A deep reinforcement learning for joint short term electricity load and price forecasting in China's smart grids: Evidence from energy economics, market volatility and Financial risk management
Energy Strategy Reviews, vol.66, 2026 (SCI-Expanded, Scopus)
- Publication Type: Article / Article
- Volume: 66
- Publication Date: 2026
- Doi Number: 10.1016/j.esr.2026.102322
- Journal Name: Energy Strategy Reviews
- Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC, Directory of Open Access Journals
- Keywords: Deep reinforcement learning, Electricity price forecasting, Financial risk Management, Short-term load forecasting
- Azerbaijan State University of Economics (UNEC) Affiliated: Yes
Abstract
The increasing integration of renewable energy, demand-side flexibility, and market-based electricity trading has intensified the complexity of short-term load and price dynamics in China’s smart grids, where forecasting errors can generate substantial operational inefficiencies, market volatility, and financial risk exposure. This study develops a deep reinforcement learning framework for joint short-term electricity load and price forecasting by integrating grid-side demand behavior, wholesale price fluctuations, renewable generation uncertainty, macro-energy indicators, and financial risk signals into a unified predictive architecture. Unlike conventional forecasting models that treat electricity load and price as independent time-series problems, the proposed framework captures their dynamic interdependence through an adaptive learning mechanism that continuously updates forecasting decisions under changing market and system conditions. The model combines deep neural feature extraction with reinforcement learning-based policy optimization to improve multi-horizon prediction accuracy while accounting for volatility clustering, peak-demand shocks, price spikes, and risk-sensitive decision environments. Using high-frequency electricity market and smart grid data from China, the study evaluates the proposed model against benchmark approaches including ARIMA, support vector regression, random forest, long short-term memory networks, gated recurrent units, and hybrid deep learning models. The empirical results are expected to show that the deep reinforcement learning framework achieves superior forecasting performance in terms of MAE, RMSE, MAPE, and directional accuracy, particularly during periods of high demand variability and electricity price instability. Further analysis links forecasting improvements to energy economics and financial risk management by demonstrating how accurate joint prediction can support optimal dispatch, demand response planning, hedging strategies, grid investment decisions, and electricity market risk mitigation. The study contributes to the literature by providing an integrated forecasting framework that connects smart grid intelligence with energy market volatility and financial risk governance in the context of China’s evolving power sector reform. The findings offer practical implications for grid operators, electricity retailers, energy investors, and policymakers seeking to enhance market efficiency, improve system reliability, and manage financial exposure under the transition toward low-carbon and digitally coordinated energy systems.