New energy power generation cluster power prediction method, device, equipment, medium and product

Abstract

The application discloses a new energy power generation cluster power prediction method, device, equipment, medium and product, relates to the power prediction technical field, and the method comprises the following steps: inputting weather forecast data at the positions of each power station in the new energy power generation cluster into a new energy power generation cluster power prediction model to obtain a power prediction result; the new energy power generation cluster power prediction model is obtained by training a target neural network through a training set, the target neural network comprises an encoder and a decoder, the encoder comprises a first layer heterogeneous graph neural network and a second layer heterogeneous graph neural network connected in sequence, each node in the first layer heterogeneous graph neural network is a power station node, each node in the second layer heterogeneous graph neural network is a region node aggregated by a plurality of power station nodes, and the decoder comprises a plurality of parallel gated recurrent units, and each gated recurrent unit outputs a power prediction value of a power station. The application can improve the power prediction accuracy of the power generation cluster of multiple types of power stations.

Description

Technical Field

[0001] This application relates to the field of power prediction technology, and in particular to a method, device, equipment, medium and product for predicting the power of a new energy power generation cluster. Background Technology

[0002] With a high proportion of wind and solar power connected to the grid, the impact of random fluctuations in wind and solar power on the safe and stable operation of the power grid is becoming increasingly significant. Power forecasting serves as an important source of reference information for power grid dispatching.

[0003] Current forecasting techniques primarily model and predict the power output of individual wind farms or photovoltaic power plants. For regional power generation clusters with a large number of power plants, repetitive modeling is inefficient, as seen with techniques like Long Short-Term Memory (LSTM) neural networks, Transformers, and Informers. Only a few techniques can simultaneously predict the power output of multiple power plants using multi-task neural networks, such as convolutional neural networks (CNNs), fully connected neural networks, and so on. However, these methods do not adequately model the correlations between different types of power generation; they only predict the power output of wind farm clusters or photovoltaic power plant clusters and cannot achieve accurate power prediction for multiple types of power plants. Summary of the Invention

[0004] The purpose of this application is to provide a method, device, equipment, medium and product for predicting the power of new energy power generation clusters, which can improve the accuracy of power prediction for power generation clusters of various types of power plants.

[0005] To achieve the above objectives, this application provides the following solution:

[0006] Firstly, this application provides a method for predicting the power output of a new energy power generation cluster, the method comprising:

[0007] Collect weather forecast data for the locations of each power station in the new energy power generation cluster; the new energy power generation cluster includes multiple power stations, each of which is a wind power station or a photovoltaic power station.

[0008] Weather forecast data for each power station location is input into the power prediction model of the new energy power generation cluster to obtain power prediction results. The power prediction results include the power prediction values ​​of each power station. The power prediction model of the new energy power generation cluster is obtained by training a target neural network with a training set. The target neural network includes an encoder and a decoder connected in sequence. The encoder includes a first layer heterogeneous graph neural network and a second layer heterogeneous graph neural network connected in sequence. Each node in the first layer heterogeneous graph neural network is a power station node, and each node in the second layer heterogeneous graph neural network is a region node formed by the aggregation of multiple power station nodes. The decoder includes multiple parallel gated recurrent units, and each gated recurrent unit outputs the power prediction value of one power station.

[0009] Optionally, each node in the first layer of the heterogeneous graph neural network and the second layer of the heterogeneous graph neural network is matched with a learning vector, which is represented as:

[0010] in, This represents the learning vector of node v after k aggregation operations, where k ≥ 1. As the initial learning vector, Let v represent the learned vector of node v after k-1 aggregation operations. intermediate parameters Aggregate k () represents the k-th aggregation operation, Combine k () represents the k-th merge operation, N(v) represents the set of neighboring nodes of node v, and u is a node in N(v).

[0011] Optionally, the target neural network is trained using a training set, specifically including:

[0012] Historical data of new energy power generation clusters are collected to construct an original dataset; each sample in the original dataset includes the power value of each power station, the location of the power station, weather forecast data, and timestamp;

[0013] Outlier processing was performed on the power values ​​and weather forecast data in the original dataset, and the missing values ​​after outlier processing were filled in using the moving average method to obtain the training set;

[0014] Based on the training set, the target neural network is trained with weather forecast data and timestamps as inputs and timestamps and power values ​​as outputs. The trained target neural network serves as the power prediction model for the new energy power generation cluster.

