Power distribution network line loss analysis method and device based on data and model dual drive

Abstract

The invention provides a power distribution network line loss analysis method and device based on data and model dual drive. The method comprises the steps that S1, original data, collected by SCADA / AMI equipment, of a power distribution network are acquired and preprocessed into a feature matrix; s2, calculating theoretical line loss by using a multi-target algorithm library, and calculating a residual error between the theoretical line loss and actually measured line loss; s3, splicing the feature matrix and the residual error into an input vector; s4, inputting the input vector into a pre-trained LSTM network to generate a parameter correction amount; s5, dynamically adjusting the current algorithm parameters of the multi-target algorithm library based on the correction amount; s6, the theoretical line loss is recalculated through the updated parameters, and a calibrated line loss value is obtained; and S7, evaluating the average absolute percentage error of the calibrated line loss value by using a sliding window, if the average absolute percentage error exceeds a preset threshold value, re-executing the steps S2 to S6, otherwise, outputting the line loss value. According to the method, high-precision and adaptive analysis is realized, manual intervention is reduced, efficiency and reliability are improved, and intelligent management of a power system is supported.

Description

Technical Field

[0001] This application belongs to the field of line loss analysis, and particularly relates to a method and device for analyzing the line loss of a distribution network driven by both data and models. Background Art

[0002] In the field of operation and management of distribution networks, line loss analysis, as a key indicator for measuring the economy, stability, and efficiency of power systems, has always received much attention. However, traditional methods for analyzing the line loss of distribution networks are unable to cope when faced with the requirements of intelligent and refined management of modern power systems.

[0003] Traditional line loss analysis methods often rely on single data statistical means, such as estimating line loss through simple averaging or trend analysis of historical data. Although this method is easy to operate, it ignores the complexity and dynamics of the real-time operating state of the distribution network, resulting in a large deviation between the line loss estimation result and the actual value. At the same time, there are also some methods that use fixed theoretical models for line loss calculation, such as formula calculations based on parameters such as resistance and inductance. However, the structure of the distribution network is complex and changeable, and the load characteristics also change with time continuously. Fixed models are difficult to adapt to these changes, making the analysis results lack accuracy and adaptability.

[0004] In addition, traditional line loss analysis methods also have obvious deficiencies in terms of intelligence and automation. These methods often require a large amount of manual intervention, such as data collection, processing, and analysis, which are not only inefficient but also error-prone. At the same time, due to the lack of intelligent means, traditional methods are difficult to achieve real-time monitoring and early warning of the line loss of the distribution network, and are unable to respond to potential problems in a timely manner, posing a hidden danger to the safe and stable operation of the power system.

[0005] With the intelligent development of power systems, a large amount of power operation data has been generated in distribution networks, including various types of data such as voltage, current, power, and temperature. These data contain rich information, which is of great significance for improving the accuracy and efficiency of line loss analysis in distribution networks. However, how to make full use of these data and combine advanced analysis models to achieve high-precision and adaptive line loss analysis of distribution networks has become an urgent problem to be solved. Summary of the Invention

[0006] The purpose of this application is to overcome the defects in the above-mentioned prior art and provide a method and device for analyzing the line loss of a distribution network driven by both data and models.

[0007] This application provides a method for analyzing the line loss of a distribution network driven by both data and models, including:

[0008] S1. Obtain the original data collected by SCADA / AMI devices in the distribution network, and preprocess the original data to obtain a feature matrix;

[0009] S2. Based on the feature matrix, calculate the theoretical line loss through a multi-objective algorithm, and calculate the residual between the theoretical line loss and the measured line loss;

[0010] S3. Concatenate each row element in the feature matrix with the residual to generate an input vector;

[0011] S4. Input the input vector into a pre-trained LSTM network to generate a parameter correction amount;

[0012] S5. Dynamically adjust the current algorithm parameters of the multi-objective algorithm library based on the parameter correction amount to obtain updated parameters;

[0013] S6. Based on the updated parameters, recalculate the theoretical line loss through the multi-objective algorithm library to obtain a calibrated line loss value;

[0014] S7. Obtain the mean absolute percentage error of the calibrated line loss value through a sliding window. When the mean absolute percentage error exceeds a preset threshold, trigger the re-execution of S2 to S6, otherwise output the calibrated line loss value.

