Artificial Neural Networks Used As a Rule Selector for Fuzzy Logic Controllers
Abstract A feedforward, three-layer, partially-connected artificial neural network (ANN) is proposed to be used as a rule selector for a rule-based fuzzy logic controller. This will allow the controller to adapt to various control modes and operating conditions for different plants. A principal advantage of an ANN over a look up table is that the ANN can make good estimates to fill in for missing data. The control modes, operating conditions, and control rule sets are encoded into binary numbers as the inputs and outputs for the ANN. The General Delta Rule is used in the backpropagation learning process to update the ANN weights. The proposed ANN has a simple topological structure and results in a simple analysis and relatively easy implementation. The average square error and the maximal absolute error are used to judge if the correct connections between neurons are set up. Computer simulations are used to demonstrate the effectiveness of this ANN as a rule selector.