Improving Long-Term Electricity Price Predictions: A Comparison of Pure GPR and Hybrid LP-GPR Models

The capital-intensive nature of renewable energy sources calls for robust assessments of expected revenues to mitigate investment risks and, in turn, lower their cost-of-capital. Consequently, long-term electricity price prediction has become an important support for decision-making in the electricity sector. This paper presents a comparative study of two electricity price dynamics models for long-term applications. The first approach relies on Fourier decomposition and Gaussian Process Regression (GPR), using only yearly regressors values as inputs. The second model combines linear-polynomial (LP) and GPR components to separately capture supra-weekly and sub-weekly dynamics, assessing whether annual regressors variations influence supra-weekly dynamics. Both models adopt Italian dayahead electricity prices for training (2011-2023) and testing (2024). The results shows that the hybrid LP-GPR model seems to achieve better accuracies than the pure GPR model, thereby demonstrating the need for a trade-off between high-time granularity input data and model simplicity.