New energy power generation prediction error correction method based on artificial intelligence

Abstract

The invention relates to a new energy power generation prediction error correction method based on artificial intelligence. The method comprises the following steps: data collection and processing; predicting by using a plurality of machine learning models; performing model integration through variational Bayesian inference to generate a final integrated prediction result; carrying out dynamic error correction by adopting stochastic differential equation modeling and extended Kalman filtering; modeling relevance between power station points by using a graph neural network, and taking the relevance as auxiliary information for error correction; self-adaptive parameter adjustment based on a feedback mechanism; global optimization and final prediction result output; according to the method, by integrating various machine learning models, implementing dynamic error correction, modeling site relevance by using the graph neural network, adaptively adjusting model parameters and executing global optimization, the accuracy, stability and adaptability of new energy power generation prediction are remarkably improved; the method has the advantages that multiple models are integrated, the dynamic error correction capability is achieved, and the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of renewable energy power generation, and in particular relates to a method for correcting prediction errors of renewable energy power generation based on artificial intelligence. Background Art

[0002] With the increasing global demand for renewable energy, new energy power generation (such as wind power and photovoltaic power generation) has occupied an increasingly important position in the energy structure. However, since the new energy power generation process is affected by many complex factors such as meteorological conditions and geographical location, how to accurately predict power generation has become a key challenge in power grid dispatching and energy management. Traditional power generation forecasting methods mainly rely on a single statistical model or a physics-based simulation method. These methods usually have limitations such as a single model, insufficient ability to handle nonlinear relationships, and inability to fully utilize multi-source data, resulting in prediction accuracy and stability that are difficult to meet actual needs. Although existing machine learning and data-driven forecasting methods have improved prediction accuracy to a certain extent, they still face many problems, such as For example, a single machine learning model is unable to cope with the complex dynamic changes and uncertainties in the renewable energy power generation process; in terms of data fusion and feature extraction, traditional methods find it difficult to fully tap the potential value of multi-source data, resulting in insufficient robustness of the prediction results; in addition, in terms of dynamic error correction, most existing methods lack a real-time update mechanism, making it difficult to effectively cope with rapidly changing environmental conditions and random fluctuations in power generation systems; these defects have seriously restricted the application effect of renewable energy power generation prediction technology, resulting in greater uncertainty in the scheduling and management of power systems; therefore, it is very necessary to provide an artificial intelligence-based renewable energy power generation prediction error correction method that integrates multiple models, has dynamic error correction capabilities, and improves prediction accuracy, stability, and adaptability. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide an artificial intelligence-based new energy power generation prediction error correction method that integrates multiple models, has dynamic error correction capabilities, and improves prediction accuracy, stability and adaptability.

[0004] The object of the present invention is achieved by: a method for correcting prediction errors of renewable energy power generation based on artificial intelligence, the method comprising the following steps:

[0005] Step 1: Collect and process historical power generation data, meteorological data, and environmental variables from multiple renewable energy power generation sites, and perform multi-source data fusion and feature extraction;

[0006] Step 2: Use multiple machine learning models to predict renewable energy power generation and generate multiple independent prediction results;

[0007] Step 3: Estimate the weight distribution of each model through the variational Bayesian inference method, perform model integration, and generate the final integrated prediction result;

[0008] Step 4: Use stochastic differential equations to model the dynamic behavior of renewable energy power generation, and use extended Kalman filtering to perform dynamic error correction on the prediction results;

[0009] Step 5: Use graph neural networks to model the correlation between power generation sites and use it as auxiliary information for error correction;

[0010] Step 6: Dynamically optimize model parameters through variational reinforcement learning based on feedback mechanism and adaptive parameter adjustment;

[0011] Step 7: Model the global optimization problem of the system as a mixed integer nonlinear programming, solve it through the global optimization algorithm, and output the final revised new energy power generation prediction result.

[0012] The multi-source data fusion and feature extraction in step 1 specifically include the following steps:

[0013] Step 1.1: Collect data from multiple renewable energy power generation sites as the original data set;

[0014] Step 1.2: Perform principal component analysis on the cleaned data to reduce the dimension and extract the main features that have the greatest impact on the prediction;

[0015] Step 1.3: Based on principal component analysis, random matrix theory is used to analyze the eigenvalue spectrum of high-dimensional data;

[0016] Step 1.4: Fusion the data features processed by principal component analysis and random matrix theory, and construct a new feature set based on the actual application requirements of renewable energy power generation;

[0017] Step 1.5: Standardize and normalize the final feature set to eliminate the dimensional differences between different features;

[0018] Step 1.6: Use feature selection algorithm to select the features that have the greatest impact on prediction and use them as input data for subsequent model training and prediction.

[0019] The machine learning models in step 2 include: long short-term memory network, convolutional neural network, autoregressive integrated moving average model and random forest.

[0020] The step 3 specifically includes the following steps:

[0021] Step 3.1: Set the initial weight distribution for each independent prediction model, assuming that the weight of each model follows a certain probability distribution, and initialize the prior distribution p(w) of the model weight;

[0022] Step 3.2: Use the variational Bayesian inference method to approximate the posterior distribution p(w|D), where D represents the training data; the posterior distribution of the weight is determined by the likelihood function p(D|w) and the prior distribution p(w), select an appropriate variational distribution q(w), and approximate the posterior distribution by minimizing the Kullback-Leibler divergence;

[0023] Step 3.3: Optimize the parameters of the variational distribution q(w) by maximizing the evidence lower bound. The expression for maximizing the evidence lower bound is: ELBO = Ε q(w) [logp(D|w)]-D KL (q(w)||p(w)), where the first term is the logarithmic expectation of the likelihood function p(D|w) of the prediction result; the second term is the Kullback-Leibler divergence between the variational distribution q(w) and the prior distribution p(w);

[0024] Step 3.4: During the optimization process, the parameters of the variational distribution are continuously updated through the optimization algorithm, and the weight distribution of each model is gradually adjusted until the maximum evidence lower bound reaches a stable convergence state, and the optimal weight distribution of each model is obtained;

[0025] Step 3.5: Apply the optimized variational distribution q(w) to the weighted combination of independent prediction results to generate the final integrated prediction result Among them, y m is the prediction result of the mth model; q(w m ) is the weight of the model;

[0026] Step 3.6: Output the final integrated prediction result and use it as input data for the subsequent error correction step.

