Short-term residential electricity consumption prediction method based on feature fusion graph neural network

Abstract

The invention discloses a short-term residential electricity consumption prediction method based on a feature fusion graph neural network, and belongs to the technical field of short-term electricity consumption prediction of a power system. The method overcomes the defects of a traditional method in the aspects of capturing the space-time dependency relationship and feature fusion, and is mainly technically characterized by comprising the steps that residential historical electricity consumption data and auxiliary feature data disclosed in the real world are collected and preprocessed, and the data are processed through an input model; building a parallel time convolution structure; constructing a three-dimensional dynamic adjacency matrix through a reverse tri-modal factor analysis method; designing a multi-feature fusion module, and dynamically introducing the time-space relationship of the auxiliary features into main feature prediction; and introducing a residual connection and jump connection structure, finally constructing a multi-feature fusion space-time diagram neural network model, and verifying the effectiveness of the model through an evaluation index. The result shows that the accuracy of power dispatching and energy consumption management can be improved, and technical support is provided for management of the smart power grid and renewable energy sources.

Description

Technical Field

[0001] The present invention relates to the technical field of short-term electricity consumption prediction for power systems, and in particular to a method for predicting short-term residential electricity consumption using a feature fusion graph neural network. Background Art

[0002] With the rapid growth of electricity demand in modern society, short-term residential electricity consumption forecasting has become a key link in power system management. Accurate electricity consumption forecasting is of great significance for grid dispatch optimization, energy consumption management, and efficient use of renewable energy. However, since residential electricity consumption patterns are affected by multiple complex factors such as weather, time, and user behavior, they have significant nonlinearity and spatiotemporal dependence, and traditional forecasting methods are difficult to effectively capture these characteristics.

[0003] At present, the prediction methods based on time series mainly focus on the changes in the time dimension of historical data, ignoring the potential spatial associations between residences. In addition, the prediction models based on a single feature usually cannot fully utilize multi-source data, such as auxiliary features such as weather, holidays and user behavior, which limits the prediction accuracy of the model.

[0004] In recent years, Graph Neural Network (GNN) has attracted widespread attention due to its ability to process non-Euclidean data structures and has shown significant advantages in modeling complex relationships between nodes. However, the existing GNN framework still has shortcomings in combining spatiotemporal characteristics and multi-feature fusion, making it difficult to balance prediction accuracy and model robustness. Summary of the invention

[0005] A method for predicting short-term residential electricity consumption based on a feature fusion graph neural network, characterized in that the method comprises the following operations:

[0006] First, using the public dataset provided by the OpenEI official website, we mainly selected the historical electricity consumption data of 15 residences in Louisiana (LA), USA, and two auxiliary feature historical data of the residences.

[0007] Secondly, the residential electricity consumption dataset and the auxiliary feature dataset are processed through the input module respectively and compiled into a set of datasets.

[0008] Furthermore, a parallel time convolution structure is built to represent the temporal relationship between electricity consumption. By introducing a three-dimensional dynamic adjacency matrix, the complex spatial relationship between electricity consumption is represented, and a dynamic graph convolution structure is constructed.

[0009] Then, a multi-feature fusion structure is designed to dynamically add auxiliary features into the prediction of electricity consumption.

[0010] Finally, the entire multi-feature fusion spatiotemporal graph neural network is constructed. The entire model is subsequently represented by DFF-GWN. The model includes input and output modules, time modules, space modules and multi-feature fusion modules. The model is then verified using public data sets, and the effectiveness of the model is judged by evaluation indicators.

[0011] The present invention is a short-term residential electricity consumption prediction method based on a feature fusion graph neural network. There are many auxiliary features in the data set used. For the selection of these auxiliary features, the correlation coefficient analysis method is used to calculate the coefficient between each auxiliary feature and the electricity consumption. The two selected auxiliary features are determined according to the size of the coefficient. The correlation coefficient comprehensively considers the Pearson coefficient and the Spearman coefficient. The Pearson coefficient and the Spearman coefficient are shown as follows: In the formula, and They are Pearson coefficient and Spearman coefficient. and are the observation values ​​of auxiliary features and electricity consumption, and are the means of auxiliary features and power consumption, yes and The ranking difference, is the time step; for electricity consumption data, all data are added together to obtain the total electricity consumption as the main feature, which is the prediction object.

[0012] The processing of electricity consumption and auxiliary features is to normalize the data separately, and then splice all the data in the feature dimension, instead of directly preprocessing all the data and compiling them into a set of data sets for easy input into the model.

