Deep learning for day‐ahead electricity price forecasting. Issue 4 (22nd April 2020)

• Record Type: Journal Article
• Authors: Zhang, Chi; Li, Ran; Shi, Heng; Li, Furong
• Is Part Of: IET smart grid. Volume 3:Issue 4(2020)
• Page Start: 462
• Page End: 469
• Publication Date: 2020-04-22
• Subjects: learning (artificial intelligence) -- power system economics -- power engineering computing -- economic forecasting -- pricing -- support vector machines -- recurrent neural nets -- power markets; deregulated electricity market -- multivariate EPF model -- New England electricity market -- deep learning -- day‐ahead electricity price forecasting -- accurate electricity price forecasting -- market participants -- price movements -- forecasting model -- electricity consumption -- natural gas price -- deep recurrent neural network method; B0240Z Other topics in statistics -- B8110B Power system management, operation and economics -- C1140Z Other topics in statistics -- C5290 Neural computing techniques -- C6170K Knowledge engineering techniques -- C7410B Power engineering computing; Smart power grids -- Periodicals; Computer science -- Periodicals; Energy industries -- Periodicals; Broadcasting -- Periodicals; 333.79110285
• DOI: 10.1049/iet-stg.2019.0258
• Languages: English
• ISSNs: 2515-2947

Abstract

Abstract : Deregulation exposes the inherent volatility of the electricity price. Accurate electricity price forecasting (EPF) could help the market participants to hedge against the price movements and maximise their profits. The existing methods have limited capability of integrating other external factors into the forecasting model, such as weather, electricity consumption and natural gas price. This study proposes a deep recurrent neural network (DRNN) method to forecast day‐ahead electricity price in a deregulated electricity market to explore the complex dependence structure of the multivariate EPF model. The proposed method can learn the indirect relationship between electricity price and external factors through its efficient diverse function and multi‐layer structure. The effectiveness of the method is validated using data from the New England electricity market. Compared with the up‐to‐date techniques, the proposed DRNN outperforms the single support vector machine (SVM) by 29.71%, and the improved hybrid SVM network by 21.04% in terms of mean absolute percentage error.