Regional electricity consumption prediction method based on hybrid model and terminal

Abstract

The invention discloses a regional electricity consumption prediction method based on a hybrid model and a terminal. The method comprises the following steps: acquiring historical electricity consumption data of a target region; performing data preprocessing according to the historical power consumption data to obtain an original data set; performing data processing on the original data set through an empirical mode decomposition model to obtain a combined data set; constructing an electric quantity prediction model based on a convolutional neural network model and a dual echo state network model, and training the electric quantity prediction model according to the combined data set; and predicting the future power consumption of the target area through the trained power consumption prediction model. According to the method, the spatial feature extraction capability of the convolutional neural network and the time sequence feature prediction capability of the echo state network are effectively fused, so that the electric quantity prediction model can effectively learn spatial features and time features from historical power consumption data; according to the electric quantity prediction model, the prediction precision is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electricity consumption prediction, and in particular to a method and terminal for predicting regional electricity consumption based on a hybrid model. Background Art

[0002] With the development of smart grids and current measurement infrastructure, short-term electricity consumption forecasting has received increasing attention. In fact, the prediction of future power load is a key method to avoid energy waste and develop effective power management strategies. Therefore, balancing power production and consumption is a pressing task, and its implementation depends largely on the means and methods of power production planning. Among them, accurate prediction of power demand is a key factor in optimizing power generation and consumption, saving energy, and determining energy prices. However, integrating highly nonlinear and seasonal energy mix scenarios, including meteorological environments, into existing power grids increases the uncertainty of power generation and brings additional challenges to accurate prediction. Early prediction methods based on machine learning are prone to data overfitting or poor prediction accuracy. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a method and terminal for predicting regional electricity consumption based on a hybrid model, which can achieve accurate prediction of regional electricity consumption.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0005] A regional electricity consumption prediction method based on a hybrid model, comprising:

[0006] Obtain historical electricity consumption data for the target area;

[0007] Performing data preprocessing according to the historical electricity consumption data to obtain an original data set;

[0008] Processing the original data set by an empirical mode decomposition model to obtain a combined data set;

[0009] Building a power prediction model based on a convolutional neural network model and a dual echo state network model, and training the power prediction model according to the combined data set;

[0010] The future electricity consumption of the target area is predicted by the trained electricity prediction model.

[0011] In order to solve the above technical problems, another technical solution adopted by the present invention is:

[0012] A regional electricity consumption prediction terminal based on a hybrid model comprises a memory, a processor and a computer program stored in the memory and running on the processor. When the processor executes the computer program, each step in the above-mentioned regional electricity consumption prediction method based on the hybrid model is implemented.

[0013] The beneficial effects of the present invention are: obtaining the historical electricity consumption data of the target area, and constructing an original data set based on the historical electricity consumption data, and processing the original data set through the empirical mode decomposition model, thereby screening the original data set, avoiding the occurrence of random errors, and improving the accuracy of subsequent model training. At the same time, the present invention constructs an electricity prediction model based on a convolutional neural network model and a dual echo state network model, effectively integrating the spatial feature extraction capability of the convolutional neural network and the temporal feature prediction capability of the echo state network, so that the electricity prediction model can effectively learn spatial features and temporal features from historical electricity consumption data. Compared with a single prediction model, the electricity prediction model of the present invention effectively improves the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A flowchart of a method for predicting regional electricity consumption based on a hybrid model provided by an embodiment of the present invention;

[0015] Figure 2 Another flow chart of a method for predicting regional electricity consumption based on a hybrid model provided by an embodiment of the present invention;

[0016] Figure 3 A schematic diagram of the architecture of a dual echo state network model provided by an embodiment of the present invention;

[0017] Figure 4 A schematic diagram of the structure of a regional power consumption prediction terminal based on a hybrid model provided by an embodiment of the present invention;

[0018] Description of labels:

[0019] 100. A regional electricity consumption prediction terminal based on a hybrid model; 101. Memory; 102. Processor. DETAILED DESCRIPTION

[0020] In order to explain the technical content, achieved objectives and effects of the present invention in detail, the following is an explanation in combination with the implementation modes and the accompanying drawings.

