Wind power prediction method, device, computer equipment and storage medium

Abstract

The present application relates to the technical field of wind power prediction, and in particular to a method, device, computer equipment and storage medium for wind power prediction. The method comprises: obtaining time series data of historical wind power, wherein the time series data includes wind power values ​​at different time points, and respectively training an extreme learning machine model, a long short-term memory neural network model and a time convolutional network model; then training a multi-objective prediction model; using a multi-objective Harris Eagle algorithm to obtain the Pareto optimal solution set of the combined weight vector of the multi-objective prediction model; using a deep deterministic policy gradient method to obtain the dynamic optimal weights of each model in the multi-objective prediction model according to the Pareto optimal solution set; according to each model in the multi-objective prediction model and the corresponding dynamic optimal weights, obtaining the prediction results of short-term wind power, and visually displaying the prediction results. The present method can improve the accuracy and flexibility of the prediction results of short-term wind power.

Description

Technical Field

[0001] The present application relates to the technical field of wind power prediction, and in particular to a wind power prediction method, device, computer equipment and storage medium. Background Art

[0002] Wind energy is a clean, renewable energy source. Its development and utilization are of great significance for promoting energy structure adjustment and reducing environmental pollution. Wind power forecasting is one of the key technologies for wind energy utilization and has an important impact on the operation of wind power plants and the stability of power systems. In order to improve the accuracy and reliability of wind power forecasting, researchers are constantly exploring new forecasting methods and technologies.

[0003] At present, a single prediction model is often used to predict short-term wind power. However, a single prediction model often finds it difficult to take into account the complex interactions between various factors, and the accuracy and flexibility of the prediction results of short-term wind power are low.

[0004] Therefore, there is an urgent need for a wind power prediction method, apparatus, computer equipment and storage medium to improve the accuracy and flexibility of short-term wind power prediction results. Summary of the invention

[0005] Based on this, it is necessary to provide a wind power prediction method, device, computer equipment and storage medium that can improve the accuracy and flexibility of short-term wind power prediction results in response to the above technical problems.

[0006] In a first aspect, the present application provides a wind power prediction method, comprising:

[0007] Acquire time series data of historical wind power, wherein the time series data includes wind power values ​​at different time points;

[0008] Normalizing the time series data of the historical wind power, and training an extreme learning machine model, a long short-term memory neural network model, and a time convolutional network model respectively according to the time series data of the historical wind power after the normalization process;

[0009] Training a multi-objective prediction model according to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model;

[0010] Using a multi-objective Harris Hawk algorithm, obtaining a Pareto optimal solution set of a combined weight vector of the multi-objective prediction model;

[0011] Using a deep deterministic policy gradient method, according to the Pareto optimal solution set, to obtain the dynamic optimal weight of each model in the multi-objective prediction model;

[0012] According to each model in the multi-objective prediction model and the corresponding dynamic optimal weight, a prediction result of short-term wind power is obtained, and the prediction result is visualized and displayed.

[0013] In one embodiment, the normalizing the time series data of the historical wind power, and training an extreme learning machine model, a long short-term memory neural network model, and a time convolutional network model respectively according to the time series data of the historical wind power after the normalization, include:

[0014] Using the sequence decomposition method CEEMDAN, the time series data of the historical wind power is decomposed to obtain a plurality of first subsequence data and a residual sequence;

[0015] The extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model are trained simultaneously through the multiple first subsequence data.

[0016] In one of the embodiments, obtaining the wind power prediction result according to each model in the multi-objective prediction model and the corresponding dynamic optimal weight, and visually displaying the prediction result includes:

[0017] Obtain time series data of target wind power;

[0018] Using a sequence decomposition method CEEMDAN, the time series data of the target wind power is decomposed to obtain a plurality of second subsequence data;

[0019] By using each model in the multi-objective prediction model, the plurality of second subsequence data are traversed and predicted at the same time to obtain corresponding first prediction results, second prediction results and third prediction results;

[0020] According to the dynamic optimal weights corresponding to each model, the first prediction result, the second prediction result and the third prediction result are weightedly calculated to obtain a prediction result of short-term wind power.

[0021] In one embodiment, the training of a multi-objective prediction model according to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model includes:

[0022] Obtain time series data of test wind power;

[0023] Using the sequence decomposition method CEEMDAN, the time series data of the test wind power is decomposed to obtain a plurality of third subsequence data;

[0024] According to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model, the plurality of third subsequence data are traversed and tested at the same time to obtain corresponding first test results, second test results and third test results;

[0025] Obtaining mean square errors and standard deviations between the actual short-term wind power and the first test result, the second test result, and the third test result, respectively;

[0026] The minimum mean square error and the minimum standard deviation are used as optimization objectives to train the multi-objective prediction model.

[0027] In one embodiment, the deep deterministic policy gradient method is used to obtain the dynamic optimal weight of each model in the multi-objective prediction model according to the Pareto optimal solution set, including:

[0028] Using a deep deterministic policy gradient method, the Pareto optimal solution set is defined as the action vector of the agent;

[0029] The weights and biases of the multi-objective prediction model are iteratively updated through the action vector of the agent to obtain the dynamic optimal weight of each model in the multi-objective prediction model.

[0030] In one embodiment, iteratively updating the weights and biases of the multi-objective prediction model through the action vector of the agent includes:

[0031] Selecting a static weight coefficient from the Pareto optimal solution set based on the minimum mean absolute percentage error principle through the action vector of the agent;

[0032] Obtaining a reference error according to the static weight coefficient and the actual short-term wind power;

[0033] Analyzing the correlation between the minimum value of the mean square error, the minimum value of the standard deviation, and the reference error;

[0034] According to the correlation, obtaining a reward function;

[0035] The weights and biases of the multi-objective prediction model are iteratively updated through the reward function.

[0036] In a second aspect, the present application also provides a wind power prediction device, comprising:

[0037] An acquisition module, used to acquire time series data of historical wind power, wherein the time series data includes wind power values ​​at different time points;

[0038] A model building module, used for normalizing the time series data of the historical wind power, and training an extreme learning machine model, a long short-term memory neural network model and a time convolutional network model respectively according to the time series data of the historical wind power after the normalization process;

[0039] The model building module is also used to train a multi-objective prediction model according to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model;

[0040] A processing module, used for obtaining a Pareto optimal solution set of a combined weight vector of the multi-objective prediction model by using a multi-objective Harris Hawk algorithm;

[0041] The processing module is further used to adopt a deep deterministic policy gradient method to obtain the dynamic optimal weight of each model in the multi-objective prediction model according to the Pareto optimal solution set;

[0042] The processing module is further used to obtain the prediction result of wind power according to each model in the multi-objective prediction model and the corresponding dynamic optimal weight;

[0043] The display module is used to visually display the prediction results.