[0013] The steps of constructing the three-dimensional dynamic adjacency matrix proposed by the present invention include:

[0014] The total time step The days are divided into multiple cycles, and the number of time slots in each cycle is , and then construct the adjacency matrix for each cycle.

[0015] Using the inverse trimodal factor analysis method, the original adjacency matrix is ​​reconstructed by constructing three factor loading matrices and one core matrix. , incorporating the temporal relationship into the construction of the adjacency matrix.

[0016] In this process, three types of adjacency matrices are generated, namely, the adjacency matrix representing the main features , represents the adjacency matrix of auxiliary features , here Only one auxiliary feature is represented, and the remaining features are used Indicates that the subsequent analysis only uses To replace all auxiliary features, it is convenient for subsequent analysis, representing the adjacency matrix of the fusion of main features and auxiliary features , Represents the number of main feature nodes, Represents the number of secondary nodes.

[0017] The three 3D dynamic graph structures generated are: main feature graph , auxiliary feature map And the main feature and auxiliary feature fusion map .

[0018] The construction process of the three-dimensional dynamic adjacency matrix specifically includes:

[0019] Step (1) utilizes the temporal correlation of residential electricity consumption in adjacent time periods to share a graph structure for the 24-hour electricity consumption relationship, thereby dividing all time steps into 24-hour cycles. , so that the original adjacency matrix The complexity is reduced to .

[0020] The goal of the trimodal factor analysis method used in step (2) is to construct a core matrix and the factor loading matrix To reconstruct the original data matrix , and the multiple matrices constructed are low-dimensional matrices, thus achieving the purpose of dimensionality reduction. The core formula is as follows: In the formula, is the slot factor loading matrix, is the input node factor loading matrix, is the output node factor loading matrix and is the core matrix, where is the number of input nodes, is the number of output nodes, is the embedding dimension, where the factor loading matrix is a two-dimensional matrix, is a three-dimensional matrix,

[0021] The above two formulas are respectively for non-negative processing and row normalization processing of the created adjacency matrix, and the three-dimensional dynamic adjacency matrix finally used is obtained.

[0022] Step (3) is to construct three three-dimensional adjacency matrices that meet the needs of the model based on the three-dimensional adjacency matrix construction method of step (2), namely the three-dimensional adjacency matrices of the main features and the auxiliary features. and , here, since multiple auxiliary features are used, only one is used to represent all auxiliary features for ease of description. It is a three-dimensional adjacency matrix used in the multi-feature fusion module. Different from the conventional method of directly splicing auxiliary features to the main features for prediction, in the multi-feature fusion module, To dynamically simulate the spatiotemporal relationship between the main features and the auxiliary features.

[0023] Step (4) generates a graph structure based on the three 3D adjacency matrices obtained in step (3). A set of nodes and a set of edges Composition, of which Represents a node, express and The edges between them are the connections between the electricity consumption of multi-family houses and their auxiliary features, representing their spatial dependencies. In this process, a diffusion graph convolutional network is used to handle the spatial dependencies between houses by aggregating the hidden states of adjacent houses. The formula is: In the formula, yes The input of the dynamic graph convolution layer, is the output after being updated by the dynamic graph convolution layer. Similarly, the feature output is , the auxiliary feature output is , for The learnable parameters of the diffusion, is the total number of diffusion steps, which produces the main feature map , auxiliary feature map And the main feature and auxiliary feature fusion map .

[0024] The multi-feature fusion structure proposed in the present invention dynamically adds auxiliary features to the prediction of electricity consumption, specifically in the prediction of residential electricity consumption. At the same time, the auxiliary features will not only affect the main features of the current time, but also affect the main features of the adjacent time. This relationship can be understood as a progressive relationship in which the auxiliary features affect the main features through mutual influence. The module takes into account the spatiotemporal dependency between the nodes of the auxiliary features. In each layer, a separate spatiotemporal block is allocated to the main feature and the auxiliary feature, and the output of the main feature and the output of the auxiliary feature are obtained, which are respectively input into their respective multi-feature fusion modules, that is, and , after the aggregation process, the final fused output is obtained, the formula is: In the formula, is the dynamic graph convolution layer, and It is the output of the main feature and auxiliary feature after the dynamic graph convolution layer. It is the output after multi-feature fusion. represents the aggregation process of two results. The purpose of using direct addition is to make the entire model calculation concise and maintain high representation ability. The adjacency matrix of time slot t is expressed as ,in is a function that gets the number of time slots in a cycle, represents the learnable parameters, Represents the output result after the multi-feature fusion module.

