Day-ahead electricity price forecasting employing a novel hybrid frame of deep learning methods: A case study in NSW, Australia

Day-ahead electricity price forecasting plays a vital role in electricity markets under liberalization and deregulation, which can provide references for participants in bidding strategies, energy trading, and risk management. However, due to various uncertain factors, electricity prices often exhibit nonlinearity, randomness, and volatility, adding technical difficulties to accurate price forecasting. To address these difficulties, A novel hybrid deep learning-based model named convolutional neural network+stacked sparse denoising auto-encoders is proposed first. Moreover, the improved complete ensemble empirical mode decomposition with adaptive noise, a decomposition method, is introduced to enhance model performance by the decomposition of complex data sequences. Each intrinsic mode function sub-component obtained by decomposition is separately predicted using the proposed hybrid model, and the forecast result of day-ahead prices is superimposed finally. Taking the Australian national electricity market as a case study, the experimental results verify that the proposed hybrid model can effectively improve prediction accuracy and stability, and shows outstanding prediction performance for price spikes. Furthermore, the proposed model can save training time for neural networks in the prediction process thanks to its faster convergence speed. Hence, the proposed deep learning-based hybrid predictive model can provide a technology-based reference for electricity market participants.

Automated summary

Day-ahead electricity price forecasting is crucial in liberalized and deregulated electricity markets, providing references for bidding strategies, energy trading, and risk management. However, due to uncertain factors, electricity prices exhibit nonlinearity, randomness, and volatility, posing challenges for accurate forecasting. This study proposes a novel hybrid deep learning model, combined with improved decomposition and prediction techniques, which effectively improves prediction accuracy and stability in the Australian electricity market, showcasing outstanding performance in predicting price spikes. Additionally, the proposed model saves training time for neural networks due to its faster convergence speed, making it a valuable technology-based reference for electricity market participants.