[0044] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0045] Acquire time series data of historical wind power, wherein the time series data includes wind power values ​​at different time points;

[0046] Normalizing the time series data of the historical wind power, and training an extreme learning machine model, a long short-term memory neural network model, and a time convolutional network model respectively according to the time series data of the historical wind power after the normalization process;

[0047] Training a multi-objective prediction model according to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model;

[0048] Using a multi-objective Harris Hawk algorithm, obtaining a Pareto optimal solution set of a combined weight vector of the multi-objective prediction model;

[0049] Using a deep deterministic policy gradient method, according to the Pareto optimal solution set, to obtain the dynamic optimal weight of each model in the multi-objective prediction model;

[0050] According to each model in the multi-objective prediction model and the corresponding dynamic optimal weight, a prediction result of wind power is obtained, and the prediction result is visualized and displayed.

[0051] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0052] Acquire time series data of historical wind power, wherein the time series data includes wind power values ​​at different time points;

[0053] Normalizing the time series data of the historical wind power, and training an extreme learning machine model, a long short-term memory neural network model, and a time convolutional network model respectively according to the time series data of the historical wind power after the normalization process;

[0054] Training a multi-objective prediction model according to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model;

[0055] Using a multi-objective Harris Hawk algorithm, obtaining a Pareto optimal solution set of a combined weight vector of the multi-objective prediction model;

[0056] Using a deep deterministic policy gradient method, according to the Pareto optimal solution set, to obtain the dynamic optimal weight of each model in the multi-objective prediction model;

[0057] According to each model in the multi-objective prediction model and the corresponding dynamic optimal weight, a prediction result of wind power is obtained, and the prediction result is visualized and displayed.

[0058] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0059] Acquire time series data of historical wind power, wherein the time series data includes wind power values ​​at different time points;

[0060] Normalizing the time series data of the historical wind power, and training an extreme learning machine model, a long short-term memory neural network model, and a time convolutional network model respectively according to the time series data of the historical wind power after the normalization process;

[0061] Training a multi-objective prediction model according to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model;

[0062] Using a multi-objective Harris Hawk algorithm, obtaining a Pareto optimal solution set of a combined weight vector of the multi-objective prediction model;

[0063] Using a deep deterministic policy gradient method, according to the Pareto optimal solution set, to obtain the dynamic optimal weight of each model in the multi-objective prediction model;

[0064] According to each model in the multi-objective prediction model and the corresponding dynamic optimal weight, a prediction result of wind power is obtained, and the prediction result is visualized and displayed.

[0065] The above-mentioned wind power prediction method, device, computer equipment and storage medium, by adopting a combination of multiple prediction models for prediction, can take into account the complex interactions between different models, thereby improving the accuracy of short-term wind power prediction results. By using multiple prediction models and adjusting the dynamic optimal weights, the weights of each model can be flexibly adjusted according to actual conditions, thereby improving the adaptability and flexibility of the prediction method. By using the multi-objective Harris Eagle algorithm to obtain the Pareto optimal solution set of the combined weight vector of the prediction model, multiple prediction strategies can be provided for selection while ensuring the accuracy of the prediction. By using the deep deterministic policy gradient method, the optimal weights of each model can be dynamically adjusted according to the real-time Pareto optimal solution set, further improving the accuracy and adaptability of the prediction. This method can improve the accuracy and flexibility of short-term wind power prediction results, and provide important support for wind energy utilization and the stability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0067] FIG1 is a diagram showing an application environment of a wind power prediction method according to an embodiment;

[0068] FIG2 is a schematic diagram of a flow chart of a wind power prediction method in one embodiment;

[0069] FIG3 is a schematic flow chart of a wind power prediction method in another embodiment;

[0070] FIG4 is a structural block diagram of a wind power prediction device in one embodiment;

[0071] FIG. 5 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0072] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0073] The wind power prediction method provided in the embodiment of the present application can be applied in the application environment shown in FIG1 . The terminal 102 communicates with the server 104 through a network. The data storage system can store data that the server 104 needs to process. The data storage system can be integrated on the server 104, or placed on the cloud or other network servers.

[0074] The server 104 obtains the time series data of historical wind power, which includes wind power values ​​at different time points; normalizes the time series data of historical wind power, and trains the extreme learning machine model, the long short-term memory neural network model and the time convolution network model according to the time series data of historical wind power after normalization; trains the multi-objective prediction model according to the extreme learning machine model, the long short-term memory neural network model and the time convolution network model; adopts the multi-objective Harris Eagle algorithm to obtain the Pareto optimal solution set of the combined weight vector of the multi-objective prediction model; adopts the deep deterministic policy gradient method to obtain the dynamic optimal weight of each model in the multi-objective prediction model according to the Pareto optimal solution set; obtains the prediction result of wind power according to each model in the multi-objective prediction model and the corresponding dynamic optimal weight. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices, and the Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers.

[0075] In an exemplary embodiment, as shown in FIG2 , a wind power prediction method is provided, which is described by taking the method applied to the server in FIG1 as an example, and includes the following steps S202 to S212. Among them:

[0076] Step S202: Acquire time series data of historical wind power, where the time series data includes wind power values ​​at different time points.

[0077] Specifically, obtaining the time series data of historical wind power means collecting and recording the data of wind power generated by wind power plants in the past period of time, which are arranged in chronological order to form a time series. Time series data contains wind power values ​​at different time points and can be used to analyze and predict the performance and behavior of wind power systems. By drawing time series graphs or applying statistical methods, the changing trend of wind power in different time periods can be observed, such as hourly, daily, quarterly or annual changes. By analyzing time series data, the normal operating mode of wind power systems can be understood, including high-load and low-load periods, periodic fluctuations, etc. By observing historical data, the power generation capacity, stability and reliability of wind power systems in different time periods can be evaluated, providing a basis for system operation and equipment maintenance. Using the time series data of historical wind power, wind resources can be evaluated and analyzed, potential wind power generation in the future can be predicted, and the layout and operation and maintenance strategies of wind power systems can be optimized.

[0078] Step S204, normalizing the time series data of historical wind power, and training an extreme learning machine model, a long short-term memory neural network model, and a temporal convolutional network model respectively based on the normalized time series data of historical wind power.

[0079] Specifically, the time series data of historical wind power is normalized. According to the normalized time series data of historical wind power, the following are the general steps for training wind power prediction using three models: Extreme Learning Machine (ELM), Long Short-Term Memory Neural Network (LSTM) and Time Convolutional Neural Network (TCN).

