基于改进空间相关法和径向基神经网络的风电场短期风速分时预测模型

Multi-interval wind speed forecast model based on improved spatial correlation and RBF neural network

空间相关法利用空间相关点风速数据之间的相似性和延时性进行风速预测,但在实际应用中存在数据收集困难的问题。提出用空间平移法对空间相关法进行改进,通过减少空间相关点的数目,可有效地降低数据收集难度。为了确定空间相关点风速与所需预测的风电场风速数据之间的非线性关系,采用径向基(RBF)神经网络,建立了基于空间相关法的分时预测模型。该方法通过对风电场与空间相关点风速时间序列之间的关联度分析,将未来预测时段分成若干个时段,在每个时段内分别选择关联度高的相关点的风速数据,作为RBF神经网络的输入数据进行训练和预测。算例表明,该方法可提高风电场风速预测的预测精度,减少了RBF神经网络的训练时间。

The spatial correlation method forecasts the wind speed based on the similarity and delay of wind speed between spatially correlative sites, but it is difficult to collect data in applications. The spatial translation method is presented to make the data collection easier by reducing the number of correlative sites ,which uses RBF（Radial Basis Function） neural network to set the nonlinear relation of wind speed between correlative sites. A multiinterval forecast model based on the spatial correlation is built,which divides the forecast period into several intervals ,analyzes the chronological correlation of wind speed between wind farm and correlative site and selects the best correlative site for each interval. The wind speed data of that site are used as the input data of RBF neural network to train the neural network and forecast the wind speed of that interval. The example calculation shows that,the wind speed forecast accuracy is improved and the training time is reduced.