[0027] The step 4 specifically comprises the following steps:

[0028] Step 4.1: Model the dynamic behavior of the renewable energy generation process as a stochastic differential equation describing the system state x t The evolution over time t, the model form is: dx t =f(x t ,t)dt+g(x t ,t)dW t , where f(x t ,t) is the drift term, which describes the deterministic change of the system; g(x t ,t) is a diffusion term, reflecting the uncertainty of the system; W t is the Wiener process, which indicates the random fluctuation of the system; d is the differential symbol;

[0029] Step 4.2: Based on the current state and the drift term f(x t ,t), use stochastic differential equations to predict the system state and get the state estimate for the next time step Among them, Δt is the time step;

[0030] Step 4.3: According to the diffusion term g(x t ,t) and the Jacobian matrix of the system, calculate the covariance matrix p of the predicted state t|t-1 : Among them, F t-1 is the drift term f(x t ,t)’s linearized Jacobian matrix; G t-1 is the coefficient matrix of the diffusion term; Q t-1 is the process noise covariance matrix;

[0031] Step 4.4: Combine the observed data y of the system t With the observation function h(x t ), using the extended Kalman filter to update the system state estimate and covariance matrix;

[0032] Step 4.5: State estimation after extended Kalman filtering The preliminary forecast results are dynamically corrected for errors, and the revised forecast value of renewable energy power generation is generated and used as the final forecast output.

[0033] The step 5 specifically comprises the following steps:

[0034] Step 5.1: Taking power generation sites as nodes and the physical distances and meteorological condition similarities between sites as edges, an undirected graph G = (V, E) is formed, where V is the site set and E is the edge set;

[0035] Step 5.2: Initialize feature vector for each node Including the historical power generation and key information of meteorological conditions of the site, each edge e∈E is given a weight, which is calculated based on the physical distance between sites or data similarity;

[0036] Step 5.3: Build a graph neural network to process the correlation between sites. The message passing process of the graph neural network is described by the following formula: in, is the feature vector of node v at the kth layer; N(v) is the set of neighbor nodes of node v; d u d v are the degrees of node u and node v respectively; W (k) is a trainable weight matrix; is the feature vector of node u at the k-1 layer; b (k) is the bias term; σ is the activation function;

[0037] Step 5.4: Perform message passing and aggregation through the multi-layer graph neural network model, update the feature representation of each node layer by layer, and the feature representation of each layer k They are updated based on the weighted sum of their neighbor node features, gradually integrating the structural information of the entire graph, and finally outputting the feature vector of the node Represents the comprehensive information of the site in the graph structure;

[0038] Step 5.5: Output the site feature vector of the graph neural network Combined with the output of the prediction model as auxiliary information for error correction;

[0039] Step 5.6: The error correction result based on the combination of graph neural network and extended Kalman filter is used as the final power generation prediction output.

[0040] The step 6 specifically comprises the following steps:

[0041] Step 6.1: Monitor the environmental status of the renewable energy power generation system in real time, collect power generation, meteorological conditions, and forecast error feedback information, and construct a feedback data set;

[0042] Step 6.2: Construct a variational reinforcement learning framework, where the state s of the system t It is composed of the current environmental conditions and model parameters. Action a t is the adjustment strategy for model parameters, and the reward r t is the prediction accuracy or error reduction of the model in its current state;

[0043] Step 6.3: Design the policy network π θ (a t |s t ), and optimizes the parameter distribution q(θ) of the policy network through variational inference. The policy network outputs the probability distribution of parameter adjustment. Variational inference approximates the posterior distribution p(θ|D) by minimizing the Kullback-Leibler divergence. Its optimization goal is: Among them, γ is the discount factor, which controls the influence weight of future rewards on the current strategy; r t is the immediate reward of the current time step; D KL (q(θ)||p(θ)) represents the Kullback-Leibler divergence, which measures the difference between the variational distribution q(θ) and the prior distribution p(θ);

[0044] Step 6.4: Based on the current environment status s t And the output of the policy network, select the optimal action a t , that is, the optimal parameter adjustment strategy;

[0045] Step 6.5: Execute the selected action a t , update and adjust model parameters in real time;

[0046] Step 6.6: Accumulate multiple rounds of feedback information and continuously optimize the policy network through multiple policy iterations;

[0047] Step 6.7: Apply the model parameters that have undergone multiple rounds of adjustment and optimization to the actual power generation forecast to generate the final forecast results.

[0048] The step 7 specifically includes the following steps:

[0049] Step 7.1: The objective function and constraints of the renewable energy generation prediction system are expressed as a mixed integer nonlinear programming problem. Let the continuous variable be x and the integer variable be z. The optimization goal is to minimize the total prediction error while satisfying the system resource and computing constraints. The specific expression is: subject to g(x,z)≤0,x∈R n ,z∈Z m , where f(x,z) is the objective function, which represents the prediction error or other system performance indicators; g(x,z) is the constraint condition, including computing resources, time or physical constraints;

[0050] Step 7.2: Generate an initial solution through a heuristic algorithm and decompose the mixed integer nonlinear programming problem into several sub-problems, each sub-problem corresponds to a nonlinear programming problem or an integer programming problem;

[0051] Step 7.3: Apply the branch and bound method to solve the problem. Use the method of partitioning the variable space to divide the mixed integer nonlinear programming problem into multiple subproblems, calculate the lower bound of each subproblem, and use the pruning strategy to exclude subregions that are unlikely to contain the optimal solution, thereby narrowing the solution space;

[0052] Step 7.4: Combine the global optimization algorithm to solve the sub-problems generated by the branch and bound method, and find the optimal solution in the global scope through global search and information sharing mechanism;

[0053] Step 7.5: Continuously monitor the change of the solution and determine whether the solution has converged by using the set convergence criteria. If the convergence condition is not met, continue to iterate the solution until the convergence criteria are met.

[0054] Step 7.6: Apply the final global optimal solution to the renewable energy power generation prediction model to generate the final prediction result after global optimization, and use the result as the output of the system.

[0055] An artificial intelligence-based new energy power generation prediction error correction device includes: a data processing module for collecting, cleaning and processing historical power generation data, meteorological data and environmental variables from multiple new energy power generation sites, and performing feature extraction and data fusion operations;

[0056] Model prediction module, which is used to select and train multiple machine learning models, generate multiple independent renewable energy power generation prediction results, and evaluate the performance of each model;

[0057] The model integration module is used to estimate the weight distribution of each model through the variational Bayesian inference method, integrate the prediction results of multiple models, and generate the final integrated prediction results;

[0058] Dynamic error correction module, which is used to model the dynamic behavior of the system based on stochastic differential equations and perform dynamic error correction on the prediction results through extended Kalman filtering;

[0059] The correlation modeling module is used to model the correlation between power generation sites using graph neural networks, and input it into the dynamic error correction module as auxiliary information to further optimize the prediction results;

[0060] Parameter optimization module, which is used to dynamically adjust and optimize model parameters based on feedback mechanism and variational reinforcement learning method to ensure that the model can adapt to environmental changes and maintain optimal performance;

[0061] The global optimization module is used to model the global optimization problem of the system as a mixed integer nonlinear programming, solve it using a global optimization algorithm, and output the final revised new energy power generation forecast results.

