基于聚类分析和优化神经网络的风电功率预测研究

Research on Wind Power Prediction Based on Cluster Analysis and Optimized Neural Network

为了提高风电功率的预测精度,减小风电随机性对电网的影响,提出一种基于K均值聚类算法和思维进化算法优化误差反向传播神经网络的风电功率短期组合预测模型。首先,采用K均值聚类算法将风速分为微风、中风和大风,风向分为正风向和反风向共六个类别以降低其随机性。然后,为各个类别分别建立神经网络预测模型,并采用思维进化算法对其初始权值和阈值寻优,再将待预测样本根据所属类别输入到相应的预测模型中,得到最终的预测值。最后利用算例仿真,证明所提的组合预测模型比其他传统预测模型具有更高的预测精度。

A short-term wind power combination prediction model was proposed based on the K-means clustering algorithm as well as optimized error back propagation neural network using the mind evolution algorithm to raise the prediction accuracy of wind power and reduce the impact of wind power randomness upon the power grid.Firstly,K-means clustering algorithm was adopted to divide wind speed into six categories:breeze,moderate wind and gale in positive and negative directions,so as to reduce the randomness.Then,neural network prediction models were established for all categories,respectively,mind evolution algorithm was used to optimize initial weights and threshold values,and the samples to be predicted were imported into corresponding prediction models to obtain final prediction value.Finally,example simulation proved that the proposed combination prediction model had higher accuracy than other traditional models.