Self-learning neurofuzzy controller

A self-learning fuzzy logic system is given for control of unknown multiple-input-multiple-output (MIMO) plants. A concise formulation of fuzzy controllers for MIMO plants is presented. Through new terminology and data types, relations among the crisp input vector, the fuzzy basis set for all linguistic input variables, the cardinality vector of fuzzy partitions in all input universes of discourse, the rule base linguistic value set, and the fuzzy inference action vector are established. The fuzzy controller can be cast into neural net structure. The integration of fuzzy logic and a neural network takes advantage of fuzzy data representation, fuzzy inference, parallel processing, and learning ability. The random optimization method is used to train the controller. The training process uses observations of plant input and output behavior, so that a model of the plant is not required.