[0015] Optionally, concatenating each row element in the feature matrix with the residual to generate an input vector includes:

[0016] Concatenate the voltage, current, resistance, and temperature feature vectors of each time step in the feature matrix with the residual value at the corresponding time point in the column direction to generate an input vector containing device status features and error feedback.

[0017] Optionally, in calculating the theoretical line loss through a multi-objective algorithm based on the feature matrix, dynamically selecting the optimal algorithm through the multi-objective algorithm library includes:

[0018] According to the historical error standard deviation of each algorithm and the residual between the current theoretical line loss and the measured line loss, perform weighted calculation using weight coefficients α and β, and select the algorithm with the smallest weight value as the current optimal algorithm.

[0019] Optionally, dynamically adjusting the current parameters of the multi-objective algorithm library based on the parameter correction amount includes:

[0020] Compare the absolute value of the residual obtained in real time with a preset threshold. When the residual exceeds the preset threshold, reduce the parameter adjustment amplitude, and when the residual is lower than the preset threshold, increase the adjustment amplitude.

[0021] Optionally, obtaining the mean absolute percentage error of the calibrated line loss value through a sliding window includes:

[0022] Within the set sliding window, calculate the absolute percentage error between the theoretical line loss and the measured line loss point by point, and obtain the average value of the errors within the window as the evaluation index.

[0023] This application also provides a distribution network line loss analysis device based on dual driving of data and model, including:

[0024] An acquisition module, which acquires the original data collected by SCADA / AMI devices in the distribution network and preprocesses the original data to obtain a feature matrix;

[0025] A calculation module, which calculates the theoretical line loss through a multi-objective algorithm based on the feature matrix, and calculates the residual between the theoretical line loss and the measured line loss;

[0026] A splicing module, which splices each row element in the feature matrix with the residual to generate an input vector;

[0027] A model module, which inputs the input vector into a pre-trained LSTM network to generate a parameter correction amount;

[0028] An adjustment module, which dynamically adjusts the current algorithm parameters of the multi-objective algorithm library based on the parameter correction amount to obtain updated parameters;

[0029] The calculation module, based on the updated parameters, recalculates the theoretical line loss through the multi-objective algorithm library to obtain a calibrated line loss value;

[0030] An output module, which obtains the mean absolute percentage error of the calibrated line loss value through a sliding window. When the mean absolute percentage error exceeds a preset threshold, it triggers the re-execution of the above modules, otherwise it outputs the calibrated line loss value.

[0031] Optionally, the splicing module splices each row element in the feature matrix with the residual to generate an input vector, including:

[0032] Splice the voltage, current, resistance, and temperature feature vectors at each time step in the feature matrix with the residual value at the corresponding time point in the column direction to generate an input vector containing device status features and error feedback.

[0033] Optionally, when the calculation module calculates the theoretical line loss through a multi-objective algorithm based on the feature matrix, it dynamically selects the optimal algorithm through the multi-objective algorithm library, including:

[0034] According to the historical error standard deviation of each algorithm and the residual between the current theoretical line loss and the measured line loss, use weight coefficients α and β for weight calculation, and select the algorithm with the smallest weight value as the current optimal algorithm.

[0035] Optionally, the adjustment module dynamically adjusts the current parameters of the multi-objective algorithm library based on the parameter correction amount, including:

[0036] Compare the absolute value of the residual obtained in real time with a preset threshold. When the residual exceeds the preset threshold, reduce the parameter adjustment amplitude; when the residual is lower than the preset threshold, increase the adjustment amplitude.

[0037] Optionally, the output module obtains the mean absolute percentage error of the calibrated line loss value through a sliding window, including:

[0038] Within the set sliding window, calculate the absolute percentage error between the theoretical line loss and the measured line loss point by point, and obtain the average value of the errors within the window as an evaluation index.

