Gas power station power prediction method

Abstract

The invention discloses a method for predicting the power of a gas power station. The method specifically comprises the steps of 1, collecting historical data of the gas power station as sample data; 2, preprocessing the sample data; 3, constructing the preprocessed data into a space-time diagram H; 4, designing a self-adaptive extensible space-time diagram convolutional network model; 5, taking the preprocessed data and the adjacent matrix as a training set, and training the adaptive extensible space-time diagram convolutional network model; and 6, predicting the power of the gas power station by adopting the trained adaptive extensible space-time diagram convolutional network model. The method provided by the invention shows higher precision, better stability, stronger trend tracking capability and better abnormal value processing capability when the gas power generation power is predicted.

Description

Technical Field

[0001] The invention belongs to the technical field of gas power plants, and in particular relates to a power prediction method for a gas power plant. Background Art

[0002] Gas power stations play an important role in modern power systems, and accurate prediction of their power output is of great significance for the stable operation and optimal dispatch of power systems. However, the power output of gas power stations is affected by many complex factors, including weather conditions, gas supply, coal production, energy demand, etc. There are complex nonlinear relationships and spatiotemporal dependencies between these factors, which makes the number and nodes of gas power stations grow dynamically, and their power prediction is more difficult.

[0003] Traditional gas power plant power forecasting methods mainly rely on time series models and machine learning models. These methods include autoregressive integrated moving average model (ARIMA), support vector regression (SVR), etc. Time series models (such as long short-term memory recursive neural network (LSTM)) make predictions by analyzing the time dependence of historical data, but they mainly assume the linear relationship of data and are difficult to capture the complex nonlinear characteristics in the data. In addition, these models often ignore the spatial correlation of gas power plant power output, resulting in limited prediction accuracy; at the same time, most of these models manually schedule model parameters, and the adaptability and scalability of power forecasting under the dynamic growth of the number of gas power plants and nodes are poor.

[0004] In recent years, with the development of deep learning technology, graph convolutional neural networks (GCN) and temporal graph convolutional neural networks (ST-GCN) have been widely used in fields such as traffic flow prediction and power load prediction. GCN can effectively capture the spatial dependencies in the data by performing convolution operations on the graph structure; ST-GCN can simultaneously capture the temporal and spatial characteristics of the data by combining temporal convolution and graph convolution. Although GCN and ST-GCN have performed well in traffic flow and power load prediction, there are no examples of using GCN and ST-GCN for prediction in the field of gas power plant power prediction. In addition, the power output of gas power plants is not only affected by external factors such as weather and gas supply, but also has significant spatiotemporal heterogeneity and dynamics. Existing methods still have limitations in dealing with these unique spatiotemporal heterogeneity and dynamics. Summary of the invention

[0005] Purpose of the invention: In order to solve the problems existing in the above-mentioned prior art, the present invention provides a power prediction method for a gas power plant.

[0006] Technical solution: The present invention provides a method for predicting power of a gas power plant, which specifically comprises the following steps: Step 1: Collect historical data of gas power plants as sample data; Step 2: Preprocess the sample data; Step 3: construct the preprocessed data into a spatiotemporal graph H; H = (V, E, A), V represents a node set, the node represents a gas power station, E represents an edge set, the edge is a connection edge between nodes, and A is an adjacency matrix; Step 4: Design an adaptive and scalable spatiotemporal graph convolutional network model; Step 5: Use the preprocessed data and adjacency matrix as training sets to train the adaptive scalable spatiotemporal graph convolutional network model; Step 6: Use the trained adaptive scalable spatiotemporal graph convolutional network model to predict the power of the gas power station.

[0007] Furthermore, the sample data in step 1 includes historical power, weather data related to power generation, and gas supply data.

[0008] Furthermore, the data preprocessing in step 2 is specifically to first clean the data and then normalize it. The data cleaning is specifically to: process the missing power data using the forward filling or backward filling method; then use the Z-score method to detect and process outliers; and convert the timestamp into a datetime type.

[0009] Furthermore, in step 3, a Gaussian kernel function is used to construct an adjacency matrix, and the expression of the adjacency matrix A is: ; Among them, i, j represent nodes i, j in the node set respectively; p i and p j Represent the geographical locations of node i and node j respectively, d ( i , j ) is the correlation coefficient between the historical powers of node i and node j, and a and k are parameters that control the sparsity of the adjacency matrix.