[0062] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the above method is implemented.

[0063] A storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned artificial intelligence-based new energy power generation prediction error correction method.

[0064] Beneficial effects of the invention: The invention is a new energy power generation prediction error correction method based on artificial intelligence. By integrating multiple machine learning models, dynamic error correction and global optimization, the accuracy, stability and adaptability of new energy power generation prediction are significantly improved. In use, the invention has the following advantages: ① The invention uses multiple machine learning models and integrates them to combine the advantages of different models and reduce the bias and error that may be caused by a single model; ② The invention adopts stochastic differential equations to model the dynamic characteristics of the new energy power generation process, and uses extended Kalman filtering to correct the prediction results in real time, which can effectively cope with the random fluctuations and environmental changes encountered by the system during actual operation; ③ The invention uses graph neural networks to model the correlation between new energy power generation sites, which can capture and utilize the complex interactive relationship between sites and provide auxiliary information for error correction; ④ The invention dynamically adjusts the model parameters through variational reinforcement learning, thereby realizing the adaptive adjustment of the model in the face of environmental changes; ⑤ The invention models the global optimization problem of the system as a mixed integer nonlinear programming, and solves it through an efficient global optimization algorithm, which can find the optimal solution based on considering multiple constraints; The invention has the advantages of integrating multiple models, having dynamic error correction capabilities, and improving prediction accuracy, stability and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a flow chart of the present invention.

[0066] Figure 2 It is a schematic diagram of the device structure of the present invention.

[0067] Figure 3 It is a schematic diagram of the computer device structure of the present invention.

[0068] In the figure: 100, data processing module 200, model prediction module 300, model integration module 400, dynamic error correction module 500, correlation modeling module 600, parameter optimization module 700, global optimization module 40, computer device 41, processor 42, memory 43, storage medium DETAILED DESCRIPTION

[0069] The present invention is a new prediction method that can integrate multiple models, has dynamic error correction capabilities, and can effectively utilize multi-source data and site correlation to improve the accuracy, stability and adaptability of renewable energy power generation prediction and meet the needs of modern power systems for high-precision prediction.

[0070] The present invention will be further described below in conjunction with the accompanying drawings.

[0071] Example 1

[0072] like Figure 1 As shown, a method for correcting prediction errors of renewable energy power generation based on artificial intelligence comprises the following steps:

[0073] S1: Data collection and processing: mainly involves multi-source data fusion and feature extraction of historical power generation data, meteorological data and environmental variables from multiple renewable energy power generation sites;

[0074] In this embodiment, this step is the basis of the entire algorithm process. It provides high-quality input data for subsequent machine learning model training and prediction. Through effective data processing and feature extraction, the prediction accuracy and robustness of the model can be significantly improved.

[0075] First, the data collected from multiple renewable energy power generation sites are used as the original data set. These data include historical power generation, temperature, humidity, wind speed, light intensity and other meteorological conditions, as well as other related environmental variables. After the data collection is completed, data cleaning is performed to eliminate or correct outliers and missing values ​​in the original data. The data cleaning method can use interpolation, mean filling or model-based missing value estimation methods to ensure the integrity and consistency of the data.

[0076] After data cleaning, principal component analysis (PCA) is performed to reduce the dimension of the cleaned data through PCA, and the main features that have the greatest impact on the prediction of renewable energy power generation are extracted; the specific implementation of PCA includes calculating the covariance matrix of the data set, then performing eigenvalue decomposition on the covariance matrix, and selecting the first several principal components, which explain most of the variance in the data, thereby achieving a dimensionality reduction effect; the mathematical expression of PCA is: Z = XW, where X is the standardized data matrix; W is the eigenvector matrix; and Z is the eigenmatrix after dimensionality reduction;

[0077] Based on the principal component analysis, this embodiment further uses random matrix theory to analyze the eigenvalue spectrum of high-dimensional data. Theorem can distinguish useful signals and noise in data, thereby optimizing the feature selection process; specifically, random matrix theory analysis is described by the following formula: Among them, λ is the eigenvalue; is the ratio of the number of samples to the number of features; σ 2 is the noise variance; max , min are the upper and lower bounds of the spectrum respectively;

[0078] The data features processed by principal component analysis and random matrix theory are fused, and a new feature set is constructed according to the actual application requirements of renewable energy power generation; in this embodiment, the fused feature set includes not only the principal components after dimensionality reduction processing, but also the signal features filtered by random matrix theory; these feature sets can effectively reduce redundant information and improve the interpretability and effectiveness of features;

[0079] Next, the final feature set is standardized and normalized to eliminate the dimensional differences between different features. In this embodiment, the standardization process is to adjust the feature data to a standard normal distribution with a mean of zero and a variance of one; the normalization process is to adjust the feature data to a specified range (such as [0,1]) to ensure the equivalence of different features in model training.

[0080] Finally, in this embodiment, the features that have the greatest impact on prediction are further screened through feature selection algorithms. These feature selection algorithms may include recursive feature elimination (RFE), selection methods based on information gain, etc., to ensure that the feature set that is ultimately retained has the strongest influence and relevance to subsequent model training and prediction.

[0081] S2: Prediction using multiple machine learning models: After completing the multi-source data fusion and feature extraction in step S1, the system enters the prediction stage, i.e., step S2; in this step, the system predicts the renewable energy power generation by selecting and training multiple machine learning models, and generates multiple independent prediction results; different models, based on their unique algorithm structure and assumptions, can estimate the power generation from different angles, providing diverse input data for subsequent model integration;

[0082] In this embodiment, firstly, according to the characteristics of the renewable energy power generation data, multiple machine learning models suitable for processing time series data and nonlinear relationships are selected; specifically, the selected models include but are not limited to long short-term memory network (LSTM), convolutional neural network (CNN), autoregressive integrated moving average model (ARIMA) and random forest (RandomForest);

[0083] ① Model training: In this embodiment, the above model is trained using the data feature set extracted and processed in step S1; during the training process, a cross-validation method is used to optimize the hyperparameter settings of each model; cross-validation divides the data set into several folds, and uses each fold as a validation set in turn, and the rest as a training set, thereby reducing the risk of overfitting of the model under a specific data distribution; the optimized hyperparameters include the hidden layer size of LSTM, the convolution kernel size of CNN, the order of ARIMA, the number of trees of random forest, etc.;

[0084] ② Independent prediction: After each model is trained, the data of renewable energy power generation to be predicted is input into each model to generate independent prediction results. Specifically, the LSTM model generates a prediction sequence with long-term dependency characteristics through step-by-step prediction of time steps. The CNN model generates prediction results that can capture local features through multi-layer convolution and pooling operations. The ARIMA model generates linear prediction results based on the trend and seasonal characteristics of historical data. The random forest model generates comprehensive prediction results by integrating the voting results of multiple decision trees.

