The invention discloses a loading robot arm control multipoint mapping intelligent control method and system. The method comprises the first step of conducting grabbing training multiple times for a loading robot arm to obtain a grabbing sample set when a loading robot completes a grabbing task in different fixed grabbing points; the second step of conducting joint classification on the grabbing sample set according to the change amplitude a% of all arm joint control values to obtain various joint sample sets; the third step of using the joint sample sets obtained in the second step to construct a prediction model of loading robot arm joint control values; the fourth step of regarding the distance between a loading robot base and the edge of the bottom end of a grabbing table as an input value, and through a prediction model, obtaining control values of all joints respectively, and outputting the control values to complete the control over the grabbing task. According to the loading robot arm control multipoint mapping intelligent control method and system, by constructing the map between the distance of the robot base and the grabbing platform and a robot arm posture, complicated construction of a kinematical equation is avoided; accurate joint control values are obtained, and the joints are flexibly controlled.