According to the shortcomings of slower convergence speed and easy to fall into the local optimal of BP neural network, a method that weights and thresholds of BP neural network are optimized by genetic algorithm which has the global search ability is put forward, using the optimized BP network by genetic algorithm, the simulation experiments are conducted using a lot of sensor data of modern strip mill hydraulic automatic gauge control system, experiments show that this algorithm has obvious superiority, the shortages of BP algorithm are avoided, the learning performances of network are improved greatly, and it has certain practicability.