This paper presents an analysis of the exhaust gas concentration of a compression ignition engine powered by diesel fuel and rapeseed oil under dynamic conditions. The measurement cycle consisted of a 100 s segment of the WLTC cycle. An attempt was then made to estimate the exhaust gas concentration using predictive algorithms based on parameters recorded using the OBD-II diagnostic interface. The model was validated based on previously unobserved measurements of the measurement cycle, and the procedure was repeated several times with random parameter changes. Due to the dynamic nature of the combustion process (taking into account its non-linearity and inertia), a delayed feature design was used. A consistent time horizon of input information was selected for the tabular and sequential models used. The results obtained indicated that Gradient-Boosted Regression Trees class algorithms achieved the highest quality of fit and were characterised by the greatest stability.
Kuszneruk et al. (Thu,) studied this question.