Abstract: Based on nonlinear prediction and evaluation, and using the BP neural network model, using genetic algorithms to optimize the network initial weights and thresholds of the network model, a new coal mine inrush water risk prediction is built by collecting different mine water inrush datum and considering a variety of water inrush factors. Using matlab programming to train the network initial datum, and to ananlyze whether the different working face floor inrush water or not and the quantity of water inrush, the results show that the model is fast convergence strong generalization ability, forecast precision.