Power line loss lean management method and system based on data driving and dynamic threshold
By constructing a data-driven correlation model and dynamic threshold to identify power line loss anomalies, and combining it with an economic transmission power model, the problems of rigid thresholds and separation of diagnosis in existing power line loss management are solved, enabling accurate determination and continuous optimization of high-loss types.
Patent Information
- Application Number
- CN202511749989.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-26
- Publication Date
- 2026-02-24
AI Technical Summary
Existing power line loss management methods are inadequate in terms of accuracy and timeliness. Problems such as rigid threshold settings, separation of cause diagnosis and treatment, and disconnect between treatment and models lead to frequent false alarms and missed alarms, and the treatment measures are not targeted enough.
A data-driven and dynamic threshold-based lean management method for power line losses is constructed. By integrating data on mechanical component wear status, power line temperature changes, and current surge events through an association model, dynamic thresholds are generated and abnormal fluctuations are identified. High-loss types are determined by combining an economic transmission power model, and the governance scheme is optimized through a multi-objective search method library.
It enables accurate identification and management of high-loss types, improves the precision and efficiency of line loss management, reduces misjudgments and slow response of management measures, and ensures that the model adapts to dynamic changes in the power grid.
Smart Images
Figure CN121562913A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power line loss technology, specifically to a data-driven and dynamic threshold-based lean management method and system for power line loss. Background Technology
[0002] Power line loss is an unavoidable loss during the transmission of electrical energy in the power grid, and its management level directly measures the operational efficiency and economic benefits of power grid companies. Currently, the industry generally adopts a "four-part" management model of "zoning, voltage, line, and distribution area," relying on platforms such as electricity consumption information collection systems for daily monitoring and management of line losses.
[0003] However, with the increasing complexity of distribution network structures, especially the high penetration rate of distributed energy resources, existing line loss management methods face severe challenges in terms of accuracy and timeliness, mainly reflected in:
[0004] Rigid threshold settings and poor adaptability: Existing methods mostly set abnormal line loss thresholds based on static historical statistical values or manual experience, failing to effectively couple with the real-time operating status of the power grid (such as load fluctuations, equipment operating conditions, distributed power generation output, etc.). Such rigid thresholds cannot accurately capture the real impact of gradual or sudden changes in operating conditions on line losses, leading to frequent false alarms and missed alarms, and seriously insufficient accuracy in anomaly detection.
[0005] The causes and diagnoses are separate, resulting in low efficiency in tracing the source: the causes of technical and management line losses are intertwined, but existing diagnostic models rely on manual retrieval of scattered multi-data information (such as equipment status, operating electrical quantities, environmental data, etc.) across systems and platforms. Due to the lack of an inherent correlation analysis and collaborative diagnostic mechanism for multi-source information, it is difficult to quickly clarify whether the root cause of high losses is a "technical" problem such as equipment condition deterioration or a "management" problem such as electricity theft or metering failure, resulting in weak targeted governance measures and slow response.
[0006] The governance measures are disconnected from the model, lacking closed-loop optimization: Existing governance measures are often independent of the line loss analysis model, and the effect data after governance is not fed back into the model for iterative optimization. This causes the model to be unable to evolve with the dynamic changes of the power grid, and its analytical accuracy and decision-making effectiveness gradually decline over long-term operation, forming a vicious cycle of "governance-failure-re-governance".
[0007] Therefore, there is an urgent need in this field for a lean line loss management method and system that can adapt to dynamic changes in the power grid, automatically and accurately distinguish high-loss types, and achieve self-learning and continuous optimization. Summary of the Invention
[0008] To address the aforementioned problems, this invention provides a data-driven and dynamic threshold-based lean management method and system for power line loss.
[0009] To achieve the above objectives, the present invention provides the following technical solution:
[0010] In a first aspect, the present invention provides a method for lean management of power line losses based on data-driven and dynamic thresholds, comprising the following steps:
[0011] Step 1. Based on the interrelationship between the wear state of mechanical components, the temperature change of power lines, and current surge events, construct a correlation model of the factors affecting power line losses;
[0012] Step 2. Using power distribution lines or transformer substations as the smallest unit, synchronously input real-time data on the wear status of mechanical components, changes in power line temperature, and current surge events into the correlation model, output the theoretical line loss rate for the day, and accumulate the theoretical line loss rate day by day to obtain a time series.
[0013] Step 3. Identify abnormal fluctuations in the time series to obtain a dual-track time series; identify the distribution pattern of the dual-track time series, extract the theoretical line loss rate corresponding to the mutation point as a dynamic threshold, and generate the optimal interval with the economic transmission power model as a constraint; the economic transmission power model is a transmission loss power calculation function constructed based on the equivalent resistance change value output by the correlation model.
[0014] Step 4. Compare the measured line loss rate with the optimal range. If it exceeds the limit, compare the difference between the measured line loss rate and the theoretical line loss rate with the difference between the theoretical line loss rate and the dynamic threshold. If the two deviations are in the same direction, it is determined to be a technically high loss; if the two deviations are in opposite directions, it is determined to be a managerially high loss.
[0015] Step 5. When the judgment result is high management loss, an automatic work order is triggered to start a multi-objective search method library to call the optimal solution for high technical loss; after implementation, the newly added running data is used to incrementally update the correlation model and dynamic threshold.
[0016] Furthermore, when constructing the correlation model of the influencing factors of power line loss, a deep learning algorithm is used to jointly train the mechanical component wear status data, power line temperature change data, and current impact event data. The training process is supervised by historical line loss data, and regularization technology is applied to optimize the network parameters, ultimately forming a correlation model that combines data learning ability and conformity to physical laws.
