Research On Short-term Electricity Price Forecasting Under Power Market

Electricity price forecasting under power market is a very complicated issue. Short-term electricity price forecasting is the focus of this paper. Electricity price is a sequence of time-series.Due to complication of relative affected factors,the electricity price curves are characterized by multicycle,fluctuation and spike. So electricity price forecasting is more difficult than load forecasting in power market.The self-organizing data mining combined forecasting is introduced to study the short-term electricity price forecasting in this paper. It can achieve the optimal combination of forecasting models.It used three forecasting models which are auto regressive integrated moving average(ARIMA) model,subtractive clustering and adaptive neuron-fuzzy inference system(ANFIS),grey model of period residual modification,and each has its own characteristics.The ARIMA model used historical data to establish time-series model of electricity price.It implied the synthetic factors which impact electricity price in the historical data.The physical meaning of model could be explained clearly and easily.The ANF1S based on Takagi-Sugeno model.Subtractive clustering algorithm is used to determine the structure of adaptive neuron-fuzzy inference system.Then the hybrid algorithm is used to train the parameter of the fuzzy inference system. Finally,inputted the related data to the trained model and anticipated electricity price is gotten.The grey model improved the initial value and applied the period residual modification. So the simulated curve is approximate to original data and the high accurancy is obtained.The method of self-organizing data mining combined forecasting which takes into account the characteristics of the three models is applied in short-term electricity price forecasting.It achieved the optimal combination of many forecasting models and used the new information of each forecasting model.The advantages of combined forecasting method are that it is a non-linear combined forecasting method and could display in direct function which can be used for analysis and interpretation.In addition, the model evaluation criteria which counld make use of new information to choose optimal complicated model.So the model has good predictive ability.California Electricity Market is used as our case study.Three single forecasting models and combined forecasting model are established.The predicted result shows the disposal in forecasting model has higher accuracy and is satisfactory.In view of the frequency of electricity price forecasting and the important reference of transaction inday-ahead power market,electricity price forecasting system has been designed andimplemented using the MALAB software and SPSS software.