A method and system for accurately reducing line loss rate in low-voltage distribution areas based on multi-technology integration

By collecting multi-dimensional data to construct dynamic graph models and digital twins, the entire link closed-loop optimization of low-voltage transformer area line loss rate is achieved, solving the problems of difficult topology identification, insufficient accuracy of line loss calculation, and lagging loss reduction strategies, and realizing efficient reduction of line loss rate.

CN123117422APending Publication Date: 2026-11-17STATE GRID SHANDONG ELECTRIC POWER CO PINGDU POWER SUPPLY CO
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611174060.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-04
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

High line loss rates in low-voltage distribution areas, difficulties in topology identification, insufficient accuracy in line loss calculation, static lag in loss reduction strategies, and lack of full-process closed-loop management lead to waste of power resources and economic losses.

Method used

By collecting high-frequency electrical quantities, non-electrical quantities, and business data, a dynamic graphical model and digital twin are constructed. Combined with multi-timescale collaborative optimization, the entire link closed-loop optimization is achieved, the equipment status is dynamically adjusted, line loss is accurately calculated, and loss reduction strategies are iteratively optimized.

Benefits of technology

It achieves high-precision topology identification, real-time line loss distribution output, and full-process closed-loop optimization, significantly reducing the line loss rate, improving the accuracy and efficiency of line loss management, and reducing operation and maintenance costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117422A_ABST
    Figure CN123117422A_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for accurately reducing line loss rate in low-voltage distribution areas based on multi-technology fusion, belonging to the field of low-voltage distribution network line loss management technology. The method first collects multi-source data on high-frequency electrical quantities, environmental and equipment non-electrical quantities, and electricity consumption in the low-voltage distribution area and performs standardized preprocessing; then, it constructs a spatiotemporal dynamic graph model, trains it using a dynamic graph neural network with a joint loss function, and outputs the optimal topology of the distribution area and the real-time line loss distribution of all branches; it integrates multi-source data to construct a digital twin of the distribution area, and combines electricity consumption behavior prediction to achieve multi-timescale collaborative optimization of loss reduction control; it reconstructs low-quality data through a generative model to complete accurate line loss calculation and anomaly tracing; finally, through comparison of loss reduction effects, it performs closed-loop correction of model parameters and iterative optimization of loss reduction strategies.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of power system line loss management technology, and in particular to a method and system for accurately reducing the line loss rate of low-voltage distribution areas based on the integration of multiple technologies. Background Technology

[0002] As the final link in the power supply network, low-voltage distribution areas directly connect to a large number of end users. Their line loss rate is a core indicator for measuring the operational efficiency, energy utilization level, and rationality of power grid planning for power supply companies. The line loss rate directly affects the economic benefits of power companies and is also closely related to energy conservation and power supply reliability. Reducing the line loss rate of low-voltage distribution areas has become a crucial breakthrough for energy conservation and quality improvement in the power industry. In recent years, industries such as animal husbandry and cold storage have rapidly emerged, and villages have developed a considerable number of poultry farmers. Coupled with the newly built natural gas pipeline pressurization stations and other new electricity loads, the electricity load in low-voltage distribution areas has continued to grow, and the scenarios have become increasingly complex, placing higher demands on line loss management. If line loss control is not strengthened, the line loss rate will continue to rise, resulting in a waste of electricity resources and significant economic losses for power supply companies.

[0003] Currently, low-voltage transformer area line loss management faces numerous challenges: First, transformer areas are widely distributed, with complex and frequently changing topologies, numerous equipment nodes, and irregular line routes. Coupled with random fluctuations in new loads, topology identification and accurate real-time line loss distribution are difficult to obtain, posing challenges to line loss calculation and anomaly location. Second, multi-source data acquisition has shortcomings. Various data sources are scattered and of varying quality. Existing calculation methods rely on traditional empirical formulas, ignoring data quality, topology changes, and the impact of new loads, resulting in insufficient accuracy. Third, loss reduction strategy optimization lags behind, often employing single static optimization methods without a closed-loop iterative mechanism. Advanced technologies are applied sporadically, making it difficult to form a complete closed-loop management process. Summary of the Invention

[0004] To thoroughly address the pain points of existing technologies, such as difficulty in topology identification, insufficient accuracy in line loss calculation, static lag in loss reduction strategies, and lack of end-to-end closed-loop management, this invention proposes a method for accurately reducing line loss rate in low-voltage distribution areas based on multi-technology fusion. This method achieves end-to-end closed-loop optimization and continuous loss reduction. The specific technical solution is as follows:

[0005] A method for accurately reducing line loss rate in low-voltage distribution areas based on multi-technology integration includes the following steps:

[0006] S1: Collect high-frequency electrical quantity data, environmental and equipment status non-electrical quantity data, and business data of electrical equipment in the low-voltage distribution area. Then, perform deduplication, outlier removal, and standardization preprocessing on each collected data.

[0007] S2: Construct a dynamic graph model using the high-frequency electrical quantity data from S1, introduce a dynamic graph neural network and design a joint loss function, and train to obtain the optimal topology and real-time line loss distribution map;

[0008] S3: Construct a user behavior model using the business data from S1, and integrate it with the high-frequency electrical quantity data, non-electrical quantity data, and equipment parameters from S1 to build a digital twin. Based on this digital twin, achieve multi-timescale collaborative optimization and dynamically adjust the equipment status.

[0009] S4: Integrate the high-frequency electrical quantity data, environmental and equipment status non-electrical quantity data, and business data collected by S1, design a generative model to reconstruct low-quality data; combine data quality and the real-time line loss distribution map output by S2 to complete the calculation of transformer area line loss and anomaly tracing.

[0010] S5: Compare the line loss calculation and anomaly tracing results in S4 with the active control effect of the digital twin in S3, correct the model parameters and iteratively optimize the loss reduction strategy to continuously reduce the line loss rate of the transformer area.