[0017] The processing steps of the correlation model are as follows: First, the input layer synchronously receives preprocessed data on the wear status of mechanical components, temperature changes in power lines, and current surge events. The wear status data of mechanical components undergoes time-series analysis to output fluctuation characteristic values reflecting changes in the contact state of connection points. Based on the power grid topology, the spatial correlation effect of conductor thermal conduction is modeled using a graph convolutional layer. The power line temperature change data is used to output the equivalent resistance change value. For the current surge event data, considering its transient characteristics, waveform distortion features are extracted using a wavelet decomposition layer, and a quantitative value representing the surge intensity is output based on the modulus of the wavelet coefficients. The three types of features are input into a dynamic gated fusion layer. Through a dynamic weight allocation strategy, the fluctuation characteristic value, the equivalent resistance change value, and the quantitative value representing the surge intensity are weighted and summed to generate the theoretical line loss rate. The dynamic weight allocation strategy is based on historical line loss data to inversely calibrate the actual contribution ratio of each data source to line loss and adjusts the weight parameters in real time with newly added operational data.
[0018] Furthermore, in the dynamic weight allocation strategy, when calibrating the actual contribution ratio of each data source to line loss based on historical line loss data, a grey relational analysis method is used. First, a reference data sequence is determined, i.e., the historical actual line loss rate sequence. Then, the grey relational degree between the mechanical component wear status data, power line temperature change data, and current impact event data and the reference data sequence is calculated. The larger the relational degree, the greater the impact of the data source on line loss. Based on the calculated grey relational degree, the initial weight of each data source is determined. In subsequent operation, as new operating data is continuously accumulated, the grey relational degree is recalculated periodically using the latest data, thereby dynamically adjusting the weight parameters to ensure that the correlation model can reflect the actual impact of each factor on line loss in real time, improving the accuracy and adaptability of the model.
[0019] Furthermore, to ensure that the output theoretical line loss rate always remains consistent with the actual physical relationship between current and resistance, the theoretical line loss rate, along with the equivalent resistance change and the real-time monitored current value, are input into a constraint function to calculate and output a theoretical line loss rate that conforms to the physical law of conductor heating; where the constraint function is... ,L theory I represents the corrected theoretical line loss rate, I represents the real-time monitored line current value, and R represents the line loss rate. eq The equivalent resistance change value and P output by the correlation model base The reference power is the rated transmission power of the line or the total power supply power of the area.
[0020] Further, the steps for identifying abnormal fluctuations in the time series are as follows: calculate the absolute difference between the theoretical line loss rates of adjacent periods; when the absolute difference exceeds the theoretical line loss fluctuation range under normal operating conditions, mark the data point as an abnormal fluctuation point; retain all original data points and associate them with abnormal marking information to form a dual-track time series containing complete original data and abnormal marking status; the dual-track time series serves as input data for distribution pattern identification.
[0021] Furthermore, when identifying anomalous fluctuations in a time series, in addition to calculating the absolute difference between the theoretical line loss rates of adjacent periods, wavelet analysis is used to decompose the time series into multiple scales. This extracts feature information at different scales to identify anomalous fluctuation points that could not be detected by comparing absolute differences, thus accurately identifying anomalous fluctuation points. Wavelet analysis can decompose the time series into different frequency components, analyzing the abrupt changes in high-frequency components and the trend changes in low-frequency components separately, which helps to capture more subtle anomalous fluctuation features. After completing the multi-scale decomposition, by setting appropriate thresholds, the coefficients at each scale are thresholded, and the time series is reconstructed, thereby obtaining more accurate anomalous fluctuation point marking results, providing a more reliable basis for subsequent distribution pattern analysis.
[0022] Furthermore, the step of extracting the dynamic threshold and generating the optimal interval includes: based on the distribution pattern analysis of the dual-track time series, when a definite abrupt change is detected in the statistical distribution pattern of the theoretical line loss rate in a continuous normal data segment, the measured value of the theoretical line loss rate corresponding to the abrupt change point is extracted as the dynamic threshold; subsequently, the dynamic threshold is input into the economic transmission power model, and the economic transmission power model calculates in real time the upper and lower limits of the technically permissible interval that meets the minimum transmission loss cost, i.e., the optimal interval, based on the current power grid topology, load period and line current carrying capacity.
[0023] Furthermore, the process of performing distribution pattern analysis on the dual-track time series is as follows: the dual-track time series is divided into two equal sub-data segments; the mean and variance of the theoretical line loss rate of the two sub-data segments are calculated respectively; if the absolute difference between the mean of the two sub-data segments exceeds the maximum value of the historical normal fluctuation record of the distribution line or transformer area, and the variance of the two sub-data segments shows a continuous increase or decrease in the same direction, then an identifiable pattern change is determined to have occurred, and the change point is located as the boundary point between the two sub-data segments; the maximum value of the historical normal fluctuation record comes from the historical statistical extreme value of the theoretical line loss fluctuation range under normal operating conditions of the equipment.
[0024] Furthermore, the economic transmission power model constructs a transmission loss power calculation function based on the equivalent resistance change value generated by the correlation model, taking distribution lines or transformer areas as objects; simultaneously, it obtains the time-of-use electricity price coefficient corresponding to the current load period and the line current margin determined through real-time monitoring; inputs the dynamic threshold into the transmission loss power calculation function, and combines it with the real-time power grid topology to establish a line loss economic optimization equation with the dynamic threshold as the baseline line loss rate, the current margin as the upper limit constraint, and the time-of-use electricity price coefficient as the economic weight coefficient; the dynamic threshold is the benchmark for the system to judge the data, and the economic transmission power model uses this benchmark to determine the optimal range; by iteratively solving the equation, the operating point that minimizes the transmission loss cost under the current margin constraint is obtained, and the line loss rate corresponding to this operating point is extracted as the lower limit of the technically permissible range; at the same time, the theoretical line loss rate corresponding to the current margin boundary current is calculated as the upper limit of the technically permissible range.
