In order to solve the problems of low efficiency and high missed detection rate caused by the difficulty of intelligent inspection method to effectively deal with complex environment, a laser point cloud positioning and early warning method for abnormal state of oil and gas pipeline inspection by intelligent robot was proposed. An intelligent inspection robot is designed, including mechanical rocker arm, sealed cabin and frame structure module. 3D laser scanner was used to collect pipeline data, and 3D laser simultaneous localization and mapping simultaneous local ization and mapping (SLAM), the algorithm of laser radar odometry and mapping system lightweight and ground and optimized lidar odometry and mapping (LeGO-LOAM), in SLAM technology is improved to realize synchronous localization and mapping of robots. Convolutional neural network is combined to evaluate the pipeline status and early warning and grading. The experimental results show that the method can accurately detect the pipeline coating condition, cracks, deformation and other abnormalities, the number of detection is consistent with the actual, the inspection rate and early warning rate are more than 99. 8%, the missing rate and false alarm rate are less than 0. 3%, the path planning is efficient, and the overall inspection performance is excellent.