[0010] In the next step, the adaptive scalable spatiotemporal graph convolutional network model is configured to perform the following operations: The spatiotemporal convolution block is used to capture the spatiotemporal characteristics of the gas power station, and the feature extraction capability is improved by stacking L layers of spatiotemporal convolution blocks. Then, the adaptive module is used to adjust the model parameters. The scalable module is used to expand according to the increase in the amount of gas power generation data and the number of nodes. Finally, the predicted power of the gas power station is output through the fully connected layer.

[0011] Furthermore, each spatiotemporal convolution block uses graph convolution operations to extract the features of each node and neighborhood of the gas power plant, uses the one-dimensional convolution operation in the time convolution to extract the time features, and uses the gated linear unit in the time convolution to capture the temporal dynamic behavior of the gas power plant. The structure of the spatiotemporal convolution block is as follows: ; in, Indicates The output features of the spatiotemporal convolutional block of the layer, TemporalConv(.) represents the temporal convolution operation, s(.) represents the activation function, and GraphConv(.) represents the graph convolution operation. Indicates The output features of the layer graph convolution, Indicates The output features of the layer temporal convolution, Indicates The bias of the 1D convolution in the layer temporal convolution, represents the normalized adjacency matrix, , where J represents the degree matrix, Indicates The weight matrix of the layer graph convolution; , , X is the sample data; The structure of graph convolution is as follows: ; in, Indicates The output features of the layer graph convolution, T k Indicates k The Chebyshev polynomial of order, K represents the total order of Chebyshev polynomial; represents the normalized graph Laplacian operator, , I represents the identity matrix, Indicates The weight matrix of the k-th order Chebyshev polynomial in the layer graph convolution, The weight matrix of K Chebyshev polynomials in the layer graph convolution constitutes the Weight matrix of layer graph convolution ; The structure of temporal convolution is as follows: ; in, Indicates The output features of the layer temporal convolution, Indicates The output features of the one-dimensional convolution in the layer time convolution, Indicates The convolution kernel of the one-dimensional convolution in the layer time convolution, Indicates The bias of the 1D convolution in the layer temporal convolution, Indicates The convolution kernel of the gated linear unit in the layer temporal convolution, Indicates Bias of the gated linear unit in the layer temporal convolution, * indicates the convolution operation.

[0012] Furthermore, the adaptive module includes M adaptive layers connected in sequence, and the adaptive layers include node feature transformation, adaptive attention weight calculation and attention weight normalization; The node feature transformation is specifically as follows: ; Among them, F (m+1) Represents the node feature matrix output by the m+1th adaptive layer, and the node feature matrix output by the 1st adaptive layer ,in represents the features output by the spatiotemporal convolution module, represents the weight matrix of the first adaptive layer, and s(.) represents the activation function; represents the weight matrix of the m+1th adaptive layer, m=1,2,3,…,M; M represents the total number of adaptive layers; The adaptive attention weight meter is specifically: ; in, represents the adaptive attention weight between node i and node j in the mth adaptive layer; LeakyReLU represents the linear rectification activation function with leakage, || represents the vector connection, and They represent the features of nodes i and j in the mth adaptive layer respectively; represents the weight matrix of the mth adaptive layer, is the set of normalized adaptive attention weights in the mth adaptive layer; The adaptive attention weight normalization is specifically: ; in, represents the normalized value of the adaptive attention weight between node i and node j in the mth adaptive layer, V(i) represents the set of neighboring nodes of node i, represents the total number of nodes in the set V(i), represents the node i in the mth adaptive layer and the node in the set V(i) Adaptive attention weights between .

[0013] Furthermore, the scalable module includes N scalable layers connected in sequence, and the scalable layers include multi-head graph convolution and layered aggregation. The multi-head graph convolution is as follows: ; in, represents the feature matrix of node i output by the n+1th layer of multi-head graph convolution, Q represents the number of attention heads, represents the weight matrix of the qth attention head in the n+1th layer of multi-head graph convolution, Represents the feature matrix of node i output by the n-th layer multi-head graph convolution, and the normalized adaptive attention in the M-th adaptive layer Decompose into Q attention heads, use the Q attention heads as the Q attention heads in the first layer of multi-head graph convolution, and the feature matrix of node i output by the first layer of multi-head graph convolution is The expression is , represents the attention weight between node i and node j calculated by the qth attention head in the first layer of multi-head graph convolution, represents the node feature matrix output by the Mth adaptive layer, .