[0025] Furthermore, in the aforementioned economic transmission power model, the impact of distributed generation on the power grid topology is considered when constructing the transmission loss power calculation function; the transmission loss power calculation function is structured as C(I)=I 2 ×R eq ×C price I represents line current, R eq The equivalent resistance change value and C generated by the correlation model price This represents the time-of-use electricity price factor corresponding to the current load period; the main constraint of the transmission loss power calculation function comes from the line's current carrying capacity, i.e.: I ≤ I max , where I max This model represents the upper limit of the current corresponding to the current carrying capacity of the line, determined through real-time monitoring. With the widespread application of distributed generation (DG) in distribution networks, its connection location and capacity affect the power flow distribution and losses of the lines. Therefore, when constructing the model, based on the real-time topology of the power grid, including the connection point and output power of the DG, the current distribution of each branch is accurately calculated, thereby obtaining the transmission loss power. Simultaneously, in the optimization equations, the output of the DG is treated as an adjustable variable. Under the condition of satisfying power grid security constraints, by optimizing the output of the DG, the transmission loss cost is further reduced, making the economic transmission power model more closely reflect the actual power grid operation and improving its practicality and economy.
[0026] Further, in step 4, the process of comparing the measured line loss rate with the optimal interval is as follows: when the measured line loss rate exceeds the boundary of the optimal interval, a dual-deviation direction collaborative diagnosis mechanism is activated; firstly, the first difference between the measured line loss rate and the theoretical line loss rate output by the current cycle correlation model is calculated, and the second difference between the theoretical line loss rate and the dynamic threshold of the median of the optimal interval is calculated independently; based on the collaborative relationship between the first difference and the second difference, the nature of high loss is determined: if the first difference and the second difference are both positive or both negative, it is determined to be a technical high loss, indicating that the abnormal line loss originates from the equipment operating state deviating from the optimal benchmark set by the economic transmission model; if the first difference and the second deviation direction are opposite, it is determined to be a management high loss, indicating that the abnormal line loss originates from the theoretical line loss rate itself deviating from the historical normal distribution pattern represented by its dynamic threshold, pointing to non-technical control factors.
[0027] Furthermore, when invoking the optimal solution from the multi-objective search method library for technically high-loss initiation, multiple optimization objectives are comprehensively considered, including reducing line loss rate, improving voltage quality, balancing three-phase load, and reducing equipment investment costs. The multi-objective search method library pre-stores various advanced optimization algorithms, such as genetic algorithms, particle swarm optimization algorithms, and simulated annealing algorithms. For different high-loss situations and optimization objectives, the most suitable algorithm is intelligently selected for solution. During the solution process, by establishing the priority and weight coefficients of the objective function, the various optimization objectives are coordinated and balanced, ultimately obtaining a comprehensive optimal governance solution. Simultaneously, to improve the search efficiency and convergence speed of the algorithm, the algorithm parameters are adaptively adjusted, dynamically modifying the parameters based on the search results of each iteration, ensuring a high-quality solution is obtained within a limited computation time, providing strong support for the effective governance of technically high losses.
[0028] Secondly, the present invention provides a data-driven and dynamic threshold-based lean management system for power line loss, used to implement the method, comprising:
[0029] Data Synchronization Injection Unit: Taking power distribution lines or transformer substations as the smallest unit, real-time data on the wear status of mechanical components, changes in power line temperature, and current surge events are synchronously injected into the correlation model; the correlation model is constructed through the interrelationships between the wear status of mechanical components inside the power detection equipment, changes in power line temperature, and current surge events.
[0030] Time series construction unit: The theoretical line loss rate output by the correlation model is accumulated day by day to obtain the time series;
[0031] Dynamic threshold unit: Identify abnormal fluctuations in the time series to obtain a dual-track time series, identify the distribution pattern of the dual-track time series, extract the theoretical line loss rate corresponding to the mutation point as the dynamic threshold, and generate the optimal interval with the economic transmission power model as a constraint.
[0032] Collaborative diagnostic unit: Compares the measured line loss rate with the optimal range. If it exceeds the limit, it compares the difference between the measured line loss rate and the theoretical line loss rate with the difference between the theoretical line loss rate and the dynamic threshold. If the two deviations are in the same direction, it is judged as technical high loss; if the two deviations are in opposite directions, it is judged as management high loss.
[0033] Management Update Unit: When the judgment result is high management loss, an automatic work order is triggered, and the optimal solution is called from the multi-objective search method library for high technical loss; after implementation, the newly added running data is used to incrementally update the correlation model and dynamic threshold.
[0034] The present invention has the following technical effects: (1) The present invention constructs a dynamic correlation model by integrating multi-source operating condition data and coupling economic transmission power constraints, thereby realizing the adaptive optimal line loss interval generation based on the real-time state of the power grid, accurately capturing the real-time impact of sudden change points in operating conditions on line loss, improving the accuracy of high loss type determination, and solving the problem of misjudgment caused by environmental interference.
[0035] (2) This invention constructs a fully automated diagnostic process by comparing measured values with theoretical values and theoretical values with dynamic thresholds in two deviation directions, replacing the reliance on manual experience and cross-platform data retrieval, thereby improving the efficiency of line loss root cause analysis and improving the automatic triggering rate of management high loss work orders.
