基于改进决策树的短期风速预测算法设计

Design of short⁃term wind speed prediction algorithm based on improved decision tree

针对现有短期风速预测算法准确度较低且计算复杂的问题,提出了一种基于改进决策树的短期风速预测算法。该算法采用改进经验模态算法进行风电数据的预处理,以获得多个固有模态分量和残差分量。同时引入改进初始学习机选择策略、自适应学习率以及梯度拟合逼近方法,弥补了传统梯度提升决策树算法中存在的学习训练效果差、计算速度低等不足。利用改进梯度提升决策对风电数据进行特征提取与学习训练,进而实现了短期风速的精准预测。算例分析结果表明,与IEMD-GDBT和EMD-GDBT算法相比,所提算法的训练时间仅为1489.5 s,预测指标RMSE、MAE和MAPE的值分别为0.2286%、0.1827%以及2.37%,在计算速度及预测准确度方面均具有显著优势,实际风速预测误差小于1 m/s。

Aiming at the problems of low accuracy and complex calculation of existing short⁃term wind speed prediction algorithms,a short⁃term wind speed prediction algorithm based on improved decision tree is proposed.The improved empirical mode algorithm is used to preprocess wind power data,and multiple natural mode components and residual components are obtained.By introducing improved initial learning machine selection strategy,adaptive learning rate and gradient fitting approximation method,the shortcomings of traditional gradient lifting decision tree algorithm,such as poor learning training effect and low computing speed,are remedied.The improved gradient lifting decision is used for feature extraction and learning training of wind power data,and the accurate prediction of short⁃term wind speed is finally realized.The numerical analysis results show that compared with the IEMD-GDBT and EMDGDBT algorithms,the training time of the proposed algorithm is only 1489.5 seconds,and the values of the prediction indicators RMSE,MAE,and MAPE are 0.2286%,0.1827%,and 2.37%,respectively.It has significant advantages in calculation speed and prediction accuracy,and the actual wind speed prediction error is less than 1 m/s.