A Temporal Convolutional Network Based Hybrid Model for Short-Term Electricity Price Forecasting

Abstract

Electricity prices have complex features, such as high frequency, multiple seasonality, and nonlinearity. These factors will make the prediction of electricity prices difficult. However, accurate electricity price prediction is important for energy producers and consumers to develop bidding strategies. To improve the accuracy of prediction by using each algorithms' advantages, this paper proposes a hybrid model that uses the Empirical Mode Decomposition (EMD), Autoregressive Integrated Moving Average (ARIMA), and Temporal Convolutional Network (TCN). EMD is used to decompose the electricity prices into low and high frequency components. Low frequency components are forecasted by the ARIMA model and the high frequency series are predicted by the TCN model. Experimental results using the realistic electricity price data from Pennsylvania-New Jersey-Maryland (PJM) electricity markets show that the proposed method has a higher prediction accuracy than other single methods and hybrid methods.

Keywords

• Autoregressive integrated moving average model
• electricity price forecasting
• empirical mode decomposition
• temporal convolutional network