A neuro fuzzy control algorithm using a new extrapolation methodology

Describes a new procedure to implement a fuzzy control algorithm, based on an approach to the well-known “fuzzy model-reference learning control” (FMRLC) algorithm, and the use of an artificial neural net (ANN) to improve its initialization. In this approach, according to the spreading matrix (extrapolator), once the current fired rules have been adapted, the adaptation is extrapolated to their neighbors. The non-explored rules near the explored ones are then prepared to be used in the future. The final rule matrix for a given input reference to the plant (a tanker in this case) could then be considered as the desired output for this input reference. So, the whole experience gained by the control could be learned by an ANN, which may be used as an expert to give a rule base initialization to be used in later situations. Various simulations and their results are shown and compared, with and without the ANN, as an example of the performance of this algorithm