Methods and Devices for Predicting Line Losses in High-Proportion New Energy Distribution Networks

By constructing a line loss prediction model that combines bidirectional gated cyclic units with long short-term memory networks, the line loss data is decomposed into high-frequency and low-frequency components, solving the instability problem of line loss prediction in high-proportion new energy distribution networks, and achieving accurate line loss prediction and adaptive enhancement.

CN117495126BActive Publication Date: 2026-11-17STATE GRID HEBEI ELECTRIC POWER RES INST +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202311455672.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-11-02
Publication Date
2026-11-17
Estimated Expiration
2043-11-02

AI Technical Summary

Technical Problem

The prediction of line losses in high-proportion renewable energy distribution networks is unstable and volatile. Existing technologies cannot reflect the current or future line loss situation in a timely and accurate manner, resulting in a lag in loss reduction measures.

Method used

A line loss prediction model combining bidirectional gated recurrent units and long short-term memory networks is adopted. By decomposing line loss data into high-frequency and low-frequency components, high-frequency sub-models and low-frequency sub-models are trained separately, and prediction is performed using key influencing factors to synthesize line loss prediction data.

Benefits of technology

It enables accurate prediction of line losses in distribution networks with a high proportion of new energy sources, adapts to changes in the proportion of different new energy sources, and improves the accuracy and stability of prediction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117495126B_ABST
    Figure CN117495126B_ABST
Patent Text Reader

Abstract

This application provides a method and apparatus for predicting line losses in high-proportion renewable energy distribution networks, belonging to the field of distribution network line loss prediction technology. The method includes: acquiring line loss data and key influencing factor data of a sample high-proportion renewable energy distribution network; decomposing the line loss data into first line loss data containing high-frequency components and second line loss data containing low-frequency components; constructing a line loss prediction model combining a bidirectional gated cyclic unit and a long short-term memory network, the line loss prediction model including a high-frequency sub-model and a low-frequency sub-model; training the high-frequency sub-model based on the key influencing factor data and the first line loss data, and training the low-frequency sub-model based on the key influencing factor data and the second line loss data, obtaining a trained line loss prediction model; and using the trained line loss prediction model to predict the line loss of the target high-proportion renewable energy distribution network. This application can more accurately predict the line loss of high-proportion renewable energy distribution networks.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application belongs to the field of distribution network line loss prediction technology, and in particular relates to a method and device for predicting line losses in a high-proportion new energy distribution network. Background Technology

[0002] Line loss is a crucial indicator of the economic operation of a power system, impacting the planning, production, and management of power companies. It is a major cause of energy consumption in distribution systems. Studies show a strong correlation between line loss and load fluctuations, with most exhibiting seasonal variations. Simultaneously, with the rapid development of clean energy generation technologies, the penetration rate of renewable energy sources such as photovoltaics and hydropower in distribution networks is continuously increasing. Furthermore, the intermittent and volatile nature of renewable energy leads to constant changes in the line loss rate of new energy distribution networks, increasing the difficulty of formulating loss reduction measures and affecting power grid development planning.

[0003] Traditional line loss calculation methods rely on overly simplified models and processing techniques, which no longer meet the current requirements of power supply companies for line loss management. Unlike traditional distribution networks, high-proportion renewable energy distribution networks primarily use renewable energy sources. Therefore, line losses in high-proportion renewable energy distribution networks exhibit considerable instability and volatility, easily affected by weather, seasons, and time factors. Calculations of line losses in high-proportion renewable energy distribution networks, whether offline or online, are inherently lagging, failing to accurately reflect current or future line losses and thus hindering timely or proactive loss reduction measures. Summary of the Invention

[0004] This application provides a method and apparatus for predicting line losses in high-proportion renewable energy distribution networks, in order to solve the problem that line losses in high-proportion renewable energy distribution networks are difficult to predict accurately.

[0005] In a first aspect, embodiments of this application provide a method for predicting line losses in a high-proportion renewable energy distribution network, comprising: acquiring line loss data and key influencing factors of line loss for a sample high-proportion renewable energy distribution network, wherein the key influencing factors include time-series characteristic parameters and non-time-series characteristic parameters; decomposing the line loss data into first line loss data containing high-frequency components and second line loss data containing low-frequency components; constructing a line loss prediction model combining a bidirectional gated cyclic unit and a long short-term memory network, wherein the line loss prediction model includes a high-frequency sub-model and a low-frequency sub-model; training the high-frequency sub-model based on the key influencing factors data and the first line loss data, and training the low-frequency sub-model based on the key influencing factors data and the second line loss data to obtain a trained line loss prediction model; and using the trained line loss prediction model to predict the line loss of a target high-proportion renewable energy distribution network.

[0006] Among them, the key influencing factors of line loss refer to some key factors affecting the line loss of high-proportion renewable energy distribution networks. Collecting line loss data and data on key influencing factors of line loss from sample high-proportion renewable energy distribution networks is for use as training data to train the line loss prediction model.

[0007] In one possible implementation, the line loss prediction model further includes a synthesis unit; the synthesis unit is used to synthesize the first output data output by the high-frequency sub-model and the second output data output by the low-frequency sub-model into total line loss prediction data.

[0008] In one possible implementation, predicting the line loss of a target high-proportion renewable energy distribution network using the trained line loss prediction model includes: acquiring historical line loss data of the target high-proportion renewable energy distribution network at multiple historical moments and historical data of the key influencing factors; decomposing the historical line loss data at multiple historical moments into third line loss data containing high-frequency components and fourth line loss data containing low-frequency components; based on the third line loss data corresponding to the multiple historical moments and the historical data of the key influencing factors, using the high-frequency sub-model to predict the high-frequency components of the line loss of the target high-proportion renewable energy distribution network at at least one prediction moment to obtain high-frequency prediction data; based on the fourth line loss data corresponding to the multiple historical moments and the historical data of the key influencing factors, using the low-frequency sub-model to predict the low-frequency components of the line loss of the target high-proportion renewable energy distribution network at the at least one prediction moment to obtain low-frequency prediction data; and synthesizing the high-frequency prediction data and the low-frequency prediction data through the synthesis unit to obtain the line loss prediction data of the target high-proportion renewable energy distribution network at the at least one prediction moment.

[0009] In one possible implementation, the synthesis unit synthesizes the first output data and the second output data into total line loss prediction data through wavelet packet reconstruction; the step of decomposing the line loss data into first line loss data containing high-frequency components and second line loss data containing low-frequency components includes: performing wavelet packet decomposition on the line loss data to obtain multiple high-frequency components and multiple low-frequency components in multiple decomposition levels; combining the multiple high-frequency components in the multiple decomposition levels to obtain the first line loss data; and combining the multiple low-frequency components in the multiple decomposition levels to obtain the second line loss data.

