Wind power short-term power prediction method based on error feedback and model dynamic optimization

Abstract

The application discloses a wind power short-term power prediction method based on error feedback and model dynamic optimization, processes wind power cluster historical data and online data, constructs a core feature sample set, adopts a Kmeans-DTW similar scene generation model, adopts sample expansion to complete online sample set construction for new input data, proposes wind power prediction based on an integrated model, adopts an initial proportion weight integrated model to predict each scene, screens potential integrated models to be optimized based on a root mean square error, proposes a comprehensive evaluation index system, and obtains a target optimization scene set according to a certain threshold, and for the target scene, based on the integrated model dynamic adjustment of the Staking algorithm, the Optuna framework is adopted to optimize integrated model super parameters, the optimal integrated model super parameters are determined, and the dynamic optimization integrated model of the typical scene of the future input wind power sample set is realized. The method realizes wind power online prediction, and provides corresponding technologies for dispatching system model optimization.

Description

Technical Field

[0001] This invention relates to a short-term wind power prediction method based on error feedback and dynamic model optimization, belonging to the field of wind power prediction. Background Technology

[0002] The development and utilization of wind power is a crucial support for balancing supply and demand, improving my country's energy consumption structure, and promoting green development and low-carbon transformation—all important strategies for high-quality development. The utilization of wind power is an inevitable trend in my country's energy industry, and wind power generation technology is an important component of the new power system, providing a power source. However, due to the volatility and randomness of wind energy, the negative temperature effect of photovoltaic power generation, and variations in irradiance, the balance of the new power system is significantly affected. Currently, AI + power energy is a goal of my country's energy transformation and development, and also an important way to ensure the safe balance of the power system and enhance grid security. Therefore, if AI technology can be combined with periodic dynamic model adjustments and optimization based on online data to provide a short-term wind power prediction method based on error feedback and dynamic model optimization, offering corresponding solutions for grid dispatch, it will be key to the safe development of the new power system.

[0003] Currently, wind power forecasting primarily relies on offline training, which fails to leverage newly added operational data for dynamic model optimization in real time. This study investigates weekly online wind power data collection, categorizes similar characteristic scenarios, examines the prediction capabilities of ensemble models, proposes appropriate evaluation metrics for model selection, and completes online model optimization—all crucial for improving wind power forecasting accuracy. Summary of the Invention

[0004] The purpose of this invention is to provide a short-term wind power prediction method based on error feedback and dynamic model optimization, in order to solve the problem that offline model training cannot be updated and optimized in real time. The method of this invention can not only perform online prediction, but also dynamically optimize and adjust the model through a corresponding system, which can improve the prediction accuracy of wind power, optimize the model information in real time, and make the prediction more reliable.

[0005] The present invention adopts the following technical solution:

[0006] The wind power short-term power prediction method based on error feedback and dynamic model optimization is characterized by the following steps:

[0007] Step 1: Construction of online sample set for wind power clusters: Historical numerical weather prediction (NWP) data and online weekly operation data of wind power clusters are collected to obtain sample data of wind power clusters. Data processing and feature mining are used to obtain the core feature sample set of wind power clusters. The sample data is then input into a wind power similarity scenario generation model based on Kmeans-DTW (Dynamic Time Warping) to obtain the sample set of each scenario. A method for constructing online sample set of wind power based on an enhanced diffusion model is proposed to obtain the online sample set of wind power clusters under each weather scenario.

[0008] Step 2, Wind power prediction based on ensemble learning: Establish an ensemble model using LSTM (Long Short-Term Memory), Transformer, LightGBM, BPNN (Back Propagation Neural Network), TCN, and CNN models. Input the online sample set of wind power clusters obtained in Step 1 into the initial ensemble model to obtain preliminary prediction results for each scenario and cluster. Screen potential optimization scenarios based on root mean square error, and calculate relevant indicators for accuracy and pass rate. Assign corresponding weights to each indicator, establish a comprehensive evaluation index, and screen the target optimization scenario set and normal scenario set according to the corresponding thresholds.