[0015] Optionally, outlier processing is performed on the power values ​​and weather forecast data in the original dataset, and missing values ​​after outlier processing are filled in using a moving average method to obtain a training set, specifically including:

[0016] Calculate the upper and lower quartiles of all power values ​​in the original dataset;

[0017] Power values ​​that exceed the upper quartile and lower quartile of all power values ​​are considered outliers and removed.

[0018] Calculate the upper quartile and lower quartile of each weather forecast data in the original dataset;

[0019] Power values ​​that fall outside the upper and lower quartiles of each weather forecast data are considered outliers and removed.

[0020] The missing values ​​after outlier removal are filled in using a moving average method to obtain the training set.

[0021] Optionally, when training the target neural network based on the training set, with weather forecast data and timestamps as inputs and timestamps and power values ​​as outputs, the mean squared error loss is used as the forward propagation error, and the model parameters of the target neural network are adjusted by the error gradient of the forward propagation error.

[0022] Optionally, the weather forecast data includes temperature, humidity, air temperature, and air pressure.

[0023] Secondly, this application provides a new energy power generation cluster power prediction device, which applies the aforementioned new energy power generation cluster power prediction method, and the new energy power generation cluster power prediction device includes:

[0024] The data acquisition module is used to collect weather forecast data at the locations of each power station in the new energy power generation cluster; the new energy power generation cluster includes multiple power stations, each of which is a wind power station or a photovoltaic power station;

[0025] The power prediction module is used to input weather forecast data at the location of each power station into the power prediction model of the new energy power generation cluster to obtain power prediction results. The power prediction results include the power prediction values ​​of each power station. The power prediction model of the new energy power generation cluster is obtained by training a target neural network with a training set. The target neural network includes an encoder and a decoder connected in sequence. The encoder includes a first layer heterogeneous graph neural network and a second layer heterogeneous graph neural network connected in sequence. Each node in the first layer heterogeneous graph neural network is a power station node, and each node in the second layer heterogeneous graph neural network is a regional node formed by the aggregation of multiple power station nodes. The decoder includes multiple parallel gated recurrent units, and each gated recurrent unit outputs the power prediction value of one power station.

[0026] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the new energy power generation cluster power prediction method described in any one of the above.

[0027] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the new energy power generation cluster power prediction method described above.

[0028] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the new energy power generation cluster power prediction method described above.

[0029] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0030] This application provides a method, device, equipment, medium, and product for predicting the power of a new energy power generation cluster. The encoder includes a first layer of heterogeneous graph neural network and a second layer of heterogeneous graph neural network connected in sequence. Each node in the first layer of heterogeneous graph neural network is a power plant node, and each node in the second layer of heterogeneous graph neural network is a regional node formed by the aggregation of multiple power plant nodes. The heterogeneous graph neural network can reflect the correlation between different types of power plants, thereby improving the accuracy of power prediction. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart illustrating a new energy power generation cluster power prediction method provided in an embodiment of this application.

[0033] Figure 2 This is a schematic diagram illustrating the principle of a new energy power generation cluster power prediction method provided in an embodiment of this application.

[0034] Figure 3 This is a schematic diagram of the structure of a heterogeneous graph neural network provided in an embodiment of this application.

[0035] Figure 4 This is a schematic diagram of the structure of a gated loop unit provided in an embodiment of this application.

[0036] Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0037] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0038] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0039] This application provides a method for predicting the power output of new energy power generation clusters, such as... Figure 1 and Figure 2 As shown, the new energy power generation cluster power prediction method includes steps 101 to 102.

[0040] Step 101: Collect weather forecast data for the locations of each power station in the new energy power generation cluster; the new energy power generation cluster includes multiple power stations, each of which is a wind power station or a photovoltaic power station.

[0041] Step 102: Input the weather forecast data of each power station location into the new energy power generation cluster power prediction model to obtain the power prediction result; the power prediction result includes the power prediction value of each power station; the new energy power generation cluster power prediction model is obtained by training a target neural network with a training set, the target neural network includes an encoder and a decoder connected in sequence, the encoder includes a first layer heterogeneous graph neural network and a second layer heterogeneous graph neural network connected in sequence, each node in the first layer heterogeneous graph neural network is a power station node, and each node in the second layer heterogeneous graph neural network is a regional node formed by the aggregation of multiple power station nodes, the decoder includes multiple parallel gated recurrent units, and each gated recurrent unit outputs the power prediction value of a power station.

