Simplicity in dynamic and competitive electricity markets: A case study on enhanced linear models versus complex deep-learning models for day-ahead electricity price forecasting

In the transitioning electricity market of China, accurate forecasting of Day-Ahead Electricity Prices (DAEP) is crucial for strategic planning and profit optimization of market participants. It plays a significant role in resource allocation and in enhancing the efficiency of the energy system. DAEP forecasting in complex electricity markets is challenging due to a multitude of factors, including end-user consumption patterns and physical elements like network losses and transmission congestion. Furthermore, DAEP bidding strategies are often entwined with strategic gaming behavior. Motivated by this, we introduce a novel enhanced linear framework designed to optimize the trade-off between preserving historical patterns (the memory function) and extending predictions to new situations (the generalization function) in DAEP forecasting. The framework employs a linear network to capture data trends and Multi-Layer Perceptron networks for the robust extraction of intricate features and generalization. The proposed enhanced linear framework is developed and evaluated using real-world data from 3 geographically distinct power plants in Guangdong, the province with the highest economic scale and electricity consumption in China. Our approach outperforms representative deep-learning methods, including the Long Short-Term Memory model and Transformer models, with improvements of RMSE up to 26.64% and 51.80%, respectively. Additionally, the results reveal that complex models do not always outperform more straightforward ones in real-world markets characterized by extensive interaction and competition. This indicates the proposed framework provides a straightforward but effective method for time-series DAEP forecasting within the competitive electricity markets. Accurate DAEP forecasting can enhance grid security, facilitate optimal resource allocation, and promote the integration of green and low-carbon power sources into the urban energy system. •The landscape of the transforming Guangdong electricity market is introduced.•An enhanced linear framework is proposed for electricity price forecasting.•The models are developed and validated using real-world data from 3 power plants.•The proposed approach outperforms the state-of-the-art models in the real power markets.•Models with higher complexity do not always ensure better forecasts.