[0036] (3) This invention links the management high-loss automatic work order system with the technical high-loss multi-objective solution library, and the newly added operation data after the treatment is fed back to the associated model and dynamic threshold for incremental updates in real time, so that the model continuously approximates the real power grid state; improves the pain point of the disconnect between measures and effects and the rigidity of the model in traditional treatment, and effectively improves the implementation of loss reduction schemes. Attached Figure Description
[0037] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation
[0038] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0039] like Figure 1 As shown, a data-driven and dynamic threshold-based lean management method for power line losses includes the following steps:
[0040] Step 1: By deeply analyzing the interrelationships between the wear state of internal mechanical components of the power detection equipment, the temperature change of the power line, and excessive current surge events, a correlation model of the influencing factors of power line loss is constructed. Historical line loss data is used for supervision, and regularization techniques are applied to optimize network parameters to train a dynamic correlation model between the wear state data of mechanical components, the temperature change data of the power line, and the current surge event data and the theoretical line loss rate.
[0041] Specifically: First, time-series analysis is performed on the wear status data of mechanical components to extract fluctuation characteristic values reflecting changes in the contact state of connection points. This process uses time-series analysis to mine a large amount of historical wear data to accurately capture its fluctuation patterns. Simultaneously, based on the physical law of conductor material resistance changing with temperature, the electric field line temperature change data is converted into corresponding resistance change values. For current impact event data, considering its transient characteristics, waveform distortion features are extracted through wavelet decomposition, and a quantized value representing the impact intensity is output based on the modulus of the wavelet coefficients, where the quantized value of impact intensity = max(|W( a,b)|); where W(a,b) is the wavelet coefficient at a specific scale a within the impact time period, and b is the time position, which depends on the deep understanding and feature extraction of the current impact waveform; finally, through a dynamic weight allocation strategy, the fluctuation characteristic value, the resistance change value and the quantized value of the impact intensity are weighted and summed, corrected, and a theoretical line loss rate is generated; this dynamic weight allocation strategy is based on the inversion calibration of the actual contribution ratio of each data source to the line loss based on historical line loss data, and the weight parameters are adjusted in real time as new operating data is added, so that the correlation model can dynamically adapt to the actual operating conditions of the power system.
[0042] Step 2: Using power distribution lines or transformer substations as the smallest unit, real-time data on the wear status of mechanical components, temperature changes, and current surge events are synchronously injected into the constructed and trained correlation model to output the theoretical line loss rate for the day. The theoretical line loss rate is then accumulated daily into a time series to provide basic data for subsequent analysis and diagnosis.
[0043] Specifically: First, the input layer of the correlation model synchronously acquires mechanical component wear status data, power line temperature change data, and current surge event data according to the acquisition cycle. The mechanical component wear status data undergoes time-series analysis to output fluctuation characteristic values reflecting changes in the contact state of connection points. Based on the power grid topology, the spatial correlation effect of conductor heat conduction is modeled through a graph convolutional layer. The power line temperature change data is output as the equivalent resistance change value. For the current surge event data, considering its transient characteristics, waveform distortion features are extracted through a wavelet decomposition layer, and a quantitative value representing the surge intensity is output based on the modulus of the wavelet coefficients. These three types of features are input into a dynamic gated fusion layer. Through a dynamic weight allocation strategy, the fluctuation characteristic value, the equivalent resistance change value, and the quantitative value representing the surge intensity are weighted and summed to generate a predicted theoretical line loss rate. To correct the generated theoretical line loss rate, the predicted theoretical line loss rate, along with the resistance change value and the real-time monitored current value, are input into a constraint function to calculate and output the theoretical line loss rate that conforms to the physical laws of conductor heating. The constraint function is... ,L theory I represents the corrected theoretical line loss rate, I represents the real-time monitored line current value, and R represents the line loss rate. eq The equivalent resistance change value and P output by the correlation model base The reference power is the rated transmission power of the line or the total power supply power of the area.
[0044] Step 3: Identify abnormal fluctuations in the time series to obtain a dual-track time series; identify the distribution pattern of the dual-track time series, extract the theoretical line loss rate corresponding to the mutation point as a dynamic threshold, and generate the optimal interval with the economic transmission power model as a constraint; the economic transmission power model is a transmission loss power calculation function constructed based on the equivalent resistance change value output by the correlation model.
[0045] Specifically: First, considering the possible abnormal fluctuations in actual operating data, abnormal fluctuations are identified in the time series. This is done by calculating the absolute difference between the theoretical line loss rates of adjacent periods and comparing it with the theoretical line loss fluctuation range under normal equipment operation. When the absolute difference exceeds this range, the data point is marked as an abnormal fluctuation point. Based on retaining all original data points, the abnormal marking information is associated to form a dual-track time series containing complete original data and abnormal marking status, providing more accurate and reliable input data for subsequent distribution pattern identification.
[0046] Next, a distribution pattern analysis is performed based on the dual-track time series: the dual-track time series is divided into two equal sub-data segments; the mean and variance of the theoretical line loss rate for each sub-data segment are calculated; if the absolute difference between the mean of the first sub-data segment and the mean of the second sub-data segment exceeds the maximum value of the historical normal fluctuation record for the distribution line or transformer area, and the variance of the first and second sub-data segments shows a continuous increase or decrease in the same direction, then an identifiable morphological change is determined, and this change point is located as the boundary point between the first and second sub-data segments; if this change point is an abnormal marker state, when an identifiable change in the statistical distribution pattern of the theoretical line loss rate is detected in a continuous normal data segment, the measured value of the theoretical line loss rate corresponding to the change point is extracted as a dynamic threshold; to more accurately identify abnormal fluctuation points, wavelet analysis is combined to decompose the time series into multiple scales and extract feature information at different scales; the maximum value of the historical normal fluctuation record comes from the historical statistical extreme value of the theoretical line loss fluctuation range under normal equipment operation conditions;
[0047] Then, an economic transmission power model is constructed: based on the equivalent resistance change value generated by the correlation model, a transmission loss power calculation function is constructed with distribution lines or transformer areas as objects. The time-of-use electricity price coefficient corresponding to the current load period and the line current margin determined by real-time monitoring are obtained simultaneously. Combined with the real-time grid topology, an economic optimization equation for line loss C(I)=I is established, with a dynamic threshold as the baseline line loss rate, current margin as the upper limit constraint of current, and time-of-use electricity price coefficient as the economic weight coefficient. 2 ×R eq ×C price I represents line current, R eq The equivalent resistance change value and C generated by the correlation model price Let be the time-of-use electricity price coefficient corresponding to the current load period. By iteratively solving this equation, the operating point that minimizes transmission loss cost under the current margin constraint is obtained. The line loss rate corresponding to this operating point is extracted as the lower limit of the technically permissible range. At the same time, the theoretical line loss rate corresponding to the current margin boundary current is calculated as the upper limit of the technically permissible range, thus forming the optimal range.
