基于非参数GARCH的时间序列模型在日前电价预测中的应用

Day-ahead Electricity Price Forecasting Using Time Series Model Based on Nonparametric Generalized Auto Regressive Conditional Heteroskedasticity

电力市场中电价序列具有较强的波动性、周期性和随机性,以致经常出现价格尖峰,这在很大程度上影响了电价预测的精度。提出了一种基于小波变换和非参数GARCH(generalized auto regressive conditional heteroskedasticity)模型的时间序列模型对日前电价进行预测。利用小波变换将历史电价序列分解重构概貌序列和细节序列,分别建立累积式自回归滑动平均(auto-regressive integrated moving average,ARIMA)模型进行预测,采用非参数GARCH模型对电价序列预测残差的随机波动率进行建模,从而提高对价格波动性的预测能力和ARIMA模型的预测精度。将该模型应用于美国宾夕法尼亚—新泽西—马里兰(Pennsylvania-New Jersey-Maryland,PJM)电力市场的日前电价预测。算例结果表明,非参数GARCH模型可以更好地拟合电价序列剧烈波动的特性,该模型能够提高电价的预测精度。

In electricity market the price sequence fluctuates frequently, periodically and stochastically, consequently price spikes often appear and it impacts the accuracy of price forecasting. Based on wavelet transform and nonparametric generalized auto regressive conditional heteroscedasticity （NPGARCH） model, a time sequence model for the forecasting of day-ahead price was proposed. Utilizing wavelet transform, the historical price sequence was decomposed and reconstructed into a general picture sequence and a detail sequence, and corresponding auto-regressive integrated moving average （ARIMA） models were built respectively for price forecasting. The NPGARCH model was used for the modeling of stochastic volatility of residual error in forecasting results of ARIMA models to improve both forecasting ability of price spikes and forecasting accuracy of ARIMA models. The proposed models were applied to day-ahead price forecasting of the electricity market of Pennsylvania-New Jersey-Maryland （PJM）. Forecasting results of the calculation example show that the proposed model can better fit the characteristics of price sequence that fluctuates intensively, and the forecasting results of day-ahead price are more accurate.