[0009] Step 3, Dynamic Optimization Technology for Wind Power Cluster Model: Based on the target optimization scenario set from Step 2 as the feature set input, the ensemble model is dynamically adjusted using the Staking algorithm, and cross-validation is used to obtain the adjusted ensemble model. Hyperparameter optimization is performed on the adjusted ensemble model using the Optuna framework, and the ideal function Trail is used for hyperparameter search to determine the optimized ensemble model. Similar typical scenarios in the new input sample set will be used to dynamically update the initial ensemble model. Finally, prediction is performed using the online optimized ensemble model to obtain the prediction result.

[0010] Furthermore, step 1 is specifically performed according to the following steps:

[0011] Step 1.1: Select wind power clusters in the same province, and collect numerical weather prediction (NWP) data and online weekly operation data of each wind farm and photovoltaic power station in the cluster. The numerical weather prediction (NWP) includes meteorological factors such as wind speed, wind direction, irradiance, temperature and humidity.

[0012] Step 1.2: Data processing and feature mining are applied to the NWP data and online weekly operation data of the wind power cluster from Step 1.1 to construct a core sample set for the wind power cluster. Data processing includes data cleaning, outlier removal, and missing data imputation. Normalization is used for data smoothing, and outlier values ​​are removed by identifying outliers, spurious points, and power curtailment points. Missing data is imputed using neighbor and average station imputation methods. Feature mining uses Pearson coefficient and SHAP value for correlation analysis to establish high-dimensional features and construct an optimized core feature sample set.

[0013] Pearson coefficient ρ X，Y The calculation formula is:

[0014]

[0015] In the formula, cov(X,Y) represents the variance. σ Let X and Y represent the standard deviation, and let X and Y represent two different input features.

[0016] Step 1.3: Divide the core feature sample set from Step 1.2 into sets based on a 96-hour time scale. Obtain sample sets for different weather scenarios using the K-means-DTW model. Use 70% of the sample sets from different weather scenarios as the training set for the enhanced diffusion model, and 30% as the test set. Input the training set into the enhanced diffusion model for sample expansion, and then input the generated results into the model training result discriminator for judgment. If the judgment is negative, continue to adjust the parameters of the model on the training set; if the judgment is positive, obtain the pre-trained model. Input the test set into the pre-trained model for result testing. When the t-distributed stochastic neighbor embedding (t-sne) and principal component analysis (PCA) of the data are consistent, output the output result of the pre-trained model, and obtain the online sample set of the wind power cluster, which serves as the data foundation for subsequent online prediction and optimization.

[0017] Furthermore, step 2 is specifically performed according to the following steps:

[0018] Step 2.1: Set fixed weights for the LSTM, Transformer, LightGBM, BPNN, TCN and CNN models respectively to build an initial ensemble model. Input the online sample set of wind power clusters generated in Step 1.3 into the initial ensemble model to obtain preliminary prediction results for each scenario and cluster.

[0019] Step 2.2: For the prediction results of each scenario and cluster, compare the difference of root mean square error, determine the prediction model of the potential scenario to be optimized, calculate the accuracy, pass rate, correlation coefficient between prediction results and error, accuracy of high wind speed segment and accuracy of low wind speed segment under the weather scenario, and statistically analyze the results of each indicator.

[0020] Step 2.3: Assign corresponding weights to the five indicators from Step 2.2, calculate the comprehensive indicator results, set the threshold to 75% to 85%, classify the wind farm data below the threshold as the target dynamic optimization scenario set, and classify the remaining wind farms as the normal scenario set.

[0021] Furthermore, step 3 is specifically performed according to the following steps:

[0022] Step 3.1: Dynamically adjust the weights of each model in the initial ensemble model based on the Stacking algorithm to obtain the adjusted ensemble model. Input the target dynamic optimization scenario set from Step 2.3 as the feature set into the adjusted ensemble model for training. The models in the adjusted ensemble model are cross-validated. When the average accuracy of the validation reaches more than 85%, the adjusted ensemble model is output.

[0023] Step 3.2: Based on the adjusted ensemble model obtained in Step 3.1, optimize the model hyperparameters based on the Optuna framework, define the search space, create the ideal function Trail, obtain the ensemble model hyperparameter optimization results, complete the optimization of the ensemble model, and obtain the optimized ensemble model for the target scenario.

[0024] Step 3.3: For new input sample sets in the future, the similar typical scene part will be dynamically updated by the initial ensemble model of the normal scene set. Finally, the prediction is made through the online optimized ensemble model to obtain the final optimized prediction result.