[0010] In one possible implementation, acquiring line loss data and key influencing factors data of a sample high-proportion renewable energy distribution network includes: collecting raw line loss data and raw data of line loss influencing factors at multiple times in the sample high-proportion renewable energy distribution network, wherein the line loss influencing factors include electrical parameters and environmental characteristic parameters; preprocessing the raw line loss data and raw data of the line loss influencing factors to obtain a dataset; classifying the line loss influencing factors into time-series factors and non-time-series factors according to their time correlation; analyzing the non-time-series factor data and corresponding line loss data in the dataset using the grey relational analysis method, selecting non-time-series factors with a grey relational degree greater than a first threshold as non-time-series feature parameters, and acquiring the data of the non-time-series feature parameters; and analyzing the time-series factor data and corresponding line loss data in the dataset using the ARIMA method, selecting time-series factors with an ARIMA correlation degree greater than a second threshold as time-series feature parameters, and acquiring the data of the time-series feature parameters.

[0011] In one possible implementation, the time-series characteristic parameters include active power supply, temperature, light intensity, rainfall, reactive power supply, humidity, and wind speed; the non-time-series characteristic parameters include the total capacity of the public transformer, line length, proportion of high-loss transformers, line cross-sectional area, reactive power compensation device capacity, and line materials.

[0012] In one possible implementation, before predicting the line loss of a target high-proportion renewable energy distribution network using the trained line loss prediction model, the method further includes: testing the trained line loss prediction model based on a test set, evaluating the trained line loss prediction model using an evaluation function, and fine-tuning the line loss prediction model based on the evaluation results; the test set is obtained based on the dataset, and the evaluation function includes mean absolute error, mean absolute percentage error, mean square error, root mean square error, and R-squared error; fine-tuning the line loss prediction model based on the evaluation results includes: adding regularization terms, adjusting the learning rate, adjusting the number of network layers, and batch normalizing layers.

[0013] In one possible implementation, during the training process of the line loss prediction model, the method further includes updating the weights and biases of the line loss prediction model according to a preset loss function and optimizer.

[0014] In one possible implementation, the collection of raw line loss data and raw data of line loss influencing factors at multiple times in the sample high-proportion renewable energy distribution network includes: using PSCAD software to simulate the distribution system of a transformer substation with a high proportion of renewable energy access, and setting multiple renewable energy configuration scenarios to obtain the sample high-proportion renewable energy distribution network under various renewable energy configuration scenarios; simulating the operation of the sample high-proportion renewable energy distribution network under the multiple renewable energy configuration scenarios, and collecting data on the electrical structure parameters, operating load, renewable energy access-related parameters, and environmental parameters of the sample high-proportion renewable energy distribution network during the simulation operation to obtain the raw data of the line loss influencing factors.

[0015] Secondly, embodiments of this application provide a line loss prediction device for a high-proportion renewable energy distribution network, comprising: an acquisition module for acquiring line loss data and key influencing factors of line loss for a sample high-proportion renewable energy distribution network, wherein the key influencing factors include time-series characteristic parameters and non-time-series characteristic parameters; a frequency division module for decomposing the line loss data into first line loss data containing high-frequency components and second line loss data containing low-frequency components; a modeling and training module for constructing a line loss prediction model combining a bidirectional gated cyclic unit and a long short-term memory network, wherein the line loss prediction model includes a high-frequency sub-model and a low-frequency sub-model; training the high-frequency sub-model based on the key influencing factors data and the first line loss data, and training the low-frequency sub-model based on the key influencing factors data and the second line loss data to obtain a trained line loss prediction model; and a line loss prediction module for using the trained line loss prediction model to predict the line loss of a target high-proportion renewable energy distribution network.

[0016] This application provides a method and apparatus for predicting line losses in a high-proportion renewable energy distribution network. The method decomposes line loss data into first line loss data containing high-frequency components and second line loss data containing low-frequency components. The first line loss data and data on key influencing factors of line loss are used to train a high-frequency sub-model in the line loss prediction model, while the second line loss data and data on key influencing factors of line loss are used to train a low-frequency sub-model. Thus, when predicting line losses in a target high-proportion renewable energy distribution network, the prediction can be divided into two scenarios: high-frequency component prediction and low-frequency component prediction. This allows for accurate prediction of line losses with different high and low frequency contents using the same line loss prediction model, thereby enabling accurate prediction of line losses in distribution networks with different renewable energy integration ratios. Furthermore, the line loss prediction model is a prediction model combining a bidirectional gated cyclic unit and a long short-term memory network. During the training process, this model can better capture the key influencing factors of line loss and their long-term dependencies, resulting in more accurate line loss prediction results. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the implementation of the high-proportion renewable energy distribution network line loss prediction method provided in the embodiments of this application;

[0019] Figure 2 This is a schematic diagram of the structure of the BIGRU-LSTM model provided in the embodiments of this application;

[0020] Figure 3 This is a schematic diagram of the process for establishing a line loss prediction model provided in an embodiment of this application;

[0021] Figure 4 This is a schematic diagram of the structure of the high-proportion new energy distribution network line loss prediction device provided in the embodiments of this application. Detailed Implementation

[0022] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0023] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0024] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0025] A high-proportion renewable energy distribution network refers to a distribution network composed of a large number of renewable energy power generation devices, such as wind turbines and solar photovoltaic generators. In practical applications, the renewable energy integration ratio of such distribution networks is not fixed. For distribution networks with a high renewable energy integration ratio, the operation of renewable energy devices has a greater impact on system stability, and the proportion of high-frequency components in the power system load is relatively increased. Conversely, for distribution networks with a relatively low renewable energy integration ratio, the operation of renewable energy devices has a relatively smaller impact on system stability, and the proportion of low-frequency components in the power system load may be larger.

[0026] The high-proportion renewable energy distribution network line loss prediction method and apparatus provided in this application establish and train a line loss prediction model combining a bidirectional gated cyclic unit and a long short-term memory network. This model includes a high-frequency sub-model for predicting high-frequency components of line loss and a low-frequency sub-model for predicting low-frequency components. Therefore, the method in this application can be applied to high-proportion renewable energy distribution networks with various configurations, including those where the renewable energy integration ratio fluctuates. Whether the renewable energy integration ratio is high or low, the method provided in this application can accurately predict line losses based on the actual situation.

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0028] Figure 1 The implementation flowchart of the high-proportion renewable energy distribution network line loss prediction method provided in the embodiments of this application is described in detail below:

[0029] S101, Obtain line loss data and key influencing factors of line loss for high-proportion renewable energy distribution networks. Key influencing factors include time-series characteristic parameters and non-time-series characteristic parameters.

[0030] Among them, the key influencing factors of line loss refer to some key factors affecting the line loss of high-proportion renewable energy distribution networks. Collecting line loss data and data on key influencing factors of line loss from sample high-proportion renewable energy distribution networks is for use as training data to train the line loss prediction model.

[0031] In some embodiments, step S101 can be implemented by the following steps S201 to S205:

[0032] S201 collects raw line loss data and raw data of line loss influencing factors at multiple times in a high-proportion renewable energy distribution network sample. Line loss influencing factors include electrical parameters and environmental characteristic parameters.