[0021] The embodiment of the present invention provides a method for predicting regional electricity consumption based on a hybrid model, including:

[0022] Obtain historical electricity consumption data for the target area;

[0023] Performing data preprocessing according to the historical electricity consumption data to obtain an original data set;

[0024] Processing the original data set by an empirical mode decomposition model to obtain a combined data set;

[0025] Building a power prediction model based on a convolutional neural network model and a dual echo state network model, and training the power prediction model according to the combined data set;

[0026] The future electricity consumption of the target area is predicted by the trained electricity prediction model.

[0027] From the above description, it can be seen that the beneficial effects of the present invention are: obtaining the historical electricity consumption data of the target area, and constructing an original data set based on the historical electricity consumption data, and processing the original data set through the empirical mode decomposition model (EMD), thereby screening the original data set to avoid the occurrence of random errors and improve the accuracy of subsequent model training. At the same time, the present invention constructs an electricity prediction model based on a convolutional neural network model (CNN) and a dual echo state network model (DESN), which effectively integrates the spatial feature extraction capability of the convolutional neural network and the temporal feature prediction capability of the echo state network, so that the electricity prediction model can effectively learn spatial features and temporal features from historical electricity consumption data. Compared with a single prediction model, the electricity prediction model of the present invention effectively improves the prediction accuracy.

[0028] Furthermore, the data preprocessing according to the historical electricity consumption data to obtain the original data set includes:

[0029] Eliminate abnormal values ​​in the historical power consumption data by using a moving average method, and normalize the historical power consumption data to obtain standard data;

[0030] The outliers in the standard data were eliminated by the interquartile range method to obtain the original data set.

[0031] From the above description, it can be seen that the historical electricity consumption data may contain outliers with sudden energy consumption peaks. These outliers will make it difficult for subsequent models to predict future energy consumption. Therefore, when constructing the original data set, the outliers in the historical electricity consumption data should be removed as much as possible to reduce the random error of the original data set and improve the training accuracy of the subsequent model.

[0032] Furthermore, the step of processing the original data set by using an empirical mode decomposition model to obtain a combined data set includes:

[0033] Decomposing the original data set by an empirical mode decomposition model to obtain an initial feature set; calculating the correlation coefficient between each feature component in the initial feature set and the original data set based on the Pearson correlation coefficient method;

[0034] Select the feature components whose correlation coefficients are greater than a preset threshold as the target feature set;

[0035] The target feature set is combined with the original data set to obtain a combined data set.

[0036] From the above description, it can be seen that the original data set is decomposed through the EMD model to obtain the initial feature set composed of the decomposed feature components, and then the correlation analysis of the decomposed feature components is performed based on the Pearson correlation coefficient method. The feature components with greater correlation with the original data set are selected as the input data of the subsequent power prediction model, which can effectively avoid the occurrence of random errors and improve the prediction accuracy of the model.

[0037] Furthermore, the power prediction model includes a plurality of hybrid prediction models with different reserve pool spectrum radii and a fully connected layer;

[0038] The step of constructing a power prediction model based on a convolutional neural network model and a dual echo state network model, and training the power prediction model according to the combined data set includes:

[0039] The convolutional neural network model is fused with multiple dual echo state network models with different reservoir spectral radii to obtain multiple hybrid prediction models;

[0040] Performing feature learning and processing on the combined data set through the hybrid prediction model respectively to obtain output data of each hybrid prediction model;

[0041] The output data of each hybrid prediction model is fused through a fully connected layer to obtain target data;

[0042] The hybrid prediction model and the fully connected layer are iteratively optimized according to the combined data set and the target data.

[0043] From the above description, we can see that the CNN model can effectively extract the spatial features in the data, while the DESN model can effectively extract and process the forward and backward dependency features of the data in the time series. Therefore, the hybrid prediction model (CDESN) based on the CNN model and the DESN model can effectively learn the spatial and temporal features of the data, thereby accurately predicting future data. At the same time, since the state of the echo state network with a larger spectral radius of the reserve pool is unstable, and the network memory ability of the echo state network with a smaller spectral radius is weak, multiple CDESN models with different spectral radii of the reserve pool are constructed, and finally the output data of each CDESN model is fused through the fully connected layer to eliminate the errors caused by different spectral radius performance and improve the prediction accuracy of the model.

[0044] Further, the dual echo state network model includes a coding echo state network, a decoding echo state network and a potential connection layer;

[0045] The combining data set is subjected to feature learning and processing by the hybrid prediction model respectively to obtain output data of each hybrid prediction model, including:

[0046] Inputting the combined data set into the convolutional neural network model in each of the hybrid prediction models to extract initial data features;

[0047] Inputting the initial data features into the coded echo state network to perform feature dimension reduction to obtain dimension-reduced data features;

[0048] The dimension-reduced data features are input into the decoding echo state network through the potential connection layer for data decoding to obtain output data of each of the hybrid prediction models.

