基于互信息属性分析与极端学习机的超短期风速预测

超短期风速预测对保证风电并网运行可靠性和维持电力系统安全稳定具有重要的意义,针对风速预测中不同因素对风速影响程度不同的特点,本文提出一种基于互信息属性分析与极端学习机的超短期风速预测方法。首先,选取与风速相关的68种候选属性因素,分别计算其相对于风速序列的互信息值,根据互信息,衡量属性对风速的影响程度,并选择输入属性;然后,由互信息值计算属性权值;之后,采用加权处理后的属性值来训练极端学习机,构建风速预测模型;最后,采用新模型预测未来4h内风速。采用北纬39.91°、西经105.29°的美国风能技术中心的实测数据开展实验,实验结果表明,新方法具有良好的预测精度,能够满足实际风速预测需要。

Ultra-short-term wind speed forecasting based on mutual information attributes analysis and extreme learning machine

Ultra?short?term wind speed forecasting has great practical significance to ensure the reliability with the wind farm connected to the grid, and is very helpful to maintain the safety and stability of power system. Different properties for wind speed forecasting have different influence degrees. A new method based on mutual information a?nalysis and extreme learning machine is proposed in this paper. Firstly, five properties used for wind speed forecas?ting are selected for wind speed forecasting. The mutual information values of each property are calculated to measure the correlations between properties and wind speed series. Then, the data of properties weighted with the values of their mutual information are used to train the extreme learning machine. Finally, the trained ELM is used to forecast the wind speed. The experiment is carried out with the data from US NWTC. The experimental results show that the new method has good forecasting accuracy, and it can meet the actual needs of wind speed forecasting.