Abstract The centralized operation of wind turbines impacts on the grid’s security and stability operation. For rational planning efficient operation of various power supply unit, the power load forecasting precision has been put forward higher requirements. The time series of grid has chaotic characteristics and it is difficult to describe its characteristics and inherent laws. We take advantage of the chaotic phase space reconstruction theory to study the power load time series sample data. Time delay and embedding dimension were obtained using the mutual information method and the CAO. Lyapunov exponent of this system is obtained so that we know the grid system has chaotic characteristics. Then reconstructed the phase space according to the time delay and embedding dimension. On the basis of phase space reconstruction, used support vector regression algorithm to predict the power load. The grid search method was used for for parameter optimization. Finally, the predicted results with the time series prediction model and BP neural network model were compared. The results show that this is a small error, high precision load forecasting method.