[0033] Among them, electrical parameters may include, but are not limited to, line length, active and reactive power supply, and line materials; environmental characteristic parameters may include, but are not limited to, temperature, humidity, light intensity, and rainfall.

[0034] In one possible implementation, the aforementioned high-proportion renewable energy distribution network can be the high-proportion renewable energy distribution network to be predicted (target high-proportion renewable energy distribution network), which is the distribution network that the line loss prediction model constructed in this embodiment aims to predict. For example, if we want to predict the line loss of high-proportion renewable energy distribution network A, we can collect the original line loss data and the original data of line loss influencing factors of distribution network A to train the line loss prediction model. In this case, distribution network A is the aforementioned sample high-proportion renewable energy distribution network. The aforementioned original line loss data and the original data of line loss influencing factors can be the corresponding data statistically collected from the past operation of the high-proportion renewable energy distribution network to be predicted.

[0035] In another possible implementation, PSCAD (Power Systems Computer-Aided Design) software can be used to simulate a distribution system with a high proportion of renewable energy access, and various renewable energy configuration scenarios can be set to obtain sample high-proportion renewable energy distribution networks under various renewable energy configuration scenarios. Here, various renewable energy configuration scenarios refer to scenarios with different configurations such as the type combination of renewable energy, the access location of renewable energy equipment, and the amount of renewable energy equipment accessed. The distribution system under various renewable energy configuration scenarios is a sample high-proportion renewable energy distribution network with different renewable energy access ratios. Then, the daily operation of the above-mentioned sample high-proportion renewable energy distribution network is simulated in PSCAD software. During the simulation operation, data on line loss-related parameters such as electrical structure parameters, operating load, renewable energy access-related parameters, and environmental parameters of the sample high-proportion renewable energy distribution network are collected to obtain raw line loss data and raw data on line loss influencing factors.

[0036] It should be understood that during the data collection process described above, data can be collected at preset time intervals, and the line loss data and data on line loss influencing factors collected at each time point can be stored together. For example, data on relevant parameters of the operating distribution network can be collected every 15 minutes, and the collected line loss data and data on line loss influencing factors can be saved according to the collection time.

[0037] By using PSCAD software to simulate the operation of power distribution systems under various renewable energy configuration scenarios, it is possible to obtain operational data such as electrical characteristic parameters of high-proportion renewable energy distribution networks under various renewable energy access ratios. This allows for accurate analysis of high-proportion renewable energy distribution networks, which in turn helps to establish more realistic line loss prediction models.

[0038] S202, preprocess the original line loss data and the original data of the factors affecting line loss to obtain the dataset.

[0039] Optionally, preprocessing includes data cleaning and data normalization. Data cleaning removes or replaces outliers with usable values. Outliers typically include abnormal values ​​and highly abnormal values. Abnormal values ​​are data that deviate from the mean by more than two standard deviations, while highly abnormal values ​​are data that deviate from the mean by more than three standard deviations. Data normalization linearly scales the data in the dataset to a specified range. In this embodiment, the min-max normalization method is used for data normalization, and the normalization formula is as follows:

[0040]

[0041] Where X is the normalized value of the data, and y is the original value of the data. max y min These are the maximum and minimum values ​​among all the original values ​​of the data.

[0042] Accordingly, the inverse normalization formula is as follows, if necessary:

[0043] y = X·(y max -y min )+y min

[0044] S203 categorizes factors affecting line loss into time-series factors and non-time-series factors based on their time correlation.

[0045] In some embodiments, various parameters among the electrical parameters and environmental characteristic parameters in the collected factors affecting line loss can be classified into time-series factors and non-time-series factors according to preset classification rules. The preset classification rules are used to determine whether the above-mentioned parameters are related to time.

[0046] In other embodiments, time correlation analysis can be performed on various parameters among the electrical parameters and environmental characteristic parameters influencing line loss to determine time-series and non-time-series factors.

[0047] S204. The grey relational analysis method is used to analyze the non-time-series factors and corresponding line loss data in the dataset. Non-time-series factors with a grey relational degree greater than the first threshold are selected as non-time-series feature parameters, and the data of non-time-series feature parameters are obtained.

[0048] In some embodiments, the implementation process of step S204 may include: establishing a correlation coefficient matrix based on the data of non-time-series factors and the corresponding line loss data, and calculating the degree of correlation between each non-time-series factor and line loss. Then, according to the magnitude of the correlation between each non-time-series factor and line loss, the non-time-series factors are sorted, and non-time-series factors with a correlation degree greater than a first threshold are selected as non-time-series characteristic parameters among the key influencing factors of line loss.

[0049] The correlation coefficient matrix refers to the matrix composed of the correlation values ​​between non-time-series data and line loss data.

[0050] Optionally, the normalized line loss data can be used as the reference sequence X. s The normalized non-time-series factor data is used as the comparison sequence X. z The number of sequences is n z Reference sequence X s And compare sequence X z The expression is as follows:

[0051]

[0052] Next, according to the definition of grey relational degree, we compare sequence X. z and reference sequence X s The grey relational degree λ(X) between them s ,X z It can be calculated using the following formula:

[0053]

[0054]

[0055] g j =X s (j)-X z (j)

[0056] Where m is the total number of sequences; g j Become the reference sequence X s The j-th parameter X in s (j) and comparison sequence X z The j-th parameter X in z The difference of (j); This represents the maximum difference between two levels of sequences; X represents the minimum difference between two levels of sequences; s and X z The gray correlation degree between them is m times ζ j (X s (j),X z(j)); ρ is the resolution coefficient, and in this embodiment, ρ is 0.5.

[0057] In a specific embodiment of this application, six non-time-series characteristic parameters were determined using the grey relational analysis method, including the total capacity of the public transformer, line length, proportion of high-loss transformers, line cross-sectional area, reactive power compensation device capacity, and line materials. The grey relational analysis results of these six non-time-series characteristic parameters and line loss are shown in Table 1.

[0058] Table 1. Calculation results of the grey relational degree between non-time-series characteristic parameters and line loss.

[0059]

[0060] As shown in Table 1, the grey correlation between the total capacity of the public transformer and line loss is the highest, with a result of 0.925. The grey correlation between line materials and line loss is the lowest among the six non-time-series characteristic parameters, with a result of 0.681. All six non-time-series characteristic parameters in Table 1 have grey correlations with line loss of no less than 0.681, indicating a high degree of correlation. Therefore, they can be selected as key influencing factors on line loss.

[0061] Compared to traditional distribution networks, the electrical characteristic parameters in high-proportion renewable energy distribution networks are more complex. By using the grey relational analysis method in the above embodiments to specifically analyze the influencing factors of line loss, characteristic parameters that are more accurately matched to the line loss of high-proportion renewable energy distribution networks can be selected. This ensures the accuracy of line loss prediction while reducing the computational load in the establishment and training of the line loss prediction model.

