Meteorological factor-based power distribution network line loss prediction method and device

Abstract

The invention is suitable for the technical field of line loss prediction, and provides a meteorological factor-based power distribution network line loss prediction method and device, and the method comprises the steps: obtaining historical line loss data and a meteorological factor data set of a power distribution network in a target region, and carrying out the preprocessing of the historical line loss data and all meteorological factor data in the meteorological factor data set; decomposing the preprocessed historical line loss data based on a preset algorithm to obtain a historical line loss component; inputting the historical line loss component and all the preprocessed meteorological factor data into a feature extraction model of a pre-constructed line loss prediction network to obtain a correlation coefficient between each meteorological factor and the line loss data; and inputting the correlation coefficient and the historical line loss component into a prediction model of the line loss prediction network to obtain a line loss prediction result of the target area power distribution network. According to the method, the meteorological factors having positive influence on the prediction precision are selected in an iteration mode, and the prediction precision of line loss prediction is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of line loss prediction, and in particular relates to a method and device for predicting line loss in a distribution network based on meteorological factors. Background Art

[0002] With the rapid development of the economy, my country's demand for electricity is growing. The rapid growth of new energy power equipment continues to affect the stability of the power system. The stable operation of the power system has a very important impact on the development of the national economy. As the proportion of new energy in modern power systems continues to increase, the volatility and nonlinearity in the line loss series are much greater than those in traditional power systems, which further increases the difficulty of line loss prediction. In addition, with the continuous popularization of intelligent meters, it also provides more possibilities and challenges for line loss prediction. Traditional prediction methods, including time series method, fuzzy regression method, support vector regression (SVR), etc., however, due to the defects of these prediction methods themselves, the prediction accuracy is low and cannot meet the line loss prediction requirements of the ever-changing smart grid.

[0003] In recent years, the Long Short-Term Memory (LSTM) network has been widely used in the field of power line loss prediction because of its good performance in prediction. For example, the LSTM network was compared with the feedforward neural network experiment, highlighting the high prediction ability and practicality of the LSTM model for short-term power line loss prediction; the Gray Wolf Optimizer (GWO) was used to optimize the LSTM model and obtain the optimal hyperparameters, which effectively improved the accuracy of nonlinear short-term line loss data prediction; the BILSTM neural network prediction model was optimized by combining the attention mechanism with the Whale Optimization Algorithm (WOA), and compared with the commonly used Auto Regressive Integrated Moving Average Model (ARIMA). By comparing the prediction results and indicators, it was found that the WOA-BILSTM model has better results for short-term line loss prediction. However, similar to other neural network models, the LSTM model has limited generalization performance and is overly dependent on empirical values ​​due to some factors, such as the number of neurons in the hidden layer and the learning rate. The WOA-BILSTM model has better results for short-term line loss prediction. Although the prediction model has been optimized, it does not effectively process irregular and volatile data. It can be seen that the traditional single line loss prediction model does not deeply consider the coupling characteristics between the factors affecting line loss, resulting in limited prediction results and too long prediction time. Therefore, how to effectively screen features to improve prediction accuracy has become a hot topic in current research.

[0004] Some studies have analyzed the characteristics of load forecasting by comprehensively considering meteorological factors, constructed optimal input features, and optimized the prediction accuracy. Other studies have proposed that changes in meteorological factors and loads may limit the accuracy of learning models, and combined the SVR model for load forecasting to improve the reliability and prediction performance of the model. However, the above studies still use the traditional inefficient correlation coefficient method for the screening of meteorological factors, and cannot effectively calculate the weight of the features, resulting in the accuracy of the prediction model is still limited. Therefore, a more accurate distribution network line loss prediction method that considers key meteorological factors is urgently needed. Summary of the invention

[0005] The embodiments of the present invention provide a method and device for predicting line loss in a distribution network based on meteorological factors, so as to solve the problem of low accuracy of the prediction model caused by the inability to obtain an accurate correlation between meteorological factors and line loss.

[0006] The present invention is achieved through the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for predicting line loss in a distribution network based on meteorological factors, comprising:

[0008] Acquire historical line loss data and a meteorological factor data set of the target area distribution network, and preprocess the historical line loss data and all meteorological factor data in the meteorological factor data set; wherein the meteorological factor data set includes multiple meteorological factors and multiple meteorological factor data corresponding to each meteorological factor;

[0009] Decomposing the preprocessed historical line loss data based on a preset algorithm to obtain historical line loss components;

[0010] Input the historical line loss components and all pre-processed meteorological factor data into a pre-built feature extraction model of a line loss prediction network to obtain a correlation coefficient between each meteorological factor and the line loss data;

[0011] The correlation coefficient and the historical line loss component are input into the prediction model of the line loss prediction network to obtain the line loss prediction result of the target area distribution network.

