New energy station generation power prediction method and system based on recurrent graph neural network

Abstract

The invention discloses a new energy station generation power prediction method and system based on a cyclic graph neural network. The method comprises the following steps: pre-constructing a graph neural network model; the graph neural network model comprises a graph neural network layer, a recurrent neural network layer and a first full connection layer; acquiring historical weather forecast data and actual power data of the target station and the source domain station, and training the graph neural network model based on the historical weather forecast data and the actual power data of the source domain station to obtain a trained graph neural network model; and freezing the parameter weight of each network layer in the trained graph neural network model and migrating the parameter weight to a target station to obtain a power prediction model, and then performing fine tuning on the power prediction model based on historical meteorological data and actual power data of the target station to obtain a final power prediction model to perform power prediction of the target station. The method has the advantages of high prediction precision and the like.

Description

Technical Field

[0001] The present invention mainly relates to the field of new energy technology, and specifically to a method and system for predicting the power generation capacity of new energy stations based on a recurrent graph neural network. Background Art

[0002] With the increasing application of renewable energy power generation systems (primarily photovoltaic and wind power), more and more renewable energy sites are being connected to the distribution network, which brings huge challenges to the planning, operation, and control of the power system. Since the output characteristics of renewable energy sites are closely related to meteorological conditions, the output power of renewable energy power generation systems is inherently random and volatile. In the case that the mismatch between power storage facilities and renewable energy grid-connected power is difficult to change in the short term, the connection of large-scale renewable energy power generation systems to the grid will have a great impact on the safe and stable operation of the power system. This is also a key technical issue that needs to be solved for the large-scale connection of renewable energy power generation to the grid. Countries around the world have successively carried out technical research on renewable energy power generation power prediction, which is of great significance to the stable operation of the power system. It helps the power system dispatching department to coordinate the power generation planning of conventional energy and renewable energy power generation and reasonably arrange the operation mode of the power grid.

[0003] Currently, most models used to predict the power generation of renewable energy stations are machine learning models and deep learning models. Existing models for predicting the power generation of renewable energy stations construct a mapping relationship between forecast weather, historical weather data, and historical power generation data and power generation data, and use this mapping relationship to predict the power generation of renewable energy stations in future time periods. Existing methods are primarily data-driven, with the model mining the data to map forecast weather, historical weather, and historical power generation data to actual power generation. In conventional scenarios, we would like to have at least one year of historical data to train the model. However, for newly operated stations, there may be cases where historical data is less than one year, and some stations even have only one or two months of historical data.

[0004] To ensure the predictive performance of the model, for stations with less than one year of historical data, or even only one or two months, the conventional processing method is to use the data of surrounding stations together with the data of the target station to train the model, that is, to increase the number of samples, or to train the model based on the historical data of surrounding stations (source stations), and then migrate the model to the target station for fine-tuning for subsequent power prediction. The target station here refers to the station that needs to configure the power prediction model, and the surrounding stations are source stations. Currently, training models for target stations based on data fusion or model migration often only uses one source station. Therefore, before training the model, data analysis and station screening of surrounding stations are required. Even if the data of a surrounding station is screened for data fusion or training the model based on the migration mechanism, the data of other surrounding stations is wasted. Summary of the Invention

[0005] In response to the technical problems existing in the prior art, the present invention provides a method and system for predicting the power generation capacity of a new energy station based on a recurrent graph neural network with high prediction accuracy.

[0006] In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0007] A method for predicting power generation of a new energy station based on a recurrent graph neural network comprises the following steps:

[0008] Pre-constructing a graph neural network model; wherein the graph neural network model includes a graph neural network layer, a recurrent neural network layer, and a first fully connected layer, wherein the graph neural network layer, the recurrent neural network layer, and the first fully connected layer are sequentially connected;

[0009] Obtain historical weather forecast data and actual power data for the target and source stations, and train a graph neural network model based on the historical weather forecast data and actual power data for the source stations to obtain a trained graph neural network model; the source stations are the surrounding stations of the target station;

[0010] The parameter weights of each network layer in the trained graph neural network model are frozen and transferred to the target station to obtain a power prediction model. The power prediction model is then fine-tuned based on the historical meteorological data and actual power data of the target station to obtain the final power prediction model for power prediction of the target station.

