Neuro-Fuzzy Identification of an Internal Combustion Engine

Abstract – Dynamic modeling and identification of an internal combustion engine (ICE) model is presented in this paper. Initially, an analytical model of an internal combustion engine simulated within SIMULINK environment is excited by pseudorandom binary sequence (PRBS) input. This random signals input is chosen to excite the dynamic behavior of the system over a large range of frequencies. The input and output data obtained from the simulation of the analytical model is used for the identification of the system. Next, a parametric modeling of the internal combustion engine using recursive least squares (RLS) technique within an auto-regressive external input (ARX) model structure and a nonparametric modeling using neuro-fuzzy modeling (ANFIS) approach are introduced. Both parametric and nonparametric models verified using one-step-ahead (OSA) prediction, mean squares error (MSE) between actual and predicted output and correlation tests. Although both methods are capable to represent the dynamic of the system very well, it is demonstrated that ANFIS gives better prediction results than RLS in terms of mean squares error achieved between the actual and predicted signals.