[0049] As can be seen from the above description, the DESN model of the present invention includes two parts: an encoding network and a decoding network, that is, an ESN network for encoding and an ESN network for decoding. The encoding part maps the combined data set to a latent space of a preset dimension, thereby achieving feature dimensionality reduction, while the decoding part reconstructs the dimensionality reduction features from the latent space and brings them back to the original space for output. Compared with the traditional ESN model, the DESN model of the present invention can effectively learn the high-level feature representation of the data, thereby better capturing the intrinsic structural characteristics of the data.

[0050] Furthermore, the initial data features are input into the coded echo state network for feature dimension reduction, and the dimension-reduced data features obtained are specifically:

[0051]

[0052] Among them, x(t+1) represents the state of the reserve pool of the coded echo state network at time t+1, and f ε represents the activation function of the coded echo state network, W in represents the input weight of the coded echo state network, u(t+1) represents the initial data features input into the reserve pool of the coded echo state network at time t+1, represents the weight of the reserve pool of the coded echo state network, x(t) represents the state of the reserve pool of the coded echo state network at time t, and b ε represents the bias vector of the coded echo state network, z(t+1) represents the dimension-reduced data feature output by the reserve pool of the coded echo state network at time t+1, Represents the input weights of the latent connection layer.

[0053] From the above description, it can be seen that the state of the reserve pool of the coded echo state network depends on the input initial data features, the state at the previous moment and the reserve pool weight of the current coded echo state network, and the dimension reduction data features of the coded echo state network are obtained based on the state of the reserve pool at the current moment and the input weight of the potential connection layer. The input weight of the potential connection layer is iteratively optimized, thereby forcing the reserve pool weight of the coded echo state network to be optimized, thereby completing the optimization training of the coded echo state network.

[0054] Furthermore, the dimension reduction data features are input into the decoding echo state network through the potential connection layer for data decoding, and the output data of each hybrid prediction model is obtained as follows:

[0055]

[0056] in, represents the state of the reserve pool of the decoded echo state network at time t+1, f D represents the activation function of the decoded echo state network, represents the output weight of the potential connection layer, z(t+1) represents the dimension-reduced data feature output by the reserve pool of the coded echo state network at time t+1, represents the weight of the reservoir of the decoding echo state network, represents the state of the reserve pool of the decoded echo state network at time t, b D represents the bias vector of the decoded echo state network, represents the output data of the reserve pool of the decoded echo state network at time t+1, W out represents the output weights of the decoded echo state network.

[0057] From the above description, it can be seen that the decoding echo state network decodes the reduced dimension data features based on its reserve pool weights and determines the state of the reserve pool of the current decoding echo state network. Then, the output data is reconstructed by combining the reserve pool state of the current decoding echo state network and the output weight obtained by the current training. Among them, the potential connection layer serves as a bridge connecting the encoding network and the decoding network, effectively adjusting the weight parameters between the two networks, so that the model parameters are optimized.

[0058] Furthermore, the historical electricity consumption data is normalized to obtain standard data as follows:

[0059]

[0060] Denormalized(a)=Normalized(a)*(max(a)-min(a))+min(a);

[0061] Among them, Normalized(a) represents standard data, x(a) represents historical electricity consumption data, min(a) and max(a) represent the minimum and maximum values ​​of historical electricity consumption data respectively, and Denormalized(a) represents converting standard data into its original value.

[0062] From the above description, it can be seen that since different types of historical electricity consumption data may have different dimensions and dimensional units, the historical electricity consumption data is normalized so that the historical electricity consumption data is limited to a certain range, thereby eliminating the adverse effects caused by singular sample data.

[0063] Furthermore, the outliers in the standard data are eliminated by the interquartile range method to obtain the original data set as follows:

[0064] Calculate the upper quartile and the lower quartile of the standard data;

[0065] Calculate the difference between the upper quartile and the lower quartile to obtain the interquartile range;

[0066] Determine an outlier range according to the upper quartile, the lower quartile, and the interquartile range;

[0067] The outliers in the standard data are determined according to the outlier value range and removed to obtain the original data set.