[0011] Preferably, the specific process of freezing the parameter weights of each network layer in the trained graph neural network model and migrating them to the target station is as follows:

[0012] Determine the distance between stations based on the latitude and longitude coordinates of the target station;

[0013] The target station is incorporated into the graph neural network layer of the graph neural network model as a new node, and the edge connection status between the target station and the existing source domain stations is determined based on the distance between each station;

[0014] A second fully connected layer is added to the graph neural network model to obtain the power prediction model.

[0015] Preferably, the specific process of fine-tuning the power prediction model based on the historical meteorological data and actual power data of the target station is as follows:

[0016] The historical meteorological data and actual power data of the target station are copied into n+1 copies and input into the graph neural network layer and recurrent neural network layer of the power prediction model to obtain the output of the recurrent neural network layer. The data dimensions of the historical weather forecast data and actual power data of the target station are [bs, n+1, T, d], where bs is the data batch size, n+1 is the number of stations, T is the sequence data length, and d is the number of features of the input data.

[0017] The output results of the recurrent neural network layer are merged to obtain data of dimension [bs*(n+1),T,d2], which is then changed to [bs*(n+1)*T,d2] and input into the first fully connected layer FC1 to obtain data of dimension [bs*(n+1)*T,1];

[0018] The output dimension of the data with the dimension [bs*(n+1)*T,1] is changed to [bs*T,n+1] and input into the second fully connected layer FC2 to obtain data with the dimension [bs*T,1];

[0019] The label data dimension is expressed as [bs*T], and the MSE loss is calculated based on the label data and the data with the dimension [bs*T,1] to fine-tune the power prediction model.

[0020] Preferably, the historical meteorological data and actual power data of the target station are copied n+1 times and input into the graph neural network layer and the recurrent neural network layer of the power prediction model. The specific process of obtaining the output result of the recurrent neural network layer is as follows:

[0021] For each time step t=1,2,...,T, perform the following operations:

[0022] Input the slice data of dimension [bs,n+1,d] into the graph neural network layer GCN2, and obtain the output of dimension [bs,n+1,d1], where d1 is determined by the output channel parameter of the graph neural network layer;

[0023] The output dimension of [bs,n+1,d1] is changed to [bs*(n+1),d1] and input into the recurrent neural network layer to obtain data of dimension [bs*(n+1),d2], where d2 is determined by the parameters of the recurrent neural network layer RNN;

[0024] If the current time step t<=T, repeat the above steps and increase the time step by 1; if t>T, proceed to the next step.

[0025] Preferably, the specific process of training the graph neural network model based on the historical weather forecast data and actual power data of the source domain station is as follows:

[0026] Select source domain stations and determine the distance between stations based on the latitude and longitude coordinates of the stations; determine the edge connection status between nodes in the graph neural network layer based on the selected source domain stations; determine the number of layers and parameters of the graph neural network layer, and determine the number of layers and parameters of the recurrent neural network layer;

[0027] The historical meteorological data and actual power data of the source domain stations are input into the graph neural network layer and the recurrent neural network layer to obtain the output of the recurrent neural network layer; the data dimensions of the historical meteorological data and actual power data of the source domain stations are [bs,n,T,d], where bs is the data batch size, n is the number of source domain stations, T is the sequence data length, and d is the number of features of the input data;

[0028] The output results of the recurrent neural network layer are merged to obtain data of dimension [bs*n,T,d2], which is then changed to [bs*n*T,d2] and input into the first fully connected layer FC1 to obtain data of dimension [bs*n*T,1];

[0029] The label data dimension is expressed as [bs*n*T], the MSE loss is calculated based on the label data and the data of dimension [bs*n*T,1], and the graph neural network model is optimized.