[0085] ③ Model performance evaluation: In this embodiment, the prediction results of each model are evaluated, including the mean square error (MSE), mean absolute error (MAE) and R 2 And other common error indicators;

[0086] Through the above steps, the system can generate multiple independent new energy power generation forecast results. These results are based on different algorithms and assumptions and are diverse, providing a rich data input basis for subsequent model integration.

[0087] S3: Model integration through variational Bayesian inference: In step S2, the system generates multiple independent renewable energy power generation forecast results through multiple machine learning models; in order to further improve the accuracy and robustness of the forecast, step S3 estimates the weight distribution of each model through the variational Bayesian inference method, and integrates the models to generate the final integrated forecast result; this step effectively integrates the advantages of each model by dynamically adjusting the weight distribution of the model, ensuring that the comprehensive performance of the final forecast result is more stable;

[0088] In this embodiment, an initial weight distribution is first set for each independent prediction model, assuming that the weight of each model obeys a certain probability distribution, such as normal distribution or uniform distribution, and the prior distribution p(w) of the weight of each model is initialized. The setting of the initial weight distribution lays the foundation for the subsequent optimization process and ensures the rationality of the weight and the effectiveness of the model integration.

[0089] Next, the variational Bayesian inference method is used to approximate the posterior distribution p(wD), where D represents the training data; the posterior distribution of the weight is jointly determined by the likelihood function p(D|w) and the prior distribution p(w). To achieve this, this embodiment selects an appropriate variational distribution q(w) to approximate the posterior distribution, and optimizes the parameters of the variational distribution by minimizing Kullback-Leibler (KL). The KL divergence is defined as: KL divergence measures the difference between the variational distribution q(w) and the true posterior distribution p(wD). By minimizing this difference, variational Bayesian inference can approach the true posterior distribution.

[0090] In the optimization process, this embodiment optimizes the parameters of the variational distribution q(w) by maximizing the evidence lower bound (ELBO), and the expression for maximizing the evidence lower bound is: ELBO = Ε q(w) [logp(Dw)]-D KL (q(w)p(w)), where the first term is the logarithmic expectation of the likelihood function p(Dw) of the prediction result, which reflects the degree of fit of the model under given data; the second term is the Kullback-Leibler divergence between the variational distribution q(w) and the prior distribution p(w), which is used to adjust the model complexity;

[0091] During the ELBO optimization process, the system continuously updates the parameters of the variational distribution q(w) through gradient descent or other optimization algorithms, gradually adjusting the weight distribution of each model until ELBO reaches a stable convergence state, and finally obtains the optimal weight distribution of each model;

[0092] Subsequently, the optimized variational distribution q(w) is applied to the weighted combination of the independent prediction results to generate the final integrated prediction result The calculation formula of the integrated prediction result is: Among them, y m is the prediction result of the mth model; q(w m ) is the weight of the model; this ensemble method can combine the prediction advantages of each model and generate a more accurate and robust prediction result globally by weighted averaging;

[0093] Finally, this embodiment will integrate the final prediction results Output and use it as input data for the subsequent error correction step; this process ensures that the integrated prediction results can effectively support the subsequent dynamic error correction and correlation modeling steps, thereby further improving the prediction performance of the entire system;

[0094] The system effectively integrates the prediction results of multiple independent models through the variational Bayesian inference method, and ultimately generates integrated prediction results with higher accuracy and stability, providing a reliable foundation for subsequent error correction and optimization.

[0095] S4: Dynamic error correction using stochastic differential equation modeling and extended Kalman filter: After completing the model integration and generating preliminary integrated prediction results in step S3, the system further enters step S4, using stochastic differential equations to model the dynamic behavior of the renewable energy power generation process, and using extended Kalman filter to perform dynamic error correction on the prediction results; the goal of this step is to further improve the accuracy and reliability of the prediction results by combining the dynamic characteristics and uncertainty of the system;

[0096] In this embodiment, the dynamic behavior of the renewable energy power generation process is firstly modeled as a stochastic differential equation (SDE) to describe the system state x t As time t evolves, the mathematical form of the stochastic differential equation is: dx t =f(x t ,t)dt+g(x t ,t)dW t , where f(x t ,t) is the drift term, which describes the deterministic change of the system; g(x t ,t) is a diffusion term, reflecting the uncertainty of the system; W t is the Wiener process, which indicates the random fluctuation of the system; d is the differential symbol;

[0097] Next, based on the current state and the drift term f(x t ,t), use stochastic differential equations to predict the system state and get the state estimate for the next time step The formula for state prediction is: Among them, Δt is the time step; through this prediction, the system can preliminarily estimate the system state at the future time, which serves as the basis for subsequent error correction;

[0098] In order to further quantify the uncertainty of the prediction, this embodiment uses the diffusion term g(x t ,t) and the Jacobian matrix of the system, calculate the covariance matrix p of the predicted state tt-1 , the calculation formula of the covariance matrix is: Among them, F t-1 is the drift term f(x t ,t)’s linearized Jacobian matrix; G t-1 is the coefficient matrix of the diffusion term; Q t-1 is the process noise covariance matrix; by calculating the covariance matrix, the system can quantify the uncertainty of the predicted state, thus providing a basis for subsequent error correction;

[0099] Combined with the system's observed data y t With the observation function h(x t ), this embodiment uses an extended Kalman filter (EKF) to update the system state estimation and covariance matrix, and the updating steps include the following:

[0100] ①Calculate the Kalman gain K t : Among them, H t is the Jacobian matrix of the observation function; R t is the observation noise covariance matrix;

[0101] ②Update state estimation: This step combines the observed data with the predicted state and corrects the initial state prediction; t is the actual observed value at time t; Based on the prior state estimate The predicted observed value of

[0102] ③ Update the covariance matrix P tt :P tt =(IK t H t ) tt-1 , the updated covariance matrix reflects the uncertainty of the state estimate after correction of the observed data.

[0103] State estimation after extended Kalman filtering This embodiment performs dynamic error correction on the preliminary prediction results to generate a revised forecast value of renewable energy power generation. The revised forecast value further improves the accuracy of the prediction based on the dynamic characteristics of the system and the observed data. Finally, the revised forecast result is used as the final output of the system to provide a reliable basis for subsequent decision-making and application.

[0104] The system can effectively capture and correct dynamic errors that occur in the process of renewable energy power generation, making the prediction results more accurate and robust in an uncertain environment.

