AI-Driven LSTM-Copula Hybrid Model for Joint Risk Dependence Modelling in Carbon–Electricity Portfolio Management: Implications for Grid Cost-Effectiveness and Stability

INTRODUCTION: The increasing coupling between carbon emission trading and electricity markets creates significant joint risk challenging grid cost-effectiveness and stability. Existing approaches apply LSTM and Copula models separately, lacking a unified framework capturing both non-linear temporal dynamics and asymmetric tail dependence. OBJECTIVES: This paper proposes an end-to-end LSTM-Copula hybrid model integrating deep learning-based marginal modeling with time-varying Copula dependence estimation for joint risk measurement and optimal portfolio allocation. METHODS: The framework employs LSTM-GARCH for conditional mean and volatility modeling, EVT-GPD for tail fitting, and probability integral transform to obtain uniform variates. A time-varying t-Copula with DCC-type evolution captures dynamic joint dependence. Monte Carlo simulation estimates VaR and CVaR, followed by Min-CVaR portfolio optimization. Empirical analysis uses Chinese carbon and electricity market data (July 2021–December 2025). RESULTS: The LSTM-GARCH model achieves RMSE reductions of 42.2% and 38.0% for carbon and electricity price prediction versus standalone GARCH. The integrated model attains a VaR failure rate of 5.2% at the 95% confidence level, outperforming GARCH-Copula, GARCH-Normal, and Historical Simulation in Kupiec and Christoffersen backtesting. CONCLUSION: The proposed model provides a unified framework for carbon–electricity joint risk modeling, offering insights for AI-driven energy portfolio optimization and grid stability enhancement.