[0068] From the above description, it can be seen that after data normalization, the interquartile range method can effectively identify outliers with large deviations, thereby eliminating outliers that are ignored in the standard data due to omission errors or deviations from the normal statistical distribution of the standard data.

[0069] Another embodiment of the present invention provides a regional electricity consumption prediction terminal based on a hybrid model, including a memory, a processor, and a computer program stored in the memory and running on the processor, and when the processor executes the computer program, it implements each step of the above-mentioned regional electricity consumption prediction method based on the hybrid model.

[0070] From the above description, it can be seen that the beneficial effects of the present invention are: obtaining the historical electricity consumption data of the target area, and constructing an original data set based on the historical electricity consumption data, and processing the original data set through the empirical mode decomposition model (EMD), thereby screening the original data set to avoid the occurrence of random errors and improve the accuracy of subsequent model training. At the same time, the present invention constructs an electricity prediction model based on a convolutional neural network model (CNN) and a dual echo state network model (DESN), which effectively integrates the spatial feature extraction capability of the convolutional neural network and the temporal feature prediction capability of the echo state network, so that the electricity prediction model can effectively learn spatial features and temporal features from historical electricity consumption data. Compared with a single prediction model, the electricity prediction model of the present invention effectively improves the prediction accuracy.

[0071] The embodiments of the present invention provide a regional power consumption prediction method and terminal based on a hybrid model, which can be applied to the power load prediction of the power system to achieve accurate prediction of regional power consumption, so as to help power system operators to reasonably plan and manage loads and power generation. The following is an explanation through specific embodiments:

[0072] Please refer to Figure 1 and Figure 2 , Embodiment 1 of the present invention is:

[0073] A regional electricity consumption prediction method based on a hybrid model, comprising:

[0074] S1. Obtain historical electricity consumption data of the target area.

[0075] In some embodiments, the historical electricity consumption data includes historical electricity consumption and related data corresponding to the historical electricity consumption. The historical electricity consumption is the electricity consumption of the target area in each period of each week and each day; the related data includes electricity consumption analysis data and meteorological data affecting electricity consumption, the analysis data includes electricity consumption trends, seasonal fluctuation trends, holiday electricity consumption effects, etc., and the meteorological data includes temperature, etc.

[0076] S2. Perform data preprocessing according to the historical electricity consumption data to obtain an original data set.

[0077] Specifically, the step S2 includes:

[0078] S21. Eliminate abnormal values ​​in the historical electricity consumption data by using a moving average method, and normalize the historical electricity consumption data to obtain standard data.

[0079] The historical electricity consumption data is normalized to obtain standard data as follows:

[0080]

[0081] Denormalized(a)=Normalized(a)*(max(a)-min(a))+min(a) (2)

[0082] Among them, Normalized(a) represents standard data, x(a) represents historical electricity consumption data, min(a) and max(a) represent the minimum and maximum values ​​of historical electricity consumption data respectively, and Denormalized(a) represents converting standard data into its original numerical value. The above normalization processing formula can be used to normalize historical electricity consumption data to the range of 0-1.

[0083] S22. Eliminate outliers in the standard data by using the interquartile range method to obtain an original data set.

[0084] It should be noted that the interquartile range refers to the difference between the fourth percentile (i.e., the upper quartile) and the third percentile (i.e., the lower quartile) of the data. Therefore, the interquartile range of the data will follow the 25% decomposition point.

[0085] In some embodiments, the upper quartile, lower quartile and interquartile range of the data are calculated based on the interquartile range method, and which data are considered to be outliers can be determined according to preset determination rules.

[0086] In an optional implementation, the step S22 is specifically:

[0087] S221. Calculate the upper quartile and the lower quartile of the standard data.

[0088] S222: Calculate the difference between the upper quartile and the lower quartile to obtain the interquartile range.

[0089] Specifically, the interquartile range IQR=Q3-Q1, where Q3 represents the upper quartile and Q1 represents the lower quartile.

[0090] S223. Determine an outlier range according to the upper quartile, the lower quartile, and the interquartile range.

[0091] S224. Determine the outliers in the standard data according to the outlier value range and remove them to obtain an original data set.

[0092] In some embodiments, data smaller than Q1-1.5IQR or larger than Q3+1.5IQR may be determined as outliers based on the upper quartile Q3, the lower quartile Q1, and the interquartile range IQR.

