Abstract: Laser welding of thick section plate is an effective measure to realize the lightweight of automobile,but in the process of laser welding,the penetration state of weld is inconsistent due to unstable process,and the welding quality of workpiece is poor.Because the laser power is easy to control and has a significant influence on the penetration depth,this paper designs the power control software and hardware system and completes the construction of the power control platform from the power control requirements in the laser welding process of the variable thickness cross-section plate.Convolutional neural network is used to classify the images of small holes obtained by coaxial image acquisition system,which are divided into incomplete penetration,moderate penetration and excessive penetration according to the area of the through holes,and PID control system is used to change the welding power in real time according to the classification results to obtain the moderate penetration state.Finally,the adaptive control experiment of laser welding power is carried out by using variable cross-section panel,and the weld with uniform back residual height is obtained,which verifies the effectiveness of the adaptive power monitoring and control algorithm.