Abstract: Aiming at the uncertainty and the nonlinear during the channel operation, a BP neural network of parallel connection on the state observer loop of channel automatic control system, through learning from the water level of actual measurement, corrects mathematics model for channel operation control system ,and a BP neural network of parallel connection on controller gain loop compensates the control gain, with this kind of method to solve the nonlinear and uncertainty problems in the channel operation control system. Comparing with linear quadratic regulation(LQR) control, the transition process is more smooth, and the time to reach steady is shortened, the vibration of gate operation is mitigated, and the excessive regulation is reduced considerably.