[0048] Finally, the dynamic threshold is input into the constructed economic transmission power model. By iteratively solving the above equations, the operating point that minimizes transmission loss cost while satisfying the current margin constraint is identified. The line loss rate corresponding to this operating point is extracted as the lower limit of the technically permissible range. Simultaneously, the theoretical line loss rate corresponding to the current margin boundary current is calculated as the upper limit of the technically permissible range. Using the dynamic threshold as the median, the optimal range is formed. The lower limit of the technically permissible range is: The upper limit of the technical allowable range is: Where C(I) is the transmission loss cost, specifically: I is the line current, R eqC is the equivalent resistance generated by the correlation model. price This represents the current time-of-use electricity price coefficient.
[0049] Step 4: Compare the measured line loss rate with the optimal range. If it exceeds the range, compare the difference between the measured line loss rate and the theoretical line loss rate with the difference between the theoretical line loss rate and the dynamic threshold. If the deviations are in the same direction, it is determined to be a technically high loss; if the deviations are in opposite directions, it is determined to be a managerially high loss. The measured line loss rate is a statistical value calculated based on existing metering system data and is a routine monitoring indicator for the power grid.
[0050] Specifically: The measured line loss rate is compared with the optimal range. When the measured line loss rate exceeds the boundary of the optimal range, a dual-deviation direction collaborative diagnosis mechanism is activated. First, the first difference between the measured line loss rate and the theoretical line loss rate output by the current cycle correlation model is calculated, and the second difference between the theoretical line loss rate and the dynamic threshold of the median of the optimal range is calculated independently. Based on the collaborative relationship between the first and second differences, the nature of the high loss is determined: if the first and second differences are both positive or both negative, it is determined to be a technical high loss, indicating that the abnormal line loss originates from the equipment operating state deviating from the optimal benchmark set by the economic transmission model; if the first and second differences are in opposite directions, it is determined to be a management high loss, indicating that the abnormal line loss originates from the theoretical line loss rate itself deviating from the historical normal distribution pattern represented by its dynamic threshold, pointing to non-technical control factors.
[0051] Step 5: For high management losses, trigger automatic work orders; for high technical losses, initiate a multi-objective search method library to call the optimal solution; after implementation, use the newly added operational data to incrementally update the correlation model and dynamic thresholds.
[0052] Specifically: when the judgment result is high management loss, the system automatically triggers an action signal to start a multi-objective search method library to call the optimal solution for high technical loss (which can be designed according to actual needs); after implementing the corresponding governance measures, the system uses the newly added operating data to incrementally update the correlation model and dynamic threshold, so that the entire power line loss lean management system can continuously adapt to the dynamic changes of the power system and continuously improve management effectiveness and accuracy.
[0053] A data-driven and dynamic threshold-based lean management system for power line losses includes:
[0054] Data Synchronization Injection Unit: Taking power distribution lines or transformer substations as the smallest unit, real-time data on the wear status of mechanical components, changes in power line temperature, and current surge events are synchronously injected into the correlation model; the correlation model is constructed through the interrelationships between the wear status of mechanical components inside the power detection equipment, changes in power line temperature, and current surge events.
[0055] Time series construction unit: The theoretical line loss rate output by the correlation model is accumulated day by day to obtain the time series;
[0056] Dynamic threshold unit: Identify abnormal fluctuations in the time series to obtain a dual-track time series, identify the distribution pattern of the dual-track time series, extract the theoretical line loss rate corresponding to the mutation point as the dynamic threshold, and generate the optimal interval with the economic transmission power model as a constraint.
[0057] Collaborative diagnostic unit: Compares the measured line loss rate with the optimal range. If it exceeds the limit, it compares the difference between the measured line loss rate and the theoretical line loss rate with the difference between the theoretical line loss rate and the dynamic threshold. If the two deviations are in the same direction, it is judged as technical high loss; if the two deviations are in opposite directions, it is judged as management high loss.
[0058] Management Update Unit: When the judgment result is high management loss, an automatic work order is triggered, and the optimal solution is called from the multi-objective search method library for high technical loss; after implementation, the newly added running data is used to incrementally update the correlation model and dynamic threshold.
[0059] Example:
[0060] This embodiment of the power line loss lean management system integrates data-driven and dynamic threshold technologies. It primarily relies on the online power grid platform and other specialized systems, such as the concurrent power consumption and line loss system, to acquire multi-dimensional structured data on grid topology, equipment attributes, and grid operating status. Simultaneously, to achieve real-time monitoring and accurate analysis of factors influencing power line losses, the system also integrates real-time sensor data from power detection equipment, including mechanical component wear status, power line temperature changes, and current surge events. For example, temperature data is collected in real-time by temperature sensors installed at key nodes of distribution lines, with a sampling frequency of once per hour, ensuring timely capture of the impact of temperature changes on line resistance, which is then reflected in fluctuations in line loss. For monitoring the wear status of equipment components, the system utilizes the equipment's own condition monitoring system to periodically output fluctuation characteristic values reflecting changes in the contact status of connection points. These characteristic values, after time-series analysis, can be used to characterize the long-term impact trend of equipment health status on line loss.
