Fan power prediction method and system based on recurrent neural network and related equipment

Abstract

The invention discloses a fan power prediction method and system based on a recurrent neural network and related equipment. The method comprises the following steps: constructing a fan power prediction fusion neural network model by combining a time sequence neural network and a graph neural network and adopting an attention mechanism; constructing a dynamic graph matrix according to wind power plant fan geographical distribution and fan working parameters, converting wind power plant fan historical operation data into a set input format, and performing staged training on the fan power prediction fusion neural network model; and utilizing the fan power prediction fusion neural network model which is trained in stages to predict fan output power at a future moment according to the weather data and the emergency data at each moment, and completing self-learning of the fan power prediction fusion neural network model. The fusion neural network model can comprehensively consider wind power plant geography and fan information for power prediction, has a self-learning function, can process a large amount of complex information and emergencies, and has high flexibility and expandability.

Description

Technical Field

[0001] The present invention belongs to the field of wind turbine power prediction, and in particular relates to a wind turbine power prediction method, system and related equipment based on a recurrent neural network. Background Art

[0002] As an important form of clean energy generation, wind farms are of great significance for promoting energy structure adjustment and achieving sustainable development. Wind turbine power prediction is an important technical task in the wind power industry and is of great significance for improving the safety, stability and economy of the power system. With the continuous advancement of technology and the continuous expansion of its application scope, wind turbine power prediction technology will play a more important role in the development of the wind power industry.

[0003] Wind turbine power forecasting refers to the prediction of wind farm power generation in the future based on multiple factors such as wind farm meteorological information, historical data, topography, and wind turbine operating status, using physical simulation calculations and scientific statistical methods. This forecast result helps the power dispatching department to reasonably arrange power generation plans and dispatching strategies, and improve the overall operating efficiency of the power system.

[0004] The existing wind turbine power prediction algorithms combine cluster prediction, probability prediction, and event prediction algorithms to greatly improve the prediction accuracy. In the new era, the research on wind turbine power prediction is of great significance for ensuring the efficient operation of smart grids, peak and frequency regulation, and cost reduction and efficiency improvement. However, the traditional prediction model algorithm is simple, has high requirements for data correctness, and its accuracy cannot meet the grid's demand for flexibility.

[0005] Specifically, traditional prediction models, such as time random number series model, smooth exponential model, quadratic linear regression, nonlinear regression, etc., are often based on simple mathematical statistics and regression analysis. These models are relatively simple and rough when dealing with complex and changeable power system loads, and it is difficult to fully capture and reflect the nonlinear, random and time-varying characteristics of load changes. Traditional prediction models have very high requirements for data quality because they usually rely on the integrity and accuracy of historical data to build prediction models. However, in practical applications, power grid load data is often affected by a variety of factors, such as sensor failure, data transmission errors, human intervention, etc., resulting in noise, missing data and outliers. These problems will seriously affect the prediction accuracy and reliability of traditional prediction models. With the continuous development of the power market and the construction of smart grids, the power grid has higher and higher requirements for the accuracy and flexibility of load forecasting. Due to the simple algorithm and insufficient consideration of external factors, traditional prediction models are often difficult to accurately predict short-term fluctuations and long-term trends of loads, and cannot meet the power grid's demand for flexibility. For example, during holidays, special events or sudden weather changes, the change pattern of load may be significantly different from usual, and traditional prediction models may not be able to capture these changes in a timely and accurate manner. Summary of the invention

[0006] The purpose of the present invention is to address the problems in the above-mentioned prior art and provide a wind turbine power prediction method, system and related equipment based on a recurrent neural network, which can handle a large amount of complex information and emergencies, and automatically adjust the predicted power according to the events, with fewer restrictions, strong flexibility and scalability, and improve the accuracy of predicted power.

[0007] In order to achieve the above object, the present invention has the following technical solutions:

[0008] In a first aspect, a method for predicting wind turbine power based on a recurrent neural network is provided, comprising:

[0009] Combining time series neural network and graph neural network and using attention mechanism to construct a fusion neural network model for wind turbine power prediction;

[0010] A dynamic graph matrix is ​​constructed based on the geographical distribution of wind turbines in the wind farm and the operating parameters of the wind turbines. The historical operating data of wind turbines in the wind farm is converted into a set input format, and the wind turbine power prediction fusion neural network model is trained in stages.