[0030] Preferably, the historical meteorological data and actual power data of the source domain station are input into the graph neural network layer and the recurrent neural network layer, and the specific process of obtaining the output result of the recurrent neural network layer is as follows:

[0031] For each time step t=1,2,...,T, perform the following operations:

[0032] Input the slice data of dimension [bs,n,d] into the graph neural network layer GCN1, and obtain the output of dimension [bs,n,d1], where d1 is determined by the output channel parameter of the graph neural network layer;

[0033] The output dimension of [bs,n,d1] is changed to [bs*n,d1] and input into the recurrent neural network RNN ​​layer to obtain data of dimension [bs*n,d2], where d2 is determined by the parameters of the recurrent neural network layer RNN;

[0034] If the current time step t<=T, repeat the above operation and increase the time step by 1; if t>T, go to the next step.

[0035] Preferably, the weather forecast data includes forecast radiation, forecast temperature, forecast humidity, forecast air pressure, forecast wind speed and forecast wind direction.

[0036] The present invention also discloses a computer program product, comprising a computer program, which executes the steps of the above method when executed by a processor.

[0037] The present invention further discloses a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method described above are executed.

[0038] The present invention also discloses a new energy station power generation prediction system based on a cyclic graph neural network, comprising an interconnected memory and a processor, wherein a computer program is stored on the memory, and when the computer program is run by the processor, the steps of the method described above are executed.

[0039] Compared with the prior art, the advantages of the present invention are:

[0040] The new energy power prediction method of the present invention adopts a composite architecture that integrates a graph neural network layer (GNN layer) and a recurrent neural network layer (RNN layer). The GNN layer models the spatial correlation between stations (such as meteorological transmission and geographic proximity effects) through topological connections, capturing the spatial coordination characteristics of the regional meteorological field; the RNN layer dynamically models time series data and analyzes the time-varying laws of meteorological factors such as wind speed and irradiance; the dual network coupling realizes the joint analysis of spatiotemporal dependencies in the meteorological-power mapping relationship, thereby improving the prediction accuracy. During the migration phase of the present invention, the core parameters of the GNN-RNN are frozen (retaining the spatial topological modeling and time series feature extraction capabilities), and only the fully connected layer is fine-tuned to adapt to the characteristics of the target station (such as the inclination angle of the photovoltaic panel and the difference in wind turbine models). The training amount is small, which improves the training efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Schematic diagram of the structure of the graph neural network layer in the present invention.

[0042] Figure 2 This is a structural diagram of the graph neural network model in the present invention.

[0043] Figure 3 It is a structural diagram of the power prediction model in the present invention.

[0044] Figure 4 This is a flow chart of the training method of the graph neural network model in the present invention.

[0045] Figure 5 This is a flow chart of the training method of the power prediction model in the present invention.

[0046] Figure 6 for Figure 1 The corresponding edge connection matrix.

[0047] Figure 7This is a structural diagram of the present invention after incorporating the target station as a new node into the existing graph neural network layer.

[0048] Figure 8 for Figure 7 The corresponding edge connection matrix.

[0049] Figure 9 This is a flow chart of an embodiment of the method for predicting power generation of a new energy station based on a recurrent graph neural network of the present invention. DETAILED DESCRIPTION

[0050] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0051] like Figure 9 As shown, the method for predicting power generation of a new energy station based on a recurrent graph neural network according to an embodiment of the present invention includes the following steps:

[0052] Pre-constructing a graph neural network model; wherein the graph neural network model includes a graph neural network layer, a recurrent neural network layer, and a first fully connected layer, wherein the graph neural network layer, the recurrent neural network layer, and the first fully connected layer are sequentially connected;

[0053] Obtain historical weather forecast data and actual power data for the target and source stations, and train a graph neural network model based on the historical weather forecast data and actual power data for the source stations to obtain a trained graph neural network model; the source stations are the surrounding stations of the target station;

[0054] The parameter weights of each network layer in the trained graph neural network model are frozen and transferred to the target station to obtain a power prediction model. The power prediction model is then fine-tuned based on the historical meteorological data and actual power data of the target station to obtain the final power prediction model for power prediction of the target station.

