Modeling Methodology And Its Application For Price Forecasting In Electricity Market

Electricity price in power market is closely linked to the interests of the participants. It is the focus that their concerns. It has profound effects to forecast price precisely for all the related sides, not only the power industry, the regulator, but also the publics. It is difficult to forecast with one model for its high volatility and obvious difference in different markets. It should be modeled respectively with special inputs selected for each kind market. The characters of electricity price and its influence factors were discussed in detail for following modeling. Then power load, the most important factor, was forecasted. Support Vector Machine (SVM) with parameters optimized by Genetic Algorithm(GA) were applied to forecast short term load; and enhanced gray methodology considering seasonal factors to forecast mid-term load.General Autoregressive Conditional Heteroskedasticity(GARCH) was applied for stationary day-ahead price forecasting. GARCH considered more volatility clustering than other time series modeling. Exogenous variable, the daily price ratios of different weekday types were led into the GARCH model to intensify its response to external influences. Wavelet was used to reduce the fluctuation of price series before modeling.For other more volatility day-ahead price, Intelligence Algorithms (IA), Artificial Neural Network(ANN) and SVM, were applied for forecasting model. Principle Component Analysis (PCA) was employed to mine main information of the inputs in ANN modeling. For the peculiar price during the time of summer peak load and holiday, Self-Organizing Mapping (SOM) was used to cluster the price and its influential factors automatically, then each cluster was modeled by a SVM model. Cases study shows that the PCA helps to improve the modeling accuracy, also the SOM cluster works well in forecasting modling of summer peak price.An integrated methodology of IA and time series modeling was applied for real time price forecasting. Time series, GARCH models were applied to adjust the error series of price forecasted by IA models, SVM-GA, eliminating their autocorrelations and heteroscedasticity effects.Under many uncertain influences, medium and long term electricity price is hard to forecast with the traditional statistics methods. This paper applied Empirical Mode Decomposition (EMD) to decomposed price into several intrinsic modes intuitively. The influence on these modes of different factors was discussed in detail, and then the influence of price was discussed from the variety of time scale. At last time series modeling and SVM modeling were performed according to the characters of the modes.