[0105] S5: Modeling the correlation between power generation sites using graph neural networks: In the previous steps, the system dynamically corrected the prediction results through stochastic differential equations and extended Kalman filters. In order to further improve the prediction accuracy, step S5 introduces a graph neural network (GNN) to model the correlation between different power generation sites and use it as auxiliary information for error correction. Through this step, the system can effectively capture the complex interactive relationship between power generation sites and use this information to further optimize the prediction results.

[0106] In this embodiment, firstly, the power generation sites are taken as nodes, and the physical distances and meteorological condition similarities between the sites are taken as edges to form an undirected graph G = (V, E), where V is a set of sites; E is a set of edges; the weights of the edges are calculated based on the physical distances or data similarities between the sites; this graph structure can effectively represent the geographical and meteorological correlations between the power generation sites;

[0107] Initialize the feature vector for each node (i.e. power generation site) The feature vector includes the historical power generation and key information of meteorological conditions of the site; each edge e∈E is given a weight, which is calculated based on the physical distance or data similarity between the sites, so as to accurately reflect the strength of association between the sites;

[0108] Next, a graph neural network (GNN) is constructed to process the correlation between sites. The message passing process of the graph neural network (GNN) is described by the following formula: in, is the feature vector of node v at the kth layer; N(v) is the set of neighbor nodes of node v; d u ,d v are the degrees of node u and node v respectively; W (k) is a trainable weight matrix; is the feature vector of node u at the k-1 layer; b (k) is the bias term; σ is the activation function; through this formula, GNN can aggregate the information of neighbor nodes in each layer and update the feature representation of nodes layer by layer;

[0109] Through the multi-layer graph neural network model, message passing and aggregation are performed to update the feature representation of each node layer by layer; the feature representation of each layer k They are updated based on the weighted sum of their neighbor node features, gradually integrating the structural information of the entire graph; the feature vector of the final output node Represents the comprehensive information of the site in the graph structure, which includes the characteristics of the site itself and the comprehensive characteristics of its neighboring nodes, and can effectively capture the mutual influence between power generation sites;

[0110] The site feature vector output by the graph neural network Combined with the output of the previous prediction model as auxiliary information for error correction; in this embodiment, these auxiliary information are combined with the extended Kalman filter results to further optimize the error correction process, thereby improving the final prediction accuracy;

[0111] Ultimately, this embodiment uses the error correction result based on the combination of graph neural network and extended Kalman filter as the final power generation prediction output; by integrating the correlation information between sites, the system can consider various influencing factors more comprehensively, thereby further improving the accuracy and reliability of renewable energy power generation prediction.

[0112] S6: Based on feedback mechanism and adaptive parameter adjustment: In the previous steps, the system has generated relatively accurate prediction results through multi-model integration, dynamic error correction and correlation modeling; in order to further improve the adaptability and prediction accuracy of the model, step S6 introduces a variational reinforcement learning method based on feedback mechanism and adaptive parameter adjustment, which ensures that the system can continue to provide high-precision prediction results in a constantly changing environment by real-time monitoring of system status and dynamic optimization of model parameters;

[0113] In this embodiment, the environmental status of the new energy power generation system is first monitored in real time, and various data related to the power generation process are collected, including power generation, meteorological conditions (such as temperature, humidity, wind speed, light intensity, etc.) and feedback information of the current prediction error. This information will be constructed into a feedback data set to provide a basis for subsequent model parameter adjustment and optimization;

[0114] Next, we construct a variational reinforcement learning framework, in which the state s of the system t It is composed of the current environmental conditions and model parameters. Action a t The reward r is the adjustment strategy for the model parameters. t It is used to evaluate the prediction accuracy or error reduction of the current strategy under specific conditions. The core of this framework is to adjust the model parameters through real-time feedback, so that the system can maintain high prediction performance under different environmental conditions.

[0115] To achieve the above objectives, this embodiment designs a strategy network π θ (a t |s t ), the network optimizes the parameter distribution q(θ) of the policy network through variational inference. The policy network outputs the probability distribution of parameter adjustment and approximates the posterior distribution p(θ|D) by minimizing the Kullback-Leibler (KL) divergence. The expression of the optimization objective function is: Among them, γ is the discount factor, which controls the influence weight of future rewards on the current strategy; r t is the immediate reward of the current time step; D KL (q(θ)||p(θ)) represents the Kullback-Leibler divergence, which measures the difference between the variational distribution q(θ) and the prior distribution p(θ). By optimizing this objective function, the policy network can gradually learn the optimal parameter adjustment strategy, so that the system can maximize the prediction accuracy under different environmental conditions.

[0116] According to the current environment status t and the output of the policy network, this embodiment selects the optimal action a t , which is the optimal parameter adjustment strategy for the current situation; then, the system executes the selected action to update and adjust the key parameters of the model in real time, including model weights, learning rate, regularization coefficient, etc.; the dynamic adjustment of these parameters ensures that the model can quickly adapt and maintain high prediction performance when the environment changes;

[0117] In order to further optimize the policy network, this embodiment accumulates multiple rounds of feedback information and continuously optimizes the policy network in multiple policy iterations. In this way, the policy network can gradually enhance its adaptability to different environmental conditions and learn more effective parameter adjustment strategies;

[0118] Finally, this embodiment applies the model parameters that have undergone multiple rounds of adjustment and optimization to the actual prediction of renewable energy power generation to generate the final prediction results. The optimized model parameters can better adapt to the environmental changes in actual operation and ensure the high accuracy and reliability of the prediction results;

[0119] The system uses feedback mechanisms and adaptive parameter adjustments to achieve dynamic model optimization based on variational reinforcement learning, ensuring that the model can maintain its optimal state in a changing environment, thereby improving the overall performance of renewable energy power generation prediction.

[0120] S7: Global optimization and final prediction result output: In the previous steps, the system has improved the accuracy and adaptability of the renewable energy power generation prediction through various methods; in order to further optimize the prediction performance and ensure the optimal performance of the system under complex constraints, step S7 models the global optimization problem of the system as a mixed integer nonlinear programming (MINLP) problem, and solves it through a global optimization algorithm to output the final revised renewable energy power generation prediction result; this step ensures that the system achieves the optimal solution globally by comprehensively considering various constraints.

