Prediction of power system marginal price based on chaos characteristics

System marginal price is the unified price of reflecting the short-term supply and demand relation of electric commodity in the electricity market. Confirming the system marginal price is the lever and core content of electricity market. At present, how to predict the system marginal price efficiently is one of the focus problems in the research of electricity market application. In the electricity market, system marginal price is not a parameter that varies according to normal rules. It not only has relations to load curve, available generation capacity, bidding mode, system and unit of the generation companies, but also is in reference to price of commodities, inflation and so on. As we can not use mathematic mode to express the diversity and randomicity of factors which reflects electricity price, it is difficult to predict electricity price accurately. In recent years, as the chaotic theory develops, especially speaking, applying chaotic theory to load forecasting offers a new tool to study the stochastic evolution of electricity price. Furthermore, people reveal the regularity that the evolution of both electricity price time series and load time series indicates the chaotic nature, which hiding in these time series themselves. It offers foundation for predicting electricity price by using phase space reconstruction. This paper analyzes the basic character of the system marginal price time series in greater depth. Then, according to the theory of phase space reconstruction, it makes use of the time series of electricity price data, adopts adding weight one-rank local region method, does not consider the stochastic factor related to electricity price, directly analyzes historical data of the electricity price that contains all factors, and carries on forestation based on acquired objective regularity. We carry on the prediction on the historical data from California electricity market, analyze and draw a detailed comparison with existing forecasting methods such as BP algorithm and LS-VSM algorithm, which shows that mean absolute percentage error reduce 22.6% and 12.8% respectively. Therefore, this method can acquire a better forecasting result.