[0011] By using the wind turbine power prediction fusion neural network model that has been trained in stages, the wind turbine output power at future moments is predicted according to the weather data and emergency data at each moment, and the self-learning of the wind turbine power prediction fusion neural network model is completed.

[0012] As a preferred solution, in the wind turbine power prediction fusion neural network model, the construction method of the time series neural network model is as follows:

[0013] Given a set of wind turbine geographic coordinates {P1, P2, P3, ..., P N}, where each coordinate contains latitude, longitude and altitude information; given a time step {T1, T2, …T M} and the corresponding input feature D, the input feature D represents the weather data with coordinates in a single time step, the maximum output power of the wind turbine, the power demand of the power grid and the characteristics of the emergency event vector; the training goal is to predict the wind turbine output power at future times given the input feature D and the historical wind turbine output power.

[0014] As a preferred solution, in the wind turbine power prediction fusion neural network model, the graph neural network model is constructed as follows:

[0015] A dynamic graph matrix is ​​used to establish nodes for each wind turbine, and all nodes are connected with an undirected graph;

[0016] A feature vector is assigned to each node as the input data of the graph neural network model to capture the connection and spatial dependency between every two nodes; the connection strength between two nodes is represented by the adjacency matrix; the similarity between nodes is captured by calculating the dot product of the node embedding and applying the activation function to improve the graph structure.

[0017] As a preferred solution, in the wind turbine power prediction fusion neural network model, the attention mechanism is constructed as follows:

[0018] A time series neural network model is used to generate an embedding for each time step to capture the impact of historical weather and historical output on future output. A self-attention mechanism is used to capture the impact of emergencies on output power. Elements are added to the input features of each node to input emergencies, including wind turbine failures and maintenance plans. A self-attention mechanism training weight matrix W is set to analyze the weight value of each element of the input feature.

[0019] As a preferred solution, in the wind turbine power prediction fusion neural network model, the loss function is the sum of the following losses: the mean square error between the output prediction value and the actual value, the distance between the dynamic graph adjacency matrix and the reconstructed adjacency matrix, and multiple additional loss functions for preventing overfitting.

[0020] As a preferred solution, the step of performing phased training on the wind turbine power prediction fusion neural network model includes:

[0021] An initial attribute graph is constructed based on the geographical distribution of wind turbines in the wind farm and the operating parameters of the wind turbines. The initial attribute graph and input features are passed to the graph neural network layer and the recurrent neural network layer of the wind turbine power prediction fusion neural network model to generate all node embeddings. All node embeddings are output to the fully connected layer, and the predicted value at the current moment is output. Through iterative calculation, the loss value calculated by the loss function is reduced, and all parameters of the wind turbine power prediction fusion neural network model are synchronously updated. When the loss value is reduced to the threshold or the trend converges, the iteration is ended and the training of the wind turbine power prediction fusion neural network model is completed.

[0022] As a preferred scheme, the step of using the wind turbine power prediction fusion neural network model that has been trained in stages to predict the wind turbine output power at future moments based on the weather data and emergency data at each moment includes: inputting the input features before a certain moment in the future into the wind turbine power prediction fusion neural network model that has been trained in stages, and predicting the wind turbine output power at multiple moments in the future.

[0023] In a second aspect, a wind turbine power prediction system based on a recurrent neural network is provided, comprising:

[0024] A wind turbine power prediction fusion neural network model building module, which is used to combine the time series neural network and the graph neural network and use the attention mechanism to construct a wind turbine power prediction fusion neural network model;

[0025] The model training module is used to construct a dynamic graph matrix based on the geographical distribution of wind turbines in the wind farm and the operating parameters of the wind turbines, and convert the historical operating data of the wind turbines in the wind farm into a set input format to perform phased training on the fusion neural network model for wind turbine power prediction;

[0026] The wind turbine output power prediction and model self-learning module is used to use the wind turbine power prediction fusion neural network model that has been trained in stages to predict the wind turbine output power at future moments based on the weather data and emergency data at each moment, and complete the self-learning of the wind turbine power prediction fusion neural network model.

[0027] According to a third aspect, an electronic device is provided, including:

[0028] A memory storing at least one instruction; and a processor executing the instruction stored in the memory to implement the wind turbine power prediction method based on a recurrent neural network.

[0029] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the wind turbine power prediction method based on a recurrent neural network is implemented.