[0121] In this embodiment, the objective function and constraints of the new energy generation prediction system are first expressed as a mixed integer nonlinear programming (MINLP) problem. Let the continuous variable be x and the integer variable be z. The optimization goal is to minimize the total prediction error while satisfying the system resource and computing constraints. The specific optimization expression is: subject to g(x,z)≤0,x∈R n ,z∈Z m , where f(x,z) is the objective function, which represents the prediction error or other system performance indicators; g(x,z) is the constraint condition, which usually includes computing resources, time or physical constraints; this modeling method can combine continuous optimization and integer optimization together, and is suitable for dealing with complex global optimization problems;

[0122] After the modeling is completed, this embodiment generates an initial solution through a heuristic algorithm and decomposes the mixed integer nonlinear programming problem into several sub-problems, each of which corresponds to a nonlinear programming problem or an integer programming problem; this decomposition method can effectively reduce the complexity of the problem, so that large-scale problems can be solved in a limited time;

[0123] Next, the branch and bound method is applied to solve the problem. The branch and bound method divides the mixed integer nonlinear programming problem into multiple sub-problems by dividing the variable space. The specific steps include: firstly, calculating the lower bound of each sub-problem, and then using the pruning strategy to exclude the sub-regions that are unlikely to contain the optimal solution, thereby narrowing the solution space. This method can effectively reduce the amount of calculation and improve the solution efficiency.

[0124] In order to further improve the solution effect, this embodiment combines a global optimization algorithm (such as simulated annealing and particle swarm optimization) to solve the sub-problems generated by the branch and bound method; the global optimization algorithm finds the optimal solution in the global scope through a global search and information sharing mechanism to avoid falling into the local optimum; this process can ensure that the system achieves the best prediction performance at the global level;

[0125] During the solution process, this embodiment continuously monitors the change of the solution and determines whether the solution converges by using the set convergence criterion; if the convergence condition is not met, the iterative solution is continued until the convergence criterion is met; the convergence criterion may include the change range of the objective function value, the consumption of computing resources, etc., to ensure the quality of the final solution and the efficiency of the algorithm;

[0126] Finally, this embodiment applies the global optimal solution finally obtained to the new energy power generation prediction model, generates the final prediction result after global optimization, and uses the result as the output of the system; through this optimization process, the system can maintain the best performance under a variety of complex constraints, ensuring that the final prediction result has the highest accuracy and robustness;

[0127] The system uses mixed integer nonlinear programming and global optimization algorithms to successfully solve complex global optimization problems and ensure the optimality and reliability of renewable energy power generation prediction results.

[0128] In general, through the artificial intelligence-based renewable energy power generation prediction error correction algorithm provided by the present invention, the system can make full use of multi-model integration, dynamic error correction, graph neural network modeling and global optimization technology to achieve accurate prediction of renewable energy power generation in complex environments; the algorithm can adapt to environmental changes through dynamic adjustment and optimization, provide more accurate and stable prediction results, and has broad application prospects.

[0129] The present invention is a new energy power generation prediction error correction method based on artificial intelligence. By integrating multiple machine learning models, dynamic error correction and global optimization, the accuracy, stability and adaptability of new energy power generation prediction are significantly improved. In use, the present invention has the following advantages: ① The present invention can combine the advantages of different models and reduce the bias and error that may be caused by a single model by using multiple machine learning models and integrating them; the integrated learning method effectively improves the accuracy and consistency of the prediction through weight adjustment and model optimization; this method makes the final prediction result not only more accurate, but also able to reflect the complementary characteristics of different models; ② The present invention uses stochastic differential equations to model the dynamic characteristics of the new energy power generation process, and uses extended Kalman filtering to correct the prediction results in real time, which can effectively cope with the random fluctuations and environmental changes encountered by the system during actual operation; This dynamic error correction method improves the stability and reliability of the prediction results, ensuring that the output results maintain high quality under different conditions; ③ The present invention uses graph neural networks to model the relationship between new energy power generation sites ④ The present invention dynamically adjusts the model parameters through variational reinforcement learning, thereby realizing the adaptive adjustment of the model in the face of environmental changes; this mechanism continuously optimizes the behavior of the model through real-time feedback and continuous learning, so that the model can continuously optimize its performance in a constantly changing operating environment, greatly improving the flexibility and practicality of the model; ⑤ The present invention models the global optimization problem of the system as a mixed integer nonlinear programming, and solves it through an efficient global optimization algorithm, which can find the optimal solution based on considering multiple constraints; this global optimization method ensures that the system can achieve the optimal prediction performance under complex operating environments and multiple operating restrictions, thereby maximizing the operating efficiency and power generation benefits of the entire system; the present invention has the advantages of integrating multiple models, having dynamic error correction capabilities, and improving prediction accuracy, stability and adaptability.

[0130] Example 2

[0131] like Figure 2 As shown, a new energy power generation prediction error correction device based on artificial intelligence includes:

[0132] The data processing module 100 is used to collect, clean and process historical power generation data, meteorological data and environmental variables from multiple renewable energy power generation sites, and perform feature extraction and data fusion operations;

[0133] The model prediction module 200 is used to select and train multiple machine learning models, generate multiple independent new energy power generation prediction results, and evaluate the performance of each model;

[0134] The model integration module 300 is used to estimate the weight distribution of each model through the variational Bayesian inference method, integrate the prediction results of multiple models, and generate the final integrated prediction result;

[0135] A dynamic error correction module 400 is used to model the dynamic behavior of the system based on stochastic differential equations and perform dynamic error correction on the prediction results through extended Kalman filtering;

[0136] A correlation modeling module 500 is used to model the correlation between power generation sites using a graph neural network, and input it into a dynamic error correction module as auxiliary information to further optimize the prediction results;

[0137] Parameter optimization module 600, used to dynamically adjust and optimize model parameters based on feedback mechanism and variational reinforcement learning method to ensure that the model can adapt to environmental changes and maintain optimal performance;

[0138] The global optimization module 700 is used to model the global optimization problem of the system as a mixed integer nonlinear programming, solve it using a global optimization algorithm, and output the final corrected new energy power generation prediction result.

[0139] In this embodiment, an artificial intelligence-based renewable energy power generation prediction error correction device provided by the present invention and an artificial intelligence-based renewable energy power generation prediction error correction method described above can correspond to each other; the device of this embodiment can be used to execute the above-mentioned artificial intelligence-based renewable energy power generation prediction error correction method embodiment, and its principles and technical effects are similar, which will not be repeated here.

[0140] The present invention is a new energy power generation prediction error correction method based on artificial intelligence. By integrating multiple machine learning models, dynamic error correction and global optimization, the accuracy, stability and adaptability of new energy power generation prediction are significantly improved. In use, the present invention has the following advantages: ① The present invention can combine the advantages of different models and reduce the bias and error that may be caused by a single model by using multiple machine learning models and integrating them; ② The present invention adopts stochastic differential equations to model the dynamic characteristics of the new energy power generation process, and uses extended Kalman filtering to correct the prediction results in real time, which can effectively cope with the random fluctuations and environmental changes encountered by the system during actual operation; ③ The present invention uses graph neural networks to model the correlation between new energy power generation sites, which can capture and utilize the complex interactive relationship between sites and provide auxiliary information for error correction; ④ The present invention dynamically adjusts model parameters through variational reinforcement learning, thereby realizing adaptive adjustment of the model in the face of environmental changes; ⑤ The present invention models the global optimization problem of the system as mixed integer nonlinear programming, and solves it through an efficient global optimization algorithm, which can find the optimal solution based on considering multiple constraints; The present invention has the advantages of integrating multiple models, having dynamic error correction capabilities, and improving prediction accuracy, stability and adaptability.