[0012] In a second aspect, an embodiment of the present invention provides a distribution network line loss prediction device based on meteorological factors, comprising:

[0013] A processing module, used to obtain historical line loss data and a meteorological factor data set of the target area distribution network, and pre-process the historical line loss data and all meteorological factor data in the meteorological factor data set; wherein the meteorological factor data set includes multiple meteorological factors and multiple meteorological factor data corresponding to each meteorological factor;

[0014] A decomposition module, used for decomposing the preprocessed historical line loss data based on a preset algorithm to obtain historical line loss components;

[0015] An extraction module, used for inputting the historical line loss components and the preprocessed data of multiple meteorological factors into a feature extraction model of a pre-built line loss prediction network to obtain a correlation coefficient between each meteorological factor and the line loss data;

[0016] The prediction module is used to input the correlation coefficient and the historical line loss component into the prediction model of the line loss prediction network to obtain the line loss prediction result of the distribution network in the target area.

[0017] The embodiment of the present invention provides a distribution network line loss prediction method and device based on meteorological factors, which obtains the historical line loss and meteorological factor data sets of the distribution network in the target area, performs preprocessing, optimizes the variational mode decomposition algorithm, decomposes the preprocessed historical line loss by the optimized algorithm, obtains the historical line loss components, constructs a line loss prediction network including a feature extraction model and a prediction model, and inputs the processed historical line loss and meteorological data sets into the feature extraction network to obtain the correlation coefficient between each meteorological factor and the line loss data, and inputs the correlation coefficient and the historical line loss components into the line loss prediction network to obtain the line loss prediction result of the target area. The preprocessed data is optimized to effectively improve the data quality, the correlation coefficient between the climate factor and the line loss data is obtained by the feature extraction network, and the line loss prediction is performed by the line loss prediction network, and the prediction accuracy of the line loss prediction is effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0019] Figure 1 It is a flow chart of a method for predicting line loss in a distribution network based on meteorological factors provided by an embodiment of the present invention;

[0020] Figure 2 It is a flow chart of a ZOA algorithm of a distribution network line loss prediction method based on meteorological factors provided by an embodiment of the present invention;

[0021] Figure 3 This is a flow chart of an implementation method of a distribution network line loss prediction method based on meteorological factors provided by an embodiment of the present invention;

[0022] Figure 4 It is a schematic diagram of the structure of a distribution network line loss prediction device based on meteorological factors provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0024] Figure 1is a flow chart of a method for predicting line loss in a distribution network based on meteorological factors provided by an embodiment of the present invention, with reference to Figure 1 , the distribution network line loss prediction method based on meteorological factors is described as follows:

[0025] S110, obtaining historical line loss data and a meteorological factor data set of the target area distribution network, and preprocessing the historical line loss data and all meteorological factor data in the meteorological factor data set; wherein the meteorological factor data set includes multiple meteorological factors and multiple meteorological factor data corresponding to each meteorological factor.

[0026] In an optional embodiment, preprocessing the historical line loss data and the meteorological factor data set includes:

[0027] Standardize the historical line loss data and meteorological factor data sets to obtain standard historical line loss data and standard meteorological factor data sets;

[0028] The standard historical line loss data and the standard meteorological factor data set are divided into a training set and a test set according to the preset division rules;

[0029] The standard historical line loss data and the standard meteorological factor data set in the training set are used as the preprocessed historical line loss data and meteorological factor data set.

[0030] The standard historical line loss data is obtained after standardization of all historical line loss data, and the standard meteorological factor data set is obtained after standardization of the data of each meteorological factor in the meteorological factor data set.

[0031] Optionally, the meteorological factors may be: temperature, humidity, azimuth, cloud opacity, dew point temperature, direct horizontal irradiance, direct normal irradiance, global horizontal irradiance, global inclined irradiance, precipitable water, relative humidity, surface pressure, wind direction, wind speed, zenith angle. Other meteorological factors may also be used, and this application does not make specific limitations here.

[0032] Among them, preprocessing refers to standardization processing. Labeling all historical line loss data and meteorological factor data can effectively eliminate the influence of dimension, so that the method of this application can adapt to various types of data sets.

[0033] The normalization formula is:

[0034]

[0035] Where X is the historical line loss data or meteorological factor data, X′ is the standardized data, μ is the mean of the historical line loss data or meteorological factor data, and σ is the standard deviation of the historical line loss data or meteorological factor data.

[0036] S120, decomposing the preprocessed historical line loss data based on a preset algorithm to obtain historical line loss components.

[0037] In an optional embodiment, the preprocessed historical line loss data is decomposed based on a preset algorithm to obtain historical line loss components, including:

[0038] Optimizing the parameters of the variational mode decomposition algorithm to obtain a preset algorithm;

[0039] The preprocessed historical line loss data is decomposed based on a preset algorithm to obtain a preset number of historical line loss components.

[0040] Optionally, the preset algorithm is a ZOA-VMD (Zebra Optimization Algorithm-Variational Modal Decomposition) algorithm, that is, a VDM algorithm after ZOA optimization.

[0041] like Figure 2 As shown in Figure 1, the optimal parameters of the VDM algorithm can be obtained by optimizing the VDM algorithm through the ZOA algorithm. The ZOA algorithm is an optimization algorithm inspired by the natural behavior of zebras in the wild. The ZOA algorithm simulates the foraging behavior of zebras and the defensive behavior of predators.