[0061] The core of the system lies in constructing a correlation model of factors affecting power line losses. This model takes distribution lines or transformer substations as the smallest unit and comprehensively considers the interrelationships among equipment component wear status data, power line temperature change data, and current surge event data. Specifically, after time-series analysis, the equipment component wear status data outputs fluctuation characteristic values reflecting changes in the contact state of connection points. Based on the power grid topology, the system models the spatial correlation effect of conductor thermal conduction through graph convolutional layers, outputting equivalent resistance change values from power line temperature change data. For current surge event data, considering its transient characteristics, waveform distortion features are extracted through wavelet decomposition layers, and a quantitative value representing the surge intensity is output based on the modulus of the wavelet coefficients. The system sets a current safety threshold of 1.2 times the rated current. When the detected current surge intensity reaches or exceeds this threshold, the system calculates the quantitative value of the surge intensity based on the wavelet coefficients and time at a specific scale within the surge time period.
[0062] The correlation model uses a dynamic weighting strategy to weight and sum the three feature values mentioned above to generate a theoretical line loss rate. This dynamic weighting strategy is based on historical line loss data to calibrate the actual contribution ratio of each data source to line loss, and adjusts the weight parameters in real time as new operational data is added. Specifically, the initial weights are determined using grey relational analysis, and the grey relational degree is periodically recalculated using the latest data during subsequent operation, thereby dynamically adjusting the weight parameters. For example, initially, grey relational analysis determines the weights of equipment component wear status data to be 30%, power line temperature change data to 40%, and current surge event data to 30%. As system operational data accumulates, if, during a certain period, recalculation reveals a significant increase in the impact of power line temperature changes on line loss, the weights of temperature change data are adjusted accordingly to 45%, equipment component wear status data to 25%, and current surge event data to 30%, ensuring that the correlation model can accurately reflect the actual impact of each factor on line loss in real time.
[0063] Based on real-time data acquisition and preprocessing, the system takes power distribution lines or transformer substations as units, inputs data from the same acquisition period into the correlation model, outputs the theoretical line loss rate corresponding to that period, and stores it in chronological order day by day to form an initial time series. For example, for a certain 10kV power distribution line, assuming that it generates a theoretical line loss rate data point every day, after a period of operation, it accumulates to form an initial time series containing data from several days.
[0064] The system identifies abnormal fluctuations in the initial time series; it calculates the absolute difference between the theoretical line loss rates of adjacent periods and compares this difference with the theoretical line loss fluctuation range under normal operating conditions. The theoretical line loss fluctuation range under normal operating conditions is determined based on historical data statistical analysis. For example, by analyzing the historical data of this 10kV distribution line over the past six months, it is found that the daily fluctuation range of its theoretical line loss rate is usually within ±0.3%. When the difference between the theoretical line loss rates of two adjacent days exceeds this range, the data point is marked as an abnormal fluctuation point. For example, if the theoretical line loss rate of a certain day is 2.8%, while the previous day was 3.5%, the difference is 0.7%, which exceeds the normal fluctuation range, then the data for that day is marked as an abnormal fluctuation point.
[0065] Simultaneously, the system combines wavelet analysis to perform multi-scale decomposition of the time series, extracting feature information at different scales to more accurately identify abnormal fluctuation points. Wavelet analysis decomposes the time series into different frequency components, analyzing abrupt changes in high-frequency components and trend changes in low-frequency components separately. By setting appropriate thresholds, the coefficients at each scale are thresholded to reconstruct the time series, resulting in more accurate abnormal fluctuation point labeling results. For example, after wavelet decomposition of the time series of the aforementioned 10kV distribution line, a threshold of 0.5% for high-frequency components and 0.3% for low-frequency components are set. After processing, abnormal fluctuation points are identified more accurately, providing reliable input data for subsequent distribution pattern analysis.
[0066] Once the time series of theoretical line loss rates accumulates to a certain extent, the system identifies the distribution pattern of the time series to detect whether there is a identifiable abrupt change in the statistical distribution. Specifically, the dual-track sequence is divided into two equal sub-data segments, and the mean and variance of the theoretical line loss rate for each sub-data segment are calculated. For example, for 30 consecutive days of theoretical line loss rate data for a certain distribution line, it is divided into two equal sub-data segments: the first 15 days and the last 15 days. The mean of the first sub-data segment is calculated to be 2.2%, with a variance of 0.012; the mean of the second sub-data segment is 2.7%, with a variance of 0.018. If the absolute difference between the mean of the first and second sub-data segments exceeds the maximum value of the historical normal fluctuation record for this distribution line (assuming a maximum value of 0.4%), and the variance of the first and second sub-data segments shows a continuous increase in the same direction, then an identifiable abrupt change is determined. This abrupt change point is located as the boundary point between the first and second sub-data segments, and the corresponding measured theoretical line loss rate is the dynamic threshold of 2.7%.
[0067] The system inputs a dynamic threshold into the economic transmission power model, which, based on the equivalent resistance change value generated by the correlation model, constructs a transmission loss power calculation function for distribution lines or transformer substations. Simultaneously, it acquires the time-of-use electricity price coefficient corresponding to the current load period and the line current margin determined through real-time monitoring. Combined with the real-time grid topology, it establishes an economic optimization equation for line loss, using the dynamic threshold as the baseline line loss rate, the current margin as the upper limit constraint, and the time-of-use electricity price coefficient as the economic weight coefficient. By iteratively solving this equation, the system identifies the operating point that minimizes transmission loss cost while satisfying the current margin constraint. The line loss rate corresponding to this operating point is extracted as the lower limit of the technically permissible range, and the theoretical line loss rate corresponding to the current margin boundary current is calculated as the upper limit of the technically permissible range, thus forming the optimal range. For example, if the calculated lower limit of the technically permissible range is 2.3% and the upper limit is 3.1%, then the optimal range is [2.3%, 3.1%].