[0141] Example 3

[0142] like Figure 3 As shown, a computer device 40 includes: a processor 41 and a memory 42, the memory 42 stores a computer program executable by the processor, and when the computer program is executed by the processor, it executes the above-mentioned artificial intelligence-based new energy power generation prediction error correction method.

[0143] The present invention further provides a storage medium 43 on which a computer program is stored. When the computer program is run by the processor 41, the method for correcting the prediction error of new energy power generation based on artificial intelligence as described above is executed.

[0144] In this embodiment, the storage medium 43 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM); electrically erasable programmable read-only memory (EEPROM); erasable programmable read-only memory (EPROM); programmable red-only memory (PROM); read-only memory (ROM); magnetic storage; flash memory; magnetic disk or optical disk.

[0145] The present invention is a new energy power generation prediction error correction method based on artificial intelligence. By integrating multiple machine learning models, dynamic error correction and global optimization, the accuracy, stability and adaptability of new energy power generation prediction are significantly improved. In use, the present invention has the following advantages: ① The present invention can combine the advantages of different models and reduce the bias and error that may be caused by a single model by using multiple machine learning models and integrating them; ② The present invention adopts stochastic differential equations to model the dynamic characteristics of the new energy power generation process, and uses extended Kalman filtering to correct the prediction results in real time, which can effectively cope with the random fluctuations and environmental changes encountered by the system during actual operation; ③ The present invention uses graph neural networks to model the correlation between new energy power generation sites, which can capture and utilize the complex interactive relationship between sites and provide auxiliary information for error correction; ④ The present invention dynamically adjusts model parameters through variational reinforcement learning, thereby realizing adaptive adjustment of the model in the face of environmental changes; ⑤ The present invention models the global optimization problem of the system as mixed integer nonlinear programming, and solves it through an efficient global optimization algorithm, which can find the optimal solution based on considering multiple constraints; The present invention has the advantages of integrating multiple models, having dynamic error correction capabilities, and improving prediction accuracy, stability and adaptability.

Claims

1. A method for correcting prediction errors of renewable energy power generation based on artificial intelligence, characterized in that: The method comprises the following steps: Step 1: Collect and process historical power generation data, meteorological data, and environmental variables from multiple renewable energy power generation sites, and perform multi-source data fusion and feature extraction; Step 2: Use multiple machine learning models to predict renewable energy power generation and generate multiple independent prediction results; Step 3: Estimate the weight distribution of each model through the variational Bayesian inference method, perform model integration, and generate the final integrated prediction result; Step 4: Use stochastic differential equations to model the dynamic behavior of renewable energy power generation, and use extended Kalman filtering to perform dynamic error correction on the prediction results; Step 5: Use graph neural networks to model the correlation between power generation sites and use it as auxiliary information for error correction; Step 6: Dynamically optimize model parameters through variational reinforcement learning based on feedback mechanism and adaptive parameter adjustment; Step 7: Model the global optimization problem of the system as a mixed integer nonlinear programming, solve it through the global optimization algorithm, and output the final revised new energy power generation prediction result.

2. The artificial intelligence-based new energy power generation prediction error correction method according to claim 1, characterized in that: The multi-source data fusion and feature extraction in step 1 specifically include the following steps: Step 1.1: Collect data from multiple renewable energy power generation sites as the original data set; Step 1.2: Perform principal component analysis on the cleaned data to reduce the dimension and extract the main features that have the greatest impact on the prediction; Step 1.3: Based on principal component analysis, random matrix theory is used to analyze the eigenvalue spectrum of high-dimensional data; Step 1.4: Fusion the data features processed by principal component analysis and random matrix theory, and construct a new feature set based on the actual application requirements of renewable energy power generation; Step 1.5: Standardize and normalize the final feature set to eliminate the dimensional differences between different features; Step 1.6: Use feature selection algorithm to select the features that have the greatest impact on prediction and use them as input data for subsequent model training and prediction.

3. The method for correcting prediction errors of renewable energy power generation based on artificial intelligence according to claim 1, characterized in that: The machine learning models in step 2 include: long short-term memory network, convolutional neural network, autoregressive integrated moving average model and random forest.

4. The method for correcting prediction errors of renewable energy power generation based on artificial intelligence according to claim 1, characterized in that: The step 3 specifically comprises the following steps: Step 3.1: Set the initial weight distribution for each independent prediction model, assuming that the weight of each model follows a certain probability distribution, and initialize the prior distribution p(w) of the model weight; Step 3.2: Use the variational Bayesian inference method to approximate the posterior distribution p(wD), where D represents the training data; the posterior distribution of the weights is determined by the likelihood function p(Dw) and the prior distribution p(w), and an appropriate variational distribution q(w) is selected to approximate the posterior distribution by minimizing the Kullback-Leibler divergence; Step 3.3: Optimize the parameters of the variational distribution q(w) by maximizing the evidence lower bound. The expression for maximizing the evidence lower bound is: ELBO = Ε q(w) [logp(Dw)]-D KL (q(w)p(w)), where the first term is the logarithmic expectation of the likelihood function p(Dw) of the prediction result; the second term is the Kullback-Leibler divergence between the variational distribution q(w) and the prior distribution p(w); Step 3.4: During the optimization process, the parameters of the variational distribution are continuously updated through the optimization algorithm, and the weight distribution of each model is gradually adjusted until the maximum evidence lower bound reaches a stable convergence state, and the optimal weight distribution of each model is obtained; Step 3.5: Apply the optimized variational distribution q(w) to the weighted combination of independent prediction results to generate the final integrated prediction result Among them, y m is the prediction result of the mth model; q(w m ) is the weight of the model; Step 3.6: Output the final integrated prediction result and use it as input data for the subsequent error correction step.

5. The method for correcting prediction errors of renewable energy power generation based on artificial intelligence according to claim 1, characterized in that: The step 4 specifically comprises the following steps: Step 4.1: Model the dynamic behavior of the renewable energy generation process as a stochastic differential equation describing the system state x t The evolution over time t, the model form is: dx t =f(x t ,t)dt+g(x t ,t)dW t , where f(x t ,t) is the drift term, which describes the deterministic change of the system; g(x t ,t) is a diffusion term, reflecting the uncertainty of the system; W t is the Wiener process, which indicates the random fluctuation of the system; d is the differential symbol; Step 4.2: Based on the current state and the drift term f(x t ,t), use stochastic differential equations to predict the system state and get the state estimate for the next time step Among them, Δt is the time step; Step 4.3: According to the diffusion term g(x t ,t) and the Jacobian matrix of the system, calculate the covariance matrix p of the predicted state tt-1 : Among them, F t-1 is the drift term f(x t ,t)’s linearized Jacobian matrix; G t-1 is the coefficient matrix of the diffusion term; Q t-1 is the process noise covariance matrix; Step 4.4: Combine the observed data y of the system t With the observation function h(x t ), using the extended Kalman filter to update the system state estimate and covariance matrix; Step 4.5: State estimation after extended Kalman filtering The preliminary forecast results are dynamically corrected for errors, and the revised forecast value of renewable energy power generation is generated and used as the final forecast output.