[0025] The entire multi-feature fusion spatiotemporal graph neural network model constructed by the present invention includes input and output modules, the data is transformed into the model through a two-dimensional convolutional network, and the data after passing through the spatiotemporal module is output through an activation function and a two-dimensional convolutional network to output the prediction result. In order to improve the performance and training efficiency of the model, residual connections and jump connection structures are added to the entire network. The jump connection structure integrates the spatial dependencies of different time scales. The formula is: In the formula, is the final residential electricity consumption forecast result, It is a splicing operation. is the output of the time module in each layer, is the number of space-time layers.

[0026] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: (1) The processing of auxiliary features in the present invention is different from the previous direct splicing. Instead, it is processed separately and an independent spatiotemporal module is allocated to capture the spatiotemporal correlation of auxiliary features; (2) The present invention uses a diffusion convolution layer to dynamically update the three-dimensional adjacency matrix, and can operate on dynamic graphs corresponding to different times; (3) The present invention uses a three-modal factor method to reconstruct the three-dimensional adjacency matrix, and also incorporates the time correlation into the construction of the adjacency matrix, thereby reducing the complexity of adjacency matrix modeling and capturing the spatial dependency between residential electricity consumption in a more delicate manner; (4) The present invention designs a multi-feature fusion module, which uses a three-dimensional adjacency matrix to dynamically aggregate the spatiotemporal relationship between the main features and the auxiliary features, thereby improving the accuracy of prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a specific flow chart of a method for predicting short-term residential electricity consumption using a feature fusion graph neural network according to the present invention;

[0028] Figure 2 It is a structural diagram of a DFF-GWN short-term residential electricity consumption prediction model in an example of the present invention;

[0029] Figure 3 It is a three-dimensional dynamic adjacency matrix construction diagram in an example of the present invention. DETAILED DESCRIPTION

[0030] The present invention is further described below in conjunction with the accompanying drawings. The embodiments are to more clearly illustrate the technical solution of the present invention, and the application scope of the present invention is not limited thereto and can be applied to many fields. Here is just one example.

[0031] like Figure 1 As shown, the present invention provides a method for predicting short-term residential electricity consumption using a feature fusion graph neural network, which is specifically divided into the following steps:

[0032] First, using the public data set provided by the OpenEI official website, we mainly selected the historical electricity consumption data of 15 residences in Louisiana (LA), USA, and the historical data of two auxiliary features of the residences. The historical electricity consumption data of these 15 residences includes the historical electricity consumption data per hour in 2012. Each household records the electricity consumption of various electrical equipment every hour, and a total of 8760 data points are recorded. The two auxiliary features are Electricity: Facility (EF) and Electricity: HVAC (EH).

[0033] Secondly, the residential electricity consumption dataset and the two auxiliary feature datasets are processed through the input module and compiled into a set of datasets. Assume that the existing residential electricity consumption dataset and the two auxiliary feature datasets are and , the two are collectively referred to as ,in Represents the historical time step of In the following example analysis, only To represent two auxiliary features, and the input module passes through a two-dimensional convolutional layer, then the process is as follows: In the formula, and It is the output of the main features and auxiliary features after 2D CNN, which will be used as the input of the time module. and are all learnable parameters.

[0034] Furthermore, a parallel time convolution structure is built to represent the temporal relationship between power consumption. By introducing a three-dimensional dynamic adjacency matrix, the complex spatial relationship between power consumption is represented, and a dynamic graph convolution structure is constructed. In the process of building a parallel time convolution structure, a dilated causal convolution structure is used, and a gated DCL is formed by controlling two parallel dilation convolution layers (DCL) through a gating mechanism, and then respectively passes through and The processing of the two activation functions is finally performed element-by-element product operation. The whole process is as follows: In the formula, yes The output of the module temporal convolutional layer, where the main feature is , the auxiliary feature is , and are the learnable parameters of the convolution filter, is the Sigmoid activation function.