[0042] In an exemplary embodiment, each node in the first layer heterogeneous graph neural network and the second layer heterogeneous graph neural network is matched with a learning vector, which is represented as:

[0043] in, This represents the learning vector of node v after k aggregation operations, where k ≥ 1. As the initial learning vector, Let v represent the learned vector of node v after k-1 aggregation operations. intermediate parameters Aggregate k () indicates the k-th aggregation operation, which aggregates information from neighboring nodes. k () indicates the k-th merge operation, which is to merge the aggregated vector representation with the original representation of the node. N(v) represents the set of neighboring nodes of node v, and u is any node in N(v).

[0044] Within adjacent regions, different power plants share natural similarities in their geographical location and meteorological environment, and their power variation processes also exhibit certain temporal or spatial correlations. Accurately grasping the relationships between multiple power plants will help improve the accuracy of power prediction for each plant. Furthermore, focusing on the specific power characteristics of each power plant influenced by its own meteorological or geographical environment will further enhance prediction accuracy.

[0045] In this application, the encoder is a new energy power generation cluster input feature encoder based on a two-layer heterogeneous graph neural network.

[0046] The first and second layers of heterogeneous graph neural networks constitute a two-layer heterogeneous graph neural network (HetGNN) model. The construction of this model is based on a combination of graph theory and neural network techniques. The first and second layers have the same structure but different inputs. The first layer's input is data from the nodes of the new energy power plant, and its output is the information from the first aggregation of each node with its neighboring nodes; therefore, it is called the feature of the region. The second layer's input is the output of the first layer, which undergoes a second aggregation; therefore, it is called the feature between regions.

[0047] The first layer of the heterogeneous graph neural network is used to mine the spatiotemporal correlations of forecast features among each renewable energy power plant. These nodes not only represent the physical location of the renewable energy power plants but also contain detailed information on their meteorological and power variation characteristics. In the second layer of the heterogeneous graph neural network, the model aggregates multiple renewable energy power plant nodes into regional nodes, revealing the transfer patterns of forecast features between regions by considering the geographical distance factors between regions. This two-layer structure design enables the model to capture both local and global load characteristics simultaneously, thereby improving the accuracy of predictions.

[0048] Heterogeneous Graph Neural Networks (GNNs) evolved from conventional Graph Neural Networks (GNNs), treating wind farms and photovoltaic power plants as different types of power nodes to facilitate the model learning targeted feature representations for different types of nodes. The basic principle of GNNs is as follows: GNNs, developed from convolutional neural networks and graph embedding concepts, are a novel extended neural network for extracting features from graph data structures. They exhibit superior performance in modeling graph data structure representations. GNNs update the current node by integrating information from neighboring nodes based on neighborhood information propagation and aggregation strategies, and capture arbitrary nearest neighbor information by stacking layers, thus effectively fusing the topological structure information of the graph. The graph representation in a GNN is defined as G = (ν, ε), where each node ν ∈ V has its vector representation X. νA graph is considered heterogeneous if the number of node types or edge types is greater than or equal to two. Graph neural networks are based on the graph topology and the initial vector X of the nodes. ν This means that the learning vector for each node is represented as h. v Graph neural networks (GNNs) are based on a neighborhood aggregation strategy. They obtain a new representation by aggregating information from neighboring nodes and then fusing this new representation with the original representation of the target node. By stacking k layers of a GNN, any node in the graph incorporates information from its k nearest neighbors and topological structure. The structure of a GNN is as follows: Figure 3 As shown, Figure 3 Type 1, Type 2, and Type k represent different node types.

[0049] The two-layer heterogeneous graph neural network model is a deep learning model that combines the characteristics of graph neural networks and heterogeneous data. This model is particularly effective when dealing with graph data with complex structures and different types of nodes, and can capture the heterogeneity between nodes and the complex relationships between them.

[0050] To adapt to the diversity of heterogeneous graphs, two-layer heterogeneous graph neural network models typically learn node representations in two steps: First, by weighted aggregation of different types of edges and nodes, local relationships between heterogeneous features are captured. Then, different graph convolution or attention mechanisms are used to further fuse the neighborhood information of nodes, thereby enhancing the expressive power of the features. This approach enables two-layer heterogeneous graph neural network models to effectively model heterogeneous relationships between different nodes and edges when processing complex graph data such as social networks, e-commerce recommendation systems, and knowledge graphs, thereby improving the model's inference and prediction performance.

