Aiming at the problem of that the rapid-exploration random trees(RRT) algorithm is difficult to solve the rapid motion planning of a manipulator in the multi-scene environment, a fast motion planning algorithm combining long and short time memory mechanism is proposed. Firstly, the paper uses the Gaussian mixture model(GMM) to build a scene model of the environment through random sampling in the initial stage of planning, and uses this model for collision detection to improve the efficiency of motion planning. Then, according to the principle of human memory mechanism, different GMMs for multiple scenes are stored according to real-time memory, short-term memory and long-term memory, and the scene matching algorithm is used to achieve fast adaptive extraction of different scene GMM to improve the adaptability to changing environments. Finally, the proposed algorithm is validated by the grab simulation experiment of the 6-DOF flexible manipulator in Matlab and ROS simulation environment. Experimental results show that the algorithm can quickly extract the memory information of the scene, effectively improve the efficiency of motion planning in multi-scene environment, and has strong adaptability.