Electricity price forecasts using a Curvelet denoising based approach

• We propose a new model incorporating the heterogeneous market structure.
• The geometric multiscale data feature is modeled by Curvelet denoising model.
• The chaotic data characteristic is modeled by Time Delay Embedding algorithm.
• Maximum entropy criteria are proposed to determine Curvelet model parameters.
• The forecasting accuracy improves at the statistically significant level.

Price movement in the electricity market can be viewed as a nonlinear and dynamic system, exhibiting significant chaotic and multiscale characteristics. To conduct more accurate analysis and forecasting, this paper proposes a new Curvelet denoising based algorithm to analyze these characteristics and predict its future movement. We project the original electricity price into its time delay embedding domain to reveal its chaotic characteristics. The Curvelet denoising method is introduced to separate and suppress the noise disruptions in the transformed phase space. Empirical studies using the typical Australian electricity market prices data show that the proposed algorithm demonstrates more robust and superior performance than the traditional benchmark models.