[0051] The decoder in this application is a power plant decoder based on a parallel gated recurrent unit (GRU) neural network.

[0052] To accurately represent the differentiated charging modes of each new energy power station node, a multi-gated recurrent unit (GRU) parallel decoding neural network model was established. GRU is a commonly used recurrent neural network structure that can effectively handle time dependencies in sequential data. In this model, each new energy power station node corresponds to an independent GRU as a decoder, and the GRUs operate in parallel without interference. Each decoder generates a power prediction result for the corresponding new energy power station based on its own input (including weather forecast data and current status of the new energy power station) and the model's internal state. This approach ensures that the prediction for each charging station node is based on its unique characteristics and patterns, thereby improving the personalization of predictions for each new energy power station.

[0053] The multi-gated recurrent unit (GRU) parallel decoding neural network model is a widely used model architecture in sequence-to-sequence learning tasks. This model combines the advantages of gated recurrent units (GRUs) and parallel computing, enabling simultaneous consideration of time dependencies and computational efficiency when processing sequential data. GRUs are a variant of recurrent neural networks (RNNs) designed to address the vanishing and exploding gradient problems encountered by traditional RNNs when processing long sequences. GRUs effectively capture long-term dependencies in sequences by introducing gating mechanisms (including reset and update gates) to control the flow of information. During decoding, each GRU generates its current output based on the context vector and previous outputs. Because the GRUs in the decoder operate in parallel, multiple outputs can be generated simultaneously, improving computational efficiency. The structure of a GRU is as follows: Figure 4 As shown, Figure 4 Chinese x t The input to the gated recurrent unit at time t is σ, and the fully connected layer with the activation function is r. t Let z be the reset gate at time t. t h is the update gate at time t. t-1 h represents the hidden state at time t-1. t Let be the hidden state at time t. Let be the candidate hidden state at time t, and tanh be the nonlinear activation function.

[0054] In the decoding process of the two-layer heterogeneous graph neural network model, each renewable energy power plant node is treated as an independent time-series data point, each possessing its own unique meteorological / power variation pattern. The multi-gated recurrent unit parallel decoding neural network model assigns an independent gated recurrent unit decoder to each renewable energy power plant node. These decoders can process their respective data streams in parallel without interference. Each gated recurrent unit decoder predicts the power output of the renewable energy power plant over a future period based on the features processed by the heterogeneous graph neural network. Through this parallel decoding approach, the model can simultaneously consider the power variation characteristics of multiple renewable energy power plant nodes and make differentiated predictions based on the unique characteristics of each node. This not only improves the accuracy of predictions but also makes the model more flexible and personalized.

[0055] In an exemplary embodiment, training the target neural network using a training set specifically includes:

[0056] 1) Collect historical data of new energy power generation clusters to construct an original dataset; each sample in the original dataset includes the power value of each power station, the location of the power station, weather forecast data and timestamp.

[0057] 2) Perform outlier processing on the power values ​​and weather forecast data in the original dataset, and use the moving average method to fill in the missing values ​​after outlier processing to obtain the training set.

[0058] 3) Based on the training set, the target neural network is trained with weather forecast data and timestamps as inputs and timestamps and power values ​​as outputs. The trained target neural network serves as the power prediction model for the new energy power generation cluster.

[0059] In an exemplary embodiment, outlier processing is performed on the power values ​​and weather forecast data in the original dataset, and missing values ​​after outlier processing are filled in using a moving average method to obtain a training set, specifically including:

[0060] 1) Calculate the upper and lower quartiles of all power values ​​in the original dataset.

[0061] 2) Power values ​​that exceed the upper quartile and lower quartile of all power values ​​are considered outliers and removed.

[0062] 3) Calculate the upper quartile and lower quartile of each weather forecast data in the original dataset.

[0063] 4) Power values ​​that exceed the range between the upper and lower quartiles of each weather forecast data are considered outliers and removed.

[0064] 5) The missing values ​​after outlier removal are filled in using the moving average method to obtain the training set.