[0035] By introducing a three-dimensional dynamic adjacency matrix, the complex spatial relationship between electricity consumption is represented, and a dynamic graph convolution structure is constructed. This process considers a matrix decomposition method - trimodal factorization method. The goal of this idea is to use the core matrix and the factor loading matrix To reconstruct the original data matrix , and decompose it into low-dimensional representations of multiple patterns. Its core formula is as follows: In the formula, is the original data matrix (three-dimensional) elements, representing sample, Variables and The observed value at a time point, is the factor loading matrix Elements in (two dimensions) that describe patterns (Sample) and derived model The relationship between (factors), is the factor loading matrix Elements in (two dimensions) that describe patterns (Variables) and inference patterns The relationship between (factors), is the factor loading matrix Elements in (two dimensions) that describe patterns (Time point) and derived model The relationship between (factors), Core Matrix The elements in (three dimensions) describe the interaction between factors. It is the core part of the model and captures the interaction between different modes.

[0036] In this invention, the reverse trimodal factor analysis method is used to reconstruct the original adjacency matrix by constructing three factor loading matrices and one core matrix. , incorporating the time relationship into the construction of the adjacency matrix. The process specifically includes:

[0037] Step (1) utilizes the temporal correlation of residential electricity consumption in adjacent time periods to share a graph structure for the 24-hour electricity consumption relationship, thereby dividing all time steps into 24-hour cycles. , so that the original adjacency matrix The complexity is reduced to .

[0038] The goal of the trimodal factor analysis method used in step (2) is to construct a core matrix and the factor loading matrix To reconstruct the original data matrix , and the multiple matrices constructed are low-dimensional matrices, thus achieving the purpose of dimensionality reduction. The core formula is as follows: In the formula, is the slot factor loading matrix, is the input node factor loading matrix, is the output node factor loading matrix and is the core matrix, where is the number of input nodes, is the number of output nodes, is the embedding dimension, where the factor loading matrix is a two-dimensional matrix, is a three-dimensional matrix, The above two formulas are respectively for non-negative processing and row normalization processing of the created adjacency matrix, and the three-dimensional dynamic adjacency matrix finally used is obtained.

[0039] Step (3) is to construct three three-dimensional adjacency matrices that meet the needs of the model based on the three-dimensional adjacency matrix construction method of step (2), namely the three-dimensional adjacency matrices of the main features and the auxiliary features. and , here, since multiple auxiliary features are used, only one is used to represent all auxiliary features for ease of description. It is a three-dimensional adjacency matrix used in the multi-feature fusion module. Different from the conventional method of directly splicing auxiliary features to the main features for prediction, in the multi-feature fusion module, To dynamically simulate the spatiotemporal relationship between the main features and the auxiliary features , Represents the number of main feature nodes, Represents the number of secondary nodes.

[0040] Step (4) generates a graph structure based on the three 3D adjacency matrices obtained in step (3). A set of nodes and a set of edges Composition, of which Represents a node, express and The edges between them are the connections between the electricity consumption of multi-family houses and their auxiliary features, representing their spatial dependencies. In this process, a diffusion graph convolutional network is used to handle the spatial dependencies between houses by aggregating the hidden states of adjacent houses. The formula is: In the formula, yes The input of the dynamic graph convolution layer, is the output after being updated by the dynamic graph convolution layer. Similarly, the feature output is , the auxiliary feature output is , for The learnable parameters of the diffusion, is the total number of diffusion steps, which produces the main feature map , auxiliary feature map And the main feature and auxiliary feature fusion map .

[0041] Then, a multi-feature fusion structure is designed. This module takes into account the spatiotemporal dependencies between the nodes of the auxiliary features. In each layer, a separate spatiotemporal block is assigned to the main feature and the auxiliary feature, and the output of the main feature and the output of the auxiliary feature are obtained, which are input into their respective multi-feature fusion modules, that is, and , after the aggregation process, the final fused output is obtained, the formula is: In the formula, is the dynamic graph convolution layer, and It is the output of the main feature and auxiliary feature after the dynamic graph convolution layer. It is the output after multi-feature fusion. represents the aggregation process of two results. The purpose of using direct addition is to make the entire model calculation concise and maintain high representation ability. The adjacency matrix of time slot t is expressed as ,in is a function that gets the number of time slots in a cycle, represents the learnable parameters, Represents the output result after the multi-feature fusion module.

[0042] Finally, the entire multi-feature fusion spatiotemporal graph neural network (DFF-GWN) is constructed, and the data after the spatiotemporal module is output through the activation function and the two-dimensional convolutional network to output the prediction result. In order to improve the performance and training efficiency of the model in the entire network, residual connection and jump connection structure are also added. The jump connection structure integrates the spatial dependencies of different time scales. The formula is: In the formula, is the final residential electricity consumption forecast result, It is a splicing operation. is the output of the time module in each layer, is the number of space-time layers. The final output is as follows: In the formula, is the output of the model after skip connection, It’s the future Forecasted residential electricity consumption for the time step.

