Calculating optimized IC-engine control maps using hinging hyperplane trees

The progress towards near-zero emission vehicles (NZEV) requires very sophisticated techniques to be applied to modern combustion engines. With the rising amount of engine settings, it becomes a more demanding task to find the combination of engine setting values which leads to an overall optimal performance of the vehicle where low consumption, low emissions, and good drivability often are opposing one another. Therefore, it seems very favorable to employ mathematical optimization algorithms. Within the short time frames, the required mathematical emission models can hardly be derived from first principles with satisfying accuracy. Neural models represent an alternative for this modeling task. This paper presents an optimization environment basing on special neural models which led to significant consumption/emission improvements compared to the production car performance.