[0062] S205. The ARIMA method is used to analyze the time series factors and corresponding line loss data in the dataset. Time series factors with ARIMA correlation greater than the second threshold are selected as time series feature parameters, and the data of time series feature parameters are obtained.

[0063] In this application embodiment, the time-series factor is a time-related parameter among the factors affecting line loss. The data of the time-series factor changes over time. Therefore, the ARIMA model can be used to select the time-series characteristic parameters that have a high contribution rate to the line loss of high-proportion new energy distribution networks.

[0064] The basic idea of ​​the ARIMA model (Autoregressive Integrated Moving Average model) is to effectively handle the trend and seasonal changes of time series data and predict future data points by using three main steps: autoregression (AR), differencing (I), and moving average (MA), based on the autocorrelation and lag relationship of time series data.

[0065] Autoregressive (AR): This step considers that the current observation may be related to one or more previous observations. In other words, the current observation can be explained by previous observations. In practical applications, values ​​from one or more past times can be used to predict future values.

[0066] Definition of the formula for a p-order autoregressive process:

[0067]

[0068] Among them, y t The current value is given, μ is a constant term, p is the order, and γ is the y-value. i ε is the autocorrelation coefficient. t The error is γ. It can be solved using maximum likelihood estimation and the least squares method. i .

[0069] Difference (I): Difference is primarily used to make non-stationary time series stationary. In the difference process, it is necessary to calculate the amount of change between consecutive observations in the time series data, thereby reflecting the trend or seasonal variation in the original time series.

[0070] Moving Average (MA): This step takes into account the potential influence of past error terms on the current observation. In other words, it is necessary to consider both the impact of previous observations on the current value and the impact of past errors on the current value.

[0071] The formula for q-order autoregression is:

[0072]

[0073] As described above, the basic idea of ​​the ARIMA model is to predict and analyze data by considering the autoregressiveness, non-stationarity, and cumulative error effects of time series data. Therefore, ARIMA can be used to capture time-series factors among the influencing factors of new energy output, thereby extracting their inherent time-series characteristics.

[0074] In this embodiment, the ARIMA method is used to analyze the time-series factors and corresponding line loss data in the dataset. First, the parameters of the ARIMA model need to be determined, including the order of the AR term, the degree of differencing, and the order of the MA term. These parameters can be estimated using tools such as autocorrelation plots (ACF) and partial autocorrelation plots (PACF), or the optimal model can be selected using criteria such as the Akaike information criterion (AIC) or the Bayesian information criterion (BIC).

[0075] Akaike Information Content Criterion Expression:

[0076] AIC = 2k - 2ln(L)

[0077] Bayesian information criterion expression:

[0078] BIC = kln(n) - 2ln(L)

[0079] In the two expressions above, k is the model parameter, n is the number of samples, and L is the likelihood function.

[0080] In some embodiments, based on the aforementioned time-series factors, time-lag features are established using the ARIMA method. These time-lag features are then used to assess the time dependence between time-series factors and line losses in high-proportion renewable energy distribution networks, thereby selecting time-series feature parameters that contribute significantly to line losses in such networks. A lag feature combines current observations with past observations to create a new feature. For example, a first-order lag feature is created by using the observation from the previous period as the new feature; a second-order lag feature is created by using the observations from the two previous periods as the new feature. Therefore, time-lag features can capture the time correlation between the influencing factors of line losses and line losses, thereby improving the accuracy and robustness of line loss prediction.

[0081] In one possible implementation, the process of analyzing the time-series data and corresponding line loss data in the dataset using the ARIMA method in step S205 includes: First, cleaning the time-series data and corresponding line loss data in the dataset, removing outliers and missing values, and performing preprocessing operations such as smoothing. Second, visualizing the data from the first step, including drawing time series plots, ACF (Autocorrelation Function), and PACF (Partial Autocorrelation Function) to understand the patterns and trends of the data. Third, observing and analyzing the visualized data from the second step, selecting an appropriate ARIMA model, establishing the ARIMA model, and optimizing the parameters. Fourth, establishing different ARIMA models, comparing and evaluating the parameters, and selecting the optimal ARIMA model. Fifth, evaluating and validating the ARIMA model selected in the fourth step through prediction and actual comparison to determine the model's accuracy and reliability. Step 6: Based on the ARIMA model prediction results from Step 5, analyze the linear and nonlinear influencing factors to identify the factors that have a significant impact on the ARIMA model prediction results. Step 7: Based on the ARIMA model prediction and analysis results from Step 6, and combined with pre-set requirements (specific business conditions and actual requirements), interpret and apply the results, and finally derive the specific calculation results of the correlation between time-series factors and line loss.

[0082] In a specific embodiment of this application, seven time-series characteristic parameters were determined using the ARIMA method, including active power supply, temperature, illuminance, rainfall, reactive power supply, humidity, and wind speed. The correlation results between the time-series characteristic parameters and line loss are shown in Table 2.

[0083] Table 2. Calculation results of the correlation between time series characteristic parameters and line loss.

[0084]

[0085] Referring to Table 2, active power supply has the strongest correlation with line loss, almost directly affecting its magnitude. Temperature and light intensity have relatively high correlations with line loss because they affect the resistance of power transmission, thus influencing line loss. While reactive power supply has a less significant impact on line loss than active power supply, it still consumes a certain amount of reactive power during distribution network operation, thus affecting line loss. Humidity and rainfall have relatively low correlations with line loss, but they still affect it to some extent. For example, excessively high humidity or rain can reduce the insulation performance of insulators, leading to increased line loss. Wind speed has the weakest correlation, with a relatively small impact on line loss.

[0086] In some embodiments, after step S205, the above-mentioned high-proportion renewable energy distribution network line loss prediction method may further include: dividing the dataset into a training set and a test set according to a preset ratio, wherein the training set is used to train the line loss prediction model established in the embodiments of this application, and the test set is used to test the line loss prediction model. For example, the dataset may be divided into training set and test set in a 7:3 ratio.

[0087] S102, decompose the line loss data into first line loss data containing high-frequency components and second line loss data containing low-frequency components.

[0088] In some embodiments, the implementation process of step S102 may include: performing wavelet packet decomposition on the line loss data to obtain multiple high-frequency components and multiple low-frequency components in multiple decomposition levels; combining the multiple high-frequency components in multiple decomposition levels to obtain first line loss data; and combining the multiple low-frequency components in multiple decomposition levels to obtain second line loss data.

[0089] In other embodiments, wavelet transform can be performed on the line loss data to obtain multiple high-frequency components and multiple low-frequency components in multiple decomposition levels; the multiple high-frequency components in multiple decomposition levels can be combined to obtain first line loss data; and the multiple low-frequency components in multiple decomposition levels can be combined to obtain second line loss data.