[0014] An electronic device / system for a gas power plant power prediction method comprises a processor and a memory, wherein the memory stores execution instructions of the processor, and the processor is configured to execute the execution instructions to implement the gas power plant power prediction method.

[0015] A computer-readable storage medium is used to store a program, and the program is executed to implement the power prediction method of the gas power plant.

[0016] Beneficial effects: The present invention combines the advantages of graph convolutional neural networks and time convolutional networks, and can effectively capture the complex spatiotemporal dependencies in the power data of gas power plants. At the same time, an adaptive module is introduced to dynamically adjust the model parameters to adapt to the changing characteristics of gas power generation, and an expandable module is introduced to enable the model to expand as the amount of data and the number of nodes in the gas power plant increase, thereby improving the power prediction accuracy of the gas power plant while enhancing the adaptability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The present invention is a flow chart of the method.

[0018] Figure 2 A block diagram of the method of the present invention.

[0019] Figure 3 This is a comparison chart of the effects of three prediction methods. DETAILED DESCRIPTION

[0020] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0021] like Figure 1 , 2 shows a method for predicting power of a gas power plant provided by the present invention, comprising the following steps: 1) Collect historical data of gas power plants, and perform data cleaning and normalization processing, wherein the historical data includes power output data, weather data, and gas supply data; For this embodiment, first collect the historical power output data of the gas power station and record the power output value every hour or every minute; collect weather data related to power generation, including temperature, humidity, wind speed, rainfall and air pressure; obtain gas supply data, including gas supply volume, pressure and temperature. Secondly, clean the collected data and use forward filling or backward filling methods to process missing power data; use the Z-score method to detect and process outliers. Convert the timestamp to datetime type to facilitate time series processing. Finally, use the minimum-maximum normalization method to normalize the data and standardize the output data.

[0022] (1) The historical data of the gas power plant can be expressed as: D={P,W,G} Among them, D represents the historical data set; P represents the power output data; W represents the weather data; and G represents the gas supply data.

[0023] (2) The data cleaning and normalization processing is expressed as: ; in, D clean Indicates data cleaning operation; f clean represents the data cleaning function; D represents the historical data set, D norm Represents data normalization operation; f norm Represents the data normalization function.

[0024] (3) The data normalization process further includes: ; in, represents the normalized data;x Represents the original data; min( x ) represents the minimum value of the data; max( x ) indicates the maximum value of the data.

[0025] 2) The different data sources of the gas power station are constructed as a spatiotemporal graph, and an adjacency matrix is ​​constructed to represent the spatial relationship between nodes.

[0026] (1) The construction of different data sources of the gas power station into a time-space diagram can be expressed as: ; Among them, H represents the spatiotemporal graph of the gas power station; V represents the node set, and one gas power station represents one node; 𝐸 represents the edge set, and the edge represents the spatiotemporal relationship between time steps; A represents the spatiotemporal correlation between gas power stations.

[0027] (2) Constructing an adjacency matrix to represent the spatial relationship between nodes can be expressed as: Using the Gaussian kernel function to construct the elements in the adjacency matrix : ; in, represents the jth element in the i-th row of the adjacency matrix A, where i and j represent nodes i and j respectively; p i and p j Represent the geographical locations of node i and node j respectively, d ( i , j ) is the correlation coefficient between the historical power of node i and node j, a and k is a parameter that controls the sparsity of the adjacency matrix.

[0028] 3) Design an adaptive and scalable spatiotemporal graph convolutional network model, including input layer, spatiotemporal convolution module, adaptive module, scalable module and fully connected layer; In the input layer, the preprocessed data is input into the model, including historical power data, weather data and gas supply data; time convolution and graph convolution are performed in the spatiotemporal convolution module to capture the correlation in time and space; an adaptive module is introduced to dynamically adjust the model parameters to adapt to the changing characteristics of gas power generation; an extensible module is designed so that the model can be expanded as the amount of data and the number of nodes increase, maintaining efficient computing.