[0080] Extreme Learning Machine (ELM) Model:

[0081] Prepare the data set: Divide the historical wind power time series data into training set and test set, and normalize the data.

[0082] Constructing the ELM model: ELM is a single hidden layer feedforward neural network with randomly initialized weight matrix and bias vector. The wind power data is input into the input layer, and the hidden layer output is calculated through the random weight matrix and bias vector.

[0083] Training model: Adjust the hidden layer weight matrix and bias vector through least squares method or regularization to make the model output as close to the actual wind power value as possible.

[0084] Model evaluation: Use the test set data to evaluate the prediction performance of the model, such as calculating indicators such as the root mean square error (RMSE).

[0085] Long Short-Term Memory Neural Network (LSTM) Model:

[0086] Prepare the data set: Convert the historical wind power time series data into input series and target series, and normalize the data.

[0087] Build an LSTM model: LSTM is a recurrent neural network suitable for sequence data. It has memory units and gating mechanisms to capture long-term dependencies in sequences. The model includes an input layer, an LSTM layer, and an output layer.

[0088] Training model: Through the back-propagation algorithm and gradient descent optimizer, the weights and biases of the model are adjusted according to the input sequence and the target sequence, so that the model can accurately predict wind power.

[0089] Model evaluation: Use the test set data to evaluate the prediction performance of the model, such as calculating indicators such as the root mean square error (RMSE).

[0090] Temporal Convolutional Network (TCN) Model:

[0091] Prepare the dataset: Convert the historical wind power time series data into input sequence and target sequence, and perform normalization.

[0092] Build TCN model: TCN is a sequence modeling method based on convolutional neural network, which captures long-term dependencies in the sequence through one-dimensional convolutional layers and residual blocks. The model includes input layer, convolutional layer, residual block and output layer.

[0093] Training model: Through the back-propagation algorithm and gradient descent optimizer, the weights and biases of the model are adjusted according to the input sequence and the target sequence, so that the model can accurately predict wind power.

[0094] Among them, when training the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model, some parameters of each model will be set based on empirical values, such as:

[0095] ELM model: The number of neurons in the hidden layer is set to 16;

[0096] LSTM model; the number of neurons in the hidden layer is set to 100; the learning rate is set to 0.001; the maximum number of rounds is set to 100; the number of convolution kernels is set to 64;

[0097] TCN model: the number of convolution kernels is set to 64; the number of training batches is set to 32; the number of iterations is set to 50; and the inactivation rate is set to 0.2.

[0098] Step S206, training a multi-target prediction model based on the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model.

[0099] Specifically, the entire multi-objective model is trained using back propagation and optimization algorithms, with the goal of minimizing the model's error on the target variable. The model can be trained using gradient descent or other optimization algorithms. The trained multi-objective prediction model is evaluated using a test dataset. The evaluation metric can be selected based on the type of target variable, such as Root Mean Squared Error (RMSE) or Mean Absolute Error (MAE). Once the model is trained and evaluated, it can be used to predict new target variable values.

[0100] Among them, the expression of the multi-objective prediction model is:

[0101]

[0102] In the formula, yes The measured power at the time, Respectively The prediction of a single model at time yes The combined weight of the moment.

[0103] Step S208, using the multi-objective Harris Hawk algorithm to obtain the Pareto optimal solution set of the combined weight vector of the multi-objective prediction model.

[0104] Specifically, the multi-objective Harris Hawk algorithm is a commonly used method for solving multi-objective optimization problems. The multi-objective prediction task is transformed into a multi-objective optimization problem, in which each target variable can be obtained by different prediction models, and an objective function can be defined to measure the optimization target of each target. This objective function may include indicators such as prediction error, generalization ability, and complexity. Determine the feasible domain of the Pareto optimal solution set. The feasible domain determines the value range of the weight vector, and constraints can be set according to actual needs. A group of eagles is generated using a random initialization method. The individual eagle represents a possible weight vector. Through continuous iteration and optimization, the group gradually converges to the optimal solution. Calculate the fitness of each eagle individual, and the fitness will evaluate the quality of each individual according to the value of the objective function, usually based on the definition of the Pareto optimal solution. Update the position of each eagle individual according to the selected update strategy. The update strategy can adopt the search strategy in the standard Harris Hawk algorithm, such as exploration and tracking, population search, and encounter and fusion. Check whether the eagle group converges to a stable Pareto optimal solution set. If the stopping condition is not met, continue iterative optimization. The final stable group is the Pareto optimal solution set, which contains the optimal solutions of multiple weight vector combinations.

[0105] Among them, the following multi-objective optimization is defined:

[0106] ;

[0107] in , are constraints. and Find the minimum value of f(x) under the condition .

[0108] Given a vector and , ( );

[0109] If and only if hour, ;

[0110] and is a Pareto optimal solution if , then the set of Pareto optimal solutions is recorded as ;

[0111] Further, It is called the Pareto frontier. All Pareto optimal solutions constitute the Pareto optimal solution set. These solutions constitute the Pareto optimal frontier or Pareto frontier surface of the problem through the mapping of the objective function. That is, the objective function value corresponding to the Pareto optimal solution is the Pareto optimal frontier.

[0112] Step S210, using a deep deterministic policy gradient method, according to the Pareto optimal solution set, obtains the dynamic optimal weight of each model in the multi-objective prediction model.

[0113] Among them, Deep Deterministic Policy Gradient (DDPG) is a commonly used reinforcement learning algorithm that can be used to learn dynamic optimal weights.

[0114] Specifically, the state space and action space of the problem are defined. In the multi-objective prediction model, the state space can be indicators such as the prediction error and generalization ability of each prediction model, and the action space can be each weight in the weight vector. Construct a DDPG network consisting of two neural networks, one is the Actor network for learning the action strategy, and the other is the Critic network for evaluating the value of the action. The Actor network outputs the action vector of each weight in the current state, and the Critic network outputs a Q value to evaluate the value of the action based on the Actor's action vector and state information. Randomly initialize the parameters of the Actor network and the Critic network. According to the DDPG strategy, use the Actor network to select and execute actions to obtain reward signals and the next state. Use these samples to update the parameters of the Critic network, and use the gradient descent method to minimize the loss function of the Critic network. At the same time, update the parameters of the Actor network according to the output of the Critic network, and use the gradient ascent method to maximize the objective function of the Actor network. Repeat sampling, optimization, and updating of network parameters until the network converges or reaches the preset number of iterations. After the DDPG training is completed, the dynamic optimal weights of each model in the multi-objective prediction model can be obtained through the Actor network. Use the Actor network to input the current state and get the output weight vector as the weight of the model.

[0115] Step S212, according to each model in the multi-objective prediction model and the corresponding dynamic optimal weight, obtain the prediction result of wind power, and visualize the prediction result.

