Real-time Modelling and Parallel Optimisation of a Gasoline Direct Injection Engine

Abstract

With the increasing complexity of engines and number of control parameters, optimal engine parameter sets need to be searched in the high dimensionality. Traditional calibration methods are too complicated, expensive and timeconsuming. The model-based optimisation is of critical importance for engine fuel efficiency improvement and exhaust emissions reduction. The optimisation highly depends on the model accuracy. In this paper, a multi-layer modelling method is proposed, which can be used to generate the engine model at arbitrary operating points in real time with high accuracy. An enhanced heuristic-algorithm-based optimiser is combined with the real-time modelling method to perform a parallel optimisation. The proposed modelling and optimisation strategy can achieve the minimal fuel consumption fast and accurately. This strategy has been successfully verified using experimental data sets.

Online neuro-fuzzy model learning of dynamic systems with measurement noise

“…To avoid time-consuming nonlinear optimisation, the incremental tree-construction algorithm [10] is used to split the input space. The input space defined by the combination of input channels in φ(k) can be treated as a hyperplane.…”

“…The information vector consisting of the weighted regressor of each local model calculated using (4), (10) and ( 11) is given by…”