[0090] The high-frequency components in the aforementioned line loss data can reflect short-term, accidental, or externally-induced fluctuations, while the low-frequency components can reflect long-term, systematic trends. In subsequent prediction processes, by predicting the high-frequency and low-frequency components separately, the overall trend of line loss can be accurately captured, while also reflecting potential short-term anomalies or fluctuations. After synthesizing the predicted high-frequency and low-frequency components, a more accurate line loss prediction result can be obtained.

[0091] S103, construct a line loss prediction model that combines a bidirectional gated recurrent unit with a long short-term memory network. The line loss prediction model includes a high-frequency sub-model and a low-frequency sub-model. The high-frequency sub-model is trained based on the data of key influencing factors and the first line loss data, and the low-frequency sub-model is trained based on the data of key influencing factors and the second line loss data to obtain a trained line loss prediction model.

[0092] It is important to note that both the high-frequency and low-frequency sub-models are models that combine bidirectional gated recurrent units (GUUs) with long short-term memory (LSTM) networks. The high-frequency sub-model is used to predict the high-frequency components of line loss, while the low-frequency sub-model is used to predict the low-frequency components of line loss.

[0093] In some embodiments, the line loss prediction model further includes a synthesis unit; this synthesis unit is used to synthesize the first output data from the high-frequency sub-model and the second output data from the low-frequency sub-model into total line loss prediction data. For example, the first and second output data can be synthesized into total line loss prediction data through wavelet packet reconstruction. Corresponding to the decomposition method of line loss data, if wavelet transform is used to decompose the line loss data, the first and second output data are synthesized into total line loss prediction data in the synthesis unit through inverse wavelet transform.

[0094] GRU (Gated Recurrent Unit) is a commonly used gated recurrent neural network model for processing sequential data. Compared to traditional RNNs (Recurrent Neural Networks), GRU introduces a gating mechanism, which can better capture long-term dependencies and alleviate the gradient vanishing problem. Compared to unidirectional GRU, BIGRU (Bidirectional Gated Recurrent Unit) has richer contextual information, more comprehensive feature representation, and better model robustness.

[0095] LSTM (Long Short-Term Memory) networks are better suited for handling time-series tasks than CNN (Convolutional Neural Networks). Furthermore, LSTM addresses the vanishing gradient problem caused by backpropagation during RNN training. In this embodiment, the use of an LSTM structure alleviates the issue that when the sequence length exceeds a certain range, previous training data cannot correctly influence subsequent prediction results.

[0096] It should be noted that, in the embodiments of this application, the line loss prediction model combining a bidirectional gated recurrent unit (BiGRU-LSTM) and a long short-term memory network has a switching output mechanism.

[0097] The output switching mechanism is a prediction strategy that divides the unknown predicted value sequence corresponding to the predicted time period into two scenarios. The purpose of this mechanism is to achieve more accurate predictions, as different scenarios may require different prediction methods. In this embodiment, there are two scenarios for line loss prediction: a high-frequency scenario and a low-frequency scenario. The distinction between high-frequency and low-frequency scenarios is achieved by decomposing the input line loss data into high and low frequencies before using the line loss prediction model. For example, wavelet transform or wavelet packet decomposition can be used to obtain the high-frequency and low-frequency components of the line loss data. Here, by decomposing the line loss data, the renewable energy access characteristics of the high-proportion renewable energy distribution network currently being predicted can be analyzed; that is, the impact of renewable energy equipment in the currently predicted high-proportion renewable energy distribution network on line loss can be analyzed.

[0098] Traditional power systems are primarily supplied by energy sources such as coal and natural gas, which generate electricity with relatively stable frequencies and waveforms. However, the output of new energy power generation equipment, such as wind and solar power, is affected by factors such as weather and the natural environment, resulting in a certain degree of instability and volatility in its power output. This instability and volatility are mainly manifested in the high-frequency components of the power signal. Therefore, as the proportion of new energy sources increases, the proportion of high-frequency components in the power system load will relatively increase. Thus, if the output of new energy in a target high-proportion distribution network is relatively small, it may be a low-frequency scenario; if the output of new energy is relatively large, it may be a high-frequency scenario. In the low-frequency scenario, the proportion of low-frequency components after decomposing line loss data is relatively high, while the proportion of high-frequency components is relatively low. By using high-frequency sub-models and low-frequency sub-models to predict the high-frequency and low-frequency components of the line loss to be predicted, the result is still a line loss with a higher proportion of low-frequency components and a lower proportion of high-frequency components. By combining the two predicted components, an accurate line loss prediction result can be obtained. Therefore, based on the above-mentioned switching output mechanism, the scenario of the target high-proportion renewable energy distribution network can be automatically determined, and then targeted prediction can be made based on the actual high and low frequency ratio, thereby improving the accuracy of line loss prediction.

[0099] BiGRU captures both forward and backward information of the sequence through its bidirectional structure, thus better understanding the interrelationships between various factors related to line loss in the data. LSTM, on the other hand, utilizes its unique gating mechanism to effectively capture long-range temporal dependencies. Combining these two structures, the BiGRU-LSTM model can predict line loss more accurately.

[0100] In some embodiments, see Figure 2 The structure of the BiGRU-LSTM model shown is shown, and the output h of BiGRU is shown. t-1 This is an input to the LSTM. According to the principle of BiGRU, the output of the hidden state of BiGRU is as follows:

[0101]

[0102] Among them, w t v t These represent the forward hidden state of the BiGRU at time t. and reverse hidden state The corresponding weight; b t This represents the bias corresponding to the hidden state at time t.

[0103] Based on the principle of LSTM, combined with Figure 2 The structure of LSTM, and the forget gate f in the LSTM structure. t Input gate and output gate o tThe expression is as follows:

[0104] f t =σ(W f [h t-1 ,x t ]+b f )

[0105]

[0106] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0107] Among them, b f b C b o Forgetting gate f t Input gate and output gate o t The corresponding bias.

[0108] After establishing the structure of the line loss prediction model, the model parameters are set and the model is initialized. Model parameters include the input feature dimension, hidden state dimension, number of network layers, and whether bias is used for BiGRU; and the input feature dimension, hidden state dimension, number of network layers, and whether bias is used for LSTM. The input feature dimension is determined based on the dimensions of the key influencing factors of line loss. If there are p key influencing factors, the input feature dimension is 2p, where the 2x relationship is due to the existence of lag features.

[0109] The line loss prediction model in this embodiment exhibits excellent adaptability to the proportion of renewable energy. If the proportion of renewable energy is high, the high-frequency component in the line loss is greater, in which case the high-frequency sub-model plays a primary role, supplemented by the low-frequency sub-model. If the proportion of renewable energy is low, the low-frequency model plays a more significant role. This allows for dynamic and accurate prediction of line losses as the proportion of renewable energy in the target high-proportion renewable energy distribution network changes, achieving a flexible and compatible prediction effect and greatly enhancing the robustness and environmental adaptability of the line loss prediction model.