[0039] The beneficial effects of this application are:

[0040] This application provides a method for analyzing distribution network line loss based on dual driving of data and model, including: S1. Obtain the original data collected by SCADA / AMI devices in the distribution network, and preprocess the original data to obtain a feature matrix; S2. Based on the feature matrix, calculate the theoretical line loss through a multi-objective algorithm, and calculate the residual between the theoretical line loss and the measured line loss; S3. Concatenate each row element in the feature matrix with the residual to generate an input vector; S4. Input the input vector into a pre-trained LSTM network to generate a parameter correction amount; S5. Dynamically adjust the current algorithm parameters of the multi-objective algorithm library based on the parameter correction amount to obtain updated parameters; S6. Based on the updated parameters, recalculate the theoretical line loss through the multi-objective algorithm library to obtain a calibrated line loss value; S7. Obtain the mean absolute percentage error of the calibrated line loss value through a sliding window. When the mean absolute percentage error exceeds the preset threshold, trigger the re-execution of S2 to S6, otherwise output the calibrated line loss value. This application realizes high-precision and adaptive analysis of distribution network line loss through dual driving of data and model, reduces manual intervention, improves analysis efficiency and reliability, and provides strong support for the intelligent management of the power system. Description of the Drawings

[0041] Figure 1 is a schematic diagram of the distribution network line loss analysis process based on dual driving of data and model in this application;

[0042] Figure 2 is a schematic diagram of the distribution network line loss analysis device based on dual driving of data and model in this application. Detailed Embodiments

[0043] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, the embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art.

[0044] A method for analyzing power line losses in a distribution network based on dual driving of data and models in this application adopts a dual-driving architecture, in which the data layer and the model layer work together, specifically as follows:

[0045] Data layer:

[0046] Acquisition module: Through SCADA (Supervisory Control and Data Acquisition) / AMI (Advanced Metering Infrastructure) data acquisition devices deployed in the distribution network, data such as voltage U, current I, line resistance R, temperature T, etc. are collected in real time at a second-level frequency.

[0047] Feature engineering module: Preprocess the collected data, including data cleaning (removing outliers and missing values) and normalization (mapping the data to the interval [0, 1]). Then, combine voltage U, current I, line resistance R, temperature T, and the corresponding timestamp t in chronological order to construct a spatio-temporal feature matrix X = [U, I, R, T, t]. By constructing the feature matrix, multi-source data are integrated, and data of different types and different time scales are converted into a unified structured data form, providing a standardized and effective data input for the algorithm calculation in the model layer, facilitating the extraction of effective features in the data and the exploration of potential relationships between the data, thereby improving the accuracy of power line loss analysis.

[0048] Model layer:

[0049] The model layer consists of a multi-objective algorithm library and a dynamic calibration engine. The multi-objective algorithm library contains various algorithms applicable to power line loss analysis, including the root mean square current method, average current method, maximum current method, etc. The dynamic calibration engine consists of a data interface module, a power line loss difference analysis module, a parameter adjustment module, and a feedback evaluation module.

[0050] Multi-objective algorithm library:

[0051] Data interface module: This module is responsible for interacting with the data layer to obtain the feature matrix and provide a data basis for the entire dynamic calibration engine to ensure that the engine can work based on the latest and most accurate power operation information.

[0052] Power line loss difference analysis module: The power line loss difference analysis module calculates the residual between the theoretical power line loss and the measured value.

[0053] Parameter adjustment module: The parameter adjustment module predicts the parameter correction amount based on the LSTM network (Long Short-Term Memory network) and adjusts the algorithm parameters, and the feedback evaluation module evaluates the calibration result.

[0054] Feedback evaluation module: Visualize the calibrated line loss analysis results in the form of charts (such as line charts, bar charts), reports, etc., and set the warning thresholds for indicators such as line loss rate and error.

[0055] Please refer to Figure 1 As shown, this application provides a method for analyzing the line loss of a distribution network based on dual driving of data and model, including:

[0056] S1. Obtain the original data collected by SCADA / AMI devices in the distribution network, and preprocess the original data to obtain a feature matrix;

[0057] The data layer collects data such as voltage U, current I, line resistance R, temperature T, etc. in real time at a second-level frequency through SCADA (Supervisory Control and Data Acquisition) / AMI (Advanced Metering Infrastructure) data acquisition devices deployed in the distribution network.

[0058] Preprocess the collected data, including data cleaning (removing outliers and missing values), and normalization (mapping the data to the [0, 1] interval).