[0055] Specifically, the historical weather forecast data and actual power data are sequence data, and the dimension is is the sequence data, n is the number of stations, T is the length of the sequence data, and d is the number of input features. The actual power (marked) corresponding to the input data is For each observation time t=1,2,...,T, take the slice data The observation time can be understood as the t-th moment in the sequence data, and then the slice data X t Expressed as a graph neural network layer, the corresponding actual power is Figure 1A graph neural network (GCN) layer consisting of 10 stations is shown, with each node representing a station. The features of each node in the GCN layer are weather forecast data and historical measured data. Whether a station is connected depends on the distance between the stations. For example, if the distance between two stations is less than 100 kilometers, the two stations are connected.

[0056] like Figure 2 As shown in the figure, the time slice data processed by the graph neural network layer is input into the recurrent neural network layer (RNN), and finally input into the first fully connected layer (FC1) to obtain the final output. Figure 2 The recurrent neural network layer in can be LSTM or GRU, and the specific parameters can be determined according to the actual situation.

[0057] Training based on data from selected source domain stations Figure 2 The model is displayed, and then the weights of the recurrent neural network layer and the existing graph neural network layer parameters in the trained model are frozen and migrated to the target station. The migration method is to combine the target station with the original graph neural network layer as a new node, and Figure 2 The second fully connected layer FC2 is added to the model framework, and the unfrozen model parameters are fine-tuned based on the data of the target station. The final adjusted model framework (power prediction model) is as follows Figure 3 As shown, the data input to the power prediction model is still sequence data, but the dimension is adjusted to And every They are all the same, namely the weather forecast data and historical actual data of the target station. Using this migration method and model framework, it is possible to achieve the goal of migrating models trained based on data from multiple source domain stations to the target station.

[0058] like Figure 4 As shown, the data training based on the selected source domain station Figure 2 The specific steps of the model (graph neural network model) are as follows:

[0059] 1. Select source stations and determine the distances between them based on their latitude and longitude coordinates;

[0060] 2. Build a graph neural network layer based on the selected source domain stations to determine the edge connection status between nodes;

[0061] 3. Determine the number of layers and parameters of the graph neural network layer, and determine the number of layers and parameters of the recurrent neural network layer;

[0062] 4. Prepare training data of dimensions [bs,n,T,d], where bs is the batch size, n is the number of source stations, T is the length of the sequence data, and d is the number of features in the input data. For each time step t = 1, 2, ..., T, perform the following operations:

[0063] (1) Input the slice data of dimension [bs,n,d] into the graph neural network layer GCN1, and obtain the output of dimension [bs,n,d1], where d1 is determined by the output channel parameter of the graph neural network layer;

[0064] (2) The output dimension of the above step is changed to [bs*n, d1], and input into the recurrent neural network layer RNN to obtain data of dimension [bs*n, d2]; where d2 is determined by the parameters of the recurrent neural network layer RNN;

[0065] (3) If the current time step t<=T, repeat the two operations (1) to (2) and increase the time step by 1; if t>T, jump to step 5;

[0066] 5. Combine the outputs from step 4 to obtain data of dimension [bs*n, T, d2], change the data dimension to [bs*n*T, d2] and input it into the first fully connected layer FC1 to obtain data of dimension [bs*n*T, 1];

[0067] 6. Represent the labeled data as [bs*n*T], calculate the MSE loss based on the labeled data and the output of step 5, and optimize the model.

[0068] like Figure 5 As shown, the source domain station data is used to train Figure 2 After the model framework is displayed, the parameters of the recurrent neural network layer and the existing graph neural network layer are frozen, and then the model is migrated to the target site and the unfrozen model parameters are fine-tuned. The specific steps are as follows:

[0069] 1. Determine the distance between stations based on the latitude and longitude coordinates of the target station;

[0070] 2. Incorporate the target station as a new node into the existing graph neural network layer, and determine the edge connection status between the target station and the existing nodes based on the station distance in step 1;

[0071] 3. In Figure 2 A second fully connected layer FC2 is added to the model framework, and the number of neurons in the second fully connected layer is 1;