[0110] In this embodiment of the application, step S103, which trains the high-frequency sub-model based on data of key influencing factors and first line loss data, and trains the low-frequency sub-model based on data of key influencing factors and second line loss data, includes:

[0111] A first time series composed of key influencing factors data and first line loss data at multiple moments within the first time period is used as input to a high-frequency sub-model, and a second time series composed of first line loss data at multiple moments within the second time period is used as label to train the high-frequency sub-model. A third time series composed of key influencing factors data and second line loss data at multiple moments within the first time period is used as input to a low-frequency sub-model, and a fourth time series composed of first line loss data at multiple moments within the second time period is used as label to train the low-frequency sub-model. The first moment within the second time period is after the first moment within the first time period. The lengths of the first and second time periods, and the time intervals between moments within each time period, can be set according to actual needs; this application does not impose any restrictions on this.

[0112] In some embodiments, the training data used in the above training process is the data in the training set.

[0113] In some embodiments, during the training of the above-mentioned line loss prediction model, the weights and biases of the line loss prediction model can be updated according to a preset loss function and optimizer to improve the prediction capability of the line loss prediction model.

[0114] In some embodiments, the above-mentioned high-proportion new energy distribution network line loss prediction method may further include: testing the trained line loss prediction model based on a test set, evaluating the trained line loss prediction model through an evaluation function, and optimizing the line loss prediction model based on the evaluation results; the test set is obtained based on a dataset, and the evaluation function includes mean absolute error, mean absolute percentage error, mean square error, root mean square error, and R-squared error.

[0115] Optionally, the line loss prediction model can be optimized based on the evaluation results, including: adding regularization terms, adjusting the learning rate, adjusting the number of network layers, and batch normalizing layers.

[0116] Mean Absolute Error (MAE) is the average of the absolute differences between predicted and actual values; it measures the overall error of a prediction model. Mean Absolute Percentage Error (MAPE) measures the average percentage difference between predicted and actual values. Mean Squared Error (MSE) measures the square of the average difference between predicted and actual values. Root Mean Squared Error (RMSE) is the square root of the MSE; it measures the average difference between predicted and actual values. R-squared Error (R... 2The evaluation function is used to measure how well the model fits the observed data. The expressions for each evaluation function are as follows:

[0117]

[0118]

[0119]

[0120]

[0121]

[0122] MAE is relatively robust and unaffected by outliers; however, it does not square the error, so it may not capture the significance of larger errors. MAPE expresses the error as a percentage, which provides a better understanding of relative error; however, when the true value y... i When the value is close to or equal to zero, the denominator becomes zero, which may cause a division-by-zero error. MSE is more sensitive to outliers; larger errors can significantly impact the result. RMSE is similar to MSE, but even more sensitive to larger errors. Compared to MSE, RMSE has a larger value and therefore may amplify with increasing errors. R 2 The range of R is between 0 and 1, with larger values ​​indicating a better model fit and a higher degree of fit. However, R... 2 It is impossible to capture the distribution of prediction errors and therefore cannot assess the accuracy of the model.

[0123] Each of the above evaluation functions has its own advantages and disadvantages. In this embodiment, all five evaluation functions are used together to evaluate the line loss prediction results, thereby obtaining a comprehensive evaluation result. Based on the evaluation results, the BiGRU-LSTM line loss prediction model is tuned to further optimize its performance.

[0124] Figure 3 This is a schematic diagram of the process for establishing a line loss prediction model provided in an embodiment of this application. For example... Figure 3As shown, the process begins with data acquisition, including data on factors influencing line loss and load data. Relevant electrical and environmental parameters are determined from the data on factors influencing line loss, and line loss data is calculated from the load data. Next, feature engineering is used to determine the characteristic parameters of line loss in high-proportion renewable energy distribution networks. This involves dividing electrical and environmental parameters into non-time-series and time-series factors. Grey relational analysis is performed on the non-time-series factors, and ARIMA analysis is performed on the time-series factors to extract the characteristic parameters of line loss (both time-series and non-time-series parameters), which are then used to train the high-frequency and low-frequency sub-models. Simultaneously, wavelet transform or wavelet packet decomposition is performed on the line loss data to obtain high-frequency and low-frequency components. The high-frequency components are imported into the high-frequency sub-model, and the low-frequency components are imported into the low-frequency sub-model. Finally, inverse wavelet transform or wavelet packet reconstruction is performed on the output data of the high-frequency and low-frequency sub-models to obtain the predicted line loss data. The line loss prediction model is trained and tested by importing data into the high-frequency sub-model and the low-frequency sub-model. Finally, the line loss prediction model is evaluated to establish a line loss prediction model that can be used for line loss prediction in high-proportion new energy distribution networks.

[0125] S104, predict the line loss of the target high-proportion renewable energy distribution network using the trained line loss prediction model. In some embodiments, step S104 can be implemented by the following steps S301 to S305:

[0126] S301: Obtain historical line loss data and historical data of key influencing factors for the target high-proportion renewable energy distribution network at multiple historical moments.

[0127] S302 decomposes historical line loss data from multiple historical moments into third line loss data containing high-frequency components and fourth line loss data containing low-frequency components.

[0128] S303, based on third-party line loss data from multiple historical moments and historical data of key influencing factors, uses a high-frequency sub-model to predict the high-frequency components of line loss in the target high-proportion renewable energy distribution network at at least one moment to be predicted, and obtains high-frequency prediction data.

[0129] S304, based on the fourth line loss data at multiple historical moments and historical data of key influencing factors, uses a low-frequency sub-model to predict the low-frequency component of the line loss of the target high-proportion renewable energy distribution network at at least one moment to be predicted, and obtains low-frequency prediction data.

[0130] S305 synthesizes high-frequency and low-frequency prediction data through a synthesis unit to obtain the line loss prediction data of the target high-proportion renewable energy distribution network at at least one time to be predicted.

[0131] In the above embodiments, various operational and environmental data of the target high-proportion renewable energy distribution network are collected. These data are preprocessed, and historical line loss data and historical data of key influencing factors are extracted. The historical line loss data is then decomposed into high-frequency and low-frequency data. A time series for high-frequency prediction is composed of the historical data of key influencing factors and the third line loss data, and a time series for low-frequency prediction is composed of the historical data of key influencing factors and the fourth line loss data. These are input into the line loss prediction model and processed by the high-frequency and low-frequency sub-models respectively. Finally, the predicted high-frequency and low-frequency prediction data are synthesized to obtain the line loss prediction data for the time to be predicted. Since line loss data is usually not collected in real time, the time to be predicted can be the current time or a future time. Therefore, the embodiments of this application can achieve accurate prediction of the current and future line losses of a high-proportion renewable energy distribution network. Because the high-frequency and low-frequency components of line loss are considered separately, flexible prediction of line losses can also be achieved based on changes in the proportion of renewable energy access.

[0132] It is important to note that the line loss prediction model can be updated and adjusted in real time during the prediction process. Since the operating environment and state of the power system change over time, the line loss prediction model can be trained and updated periodically using the latest operating and environmental data to maintain its predictive performance.