[0043] In the example of the present invention, the effectiveness of the present invention is verified by using common evaluation indicators in the neural network verification method, and the mean absolute percentage error (MAPE [%]) and the mean absolute error (MAE [kW h]) are selected for verification. The formulas of the two are: In the formula, yes The predicted residential electricity consumption at the time, yes The actual value of residential electricity consumption at the time, is the number of time steps.

[0044] The prediction results are shown in Table 1. When other model conditions are the same, the model proposed in the present invention is compared with three advanced models, namely LSTM, CNN-GRU and Ada-GWN, to predict the average electricity consumption of 15 households in the next hour, and the evaluation index is used to judge the effectiveness of the model.

[0045] Table 1 Model evaluation index results Model MAPE [%] MAE [kW h] LSTM 8.68 0.78 CNN-GRU 9.86 0.89 Ada-GWN 7.62 0.72 Prediction model of the present invention 6.36 0.62 From the results in Table 1, it can be seen that the prediction model GFF-GWN of the present invention has the best effect, which can prove the advantages of the present invention.

Claims

1. A method for predicting short-term residential electricity consumption based on feature fusion graph neural network, characterized in that: The method comprises the following operations: first, using the public data set provided by the OpenEI official website, the historical electricity consumption data of 15 residences in Louisiana (LA) of the United States and two auxiliary feature historical data of the residences are mainly selected; Secondly, the residential electricity consumption dataset and the auxiliary feature dataset are processed through the input module respectively and compiled into a set of datasets; then, a parallel time convolution structure is built to represent the temporal relationship between electricity consumption. By introducing a three-dimensional dynamic adjacency matrix, the complex spatial relationship between electricity consumption is represented, and a dynamic graph convolution structure is constructed; then, a multi-feature fusion structure is designed to dynamically add auxiliary features to the prediction of electricity consumption; finally, the entire multi-feature fusion spatiotemporal graph neural network is constructed, and the entire model is subsequently represented by DFF-GWN. The model includes input and output modules, time modules, space modules and multi-feature fusion modules, and then the model is verified using public datasets, and the effectiveness of the model is judged by evaluation indicators.

2. According to the method for predicting short-term residential electricity consumption using a feature fusion graph neural network according to claim 1, it is characterized in that: There are many auxiliary features in the data set used. For the selection of these auxiliary features, the correlation coefficient analysis method is used to calculate the coefficient between each auxiliary feature and the power consumption. The two auxiliary features selected are determined according to the size of the coefficient. The correlation coefficient comprehensively considers the Pearson coefficient and the Spearman coefficient. The Pearson coefficient and the Spearman coefficient are shown as follows: In the formula, and They are Pearson coefficient and Spearman coefficient. and are the observation values ​​of auxiliary features and electricity consumption, and are the means of auxiliary features and power consumption, yes and The ranking difference, is the time step; for electricity consumption data, all data are added together to obtain the total electricity consumption as the main feature, which is the prediction object.

3. According to the method of claim 1, the feature fusion graph neural network short-term residential electricity consumption prediction method is characterized in that: The processing of the short-term residential electricity consumption and its auxiliary characteristics is to perform data normalization processing separately, and then splice all the data in the feature dimension, rather than directly pre-processing all the data and compiling them into a set of data sets for easy input into the model.

4. The method for predicting short-term residential electricity consumption using a feature fusion graph neural network according to claim 1 is characterized in that: The steps of constructing the three-dimensional dynamic adjacency matrix include: (1) The total time step The days are divided into multiple cycles, and the number of time slots in each cycle is , and then construct the adjacency matrix for each cycle; (2) Using the inverse trimodal factor analysis method, the original adjacency matrix is ​​reconstructed by constructing three factor loading matrices and one core matrix. , incorporating temporal relations into the construction of the adjacency matrix; (3) In this process, three types of adjacency matrices are generated, namely, the adjacency matrix representing the main features , represents the adjacency matrix of auxiliary features , here Only one auxiliary feature is represented, and the remaining features are used Indicates that only one letter is used in subsequent analysis To replace all auxiliary features, it is convenient for subsequent analysis, representing the adjacency matrix of the fusion of main features and auxiliary features , Represents the number of main feature nodes, Represents the number of auxiliary nodes; (4) The three types of 3D dynamic graph structures generated: Main feature graph , auxiliary feature map And the main feature and auxiliary feature fusion map .