[0093] S3. Processing the original data set by using an empirical mode decomposition model to obtain a combined data set.

[0094] Specifically, Figure 2 As shown, step S3 includes:

[0095] S31. Decomposing the original data set by using an empirical mode decomposition model to obtain an initial feature set.

[0096] In some embodiments, the original data set X is processed by the EMD model. s After data decomposition, we can get the initial feature set X, X = {imf1, imf2, imf3, ... imf n-1 ,Res}, where imf i and Res represent the decomposed IMF components (Intrinsic Mode Functions) and residuals, respectively, and n represents the total number of decomposed time series components.

[0097] S33. Calculate the correlation coefficient between each feature component in the initial feature set and the original data set based on the Pearson correlation coefficient method.

[0098] As a filtering feature selection method, the Pearson correlation coefficient has strong versatility and low complexity, and has a strong advantage in processing large-scale data sets, and can eliminate a large number of irrelevant features in a short time. Therefore, the present invention adopts the Pearson correlation coefficient method to select and filter the initial feature set obtained by EMD decomposition, wherein the calculation formula of the Pearson correlation coefficient is specifically:

[0099]

[0100] Among them, γ represents the correlation coefficient, that is, the correlation between different feature components and the original data set. i and i Represent different feature components and original data sets respectively, and They represent the average values ​​of different feature components and the original data set respectively, and n represents the number of samples.

[0101] S34, selecting the feature components whose correlation coefficients are greater than a preset threshold as a target feature set.

[0102] In some embodiments, if the correlation coefficient r of the feature component i If it is greater than the preset threshold t, the feature component is added to the target feature set X f ; If the correlation coefficient r of the characteristic component i If the correlation is less than or equal to the preset threshold t, the feature component is removed, thereby filtering the feature components in the initial feature set whose correlation is less than the threshold.

[0103] S35: Merge the target feature set with the original data set to obtain a combined data set.

[0104] In some embodiments, the combined data set Data = {X s , X f}.

[0105] Empirical mode decomposition (EMD) is a signal processing technique that can decompose non-stationary signals into a series of intrinsic mode functions (IMFs). The present invention uses EMD to decompose the collected original data set to extract its inherent patterns and trends. EMD can decompose signals of different frequencies step by step according to the characteristics of the data itself, thereby obtaining several periodic and trendy orthogonal signals. At present, the general prediction method based on EMD is to establish prediction models for the decomposed subsequences respectively, and then superimpose the output of each prediction model to obtain the final prediction result. Although this method improves the prediction accuracy, it will introduce multiple random errors. At this time, if a prediction model is established for each subsequence respectively, some high-frequency noise components will exist in the decomposed subsequences, which will cause these subsequences to generate larger prediction errors when modeling, affecting the overall prediction accuracy. At the same time, due to the establishment of multiple prediction models, the overall complexity of the model will increase. The EMD model of the present invention will perform correlation analysis on the decomposed subsequences, thereby selecting components with greater correlation with the original sequence as the input features of the model to avoid the occurrence of random errors and improve the accuracy of subsequent prediction models.

[0106] S4. Constructing a power prediction model based on a convolutional neural network model and a dual echo state network model, and training the power prediction model according to the combined data set.

[0107] S5. Predicting the future electricity consumption of the target area by using the trained electricity prediction model.

[0108] In some embodiments, based on the trained power prediction model, it is possible to predict the power consumption of the target area on the same day, that is, to predict the short-term energy consumption of the target area.

[0109] Please refer to Figure 2 and Figure 3 , Embodiment 2 of the present invention is:

[0110] A method for predicting regional electricity consumption based on a hybrid model is different from the first embodiment in that the specific steps of step S4 are limited.

[0111] The power prediction model includes a plurality of hybrid prediction models with different reserve pool spectrum radius and a fully connected layer, and the hybrid prediction model is represented as C-DESN. Then, step S4 includes:

[0112] S41. The convolutional neural network model is respectively integrated with a plurality of dual echo state network models with different reserve pool spectrum radii to obtain a plurality of hybrid prediction models.