[0116] According to the current wind input and the characteristics of the prediction model, use each prediction model to obtain its own prediction results. Apply the corresponding dynamic optimal weight to the prediction results of each prediction model. Multiply each prediction result with the corresponding dynamic optimal weight to obtain a weighted prediction result. Add the weighted prediction results to obtain the final wind power prediction result.

[0117] In the above-mentioned wind power prediction method, by adopting a combination of multiple prediction models for prediction, the complex interactions between different models can be taken into account, thereby improving the accuracy of short-term wind power prediction results. By using multiple prediction models and adjusting the dynamic optimal weights, the weights of each model can be flexibly adjusted according to actual conditions, thereby improving the adaptability and flexibility of the prediction method. By using the multi-objective Harris Eagle algorithm to obtain the Pareto optimal solution set of the combined weight vector of the prediction model, multiple prediction strategies can be provided for selection while ensuring the accuracy of the prediction. By using the deep deterministic policy gradient method, the optimal weights of each model can be dynamically adjusted according to the real-time Pareto optimal solution set, further improving the accuracy and adaptability of the prediction. This method can improve the accuracy and flexibility of short-term wind power prediction results, and provide important support for wind energy utilization and the stability of the power system.

[0118] In an exemplary embodiment, the time series data of historical wind power is normalized, and according to the time series data of historical wind power after normalization, an extreme learning machine model, a long short-term memory neural network model and a time convolutional network model are trained respectively, including:

[0119] The sequence decomposition method CEEMDAN is used to decompose the time series data of historical wind power to obtain multiple first subsequence data and residual sequences. The extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model are trained simultaneously through multiple first subsequence data.

[0120] Specifically, the sequence decomposition method CEEMDAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise) is used to decompose the time series data of historical wind power to obtain multiple first subsequence data. CEEMDAN can effectively decompose the original sequence into multiple subsequences, each of which represents a different frequency component or mode. The multiple first subsequence data obtained by decomposition are used as input, and the corresponding training set, validation set, and test set are set according to the needs to ensure that the data division is reasonable.

[0121] Compared with traditional decomposition methods (EMD, VMD and WPD), CEEMDAN overcomes the problem of modal aliasing in empirical mode decomposition and the problem of random residual noise in aggregated empirical mode decomposition. It adds Gaussian white noise at each stage of decomposition and then performs EMD decomposition, which solves the problem of residual noise and can fully decompose time series with complex changing characteristics such as wind power.

[0122] In this embodiment, the ELM model, LSTM model and TCN model are trained respectively according to the multiple first subsequence data decomposed by the CEEMDAN method, so that the most suitable model can be selected according to the needs, and the wind power can be predicted according to the historical wind power data.

[0123] In an exemplary embodiment, as shown in FIG3 , according to each model in the multi-objective prediction model and the corresponding dynamic optimal weight, the prediction result of wind power is obtained, including:

[0124] Step S302, obtaining time series data of target wind power;

[0125] Step S304, using a sequence decomposition method CEEMDAN to decompose the time series data of the target wind power to obtain a plurality of second subsequence data;

[0126] Step S306, using each model in the multi-objective prediction model, traversing and predicting the plurality of second subsequence data at the same time, to obtain corresponding first prediction results, second prediction results, and third prediction results;

[0127] Step S308: performing weighted calculation on the first prediction result, the second prediction result and the third prediction result according to the dynamic optimal weights corresponding to each model to obtain a prediction result of the target wind power.

[0128] Specifically, the time series data of the target wind power is obtained from a reliable data source, sensor or historical record. The integrity and accuracy of the data are ensured. The time series data of the target wind power is decomposed using the CEEMDAN method to generate multiple subsequence data. CEEMDAN first extracts the low-frequency components in the data in an iterative manner, and then gradually extracts the high-frequency components to form a series of intrinsic mode functions (IMFs). These subsequences represent components in different frequency ranges. For each second subsequence data, multiple prediction models are used for traversal prediction. The extreme learning machine model, the long short-term memory neural network model and the time convolution network model are used to predict the input second subsequence data to obtain the corresponding first prediction result, the second prediction result and the third prediction result. According to the dynamic optimal weights of each model, the first prediction result, the second prediction result and the third prediction result are weighted. The weights can be determined according to the model performance or other indicators (such as the root mean square error). The weights can be evenly distributed or dynamically distributed according to the model performance. The prediction result after weighted calculation is used as the final prediction result of the target wind power. The result can provide an estimate of the target wind power at a future time point.

[0129] In this embodiment, by using a combination of multiple prediction models to perform prediction, the complex interactions between different models can be taken into account, thereby improving the accuracy of short-term wind power prediction results. By using multiple prediction models and adjusting the dynamic optimal weights, the weights of each model can be flexibly adjusted according to actual conditions, thereby improving the adaptability and flexibility of the prediction method.

[0130] In an exemplary embodiment, a multi-objective prediction model is trained according to an extreme learning machine model, a long short-term memory neural network model, and a temporal convolutional network model, including:

[0131] Obtain time series data of test wind power;

[0132] The sequence decomposition method CEEMDAN is used to decompose the time series data of the test wind power to obtain multiple third subsequence data;

[0133] According to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model, a plurality of third subsequence data are traversed and tested at the same time to obtain corresponding first test results, second test results and third test results;

[0134] Obtaining mean square errors and standard deviations between the actual wind power and the first test result, the second test result, and the third test result, respectively;

[0135] The minimum mean square error and the minimum standard deviation are used as optimization objectives to train the multi-objective prediction model.

[0136] Specifically, firstly, it is necessary to obtain time series data of the test wind power, which can be historical data or real-time measured data.

[0137] Then, the sequence decomposition method CEEMDAN is used to decompose the time series data of the test wind power. CEEMDAN is a data decomposition method that can decompose non-stationary time series data into multiple subsequences. Using the extreme learning machine (ELM), long short-term memory neural network (LSTM) model and temporal convolutional network (TCN) model, these subsequence data are traversed and predicted to obtain the corresponding first test results, second test results and third test results.

[0138] Then, the actual wind power is compared with the first test result, the second test result and the third test result, and the mean square error and standard deviation between them are calculated. These indicators measure the degree of difference between the predicted results and the actual values.

[0139] Finally, the minimum value of the mean square error and the minimum value of the standard deviation are used as optimization objectives to train a multi-objective prediction model. This model can comprehensively consider the minimization of these two indicators according to a specific optimization algorithm and objective function to improve the accuracy and stability of the prediction.

[0140] In this embodiment, using a multi-objective optimization framework can help select the best prediction model and provide more reliable and accurate wind power prediction results.

