基于多重离群点平滑转换自回归模型的短期风电功率预测

基于对风电功率时间序列波动性多重机制的研究,提出一种基于多重离群点平滑转换自回归模型(M-OSTAR)的风电功率预测方法。运用一种改进条件极大似然估计方法,获得M-OSTAR模型的参数估计。考虑风电波动性的厚尾效应,将M-OSTAR模型推广为厚尾形式。进一步借助所提模型的机制转换参数,描述了风电时间序列的多重离群点效应。此外,给出了一种新型的波动性分析工具——标准信息冲击曲面,分析了风电时间序列条件方差的动态变化特征。基于实际风电数据的算例验证了基于M-OSTAR族模型预测方法的可行性与有效性。

Short-term wind power forecast based on MOSTAR model

Based onthe analysis on the different regimes in the volatility of wind power time series, a prospective wind power forecasting method based on Multiple Outlier Smooth Transition Autoregressive (MOSTAR) type models is presented. By modifying Conditional Maximum Likelihood Estimate (CMLE), the parameters of the MOSTAR models are obtained. Considering the fat-tail effect in the volatility of wind power time series, MOSTAR models with fat-tail distribution are proposed for generalization. Moreover, with the regime switching parameter of the proposed model, the multiple outlier effect of real case is depicted more rigorously. In addition, Standard News Impact Surface (SNIS), a refined tool for volatility analysis is provided to analyze the dynamic varying characteristics of conditional variance. Case studies on a practical wind power data validate the feasibility and effectiveness of M-OSTAR type model. This work is supported byNational Natural Science Foundation of China (No. 51577025) and Natural Science Fund for Colleges and Universities in Jiangsu Province (No. 14KJB470003).