[0011] Preferably, in step S1, the high-frequency electrical quantity data includes the three-phase instantaneous voltage, three-phase instantaneous current, active power, reactive power, power factor, three-phase unbalance, harmonic content, zero-sequence current, and line impedance value of the distribution transformer outlet and each branch. The sampling frequency is not less than 1kHz, and the sampling timestamps of all nodes are kept synchronized.

[0012] The non-electrical data of the environment and equipment status include the temperature and humidity of the transformer area, the level of environmental pollution, the temperature of the equipment casing, the temperature of the iron core, the insulation resistance value, the vibration amplitude of the equipment, the switch opening and closing status, the service life of the equipment, and the historical defect records of the equipment.

[0013] The business data includes transformer substation relationship files, user electricity categories, installed capacity, historical daily and hourly meter reading data, peak and off-peak electricity consumption records, records of illegal electricity use and electricity theft investigations, historical line loss statistics for the transformer substation, and equipment maintenance operation records.

[0014] Preferably, in step S2, the specific operation of constructing the dynamic graph model is as follows: taking the distribution transformers, branch boxes, metering boxes, and user energy meters in the low-voltage distribution area as graph nodes, the low-voltage power supply lines and electrical connection relationships between nodes as graph edges, the high-frequency electrical quantity data collected at each time as node features, and the measured impedance value of the line as edge weights, constructing the single-time graph structure for each time step according to a fixed sampling time step, and connecting the graph structures of all times in a time series to form a spatiotemporal dynamic graph model.

[0015] Preferably, in step S2, the specific operations for training the optimal topology and real-time line loss distribution map are as follows: Input the time-continuous dynamic graph model into the dynamic graph neural network, extract the spatial electrical correlation and time series evolution features between nodes, and output the predicted value of the topology adjacency matrix and the predicted value of line loss for each branch; design a joint loss function that includes topology reconstruction loss, line loss regression loss and L2 regularization loss, set weight coefficients for the three loss terms and the sum of the weight coefficients is 1, iteratively train the model with the goal of minimizing the joint loss function, and output the optimal identification topology of the transformer area and the real-time line loss distribution map of all branches after the model loss value converges to a preset threshold.

[0016] Preferably, step S3 includes the following sub-steps:

[0017] S31: Based on the electrical topology drawings, equipment nameplate parameters, historical operating data and power grid dispatching rules of the low-voltage distribution area, construct a basic digital twin model of the distribution area and complete the four-dimensional accurate mapping between the virtual model and the physical entity of the distribution area.

[0018] S32: Extract electricity consumption behavior features from user service data in S1, construct a user electricity consumption behavior prediction model, and output load prediction values ​​for users at multiple time scales; input real-time high-frequency electrical quantity data, non-electrical quantity data, and service data from S1 into the basic digital twin model, combine the user load prediction values ​​to complete model calibration and update, and construct a digital twin of the distribution area.

[0019] S33: Based on the digital twin of the distribution area, with the minimization of the line loss rate of the distribution area as the core objective, it divides the optimization into three levels: long-term scale, short-term scale, and real-time scale, and constructs the corresponding optimization objective function for each scale, and matches the corresponding adjustable equipment type and safe operation constraints.

[0020] S34: An improved particle swarm optimization algorithm is used to solve the optimization objective function at each time scale in parallel to obtain the optimal control strategy for each adjustable device. The control strategy is input into a digital twin simulation for verification. After confirming that there are no safety risks, control commands are sent to the physical devices in the distribution area to dynamically adjust the operating status of the devices.

[0021] Preferably, in step S32, the specific operations for constructing the user electricity consumption behavior prediction model are as follows: preprocess the user's historical time-series electricity consumption data, remove abnormal data and align the time series, extract features such as daily load curve, peak-valley electricity consumption difference, load factor, and industry electricity consumption characteristics, and integrate them into the user electricity consumption feature vector; construct the prediction model using a bidirectional long short-term memory network Bi-LSTM, input the user electricity consumption feature vector into the model according to the time series, train the model with historical measured electricity consumption data as labels, and output the predicted values ​​of user electricity load at multiple time scales in the future after training.

[0022] Preferably, in step S33, the specific operation of constructing the multi-time-scale collaborative optimization objective function is as follows: Three optimization time scales are defined: a long-term optimization period of 24 hours, optimizing the switching positions of the reactive power compensation device and the configuration parameters of the three-phase imbalance regulation device; a short-term optimization period of 15 minutes, optimizing the charging and discharging plans of distributed photovoltaic and energy storage equipment; and a real-time optimization period of 1 minute, optimizing the output commands of the active voltage regulator. With minimizing the total line loss over the entire cycle as the core, voltage qualification rate, three-phase imbalance, and equipment adjustment times are introduced as auxiliary objectives. Weight coefficients for each sub-objective are set to construct a multi-objective comprehensive optimization objective function. Simultaneously, safe operation constraints such as node voltage, three-phase imbalance capacity, reactive power compensation capacity, energy storage charging and discharging power, and state of charge are set.

[0023] Preferably, in step S4, the line loss calculation and anomaly tracing of the transformer area are achieved in the following way: a data reconstruction model based on an improved conditional generative adversarial network (CGAN) is constructed. The complete and high-quality multi-source data of the transformer area is used as training samples. After adversarial training between the generator and the discriminator, the generator outputs a reconstructed high-confidence complete dataset based on low-quality data. Based on this reconstructed complete dataset, combined with the real-time line loss distribution map output in S2, the branch loss superposition method is used to calculate the loss value of each branch layer by layer from the end user node to the distribution transformer outlet. The total loss of the transformer area and the loss results of each branch component are obtained by superposition and summarization, thus completing the line loss calculation. The line loss contribution of each branch is calculated and sorted from high to low according to the contribution. Combined with the reverse gradient tracing of multi-source data, the branches and nodes with abnormal line loss are located and the abnormal causes are matched, thus completing the anomaly tracing.