[0133] This application embodiment establishes and trains a BiGRU-LSTM line loss prediction model with a switching output conversion mechanism, and decomposes the line loss data into high and low frequencies to achieve accurate prediction of the high-frequency and low-frequency components of line loss. Thus, under the dynamic change of the proportion of new energy access in a high-proportion new energy distribution network, the line loss prediction model can achieve flexible prediction of line loss, ensuring the accuracy of the line loss prediction data.

[0134] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0135] The following are device embodiments of this application. For details not described in detail, please refer to the corresponding method embodiments described above.

[0136] Figure 4 A schematic diagram of the high-proportion renewable energy distribution network line loss prediction device provided in an embodiment of this application is shown. For ease of explanation, only the parts related to the embodiment of this application are shown, and are described in detail below:

[0137] like Figure 4As shown, the high-proportion new energy distribution network line loss prediction device 400 includes: an acquisition module 410, a frequency division module 420, a modeling and training module 430, and a line loss prediction module 440.

[0138] The acquisition module 410 is used to acquire line loss data and key influencing factors of line loss in the high-proportion renewable energy distribution network sample. The key influencing factors include time-series characteristic parameters and non-time-series characteristic parameters.

[0139] Frequency division module 420 is used to decompose line loss data into first line loss data containing high frequency components and second line loss data containing low frequency components.

[0140] The modeling and training module 430 is used to construct a line loss prediction model that combines a bidirectional gated recurrent unit and a long short-term memory network. The line loss prediction model includes a high-frequency sub-model and a low-frequency sub-model. The high-frequency sub-model is trained based on the data of key influencing factors and the first line loss data, and the low-frequency sub-model is trained based on the data of key influencing factors and the second line loss data to obtain a trained line loss prediction model.

[0141] The line loss prediction module 440 is used to predict the line loss of a target high-proportion renewable energy distribution network using a trained line loss prediction model.

[0142] In one possible implementation, the acquisition module 410 is specifically used to: collect raw line loss data and raw data of line loss influencing factors at multiple times in a sample high-proportion renewable energy distribution network, the line loss influencing factors including electrical parameters and environmental characteristic parameters; preprocess the raw line loss data and raw data of line loss influencing factors to obtain a dataset; classify the line loss influencing factors into time-series factors and non-time-series factors according to their time correlation; analyze the data of non-time-series factors and the corresponding line loss data in the dataset using the grey relational analysis method, select non-time-series factors with a grey relational degree greater than a first threshold as non-time-series feature parameters, and acquire the data of non-time-series feature parameters; analyze the data of time-series factors and the corresponding line loss data in the dataset using the ARIMA method, select time-series factors with an ARIMA correlation degree greater than a second threshold as time-series feature parameters, and acquire the data of time-series feature parameters.

[0143] In one possible implementation, the acquisition module 410 is specifically used to: simulate a distribution system with a high proportion of renewable energy access using PSCAD software, and set up multiple renewable energy configuration scenarios to obtain sample high-proportion renewable energy distribution networks under various renewable energy configuration scenarios; simulate the operation of the sample high-proportion renewable energy distribution networks under multiple renewable energy configuration scenarios, and collect data on the electrical structure parameters, operating load, renewable energy access-related parameters, and environmental parameters of the sample high-proportion renewable energy distribution networks during the simulation operation to obtain raw line loss data and raw data on line loss influencing factors.

[0144] In one possible implementation, the frequency division module 420 is specifically used to: perform wavelet packet decomposition on the line loss data to obtain multiple high-frequency components and multiple low-frequency components in multiple decomposition levels; combine the multiple high-frequency components in multiple decomposition levels to obtain first line loss data; and combine the multiple low-frequency components in multiple decomposition levels to obtain second line loss data.

[0145] In one possible implementation, the line loss prediction model 440 is specifically used to: acquire historical line loss data and historical data of key influencing factors for the target high-proportion renewable energy distribution network at multiple historical moments; for the historical line loss data at each historical moment, decompose it into third line loss data containing high-frequency components and fourth line loss data containing low-frequency components; based on the third line loss data and historical data of key influencing factors at multiple historical moments within the historical period, use a high-frequency sub-model to predict the high-frequency components of the line loss of the target high-proportion renewable energy distribution network at at least one moment to be predicted, to obtain high-frequency prediction data; based on the fourth line loss data and historical data of key influencing factors at multiple historical moments within the historical period, use a low-frequency sub-model to predict the low-frequency components of the line loss of the target high-proportion renewable energy distribution network at at least one moment to be predicted, to obtain low-frequency prediction data; and synthesize the high-frequency prediction data and low-frequency prediction data through a synthesis unit to obtain the line loss prediction data of the target high-proportion renewable energy distribution network at at least one moment to be predicted.

[0146] In one possible implementation, the aforementioned high-proportion renewable energy distribution network line loss prediction device may further include:

[0147] The testing and evaluation module is used to test the trained line loss prediction model based on the test set, evaluate the trained line loss prediction model through the evaluation function, and fine-tune the line loss prediction model based on the evaluation results. The test set is obtained based on the dataset, and the evaluation function includes mean absolute error, mean absolute percentage error, mean square error, root mean square error, and R-squared error. The fine-tuning of the line loss prediction model based on the evaluation results includes: adding regularization terms, adjusting the learning rate, adjusting the number of network layers, and batch normalizing layers.

[0148] In one possible implementation, the aforementioned high-proportion renewable energy distribution network line loss prediction device may further include:

[0149] The training and optimization module is used to update the weights and biases of the line loss prediction model based on a preset loss function and optimizer.

