Search

Ensemble Meta-learner for Multi-Scale Electricity Price Forecasting

Author(s)

Zhuyi Jiang*

Affiliation(s)

Xi’an Jiaotong Liverpool University, Suzhou, China *Corresponding Author

Abstract

The integration of advanced machine learning techniques with traditional statistical models has become pivotal in addressing the volatility and complexity of modern electricity markets. This study introduces a hybrid framework that synergizes gradient-boosted decision trees (LightGBM) and temporal convolutional networks (TCN) to enhance electricity price forecasting accuracy, efficiency, and robustness. By leveraging LightGBM’s feature interaction optimization through gradient-based one-side sampling (GOSS) and TCN’s multi-scale temporal modelling via dilated causal convolutions, the framework resolves critical challenges in nonlinear dynamics, high-dimensional feature integration, and extreme event adaptability. Evaluated on real-world data from the PJM Interconnection Market (2018–2023) and the 2022 European energy crisis, the proposed LightGBM-TCN model achieves a mean absolute percentage error (MAPE) of 3.51% and reduces training time by 81.5% compared to LSTM baselines. The framework demonstrates superior performance during high-volatility periods, with a 36.7% lower MAPE than traditional wavelet-ARIMA-GARCH hybrids, while enabling real-time inference through parallelized computations. Additionally, the study highlights the broader applicability of ensemble learning in intelligent systems, bridging gaps in domain adaptability and computational scalability. These advancements provide a robust analytical tool for energy market participants, offering actionable insights to mitigate risks in decarbonizing grids and volatile geopolitical landscapes. Future work will focus on uncertainty quantification and decentralized energy system integration to further enhance predictive reliability.

Keywords

Hybrid Machine Learning Models; Electricity Price Forecasting; Multi-Scale Temporal Modelling; Computational Efficiency

References

[1] T. Ahmad, D. Zhang, and C. Huang, “Challenges of integrating renewable energy into power grids: A review,” Renew. Energy, vol. 156, pp. 1072–1085, 2020, doi: 10.1016/j.renene.2020.04.147. [2] R. Jiang, J. Lu, and Y. Xue, “Geopolitical risks and electricity market instability: Evidence from European energy crises,” Energy Policy, vol. 158, p. 112567, 2021, doi: 10.1016/j.enpol.2021.112567. [3] J. Contreras et al., “Day-ahead electricity price forecasting using wavelet transform combined with ARIMA and GARCH models,” Appl. Energy, vol. 87, no. 11, pp. 3605–3612, 2010, doi: 10.1016/j.apenergy.2010.05.012. [4] D. Niu et al., “A hybrid deep learning model for electricity price volatility prediction under renewable energy integration,” Appl. Energy, vol. 332, p. 120537, 2023, doi: 10.1016/j.apenergy.2022.120537. [5] K. Yu et al., “Historical information-assisted dynamic response integration for dynamic multimodal optimization,” IEEE Trans. Evol. Comput., 2025, doi: 10.1109/TEVC.2025.3527478. [6] I. Peña, I. L. Azevedo, and L. A. F. Ferreira, "Economic analysis of the profitability of existing wind parks in Portugal," Energy Economics, vol. 45, pp. 353–363, Sep. 2014. doi: 10.1016/j.eneco.2014.06.013. [7] J. Lago, F. De Ridder, and B. De Schutter, "Forecasting spot electricity prices: Deep learning approaches and empirical comparison of traditional algorithms," Applied Energy, vol. 221, pp. 386–405, Jul. 2018. doi: 10.1016/j.apenergy.2018.02.069. [8] S. Hochreiter and J. Schmidhuber, "Long short-term memory," Neural Computation, vol. 9, no. 8, pp. 1735–1780, Nov. 1997, doi: 10.1162/neco.1997.9.8.1735. [9] J. Lago, F. De Ridder, and B. De Schutter, "Forecasting spot electricity prices: Deep learning approaches and empirical comparison of traditional algorithms," Applied Energy, vol. 221, pp. 386-405, Jul. 2018. doi: 10.1016/j.apenergy.2018.02.069. [10] Ke G, Meng Q, Finley T, et al. Lightgbm: A highly efficient gradient boosting decision tree[J]. Advances in neural information processing systems, 2017, 30. [11] S. Bai, J. Z. Kolter, and V. Koltun, "An empirical evaluation of generic convolutional and recurrent networks for sequence modeling," arXiv preprint arXiv:1803.01271, Apr. 2018. doi: 10.48550/arXiv.1803.01271