[0030] Compared with the prior art, the present invention has at least the following beneficial effects:

[0031] The wind turbine power prediction fusion neural network model is constructed by combining the time series neural network and the graph neural network and adopting the attention mechanism. The wind turbine power prediction fusion neural network model can predict the data after multiple time steps based on historical data, and can understand the interaction and dependency between different wind turbine positions, effectively solving the drawbacks of the previous cluster prediction method, and can understand a variety of emergencies, and automatically adjust the predicted power according to the event. The wind turbine power prediction fusion neural network model of the present invention has two working modes: operation state and training state. It can use the staged training results for operation prediction, and continue to train the model with the input data as the data set, so as to realize the self-learning function. The wind turbine power prediction fusion neural network model of the present invention is deployed quickly, and can comprehensively consider the wind farm geography and wind turbine information for power prediction. It has a self-learning function, and the longer the use time, the closer the output result is to the accurate value. At the same time, it can handle a large amount of complex information and emergencies, and has strong flexibility and scalability, providing a way to further improve the accuracy of wind turbine power prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 Flow chart of the training mode of the fusion neural network model for predicting wind turbine power according to an embodiment of the present invention;

[0034] Figure 2 Flow chart of the prediction mode of the wind turbine power prediction fusion neural network model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, ordinary technicians in this field can also obtain other embodiments without making creative work.

[0036] There are many wind turbine power prediction methods available, mainly including physical methods, statistical methods and comprehensive methods. Among them, the physical method is based on the theory of atmospheric boundary layer dynamics and boundary layer meteorology, using numerical weather forecast (NWP) data, combined with the actual terrain and geomorphic conditions of the wind farm, to predict the wind speed and wind direction of the wind farm, and then obtain the predicted power through the power curve of the wind turbine. This method is suitable for newly built wind farms or in the absence of historical data. The statistical method is based on historical data, real-time data and numerical weather forecast data, using neural networks, support vector machines, fuzzy logic and other algorithms to establish a prediction model to predict the output power of the wind farm. This method has high prediction accuracy when the data is complete, but it requires the support of a large amount of historical data. The comprehensive method combines the advantages of physical methods and statistical methods, and uses multiple models and algorithms for combined prediction to improve prediction accuracy and reliability.

[0037] Wind turbine power prediction is of great significance to the operation of wind farms and power systems, including:

[0038] (1) Improve the safety of the power system. By accurately predicting the output power of wind farms, the power dispatching department can reasonably arrange power generation plans and dispatching strategies to avoid the impact of wind farm output power fluctuations on power grid stability.

[0039] (2) Improve the utilization rate of wind power. By predicting the power generation of wind farms in the future, the power dispatching department can optimize the power generation plan and improve the utilization rate and economic benefits of wind power.

[0040] (3) Promoting the development of the wind power industry. The development and application of wind turbine power prediction technology will help promote the rapid development of the wind power industry and technological progress, and increase the proportion of wind power in the energy structure.

[0041] See also Figure 1 and Figure 2 The embodiment of the present invention proposes a wind turbine power prediction method based on a recurrent neural network, which combines a time series neural network and a graph neural network and adopts an attention mechanism to construct a wind turbine power prediction fusion neural network model. The wind turbine power prediction fusion neural network model has two working modes: operation and training. The phased training results can be used for operation, while the model can be continuously trained with input data as a data set, thereby realizing a self-learning function.

[0042] Specifically, the wind turbine power prediction method based on a recurrent neural network in an embodiment of the present invention includes the following steps:

[0043] Step 1: Combine the time series neural network and graph neural network and use the attention mechanism to construct a fusion neural network model for wind turbine power prediction;

[0044] Step 2: construct a dynamic graph matrix according to the geographical distribution of wind turbines in the wind farm and the working parameters of the wind turbines, convert the historical operation data of the wind turbines in the wind farm into a set input format, and conduct phased training on the wind turbine power prediction fusion neural network model;

[0045] Step 3: Using the wind turbine power prediction fusion neural network model that has been trained in stages, predict the wind turbine output power at future times based on the weather data and emergency data at each moment, and complete the self-learning of the wind turbine power prediction fusion neural network model.