[0024] Preferably, in step S5, when comparing the line loss calculation and anomaly tracing results in S4 with the active control effect of the digital twin in S3, a multi-dimensional evaluation system for the line loss reduction effect of the transformer area is established. The evaluation indicators of this system include line loss rate, branch loss deviation, node voltage qualification rate, three-phase imbalance, and equipment operating status. The line loss calculation and anomaly tracing results in S4 are used as measured feedback values, and the active control simulation results of the digital twin in S3 are used as predicted values. The line loss rate deviation, branch loss deviation, and node voltage deviation between the two are calculated. Then, the adaptive moment estimation Adam algorithm is used to correct the core trainable parameters of the dynamic graph neural network, user electricity consumption behavior prediction model, digital twin, and data reconstruction generative model based on the above deviation values. At the same time, based on the deviation tracing results, the optimization target weight, constraint boundary, and execution cycle of the loss reduction control strategy are iteratively optimized to continuously reduce the line loss rate of the transformer area.

[0025] A system for accurately reducing line loss rate in low-voltage transformer areas based on multi-technology integration includes a multi-source data preprocessing virtual module, a dynamic graph topology and line loss distribution virtual module, a digital twin collaborative optimization virtual module, a line loss calculation and anomaly tracing virtual module, and a closed-loop iterative optimization virtual module. The multi-source data preprocessing virtual module is used for the collection and standardized preprocessing of multi-source data from the transformer area. The dynamic graph topology and line loss distribution virtual module is used for constructing a dynamic graph model of the transformer area, topology identification, and real-time line loss distribution output. The digital twin collaborative optimization virtual module is used for constructing a digital twin of the transformer area and optimizing loss reduction control across multiple time scales. The line loss calculation and anomaly tracing virtual module is used for low-quality data reconstruction, transformer area line loss calculation, and anomaly tracing and location. The closed-loop iterative optimization virtual module is used for comparing loss reduction effects, correcting model parameters, and iteratively optimizing loss reduction strategies.

[0026] The beneficial effects of this invention are as follows:

[0027] 1. This invention simultaneously collects high-frequency electrical quantities, non-electrical quantities, and business data across all dimensions. It captures transient electrical characteristics through synchronous sampling across the entire network at frequencies above 1kHz, and is equipped with a standardized preprocessing process. This solves the pain points of low sampling accuracy and poor data synchronization in traditional methods, providing a highly reliable data source for the entire process of line loss management.

[0028] 2. This invention constructs a spatiotemporal dynamic graph model of the transformer area and introduces a dynamic graph neural network with a multi-dimensional joint loss function to achieve high-precision adaptive topology identification and refined output of real-time line loss distribution of all branches. It solves the industry problem of inaccurate topology and inability to locate line loss branches in traditional methods, and provides accurate spatial basis for targeted loss reduction.

[0029] 3. This invention constructs a full-element digital twin that is precisely mapped to the physical entity of the transformer area, and builds a three-level multi-timescale collaborative optimization system by combining accurate prediction of electricity consumption behavior, so as to realize the collaborative control of multiple types of loss reduction equipment, which not only avoids the risk of power grid operation, but also significantly improves the accuracy and effectiveness of proactive loss reduction.

[0030] 4. This invention constructs a data reconstruction model based on an improved CGAN, which completes low-quality and missing abnormal data with high confidence, solving the problem of large deviations in line loss calculation caused by poor data quality at the end of the distribution area; it completes accurate calculation by combining real-time line loss distribution, and quickly locates anomalies through reverse gradient tracing, greatly improving the efficiency of troubleshooting.

[0031] 5. This invention establishes a closed-loop optimization system covering the entire process of “simulation control - measured feedback - parameter correction - strategy iteration”. Based on the deviation between the measured line loss results and the simulation prediction results, it adaptively corrects the parameters of the entire link model and iteratively optimizes the loss reduction strategy, solving the problem that traditional strategies are static and cannot adapt to dynamic operating conditions, and achieving a continuous and stable reduction in the line loss rate of the transformer area.

[0032] 6. This invention is equipped with a modular line loss management system that covers the core functions of the entire line loss management process. The modules are interconnected and work together, which can be flexibly adapted to low-voltage distribution areas of different types and sizes. This greatly reduces the threshold for implementing intelligent line loss management and the operation and maintenance costs, and has great value for engineering application and promotion. Attached Figure Description

[0033] Figure 1 This is a schematic diagram illustrating the steps of a method for accurately reducing line loss rate in low-voltage distribution areas based on the fusion of multiple technologies, as described in this invention. Detailed Implementation

[0034] A method for accurately reducing line loss rate in low-voltage distribution areas based on multi-technology integration, characterized by the following steps:

[0035] S1: Collect high-frequency electrical quantity data, environmental and equipment status non-electrical quantity data, and business data of electrical equipment in the low-voltage distribution area. Then, perform deduplication, outlier removal, and standardization preprocessing on each collected data.

[0036] S2: Construct a dynamic graph model using the high-frequency electrical quantity data from S1, introduce a dynamic graph neural network and design a joint loss function, and train to obtain the optimal topology and real-time line loss distribution map;

[0037] S3: Integrate the high-frequency electrical quantity data, non-electrical quantity data, and equipment parameters from S1 to construct a digital twin. Based on this digital twin, achieve multi-timescale collaborative optimization and dynamically adjust the equipment status.

[0038] S4: Integrate the high-frequency electrical quantity data, environmental and equipment status non-electrical quantity data, and business data collected by S1, design a generative model to reconstruct low-quality data; combine data quality and the real-time line loss distribution map output by S2 to complete the calculation of transformer area line loss and anomaly tracing.

[0039] S5: Compare the line loss calculation and anomaly tracing results in S4 with the active control effect of the digital twin in S3, correct the model parameters and iteratively optimize the loss reduction strategy to continuously reduce the line loss rate of the transformer area.

