基于自回归条件异方差-反向传播网络模型的日前边际电价预测

针对日前电力市场提出了一种基于自回归条件异方差分析的改进神经网络模型。首先利用自回归条件异方差分析得到边际电价序列的条件方差，然后以条件方差作为电价波动风险指标，建立基于历史电价、历史负荷和历史电价条件方差等输入量的自回归条件异方差-反向传播网络模型，并利用该模型对美国PJM电力市场的日前边际电价进行了预测。结果表明，引入自回归条件异方差分析可以有效提高传统反向传播网络的预测精度。

Day-Ahead Marginal Price Forecasting Based on Autoregressive Conditional Heteroskedasticity-Back Propagation Network Model

An improved neural network model based on autoregressive conditional heteroskedasticity (ARCH) analysis is proposed for day-ahead electricity market. Firstly,by use of ARCH analysis the conditional variance of marginal price series is obtained; then taking the conditional variance as the risk index of price fluctuation,an ARCH-back propagation networks (BPN) model,which is based on historical prices,historical loads and conditional variances of historical prices,is built,and by use of the built model the day-ahead marginal prices of Pennsylvania-New Jersey-Maryland (PJM) electricity market in United States are forecasted. Forecasting results show that by means of leading in ARCH the forecasting accuracy of traditional BPN can be effectively improved.