This paper proposes a novel price forecasting model， which can help market participants operate effectively， make their development planning and lower risks associated with price uncertainty in a competitive power market. Firstly， based on historical market information， the impacts of some major factors on price distribution are analyzed in detail. These mainly include demand load， power supply， dispatchable load， power change with neighboring grid and corresponding period. A base price value is introduced to split electricity price into normal and high price parts. Then， the proximal point and support vector machine（SVM） techniques are used to classify future prices， which will be forecasted by different methods. If one future price is judged to be a normal one， it will be estimated using traditional time serial methods， otherwise it will be estimated using weighted historical high price data. The proposed model reduces the price dependence on time by classified forecasting， which is based on price distribution. It can be applied to short-term forecasting， and in the mean time， provides useful information for medium- and long-term price forecasting as well. Finally， the actual market data of Queensland in Australia power market are used to testify its validity and rationality. Moreover， the sensitivity of the base price is analyzed to investigate its influence on classification precision of the presented scheme.