[0042] First, initialize the optimization object.

[0043] Second, initialize the zebra population. From a mathematical point of view, each zebra is a candidate solution to the problem, and the plain where the zebra is located is the search space of the problem. The position of each zebra in the search space determines the value of the decision variable. Therefore, each zebra participating in the ZOA algorithm can be modeled by a vector. The elements of this vector represent the value of the problem variable. The number of zebras can be mathematically modeled by a matrix. The initial position of the zebra in the search space is randomly assigned. The population matrix of ZOA is as follows:

[0044]

[0045] Among them, Z is the zebra population, Z1 is the first zebra, and Z i is the i-th zebra, Z N is the Nth zebra, z 1,1 The first question for the first zebra, z 1,j The jth question asked by the first zebra, z 1,m The mth question asked by the first zebra, z i,1 is the ith question raised by the ith zebra, z i,jFor the jth question raised by the i-th zebra, z i,m The mth question asked by the i-th zebra, z N,1 The nth question asked by the Nth zebra, z N,j For the jth question raised by the Nth zebra, z N,m This is the mth question asked by the Nth zebra, where N is the number of members in the population and m is the number of decision variables.

[0046] Third, an evaluation function is established, and the optimal zebra is updated according to the evaluation function. In each iteration, the position of the zebra and the value of the objective function are updated, and the members participating in the ZOA algorithm are updated through two natural behaviors of zebras in the wild. Among them, these two natural behaviors include foraging and defense against predators. In the first stage, the population members are updated according to the foraging behavior simulation of zebras. Among the members participating in the ZOA algorithm, the best member of the population is considered to be the pioneer zebra and leads other population members to its search space position in the population. Therefore, the update of the position of the zebra in the foraging stage can be mathematically modeled using the following formula.

[0047]

[0048] in, is the new state of the kth generation zebra based on the first stage, y (k,l) is the value of the kth generation zebra in the lth dimension, P1 represents the first stage, is the value of the lth dimension of the kth generation zebra based on the first stage, is the value of the objective function, PZ is the pioneer zebra, r is a random number in the interval [0,1], I = rand (1 + rand), where rand is another random number in the interval [0,1], I∈[1,2], the closer the parameter is to 2, the more drastic the change.

[0049] Fourth, in the second stage, the positions of ZOA population members in the search space are updated by simulating the defensive behavior of zebras against predator attacks. Assume that one of the following two situations occurs with equal probability:

[0050] Situation (1): A lion attacks a zebra and the zebra runs away.

[0051] Case (2): Other predators attack the zebra, and the zebra chooses to attack.

[0052] In case (1), when the zebra is attacked by a lion, the zebra will avoid the lion attack near their environment. The position state update can be modeled as follows:

[0053]

[0054] in, is the value of the lth dimension of the kth generation zebra based on the second stage, P2 is the second stage, y (k,l) is the value of the kth generation zebra in the lth dimension, t is the maximum number of iterations, T is the total number of iterations of the zebra population, R is a constant of 0.01, r is a random number in the interval [0,1], P s is a random number in the interval [0,1].

[0055] In case (2), when other predators attack one of the zebras in the population, the other zebras in the herd will move towards the attacked zebra and try to attack the predator by establishing a defense mechanism. At this time, the position state update can be modeled by the following formula:

[0056]

[0057] in, is the value of the lth dimension of the kth generation zebra based on the second stage, P2 is the second stage, y (k,l) is the value of the kth generation zebra in the lth dimension, AZ is the state of the attacked zebra, r is a random number in the interval [0,1], P s is a random number in the interval [0,1].

[0058] Fifth, the optimal position is updated according to the calculation results, and it is determined whether the updated optimal position meets the iteration conditions. The iteration conditions can be reaching the maximum number of iterations, the optimal position meets the accuracy requirements, and the fitness does not change. It can also be other iteration conditions, which are not specifically limited. If the updated optimal position meets the iteration conditions, the optimal solution is saved. If the updated optimal position does not meet the iteration conditions, it returns to the first stage and iteratively updates the optimal position.

[0059] Optionally, the historical line loss can be decomposed and defined as:

[0060]

[0061] Among them, w k is the kth center frequency; u k is the kth line loss modal component; f(t) is the original load or line loss sequence; is the partial derivative operator; δ t is the unit pulse function.

[0062] According to the process and principle of VDM, the number of decompositions and bandwidth parameters have a significant impact on the results. In this application, the minimum envelope entropy is selected as the fitness function to optimize the number of decompositions and bandwidth parameters. The envelope entropy formula is:

[0063]

[0064] Among them, E p is the minimum envelope entropy, p j is the proportion or weight of the modal component in the entire signal, m is the number of samples, and a(j) is the Hirt-modulated envelope sequence of the VDM modal component.