[0065] The missing values ​​after outlier removal are filled using a moving average method, specifically including:

[0066] ① For missing values ​​at position t, calculate the average value within the [tl, t-1] window and use it as the restored value of the data at position t, where l is a constant.

[0067] ②Then fill in the missing values ​​after t using the method in step ①.

[0068] Each sample in the training set consists of two parts: input data and output data. The input data for one training sample is the weather forecast data for the location of the wind and solar power station and the corresponding time label. The meteorological elements in the weather forecast data include temperature, humidity, air temperature, and air pressure, with a time series length of 72 hours and a time resolution of 15 minutes. The output data is the power generation value of each new energy power station within the corresponding time period, with a time series length of 72 hours and a time resolution of 15 minutes.

[0069] In an exemplary embodiment, when training the target neural network based on the training set, with weather forecast data and timestamps as inputs and timestamps and power values ​​as outputs, the mean squared error loss is used as the forward propagation error, and the model parameters of the target neural network are adjusted by the error gradient of the forward propagation error.

[0070] In one exemplary embodiment, the weather forecast data includes temperature, humidity, air temperature, and air pressure.

[0071] This application's new energy power generation cluster power prediction model can perform real-time power prediction for new energy power plants. The input information remains the weather forecast data for the period to be predicted at the location of the wind and solar power plant. After inputting the weather forecast data into the new energy power generation cluster power prediction model, the power prediction result for the location of the new energy power plant can be obtained.

[0072] The technical effects of this application are as follows.

[0073] The power prediction method for new energy power generation clusters proposed in this application is a spatiotemporal correlation modeling method for new energy power plants based on heterogeneous graph neural networks. It can effectively consider the differences and connections between similar and dissimilar power plants in terms of the spatiotemporal correlation of prediction features, which helps to improve prediction accuracy.

[0074] The power prediction method for new energy power generation clusters in this application is specifically a multi-scale power prediction feature representation method for new energy power plants using a two-layer graph neural network. By establishing the correlation between forecast features between power plant nodes and between different regions within the cluster, the precision of the power prediction features is further improved.

[0075] The power prediction method for new energy power generation clusters in this application is specifically a parallel power prediction method for multiple new energy power plants based on multiple gated cyclic units. A decoding neural network is established for each new energy power plant, which effectively improves the personalized representation of the power generation characteristics of the power plant.

[0076] Based on the same inventive concept, this application also provides a new energy power generation cluster power prediction device for implementing the aforementioned new energy power generation cluster power prediction method. The solution provided by this device is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more embodiments of the new energy power generation cluster power prediction device provided below can be found in the limitations of the new energy power generation cluster power prediction method described above, and will not be repeated here.

[0077] In one exemplary embodiment, this application provides a new energy power generation cluster power prediction device that applies the aforementioned new energy power generation cluster power prediction method. The new energy power generation cluster power prediction device includes:

[0078] The data acquisition module is used to collect weather forecast data at the locations of each power station in the new energy power generation cluster; the new energy power generation cluster includes multiple power stations, each of which is a wind power station or a photovoltaic power station.

[0079] The power prediction module is used to input weather forecast data at the location of each power station into the power prediction model of the new energy power generation cluster to obtain power prediction results. The power prediction results include the power prediction values ​​of each power station. The power prediction model of the new energy power generation cluster is obtained by training a target neural network with a training set. The target neural network includes an encoder and a decoder connected in sequence. The encoder includes a first layer heterogeneous graph neural network and a second layer heterogeneous graph neural network connected in sequence. Each node in the first layer heterogeneous graph neural network is a power station node, and each node in the second layer heterogeneous graph neural network is a regional node formed by the aggregation of multiple power station nodes. The decoder includes multiple parallel gated recurrent units, and each gated recurrent unit outputs the power prediction value of one power station.

[0080] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 5 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores power prediction data for new energy power generation clusters. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a power prediction method for new energy power generation clusters.

[0081] Those skilled in the art will understand that Figure 5The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0082] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0083] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0084] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0085] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0086] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units, etc., and are not limited to these.