6. The artificial intelligence-based new energy power generation prediction error correction method according to claim 1, characterized in that: The step 5 specifically comprises the following steps: Step 5.1: Taking power generation sites as nodes and the physical distances and meteorological condition similarities between sites as edges, an undirected graph G = (V, E) is formed, where V is the site set and E is the edge set; Step 5.2: Initialize feature vector for each node Including the historical power generation and key information of meteorological conditions of the site, each edge e∈E is given a weight, which is calculated based on the physical distance between sites or data similarity; Step 5.3: Build a graph neural network to process the correlation between sites. The message passing process of the graph neural network is described by the following formula: in, is the feature vector of node v at the kth layer; N(v) is the set of neighbor nodes of node v; d u d v are the degrees of node u and node v respectively; W (k) is a trainable weight matrix; is the feature vector of node u at the k-1 layer; b (k) is the bias term; σ is the activation function; Step 5.4: Perform message passing and aggregation through the multi-layer graph neural network model, update the feature representation of each node layer by layer, and the feature representation of each layer k They are updated based on the weighted sum of their neighbor node features, gradually integrating the structural information of the entire graph, and finally outputting the feature vector of the node Represents the comprehensive information of the site in the graph structure; Step 5.5: Output the site feature vector of the graph neural network Combined with the output of the prediction model as auxiliary information for error correction; Step 5.6: The error correction result based on the combination of graph neural network and extended Kalman filter is used as the final power generation prediction output.

7. The artificial intelligence-based new energy power generation prediction error correction method according to claim 1, characterized in that: The step 6 specifically comprises the following steps: Step 6.1: Monitor the environmental status of the renewable energy power generation system in real time, collect power generation, meteorological conditions, and forecast error feedback information, and construct a feedback data set; Step 6.2: Construct a variational reinforcement learning framework, where the state s of the system t It is composed of the current environmental conditions and model parameters. Action a t is the adjustment strategy for model parameters, and the reward r t is the prediction accuracy or error reduction of the model in its current state; Step 6.3: Design the policy network π θ (a t s t ), and optimizes the parameter distribution q(θ) of the policy network through variational inference. The policy network outputs the probability distribution of parameter adjustment. Variational inference approximates the posterior distribution p(θD) by minimizing the Kullback-Leibler divergence. Its optimization goal is: Among them, γ is the discount factor, which controls the influence weight of future rewards on the current strategy; r t is the immediate reward of the current time step; D KL (q(θ)p(θ)) represents the Kullback-Leibler divergence, which measures the difference between the variational distribution q(θ) and the prior distribution p(θ); Step 6.4: Based on the current environment status s t And the output of the policy network, select the optimal action a t , that is, the optimal parameter adjustment strategy; Step 6.5: Execute the selected action a t , update and adjust model parameters in real time; Step 6.6: Accumulate multiple rounds of feedback information and continuously optimize the policy network through multiple policy iterations; Step 6.7: Apply the model parameters that have undergone multiple rounds of adjustment and optimization to the actual power generation forecast to generate the final forecast results.

8. The artificial intelligence-based new energy power generation prediction error correction method according to claim 1, characterized in that: The step 7 specifically includes the following steps: Step 7.1: The objective function and constraints of the renewable energy generation prediction system are expressed as a mixed integer nonlinear programming problem. Let the continuous variable be x and the integer variable be z. The optimization goal is to minimize the total prediction error while satisfying the system resource and computing constraints. The specific expression is: subject to g(x,z)≤0,x∈R n ,z∈Z m , where f(x,z) is the objective function, which represents the prediction error or other system performance indicators; g(x,z) is the constraint condition, including computing resources, time or physical constraints; Step 7.2: Generate an initial solution through a heuristic algorithm and decompose the mixed integer nonlinear programming problem into several sub-problems, each sub-problem corresponds to a nonlinear programming problem or an integer programming problem; Step 7.3: Apply the branch and bound method to solve the problem. Use the method of partitioning the variable space to divide the mixed integer nonlinear programming problem into multiple subproblems, calculate the lower bound of each subproblem, and use the pruning strategy to exclude subregions that are unlikely to contain the optimal solution, thereby narrowing the solution space; Step 7.4: Combine the global optimization algorithm to solve the sub-problems generated by the branch and bound method, and find the optimal solution in the global scope through global search and information sharing mechanism; Step 7.5: Continuously monitor the change of the solution and determine whether the solution has converged by using the set convergence criteria. If the convergence condition is not met, continue to iterate the solution until the convergence criteria are met. Step 7.6: Apply the final global optimal solution to the renewable energy power generation prediction model to generate the final prediction result after global optimization, and use the result as the output of the system.

9. A device for correcting prediction errors of renewable energy power generation based on artificial intelligence, used to implement a method for correcting prediction errors of renewable energy power generation based on artificial intelligence as claimed in any one of claims 1 to 8, characterized in that: The device includes: a data processing module for collecting, cleaning and processing historical power generation data, meteorological data and environmental variables from multiple new energy power generation sites, and performing feature extraction and data fusion operations; Model prediction module, which is used to select and train multiple machine learning models, generate multiple independent renewable energy power generation prediction results, and evaluate the performance of each model; The model integration module is used to estimate the weight distribution of each model through the variational Bayesian inference method, integrate the prediction results of multiple models, and generate the final integrated prediction results; Dynamic error correction module, which is used to model the dynamic behavior of the system based on stochastic differential equations and perform dynamic error correction on the prediction results through extended Kalman filtering; The correlation modeling module is used to model the correlation between power generation sites using graph neural networks, and input it into the dynamic error correction module as auxiliary information to further optimize the prediction results; Parameter optimization module, which is used to dynamically adjust and optimize model parameters based on feedback mechanism and variational reinforcement learning method to ensure that the model can adapt to environmental changes and maintain optimal performance; The global optimization module is used to model the global optimization problem of the system as a mixed integer nonlinear programming, solve it using a global optimization algorithm, and output the final revised new energy power generation forecast results.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the artificial intelligence-based new energy power generation prediction error correction method as described in any one of claims 1 to 8.

Images