风电场风速短期多步预测改进算法

Optimization Algorithm of Short-term Multi-step Wind Speed Forecast

对风电场风速实现较准确的预测,可以有效减轻并网后风电对电网的影响,提高风电市场竞争力。该文运用时间序列法对我国某风电场测风站实测风速建立时序求和自回归滑动平均(auto regress iveintegrated moving average,ARIMA)ARIMA(11,1,0)模型,并进行风速预测。针对模型在超前1步预测时出现的延时问题,引入卡尔曼预测法加以改进,提出卡尔曼时间序列法。针对时序模型超前多步预测精度低的问题,提出滚动式时间序列法。对提出的两种改进方法进行实例验证,结果表明,卡尔曼时间序列法不仅改善了预测延时问题,而且把超前1步预测的平均相对预测误差从6.49%降低为3.19%;滚动式时间序列法改善了多步预测的精度问题,模型超前3、5、10步预测的平均相对预测误差分别仅为7.01%,7.63%,8.42%。两种改进方法都没有明显增加时间序列法的建模计算量。

Giving a high precise wind speed forecast for wind farms, which can effectively relieves disadvantageous impact of wind power plants on power systems, enhances the competitive ability of wind power in electricity market. Using time series method to establish ARIMA （11,1,0） model for some wind speed directly measured from wind farms＇ certain station in China. Then performed forecasting simulation by the established model. Aimed at the time delay of one-step forecast by ARIMA （11,1,0） model, authors proposed an improved algorithm named Kalman time-series method. Aimed also at the low accuracy of multi-step forecast by ARIMA （11,1,0） model, in addition authors proposed an improved algorithm named Rolling Amend Time-series method. Using the two improved methods to make calculative examples, which show that：① Kalman time-series method not only solved the time-delay problem in some degree, but also has the mean forecast error of one-step forecast reducing from 6.49% to 3.19%; ②Rolling Amend Time-series method improved the accuracy of multi-step forecast, the mean absolute relative error of three-step, five-step, ten-step forecast respectively are only 7.01%, 7.63% and 8.42%. More important, the two proposed methods did not significantly increase the computation complexity.