[0046] In a possible implementation manner, when constructing the wind turbine power prediction fusion neural network model, the construction method of the time series neural network model is as follows:

[0047] Given a set of wind turbine geographic coordinates {P1, P2, P3, ..., P N}, where each coordinate contains latitude, longitude and altitude information; given a time step {T1, T2, …T M} and the corresponding input feature D, which represents the weather data with coordinates at a single time step, the maximum output power of the wind turbine, the power demand of the power grid, and other features that are helpful for prediction, such as the emergency event vector. The training goal is to predict the wind turbine output power at a future moment given the input feature D and the historical wind turbine output power.

[0048] In a possible implementation manner, when constructing the wind turbine power prediction fusion neural network model, the graph neural network model is constructed as follows:

[0049] A dynamic graph matrix is ​​used to establish nodes for each wind turbine, and all nodes are connected with an undirected graph;

[0050] A feature vector is assigned to each node as the input data of the graph neural network model to capture the connection and spatial dependency between every two nodes; the connection strength between two nodes is represented by the adjacency matrix; the similarity between nodes is captured by calculating the dot product of the node embedding and applying the activation function to improve the graph structure.

[0051] In a possible implementation manner, when constructing the wind turbine power prediction fusion neural network model, the attention mechanism is constructed as follows:

[0052] A time series neural network model is used to generate an embedding for each time step to capture the impact of historical weather and historical output on future output. A self-attention mechanism is used to capture the impact of emergencies on output power. A sufficient number of elements are added to the input features of each node to input emergencies, including wind turbine failures and maintenance plans. A self-attention mechanism training weight matrix W is set to analyze the weight value of each element of the input feature.

[0053] In one possible implementation, when constructing the wind turbine power prediction fusion neural network model, the loss function is the sum of the following losses: the mean square error between the output prediction value and the actual value, the distance between the dynamic graph adjacency matrix and the reconstructed adjacency matrix, and multiple additional loss functions for preventing overfitting.

[0054] In a possible implementation, the step of performing phased training on the wind turbine power prediction fusion neural network model includes:

[0055] S1. Construct an initial attribute map based on the geographical distribution of wind turbines in the wind farm and the operating parameters of the wind turbines;

[0056] S2, passing the initial attribute graph and input features to the graph neural network layer and the recurrent neural network layer of the wind turbine power prediction fusion neural network model to generate all node embeddings;

[0057] S3, embed all nodes and output them to the fully connected layer, and output the predicted value at the current moment;

[0058] S4. Through continuous iterative calculation, the loss value calculated by the loss function is reduced, and all parameters of the wind turbine power prediction fusion neural network model are synchronously updated;

[0059] S5. When the loss value drops to the threshold or the trend converges, the iteration ends and the training of the wind turbine power prediction fusion neural network model is completed.

[0060] In a possible implementation, the step of using the wind turbine power prediction fusion neural network model completed by staged training to predict the wind turbine output power at a future time according to the weather data and emergency data at each time includes: t-1 , D t-2 …is input into the wind turbine power prediction fusion neural network model that has been trained in stages, and the wind turbine output power at multiple moments in the future is predicted.

[0061] The working process of the wind turbine power prediction fusion neural network model in the embodiment of the present invention includes:

[0062] 1. Construct a dynamic graph matrix based on the geographical planning information of the wind farm and the operating parameters of the wind turbines; specifically, the operating data of the past year can be converted into a predetermined input format for training the neural network.

[0063] 2. The model that has been trained in stages is used for actual operation prediction. The weather data, emergency data and other data at each moment are converted into input data in a predetermined format through a program to predict the output power information of the next time step.

[0064] 3. The weather data, emergency data and other data at each moment are converted into input data in a predetermined format through a program, which is used to continuously train the model to achieve self-learning function.

[0065] The wind turbine power prediction fusion neural network model of the embodiment of the present invention can be deployed quickly, and can comprehensively consider the wind farm geography and wind turbine information for power prediction. It has a self-learning function, and the longer it is used, the closer the output result is to the accurate value. At the same time, it can handle a large amount of complex information and emergencies, and has strong flexibility and scalability.

[0066] The embodiment of the present invention combines time series neural networks (such as LSTM or BiLSTM, etc.), graph neural networks (GNNs) and attention mechanisms to construct a fused neural network, which is a challenging task. This fusion model aims to simultaneously capture the temporal dependencies in time series data, the complex relationships in graph data, and the important features in the input data.

