Novel deep hybrid model for electricity price prediction based on dual decomposition

Abstract

Electricity price (𝐸𝑃 ) forecasting is vital for effective market operation, strategic planning, and risk management in deregulated energy systems. However, the inherent volatility and complexity of electricity prices, shaped by demand supply dynamics, weather variability, and regulatory interventions, pose substantial challenges to accurate prediction. This study introduces a novel hybrid framework designed to improve forecasting accuracy by leveraging both signal decomposition and deep learning techniques. Specifically, the method integrates Variational Mode Decomposition (VMD) and Empirical Wavelet Transform (EWT) for noise reduction and feature extraction, followed by a Multi Resolution Convolution (MRC) layer and a Bidirectional Long Short Term Memory (BiLSTM) network to capture multiscale temporal patterns in electricity price data. The model is applied to half hourly electricity price data from South Australia spanning January 2018 to December 2022. Its performance is benchmarked against a suite of traditional and hybrid models using a comprehensive set of twelve evaluation metrics. The results reveal that the proposed hybrid model consistently outperforms all baselines across seasons and forecast horizons. Notably, during the spring period, it achieved a Normalized Root Mean Square Error of ≈ 4.87 %, a Mean Absolute Percentage Error of ≈ 12.09 %, and a Global Performance Index of ≈ 3.22. These improvements demonstrate the model’s ability to effectively handle the non-linear and nonstationary nature of 𝐸𝑃 . Overall, the findings underscore the potential of combining advanced decomposition methods with deep learning architectures to deliver more accurate and robust 𝐸𝑃 forecasts, thereby offering valuable support for decision making in complex and evolving energy markets.