[0141] In an exemplary embodiment, a deep deterministic policy gradient method is used to obtain the dynamic optimal weight of each model in the multi-objective prediction model according to the Pareto optimal solution set, including:

[0142] Using a deep deterministic policy gradient method, the Pareto optimal solution set is defined as the action vector of the agent;

[0143] Through the action vector of the intelligent agent, the weights and biases of the multi-objective prediction model are iteratively updated to obtain the dynamic optimal weights of each model in the multi-objective prediction model.

[0144] Specifically, in a multi-objective prediction model, the Pareto optimal solution set refers to a set of solutions that cannot be optimized by other solutions on multiple objectives. In order to obtain the dynamic optimal weights of each model in the multi-objective prediction model, the Deep Deterministic Policy Gradient (DDPG) method can be used. First, the Pareto optimal solution set is defined as the action vector of the agent. The agent here refers to a trained reinforcement learning-based algorithm used to optimize the weights and biases of the multi-objective prediction model. The action vector refers to the value of the model weights and biases selected at each time step. Then, by using the DDPG algorithm, the weights and biases of the multi-objective prediction model are iteratively updated to obtain the dynamic optimal weights of each model in the multi-objective prediction model. The DDPG algorithm is a deep reinforcement learning algorithm that uses two neural networks, one is the Actor network for estimating the action value function, and the other is the Critic network for estimating the state value function. By training these two networks and updating the network parameters using the gradient descent method, a strategy for selecting the optimal action in the current state can be obtained. In the DDPG algorithm, the agent interacts with the environment, selects actions based on the current state, and collects reward signals. Then, the policy gradient method is used to calculate the gradient of the action, and the gradient ascent method is used to update the parameters of the policy network. At the same time, the gradient of the value function is used to update the parameters of the value function network through the gradient descent method. Through this iterative update method, the dynamic optimal weights of each model in the multi-objective prediction model can be gradually obtained.

[0145] In this embodiment, a deep deterministic policy gradient method is used to define the Pareto optimal solution set as the action vector of the agent, and the dynamic optimal weights of each model in the multi-objective prediction model can be obtained by iteratively updating the weights and biases of the multi-objective prediction model. This can help find the optimal model combination to achieve better multi-objective prediction results.

[0146] In an exemplary embodiment, the weights and biases of the multi-objective prediction model are iteratively updated through the action vector of the agent, including:

[0147] Through the action vector of the intelligent agent, based on the principle of minimum mean absolute percentage error, the static weight coefficient is selected from the Pareto optimal solution set; according to the static weight coefficient and the actual wind power, the reference error is obtained; the correlation between the minimum mean square error, the minimum standard deviation and the reference error is analyzed; according to the correlation, the reward function is obtained; through the reward function, the weights and biases of the multi-objective prediction model are iteratively updated.

[0148] Specifically, through the action vector of the agent, the static weight coefficient can be selected from the Pareto optimal solution set based on the minimum mean absolute percentage error principle. The minimum mean absolute percentage error is a commonly used indicator to measure the prediction error. In the multi-objective prediction model, the average absolute percentage error between the prediction results of each model and the actual wind power can be calculated, and the weight coefficient that minimizes this error can be selected.

[0149] Based on the static weight coefficients and the actual wind power, the reference error can be calculated. The reference error is the difference between the actual wind power and the weighted prediction result of the multi-objective prediction model. By calculating the reference error, you can understand the prediction performance of the model and determine whether the model weights and biases need to be adjusted.

[0150] Next, analyze the correlation between the minimum value of the mean square error, the minimum value of the standard deviation, and the reference error. The mean square error refers to the mean of the square of the difference between the predicted value and the true value, and the standard deviation is the square root of the mean square error. Correlation refers to the degree of linear relationship between two variables. By analyzing these indicators, we can understand the prediction accuracy of the model and the relationship between the prediction results.

[0151] Based on the correlation, a reward function can be designed. A reward function is a function used to guide the behavior of an agent in reinforcement learning. By associating reference error with other metrics, a reward function can be designed to encourage the model to choose more accurate predictions and penalize inaccurate predictions.

[0152] Finally, through the reward function, the weights and biases of the multi-objective prediction model can be iteratively updated. Through interaction with the environment, the agent selects actions based on the current state, collects reward signals, and updates the parameters of the model using the gradient method. Through iterative updates, a multi-objective prediction model with better prediction performance can be gradually obtained.

[0153] Specifically, the DDPG algorithm is used to solve the dynamic optimal weight of the wind power combination prediction. For the wind power combination prediction problem, the following definition is given: It consists of continuous historical wind power characteristic data. ,in is the input length of the model. Action Describe the selection of Pareto optimal solution sets, ,in is the total number of possible actions that the agent can take, that is, the number of Pareto optimal solutions. When there is a solution, .

[0154] The reward function is based on the minimum mean absolute percentage error principle, and selects a set of static weight coefficients from the Pareto optimal solution set of MOHHO. , which is used to calculate the reference error (denoted as ). Then for the action , weight and prediction error It can be expressed as:

[0155] ;

[0156] In the formula, It is The dynamic weights of the model, is the actual power. Reward function The expression is as follows:

[0157] ;

[0158] In the formula, Is a fixed value.

[0159] In this embodiment, the static weight coefficient is selected based on the minimum mean absolute percentage error principle, the reward function is designed by analyzing the correlation between the mean square error, standard deviation and reference error, and the weights and biases of the multi-objective prediction model are iteratively updated through the reward function to improve the prediction performance of the model.

[0160] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0161] Based on the same inventive concept, the embodiment of the present application also provides a wind power prediction device for implementing the wind power prediction method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more wind power prediction device embodiments provided below can refer to the limitations of the wind power prediction method above, and will not be repeated here.

[0162] In an exemplary embodiment, as shown in FIG4 , a wind power prediction device is provided, comprising:

[0163] An acquisition module 402 is used to acquire time series data of historical wind power, wherein the time series data includes wind power values ​​at different time points;

[0164] A model building module 404 is used to normalize the time series data of historical wind power, and train an extreme learning machine model, a long short-term memory neural network model, and a time convolutional network model according to the normalized time series data of historical wind power;

[0165] The model building module 404 is also used to train a multi-objective prediction model based on an extreme learning machine model, a long short-term memory neural network model, and a temporal convolutional network model;

[0166] Processing module 406, used to obtain a Pareto optimal solution set of a combined weight vector of a multi-objective prediction model using a multi-objective Harris Hawk algorithm;

[0167] The processing module 406 is further used to obtain the dynamic optimal weight of each model in the multi-objective prediction model according to the Pareto optimal solution set by using a deep deterministic policy gradient method;

[0168] The processing module 406 is also used to obtain the wind power prediction result according to each model in the multi-objective prediction model and the corresponding dynamic optimal weight.