[0072] 4. Prepare training data of dimensions [bs,n+1,T,d], where bs is the batch size, n+1 is the number of stations, T is the sequence length, and d is the number of features in the input data. Since only the target station data is used to fine-tune the model, the input data of the target station needs to be replicated n+1 times. For each time step t=1,2,...,T, perform the following operations:

[0073] (1) Input the slice data of dimension [bs,n+1,d] into the graph neural network layer GCN2, and obtain the output of dimension [bs,n+1,d1], where d1 is determined by the output channel parameter of the graph neural network layer;

[0074] (2) The output dimension of the above step is changed to [bs*(n+1), d1], and input into the recurrent neural network layer RNN to obtain data with a dimension of [bs*(n+1), d2], where d2 is determined by the parameters of the recurrent neural network layer RNN;

[0075] (3) If the current time step t<=T, repeat the two operations (1) to (2) and increase the time step by 1; if t>T, jump to step 5;

[0076] 5. Combine the outputs from step 4 to obtain data of dimension [bs*(n+1),T,d2], change the data dimension to [bs*(n+1)*T,d2] and input it into the first fully connected layer FC1 to obtain data of dimension [bs*(n+1)*T,1];

[0077] 6. The output obtained in step 5 is converted into data of [bs*T,n+1] and input into the second fully connected layer FC2 to obtain data of dimension [bs*T,1];

[0078] 7. Represent the labeled data as [bs*T], calculate the MSE loss based on the labeled data and the output of step 6, and optimize the model.

[0079] Based on target station data, Figure 3 The model framework (power prediction model) and Figure 5 The final power prediction model obtained by fine-tuning the process shown can be used to predict the power generation of the target station in the future time period. Figure 3 The graph neural network model of the model framework shown uses n+1 nodes. Therefore, each time inference is performed, the input data of the target station needs to be copied n+1 times to obtain input data of dimension [1, n+1, T, d]. This data is input into the model to obtain predicted power data of length T.

[0080] The new energy power prediction method of the present invention adopts a composite architecture that integrates a graph neural network layer (GNN layer) and a recurrent neural network layer (RNN layer). The GNN layer models the spatial correlation between stations (such as meteorological transmission and geographic proximity effects) through topological connections, capturing the spatial coordination characteristics of the regional meteorological field; the RNN layer dynamically models time series data and analyzes the time-varying laws of meteorological factors such as wind speed and irradiance; the dual network coupling realizes the joint analysis of spatiotemporal dependencies in the meteorological-power mapping relationship, thereby improving the prediction accuracy. During the migration phase of the present invention, the core parameters of the GNN-RNN are frozen (retaining the spatial topological modeling and time series feature extraction capabilities), and only the fully connected layer is fine-tuned to adapt to the characteristics of the target station (such as the inclination angle of the photovoltaic panel and the difference in wind turbine models). The training amount is small, which improves the training efficiency.

[0081] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0082] The patent of this invention is implemented based on the pytorch framework and the torch_geometric framework, where the recurrent neural network layer uses torch.nn.LSTM or torch.nn.GRU, and the graph neural network layer uses torch_geometric.nn.GCNConv.

[0083] Assume that 10 source domain stations have been selected for training Figure 2 The model framework is shown, and the connections between stations are as follows: Figure 1 As shown, n takes the value of 10. The data input to the model is weather forecast data, which specifically includes forecast irradiance, forecast temperature, forecast humidity, forecast air pressure, forecast wind speed and forecast wind direction, so d takes the value of 6. The length of the sequence data is 96, so T takes the value of 96, the time resolution of the data is 15 minutes, and the length of the output data is 96, that is, the power generation power at 96 moments in 24 hours a day is predicted based on one day's forecast data. The dimension of the input data obtained based on the source domain station data is [bs,10,96,6]. Since the graph neural network model is used, it is necessary to build edge connections based on the connection between nodes. Based on Figure 1 The node connections shown build the edge matrix as Figure 6 As shown, the values ​​in the upper and lower rows of the matrix represent the node numbers respectively, and there is a connection between the two nodes. Figure 1 Display Figure 1 There are 16 edges in total, so the edge matrix dimension created is [2,32]. Figure 6 The edge connection matrix and input data can be trained Figure 2 Model on display.

