In many industrial control applications, different controller structures and parameters should be used for different operating scenarios. These changes can be due to plant or environmental variations. The classical control approaches are either the use of some sort of gain scheduling to cover different operation modes, or to adapt the controller parameters according to the process behavior. In any case, the learning capabilities are basically restricted to select one among some predefined options or to adapt the current controller. In the paper, a learning control structure is used to avoid the load swinging of a crane, such as the one used in harbor installations. There, an expert human operator is usually required because of his experience in the selection of proper crane input commands. The operator knows how to transport different loads over different length trajectories to the destination with few oscillations, in addition to compensate unmeasurable disturbances, like the wind force and others. A similar automated learning process is proposed here.