“The application of large-scale models in the field of power grid regulation faces challenges such as adaptation difficulties and illusion risks. The research team deeply integrates domain knowledge and large-scale model capabilities, designs an integrated collaborative strategy of "domain low rank adaptive fine-tuning - retrieval enhancement, knowledge generation enhancement - optimization ranking", constructs a double-layer hybrid retrieval architecture for power grid spatiotemporal - equipment strong constraints, and establishes a triple reliability guarantee mechanism of "evidence threshold manual". The accuracy of answers in high-risk scenarios reaches 98.1%, providing a feasible engineering solution for the safe and controllable application of large-scale models in high-risk scenarios of power grid regulation.”