[0113] Specifically, the dual echo state network model (DESN, i.e., Dual-ESN) of the present invention includes two echo state networks (ESNs), one of which is used to encode and map data, and the other is used to decode and map data, and a trainable code layer is introduced between the two ESNs to generate an optimized potential representation of the input. Therefore, the dual echo state network model includes an encoding echo state network, a decoding echo state network, and a potential connection layer. Figure 3 As shown, the dual echo state network model (DESN) also includes an input layer and an output layer, and the input layer is sequentially connected to the encoding echo state network (hereinafter referred to as the encoding part), the potential connection layer, the decoding echo state network (hereinafter referred to as the decoding part) and the output layer. Among them, the encoding part maps the original data to an m-dimensional latent space (m is smaller than the dimension of the original input data), the decoding part reconstructs the input from the latent space and brings it back to the original space, and the latent connection layer is a trainable code layer.

[0114] Among them, ESN is a special type of recurrent neural network (RNN) that replaces the hidden layer in RNN with a reserve pool. Compared with the RNN model, the ESN model uses a fixed recursive neural network as a repository and uses a linear learning method to simply modify the output weight matrix, effectively improving computational efficiency.

[0115] Each ESN model includes a memory, an input layer, and an output layer. If the number of input neurons is represented by K, the number of storage neurons is represented by N, and the number of output neurons is represented by L, W in , W, W back , and W out Represent the weights of the input matrix, storage matrix, feedback matrix and output matrix respectively. The sizes of the input matrix, storage matrix, feedback matrix and output matrix are N×K, N×N, N×L and L×(K+N) respectively. The ESN model is trained through a supervised learning process, and its training process mainly includes updating the state of the reserve pool and converting the weight matrix of the reserve pool into the weight matrix W of the output layer. out .

[0116] Let u = u(n) represent the external input of the ESN model, x = x(n) represent the reserve state of the ESN model, and y = y(n) represent the output vector of the ESN model. For each ESN model, the calculation process of the input signal from the input layer to the reserve pool is as follows:

[0117] x(t+1)=f(W in u(t+1),Wx(t)+Wback y(t)) (4)

[0118] Then the output of the ESN model at time t+1 is:

[0119] y(t+1)=f out (W out x(t+1)) (5)

[0120] Among them, f represents the sigmoid function in the reserve layer, f out Represents the output layer activation function.

[0121] Therefore, the DESN model can be expressed as:

[0122] z=f ε (W ε u+b ε ) (6)

[0123]

[0124] Among them, formula (6) describes the calculation method of the encoding part in the DESN model, f ε represents the activation function of the encoding part, W ε represents the weight matrix of the encoding part. Formula (7) describes the calculation method of the decoding part in the DESN model, f D represents the activation function of the decoding part, W D represents the weight matrix of the decoding part. Formula (8) describes the loss function of the DESN model It calculates the original input u and the reconstructed input The norm between .

[0125] Based on the architecture setting of the DESN model of the present invention, it has the advantage of keeping most of the weights in the network fixed when training the network to reconstruct the sequence passed as input.

[0126] S42, performing feature learning and processing on the combined data set through the hybrid prediction model respectively to obtain output data of each hybrid prediction model.

[0127] Specifically, the step S42 includes:

[0128] S421. Input the combined data set into the convolutional neural network model in each of the hybrid prediction models to extract initial data features.

[0129] In some embodiments, a convolutional neural network model (CNN) includes two 1-dimensional convolutional layers, a pooling layer, and a Flatten layer. The CNN model is used to extract corresponding initial data features from the lag sequence.

[0130] In some embodiments, since the initial data features extracted by the CNN model are two-dimensional data, and the ESN model processes time series, the initial data features need to be converted from two-dimensional data to one-dimensional data before inputting them into the encoded ESN.

[0131] S422: Input the initial data features into the coded echo state network to perform feature dimension reduction to obtain dimension-reduced data features.

[0132] The step S422 is specifically as follows:

[0133]

[0134]

[0135] Among them, x(t+1) represents the state of the reserve pool of the coded echo state network at time t+1, and f ε represents the activation function of the coded echo state network, W in represents the input weight of the coded echo state network, u(t+1) represents the input data of the reserve pool of the coded echo state network at time t+1, represents the weight of the reserve pool of the coded echo state network, x(t) represents the state of the reserve pool of the coded echo state network at time t, and b ε represents the bias vector of the coded echo state network, z(t+1) represents the dimension-reduced data feature output by the reserve pool of the coded echo state network at time t+1, Represents the input weights of the latent connection layer.

[0136] S423: Input the dimension-reduced data features into the decoding echo state network through the potential connection layer for data decoding to obtain output data of each of the hybrid prediction models.

