Development of Neuro-Fuzzy Controller for Applications to HVAC System, Inverted Pendulum and Other Processes

Abstract — Linguistic modeling of complex nonlinear systems constitutes the heart of many control and decision-making processes and fuzzy logic represents one of the most effective algorithms to build such linguistic models. In the first part of this paper, only thirty-six fuzzy rules are developed using fuzzy triangular membership function. The objective of using less number of fuzzy rules is leading to a smaller amount of computational time. In the second part few of these rules are used to train a neural network by back propagation neural network (BPNN) to develop a neuro-fuzzy controller (NFC). In the designed three layer neural network, the first layer contains two neurons for two inputs: error (e) and change of error (∆e) respectively, whereas the second layer or the hidden layer contains thirty neurons and the third layer use only one neuron to get the output (u) value. The third part of the paper is used to study the performance of neuro-fuzzy controller for a linear and a nonlinear process respectively. In the pilot study part proposed NFC is applied to an Inverted Pendulum and to the supply air pressure control loop of Heating, Ventilation and Air-conditioning (HVAC) system. The tuning procedure of the proposed NFC is very easy. Simulation results show that the Neuro-Fuzzy controller performances are comparable to the other methods under normal conditions. Copyright c © 2008 Yang’