Wind power, as a new energy generation technology, has been applying widely and growing rapidly, which make it become the main force of renewable energy. However, wind speed sequence has its own character of the intermittent and uncertainty, which brings a great challenge to the safety and stability of the power grid, one of the valid ways solving the problem is improving the wind speed predicting accuracy. Therefore, given atmospheric disturbances, we firstly used empirical mode decomposition (EMD) to deal with the non-linear wind speed sequence, and combined with strong adaptive and self-learning ability of BP neural network, then, a wind speed prediction model, EMD-BP neural network based on Lorenz disturbance, was proposed. Finally, it was to made use of actual wind speed data to take a simulation experiment and explored the improvement effect of the preliminary forecasting sequence of wind speed influenced by Lorenz equation in the transient chaos and chaos. The results show that, the improved model weakened the random fluctuation of wind speed sequence, effectively corrected the wind speed sequences initial prediction values, and made a great improvement for the short-term wind speed prediction precision. This research work will help the power system dispatching department adjust the dispatching plan in time, formulate the wind farm control strategy reasonably, reduce the impact brought by wind power grid connection, increase the wind power penetration rate, and then promote the global energy power market innovation.