Low-cost, zero-carbon microgrid electricity price forecasting and dispatch strategy with hybrid hydrogen gas turbines

This paper addresses the optimal scheduling of low-cost, zero-carbon microgrids by proposing a novel ensemble deep learning-based electricity price prediction algorithm, BiLSTM-Adaboost. Under controlled conditions with the same network depth and 500 training rounds, our approach outperforms over ten classical and advanced electricity price prediction methods in terms of accuracy. Building on these high-precision forecasts, we introduce an advanced multi-objective optimization algorithm, A-MOEA/DD, specifically designed for zero-carbon microgrid scheduling. The integrated system encompasses wind and photovoltaic power generation, water electrolysis hydrogen production, hydrogen storage tanks, advanced hybrid hydrogen gas turbines, and carbon sequestration. Focusing on minimizing both overall operating costs and power imbalances, the algorithm executes 1,000,000 evaluations with a population size of 100, yielding 71 Pareto front solutions, thereby surpassing the performance of more than 10 state-of-the-art multi-objective optimization algorithms. Notably, the combined process of electricity price prediction and microgrid scheduling optimization is completed in under 15 minutes, demonstrating its practical applicability as an effective and reliable solution for real-time, low-cost, zero-carbon microgrid operations. To meet this practical need, this study proposes a reliability-oriented decision-making scheme based on the Pareto front. Meanwhile, three extended experimental studies were conducted to further investigate the generalization capability of the proposed prediction algorithm on datasets of varying sizes, and to elucidate the rationale for parameter choices in the prediction and optimization processes.