[0087] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0088] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting the power output of a new energy power generation cluster, characterized in that, The method for predicting the power output of new energy power generation clusters includes: Collect weather forecast data for the locations of each power station in the new energy power generation cluster; the new energy power generation cluster includes multiple power stations, each of which is a wind power station or a photovoltaic power station. Weather forecast data for each power station location is input into the power prediction model of the new energy power generation cluster to obtain power prediction results. The power prediction results include the power prediction values ​​of each power station. The power prediction model of the new energy power generation cluster is obtained by training a target neural network with a training set. The target neural network includes an encoder and a decoder connected in sequence. The encoder includes a first layer heterogeneous graph neural network and a second layer heterogeneous graph neural network connected in sequence. Each node in the first layer heterogeneous graph neural network is a power station node, and each node in the second layer heterogeneous graph neural network is a region node formed by the aggregation of multiple power station nodes. The decoder includes multiple parallel gated recurrent units, and each gated recurrent unit outputs the power prediction value of one power station.

2. The method for predicting the power output of a new energy power generation cluster according to claim 1, characterized in that, In both the first and second heterogeneous graph neural networks, each node is matched with a learning vector, which is represented as follows: in, This represents the learning vector of node v after k aggregation operations, where k ≥ 1. As the initial learning vector, Let v represent the learned vector of node v after k-1 aggregation operations. intermediate parameters Aggregate k () represents the k-th aggregation operation, Combine k () represents the k-th merge operation, N(v) represents the set of neighboring nodes of node v, and u is a node in N(v).

3. The method for predicting the power output of a new energy power generation cluster according to claim 1, characterized in that, Training the target neural network using a training set specifically includes: Historical data of new energy power generation clusters are collected to construct an original dataset; each sample in the original dataset includes the power value of each power station, the location of the power station, weather forecast data, and timestamp; Outlier processing was performed on the power values ​​and weather forecast data in the original dataset, and the missing values ​​after outlier processing were filled in using the moving average method to obtain the training set; Based on the training set, the target neural network is trained with weather forecast data and timestamps as inputs and timestamps and power values ​​as outputs. The trained target neural network serves as the power prediction model for the new energy power generation cluster.

4. The new energy power generation cluster power prediction method according to claim 3, characterized in that, Outlier handling is performed on the power values ​​and weather forecast data in the original dataset, and missing values ​​are imputed using a moving average method to obtain the training set, which specifically includes: Calculate the upper and lower quartiles of all power values ​​in the original dataset; Power values ​​that exceed the upper quartile and lower quartile of all power values ​​are considered outliers and removed. Calculate the upper quartile and lower quartile of each weather forecast data in the original dataset; Power values ​​that fall outside the upper and lower quartiles of each weather forecast data are considered outliers and removed. The missing values ​​after outlier removal are filled in using a moving average method to obtain the training set.

5. The new energy power generation cluster power prediction method according to claim 3, characterized in that, Based on the training set, when training the target neural network with weather forecast data and timestamps as inputs and timestamps and power values ​​as outputs, the mean squared error loss is used as the forward propagation error, and the model parameters of the target neural network are adjusted by the error gradient of the forward propagation error.

6. The method for predicting the power output of a new energy power generation cluster according to claim 1, characterized in that, The weather forecast data includes temperature, humidity, air temperature, and air pressure.

7. A power prediction device for a new energy power generation cluster, characterized in that, The new energy power generation cluster power prediction device applies the new energy power generation cluster power prediction method according to any one of claims 1-6, and the new energy power generation cluster power prediction device comprises: The data acquisition module is used to collect weather forecast data at the locations of each power station in the new energy power generation cluster; the new energy power generation cluster includes multiple power stations, each of which is a wind power station or a photovoltaic power station. The power prediction module is used to input weather forecast data at the location of each power station into the power prediction model of the new energy power generation cluster to obtain power prediction results. The power prediction results include the power prediction values ​​of each power station. The power prediction model of the new energy power generation cluster is obtained by training a target neural network with a training set. The target neural network includes an encoder and a decoder connected in sequence. The encoder includes a first layer heterogeneous graph neural network and a second layer heterogeneous graph neural network connected in sequence. Each node in the first layer heterogeneous graph neural network is a power station node, and each node in the second layer heterogeneous graph neural network is a regional node formed by the aggregation of multiple power station nodes. The decoder includes multiple parallel gated recurrent units, and each gated recurrent unit outputs the power prediction value of one power station.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the new energy power generation cluster power prediction method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the new energy power generation cluster power prediction method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the new energy power generation cluster power prediction method as described in any one of claims 1-6.

Images