[0040] In step S1, the high-frequency electrical quantity data includes the three-phase instantaneous voltage, three-phase instantaneous current, active power, reactive power, power factor, three-phase unbalance, harmonic content, zero-sequence current, and line impedance value of the distribution transformer outlet and each branch. The sampling frequency is not less than 1kHz, and the sampling timestamps of all nodes are kept synchronized.

[0041] The non-electrical data of the environment and equipment status include the temperature and humidity of the transformer area, the level of environmental pollution, the temperature of the equipment casing, the temperature of the iron core, the insulation resistance value, the vibration amplitude of the equipment, the switch opening and closing status, the service life of the equipment, and the historical defect records of the equipment.

[0042] The business data includes transformer substation relationship files, user electricity categories, installed capacity, historical daily and hourly meter reading data, peak and off-peak electricity consumption records, records of illegal electricity use and electricity theft investigations, historical line loss statistics for the transformer substation, and equipment maintenance operation records.

[0043] In step S2, a dynamic graph model is constructed: the distribution transformers, branch boxes, metering boxes, and user energy meters in the low-voltage distribution area are used as graph nodes, the low-voltage power supply lines and electrical connections between nodes are used as graph edges, the high-frequency electrical quantity data collected at each time moment are used as node features, and the measured impedance value of the line is used as the edge weight. The single-time graph structure of each time moment is constructed according to a fixed sampling time step, and the graph structures of all time moments are connected in series according to the time sequence to form a spatiotemporal dynamic graph model.

[0044] In step S2, when training the optimal topology and real-time line loss distribution map, the time-continuous dynamic graph model is input into the dynamic graph neural network to extract the spatial electrical correlation and time series evolution features between nodes, and output the predicted value of the topology adjacency matrix and the predicted value of line loss for each branch. A joint loss function including topology reconstruction loss, line loss regression loss and L2 regularization loss is designed, and weight coefficients are set for the three loss terms, with the sum of the weight coefficients being 1. The model is iteratively trained with the goal of minimizing the joint loss function. After the model loss value converges to a preset threshold, the optimal identification topology of the transformer area and the real-time line loss distribution map of all branches are output.

[0045] Step S3 includes the following sub-steps:

[0046] S31: Based on the electrical topology drawings, equipment nameplate parameters, historical operating data and power grid dispatching rules of the low-voltage distribution area, construct a basic digital twin model of the distribution area and complete the four-dimensional accurate mapping between the virtual model and the physical entity of the distribution area.

[0047] S32: Extract electricity consumption behavior features from user service data in S1, construct a user electricity consumption behavior prediction model, and output load prediction values ​​for users at multiple time scales; input real-time high-frequency electrical quantity data, non-electrical quantity data, and service data from S1 into the basic digital twin model, combine the user load prediction values ​​to complete model calibration and update, and construct a digital twin of the distribution area.

[0048] S33: Based on the digital twin of the distribution area, with the minimization of the line loss rate of the distribution area as the core objective, it divides the optimization into three levels: long-term scale, short-term scale, and real-time scale, and constructs the corresponding optimization objective function for each scale, and matches the corresponding adjustable equipment type and safe operation constraints.

[0049] S34: An improved particle swarm optimization algorithm is used to solve the optimization objective function at each time scale in parallel to obtain the optimal control strategy for each adjustable device. The control strategy is input into a digital twin simulation for verification. After confirming that there are no safety risks, control commands are sent to the physical devices in the distribution area to dynamically adjust the operating status of the devices.

[0050] In step S32, the specific operations for constructing the user electricity consumption behavior prediction model are as follows: preprocess the user's historical time-series electricity consumption data, remove abnormal data and align the time series, extract features such as daily load curve, peak-valley electricity consumption difference, load factor, and industry electricity consumption characteristics, and integrate them into the user electricity consumption feature vector; construct the prediction model using a bidirectional long short-term memory network Bi-LSTM, input the user electricity consumption feature vector into the model according to the time series, train the model with historical measured electricity consumption data as labels, and output the predicted values ​​of user electricity load at multiple time scales in the future after training.

[0051] In step S33, the specific operations for constructing the multi-time-scale collaborative optimization objective function are as follows: Three optimization time scales are defined: a long-term optimization period of 24 hours, optimizing the switching positions of the reactive power compensation device and the configuration parameters of the three-phase imbalance regulation device; a short-term optimization period of 15 minutes, optimizing the charging and discharging plans of distributed photovoltaic and energy storage equipment; and a real-time optimization period of 1 minute, optimizing the output commands of the active voltage regulator. With minimizing the total line loss over the entire cycle as the core objective, voltage qualification rate, three-phase imbalance, and equipment adjustment times are introduced as auxiliary objectives. Weight coefficients for each sub-objective are set to construct a multi-objective comprehensive optimization objective function. Simultaneously, safe operation constraints such as node voltage, three-phase imbalance capacity, reactive power compensation capacity, energy storage charging and discharging power, and state of charge are set.

[0052] In step S4, during the calculation of line loss and anomaly tracing in the transformer substation area, a data reconstruction model based on an improved conditional generative adversarial network (CGAN) is constructed. Using complete, high-quality multi-source data of the transformer substation area as training samples, and after adversarial training between the generator and discriminator, the generator outputs a reconstructed high-confidence complete dataset based on low-quality data. Then, based on this reconstructed complete dataset and combined with the real-time line loss distribution map output in S2, the branch loss superposition method is used to calculate the loss value of each branch layer by layer from the end user node to the distribution transformer outlet. The total loss of the transformer substation area and the loss results of each branch component are obtained by superposition and summarization, completing the line loss calculation. The line loss contribution of each branch is calculated and sorted from high to low contribution. Combined with reverse gradient tracing of multi-source data, the branches and nodes with abnormal line loss are located and the anomaly causes are matched, completing the anomaly tracing.

