基于核主成分分析和食肉植物算法优化随机森林的风电功率短期预测

Short-term Wind Power Prediction Based on Random Forest Optimized by Kernel Principal Component Analysis and Carnivorous Plant Algorithm

为提高风电功率短期预测的精度,提出一种基于核主成分分析和食肉植物算法(carnivorous plant algorithm,CPA)优化随机森林(random forest,RF)的风电功率短期预测方法。首先,利用核主成分分析从13个气象因素中提取出8个与风电功率相关的气象因素,将这8个气象因素输入到预测模型中。然后,利用CPA优化RF构建CPA-RF预测模型解决RF预测模型预测精度不够高的问题。最后,选取实际风电功率数据进行测试,测试结果表明,利用核主成分分析选取8个气象因素作为输入的效果要优于直接输入13个气象因素的效果,CPA-RF预测模型的预测精度高于长短期记忆网络(long short-term memory,LSTM)预测模型、双向长短期记忆神经网络(bidirectional long short-term memory,BiLSTM)预测模型和RF预测模型。该方法可为提升风电功率短期预测精度提供参考。

In order to improve the accuracy of short-term wind power prediction,a short-term wind power forecasting method based on kernel principal component analysis and carnivorous plant algorithm(CPA)optimized random forest(RF)was proposed.Firstly,8 meteorological factors related to wind power were extracted from 13 meteorological factors by kernel principal component analysis,and then these 8 meteorological factors were input into the prediction model.Then,the carnivorous plant algorithm was used to optimize the random forest,and to construct the CPA-RF prediction model,which can solve the problem that the prediction accuracy of the RF prediction model is not high enough.Finally,The actual wind power data was selected for testing.The test results indicate that 8 meteorological factors which are extracted through kernel principal component analysis method,is used as input.The effect is better than that of 13 meteorological factors directly inputted.The CPA-RF prediction model with higher prediction accuracy,significantly outperforms LSTM prediction model as well as other comparable models including BiLSTM and RF prediction model.This method can provide a reference for accuracy improvement of the short-term wind power prediction.