Abstract: Automation of engine model calibration proceduresis a very challenging task because (1) the calibrationprocess searches for a goal state in a huge, continuousstate space, (2) calibration is often a lengthy andfrustrating task because of complicated mutualinterference among the target parameters, and (3) thecalibration problem is heuristic by nature, and oftenheuristic knowledge for constraining a search cannot beeasily acquired from domain experts. A combinedheuristic and machine learning approach has, therefore,been adopted to improve the efficiency of modelcalibration. We developed an intelligent calibrationprogram called ICALIB. It has been used on a dailybasis for engine model applications, and has reducedthe time required for model calibrations from manyhours to a few minutes on average. In this paper, wedescribe the heuristic control strategies employed inICALIB such as a hill-climbing search based on a statedistance estimation function, incremental problemsolution refinement by using a dynamic tolerancewindow, and calibration target parameter ordering forguiding the search. In addition, we present theapplication of a machine learning program calledGID3$+*$/ for automatic acquisition of heuristic rulesfor ordering target parameters. !13
展开▼