[0053] In step S5, when comparing the line loss calculation and anomaly tracing results from S4 with the active control effect of the digital twin in S3, a multi-dimensional evaluation system for the line loss reduction effect of the transformer area is constructed. The evaluation indicators of this system include line loss rate, branch loss deviation, node voltage qualification rate, three-phase imbalance, and equipment operating status. The line loss calculation and anomaly tracing results from S4 are used as measured feedback values, and the active control simulation results of the digital twin in S3 are used as predicted values. The line loss rate deviation, branch loss deviation, and node voltage deviation between the two are calculated. Then, the adaptive moment estimation Adam algorithm is used to correct the core trainable parameters of the dynamic graph neural network, user electricity consumption behavior prediction model, digital twin, and data reconstruction generative model based on the above deviation values. At the same time, based on the deviation tracing results, the optimization target weight, constraint boundary, and execution cycle of the loss reduction control strategy are iteratively optimized to continuously reduce the line loss rate of the transformer area.

[0054] The system includes a multi-source data preprocessing virtual module, a dynamic graph topology and line loss distribution virtual module, a digital twin collaborative optimization virtual module, a line loss calculation and anomaly tracing virtual module, and a closed-loop iterative optimization virtual module. The multi-source data preprocessing virtual module is used for the collection and standardized preprocessing of multi-source data from the transformer area. The dynamic graph topology and line loss distribution virtual module is used for constructing a dynamic graph model of the transformer area, topology identification, and real-time line loss distribution output. The digital twin collaborative optimization virtual module is used for constructing a digital twin of the transformer area and multi-timescale loss reduction optimization control. The line loss calculation and anomaly tracing virtual module is used for low-quality data reconstruction, transformer area line loss calculation, and anomaly tracing and location. The closed-loop iterative optimization virtual module is used for comparing loss reduction effects, correcting model parameters, and iteratively optimizing loss reduction strategies.

[0055] The embodiments of the present invention are as follows:

[0056] A method for accurately reducing line loss rate in low-voltage distribution areas based on multi-technology integration includes the following steps:

[0057] S1: Collect high-frequency electrical quantity data, environmental and equipment status non-electrical quantity data, and business data of electrical equipment in the low-voltage distribution area. Then, perform deduplication, outlier removal, and standardization preprocessing on each collected data. The high-frequency electrical quantity data includes three-phase instantaneous voltage, three-phase instantaneous current, active power, reactive power, power factor, three-phase unbalance, harmonic content, zero-sequence current, and line impedance value of the distribution transformer outlet and each branch. The sampling frequency is not less than 1kHz, and the sampling timestamps of all nodes are kept synchronized.

[0058] The non-electrical data of the environment and equipment status include the temperature and humidity of the transformer area, the level of environmental pollution, the temperature of the equipment casing, the temperature of the iron core, the insulation resistance value, the vibration amplitude of the equipment, the switch opening and closing status, the service life of the equipment, and the historical defect records of the equipment.

[0059] The business data includes transformer substation relationship files, user electricity categories, installed capacity, historical daily and hourly meter reading data, peak and off-peak electricity consumption records, records of illegal electricity use and electricity theft investigations, historical line loss statistics for the transformer substation, and equipment maintenance operation records.

[0060] S2: A dynamic graph model is constructed using the high-frequency electrical quantity data from S1. A dynamic graph neural network is introduced and a joint loss function is designed. The optimal topology and real-time line loss distribution map are obtained through training. The specific operation of constructing the dynamic graph model is as follows: The distribution transformers, branch boxes, metering boxes, and user energy meters in the low-voltage distribution area are used as graph nodes. The low-voltage power supply lines and electrical connections between nodes are used as graph edges. The high-frequency electrical quantity data collected at each time moment are used as node features. The measured impedance value of the line is used as the edge weight. The single-time graph structure of each time moment is constructed according to a fixed sampling time step. The graph structures of all time moments are connected in series according to the time sequence to form a spatiotemporal dynamic graph model.

[0061] The expression for the dynamic graph model is as follows:

[0062]

[0063] In the formula, This is the dynamic graph model at time t. Let N be the set of electrical nodes in the transformer substation at time t, and N be the total number of nodes in the substation substation. Let be the set of electrical connection edges between nodes at time t. This indicates that node i and node j have a direct electrical connection. This indicates that there is no direct electrical connection. Let F be the node feature matrix at time t, and F be the feature dimension of a single node, corresponding to the number of features of the high-frequency electrical quantity.

[0064] The specific operations for training the optimal topology and real-time line loss distribution map are as follows: Input the time-continuous dynamic graph model into the dynamic graph neural network, extract the spatial electrical correlation and time series evolution features between nodes, and output the predicted value of the topology adjacency matrix and the predicted value of line loss for each branch; design a joint loss function including topology reconstruction loss, line loss regression loss and L2 regularization loss, set weight coefficients for the three loss terms and the sum of the weight coefficients is 1, and iteratively train the model with the goal of minimizing the joint loss function. After the model loss value converges to the preset threshold, output the optimal identification topology of the transformer area and the real-time line loss distribution map of all branches;

[0065] The expression for the joint loss function is as follows:

[0066]

[0067]

[0068]

[0069]

[0070] In the formula, For the joint loss function value, These are the weight coefficients for the topology reconstruction loss, line loss regression loss, and regularization loss, respectively. ; The topology reconstruction loss is calculated using binary cross-entropy loss. This represents the true value of the topological adjacency matrix. The predicted value of the adjacency matrix output by the model; The regression loss is calculated using the mean square error loss, where M is the total number of branches in the transformer area. This represents the measured line loss value for the k-th branch. This is the predicted line loss value for the kth branch output by the model; Let W be the L2 regularization loss, W be the set of trainable weight parameters of the model, and w be a single weight parameter.

