Dual-Agent LLMS with Genetic Evolution for Automated Bidding Strategy Optimization in Electricity Markets

As renewable energy penetration increases, electricity markets face unprecedented volatility that challenges traditional bidding strategies for power generators. However, traditional approaches struggle to adapt to the dynamic nature of renewable-driven price fluctuations and market uncertainty. To deal with these problems, this paper proposes EvoBid, a dual-Agent large language model (LLM) framework integrated with genetic evolution filtering for automated bidding strategy optimization. The framework separates market analysis from code generation through specialized agent coordination, where a modeling agent performs scenario interpretation and strategic planning, while a coder agent translates insights into executable optimization algorithms. A genetic evolution filtering mechanism systematically evaluates, selects and refines candidate strategies across multiple generations using a comprehensive fitness function that balances technical feasibility, economic performance, and strategic soundness to retain elites and repopulate candidates. Experimental validation on 30 representative operating days demonstrates that the system employing EvoBid consistently boosts execution reliability and economic performance, with dual-agent specialization enhancing feasibility and stability while genetic evolution further refines biddings and suppresses outliers for superior, reproducible outcomes. The proposed framework effectively addresses renewable energy volatility challenges and provides a practical solution for automated bidding in dynamic electricity markets.