[0029] The input layer described in (1) can be expressed as: The preprocessed data are input into the model, including historical power data, weather data and gas supply data.

[0030] (2) The spatiotemporal convolution module includes L spatiotemporal convolution blocks connected in sequence: The spatiotemporal convolution block uses graph convolution operations to extract the spatial features of gas power generation.

[0031] To ensure the stability of the convolution operation, the adjacency matrix A needs to be normalized. The normalized adjacency matrix is : ; Where J represents the degree matrix, .

[0032] Then calculate the graph Laplace operator, the normalized graph Laplace operator is:

[0033] in, represents the normalized graph Laplacian operator; I Represents the identity matrix.

[0034] The Chebyshev polynomial approximation is used to implement localized convolution and extract the features of each node and its neighborhood in the gas power station: ; in, Indicates The output features of the layer graph convolution, T k Indicates k The Chebyshev polynomial of order, K represents the total order of Chebyshev polynomial; Indicates The weight matrix of the k-th order Chebyshev polynomial in the layer graph convolution, Indicates Output features of the layer graph convolution.

[0035] Then, the temporal features are extracted through a one-dimensional convolution operation, and a gated linear unit is used to capture the temporal dynamic behavior of the gas power station.

[0036] The one-dimensional convolution operation can be expressed as: ; in, represents the initial input data of the gated linear unit; s represents the activation function; * represents the convolution operation; represents element-wise multiplication; Indicates The feature matrix of the gated linear unit output in the layer temporal convolution, Indicates The convolution kernel of the gated linear unit in the layer temporal convolution is specifically used to generate the gating signal. Indicates Bias of gated linear units in layer temporal convolution; For the The output matrix of the layer's temporal convolution.

[0037] Combining the above graph convolution operation with the temporal convolution operation, a spatiotemporal convolution block is formed to capture the spatiotemporal characteristics of the gas power station, and the feature extraction capability is improved by stacking multiple spatiotemporal convolution blocks. Each spatiotemporal convolution block includes a graph convolution and a temporal convolution. Graph convolution is used to extract spatial features, and temporal convolution is used to extract temporal features. In order to enhance the expressive power of the model, multiple spatiotemporal convolution blocks are stacked. The output of each convolution block is used as the input of the next convolution block. The activation function (ReLU) and regularization method (Dropout) are used to prevent overfitting.

[0038] The spatiotemporal convolution block captures the spatiotemporal characteristics of the gas power station and is expressed as: ; in, Indicates The feature matrix output by the spatiotemporal convolutional block; Indicates The weight matrix of the layer graph convolution, Weight matrices of K Chebyshev polynomials in layer graph convolution Composition Weight matrix of layer graph convolution ; Z (l) Indicates The feature matrix of the layer time convolution output, Indicates Output features of layer graph convolution; Indicates The bias of the one-dimensional convolution in the layer temporal convolution, GraphConv(.) represents the graph convolution operation, TemporalConv(.) represents the temporal convolution operation, ; .

[0039] 3) The adaptive module can be expressed as: dynamically adjusting the model parameters through the adaptive graph attention mechanism to adapt to the changing characteristics of gas power generation. The adaptive module includes node feature transformation, adaptive attention weight calculation and attention weight normalization process.

[0040] Among them, the node feature transformation can be expressed as: ; Among them, F (1) 、F (m) 、F (m+1) Respectively represent the 1st, m ,m +1 node feature matrix output by the adaptive layer; B (1) , B (m+1) Indicates the first m +1 weight matrix of the adaptive layer.

[0041] Adaptive attention weight calculation: ; in, represents the adaptive attention weight between node i and node j in the mth adaptive layer; LeakyReLU represents the leaky linear rectification activation function; is the set of normalized adaptive attention weights in the mth adaptive layer, concatenating the features of node i and node j; and They represent the features of nodes i and j in the mth adaptive layer, Represents the weight matrix of the mth adaptive layer.

[0042] Attention weight normalization: ; in, represents the normalized value of the adaptive attention weight between node i and node j in the mth adaptive layer, V(i) represents the set of neighboring nodes of node i, represents the total number of nodes in the set V(i), represents the node i in the mth adaptive layer and the node in the set V(i) Adaptive attention weights between .

