Research On Hybrid Forecasting Model Of Short Term Electricity Price In Electricity Market Environment

With the deregulation of power industry, establishing the mechanism of market competition is the trend of power industry. In the electricity market environment, accurate short term electricity price forecasting has become a common issue for all market participants. Generation companies can use price forecasting to optimize the bidding strategies to maximize their profits; Consumers can use price forecasting to optimize the power purchase portfolio to minimize production cost; Price forecasting can also help market regulators to promote the healthy, stable and ordered development of electricity market. However, due to the complicated factors affecting electricity price, the prices present the following complex behavior characteristics, which combine to make forecasting very challenging:multiple seasonality, high volatility, seasonality, randomness, etc.Based on the analysis of research status of short term electricity price forecasting, electricity price affecting factors, electricity price characteristics, electricity price prediction accuracy factors, this paper propose the hybrid forecasting model of electricity price according to different markets, to improve the forecasting accuracy. The main contents of this paper can be summarized as follows:(1) Selection of input variables. Although electricity prices are impacted by many factors, too many or not important factors will reduce forecasting accuracy. Thus, correlation analysis technique is used as the tool to select input variables of electricity price forecasting model, and the bigger correlation coefficient of factors are selected as imput variables.(2) Processing of input data. Directly using initial data does not produce a better result, because of the complicated features of electricity prices. Therefore, wavelet transform is utilized because it can produce a good local representation of the initial data in both time and frequency domains.(3) Selection of sample length. To test the accuracy fo forecasting model, historical datas are usually divided into training samples and testing samples. The length of traning samples will affect the prediction accuracy. Thus, this paper uses sensitivity analysis method to decide the final traning samples length according to different markets.(4) Selection of forecasting model. It has been proved that a single model has difficulty in improving short term electricity price forecating accuracy. Therefore, some researchers proposed combined forecasting models, but these models are difficult to determine the weight coefficien. Based on this, this paper proposes the hybrid forecasting model of electricity price, which can be divided into the following:Hybrid forecasting model based on time series model. Due to the advantages of time series model, such as easy to understand, powerful explanatory ability and lower requirement for new data, this hybrid forecasting model is suitable for the day ahead electricity market, where the electricity prices fluctuate smoothly. A hybrid forecasting model is proposed in this paper, which is based on wavelet transform, autoregressive moving average exogenous model, seasonal autoregressive integrated moving average model, autoregressive integrated moving average model and generalized autoregressive conditional heteroskedasticity model. The proposed model is examined on the California and Ontario electricity market. Case studies testify the validity of the proposed model.Hybrid forecasting model based on time series model and artificial intelligent model. Although time series model can be easily established, it can not describe the relationship between price and its input variables. Besides, the time series model does not have the adaptive learning ability. Thus, the artificial intelligent model is needed, which has the strong nonlinear approximation and self learning ability. However, directly applying artificial intelligent model does not produce a better result. The reason is that electricity price series is composed of a linear autocorrelation structure and a nonlinear componet. Thus, a hybrid model based on time series model and artificial intelligent model is proposed, which is suitable for the real time electricity market, where the electricity prices fluctuate sharply. A hybrid model based on wavelet transform, least squares support vector machine optimized by particle swarm optimization and autoregressive integrated moving average model is presented in this paper. This model is examined by using the data of Australian national electricity market, New South Wales and Queensland. Empirical testing indicates that the proposed method can provide more accurate and effective results.Hybrid forecasting model based on time series model, artificial intelligent model and chaos theory. In recent years, some researches have found that electricity price series has a chaotic property, which indicates that electricity price series is a nonlinear, dynamic evolution process. Thus, the traditional models can not capture this complicated features. A hybrid forecasting model based on time series model, artificial intelligent model and chaos theory is needed, which can improve prediction accuarcy. The hybrid forecasting model is suitable for the electricity market, where electricity prices have a chaotic property. A new hybrid forecasting model based on wavlet transform, radial basis function network, least squares support vector machine, exponential generalized autoregressive conditional heteroskedasticity model and chaos theory is proposed for electricity price forecasting. The superiority of this proposed model is examined by using the data acquired from Spanish and PJM electricity market. Simulation results show that this proposed model has higher prediction accuracy.