[0065] When using ZOA to optimize the number of decompositions and bandwidth parameters with the minimum envelope entropy as the fitness function, if the ZOA parameters are set as shown in Table 1:

[0066] Table 1 ZOA parameter settings

[0067] Bandwidth parameters The number of decompositions Optimizing the number of variables Maximum number of iterations Population [200,2000] [2,10] 2 30 25

[0068] After the optimization process of ZOA-VDM, it can be concluded that it tends to be stable when the number of iterations is 12. At this time, the optimal value of the number of decompositions is 6, the optimal value of the bandwidth parameter is 1250, and the minimum envelope entropy is 0.31.

[0069] like Figure 2 As shown in the figure, the historical line loss data after ZOA-VMD decomposition can show significant periodicity, and the volatility and randomness of the data are significantly reduced, which improves the accuracy of the prediction. The accuracy of signal decomposition can be verified by reconstruction error. The calculation formula of reconstruction error is as follows:

[0070]

[0071] Among them, RE is the reconstruction error, x(t) is the original signal, is the reconstructed signal, Q is the signal length, and t is the time scale.

[0072] In traditional line loss prediction, VDM is usually used to decompose historical line loss data. However, the setting of VDM parameters is often performed through , which cannot fully extract the periodicity and trend information in the historical line loss data. This application uses the ZOA algorithm to optimize the parameters of VDM using envelope entropy as the objective function, which can decompose a single set of historical line loss data into multiple modal components. When the data set is small, it can learn enough effective information from the historical line loss data, effectively improve the robustness of the prediction framework, and lay a solid foundation for subsequent data prediction.

[0073] S130, inputting the historical line loss components and all pre-processed meteorological factor data into a feature extraction model of a pre-built line loss prediction network to obtain a correlation coefficient between each meteorological factor and the line loss data.

[0074] In an optional embodiment, the pre-built line loss prediction network includes a feature extraction model and a prediction model.

[0075] Optionally, the feature extraction model is CNN-LSTM-Attention (Convolutional Neural Network-Long Short-Term Memory-Attention, convolution-long short-term memory neural network combined with attention mechanism), where the LSTM network is the basis of the feature extraction model, which can specifically process time series data and better capture the long-term dependencies in the sequence data through memory and forgetting mechanisms. The input gate is used to update the unit information, the forget gate is used to decide whether to retain or reset the information, and the output gate is used to control the output of the information. The calculation process of the LSTM information flow is as follows:

[0076] i t =σ(W i [h t-1 ,x t ]+b i )

[0077] f t =σ(W f [h t-1 ,x t ]+b f )

[0078] o t =σ(W o [h t-1 ,x t ]+b o )

[0079]

[0080] h t =o t tanh(C t )

[0081] Among them, i t is the output of the input gate, which determines how much information of the current input is written into the cell state; t Indicates the input at the current moment; h t-1 is the hidden state at the previous moment; f t is the output of the forget gate, which determines how much information in the unit at the previous moment should be forgotten; t is the output at the current moment; is the candidate unit state; C t is the current unit state; h t is the hidden state of the output; W i , W f , W o , W C is the weight matrix of the model, bi 、b f 、b o 、b c is the offset term of the model; σ is the sigmoid activation function, and tanh is the hyperbolic tangent activation function.

[0082] In the feature extraction model, the CNN layer can perform feature extraction and recognition, which is used to learn the local trend characteristics of time series. This application mainly uses the convolution layer to effectively extract nonlinear local features of meteorological data, and uses the pooling layer to reduce the dimension of the data to generate key feature information. After fusing the CNN network, the CNN network is used for feature extraction to capture the correlation in the input data. This upper-layer network that integrates CNN and LSTM simplifies the feature engineering part in the unified line loss prediction, while achieving a high degree of parallelism in the feature extraction process, improving the efficiency of training and feature selection. The one-dimensional convolution formula is as follows:

[0083]

[0084] Among them, y t is the output data, k is the data length, ω k is the convolution kernel, x t-k+1 is the input data and b is the bias.

[0085] In the feature extraction process of time series, since the line loss time series has long-term dependence and seasonal periodicity, adding a self-attention mechanism in the process of extracting feature factors can automatically learn to assign different weights according to the importance of different parts of the input sequence. This can more effectively capture the key information in the sequence data. The calculation formula of the attention mechanism is as follows:

[0086]

[0087] Among them, k is the length of the data sequence, h * is the final result of the calculation, α i Calculate the weight value for the attention mechanism, h i For existing data.

[0088] The combined neural network of CNN-LSTM-Attention is used as the feature extraction model, which has better robustness and adaptability than traditional feature extraction methods. This method can accurately extract features from various types of historical line loss data, increase the adaptability of the entire prediction framework, and reduce the complexity of the line loss prediction framework.

[0089] In an optional embodiment, the historical line loss components and the preprocessed meteorological factor data set are input into a feature extraction model of a pre-built line loss prediction network to obtain a correlation coefficient between each meteorological factor in the meteorological factor data set and the line loss data, including:

[0090] The historical line loss components and all pre-processed meteorological factor data are converted into a time series; each time point of the time series includes the value of each historical line loss component and the value of each meteorological factor data;

[0091] The control feature extraction model is learned according to the time series to obtain the correlation coefficient between each meteorological factor and each historical line loss component.