[0169] The display module 408 is used to visually display the prediction results.

[0170] In one embodiment, the processing module 406 is further used to decompose the time series data of historical wind power using a sequence decomposition method CEEMDAN to obtain a plurality of first subsequence data and a residual sequence;

[0171] The processing module 406 is further used to simultaneously train the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model using multiple first subsequence data.

[0172] In one embodiment, the acquisition module 402 is further used to acquire time series data of target wind power;

[0173] The sequence decomposition method CEEMDAN is used to decompose the time series data of the target wind power to obtain multiple second subsequence data;

[0174] Each model in the multi-objective prediction model simultaneously performs traversal prediction on multiple second subsequence data to obtain corresponding first prediction results, second prediction results and third prediction results;

[0175] The processing module 406 is further used to perform weighted calculation on the first prediction result, the second prediction result and the third prediction result according to the dynamic optimal weights corresponding to each model to obtain a short-term wind power prediction result.

[0176] In one embodiment, the acquisition module 402 is further used to acquire time series data of the test wind power;

[0177] The processing module 406 is further used to decompose the time series data of the test wind power by using the sequence decomposition method CEEMDAN to obtain a plurality of third subsequence data;

[0178] The processing module 406 is further used to perform traversal tests on the plurality of third subsequence data at the same time according to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model to obtain corresponding first test results, second test results and third test results;

[0179] The processing module 406 is further used to obtain the mean square error and standard deviation between the actual short-term wind power and the first test result, the second test result and the third test result respectively;

[0180] The model building module 404 is also used to train the multi-objective prediction model by taking the minimum value of the mean square error and the minimum value of the standard deviation as optimization targets.

[0181] In one embodiment, the processing module 406 is further configured to adopt a deep deterministic policy gradient method to define a Pareto optimal solution set as an action vector of the agent;

[0182] The processing module 406 is also used to iteratively update the weights and biases of the multi-objective prediction model through the action vector of the intelligent agent to obtain the dynamic optimal weights of each model in the multi-objective prediction model.

[0183] In one embodiment, the processing module 406 is further configured to select a static weight coefficient from a Pareto optimal solution set based on a minimum mean absolute percentage error principle using the action vector of the agent;

[0184] The processing module 406 is further used to obtain a reference error according to the static weight coefficient and the actual short-term wind power;

[0185] The processing module 406 is further used to analyze the correlation between the minimum value of the mean square error, the minimum value of the standard deviation and the reference error;

[0186] The processing module 406 is further used to obtain a reward function according to the correlation;

[0187] The model building module 404 is also used to iteratively update the weights and biases of the multi-objective prediction model through a reward function.

[0188] Each module in the above wind power prediction device can be implemented in whole or in part by software, hardware and a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module above.

[0189] In an exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be shown in FIG5. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a wind power prediction method is implemented.

[0190] Those skilled in the art will understand that the structure shown in FIG. 5 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0191] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0192] Acquire time series data of historical wind power, wherein the time series data includes wind power values ​​at different time points;

[0193] The time series data of historical wind power are normalized, and an extreme learning machine model, a long short-term memory neural network model, and a temporal convolutional network model are trained respectively according to the time series data of historical wind power after normalization.

[0194] Train multi-target prediction models based on extreme learning machine models, long short-term memory neural network models, and temporal convolutional network models;

[0195] The multi-objective Harris Eagle algorithm is used to obtain the Pareto optimal solution set of the combined weight vector of the multi-objective prediction model;

[0196] The deep deterministic policy gradient method is used to obtain the dynamic optimal weights of each model in the multi-objective prediction model according to the Pareto optimal solution set;

[0197] According to each model in the multi-objective prediction model and the corresponding dynamic optimal weight, the prediction results of short-term wind power are obtained and the prediction results are displayed visually.

[0198] In one embodiment, when the processor executes the computer program, the following steps are also implemented:

[0199] The time series data of historical wind power are normalized, and according to the normalized time series data of historical wind power, an extreme learning machine model, a long short-term memory neural network model and a time convolutional network model are trained respectively, including:

[0200] The sequence decomposition method CEEMDAN is used to decompose the time series data of historical wind power to obtain multiple first subsequence data and residual sequences;

[0201] The extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model are trained simultaneously through multiple first subsequence data.

[0202] In one embodiment, when the processor executes the computer program, the following steps are also implemented:

[0203] According to each model in the multi-objective prediction model and the corresponding dynamic optimal weight, the wind power prediction results are obtained, including:

[0204] Obtain time series data of target wind power;

[0205] The sequence decomposition method CEEMDAN is used to decompose the time series data of the target wind power to obtain multiple second subsequence data;

[0206] By using each model in the multi-objective prediction model, traversing and predicting the plurality of second subsequence data at the same time, a corresponding first prediction result, a second prediction result and a third prediction result are obtained;

[0207] According to the dynamic optimal weights corresponding to each model, the first prediction result, the second prediction result and the third prediction result are weightedly calculated to obtain the prediction result of short-term wind power.

[0208] In one embodiment, when the processor executes the computer program, the following steps are also implemented:

[0209] According to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model, a multi-objective prediction model is trained, including:

[0210] Obtain time series data of test wind power;

[0211] The sequence decomposition method CEEMDAN is used to decompose the time series data of the test wind power to obtain multiple third subsequence data;

[0212] According to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model, a plurality of third subsequence data are traversed and tested at the same time to obtain corresponding first test results, second test results and third test results;

[0213] Obtaining mean square errors and standard deviations between the actual short-term wind power and the first test result, the second test result, and the third test result, respectively;

[0214] The minimum mean square error and the minimum standard deviation are used as optimization objectives to train the multi-objective prediction model.

[0215] In one embodiment, when the processor executes the computer program, the following steps are also implemented:

[0216] The deep deterministic policy gradient method is used to obtain the dynamic optimal weights of each model in the multi-objective prediction model according to the Pareto optimal solution set, including:

[0217] Using a deep deterministic policy gradient method, the Pareto optimal solution set is defined as the action vector of the agent;

[0218] Through the action vector of the intelligent agent, the weights and biases of the multi-objective prediction model are iteratively updated to obtain the dynamic optimal weights of each model in the multi-objective prediction model.

[0219] In one embodiment, when the processor executes the computer program, the following steps are also implemented:

[0220] The weights and biases of the multi-target prediction model are iteratively updated through the agent’s action vector, including:

[0221] Through the action vector of the intelligent agent, based on the principle of minimum mean absolute percentage error, the static weight coefficient is selected from the Pareto optimal solution set;

[0222] Obtain reference error based on static weight coefficient and actual short-term wind power;

[0223] Analyze the correlation between the minimum value of mean square error, the minimum value of standard deviation and reference error;

[0224] According to the correlation, obtain the reward function;

[0225] Through the reward function, the weights and biases of the multi-objective prediction model are iteratively updated.