[0059] Then, combine voltage U, current I, line resistance R, temperature T, and the corresponding timestamp t in chronological order to construct a spatio-temporal feature matrix X = [U, I, R, T, t].

[0060] The data interface module in the model layer interacts with the data layer to obtain the feature matrix and send it to the multi-line loss difference analysis module.

[0061] S2. Based on the feature matrix, calculate the theoretical line loss through a multi-objective algorithm, and calculate the residual between the theoretical line loss and the measured line loss;

[0062] Multi-objective algorithm library: Includes a variety of algorithms applicable to line loss analysis, including the root mean square current method, average current method, maximum current method, etc.

[0063] Dynamically select the optimal algorithm through the entropy weight method, dynamically select the line loss calculation algorithm most suitable for the current operation state of the distribution network, and send it to the line loss difference analysis module.

[0064] First, according to the historical data samples, for each algorithm, calculate the historical error standard deviation σ of each algorithm according to the historical calculation results and the measured line loss data according to the following formula:

[0065]

[0066] Among them, m is the number of historical data samples, is the theoretical line loss calculated by the algorithm for the j-th sample, is the measured line loss of the j-th sample.

[0067] Then, in the current calculation, the theoretical line loss is calculated using the root mean square current method, the average current method, and the maximum current method respectively .

[0068] Obtain the measured line loss .

[0069] Next, calculate the weights of each algorithm according to the formula :

[0070]

[0071] Where and are weight coefficients ( takes values between 0.6 - 0.8, takes values between 0.2 - 0.4).

[0072] Finally, select the algorithm with the smallest weight as the current optimal algorithm.

[0073] This mechanism can dynamically select the line loss calculation algorithm that is most suitable for the current operation state of the distribution network according to the actual data and the historical performance of the algorithm, avoiding the limitations brought by fixed algorithms, and improving the accuracy and adaptability of line loss calculation.

[0074] Line loss difference analysis module:

[0075] Obtain the theoretical line loss calculated by the optimal algorithm in the multi-objective algorithm library and the measured line loss

[0076] Calculate the residual ε between the two:

[0077]

[0078] By calculating the residual, the difference between the theoretical calculation result and the actual measurement result can be quantified, intuitively reflecting the magnitude of the current line loss calculation error, providing a clear basis for subsequent parameter adjustment, so as to optimize the algorithm parameters specifically.

[0079] S3. Concatenate each row element in the feature matrix with the residual to generate an input vector;

[0080] The parameter concatenates the feature matrix X constructed by the feature engineering module and the residual calculated by the difference analysis module Perform splicing to form a new input vector [X, .

[0081] S4. Input the input vector into a pre-trained LSTM network to generate a parameter correction amount;

[0082] Input the input vector into the LSTM network. The LSTM network calculates the parameter correction amount by learning the time series features and error relationships in historical data :

[0083]

[0084] where, is the weight matrix, b is the bias vector, is the activation function. According to the current data features and calculation errors, predict the correction amount required for adjusting the algorithm parameters.

[0085] S5. Dynamically adjust the current algorithm parameters of the multi-objective algorithm library based on the parameter correction amount to obtain updated parameters;

[0086] Apply the correction amount to the relevant algorithm parameters in the multi-objective algorithm library to dynamically adjust the algorithm and further improve the accuracy of line loss calculation.

[0087] S6. Based on the updated parameters, recalculate the theoretical line loss through the multi-objective algorithm library to obtain a calibrated line loss value;

[0088] Specifically, re-execute step S2 above.

[0089] S7. Obtain the mean absolute percentage error of the calibrated line loss value through a sliding window. When the mean absolute percentage error exceeds a preset threshold, trigger the re-execution of S2 to S6, otherwise output the calibrated line loss value.

[0090] Establish a sliding window with a length of n (n ranges from 10 to 20). As new data is continuously generated, the window slides forward in the time series.

[0091] Within each window, according to the formula:

[0092]

[0093] Calculate the mean absolute percentage error.