[0092] Optionally, the feature extraction model reads the historical line loss component values ​​and the data of each meteorological factor at each time point step by step according to the time series, learns to capture the long-term dependency between each meteorological factor and the historical line loss data, and outputs the correlation coefficient between the meteorological factor and the line loss data.

[0093] S140, inputting the correlation coefficient and the historical line loss component into the prediction model of the line loss prediction network to obtain the line loss prediction result of the distribution network in the target area.

[0094] In an optional embodiment, the correlation coefficient and the historical line loss component are input into a prediction model of a line loss prediction network to obtain a line loss prediction result of a distribution network in a target area, including:

[0095] The correlation coefficient is converted into a weight parameter through a conversion formula;

[0096] The weight parameters and historical line loss components are input into the prediction model of the line loss prediction network to obtain multiple line loss component prediction values;

[0097] The predicted values ​​of multiple line loss components are superimposed to obtain the line loss prediction result of the distribution network in the target area.

[0098] In an optional embodiment, the conversion formula is:

[0099]

[0100] Among them, W i is the weight parameter of the ith meteorological factor, c i is the correlation coefficient of the ith meteorological factor, and n is the total number of meteorological factors.

[0101] Optionally, the prediction model is GA-BIGRU-LSTM (Genetic Algorithm-Bidirectional Gated Recurrent Unit-Long Short-Term Memory). First, the genetic algorithm generates an initial population, which consists of multiple individuals, each of which represents a potential solution for the prediction model parameters. By evaluating the fitness of each individual, that is, the prediction evaluation function of the prediction model, it is determined which individuals will be selected to generate the next generation. Secondly, the crossover process is introduced to generate offspring individuals by exchanging chromosome fragments of parent individuals, and then random mutations are introduced into the population through mutation operations to increase the diversity of the population and prevent falling into a local optimal solution. Finally, after completing the crossover and mutation operations, the fitness of the generated offspring individuals is evaluated, and they are used to replace a part of the individuals in the current population to ensure that the population evolves in the direction of the optimal solution. After repeated selection, crossover, mutation and replacement until the maximum number of iterations is met, the genetic algorithm will continuously optimize the population and gradually calculate the optimal solution for the prediction model parameters.

[0102] The BIGRU layer contains a two-directional loop structure: one processes the sequence from front to back, and the other processes the sequence from back to front, and simultaneously updates and outputs the hidden states in both directions, and connects them as the final feature representation. LSTM, as a subsequent processing layer, further extracts complex models and dependencies in the sequence, and controls the flow of information through three gating units: input gate, forget gate, and output gate, effectively dealing with the gradient vanishing and gradient explosion problems, and improving the prediction accuracy of the second-layer prediction model for long line loss sequences. After obtaining the prediction results of each component, linear superposition is performed in the full-link layer to obtain the final line loss prediction result. The evaluation index is calculated using the line loss prediction result and back-propagated to the first-layer feature extraction network to complete the update of the weight coefficient.

[0103] GRU combines the forget gate and the output gate into an update gate, which can effectively save time when training large amounts of data. The calculation formulas for the update gate, reset gate, and hidden state in the GRU network are as follows:

[0104] ν t =σ(W ν ·(Y t-1 ,Z t )+b ν )

[0105] r t =σ(W t ·(Y t-1 ,Z t )+b t .

[0106] hr =tanh(W i ·(r·Y t-1 ,Z t )+b i )

[0107] Y t =(1-ν t )·Y t-1 +ν t ·h t .

[0108] Among them, ν t The output of the update gate in the model determines how much information of the hidden state at the previous moment needs to be retained, and combines the information at the current moment to generate a new hidden state; Z t is the input at the current moment; r t The output of the reset gate controls how much information of the hidden state at the previous moment will be forgotten; h r is the candidate hidden state, which is calculated by the current input and the previous hidden state before resetting the control of the gate; Y t is the final hidden state, which is controlled by the update gate and makes a weighted average between the hidden state information at the previous moment and the candidate hidden state information; W ν , W t , W i is the weight matrix of the model, b ν 、b t 、b i is the bias term of the model; σ is the sigmoid activation function, and tanh is the hyperbolic tangent activation function.

[0109] The selection of various parameters in BIGRU-LSTM is the key to influencing the prediction accuracy. It is impossible to select the optimal network hyperparameters based on experience alone. Compared with other optimization algorithms, the GA algorithm has some unique advantages and characteristics in selecting hyperparameters. It can explore a wide range of search spaces, has strong global search capabilities, and performs well in complex, multi-peak optimization problems. The algorithm that combines GA with the prediction network can quickly and accurately find the optimal network parameters.

[0110] In an optional embodiment, the method further comprises:

[0111] Determine an error value based on the line loss prediction result;

[0112] If the error value does not meet the preset conditions, the historical line loss components and the multiple meteorological factor data corresponding to the meteorological factors corresponding to the target correlation coefficient are input into the line loss prediction network according to the preset rules to re-predict the line loss, obtain the line loss prediction result, and return to the step of determining the error value based on the line loss prediction result;

[0113] If the error value meets the preset conditions, the last line loss prediction result will be used as the final line loss prediction result.