[0137] The step S423 is specifically as follows:

[0138]

[0139] in, represents the state of the reserve pool of the decoded echo state network at time t+1, f D represents the activation function of the decoded echo state network, represents the output weight of the potential connection layer, z(t+1) represents the dimension-reduced data feature output by the reserve pool of the coded echo state network at time t+1, represents the weight of the reservoir of the decoding echo state network, represents the state of the reserve pool of the decoded echo state network at time t, b Drepresents the bias vector of the decoded echo state network, represents the output data of the reserve pool of the decoded echo state network at time t+1, W out represents the output weights of the decoded echo state network.

[0140] It should be noted that the DESN model of the present invention uses two separate ESN models, so that the encoding process and decoding process of the DESN model are decoupled, thereby The weights are responsible for learning the input dynamics and their encoding, while The weights are used for decoding and input reconstruction. At the same time, the potential connection layer is introduced as a bridge between the encoding part and the decoding part. The weights force the encoding part to generate an optimized encoding representation. The number of neurons in the potential connection layer is fixed to m to control the number of encoded features.

[0141] S43. The output data of each hybrid prediction model is fused through a fully connected layer to obtain target data.

[0142] S44. Iteratively optimize the hybrid prediction model and the fully connected layer according to the combined data set and the target data.

[0143] like Figure 2 As shown, the present invention constructs an original data set based on historical electricity consumption data, and decomposes the original data set through the EMD model to obtain an initial feature set composed of IMF components and residual Res, and then determines the correlation between each feature component in the initial feature set and the original data set based on the Pearson correlation coefficient method, so as to select feature components with greater correlation with the original data set as the target feature set, and merge the original data set and the target feature set to obtain a combined data set. At the same time, the combined data set is respectively input into the C-DESN model obtained by fusing the CNN model with the DESN model of different reserve pool spectrum radius for processing, and the prediction results of each C-DESN model are merged by the fully connected layer to obtain the final electricity consumption prediction result. The above-mentioned C-DESN model and the fully connected layer are iteratively optimized based on the prediction results to obtain the optimal electricity prediction model.

[0144] Please refer to Figure 4 , Embodiment 3 of the present invention is:

[0145] A regional electricity consumption prediction terminal 100 based on a hybrid model includes a memory 101, a processor 102, and a computer program stored in the memory 101 and running on the processor 102. When the processor 102 executes the computer program, each step in a regional electricity consumption prediction method based on a hybrid model as described in Example 1 is implemented.

[0146] In summary, the present invention provides a method and terminal for predicting regional electricity consumption based on a hybrid model, obtains historical electricity consumption data of a target area, and constructs an original data set based on the historical electricity consumption data, processes the original data set through an EMD model, obtains a target feature set with a high correlation with the original data set, and merges the original data set and the target feature set to obtain a combined data set, thereby avoiding the occurrence of random errors and improving the accuracy of subsequent model training. At the same time, the combined data set is respectively input into a C-DESN model obtained by fusing a CNN model with a DESN model of different reserve pool spectrum radius for processing, and the prediction results of each C-DESN model are fused by a fully connected layer to obtain a final prediction result. Based on the prediction results, the above C-DESN model and the fully connected layer are iteratively optimized to obtain the optimal electricity prediction model. The electricity prediction model of the present invention combines the CNN model with the DESN model, which can not only effectively learn spatial features from the combined data set, but also extract the temporal features of the data. In addition, a dual reserve pool network is introduced in the DESN model to perform encoding and decoding operations respectively, so that the model can better learn the high-level feature representation of the data, thereby better capturing the intrinsic structure of the data. Compared with a single prediction model, the power prediction model of the present invention effectively improves the prediction accuracy.

[0147] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for predicting regional electricity consumption based on a hybrid model, characterized in that: include: Obtain historical electricity consumption data for the target area; Performing data preprocessing according to the historical electricity consumption data to obtain an original data set; Processing the original data set by an empirical mode decomposition model to obtain a combined data set; Building a power prediction model based on a convolutional neural network model and a dual echo state network model, and training the power prediction model according to the combined data set; The future electricity consumption of the target area is predicted by the trained electricity prediction model.

2. The method according to claim 1, characterized in that: The performing data preprocessing according to the historical electricity consumption data to obtain the original data set comprises: Eliminate abnormal values ​​in the historical power consumption data by using a moving average method, and normalize the historical power consumption data to obtain standard data; The outliers in the standard data were eliminated by the interquartile range method to obtain the original data set.