[0025] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention provides an online prediction method for short-term wind power based on comprehensive evaluation indicators, which can perform online prediction and model optimization for wind power and photovoltaic clusters. By constructing a set of online samples, and utilizing dynamic adjustment and hyperparameter optimization of the model, it provides dynamic updates for similar future scenarios, effectively improving prediction accuracy. It also selects typical scenarios for optimization, thereby improving the computational speed. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the overall process structure of the present invention. Detailed Implementation

[0027] The present invention will be further described below with reference to the accompanying drawings and examples. The purpose of the present invention is to provide a method for short-term wind power prediction based on error feedback and dynamic model optimization.

[0028] This example uses a wind power cluster from Ningxia Hui Autonomous Region in Northwest China. The dataset contains wind power data from 2015 to 2017. Historical, online, and real-time data are simulated using training and testing sets. The wind power cluster comprises 16 wind farms, with some differences between them. The time window for the sample set is 15 minutes, and the prediction time scale is 96 hours. Root mean square error, mean error, accuracy, and pass rate are selected as indicators to evaluate the prediction performance.

[0029] like Figure 1 As shown, the present invention provides a short-term wind power prediction method based on error feedback and dynamic model optimization, characterized by the following steps:

[0030] Step 1: Construction of the online sample set for wind power clusters: Historical numerical weather prediction (NWP) data and online weekly operational data of wind power clusters are collected to obtain sample data for the wind power clusters. Data processing and feature mining are used to obtain a core feature sample set for the wind power clusters. This sample data is then input into a wind power similarity scenario generation model based on Kmeans-DTW (Dynamic Time Warping) to obtain sample sets for each scenario. A method for constructing the online sample set for wind power based on an enhanced diffusion model is proposed to obtain the online sample set of wind power clusters under each weather scenario. The specific steps are as follows:

[0031] Step 1.1: Select wind power clusters in the same province, and collect the numerical weather prediction (NWP) data and online weekly operation data of each wind farm and photovoltaic farm in the cluster, which together have n dimensions. The numerical weather prediction (NWP) includes meteorological factors such as wind speed, wind direction, irradiance, temperature and humidity.

[0032] Step 1.2: Data processing and feature mining are applied to the NWP data and online weekly operation data of the wind power cluster from Step 1.1 to construct a core sample set for the wind power cluster. Data processing includes data cleaning, outlier removal, and missing data imputation. Normalization is used for data smoothing, and outlier values ​​are removed by identifying outliers, spurious points, and power curtailment points. Missing data is imputed using neighbor and average station imputation methods. Feature mining uses Pearson coefficient and SHAP value for correlation analysis to establish high-dimensional features and construct an optimized core feature sample set.

[0033] Pearson coefficient ρ X，Y The calculation formula is:

[0034]

[0035] In the formula, cov(X,Y) represents the variance. σ Let X and Y represent the standard deviation, and let X and Y represent two different input features.

[0036] Step 1.3: Divide the core feature sample set from Step 1.2 into sets based on a 96-hour time scale. Obtain sample sets for different weather scenarios using the K-means-DTW model. Use 70% of the sample sets from different weather scenarios as the training set for the enhanced diffusion model, and 30% as the test set. Input the training set into the enhanced diffusion model for sample expansion, and then input the generated results into the model training result discriminator for judgment. If the judgment is negative, continue to adjust the parameters of the model on the training set; if the judgment is positive, obtain the pre-trained model. Input the test set into the pre-trained model for result testing. When the t-distributed stochastic neighbor embedding (t-sne) and principal component analysis (PCA) of the data are consistent, output the output result of the pre-trained model, and obtain the online sample set of the wind power cluster, which serves as the data foundation for subsequent online prediction and optimization.

