Application Of PSO Optimized GM(1,1) Power-neural Network On Short-term Power Grid Price Forecasting

Electricity price is the key index evaluate the market competition efficiency and reflects the operation condition of electricity market for electricitr market decision-making. It means great significance that forecasting electricity price to guide investment, spontaneous market allocation of resources, achieving a basic balance between supply and demand of electricity to meet the service goals. More accurate short-term price forecasting will help power suppliers to obtain maximum profits, provide optimal bidding strategy, and also allow the dynamic power purchase on cost control is possible. At the time, it will provide real-time supervision and the important scientific basis for regulatory authorities to ensure the normal operation of the electricity market.This paper briefly describes the existing several major price forecast methods. Such as:statistical methods, neural networks, the market simulation and the gray system theory forecasting methods. Further analysis of existing methods based on particle swarm optimization focuses on the GM (1,1) power and BP neural network forecasting model is suggested in this paper.The main results are as follows:1. According to the characteristics of price data in the first hours of electricity sub-done deal, at all points in time a linear regression model is built to fit the period known and to predict the price unknown;2. The linear regression data is compared with the original price data. Then the linear term is removed;3. The use of GM (1,1) power model and BP neural networks are applicable to the handling characteristics of nonlinear problems. The model is structured by GM (1,1) model and neural network model and optimised by particle swarm optimization; 4. At last, the PSO of GM (1,1) power and neural network model is applicated to predict electricity prices, adding the linear term removed, get the forecast electricity demand.5. Using New South Wales in Australia electricity market historical data, its price in February 2009 is predicted by the PSO using GM (1,1) power and neural network model, and the mean absolute percent error is 19.1261%.