3. The method according to claim 1, characterized in that The step of processing the original data set by using the empirical mode decomposition model to obtain the combined data set includes: Performing data decomposition on the original data set by using an empirical mode decomposition model to obtain an initial feature set; Calculate the correlation coefficient between each feature component in the initial feature set and the original data set based on the Pearson correlation coefficient method; Select the feature components whose correlation coefficients are greater than a preset threshold as the target feature set; The target feature set is combined with the original data set to obtain a combined data set.

4. The method according to claim 1, characterized in that: The power prediction model includes a plurality of hybrid prediction models with different reserve pool spectrum radius and a fully connected layer; The step of constructing a power prediction model based on a convolutional neural network model and a dual echo state network model, and training the power prediction model according to the combined data set includes: The convolutional neural network model is fused with multiple dual echo state network models with different reservoir spectral radii to obtain multiple hybrid prediction models; Performing feature learning and processing on the combined data set through the hybrid prediction model respectively to obtain output data of each hybrid prediction model; The output data of each hybrid prediction model is fused through a fully connected layer to obtain target data; The hybrid prediction model and the fully connected layer are iteratively optimized according to the combined data set and the target data.

5. The method according to claim 4, characterized in that The dual echo state network model includes an encoding echo state network, a decoding echo state network and a potential connection layer; The combining data set is subjected to feature learning and processing by the hybrid prediction model respectively to obtain output data of each hybrid prediction model, including: Inputting the combined data set into the convolutional neural network model in each of the hybrid prediction models to extract initial data features; Inputting the initial data features into the coded echo state network to perform feature dimension reduction to obtain dimension-reduced data features; The dimension-reduced data features are input into the decoding echo state network through the potential connection layer for data decoding to obtain output data of each of the hybrid prediction models.

6. The method according to claim 5, characterized in that The initial data features are input into the coded echo state network to perform feature dimension reduction, and the dimension-reduced data features are obtained as follows: Among them, x(t+1) represents the state of the reserve pool of the coded echo state network at time t+1, and f ε represents the activation function of the coded echo state network, W in represents the input weight of the coded echo state network, u(t+1) represents the input data of the reserve pool of the coded echo state network at time t+1, represents the weight of the reserve pool of the coded echo state network, x(t) represents the state of the reserve pool of the coded echo state network at time t, and b ε represents the bias vector of the coded echo state network, z(t+1) represents the dimension-reduced data feature output by the reserve pool of the coded echo state network at time t+1, Represents the input weights of the latent connection layer.

7. The method according to claim 5, characterized in that The dimension reduction data features are input into the decoded echo state network through the potential connection layer for data decoding, and the output data of each hybrid prediction model is obtained as follows: in, represents the state of the reserve pool of the decoded echo state network at time t+1, f D represents the activation function of the decoded echo state network, represents the output weight of the potential connection layer, z(t+1) represents the dimension-reduced data feature output by the reserve pool of the coded echo state network at time t+1, represents the weight of the reservoir of the decoding echo state network, represents the state of the reserve pool of the decoded echo state network at time t, b D represents the bias vector of the decoded echo state network, represents the output data of the reserve pool of the decoded echo state network at time t+1, W out represents the output weights of the decoded echo state network.

8. The method according to claim 2, characterized in that: The historical electricity consumption data is normalized to obtain standard data as follows: Denormalized(a)=Normalized(a)*(max(a)-min(a))+min(a); Among them, Normalized(a) represents standard data, x(a) represents historical electricity consumption data, min(a) and max(a) represent the minimum and maximum values ​​of historical electricity consumption data respectively, and Denormalized(a) represents converting standard data into its original value.

9. The method according to claim 2, characterized in that: The outliers in the standard data are removed by the interquartile range method to obtain the original data set as follows: Calculate the upper quartile and the lower quartile of the standard data; Calculate the difference between the upper quartile and the lower quartile to obtain the interquartile range; Determine an outlier range according to the upper quartile, the lower quartile, and the interquartile range; The outliers in the standard data are determined according to the outlier value range and removed to obtain the original data set.

10. A regional electricity consumption prediction terminal based on a hybrid model, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the processor implements the various steps in the method for predicting regional electricity consumption based on a hybrid model as described in any one of claims 1 to 9.

Images