采用最近历史观测值和PLSR进行空间相关性超短期风速预测

Ultra-Short-Term Wind Speed Prediction With Spatial Correlation Using Recent Historical Observations and PLSR

为提高超短期风速预测的可靠性和准确性,将被预测地点(本地)周边测风塔风速风向等当前和最近历史观测值作为基础数据,采用空间相关性来预测本地的未来风速。首先,依据风向和风速的延迟相关性,挑选出上游测风塔。之后,结合最优延迟时间,利用各上游和本地最近的风速观测值来训练预测模型。最后,将各上游风速的当前观测值输入模型,即可得到本地的风速预测值。以偏最小二乘回归(partial least squares regression,PLSR)为主要模型,并采用线性回归(linear regression,LR)、最小二乘支持向量回归等模型进行对照。以冬季风时期的荷兰Huibertgat和天津为被预测地点,进行了PLSR、LR预测误差与模型阶数、样本容量之间关系的数值实验。研究表明,在冬季风时期,当样本容量达到一定程度后,预测误差的变化对阶数、样本容量和模型的类型均不再敏感。这表明空间相关性是一种较为可靠的超短期风速预测方法。

In order to improve reliability and accuracy of ultra-short-term wind speed prediction, wind speeds at prediction site are predicted with a spatial correlation method, using current and recent historical wind speed and direction observations of surrounding anemometer towers. Firstly, wind speed upstream sites are sorted out with their wind directions and lag similarities of wind speeds. Secondly, with optimal lag times taken into account, a prediction model is trained with recent historical observations of wind speed at the prediction site and upstream sites. Lastly, predicted wind speeds of the prediction site are obtained when current wind speed observations of upstream sites are put into the model.The main prediction model is partial least squares regression （PLSR）. Contrast prediction models are a linear regression （LR）, the least squares support vector regression （LSSVR）, etc. Taking Huibertgat in Netherlands and Tianjin as prediction sites, for future wind speeds in winter monsoon period, relations among prediction errors, order of the model and sample size are generated with numerical experiments. Simulation results show that the prediction errors are not sensitive to the order, the sample size and model types during winter monsoon, if the sample size is large enough. These results prove that spatial correlation is a reliable approach to predict future ultra-short- term wind speeds.