[0094] By calculating , quantitatively evaluate the calibrated line loss calculation result and monitor the accuracy and reliability of line loss analysis in real time. When When it exceeds the preset threshold (e.g., 10%), the recalibration mechanism is triggered, and the residual analysis and parameter calibration are performed again to ensure that the line loss calculation results always maintain high accuracy. If the threshold is not exceeded, the calibrated line loss analysis results are visually displayed. When the results exceed the warning thresholds such as the line loss rate and error, the system issues an anomaly warning.

[0095] As Figure 2 shown, the present application also provides a distribution network line loss analysis device based on data and model dual-driving, including:

[0096] An acquisition module 201, which acquires the original data collected by SCADA / AMI devices in the distribution network and preprocesses the original data to obtain a feature matrix;

[0097] A calculation module 202, which calculates the theoretical line loss through a multi-objective algorithm based on the feature matrix and calculates the residual between the theoretical line loss and the measured line loss;

[0098] A splicing module 203, which splices each row element in the feature matrix with the residual to generate an input vector;

[0099] A model module 204, which inputs the input vector into a pre-trained LSTM network to generate a parameter correction amount;

[0100] An adjustment module 205, which dynamically adjusts the current algorithm parameters of the multi-objective algorithm library based on the parameter correction amount to obtain updated parameters;

[0101] The calculation module 202 recalculates the theoretical line loss through the multi-objective algorithm library based on the updated parameters to obtain a calibrated line loss value;

[0102] An output module 206, which obtains the mean absolute percentage error of the calibrated line loss value through a sliding window. When the mean absolute percentage error exceeds the preset threshold, it triggers the re-execution of the above modules, otherwise it outputs the calibrated line loss value.

[0103] Further, splicing each row element in the feature matrix with the residual to generate an input vector includes:

[0104] Splicing the voltage, current, resistance, and temperature feature vectors of each time step in the feature matrix with the residual value at the corresponding time point in the column direction to generate an input vector containing device status features and error feedback.

[0105] Further, in the calculation module calculating the theoretical line loss through a multi-objective algorithm based on the feature matrix, dynamically selecting the optimal algorithm through the multi-objective algorithm library includes:

[0106] Based on the historical error standard deviation of each algorithm and the residual between the current theoretical line loss and the measured line loss, the weight coefficients α and β are used for weight calculation, and the algorithm with the smallest weight value is selected as the current optimal algorithm.

[0107] Further, based on the parameter correction amount, the current parameters of the multi-objective algorithm library are dynamically adjusted, including:

[0108] Compare the absolute value of the residual obtained in real time with a preset threshold. When the residual exceeds the preset threshold, the parameter adjustment amplitude is reduced; when the residual is lower than the preset threshold, the adjustment amplitude is increased.

[0109] Further, the mean absolute percentage error of the calibrated line loss value is obtained through a sliding window, including:

[0110] Within the set sliding window, calculate the absolute percentage error between the theoretical line loss and the measured line loss point by point, and obtain the average value of the errors within the window as the evaluation index.

[0111] The above description of the embodiments is for the convenience of those of ordinary skill in the art to understand and apply the present application. It is obvious that those skilled in the art can easily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without creative labor. Therefore, the present application is not limited to the above embodiments, and the improvements and modifications made by those skilled in the art to the present application should be within the protection scope of the present application.

Claims

1. A method for analyzing distribution network line loss based on dual driving of data and model, characterized in that Including: S1. Obtain the original data collected by SCADA / AMI devices in the distribution network, and preprocess the original data to obtain a feature matrix; S2. Based on the feature matrix, calculate the theoretical line loss through a multi-objective algorithm, and calculate the residual between the theoretical line loss and the measured line loss; S3. Concatenate each row element in the feature matrix with the residual to generate an input vector; S4. Input the input vector into a pre-trained LSTM network to generate a parameter correction amount; S5. Dynamically adjust the current parameters of the multi-objective algorithm library based on the parameter correction amount to obtain updated parameters; S6. Based on the updated parameters, recalculate the theoretical line loss through the multi-objective algorithm library to obtain a calibrated line loss value; S7. Obtain the mean absolute percentage error of the calibrated line loss value through a sliding window. When the mean absolute percentage error exceeds a preset threshold, re-execute steps S2 to S6; otherwise, output the calibrated line loss value.