[0037] Step 2, Wind Power Prediction Based on Ensemble Learning: An ensemble model is established using LSTM, Transformer, LightGBM, BPNN, TCN, and CNN models. The online sample set of wind power clusters obtained in Step 1 is input into the initial ensemble model to obtain preliminary prediction results for each scenario and cluster. Potential optimization scenarios are screened based on the root mean square error, and relevant indicators of accuracy and pass rate are calculated. Appropriate weights are assigned to each indicator to establish a comprehensive evaluation index. The target optimization scenario set and the normal scenario set are selected based on corresponding thresholds. Specifically, the following steps are performed:

[0038] Step 2.1: Set fixed weights for the LSTM, Transformer, LightGBM, BPNN, TCN and CNN models respectively to build an initial ensemble model. Input the online sample set of wind power clusters generated in Step 1.3 into the initial ensemble model to obtain preliminary prediction results for each scenario and cluster.

[0039] Step 2.2: For the prediction results of each scenario and cluster, compare the difference of root mean square error, determine the prediction model of the potential scenario to be optimized, calculate the accuracy, pass rate, correlation coefficient between prediction results and error, accuracy of high wind speed segment and accuracy of low wind speed segment under the weather scenario, and statistically analyze the results of each indicator.

[0040] Step 2.3: Assign corresponding weights to the five indicators from Step 2.2, calculate the comprehensive indicator result, and set the threshold to 80%. Of course, the threshold only needs to be between 75% and 85%. In this embodiment, the threshold is set to 80%. Wind farm data below the threshold are classified into the target dynamic optimization scenario set; the remaining wind farms are classified into the normal scenario set.

[0041] Step 3, Dynamic Optimization Technology for Wind Power Cluster Model: Based on the target optimization scenario set from Step 2 as the feature set input, the ensemble model is dynamically adjusted using the Staking algorithm, and cross-validation is used to obtain the adjusted ensemble model. Hyperparameter optimization is performed on the adjusted ensemble model using the Optuna framework, employing the ideal function Trail for hyperparameter search to determine the optimized ensemble model. Similar typical scenarios in the new input sample set are used to dynamically update the initial ensemble model. Finally, prediction is performed using the online optimized ensemble model to obtain the prediction result. Specifically, the following steps are followed:

[0042] Step 3.1: Dynamically adjust the weights of each model in the initial ensemble model based on the Stacking algorithm to obtain the adjusted ensemble model. Input the target dynamic optimization scenario set from Step 2.3 as the feature set into the adjusted ensemble model for training. The models in the adjusted ensemble model are cross-validated. When the average accuracy of the validation reaches more than 85%, the adjusted ensemble model is output.

[0043] Step 3.2: Based on the adjusted ensemble model obtained in Step 3.1, optimize the model hyperparameters based on the Optuna framework, define the search space, create the ideal function Trail, obtain the ensemble model hyperparameter optimization results, complete the optimization of the ensemble model, and obtain the optimized ensemble model for the target scenario.

[0044] Step 3.3: For new input sample sets in the future, the similar typical scene part will be dynamically updated by the initial ensemble model of the normal scene set. Finally, the prediction is made through the online optimized ensemble model to obtain the final optimized prediction result.

Claims

1. A method for short-term wind power prediction based on error feedback and dynamic model optimization, characterized in that... Follow these steps: Step 1, Construction of online sample set of wind power cluster: Collect historical numerical weather forecast (NWP) data and online weekly operation data of wind power cluster to obtain sample data of wind power cluster. Use data processing and feature mining to obtain sample data of wind power cluster core feature sample set of wind power cluster. Inputting the data into a wind power similarity scene generation model based on Kmeans-DTW, we obtain sample sets for each scene. We propose a method for constructing online wind power sample sets based on an enhanced diffusion model, and obtain online sample sets of wind power clusters under various weather scenarios. Step 2, Wind power prediction based on ensemble learning: Establish an ensemble model using LSTM, Transformer, LightGBM, BPNN, TCN, and CNN models. Input the online sample set of wind power clusters obtained in Step 1 into the initial ensemble model to obtain preliminary prediction results for each scenario and cluster. Screen potential optimization scenarios based on root mean square error, and calculate relevant indicators for accuracy and pass rate. Assign corresponding weights to each indicator, establish a comprehensive evaluation index, and screen the target optimization scenario set and normal scenario set based on the corresponding thresholds. Step 3, Dynamic Optimization Technology for Wind Power Cluster Model: Based on the target optimization scenario set from Step 2 as the feature set input, the ensemble model is dynamically adjusted using the Staking algorithm, and cross-validation is used to obtain the adjusted ensemble model. Hyperparameter optimization is performed on the adjusted ensemble model using the Optuna framework, and the ideal function Trail is used for hyperparameter search to determine the optimized ensemble model. Similar scenario parts in the new input sample set will be used to dynamically update the initial ensemble model. Finally, the prediction result is obtained by using the online optimized ensemble model.

