Data-Driven Design of Takagi-Sugeno Fuzzy Systems for Predicting NOx Emissions

New emission abatement technologies for the internal combustion engine, like selective catalyst systems or diesel particulate filters, need of accurate, predictive emission models. These models are not only used in the system calibration phase, but can be integrated for the engine control and on-board diagnosis tasks. In this paper, we are investigating a data-driven design of prediction models for NOx emissions with the help of (regression-based) Takagi-Sugeno fuzzy systems, which are compared with analytical physical-oriented models in terms of practicability and predictive accuracy based on high-dimensional engine data recorded during steady-state and dynamic engine states. For training the fuzzy systems from data, the FLEXFIS approach (short for FLEXible Fuzzy Inference Systems) is applied, which automatically finds an appropriate number of rules by an incremental and evolving clustering approach and estimates the consequent parameters with the local learning approach in order to optimize the weighted least squares functional.