[0226] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0227] Acquire time series data of historical wind power, wherein the time series data includes wind power values ​​at different time points;

[0228] The time series data of historical wind power are normalized, and an extreme learning machine model, a long short-term memory neural network model, and a temporal convolutional network model are trained respectively according to the time series data of historical wind power after normalization.

[0229] Train multi-target prediction models based on extreme learning machine models, long short-term memory neural network models, and temporal convolutional network models;

[0230] The multi-objective Harris Eagle algorithm is used to obtain the Pareto optimal solution set of the combined weight vector of the multi-objective prediction model;

[0231] The deep deterministic policy gradient method is used to obtain the dynamic optimal weights of each model in the multi-objective prediction model according to the Pareto optimal solution set;

[0232] According to each model in the multi-objective prediction model and the corresponding dynamic optimal weight, the prediction results of short-term wind power are obtained and the prediction results are displayed visually.

[0233] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0234] The time series data of historical wind power are normalized, and according to the normalized time series data of historical wind power, an extreme learning machine model, a long short-term memory neural network model and a time convolutional network model are trained respectively, including:

[0235] The sequence decomposition method CEEMDAN is used to decompose the time series data of historical wind power to obtain multiple first subsequence data and residual sequences;

[0236] The extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model are trained simultaneously through multiple first subsequence data.

[0237] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0238] According to each model in the multi-objective prediction model and the corresponding dynamic optimal weight, the prediction result of wind power is obtained, and the prediction result is visualized and displayed, including:

[0239] Obtain time series data of target wind power;

[0240] The sequence decomposition method CEEMDAN is used to decompose the time series data of the target wind power to obtain multiple second subsequence data;

[0241] By using each model in the multi-objective prediction model, traversing and predicting the plurality of second subsequence data at the same time, a corresponding first prediction result, a second prediction result and a third prediction result are obtained;

[0242] According to the dynamic optimal weights corresponding to each model, the first prediction result, the second prediction result and the third prediction result are weightedly calculated to obtain the prediction result of short-term wind power.

[0243] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0244] According to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model, a multi-objective prediction model is trained, including:

[0245] Obtain time series data of test wind power;

[0246] The sequence decomposition method CEEMDAN is used to decompose the time series data of the test wind power to obtain multiple third subsequence data;

[0247] According to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model, a plurality of third subsequence data are traversed and tested at the same time to obtain corresponding first test results, second test results and third test results;

[0248] Obtaining mean square errors and standard deviations between the actual short-term wind power and the first test result, the second test result, and the third test result, respectively;

[0249] The minimum mean square error and the minimum standard deviation are used as optimization objectives to train the multi-objective prediction model.

[0250] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0251] The deep deterministic policy gradient method is used to obtain the dynamic optimal weights of each model in the multi-objective prediction model according to the Pareto optimal solution set, including:

[0252] Using a deep deterministic policy gradient method, the Pareto optimal solution set is defined as the action vector of the agent;

[0253] Through the action vector of the intelligent agent, the weights and biases of the multi-objective prediction model are iteratively updated to obtain the dynamic optimal weights of each model in the multi-objective prediction model.

[0254] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0255] The weights and biases of the multi-target prediction model are iteratively updated through the agent’s action vector, including:

[0256] Through the action vector of the intelligent agent, based on the principle of minimum mean absolute percentage error, the static weight coefficient is selected from the Pareto optimal solution set;

[0257] Obtain reference error based on static weight coefficient and actual short-term wind power;

[0258] Analyze the correlation between the minimum value of mean square error, the minimum value of standard deviation and reference error;

[0259] According to the correlation, obtain the reward function;

[0260] Through the reward function, the weights and biases of the multi-objective prediction model are iteratively updated.

[0261] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0262] Acquire time series data of historical wind power, wherein the time series data includes wind power values ​​at different time points;

[0263] The time series data of historical wind power are normalized, and an extreme learning machine model, a long short-term memory neural network model, and a temporal convolutional network model are trained respectively according to the time series data of historical wind power after normalization.

[0264] Train multi-target prediction models based on extreme learning machine models, long short-term memory neural network models, and temporal convolutional network models;

[0265] The multi-objective Harris Eagle algorithm is used to obtain the Pareto optimal solution set of the combined weight vector of the multi-objective prediction model;

[0266] The deep deterministic policy gradient method is used to obtain the dynamic optimal weights of each model in the multi-objective prediction model according to the Pareto optimal solution set;

[0267] According to each model in the multi-objective prediction model and the corresponding dynamic optimal weight, the prediction results of short-term wind power are obtained and the prediction results are displayed visually.

[0268] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0269] The time series data of historical wind power are normalized, and according to the normalized time series data of historical wind power, an extreme learning machine model, a long short-term memory neural network model and a time convolutional network model are trained respectively, including:

[0270] The sequence decomposition method CEEMDAN is used to decompose the time series data of historical wind power to obtain multiple first subsequence data and residual sequences;

[0271] The extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model are trained simultaneously through multiple first subsequence data.

[0272] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0273] According to each model in the multi-objective prediction model and the corresponding dynamic optimal weight, the wind power prediction results are obtained, including:

[0274] Obtain time series data of target wind power;

[0275] The sequence decomposition method CEEMDAN is used to decompose the time series data of the target wind power to obtain multiple second subsequence data;

[0276] By using each model in the multi-objective prediction model, traversing and predicting the plurality of second subsequence data at the same time, a corresponding first prediction result, a second prediction result and a third prediction result are obtained;

[0277] According to the dynamic optimal weights corresponding to each model, the first prediction result, the second prediction result and the third prediction result are weightedly calculated to obtain the prediction result of short-term wind power.

[0278] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0279] According to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model, a multi-objective prediction model is trained, including:

[0280] Obtain time series data of test wind power;

[0281] The sequence decomposition method CEEMDAN is used to decompose the time series data of the test wind power to obtain multiple third subsequence data;

[0282] According to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model, a plurality of third subsequence data are traversed and tested at the same time to obtain corresponding first test results, second test results and third test results;

[0283] Obtaining mean square errors and standard deviations between the actual short-term wind power and the first test result, the second test result, and the third test result, respectively;

[0284] The minimum mean square error and the minimum standard deviation are used as optimization objectives to train the multi-objective prediction model.

