Research On Artificial Intellgence Electricity Price Method In Deregulated Markets

In deregulated electricity market,as marketization is introduced into the electric power industry,electricity price forecasting becomes the most value tools for all market participants.One side,accurate price forecasting help the power supply establish the bid strategy to obtain the maximum benefit;on the other side,a valuable price forecasting make producer avoid risk due to the price fluctuation and optimize the production plan.Long time price forecasting affects the decision for power transmission expand,distribute plan and district energy trade.The purpose of the medium-term forecasting is to provide negotitations on bilateral contracts to suppliers and consumers successfully.Short-time price forecasting can help company to realize the maximization of profit in the spot market.Therefore,an effective and perfect price forecasting method is necessarily to company in power market transaction.Due to the complex bid strategy and various factors,the electricity price has the following characteristics: high frequency,non-stationary,multi-seasonality,mean reversion,calendar effect,high level of volatility,high percentage of unusual price movement and limited information for market participants,accurate price forecasting becomes a challenging task.To investigate effective model which can be improve the price prediction accuracy,under base the research on price forecasting method and classify,the factor and characteristics of price,the process of the price forecasting and data pre-process methods,in this paper,we proposed the several kind of hybrid electricity price forecasting methods based no artificial intelligence neural network.First,we research a few of common artificial intelligence neural network methods.A single method produces larger prediction errors.When the artificial neural networks are optimized by genetic algorithm and fish algorithm,the prediction error can be improved.Second,in order to capture the hidden information in electricity prices,we propose a hybrid price forecasting method based on wavelet transform and extreme learning machine.The volatility of the original price data reduces after wavelet transform.The new forward back single hidden layer neural network called extreme learning machine can be avoid some difficulties faced by gradient-based learning methods such as learning rate,learning speed,local minima,over-fitting,and low efficiency.Through Australia,PJM and California price forecasting,the proposed method can reduce the error and improve the prediction accuracy.Duo to the complex price behavior,single prediction model may not capture the decomposed price information by wavelet transform completely.Therefore,in order to obtain more accurate predictions,we proposed a new hybrid model based on ARMA,kernel-based extreme learning machine and adaptive particle swarm optimized algorithm.Through testing the stationary of decomposed price series,the ARMA is used to predict stationary price series.Because the kernel based extreme learning machine use kernel function to replace the hidden layer function,it reduce the randomness of normal extreme learning machine,and adaptive particle swarm optimized algorithm make kernel based extreme learning machine has more stable generalization ability.An adaptive particle swarm optimized algorithm combined the kernel extreme learning machine is used to forecast the non stationary decomposed price series.The price data from PJM,Australia and Spanish are used to evaluate the prediction performance of the proposed model.The experimental results show the proposed model can obtain accurate,rational,and effective electricity price prediction.The contributions and major achievements in this paper is that under detailed analysis the mechanism,characteristices,process of electricity price,we combined the artificial neural network,extreme learning machine,kernel based extreme learning machine,artificial intelligence optimized algorithm,wavelet transform,ARMA,and proposed a few of different price prediction models.Enough data to verify the effectiveness of the proposed model,and the proposed model is applied to price forecasting of the world major power markets suceessfully.