[0114] The error value calculation formula is:

[0115]

[0116] Among them, MAE is the error value, n is the total number of data points, y i is the actual line loss value, is the predicted value of line loss.

[0117] Optionally, the preset condition is: the difference between the MAE obtained this time and the MAE calculated last time is calculated, and the MAE obtained this time is greater than the MAE calculated last time. That is, when the MAE value shows an upward trend, the line loss prediction result is used as the final line loss prediction result.

[0118] In an optional embodiment, according to a preset rule, the meteorological factor data corresponding to the historical line loss components and the target correlation coefficient are input into the line loss prediction network to re-predict the line loss, including:

[0119] Arrange the correlation coefficients in ascending order, and determine the number of each meteorological factor based on the sorting result;

[0120] The number of times the line loss prediction is performed is obtained, and a target correlation coefficient is determined based on the number of times;

[0121] The meteorological factor data corresponding to the historical line loss components and the target correlation coefficient are input into the line loss prediction network to re-predict the line loss.

[0122] In this embodiment, the relationship between the number of times the line loss prediction is performed and the target correlation coefficient is: the target correlation coefficient of the nth line loss prediction is one or more correlation coefficients ranked in the top n-1.

[0123] After obtaining the line loss prediction results, the prediction results can be evaluated using the following formulas:

[0124]

[0125] Where MAPE is the mean absolute percentage error, RMSE is the root mean square error, n is the number of measurements, and y is the i is the actual line loss value, is the predicted value of line loss.

[0126] The weighted correlation coefficient output by the feature extraction model is combined with the historical line loss components after ZOA-VMD decomposition as the input of the prediction model. The BIGRU layer contains two directional loop structures: one processes the sequence from front to back, and the other processes the sequence from back to front, and simultaneously updates and outputs the hidden states in both directions, and connects them as the final feature representation. LSTM, as a subsequent processing layer, further extracts the complex models and dependencies in the sequence, and controls the flow of information through three gating units: input gate, forget gate, and output gate, effectively dealing with the gradient vanishing and gradient explosion problems, and improving the prediction accuracy of the prediction module for long line loss sequences. After obtaining the prediction results of each component, the prediction results of each component are linearly superimposed to obtain the final line loss prediction results. Finally, the line loss prediction results are used to calculate the evaluation index and back-propagate to the feature extraction model to update the correlation coefficient and weight coefficient. It can exclude meteorological factors with weak correlation and meteorological factors that have an adverse effect on the prediction results, making the line loss prediction results more accurate.

[0127] In traditional prediction networks, due to the lack of feature engineering, it is impossible to filter and optimize input features by itself, resulting in large deviations in prediction accuracy and failure to adapt well to new data sets. The distribution network line loss prediction method based on meteorological factors proposed in this application can adaptively perform feature screening, and can feed back the correlation between meteorological factors and line loss data to the prediction model to re-predict the data. When the data set changes, the line loss prediction method provided in this application can better complete the line loss prediction work and achieve the desired prediction effect. Since the original data is preprocessed and the line loss prediction results can be verified and updated, this application can adapt to a variety of data sets.

[0128] like Figure 3 As shown, the implementation process of this application is as follows:

[0129] First, obtain the historical line loss data and meteorological factor dataset of the distribution network in the target area.

[0130] Second, all the historical line loss data and all the data in the meteorological factor data set are preprocessed to obtain standard historical line loss data and standard meteorological factor data set.

[0131] Third, the standard historical line loss data and standard meteorological factor data are divided into a training set and a test set. The standard historical line loss data in the test set is used as the preprocessed historical line loss data, and the standard meteorological factor data set in the test set is used as the preprocessed meteorological factor data set.

[0132] Fourth, the ZOA algorithm is used to optimize the VDM algorithm, and the ZOA-VDM algorithm is used to decompose the preprocessed historical line loss data to obtain multiple historical line loss components.

[0133] Fifth, the historical line loss components and all meteorological factor data in the preprocessed meteorological factor data set are input into the line loss prediction network, and the correlation coefficient between each meteorological factor and the line loss data is obtained through the feature extraction model of the line loss prediction network (that is, CNN-LSTM-Attention). The correlation is converted into a weight parameter, and the line loss prediction result of the distribution network in the target area is obtained through the prediction model of the line loss prediction network (that is, GA-BIGRU-LSTM), and the error value is calculated.

[0134] Sixth, sort the correlation coefficients between meteorological factor data and line loss data from large to small.

[0135] Seventh, the preprocessed meteorological factor data corresponding to the first-ranked correlation coefficient and the preprocessed historical line loss data are re-input into the line loss prediction network to obtain a new line loss prediction result and calculate a new error value.

[0136] Eighth, compare the new error value with the error value of the line loss prediction result obtained by the last line loss prediction.