[0043] Through this adaptive attention mechanism, the model can dynamically adjust the importance of each node in the graph to adapt to the changing characteristics of gas power generation.

[0044] 4) The scalable module can be expressed as follows: the scalable module includes N scalable layers, and the scalable layers enable the model to expand as the amount of gas power generation data and the number of nodes increase through multi-head graph convolution and layered graph convolution, while maintaining efficient computing.

[0045] Through the multi-head attention mechanism, the model can capture different graph structure information, thereby enhancing the model's expressiveness and scalability. Among them, the multi-head graph convolution can be expressed as: ; in, , , Respectively represent nodes i In the 1st, n , n+1 layer of multi-head graph convolution feature matrix; Q represents the number of attention heads; the normalized adaptive attention Decompose into Q attention heads, and use the Q attention heads as the Q attention heads in the first layer of multi-head graph convolution, so that each attention head focuses on different aspects or features of the input data. Represents the node i and node q calculated by the qth attention head in the first layer of multi-head graph convolution j The attention weights between represents the node feature matrix output by the Mth adaptive layer, , Represents the weight matrix of the qth attention head in the n+1th layer of multi-head graph convolution.

[0046] Through hierarchical aggregation, the model can capture high-order information in the graph structure layer by layer, thereby achieving the scalability of the model.

[0047] Hierarchical aggregation can be expressed as: ; Finally, a fully connected layer is used to output the predicted power.

[0048] 4) Perform model training on the model in step 3; The accuracy of the prediction is evaluated by defining the mean square error (MSE) and the root mean square error (RMSE): ; Among them, n1 represents the number of samples, represents the predicted value of the i1th sample; Represents the actual value of the i1th sample.

[0049] Use historical data to train the model, input the training data, and perform forward propagation through the spatiotemporal convolution layer, adaptive module, scalable module, and output layer to calculate the predicted value. Calculate the loss between the predicted value and the actual value according to the loss function and calculate the gradient of the loss relative to the model parameters. Use the gradient descent optimization algorithm to update the model parameters: ; in, q represents the model parameters, h represents the learning rate, represents the gradient of the loss function.

[0050] Repeat forward propagation, loss calculation, back propagation, and parameter update until the model converges or reaches the maximum number of iterations. Use the validation set data to evaluate the performance of the model to prevent overfitting.

[0051] 5) Input the new gas power generation data into the model in step 4 and output the predicted gas power plant power value.

[0052] The new weather data and gas supply data are input into the model, and the gas power generation power is predicted through the trained ASGCN model, and finally the predicted gas power station power value is output.

[0053] For the above test system, this embodiment uses four estimation methods, namely, regression prediction of the traditional least squares (LR) algorithm, regression prediction of the long short-term memory recursive neural network (LSTM) method, regression prediction of the spatiotemporal graph convolutional neural network (ST-GCN) method, and adaptive scalable spatiotemporal graph convolutional network (ASGCN) proposed by the present invention, for simulation comparative analysis. Table 1 shows the statistical regression prediction effect and algorithm calculation efficiency of the four methods after 20 simulations.

[0054] Table 1 method MSE RMSE Average inference time (s) Floating point operations per second LR 10.24 3.20 1.32 / LSTM 6.79 2.61 3.28 2.45 ST-GCN 5.32 2.31 2.51 1.89 ASGCN 3.26 1.81 1.08 0.92 It can be seen from the data in the table that the traditional least squares estimation method lacks the ability to predict the dynamic changes of meteorological and gas power plant topological systems, and the estimation results show a large RMSE, which indicates that the estimation results deviate a lot from the actual state. The prediction accuracy of the long short-term memory recursive neural network method has been greatly improved compared with the static estimation. At the same time, the ST-GCN algorithm can better consider the characteristics of each node and its neighborhood of the gas power plant, and the estimation result accuracy and algorithm calculation efficiency are better than LSTM. The method proposed in the present invention considers the number of gas power plants and the dynamic growth factors of nodes on the basis of ST-GCN, which greatly improves the accuracy of the estimation results, and because of the improved structure of the convolution kernel, it does not affect the algorithm calculation efficiency too much.

