Artificial Intelligence Techniques for the Estimation of Direct Methanol Fuel Cell Performance

Artificial neural networks and neuro-fuzzy inference systems are well known artificial intelligence techniques used for black-box modelling of complex systems. In this study, Feed-forward artificial neural networks (ANN) and adaptive neuro-fuzzy inference system (ANFIS) are used for modelling the performance of direct methanol fuel cell (DMFC). Current density (I), fuel cell temperature (T), methanol concentration (C), liquid flow-rate (q) and air flow-rate (Q) are selected as input variables to predict the cell voltage. Polarization curves are obtained for 35 different operating conditions according to a statistically designed experimental plan. In modelling study, various subsets of input variables and various types of membership function are considered. A feed -forward architecture with one hidden layer is used in ANN modelling. The optimum performance is obtained with the input set (I, T, C, q) using twelve hidden neurons and sigmoidal activation function. On the other hand, first order Sugeno inference system is applied in ANFIS modelling and the optimum performance is obtained with the input set (I, T, C, q) using sixteen fuzzy rules and triangular membership function. The test results show that ANN model estimates the polarization curve of DMFC more accurately than ANFIS model.