagnosis and autonomous management method according to claim 1, characterized in that, The step S6 includes: Step S6.1: collecting power supply and power sales data after governance, and calculating a real-time line loss rate value; Step S6.2: updating a line feature vector according to the real-time line loss rate value; Step S6.3: inputting the updated feature vector into a clustering analysis model to regenerate a new version of the line loss gene map; Step S6.4: feed the new line loss gene map to step S4 for a new round of anomaly diagnosis.

7. The power distribution network line loss intelligent diagnosis and autonomous management method according to claim 4, characterized in that, The step S4.3 includes: Construct a user electricity consumption behavior game matrix, including normal electricity consumption behavior strategy and electricity stealing behavior strategy; Calculate the regret value of false positive cost and false negative cost; When the minimum maximum regret value exceeds the preset safety threshold, start multi-evidence chain joint verification: Compare the similarity of historical electricity consumption patterns, Check the meter seal state image, Combine the current phase offset for comprehensive judgment; Finally output the high confidence electricity stealing suspect user identification and write it into the device level anomaly report.

8. The power distribution network line loss intelligent diagnosis and autonomous management method according to claim 1, characterized in that, The diagnostic model running on the edge terminal is deployed by the following steps: Train the contact resistance diagnosis deep neural network model on the cloud platform; Use knowledge distillation technology to compress the deep neural network model into a lightweight model; Convert the compressed lightweight model into an executable format for the edge computing engine; Deploy to the power distribution Internet of Things edge terminal and set up a periodic parameter synchronization mechanism.

9. A power distribution network line loss intelligent diagnosis and autonomous management system based on the power distribution network line loss intelligent diagnosis and autonomous management method of any one of claims 1-8, characterized in that, Including, The data acquisition layer includes smart meters, distributed power supply monitoring units, and infrared thermometers for collecting raw monitoring data; The edge processing layer includes, The blockchain verification module, the input end is connected with the data acquisition layer, and the spatial consistency verification and timestamp alignment are performed; The diagnostic analysis module, the input end is connected with the blockchain verification module, and the abnormal report is output based on the line loss gene map; The cloud decision layer: The strategy generation module, the input end is connected with the edge processing layer, generates reactive power compensation control, transformer voltage regulation and topology reconstruction instructions; The dynamic evaluation module, the input end is connected with the strategy generation module, updates the line loss gene map and feeds back to the edge processing layer; The execution feedback layer includes reactive power compensation devices, transformer tap adjusters, and tie switches, which receive and execute the instructions of the cloud decision layer.

10. The power distribution network line loss intelligent diagnosis and autonomous management system according to claim 9, characterized in that, The edge processing layer includes: The data cleaning unit uses the long short-term memory network to predict and complete the missing data; The blockchain consensus unit deploys an intelligent contract for automatic correction of the account book; The gene map storage unit stores the cluster center vectors of type I, type II and type III line loss genes; The edge diagnosis unit inputs current time series data and temperature gradient data and outputs the contact resistance fault point coordinates.

Images