5. The method for predicting short-term residential electricity consumption using a feature fusion graph neural network according to claim 4 is characterized in that: The construction process of the three-dimensional dynamic adjacency matrix specifically includes: Step (1) utilizes the temporal correlation of residential electricity consumption in adjacent time periods to share a graph structure for the 24-hour electricity consumption relationship, thereby dividing all time steps into 24-hour cycles. , so that the original adjacency matrix The complexity is reduced to ; The goal of the trimodal factor analysis method used in step (2) is to construct a core matrix and the factor loading matrix To reconstruct the original data matrix , and the multiple matrices constructed are low-dimensional matrices, thereby achieving the purpose of dimensionality reduction. The core formula is as follows: In the formula, is the slot factor loading matrix, is the input node factor loading matrix, is the output node factor loading matrix and is the core matrix, where is the number of input nodes, is the number of output nodes, is the embedding dimension, where the factor loading matrix is a two-dimensional matrix, is a three-dimensional matrix, The above two formulas are respectively the non-negative processing and row normalization processing of the created adjacency matrix, and the final three-dimensional dynamic adjacency matrix is ​​obtained; Step (3) is to construct three three-dimensional adjacency matrices that meet the needs of the model based on the three-dimensional adjacency matrix construction method of step (2), namely the three-dimensional adjacency matrices of the main features and the auxiliary features. and , here, since multiple auxiliary features are used, only one is used to represent all auxiliary features for ease of description. It is a three-dimensional adjacency matrix used in the multi-feature fusion module. Different from the conventional method of directly splicing auxiliary features to the main features for prediction, in the multi-feature fusion module, To dynamically simulate the spatiotemporal relationship between the main features and the auxiliary features; Step (4) generates a graph structure based on the three 3D adjacency matrices obtained in step (3). A set of nodes and a set of edges Composition, of which Represents a node, express and The edges between them are the connections between the electricity consumption of multi-family houses and their auxiliary features, representing their spatial dependencies. In this process, a diffusion graph convolutional network is used to handle the spatial dependencies between houses by aggregating the hidden states of adjacent houses. The formula is: In the formula, yes The input of the dynamic graph convolution layer, is the output after being updated by the dynamic graph convolution layer. Similarly, the feature output is , the auxiliary feature output is , for The learnable parameters of the diffusion, is the total number of diffusion steps, which produces the main feature map , auxiliary feature map And the main feature and auxiliary feature fusion map .

6. The method for predicting short-term residential electricity consumption using a feature fusion graph neural network according to claim 1 is characterized in that: The multi-feature fusion structure dynamically adds auxiliary features to the prediction of electricity consumption. Specifically, in the prediction of residential electricity consumption, the auxiliary features at the same time will not only affect the main features at the current time, but also affect the main features at adjacent times. This relationship can be understood as a progressive relationship in which the auxiliary features affect the main features through mutual influence. The module takes into account the spatiotemporal dependency between the nodes of the auxiliary features. In each layer, a separate spatiotemporal block is allocated to the main feature and the auxiliary feature, and the output of the main feature and the output of the auxiliary feature are obtained, which are respectively input into their respective multi-feature fusion modules, that is, and , after the aggregation process, the final fused output is obtained, the formula is: In the formula, is the dynamic graph convolution layer, and It is the output of the main feature and auxiliary feature after the dynamic graph convolution layer. is the output after multi-feature fusion, represents the aggregation process of two results. The purpose of using direct addition is to make the entire model calculation concise and maintain high representation ability. The adjacency matrix of time slot t is expressed as ,in is a function that gets the number of time slots in a cycle, represents the learnable parameters, Represents the output result after the multi-feature fusion module.

7. The method for predicting short-term residential electricity consumption using a feature fusion graph neural network according to claim 1 is characterized in that: The entire multi-feature fusion spatiotemporal graph neural network model constructed includes input and output modules, which transform the data into a model through a two-dimensional convolutional network and output the prediction results of the data after passing through the spatiotemporal module through an activation function and a two-dimensional convolutional network. In order to improve the performance and training efficiency of the model, residual connections and jump connection structures are added to the entire network. The jump connection structure integrates spatial dependencies of different time scales. The formula is: In the formula, is the final residential electricity consumption forecast result, It is a splicing operation. is the output of the time module in each layer, is the number of space-time layers.

Images