Intelligent dispatch method for power systems based on LLM knowledge injection

This paper proposes an intelligent dispatch method for power systems based on the injection of knowledge from large language models. First, data related to the operation of the power system, including system states, environmental factors, and historical information, are collected. Then, the power system dispatch is abstracted as an optimization problem under multiple constraints, where the objective function considers cost, renewable energy utilization, and carbon emissions. The constraints include load balance constraints and operational equipment constraints. Furthermore, under the open-source large model framework, state space, action space, and target reward functions specific to the power system are designed. Finally, the unknown environment is learned through large model knowledge injection, with continuous iterative self-training to gain experience and obtain the optimal power system dispatch strategy. Experimental results show that the intelligent dispatch large model trained by this method, combined with large model knowledge injection, can effectively solve major issues such as low control accuracy and high latency in traditional power dispatching.