[0137] If the new error value is smaller than the previous error value, the calculation times n are obtained, and the data corresponding to the top n-1 meteorological factors with the highest correlation are input into the line loss prediction network to re-predict the line loss, and the final line loss prediction result is obtained, and the final line loss prediction result is used as the line loss prediction result of the distribution network in the target area.

[0138] If the new error value is larger than the previous error value, the weight is re-determined according to the correlation coefficient of the previous error prediction, and the weight and the historical line loss component are input into the prediction model for prediction.

[0139] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean 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 embodiment of the present invention.

[0140] Corresponding to the distribution network line loss prediction method based on meteorological factors described in the above embodiment, Figure 4 A schematic structural diagram of a distribution network line loss prediction device based on meteorological factors provided by an embodiment of the present invention is shown. For ease of explanation, only the parts related to the embodiment of the present invention are shown.

[0141] See also Figure 4In an embodiment of the present invention, a distribution network line loss prediction device 4 based on meteorological factors may include:

[0142] The processing module 41 is used to obtain the historical line loss data and meteorological factor data set of the target area distribution network, and pre-process the historical line loss data and all meteorological factor data in the meteorological factor data set; wherein the meteorological factor data set includes multiple meteorological factors and multiple meteorological factor data corresponding to each meteorological factor;

[0143] A decomposition module 42, used to decompose the preprocessed historical line loss data based on a preset algorithm to obtain historical line loss components;

[0144] An extraction module 43 is used to input the historical line loss components and the pre-processed data of multiple meteorological factors into a feature extraction model of a pre-built line loss prediction network to obtain a correlation coefficient between each meteorological factor and the line loss data;

[0145] The prediction module 44 is used to input the correlation coefficient and the historical line loss component into the prediction model of the line loss prediction network to obtain the line loss prediction result of the distribution network in the target area.

[0146] In a possible implementation, the processing module 41 is specifically configured to:

[0147] Standardize the historical line loss data and meteorological factor data sets to obtain standard historical line loss data and standard meteorological factor data sets;

[0148] The standard historical line loss data and the standard meteorological factor data set are divided into a training set and a test set according to the preset division rules;

[0149] The standard historical line loss data and the standard meteorological factor data set in the training set are used as the preprocessed historical line loss data and meteorological factor data set.

[0150] In a possible implementation, the pre-built line loss prediction network includes a feature extraction model and a prediction model.

[0151] In a possible implementation, the decomposition module 42 is specifically configured to:

[0152] Optimizing the parameters of the variational mode decomposition algorithm to obtain a preset algorithm;

[0153] The preprocessed historical line loss data is decomposed based on a preset algorithm to obtain a preset number of historical line loss components.

[0154] In a possible implementation, the extraction module 43 is specifically configured to:

[0155] The historical line loss components and all pre-processed meteorological factor data are converted into a time series; each time point of the time series includes the value of each historical line loss component and the value of each meteorological factor data;

[0156] The control feature extraction model is learned according to the time series to obtain the correlation coefficient between each meteorological factor and each historical line loss component.

[0157] In a possible implementation, the prediction module 44 is specifically configured to:

[0158] The correlation coefficient is converted into a weight parameter through a conversion formula;

[0159] The weight parameters and historical line loss components are input into the prediction model of the line loss prediction network to obtain multiple line loss component prediction values;

[0160] The predicted values ​​of multiple line loss components are superimposed to obtain the line loss prediction result of the distribution network in the target area.

[0161] In a possible implementation, the conversion formula is:

[0162]

[0163] Among them, W i is the weight parameter of the ith meteorological factor, c i is the correlation coefficient of the ith meteorological factor, and n is the total number of meteorological factors.

[0164] In a possible implementation, the prediction module 44 is further configured to:

[0165] Determine an error value based on the line loss prediction result;

[0166] If the error value does not meet the preset conditions, the historical line loss components and the multiple meteorological factor data corresponding to the meteorological factors corresponding to the target correlation coefficient are input into the line loss prediction network according to the preset rules to re-predict the line loss, obtain the line loss prediction result, and return to the step of determining the error value based on the line loss prediction result;

[0167] If the error value meets the preset conditions, the last line loss prediction result will be used as the final line loss prediction result.

[0168] In a possible implementation, the prediction module 44 is further configured to:

[0169] Arrange the correlation coefficients in ascending order, and determine the number of each meteorological factor based on the sorting result;

[0170] The number of times the line loss prediction is performed is obtained, and a target correlation coefficient is determined based on the number of times;

[0171] The meteorological factor data corresponding to the historical line loss components and the target correlation coefficient are input into the line loss prediction network to re-predict the line loss.

[0172] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0173] Those of ordinary skill in the art will appreciate that the templates, units, and algorithm steps of each example 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 performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0174] If the module / unit is implemented in the form of 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, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of the above-mentioned various distribution network line loss prediction method embodiments based on meteorological factors. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electrical carrier signal, telecommunication signal and software distribution medium.