[0071] S3: Construct a user behavior model using the business data from S1, and integrate it with the high-frequency electrical quantity data, non-electrical quantity data, and equipment parameters from S1 to build a digital twin. Based on this digital twin, achieve multi-timescale collaborative optimization and dynamically adjust the equipment status. Step S3 includes the following sub-steps:

[0072] S31: Based on the electrical topology drawings, equipment nameplate parameters, historical operating data and power grid dispatching rules of the low-voltage distribution area, construct a basic digital twin model of the distribution area and complete the four-dimensional accurate mapping between the virtual model and the physical entity of the distribution area.

[0073] S32: Based on the user service data in S1, extract electricity consumption behavior features, construct a user electricity consumption behavior prediction model, and output the user's load prediction values ​​at multiple time scales; input the real-time high-frequency electrical quantity data, non-electrical quantity data, and service data from S1 into the basic digital twin model, and combine the user load prediction values ​​to complete model calibration and update, constructing a digital twin of the distribution area; the specific operations for constructing the user electricity consumption behavior prediction model in step S32 are as follows: preprocess the user's historical time-series electricity consumption data, remove abnormal data and align the time series, extract features such as daily load curves, peak-valley electricity consumption difference, load factor, and industry electricity consumption characteristics, and integrate them into a user electricity consumption feature vector; construct the prediction model using a bidirectional long short-term memory network Bi-LSTM, input the user electricity consumption feature vector into the model according to the time series, train the model with historical measured electricity consumption data as labels, and output the user's electricity load prediction values ​​at multiple time scales after training.

[0074] The cell state update expression of the Bi-LSTM network is as follows:

[0075]

[0076]

[0077]

[0078]

[0079]

[0080]

[0081] In the formula, The output of the forget gate at time t. The input gate output is at time t. The output of the gate at time t is the output of the gate. It is the sigmoid activation function. These are the weight matrices for the corresponding gating. These are the bias terms for the corresponding gating. This is the output of the hidden layer at time t-1. Let t be the user electricity consumption feature vector input at time t. Let be the state of the candidate cell at time t. Let be the cell state at time t. For Hadama accumulation, This is the hidden layer output at time t, used for calculating the final load forecast.

[0082] S33: Based on the digital twin of the transformer substation, with the minimization of line loss rate as the core objective, it divides the optimization into three levels: long-term, short-term, and real-time. Optimization objective functions are constructed for each level, and corresponding adjustable equipment types and safe operation constraints are matched. The specific operation of constructing the multi-time-scale collaborative optimization objective function is as follows: Three optimization time scales are defined: the long-term optimization cycle is 24 hours, optimizing the switching positions of reactive power compensation devices and the configuration parameters of three-phase imbalance adjustment devices; the short-term optimization cycle is 15 minutes, optimizing the charging and discharging plans of distributed photovoltaic and energy storage devices; and the real-time optimization cycle is 1 minute, optimizing the output commands of active voltage regulators. With the minimization of total line loss over the entire cycle as the core objective, voltage qualification rate, three-phase imbalance, and equipment adjustment times are introduced as auxiliary objectives. Weight coefficients for each sub-objective are set to construct a multi-objective comprehensive optimization objective function. Simultaneously, safe operation constraints such as node voltage, three-phase imbalance, reactive power compensation capacity, energy storage charging and discharging power, and state of charge are set.

[0083] The expressions for the multi-objective optimization objective function and constraints are as follows:

[0084]

[0085]

[0086] In the formula, F is the comprehensive objective value of multi-objective optimization. These are the weighting coefficients for each sub-objective; To optimize the total active power line loss of the transformer area within the cycle, To optimize the total power supply of the transformer substation within the cycle, The line loss rate of the transformer area; To optimize the voltage qualification rate of transformer substation nodes within the cycle; The three-phase current imbalance in the transformer substation area; To optimize the total number of equipment adjustments within the cycle;

[0087] In the constraints, These are the lower and upper limits of the allowable node voltage, respectively. Let be the voltage value at node i; This is the upper limit of the allowable three-phase imbalance. These are the minimum and maximum output capacities of the reactive power compensation device, respectively. These are the maximum charging and discharging power of the energy storage device, respectively. Provides real-time charging and discharging power for energy storage devices; These are the lower and upper limits of the state of charge (SOC) of the energy storage device, respectively. SOC is the real-time state of charge of the energy storage device.

[0088] S34: An improved particle swarm optimization algorithm is used to solve the optimization objective function at each time scale in parallel to obtain the optimal control strategy for each adjustable device. The control strategy is input into a digital twin simulation for verification. After confirming that there are no safety risks, control commands are sent to the physical devices in the distribution area to dynamically adjust the operating status of the devices.

[0089] S4: Integrating high-frequency electrical quantity data, environmental and equipment status non-electrical quantity data, and business data collected by S1, a generative model is designed to reconstruct low-quality data. Combining data quality and the real-time line loss distribution map output by S2, line loss calculation and anomaly tracing for the transformer substation area are completed. Line loss calculation and anomaly tracing for the transformer substation area are achieved through the following methods: A data reconstruction model based on an improved conditional generative adversarial network (CGAN) is constructed. Using complete, high-quality multi-source data of the transformer substation area as training samples, after adversarial training between the generator and discriminator, the generator outputs a reconstructed high-confidence complete dataset based on low-quality data. Then, based on this reconstructed complete dataset and combined with the real-time line loss distribution map output by S2, the branch loss superposition method is used to calculate the loss value of each branch layer by layer from the end user node to the distribution transformer outlet. The total loss of the transformer substation area and the loss results of each branch component are obtained by superposition and summarization, completing the line loss calculation. The line loss contribution of each branch is calculated, sorted from high to low contribution, and combined with multi-source data for reverse gradient tracing to locate branches and nodes with abnormal line loss and match the anomaly causes, completing the anomaly tracing.

[0090]

[0091] In the formula, The contribution of the line loss to the k-th branch. Let be the active power loss value of the k-th branch. This represents the total active power loss value of the transformer area. The higher the contribution, the greater the impact of the branch on line loss anomalies.