2. The wind power short-term power prediction method based on error feedback and model dynamic optimization according to claim 1, characterized in that, Step 1 is specifically performed as follows: Step 1.1: Select wind power clusters in the same province, and collect numerical weather prediction (NWP) data and online weekly operation data of each wind farm and photovoltaic power station in the cluster. The numerical weather prediction (NWP) includes meteorological factors such as wind speed, wind direction, irradiance, temperature and humidity. Step 1.2: Data processing and feature mining are applied to the NWP data and online weekly operation data of the wind power cluster from Step 1.1 to construct a core sample set for the wind power cluster. Data processing includes data cleaning, outlier removal, and missing data imputation. Normalization is used for data smoothing, and outlier values ​​are removed by identifying outliers, spurious points, and power curtailment points. Missing data is imputed using neighbor and average station imputation methods. Feature mining uses Pearson coefficient and SHAP value for correlation analysis to establish high-dimensional features and construct an optimized core feature sample set. Pearson coefficient The calculation formula is: In the formula, Represents variance. Let X and Y represent the standard deviation, and let X and Y represent two different input features. Step 1.3: Divide the core feature sample set from Step 1.2 into sets based on a 96-hour time scale. Obtain sample sets for different weather scenarios based on the Kmeans-DTW model. Use 70% of the sample sets for different weather scenarios as the training set for the enhanced diffusion model and 30% as the test set for the enhanced diffusion model. Input the training set into the enhanced diffusion model to expand the samples. Then input the generated results into the model training result discriminator for judgment. If the judgment is negative, continue to adjust the parameters of the model on the training set. If the result is positive, a pre-trained model is obtained. The test set is then input into the pre-trained model to test the results. If the random neighborhood embedding and principal component analysis of the data t-distribution are consistent, the output of the pre-trained model is output, resulting in an online sample set of wind power clusters, which serves as the data basis for subsequent online prediction and optimization.

3. The wind power short-term power prediction method based on error feedback and model dynamic optimization according to claim 2, characterized in that, Step 2 is specifically performed as follows: Step 2.1: Set fixed weights for the LSTM, Transformer, LightGBM, BPNN, TCN and CNN models respectively to build an initial ensemble model. Input the online sample set of wind power clusters generated in Step 1.3 into the initial ensemble model to obtain preliminary prediction results for each scenario and cluster. Step 2.2: For the prediction results of each scenario and cluster, compare the difference of root mean square error, determine the prediction model of the potential scenario to be optimized, calculate the accuracy, pass rate, correlation coefficient between prediction results and error, accuracy of high wind speed segment and accuracy of low wind speed segment under the weather scenario, and statistically analyze the results of each indicator. Step 2.3: Assign corresponding weights to the five indicators from Step 2.2, calculate the comprehensive indicator results, set the threshold to 75% to 85%, classify the wind farm data below the threshold as the target dynamic optimization scenario set, and classify the remaining wind farms as the normal scenario set.

4. The short-term wind power prediction method based on error feedback and dynamic model optimization according to claim 3, characterized in that, Step 3 is specifically performed as follows: Step 3.1: Dynamically adjust the weights of each model in the initial ensemble model based on the Stacking algorithm to obtain the adjusted ensemble model. Input the target dynamic optimization scenario set from Step 2.3 as the feature set into the adjusted ensemble model for training. The models in the adjusted ensemble model are cross-validated. When the average accuracy of the validation reaches more than 85%, the adjusted ensemble model is output. Step 3.2: Based on the adjusted ensemble model obtained in Step 3.1, optimize the model hyperparameters based on the Optuna framework, define the search space, create the ideal function Trail, obtain the ensemble model hyperparameter optimization results, complete the optimization of the ensemble model, and obtain the optimized ensemble model for the target scenario. Step 3.3: For new input sample sets in the future, the similar scene part will be dynamically updated by the initial ensemble model of the normal scene set. Finally, the prediction will be made through the online optimized ensemble model to obtain the final optimized prediction result.

Images