[0175] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A distribution network line loss prediction method based on meteorological factors, characterized in that: include: Acquire historical line loss data and a meteorological factor data set of the target area distribution network, and preprocess the historical line loss data and all meteorological factor data in the meteorological factor data set; wherein the meteorological factor data set includes multiple meteorological factors and multiple meteorological factor data corresponding to each meteorological factor; Decomposing the preprocessed historical line loss data based on a preset algorithm to obtain historical line loss components; Input the historical line loss components and all pre-processed meteorological factor data into a pre-built feature extraction model of a line loss prediction network to obtain a correlation coefficient between each meteorological factor and the line loss data; The correlation coefficient and the historical line loss component are input into the prediction model of the line loss prediction network to obtain the line loss prediction result of the target area distribution network.

2. The method for predicting line loss in a distribution network based on meteorological factors according to claim 1, characterized in that: The method further comprises: Determining an error value based on the line loss prediction result; If the error value does not meet the preset condition, then according to the preset rule, the historical line loss component and the multiple meteorological factor data corresponding to the meteorological factor corresponding to the target correlation coefficient are input into the line loss prediction network to re-perform the line loss prediction, to obtain the line loss prediction result, and return to the step of determining the error value based on the line loss prediction result; If the error value meets the preset condition, the last line loss prediction result is used as the final line loss prediction result.

3. The distribution network line loss prediction method based on meteorological factors according to claim 1, characterized in that: The step of inputting the historical line loss components and the preprocessed meteorological factor data set into a feature extraction model of a pre-built line loss prediction network to obtain a correlation coefficient between each meteorological factor in the meteorological factor data set and the line loss data includes: Converting the historical line loss components and all the pre-processed meteorological factor data into a time series; each time point of the time series includes the value of each historical line loss component and the value of each meteorological factor data; The feature extraction model is controlled to learn according to the time series to obtain the correlation coefficient between each meteorological factor and each historical line loss component.

4. The method for predicting line loss in a distribution network based on meteorological factors according to claim 1, characterized in that: The step of inputting the correlation coefficient and the historical line loss component into the prediction model of the line loss prediction network to obtain the line loss prediction result of the target area distribution network includes: The correlation coefficient is converted into a weight parameter by a conversion formula; Inputting the weight parameter and the historical line loss component into the prediction model of the line loss prediction network to obtain a plurality of line loss component prediction values; The multiple line loss component prediction values ​​are superimposed to obtain a line loss prediction result of the target area distribution network.

5. The method for predicting line loss in a distribution network based on meteorological factors according to claim 2, characterized in that: According to the preset rules, the meteorological factor data corresponding to the historical line loss components and the target correlation coefficient are input into the line loss prediction network to re-predict the line loss, including: Arrange the correlation coefficients in ascending order, and determine the number of each meteorological factor based on the sorting result; Obtaining the number of times line loss prediction is performed, and determining a target correlation coefficient based on the number of times; The meteorological factor data corresponding to the historical line loss components and the target correlation coefficient are input into the line loss prediction network to re-predict the line loss.

6. The method for predicting line loss in a distribution network based on meteorological factors according to claim 1, characterized in that: The preprocessing of the historical line loss data and the meteorological factor data set includes: Standardizing the historical line loss data and the meteorological factor data set to obtain standard historical line loss data and a standard meteorological factor data set; Dividing the standard historical line loss data and the standard meteorological factor data set into a training set and a test set according to a preset division rule; The standard historical line loss data and the standard meteorological factor data set in the training set are used as the preprocessed historical line loss data and the meteorological factor data set.

7. The method for predicting line loss in a distribution network based on meteorological factors according to claim 1, characterized in that: Decomposing the preprocessed historical line loss data based on a preset algorithm to obtain historical line loss components includes: Optimizing the parameters of the variational mode decomposition algorithm to obtain a preset algorithm; The preprocessed historical line loss data is decomposed based on the preset algorithm to obtain a preset number of historical line loss components.

8. The method for predicting line loss in a distribution network based on meteorological factors according to claim 1, characterized in that: The pre-built line loss prediction network includes a feature extraction model and a prediction model.

9. The method for predicting line loss in a distribution network based on meteorological factors according to claim 4, characterized in that: The conversion formula is: Among them, W i is the weight parameter of the ith meteorological factor, c i is the correlation coefficient of the ith meteorological factor, and n is the total number of meteorological factors.

10. A distribution network line loss prediction device based on meteorological factors, characterized in that: include: A processing module, used to obtain historical line loss data and a meteorological factor data set of the target area distribution network, and pre-process the historical line loss data and all meteorological factor data in the meteorological factor data set; wherein the meteorological factor data set includes multiple meteorological factors and multiple meteorological factor data corresponding to each meteorological factor; A decomposition module, used for decomposing the preprocessed historical line loss data based on a preset algorithm to obtain historical line loss components; An extraction module, used for inputting the historical line loss components and the preprocessed data of multiple meteorological factors into a feature extraction model of a pre-built line loss prediction network to obtain a correlation coefficient between each meteorological factor and the line loss data; The prediction module is used to input the correlation coefficient and the historical line loss component into the prediction model of the line loss prediction network to obtain the line loss prediction result of the distribution network in the target area.

Images