[0092] S5: Compare the line loss calculation and anomaly tracing results of S4 with the active control effect of the digital twin in S3, correct the model parameters, and iteratively optimize the loss reduction strategy to continuously reduce the line loss rate of the transformer area. In step S5, when comparing the line loss calculation and anomaly tracing results of S4 with the active control effect of the digital twin in S3, a multi-dimensional evaluation system for the line loss reduction effect of the transformer area is constructed. The evaluation indicators of this system include line loss rate, branch loss deviation, node voltage qualification rate, three-phase imbalance, and equipment operating status. Using the line loss calculation and anomaly tracing results of S4 as measured feedback values ​​and the active control simulation results of the digital twin in S3 as predicted values, the line loss rate deviation, branch loss deviation, and node voltage deviation are calculated. Then, the adaptive moment estimation Adam algorithm is used to correct the core trainable parameters of the dynamic graph neural network, user electricity consumption behavior prediction model, digital twin, and data reconstruction generative model based on the above deviation values. At the same time, based on the deviation tracing results, the optimization target weight, constraint boundary, and execution cycle of the loss reduction control strategy are iteratively optimized to continuously reduce the line loss rate of the transformer area. The expression for iteratively updating the model parameters is as follows:

[0093]

[0094]

[0095]

[0096]

[0097]

[0098]

[0099] In the formula, Let be the gradient of the objective function at time t. This is the set of trainable parameters for the model. Let be the loss function with the objective of minimizing the deviation of the line loss rate. This is a first-moment estimate of the gradient. For the second moment estimation of the gradient, The exponential decay rate is estimated by moments. These are the deviation correction values ​​for the first and second moments, respectively. For adaptive learning rate, To prevent smooth terms with a denominator of 0, These are the model parameters updated at time t.

Claims

1. A method for accurately reducing line loss rate in low-voltage distribution areas based on multi-technology integration, characterized in that, Includes the following steps: S1: Collect high-frequency electrical quantity data, environmental and equipment status non-electrical quantity data, and business data of electrical equipment in the low-voltage distribution area. Then, perform deduplication, outlier removal, and standardization preprocessing on each collected data. S2: Construct a dynamic graph model using the high-frequency electrical quantity data from S1, introduce a dynamic graph neural network and design a joint loss function, and train to obtain the optimal topology and real-time line loss distribution map; S3: Integrate the high-frequency electrical quantity data, non-electrical quantity data and equipment parameters of S1 to construct a digital twin, and realize multi-time scale collaborative optimization based on the digital twin to dynamically adjust the equipment status; S4: Integrate the high-frequency electrical quantity data, environmental and equipment status non-electrical quantity data, and business data collected by S1, design a generative model to reconstruct low-quality data; combine data quality and the real-time line loss distribution map output by S2 to complete the calculation of transformer area line loss and anomaly tracing. S5: Compare the line loss calculation and anomaly tracing results in S4 with the active control effect of the digital twin in S3, correct the model parameters and iteratively optimize the loss reduction strategy to continuously reduce the line loss rate of the transformer area.

2. The method for accurately reducing line loss rate in low-voltage distribution areas based on multi-technology integration as described in claim 1, characterized in that, In step S1, the high-frequency electrical quantity data includes the three-phase instantaneous voltage, three-phase instantaneous current, active power, reactive power, power factor, three-phase unbalance, harmonic content, zero-sequence current, and line impedance value of the distribution transformer outlet and each branch. The sampling frequency is not less than 1kHz, and the sampling timestamps of all nodes are kept synchronized. The non-electrical data of the environment and equipment status include the temperature and humidity of the transformer area, the level of environmental pollution, the temperature of the equipment casing, the temperature of the iron core, the insulation resistance value, the vibration amplitude of the equipment, the switch opening and closing status, the service life of the equipment, and the historical defect records of the equipment. The business data includes transformer substation relationship files, user electricity categories, installed capacity, historical daily and hourly meter reading data, peak and off-peak electricity consumption records, records of illegal electricity use and electricity theft investigations, historical line loss statistics for the transformer substation, and equipment maintenance operation records.

3. The method for accurately reducing line loss rate in low-voltage distribution areas based on multi-technology integration as described in claim 1, characterized in that, In step S2, a dynamic graph model is constructed: the distribution transformers, branch boxes, metering boxes, and user energy meters in the low-voltage distribution area are used as graph nodes, the low-voltage power supply lines and electrical connections between nodes are used as graph edges, the high-frequency electrical quantity data collected at each time moment are used as node features, and the measured impedance value of the line is used as the edge weight. The single-time graph structure of each time moment is constructed according to a fixed sampling time step, and the graph structures of all time moments are connected in series according to the time sequence to form a spatiotemporal dynamic graph model.

4. The method for accurately reducing line loss rate in low-voltage distribution areas based on multi-technology integration as described in claim 1, characterized in that, In step S2, when training the optimal topology and real-time line loss distribution map, the time-continuous dynamic graph model is input into the dynamic graph neural network to extract the spatial electrical correlation and time series evolution features between nodes, and output the predicted value of the topology adjacency matrix and the predicted value of line loss for each branch. A joint loss function including topology reconstruction loss, line loss regression loss and L2 regularization loss is designed, and weight coefficients are set for the three loss terms, with the sum of the weight coefficients being 1. The model is iteratively trained with the goal of minimizing the joint loss function. After the model loss value converges to a preset threshold, the optimal identification topology of the transformer area and the real-time line loss distribution map of all branches are output.