[0084] After the model is trained based on the data of the source station, it needs to be migrated to the target station and fine-tuned. Specifically, the target station is incorporated into the existing graph neural network layer as the 11th node, and the following is obtained: Figure 7 The new graph is shown. Since a new node has been added, the edge matrix needs to be modified. The updated edge matrix is ​​as follows Figure 8 As shown in the figure, a new node 10 is added, along with its connections to surrounding stations.

[0085] Freeze the parameters of the trained recurrent neural network layer and graph neural network layer, and add a second fully connected layer FC2. The dimension of the input data obtained based on the target station data is [bs, 11, 96, 6]. Since only the target station is used to fine-tune the model parameters, the input data of the 11 nodes are all the data of the target station, that is, the data of the dimension [bs, 96, 6] is copied 11 times to obtain the input data of the dimension [bs, 11, 96, 6]. Figure 5 The demonstrated process can fine-tune the model parameters based on the data of the target site, and use the adjusted model to predict the power generation power of the target site at a later time.

[0086] The present invention also discloses a computer program product, including a computer program, which executes the steps of the above method when executed by a processor. The present invention further discloses a computer-readable storage medium, on which a computer program is stored, which executes the steps of the above method when executed by a processor. The present invention also discloses a new energy station power generation prediction system based on a cyclic graph neural network, including an interconnected memory and a processor, the memory storing a computer program, which executes the steps of the above method when executed by the processor. The product, medium and system of the present invention correspond to the above method and also have the advantages of the above method.

[0087] The present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by hardware related to computer program instructions. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, it can implement the steps of the above-mentioned method embodiment. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. Computer-readable storage media include: any entity or device that can carry computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. The memory is used to store computer programs and / or modules, and the processor implements various functions by running or executing computer programs and / or modules stored in the memory, and calling data stored in the memory. The memory may include high-speed random access memory and non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0088] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for predicting power generation of new energy stations based on recurrent graph neural networks, characterized in that: Including steps: Pre-constructing a graph neural network model; wherein the graph neural network model includes a graph neural network layer, a recurrent neural network layer, and a first fully connected layer, wherein the graph neural network layer, the recurrent neural network layer, and the first fully connected layer are sequentially connected; Obtain historical weather forecast data and actual power data for the target and source stations, and train a graph neural network model based on the historical weather forecast data and actual power data for the source stations to obtain a trained graph neural network model; the source stations are the surrounding stations of the target station; The parameter weights of each network layer in the trained graph neural network model are frozen and transferred to the target station to obtain a power prediction model. The power prediction model is then fine-tuned based on the historical meteorological data and actual power data of the target station to obtain the final power prediction model for power prediction of the target station.

2. The method for predicting power generation of new energy stations based on recurrent graph neural networks according to claim 1 is characterized in that: The specific process of freezing the parameter weights of each network layer in the trained graph neural network model and migrating them to the target station is as follows: Determine the distance between stations based on the latitude and longitude coordinates of the target station; The target station is incorporated into the graph neural network layer of the graph neural network model as a new node, and the edge connection status between the target station and the existing source domain stations is determined based on the distance between each station; A second fully connected layer is added to the graph neural network model to obtain the power prediction model.

3. The method for predicting power generation of new energy stations based on recurrent graph neural networks according to claim 2 is characterized in that: The specific process of fine-tuning the power prediction model based on the historical meteorological data and actual power data of the target station is as follows: Copy n+1 copies of the target station’s historical meteorological data and actual power data, input them into the graph neural network layer and recurrent neural network layer of the power prediction model, and obtain the output of the recurrent neural network layer; The data dimensions of the historical weather forecast data and actual power data of the target station are [bs,n+1,T,d], where bs is the data batch size, n+1 is the number of stations, T is the sequence data length, and d is the number of features of the input data; The output results of the recurrent neural network layer are merged to obtain data of dimension [bs*(n+1),T,d2], which is then changed to [bs*(n+1)*T,d2] and input into the first fully connected layer FC1 to obtain data of dimension [bs*(n+1)*T,1]; The output dimension of the data with the dimension [bs*(n+1)*T,1] is changed to [bs*T,n+1] and input into the second fully connected layer FC2 to obtain data with the dimension [bs*T,1]; The label data dimension is expressed as [bs*T], and the MSE loss is calculated based on the label data and the data with the dimension [bs*T,1] to fine-tune the power prediction model.

