Incorporating physical knowledge about the formation of nitric oxides into evolutionary system identification

Genetic programming (GP) is an evolutionary optimization method that has already been used successfully for solving data mining problems in the context of several scientific domains. For example, the identification of models describing the nitric oxides (NO_x) emissions of diesel engines has been investigated intensively, very promising results were obtained using GP. In the standard GP process, all model structures (as well as parameter settings) of models are created during an evolutionary process; populations of models are evolved using the genetic operators crossover, mutation and selection. In this paper we discuss several possibilities how a priori knowledge can be integrated into the GP process; we have used physical knowledge about the formation of NO_x emissions in a BMW diesel engine, test results are given in the empirical tests section.