Self-learning of a fuzzy controller using cell mapping in the sense of a quadratic optimum

[[abstract]]A self-learning method based on the concept of cell mapping is presented to design automatically, a fuzzy controller in the sense of a quadratic optimum for an unknown system. The construction of the system's cell mapping is directly obtained fi om experimental data instead of numerical calculations from system equations. The continuous state and control spaces of the system are first divided into many discrete fuzzy cells. A series of learning signal's is applied to excite the system for training, and then all the stales and control signals are gathered in a data buffer. Then, the real-valued data in the buffer are transformed to the fuzzy-valued cell chains which consist of the next state cell, initial control cell, quadratic cost and degree for optimization. A fuzzy control table is created to store these fuzzy cell chains with less cost and higher degree and is optimized through accumulated learning experience. Finally, the well learned fuzzy control table is used as an optimal fuzzy controller to transfer globally, the system from any initial state cell to the target state cell. Once the state reaches the target state cell, the controller is switched from the fuzzy controller to a state feedback controller to the system within the target state cell.[[fileno]]2020225010070[[department]]動機