In the rapid development of the petrochemical industry in the context of the increasing number of large tanks, large tanks in the production and installation process encountered the first problem the pontoon plate welding deformation and weld seam is too large and other defects, through the construction of the ship plate welding special robotic arm of the intelligent perception and decision-making system to solve such defects. Summarize the Structure of crude oil storage tanks and welding process characteristics based on the ultrasonic distance measurement in the distance sensor to obtain the characteristic data set, combined with the distinct data and computer programming language to build the intelligent perception and decision-making system of the unique robotic arm for ship plate welding. They set the relevant parameter values based on simulation experiments on the welding of a particular robot arm perception and decision-making system for the application of example analysis. Compared to the traditional LSTM algorithm, the YOLO-v5 network-driven weld target detection algorithm has a 5.5% higher accuracy and a 9.1% higher success rate in time robustness evaluation. In addition, the welding decision model based on the DDPG algorithm possesses excellent computational accuracy, and the average error of each parameter in the welding decision of the ship plate is controlled within 5%. This study greatly improves the welding quality of the pontoon plate of crude oil storage tanks and provides solid technical support for the successful completion of the project.