[0067] In a possible model structure design, the input layer of the time series neural network part (such as BiLSTM) is used to receive the preprocessed time series data. The BiLSTM layer uses a bidirectional long short-term memory network (BiLSTM) to capture the forward and backward dependencies in the sequence data. An attention layer is added after the BiLSTM layer to calculate the weight of each time step and perform weighted summation on the output of the BiLSTM layer to highlight important time step information.

[0068] The graph embedding layer of the graph neural network part (such as Graph Attention Network, GAT) converts the nodes in the graph data into low-dimensional vector representations. The graph attention layer uses the graph attention network (GAT) to calculate the attention weights between the node and its neighbors, and updates the node representation according to these weights to capture the complex relationships in the graph.

[0069] The fusion layer is used for feature fusion, fusing the output of the temporal neural network (BiLSTM+Attention) with the output of the graph neural network (GAT). This can be achieved through concatenation, addition, or more complex operations (such as attention mechanisms). The output layer, based on the fused features, uses an appropriate output layer (such as the softmax function) for the final classification or regression task.

[0070] When training and optimizing the fused neural network model, establish a loss function and select an appropriate loss function according to the specific task, such as cross entropy loss (classification task) or mean square error loss (regression task). Determine the optimization algorithm and use the gradient descent method (such as Adam optimizer) to optimize the model parameters. At the same time, regularization and overfitting processing may also be involved: use L1 / L2 regularization, Dropout and other methods to prevent overfitting.

[0071] Another embodiment of the present invention further provides a wind turbine power prediction system based on a recurrent neural network, comprising:

[0072] A wind turbine power prediction fusion neural network model building module, which is used to combine the time series neural network and the graph neural network and use the attention mechanism to construct a wind turbine power prediction fusion neural network model;

[0073] The model training module is used to construct a dynamic graph matrix based on the geographical distribution of wind turbines in the wind farm and the operating parameters of the wind turbines, and convert the historical operating data of the wind turbines in the wind farm into a set input format to perform phased training on the fusion neural network model for wind turbine power prediction;

[0074] The wind turbine output power prediction and model self-learning module is used to use the wind turbine power prediction fusion neural network model that has been trained in stages to predict the wind turbine output power at future moments based on the weather data and emergency data at each moment, and complete the self-learning of the wind turbine power prediction fusion neural network model.

[0075] Another embodiment of the present invention further provides an electronic device, comprising:

[0076] A memory storing at least one instruction; and a processor executing the instruction stored in the memory to implement the wind turbine power prediction method based on a recurrent neural network.

[0077] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the wind turbine power prediction method based on a recurrent neural network is implemented.

[0078] Exemplarily, the instructions stored in the memory may be divided into one or more modules / units, which are stored in a computer-readable storage medium and executed by the processor to complete the wind turbine power prediction method based on a recurrent neural network of the present invention. The one or more modules / units may be a series of computer-readable instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the server.

[0079] The electronic device may be a computing device such as a smart phone, a notebook, a PDA, and a cloud server. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the electronic device may also include more or fewer components, or a combination of certain components, or different components, for example, the electronic device may also include an input / output device, a network access device, a bus, etc.

[0080] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0081] The memory may be an internal storage unit of the server, such as a hard disk or memory of the server. The memory may also be an external storage device of the server, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the server. Furthermore, the memory may include both an internal storage unit of the server and an external storage device. The memory is used to store the computer-readable instructions and other programs and data required by the server. The memory may also be used to temporarily store data that has been output or is to be output.

[0082] It should be noted that the information interaction, execution process and other contents between the above-mentioned module units are based on the same concept as the method embodiment. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0083] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0084] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. 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. The computer-readable medium may at least include: any entity or device that can carry the computer program code to the camera device / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk.

[0085] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0086] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A wind turbine power prediction method based on recurrent neural network, It is characterized in that include: Combining time series neural network and graph neural network and using attention mechanism to construct a fusion neural network model for wind turbine power prediction; A dynamic graph matrix is ​​constructed based on the geographical distribution of wind turbines in the wind farm and the operating parameters of the wind turbines. The historical operating data of wind turbines in the wind farm is converted into a set input format, and the wind turbine power prediction fusion neural network model is trained in stages. By using the wind turbine power prediction fusion neural network model that has been trained in stages, the wind turbine output power at future moments is predicted according to the weather data and emergency data at each moment, and the self-learning of the wind turbine power prediction fusion neural network model is completed.