5. The method for accurately reducing line loss rate in low-voltage distribution areas based on multi-technology integration as described in claim 1, characterized in that, Step S3 includes the following sub-steps: S31: Based on the electrical topology drawings, equipment nameplate parameters, historical operating data and power grid dispatching rules of the low-voltage distribution area, construct a basic digital twin model of the distribution area and complete the four-dimensional accurate mapping between the virtual model and the physical entity of the distribution area. S32: Extract electricity consumption behavior features from user service data in S1, construct a user electricity consumption behavior prediction model, and output load prediction values ​​for users at multiple time scales; input real-time high-frequency electrical quantity data, non-electrical quantity data, and service data from S1 into the basic digital twin model, combine the user load prediction values ​​to complete model calibration and update, and construct a digital twin of the distribution area. S33: Based on the digital twin of the distribution area, with the minimization of the line loss rate of the distribution area as the core objective, it divides the optimization into three levels: long-term scale, short-term scale, and real-time scale, and constructs the corresponding optimization objective function for each scale, and matches the corresponding adjustable equipment type and safe operation constraints. S34: An improved particle swarm optimization algorithm is used to solve the optimization objective function at each time scale in parallel to obtain the optimal control strategy for each adjustable device. The control strategy is input into a digital twin simulation for verification. After confirming that there are no safety risks, control commands are sent to the physical devices in the distribution area to dynamically adjust the operating status of the devices.

6. The method for accurately reducing line loss rate in low-voltage distribution areas based on multi-technology integration as described in claim 5, characterized in that, In step S32, the specific operations for constructing the user electricity consumption behavior prediction model are as follows: preprocess the user's historical time-series electricity consumption data, remove abnormal data and align the time series, extract features such as daily load curve, peak-valley electricity consumption difference, load factor, and industry electricity consumption characteristics, and integrate them into the user electricity consumption feature vector; construct the prediction model using a bidirectional long short-term memory network Bi-LSTM, input the user electricity consumption feature vector into the model according to the time series, train the model with historical measured electricity consumption data as labels, and output the predicted values ​​of user electricity load at multiple time scales in the future after training.

7. The method for accurately reducing line loss rate in low-voltage distribution areas based on multi-technology integration as described in claim 5, characterized in that, In step S33, the specific operations for constructing the multi-time-scale collaborative optimization objective function are as follows: Three optimization time scales are defined: a long-term optimization period of 24 hours, optimizing the switching positions of the reactive power compensation device and the configuration parameters of the three-phase imbalance regulation device; a short-term optimization period of 15 minutes, optimizing the charging and discharging plans of distributed photovoltaic and energy storage equipment; and a real-time optimization period of 1 minute, optimizing the output commands of the active voltage regulator. With minimizing the total line loss over the entire cycle as the core objective, voltage qualification rate, three-phase imbalance, and equipment adjustment times are introduced as auxiliary objectives. Weight coefficients for each sub-objective are set to construct a multi-objective comprehensive optimization objective function. Simultaneously, safe operation constraints such as node voltage, three-phase imbalance capacity, reactive power compensation capacity, energy storage charging and discharging power, and state of charge are set.

8. The method for accurately reducing line loss rate in low-voltage distribution areas based on multi-technology integration as described in claim 1, characterized in that, In step S4, during the calculation of line loss and anomaly tracing in the transformer substation area, a data reconstruction model based on an improved conditional generative adversarial network (CGAN) is constructed. Using complete, high-quality multi-source data of the transformer substation area as training samples, and after adversarial training between the generator and discriminator, the generator outputs a reconstructed high-confidence complete dataset based on low-quality data. Then, based on this reconstructed complete dataset and combined with the real-time line loss distribution map output in S2, the branch loss superposition method is used to calculate the loss value of each branch layer by layer from the end user node to the distribution transformer outlet. The total loss of the transformer substation area and the loss results of each branch component are obtained by superposition and summarization, completing the line loss calculation. The line loss contribution of each branch is calculated and sorted from high to low contribution. Combined with reverse gradient tracing of multi-source data, the branches and nodes with abnormal line loss are located and the anomaly causes are matched, completing the anomaly tracing.

9. The method for accurately reducing line loss rate in low-voltage distribution areas based on multi-technology integration as described in claim 1, characterized in that, In step S5, when comparing the line loss calculation and anomaly tracing results from S4 with the active control effect of the digital twin in S3, a multi-dimensional evaluation system for the line loss reduction effect of the transformer area is constructed. The evaluation indicators of this system include line loss rate, branch loss deviation, node voltage qualification rate, three-phase imbalance, and equipment operating status. The line loss calculation and anomaly tracing results from S4 are used as measured feedback values, and the active control simulation results of the digital twin in S3 are used as predicted values. The line loss rate deviation, branch loss deviation, and node voltage deviation between the two are calculated. Then, the adaptive moment estimation Adam algorithm is used to correct the core trainable parameters of the dynamic graph neural network, user electricity consumption behavior prediction model, digital twin, and data reconstruction generative model based on the above deviation values. At the same time, based on the deviation tracing results, the optimization target weight, constraint boundary, and execution cycle of the loss reduction control strategy are iteratively optimized to continuously reduce the line loss rate of the transformer area.

10. A system for precisely reducing line loss rate in low-voltage distribution areas based on multi-technology integration, characterized in that, The system includes a multi-source data preprocessing virtual module, a dynamic graph topology and line loss distribution virtual module, a digital twin collaborative optimization virtual module, a line loss calculation and anomaly tracing virtual module, and a closed-loop iterative optimization virtual module. The multi-source data preprocessing virtual module is used for the collection and standardized preprocessing of multi-source data from the transformer area. The dynamic graph topology and line loss distribution virtual module is used for constructing a dynamic graph model of the transformer area, topology identification, and real-time line loss distribution output. The digital twin collaborative optimization virtual module is used for constructing a digital twin of the transformer area and multi-timescale loss reduction optimization control. The line loss calculation and anomaly tracing virtual module is used for low-quality data reconstruction, transformer area line loss calculation, and anomaly tracing and location. The closed-loop iterative optimization virtual module is used for comparing loss reduction effects, correcting model parameters, and iteratively optimizing loss reduction strategies.