[0055] Figure 3The following is a comparison chart of the effects of three different prediction methods. As shown in Figure 3, in terms of prediction accuracy, the curve of ASGCN (red) is close to the actual value (black), and can track the trend of actual value changes well in most time periods; in contrast, the fluctuation of LSTM (blue) is large, especially in some time periods, it deviates greatly from the actual value; the predicted value of ST-GCN (purple) deviates more at some peaks and valleys. In terms of stability, the curve of ASGCN is relatively smooth, with small fluctuations, showing good stability, which is particularly important for gas power generation scenarios; the curve of LSTM fluctuates greatly, and the prediction results are sometimes too drastic, not as stable as ASGCN; ST-GCN also fluctuates greatly in certain periods, and the predicted value is not stable enough. In terms of tracking trends, ASGCN tracks the overall trend of actual values ​​well throughout the entire time period, especially in the middle and late stages; LSTM fails to track the actual trend well in some time periods, and some large deviations occur; although ST-GCN performs well in some time periods, its overall trend tracking ability is not as good as ASGCN. In terms of outlier processing, ASGCN is relatively stable in processing outliers (such as the peaks and valleys in the figure) without any significant deviations. When facing outliers, the curve of LSTM fluctuates greatly, showing its insufficient ability to handle outliers. ST-GCN also fluctuates greatly at some outliers and fails to handle outliers well.

[0056] In summary, the method proposed in the present invention shows higher accuracy, better stability, stronger trend tracking ability and better outlier processing ability in predicting gas power generation. At the same time, the introduction of adaptive and scalable modules enables ASGCN to adapt to the changing characteristics of gas power generation, and can be expanded with the increase of gas power generation data volume and the number of nodes. It can handle the nonlinear and dynamic changes in the power data of gas power plants, improves the robustness and adaptability of the model, and has obvious advantages over LSTM and ST-GCN.

[0057] The above description is only a preferred embodiment of the present invention and is not intended to be a further limitation of the present invention. All equivalent changes made using the contents of the present specification and drawings are within the protection scope of the present invention.

Claims

1. A method for predicting power of a gas power plant, characterized in that: The specific steps include: Step 1: Collect historical data of gas power plants as sample data; Step 2: Preprocess the sample data; Step 3: construct the preprocessed data into a spatiotemporal graph H; H = (V, E, A), V represents a node set, the node represents a gas power station, E represents an edge set, the edge is a connection edge between nodes, and A is an adjacency matrix; Step 4: Design an adaptive and scalable spatiotemporal graph convolutional network model; Step 5: Use the preprocessed data and adjacency matrix as training sets to train the adaptive scalable spatiotemporal graph convolutional network model; Step 6: Use the trained adaptive scalable spatiotemporal graph convolutional network model to predict the power of the gas power station.

2. A method for predicting power of a gas power plant according to claim 1, characterized in that: The sample data in step 1 includes historical power, weather data related to power generation, and gas supply data.

3. A method for predicting power of a gas power plant according to claim 1, characterized in that: The data preprocessing in step 2 specifically includes first cleaning the data and then normalizing it. The data cleaning specifically includes: processing the missing power data using the forward filling or backward filling method; then using the Z-score method to detect and process outliers; and converting the timestamp into a datetime type.

4. A method for predicting power of a gas power plant according to claim 1, characterized in that: In step 3, a Gaussian kernel function is used to construct an adjacency matrix, and the expression of the adjacency matrix A is: ； Among them, i, j represent nodes i, j in the node set respectively; p i and p j Represent the geographical locations of node i and node j respectively, d ( i , j ) is the correlation coefficient between the historical power of node i and node j, a and k is a parameter that controls the sparsity of the adjacency matrix.

5. A method for predicting power of a gas power plant according to claim 1, characterized in that: The adaptive scalable spatiotemporal graph convolutional network model is configured to perform the following operations: The spatiotemporal convolution block is used to capture the spatiotemporal characteristics of the gas power station, and the feature extraction capability is improved by stacking L layers of spatiotemporal convolution blocks. Then, the adaptive module is used to adjust the model parameters. The scalable module is used to expand according to the increase in the amount of gas power generation data and the number of nodes. Finally, the predicted power of the gas power station is output through the fully connected layer.

