摘要: |
研究了负荷时间序列波动性,考虑方差时变特征,提出了基于随机波动(SV)模型的短期负荷预测方法。引入伪极大似然估计解决SV参数估计问题,进而将模型转换为状态空间方程,利用卡尔曼滤波获取标准SV模型参数。另外,还将模型推广为非高斯假设SV模型。利用动态波动曲线的构建,讨论了负荷时间序列条件方差的时变性特征。基于日用电量数据建立了SV族日负荷预测模型,并利用平均绝对百分误差、均方误差、TIC 3种指标将SV族模型预测结果与广义自回归条件异方差(GARCH)模型做了比较,得到SV族模型的前2种指标均小于GARCH模型,而且SV模型的TIC指标更接近于零。算例分析表明了SV族负荷预测模型的可行性和有效性。 |
关键词: 双伽马函数 厚尾 卡尔曼滤波 负荷预测 伪极大似然估计 状态空间 随机波动模型 |
DOI: |
分类号: |
基金项目: |
|
Short-term load forecasting based on SV model |
CHEN Hao1, WANG Yurong2
|
1.Jiangsu Nanjing Power Supply Company,Nanjing 210008,China;2.Southeast University,Nanjing 210096,China
|
Abstract: |
The volatility of load time series is analyzed,and the short-term load forecasting based on SV(Stochastic Volatility) models is presented with the consideration of the time-varying characteristics. The QMLE(Quasi Maximum Likelihood Estimate) is introduced to estimate the SV parameters and the model is then transformed into state space equations. The Kalman filter is employed to obtain the standard SV parameters and the extended non-gaussian SV model is proposed. The dynamic volatility curve is constructed to discuss the time-varying characteristics of the load time series. The SV class models are established based on daily load data. Three indices are compared between SV model and GARCH model:MAPE(Mean Absolute Percentage Error),RMSE(Root Mean Squared Error) and TIC(Theil Inequality Coefficient). The MAPE and RMSE of SV model are less than those of GARCH model and the TIC of SV model is nearer to zero. Case study verifies the validity and feasibility of SV class models. |
Key words: digamma function fat-tail Kalman filter load forecast quasi maximum likelihood estimate state space stochastic volatility model |