[0150] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0151] Those skilled in the art will recognize that the templates, units, and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0152] If the module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above-described embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various embodiments of the high-proportion new energy distribution network line loss prediction method. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory, a random access memory, an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0153] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A high-proportion new energy power distribution network line loss prediction method, characterized in that, include: Obtain line loss data and key influencing factors of line loss in a high-proportion renewable energy distribution network sample. The key influencing factors include time-series characteristic parameters and non-time-series characteristic parameters. The line loss data is decomposed into first line loss data containing high-frequency components and second line loss data containing low-frequency components; A line loss prediction model combining a bidirectional gated recurrent unit (BRU) and a long short-term memory (LSM) network is constructed. The output of the BRU is the input of the LSM network. The line loss prediction model includes a high-frequency sub-model and a low-frequency sub-model. The high-frequency sub-model is trained based on the data of the key influencing factors and the first line loss data, and the low-frequency sub-model is trained based on the data of the key influencing factors and the second line loss data to obtain a trained line loss prediction model. The trained line loss prediction model is used to predict the line loss of the target high-proportion renewable energy distribution network. The line loss prediction model further includes a synthesis unit; the synthesis unit is used to synthesize the first output data output by the high-frequency sub-model and the second output data output by the low-frequency sub-model into total line loss prediction data; The synthesis unit combines the first output data and the second output data into total line loss prediction data through wavelet packet reconstruction. The step of decomposing the line loss data into first line loss data containing high-frequency components and second line loss data containing low-frequency components includes: The line loss data is decomposed using wavelet packet decomposition to obtain multiple high-frequency components and multiple low-frequency components at multiple decomposition levels. The first line loss data is obtained by combining multiple high-frequency components from the multiple decomposition levels. The second line loss data is obtained by combining multiple low-frequency components from the multiple decomposition levels. The acquisition of line loss data and key influencing factors of line loss for high-proportion renewable energy distribution networks includes: The raw line loss data and raw data of line loss influencing factors were collected at multiple times in the sample high-proportion new energy distribution network. The line loss influencing factors include electrical parameters and environmental characteristic parameters. The original line loss data and the original data of the factors affecting line loss are preprocessed to obtain a dataset; The factors affecting line loss are categorized into time-series factors and non-time-series factors based on their time correlation. The grey relational analysis method is used to analyze the non-time-series factors and corresponding line loss data in the dataset. Non-time-series factors with a grey relational degree greater than a first threshold are selected as non-time-series feature parameters, and the data of the non-time-series feature parameters are obtained. The ARIMA method was used to analyze the time-series factors and corresponding line loss data in the dataset. Time-series factors with ARIMA correlation greater than a second threshold were selected as time-series feature parameters, and the data of the time-series feature parameters were obtained. 2.The high-proportion new energy power distribution network line loss prediction method of claim 1, wherein, The method of predicting line losses in a target high-proportion renewable energy distribution network using the trained line loss prediction model includes: Acquire historical line loss data and historical data of the key influencing factors for the target high-proportion renewable energy distribution network at multiple historical moments; The historical line loss data at the multiple historical moments are decomposed into third line loss data containing high-frequency components and fourth line loss data containing low-frequency components. Based on the third line loss data corresponding to the multiple historical moments and the historical data of the key influencing factors, the high-frequency sub-model is used to predict the high-frequency components of the line loss of the target high-proportion new energy distribution network at at least one moment to be predicted, so as to obtain high-frequency prediction data. Based on the fourth line loss data corresponding to the multiple historical moments and the historical data of the key influencing factors, the low-frequency sub-model is used to predict the low-frequency component of the line loss of the target high-proportion new energy distribution network at at least one prediction moment, and low-frequency prediction data is obtained. The high-frequency prediction data and the low-frequency prediction data are synthesized by the synthesis unit to obtain the line loss prediction data of the target high-proportion new energy distribution network at at least one prediction time.

3. The method for predicting line losses in high-proportion renewable energy distribution networks as described in claim 1 or 2, characterized in that, The time-series characteristic parameters include active power supply, temperature, light intensity, rainfall, reactive power supply, humidity, and wind speed; the non-time-series characteristic parameters include the total capacity of the public transformer, line length, proportion of high-loss transformers, line cross-sectional area, reactive power compensation device capacity, and line materials.

4. The method for predicting line losses in high-proportion new energy distribution networks as described in claim 1, characterized in that, Before using the trained line loss prediction model to predict the line loss of the target high-proportion renewable energy distribution network, the method further includes: The trained line loss prediction model is tested based on the test set, and evaluated using an evaluation function. The model is then tuned based on the evaluation results. The test set is obtained from the dataset, and the evaluation function includes mean absolute error, mean absolute percentage error, mean square error, root mean square error, and R-squared error. The optimization of the line loss prediction model based on the evaluation results includes: adding a regularization term, adjusting the learning rate, adjusting the number of network layers, and batch normalizing layers.

5. The method for predicting line losses in high-proportion new energy distribution networks as described in claim 1, characterized in that, During the training process of the line loss prediction model, the method further includes: The weights and biases of the line loss prediction model are updated according to the preset loss function and optimizer.

6. The method for predicting line losses in high-proportion new energy distribution networks as described in claim 1, characterized in that, The collection of raw line loss data and raw data on line loss influencing factors at multiple times in the sample high-proportion renewable energy distribution network includes: The PSCAD software was used to simulate the distribution system of the transformer area with a high proportion of renewable energy access, and various renewable energy configuration scenarios were set to obtain the sample high-proportion renewable energy distribution network under various renewable energy configuration scenarios. The sample high-proportion renewable energy distribution network is simulated and operated under various renewable energy configuration scenarios. During the simulation, data on the electrical structure parameters, operating load, renewable energy access-related parameters, and environmental parameters of the sample high-proportion renewable energy distribution network are collected to obtain the original line loss data and the original data on the factors affecting line loss.

7. A high-proportion renewable energy distribution network line loss prediction device, characterized in that, include: The acquisition module is used to acquire line loss data and key influencing factors of line loss in a sample high-proportion renewable energy distribution network. The key influencing factors include time-series characteristic parameters and non-time-series characteristic parameters. The frequency division module is used to decompose the line loss data into first line loss data containing high frequency components and second line loss data containing low frequency components. A modeling and training module is used to construct a line loss prediction model combining a bidirectional gated recurrent unit (BRU) and a long short-term memory (LSM) network. The output of the BRU is the input of the LSM network. The line loss prediction model includes a high-frequency sub-model and a low-frequency sub-model. The high-frequency sub-model is trained based on the data of the key influencing factors and the first line loss data, and the low-frequency sub-model is trained based on the data of the key influencing factors and the second line loss data to obtain a trained line loss prediction model. The line loss prediction model also includes a synthesis unit. The synthesis unit is used to synthesize the first output data of the high-frequency sub-model and the second output data of the low-frequency sub-model into total line loss prediction data. The synthesis unit synthesizes the first output data and the second output data into total line loss prediction data through wavelet packet reconstruction. The line loss prediction module is used to predict the line loss of the target high-proportion renewable energy distribution network using the trained line loss prediction model. The acquisition module is specifically used for: The raw line loss data and raw data of line loss influencing factors were collected at multiple times in the sample high-proportion new energy distribution network. The line loss influencing factors include electrical parameters and environmental characteristic parameters. The original line loss data and the original data of the factors affecting line loss are preprocessed to obtain a dataset; The factors affecting line loss are categorized into time-series factors and non-time-series factors based on their time correlation. The grey relational analysis method is used to analyze the non-time-series factors and corresponding line loss data in the dataset. Non-time-series factors with a grey relational degree greater than a first threshold are selected as non-time-series feature parameters, and the data of the non-time-series feature parameters are obtained. The ARIMA method was used to analyze the time-series factors and corresponding line loss data in the dataset. Time-series factors with ARIMA correlation greater than the second threshold were selected as time-series feature parameters, and the data of the time-series feature parameters were obtained. The frequency division module is specifically used for: The line loss data is decomposed using wavelet packet decomposition to obtain multiple high-frequency components and multiple low-frequency components at multiple decomposition levels. The first line loss data is obtained by combining multiple high-frequency components from the multiple decomposition levels. The second line loss data is obtained by combining multiple low-frequency components from the multiple decomposition levels.