6. A method for predicting power of a gas power plant according to claim 5, characterized in that: Each spatiotemporal convolution block uses graph convolution operations to extract the features of each node and neighborhood of the gas power plant, uses the one-dimensional convolution operation in the time convolution to extract the time features, and uses the gated linear unit in the time convolution to capture the temporal dynamic behavior of the gas power plant. The structure of the spatiotemporal convolution block is as follows: ； in, Indicates The output features of the spatiotemporal convolutional block of the layer, TemporalConv(.) represents the temporal convolution operation, s(.) represents the activation function, and GraphConv(.) represents the graph convolution operation. Indicates The output features of the layer graph convolution, Indicates The output features of the layer temporal convolution, Indicates The bias of the 1D convolution in the layer temporal convolution, represents the normalized adjacency matrix, , where J represents the degree matrix, Indicates The weight matrix of the layer graph convolution; , , X is the sample data; ； in, Indicates The output features of the layer graph convolution, T k Indicates k The Chebyshev polynomial of order, K represents the total order of Chebyshev polynomial; represents the normalized graph Laplacian operator, , I represents the identity matrix, Indicates The weight matrix of the k-th order Chebyshev polynomial in the layer graph convolution, The weight matrix of K Chebyshev polynomials in the layer graph convolution constitutes the Weight matrix of layer graph convolution ; The structure of temporal convolution is as follows: ； in, Indicates The output features of the layer temporal convolution, Indicates The output features of the one-dimensional convolution in the layer time convolution, Indicates The convolution kernel of the one-dimensional convolution in the layer time convolution, Indicates Bias of 1D convolution in layer-wise temporal convolution; Indicates The convolution kernel of the gated linear unit in the layer temporal convolution, Indicates Bias of the gated linear unit in the layer temporal convolution, * indicates the convolution operation.

7. A method for predicting power of a gas power plant according to claim 5, characterized in that: The adaptive module includes M adaptive layers connected in sequence, wherein the adaptive layers include node feature transformation, adaptive attention weight calculation and attention weight normalization; The node feature transformation is specifically as follows: ； Among them, F (m+1) Represents the node feature matrix output by the m+1th adaptive layer, and the node feature matrix output by the 1st adaptive layer ,in represents the features output by the spatiotemporal convolution module, represents the weight matrix of the first adaptive layer, and s(.) represents the activation function; represents the weight matrix of the m+1th adaptive layer, m=1,2,3,…,M; The adaptive attention weight meter is specifically: ； in, represents the adaptive attention weight between node i and node j in the mth adaptive layer; LeakyReLU represents the linear rectification activation function with leakage, || represents the vector connection, and They represent the features of nodes i and j in the mth adaptive layer respectively; represents the weight matrix of the mth adaptive layer, is the set of normalized adaptive attention weights in the mth adaptive layer; The adaptive attention weight normalization is specifically: ； in, represents the normalized value of the adaptive attention weight between node i and node j in the mth adaptive layer, V(i) represents the set of neighboring nodes of node i, represents the total number of nodes in the set V(i), represents the node i in the mth adaptive layer and the node in the set V(i) Adaptive attention weights between .

8. A method for predicting power of a gas power plant according to claim 7, characterized in that: The scalable module includes N scalable layers connected in sequence, and the scalable layers include multi-head graph convolution and layered aggregation. The multi-head graph convolution is as follows: ； in, represents the feature matrix of node i output by the n+1th layer of multi-head graph convolution, Q represents the number of attention heads, represents the weight matrix of the qth attention head in the n+1th layer of multi-head graph convolution, Represents the feature matrix of node i output by the n-th layer multi-head graph convolution, and the normalized adaptive attention in the M-th adaptive layer Decompose into Q attention heads, and use the Q attention heads as the Q attention heads in the first layer of multi-head graph convolution. The feature matrix of node i output by the first layer of multi-head graph convolution is The expression is , Represents the node i and node q calculated by the qth attention head in the first layer of multi-head graph convolution j The attention weights between represents the node feature matrix output by the Mth adaptive layer, .

9. An electronic device / system for a method for predicting power of a gas power plant, characterized in that: It comprises a processor and a memory, wherein the memory stores execution instructions of the processor, and the processor is configured to execute the execution instructions to implement the power prediction method for a gas power station as claimed in any one of claims 1 to 8.

10. A computer-readable storage medium for storing a program, characterized in that: Execute the program to implement the gas power plant power prediction method described in any one of claims 1-8.

Images