2. The method for analyzing power line losses of a distribution network based on dual drive of data and model according to claim 1, characterized in that Concatenating each row element in the feature matrix with the residual to generate an input vector includes: Concatenate the voltage, current, resistance, and temperature feature vectors at each time step in the feature matrix with the residual value at the corresponding time point in the column direction to generate an input vector containing device status features and error feedback.

3. A method for analyzing the line loss of a distribution network based on dual-drive of data and model according to claim 1, characterized in that, In calculating the theoretical line loss through a multi-objective algorithm based on the feature matrix, dynamically selecting the optimal algorithm through a multi-objective algorithm library includes: According to the historical error standard deviation of each algorithm and the residual between the current theoretical line loss and the measured line loss, calculate the weight value using weight coefficients α and β, and select the algorithm with the smallest weight value as the current optimal algorithm.

4. A method for analyzing the line loss of a distribution network based on dual-driving of data and model according to claim 1, characterized in that, Dynamically adjusting the current parameters of the multi-objective algorithm library based on the parameter correction amount includes: Compare the absolute value of the residual obtained in real time with a preset threshold. When the residual exceeds the preset threshold, reduce the parameter adjustment amplitude; when the residual is lower than the preset threshold, increase the adjustment amplitude.

5. A method for analyzing power line losses in a distribution network based on dual-driving of data and models according to claim 1, characterized in that, Obtaining the mean absolute percentage error of the calibrated line loss value through a sliding window includes: Within the set sliding window, calculate the absolute percentage error between the theoretical line loss and the measured line loss point by point, and obtain the average value of the errors within the window as an evaluation index.

6. A power distribution network line loss analysis device based on dual-drive of data and model, characterized in that, Including: An acquisition module that acquires the original data collected by SCADA / AMI devices in the distribution network and preprocesses the original data to obtain a feature matrix; A calculation module that calculates the theoretical line loss through a multi-objective algorithm based on the feature matrix and calculates the residual between the theoretical line loss and the measured line loss; A concatenation module that concatenates each row element in the feature matrix with the residual to generate an input vector; A model module that inputs the input vector into a pre-trained LSTM network to generate a parameter correction amount; An adjustment module that dynamically adjusts the current algorithm parameters of the multi-objective algorithm library based on the parameter correction amount to obtain updated parameters; The calculation module that recalculates the theoretical line loss through the multi-objective algorithm library based on the updated parameters to obtain a calibrated line loss value; The output module obtains the mean absolute percentage error of the calibrated line loss value through a sliding window. When the mean absolute percentage error exceeds a preset threshold, it triggers the re - execution of the above - mentioned module; otherwise, it outputs the calibrated line loss value.

7. The device for analyzing the line loss of a distribution network based on dual driving of data and models according to claim 6, characterized in that, The splicing module splices each row element in the feature matrix with the residual to generate an input vector, including: Splicing the voltage, current, resistance, and temperature feature vectors at each time step in the feature matrix with the residual value at the corresponding time point in the column direction to generate an input vector containing device status features and error feedback.

8. The device for analyzing the line loss of a distribution network based on dual drive of data and model according to claim 6, characterized in that The calculation module calculates the theoretical line loss based on the feature matrix through a multi - objective algorithm, and dynamically selects the optimal algorithm through a multi - objective algorithm library, including: According to the historical error standard deviation of each algorithm and the residual between the current theoretical line loss and the measured line loss, weight calculation is performed using weight coefficients α and β, and the algorithm with the smallest weight value is selected as the current optimal algorithm.

9. The device for analyzing the line loss of a distribution network based on dual drive of data and model according to claim 6, characterized in that, The adjustment module dynamically adjusts the current parameters of the multi - objective algorithm library based on the parameter correction amount, including: Comparing the absolute value of the residual obtained in real - time with a preset threshold. When the residual exceeds the preset threshold, the parameter adjustment amplitude is reduced; when the residual is lower than the preset threshold, the adjustment amplitude is increased.

10. A power distribution network line loss analysis device based on dual-drive of data and model according to claim 6, characterized in that The output module obtains the mean absolute percentage error of the calibrated line loss value through a sliding window, including: Within the set sliding window, calculate the absolute percentage error between the theoretical line loss and the measured line loss point by point, and obtain the average value of the errors within the window as an evaluation index.

Images