[0068] When constructing the economic transmission power model, the impact of distributed generation on the grid topology is fully considered. For example, if a distribution line is connected to a 500kW distributed photovoltaic power source, the system can accurately calculate the current distribution of each branch by monitoring its output power in real time and combining it with the grid topology, thereby optimizing the transmission loss power. In the optimization equation, the output of the distributed power source is treated as an adjustable variable. Under the condition of meeting the grid security constraints, by optimizing the output of the distributed power source, the transmission loss cost is further reduced, making the economic transmission power model more in line with the actual grid operation and improving its practicality and economy.
[0069] The system compares the measured line loss rate with the optimal range. When the measured line loss rate exceeds the boundary of the optimal range, a dual-deviation direction collaborative diagnosis mechanism is activated. First, the system calculates the first difference between the measured line loss rate and the theoretical line loss rate output by the current cycle correlation model, and then independently calculates the second difference between the theoretical line loss rate and the dynamic threshold of the median of the optimal range. For example, assuming the dynamic threshold of the median of the optimal range is 2.7%, the measured line loss rate at a certain moment is 3.3%, and the theoretical line loss rate output by the correlation model is 2.9%. The first difference between the measured line loss rate and the theoretical line loss rate is +0.4%, and the second difference between the theoretical line loss rate and the dynamic threshold is +0.2%. Both are positive, and the system determines that it is a technically high loss, indicating that the abnormal line loss is caused by the equipment operating state deviating from the optimal benchmark set by the economic transmission model.
[0070] Based on the high-loss determination results, the system triggers automatic work orders for managerial high losses and initiates the optimal solution from the multi-objective search method library for technical high losses. For example, for the aforementioned technical high-loss situation, the system comprehensively considers multiple optimization objectives such as reducing line loss rate, improving voltage quality, balancing three-phase load, and reducing equipment investment costs, and calls a genetic algorithm from the multi-objective search method library to solve the problem. During the solution process, by establishing the priority and weight coefficients of the objective function, the optimization objectives are coordinated and balanced, ultimately obtaining a comprehensive optimal solution. Assuming that in a certain solution, the weight coefficient for reducing line loss rate is 0.4, the weight coefficient for improving voltage quality is 0.3, the weight coefficient for balancing three-phase load is 0.2, and the weight coefficient for reducing equipment investment costs is 0.1; after iterative solution, a comprehensive optimal solution is obtained, which is expected to reduce line loss rate by 0.8%, improve voltage quality by 5%, balance three-phase load by 10%, and reduce equipment investment costs by 15%.
[0071] Meanwhile, after implementing corresponding governance measures, the system uses newly added operational data to incrementally update the correlation model and dynamic thresholds, enabling the entire power line loss lean management system to continuously adapt to the dynamic changes of the power system and continuously improve management effectiveness and accuracy. For example, after a period of governance, new operational data was collected, revealing that the fluctuation characteristics of equipment component wear status data have decreased, the resistance change value of temperature change data has tended to stabilize, and the impact intensity quantification value of current surge event data has decreased. Based on this new data, the system adjusts the weight parameters of the correlation model and recalculates the dynamic thresholds, enabling the system to better reflect the actual operating status of the current power grid and provide more accurate decision support for subsequent line loss management.
[0072] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A lean management method for power line losses based on data-driven approaches and dynamic thresholds, characterized in that, Includes the following steps: Step 1. Based on the interrelationship between the wear state of mechanical components, the temperature change of power lines, and current surge events, construct a correlation model of the factors affecting power line losses; Step 2. Using power distribution lines or transformer substations as the smallest unit, synchronously input real-time data on the wear status of mechanical components, changes in power line temperature, and current surge events into the correlation model, output the theoretical line loss rate for the day, and accumulate the theoretical line loss rate day by day to obtain a time series. Step 3. Identify abnormal fluctuations in the time series to obtain a dual-track time series; identify the distribution pattern of the dual-track time series, extract the theoretical line loss rate corresponding to the mutation point as a dynamic threshold, and generate the optimal interval with the economic transmission power model as a constraint; the economic transmission power model is a transmission loss power calculation function constructed based on the equivalent resistance change value output by the correlation model. Step 4. Compare the measured line loss rate with the optimal range. If it exceeds the limit, compare the difference between the measured line loss rate and the theoretical line loss rate with the difference between the theoretical line loss rate and the dynamic threshold. If the two deviations are in the same direction, it is determined to be a technically high loss; if the two deviations are in opposite directions, it is determined to be a managerially high loss. Step 5. When the judgment result is high management loss, an automatic work order is triggered to start a multi-objective search method library to call the optimal solution for high technical loss; after implementation, the newly added running data is used to incrementally update the correlation model and dynamic threshold.
2. The method for lean management of power line losses based on data-driven and dynamic thresholds according to claim 1, characterized in that, In step 2, the processing steps of the correlation model are as follows: First, the input layer synchronously receives the preprocessed data of mechanical component wear status, power line temperature change and current impact event, and outputs the fluctuation characteristic value by performing time series analysis on the mechanical component wear status data; Based on the power grid topology, the spatial correlation effect of thermal conduction in conductors is modeled through graph convolutional layers, and the equivalent resistance change value is output from the power line temperature change data. The data of current surge events are analyzed for transient characteristics. Waveform distortion features are extracted through a wavelet decomposition layer, and a quantitative value representing the surge intensity is output based on the modulus of the wavelet coefficients. The three types of features are input into a dynamic gated fusion layer. Through a dynamic weight allocation strategy, the fluctuation feature value, the equivalent resistance change value, and the quantitative value representing the surge intensity are weighted and summed to generate the theoretical line loss rate.
