This paper proposes a neural network-based integrated reactive power optimisation method for power grids containing large-scale wind power, which improves the generalisation capability of the neural network by constructing a typical wind-load scenario, and solves the integrated reactive power optimisation problem for a typical wind-load scenario after it is connected to the system using the Harris Hawk Optimisation algorithm (HHO) The method is designed to reduce the computational effort and decision time of reactive power optimization by deeply fitting the mapping relationship between grid operation state and comprehensive reactive power optimization strategy through neural networks.