2. According to the wind turbine power prediction method based on recurrent neural network according to claim 1, It is characterized in that In the wind turbine power prediction fusion neural network model, the construction method of the time series neural network model is as follows: Given a wind farm wind turbine geographic coordinate set {P 1 ,P 2 ,P 3 ,…,P N }, where each coordinate contains latitude, longitude and altitude information; given a time step {T 1 ,T 2 ,…T M } and the corresponding input feature D, the input feature D represents the weather data with coordinates in a single time step, the maximum output power of the wind turbine, the power demand of the power grid and the characteristics of the emergency event vector; the training goal is to predict the wind turbine output power at future times given the input feature D and the historical wind turbine output power.

3. According to the wind turbine power prediction method based on recurrent neural network according to claim 1, It is characterized in that In the wind turbine power prediction fusion neural network model, the graph neural network model is constructed as follows: A dynamic graph matrix is ​​used to establish nodes for each wind turbine, and all nodes are connected with an undirected graph; A feature vector is assigned to each node as the input data of the graph neural network model to capture the connection and spatial dependency between every two nodes; the connection strength between two nodes is represented by the adjacency matrix; the similarity between nodes is captured by calculating the dot product of the node embedding and applying the activation function to improve the graph structure.

4. According to the wind turbine power prediction method based on recurrent neural network according to claim 1, It is characterized in that In the wind turbine power prediction fusion neural network model, the attention mechanism is constructed as follows: Use a time series neural network model to generate an embedding for each time step to capture the impact of historical weather and historical output on future output, and use a self-attention mechanism to capture the impact of emergencies on output power; add elements to the input features of each node to input emergencies, including wind turbine failures and maintenance plans; Set the self-attention mechanism training weight matrix W to analyze the weight value of each element of the input feature.

5. According to the wind turbine power prediction method based on recurrent neural network according to claim 1, It is characterized in that In the wind turbine power prediction fusion neural network model, the loss function is the sum of the following losses: the mean square error between the output prediction value and the actual value, the distance between the dynamic graph adjacency matrix and the reconstructed adjacency matrix, and multiple additional loss functions for preventing overfitting.

6. According to the wind turbine power prediction method based on recurrent neural network according to claim 1, It is characterized in that The step of performing phased training on the wind turbine power prediction fusion neural network model comprises: An initial attribute graph is constructed based on the geographical distribution of wind turbines in the wind farm and the operating parameters of the wind turbines. The initial attribute graph and input features are passed to the graph neural network layer and the recurrent neural network layer of the wind turbine power prediction fusion neural network model to generate all node embeddings. All node embeddings are output to the fully connected layer, and the predicted value at the current moment is output. Through iterative calculation, the loss value calculated by the loss function is reduced, and all parameters of the wind turbine power prediction fusion neural network model are synchronously updated. When the loss value is reduced to the threshold or the trend converges, the iteration is ended and the training of the wind turbine power prediction fusion neural network model is completed.

7. According to the wind turbine power prediction method based on recurrent neural network according to claim 1, It is characterized in that The step of using the wind turbine power prediction fusion neural network model that has been trained in stages to predict the wind turbine output power at future moments based on weather data and emergency data at each moment includes: inputting input features before a certain moment in the future into the wind turbine power prediction fusion neural network model that has been trained in stages to predict the wind turbine output power at multiple moments in the future.

8. A wind turbine power prediction system based on recurrent neural network, It is characterized in that include: A wind turbine power prediction fusion neural network model building module, which is used to combine the time series neural network and the graph neural network and use the attention mechanism to construct a wind turbine power prediction fusion neural network model; The model training module is used to construct a dynamic graph matrix based on the geographical distribution of wind turbines in the wind farm and the operating parameters of the wind turbines, and convert the historical operating data of the wind turbines in the wind farm into a set input format to perform phased training on the fusion neural network model for wind turbine power prediction; The wind turbine output power prediction and model self-learning module is used to use the wind turbine power prediction fusion neural network model that has been trained in stages to predict the wind turbine output power at future moments based on the weather data and emergency data at each moment, and complete the self-learning of the wind turbine power prediction fusion neural network model.

9. An electronic device, It is characterized in that include: A memory storing at least one instruction; and A processor executes instructions stored in the memory to implement the wind turbine power prediction method based on a recurrent neural network as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program. It is characterized in that When the computer program is executed by a processor, the wind turbine power prediction method based on a recurrent neural network as described in any one of claims 1 to 7 is implemented.

Images