3. The method for lean management of power line losses based on data-driven and dynamic thresholds according to claim 2, characterized in that, The dynamic weight allocation strategy is based on historical line loss data to invert and calibrate the actual contribution ratio of each data source to line loss, and adjusts the weight parameters in real time as new operational data is added.
4. The method for lean management of power line losses based on data-driven and dynamic thresholds according to claim 2, characterized in that, To ensure that the output theoretical line loss rate always remains consistent with the actual physical relationship between current and resistance, the theoretical line loss rate, along with the equivalent resistance change value and the real-time monitored current value, are input into the constraint function to calculate and output the theoretical line loss rate that conforms to the physical law of conductor heating. Where the constraint function is: , L theory I represents the corrected theoretical line loss rate, I represents the real-time monitored line current value, and R represents the line loss rate. eq The equivalent resistance change value and P output by the correlation model base The reference power is the rated transmission power of the line or the total power supply power of the area.
5. The method for lean management of power line losses based on data-driven and dynamic thresholds according to claim 1, characterized in that, In step 3, the steps for identifying abnormal fluctuations in the time series are as follows: calculate the absolute difference between the theoretical line loss rates of adjacent periods. When the absolute difference exceeds the theoretical line loss fluctuation range under normal operating conditions, mark the data point as an abnormal fluctuation point; retain all original data points and associate them with abnormal marking information to form a dual-track time series containing complete original data and abnormal marking status; the dual-track time series serves as input data for distribution pattern identification.
6. The method for lean management of power line losses based on data-driven and dynamic thresholds according to claim 1, characterized in that, The steps of extracting the dynamic threshold and generating the optimal interval include: based on the distribution pattern analysis of the dual-track time series, when a definite abrupt change is detected in the statistical distribution pattern of the theoretical line loss rate in a continuous normal data segment, the measured value of the theoretical line loss rate corresponding to the abrupt change point is extracted as the dynamic threshold; subsequently, the dynamic threshold is input into the economic transmission power model, and the economic transmission power model calculates in real time the upper and lower limits of the technically permissible interval that meets the minimum transmission loss cost, i.e., the optimal interval, based on the current power grid topology, load period and line current carrying capacity.
7. The method for lean management of power line losses based on data-driven and dynamic thresholds according to claim 1, characterized in that, The process of performing distribution pattern analysis on the dual-track time series is as follows: the dual-track time series is divided into two equal sub-data segments; the mean and variance of the theoretical line loss rate of the two sub-data segments are calculated respectively; if the absolute difference between the mean of the two sub-data segments exceeds the maximum value of the historical normal fluctuation record of the distribution line or transformer area, and the variance of the two sub-data segments shows a continuous increase or decrease in the same direction, then an identifiable pattern change is determined to have occurred, and the change point is located as the boundary point between the two sub-data segments; the maximum value of the historical normal fluctuation record comes from the historical statistical extreme value of the theoretical line loss fluctuation range under normal operating conditions of the equipment.
8. The method for lean management of power line losses based on data-driven and dynamic thresholds according to claim 1, characterized in that: The economic transmission power model is based on the equivalent resistance change value generated by the correlation model. It simultaneously obtains the time-of-use electricity price coefficient corresponding to the current load period and the line current margin determined by real-time monitoring, and constructs a transmission loss power calculation function for distribution lines or transformer areas. By iteratively solving for the operating point that minimizes transmission loss cost under the current margin constraint, the line loss rate corresponding to the operating point is extracted as the lower limit of the technically permissible range. At the same time, the theoretical line loss rate corresponding to the current margin boundary current is calculated as the upper limit of the technically permissible range.
9. A method for lean management of power line losses based on data-driven and dynamic thresholds according to claim 1, characterized in that, In step 4, the process of comparing the measured line loss rate with the optimal interval is as follows: when the measured line loss rate exceeds the boundary of the optimal interval, the dual-bias direction collaborative diagnosis mechanism is activated; the first difference between the measured line loss rate and the theoretical line loss rate output by the current cycle correlation model is calculated, and the second difference between the theoretical line loss rate and the dynamic threshold is calculated independently; if the first difference and the second difference are both positive or both negative, it is determined to be a technically high loss; if the first difference and the second difference are opposite, it is determined to be a managerially high loss.
10. A data-driven and dynamic threshold-based lean management system for power line loss, used to implement the method described in any one of claims 1-9, characterized in that, include: Data Synchronization Injection Unit: Taking power distribution lines or transformer substations as the smallest unit, real-time data on the wear status of mechanical components, changes in power line temperature, and current surge events are synchronously injected into the correlation model; the correlation model is constructed through the interrelationships between the wear status of mechanical components inside the power detection equipment, changes in power line temperature, and current surge events. Time series construction unit: The theoretical line loss rate output by the correlation model is accumulated day by day to obtain the time series; Dynamic threshold unit: Identify abnormal fluctuations in the time series to obtain a dual-track time series, identify the distribution pattern of the dual-track time series, extract the theoretical line loss rate corresponding to the mutation point as the dynamic threshold, and generate the optimal interval with the economic transmission power model as a constraint. Collaborative diagnostic unit: Compares the measured line loss rate with the optimal range. If it exceeds the limit, it compares the difference between the measured line loss rate and the theoretical line loss rate with the difference between the theoretical line loss rate and the dynamic threshold. If the two deviations are in the same direction, it is judged as technical high loss; if the two deviations are in opposite directions, it is judged as management high loss. Management Update Unit: When the judgment result is high management loss, an automatic work order is triggered, and the optimal solution is called from the multi-objective search method library for high technical loss; after implementation, the newly added running data is used to incrementally update the correlation model and dynamic threshold.