[0285] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0286] The deep deterministic policy gradient method is used to obtain the dynamic optimal weights of each model in the multi-objective prediction model according to the Pareto optimal solution set, including:

[0287] Using a deep deterministic policy gradient method, the Pareto optimal solution set is defined as the action vector of the agent;

[0288] Through the action vector of the intelligent agent, the weights and biases of the multi-objective prediction model are iteratively updated to obtain the dynamic optimal weights of each model in the multi-objective prediction model.

[0289] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0290] The weights and biases of the multi-target prediction model are iteratively updated through the agent’s action vector, including:

[0291] Through the action vector of the intelligent agent, based on the principle of minimum mean absolute percentage error, the static weight coefficient is selected from the Pareto optimal solution set;

[0292] Obtain reference error based on static weight coefficient and actual short-term wind power;

[0293] Analyze the correlation between the minimum value of mean square error, the minimum value of standard deviation and reference error;

[0294] According to the correlation, obtain the reward function;

[0295] Through the reward function, the weights and biases of the multi-objective prediction model are iteratively updated.

[0296] 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, stored data, displayed data, 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 relevant data must comply with relevant regulations.

[0297] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium 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), magnetoresistive 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. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0298] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, 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.

[0299] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A wind power prediction method, characterized in that: The method comprises: Acquire time series data of historical wind power, wherein the time series data includes wind power values ​​at different time points; Normalizing the time series data of the historical wind power, and training an extreme learning machine model, a long short-term memory neural network model, and a time convolutional network model respectively according to the time series data of the historical wind power after the normalization process; Obtain time series data of test wind power; Using the sequence decomposition method CEEMDAN, the time series data of the test wind power is decomposed to obtain a plurality of third subsequence data; According to the extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model, the plurality of third subsequence data are traversed and tested at the same time to obtain corresponding first test results, second test results and third test results; Obtaining mean square errors and standard deviations between the actual wind power and the first test result, the second test result, and the third test result, respectively; The minimum mean square error and the minimum standard deviation are used as optimization objectives to train the multi-objective prediction model; Using a multi-objective Harris Hawk algorithm, obtaining a Pareto optimal solution set of a combined weight vector of the multi-objective prediction model; Using a deep deterministic policy gradient method, the Pareto optimal solution set is defined as the action vector of the agent; Iteratively updating the weights and biases of the multi-objective prediction model through the action vector of the agent to obtain the dynamic optimal weights of each model in the multi-objective prediction model; According to each model in the multi-objective prediction model and the corresponding dynamic optimal weight, a prediction result of wind power is obtained, and the prediction result is visualized and displayed.

2. The method according to claim 1, characterized in that The normalizing process is performed on the time series data of the historical wind power, and according to the normalized time series data of the historical wind power, respectively training an extreme learning machine model, a long short-term memory neural network model and a time convolutional network model, including: Using the sequence decomposition method CEEMDAN, the time series data of the historical wind power is decomposed to obtain a plurality of first subsequence data and a residual sequence; The extreme learning machine model, the long short-term memory neural network model and the temporal convolutional network model are trained simultaneously through the multiple first subsequence data.

3. The method according to claim 1, characterized in that The step of obtaining a wind power prediction result according to each model in the multi-objective prediction model and the corresponding dynamic optimal weight includes: Obtain time series data of target wind power; Using a sequence decomposition method CEEMDAN, the time series data of the target wind power is decomposed to obtain a plurality of second subsequence data; By using each model in the multi-objective prediction model, the plurality of second subsequence data are traversed and predicted at the same time to obtain corresponding first prediction results, second prediction results and third prediction results; According to the dynamic optimal weights corresponding to each model, weighted calculation is performed on the first prediction result, the second prediction result and the third prediction result to obtain a prediction result of the target wind power.

4. The method according to claim 1, wherein the step of iteratively updating the weights and biases of the multi-objective prediction model through the action vector of the agent comprises: Selecting a static weight coefficient from the Pareto optimal solution set based on the minimum mean absolute percentage error principle through the action vector of the agent; Obtaining a reference error according to the static weight coefficient and the actual wind power; Analyzing the correlation between the minimum value of the mean square error, the minimum value of the standard deviation, and the reference error; According to the correlation, obtaining a reward function; The weights and biases of the multi-objective prediction model are iteratively updated through the reward function.

5. A wind power prediction device, characterized in that: The device comprises: An acquisition module, used to acquire time series data of historical wind power, wherein the time series data includes wind power values ​​at different time points; A model building module, used for normalizing the time series data of the historical wind power, and training an extreme learning machine model, a long short-term memory neural network model and a time convolutional network model respectively according to the time series data of the historical wind power after the normalization process; The model building module is also used to obtain time series data of the test wind power; use the sequence decomposition method CEEMDAN to decompose the time series data of the test wind power to obtain multiple third subsequence data; according to the extreme learning machine model, the long short-term memory neural network model and the time convolution network model, traverse the multiple third subsequence data at the same time to obtain corresponding first test results, second test results and third test results; obtain the mean square error and standard deviation between the actual wind power and the first test result, the second test result and the third test result respectively; take the minimum value of the mean square error and the minimum value of the standard deviation as optimization goals, and train the multi-objective prediction model; A processing module, used for obtaining a Pareto optimal solution set of a combined weight vector of the multi-objective prediction model by using a multi-objective Harris Hawk algorithm; The processing module is further used to define the Pareto optimal solution set as the action vector of the intelligent agent by adopting the deep deterministic policy gradient method; iteratively update the weights and biases of the multi-objective prediction model through the action vector of the intelligent agent to obtain the dynamic optimal weights of each model in the multi-objective prediction model; The processing module is further used to obtain the prediction result of wind power according to each model in the multi-objective prediction model and the corresponding dynamic optimal weight; The display module is used to visually display the prediction results.

6. The device according to claim 5, characterized in that The processing module is also used to decompose the time series data of the historical wind power by using the sequence decomposition method CEEMDAN to obtain multiple first subsequence data and residual sequences; and to simultaneously train an extreme learning machine model, a long short-term memory neural network model and a time convolutional network model through the multiple first subsequence data.

7. The device according to claim 5, characterized in that The acquisition module is also used to acquire the time series data of the target wind power; the time series data of the target wind power is decomposed by using the sequence decomposition method CEEMDAN to obtain a plurality of second subsequence data; each model in the multi-objective prediction model simultaneously performs traversal prediction on the plurality of second subsequence data to obtain corresponding first prediction results, second prediction results and third prediction results; The processing module is also used to perform weighted calculation on the first prediction result, the second prediction result and the third prediction result according to the corresponding dynamic optimal weights of each model to obtain the prediction result of the target wind power.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

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 steps of the method according to any one of claims 1 to 4 are implemented.

10. 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 4 are implemented.

Images