4. The method for predicting power generation of new energy stations based on recurrent graph neural network according to claim 3 is characterized in that: The historical meteorological data and actual power data of the target station are copied n+1 times and input into the graph neural network layer and recurrent neural network layer of the power prediction model. The specific process of obtaining the output result of the recurrent neural network layer is as follows: For each time step t=1,2,...,T, perform the following operations: Input the slice data of dimension [bs,n+1,d] into the graph neural network layer GCN2, and obtain the output of dimension [bs,n+1,d1], where d1 is determined by the output channel parameter of the graph neural network layer; The output dimension of [bs,n+1,d1] is changed to [bs*(n+1),d1] and input into the recurrent neural network layer to obtain data of dimension [bs*(n+1),d2], where d2 is determined by the parameters of the recurrent neural network layer RNN; If the current time step t<=T, repeat the above steps and increase the time step by 1; if t>T, proceed to the next step.

5. The method for predicting power generation of a new energy station based on a recurrent graph neural network according to any one of claims 1 to 4, characterized in that: The specific process of training the graph neural network model based on the historical weather forecast data and actual power data of the source domain station is as follows: Select source domain stations and determine the distances between the stations based on their latitude and longitude coordinates; determine the edge connection states between nodes in the graph neural network layer based on the selected source domain stations; Determine the number of layers and parameters of the graph neural network layer, and determine the number of layers and parameters of the recurrent neural network layer; The historical meteorological data and actual power data of the source domain stations are input into the graph neural network layer and the recurrent neural network layer to obtain the output of the recurrent neural network layer; the data dimensions of the historical meteorological data and actual power data of the source domain stations are [bs,n,T,d], where bs is the data batch size, n is the number of source domain stations, T is the sequence data length, and d is the number of features of the input data; The output results of the recurrent neural network layer are merged to obtain data of dimension [bs*n,T,d2], which is then changed to [bs*n*T,d2] and input into the first fully connected layer FC1 to obtain data of dimension [bs*n*T,1]; The label data dimension is expressed as [bs*n*T], the MSE loss is calculated based on the label data and the data of dimension [bs*n*T,1], and the graph neural network model is optimized.

6. The method for predicting power generation of new energy stations based on recurrent graph neural networks according to claim 5 is characterized in that: The historical meteorological data and actual power data of the source domain station are input into the graph neural network layer and the recurrent neural network layer. The specific process of obtaining the output result of the recurrent neural network layer is as follows: For each time step t=1,2,...,T, perform the following operations: Input the slice data of dimension [bs,n,d] into the graph neural network layer GCN1, and obtain the output of dimension [bs,n,d1], where d1 is determined by the output channel parameter of the graph neural network layer; The output dimension of [bs,n,d1] is changed to [bs*n,d1] and input into the recurrent neural network RNN ​​layer to obtain data of dimension [bs*n,d2], where d2 is determined by the parameters of the recurrent neural network layer RNN; If the current time step t<=T, repeat the above operation and increase the time step by 1; if t>T, go to the next step.

7. The method for predicting power generation of a new energy station based on a recurrent graph neural network according to any one of claims 1 to 4, characterized in that: The weather forecast data includes forecast radiation, forecast temperature, forecast humidity, forecast air pressure, forecast wind speed and forecast wind direction.

8. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are performed.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 7.

10. A new energy station power generation prediction system based on a recurrent graph neural network, comprising a memory and a processor connected to each other, wherein a computer program is stored on the memory, characterized in that: When the computer program is executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 7.

Images