A Novel Hybrid Forecasting Method Using GRNN Combined With Wavelet Transform and a GARCH Model

Abstract

To improve the accuracy of electricity price forecasting, a novel hybrid forecasting method using a generalized regression neural network (GRNN) combined with wavelet transform and a generalized autoregressive conditional heteroskedastic (GARCH) model was proposed. The hourly price series usually contains nonlinearity and volatility components. By wavelet decomposition, the price series can efficiently be decomposed into its components. Then, the nonlinearity component is predicted by GRNN and the volatility component is predicted by a GARCH model. The final forecast is obtained by composing the forecasted results of each component. This proposed method was applied in the Spanish electricity market and compared with some other forecasting methods. Results show that the proposed method presents better forecasting performance.

Keywords:

• electricity price forecasting
• wavelet transform
• GARCH
• GRNN

Additional information

Funding

This work was supported by the China National Science Foundation (No.71071053, 70971038) and the Construction Project of Beijing Municipal Commission of Education.