To accurately assessing the demand response potential of meteorologically sensitive loads in the power grid, an assessment method for the demand response potential of meteorologically sensitive load is proposed based on the fusion of mechanism-data hybrid driven methods. At first, the meteorologically sensitive load power at the grid level is estimated based on a data-driven two-branch neural network model. Secondly, based on the equivalent thermal parameters model, a physical aggregation model of meteorological sensitive loads is established. With the goal of minimizing the error between the aggregate power and the estimated power, the control variable method and the particle swarm optimization (PSO) algorithm are used to optimize the equivalent parameters of the physical aggregation model, including the number of temperature control load devices and the temperature setting values. Finally, the user’s comfort and willingness are considered to evaluate the demand response potential of meteorologically sensitive loads. The case analysis shows that the meteorologically sensitive load power estimated by the proposed method has a significant correlation with temperature, and the obtained equivalent parameters of the physical model and demand response potential are consistent and reasonable with the actual situation.