Near Boundary Control of Automotive Engine using Machine Learning

We aim to develop a control method for an automotive engine to operate inside the boundary that present sbetween the admissible and violated operation domain. These boundaries can be represented by a mathematical model. For those boundaries, we design a near-boundary control using State-Dependent Riccati Equation, which performs to make the engine work to satisfy the driver demand when the engine operates inside the admissible zone, and which performs to make the engine not exceed the boundaries when the driver demand indicates inside the violated zone. Such a controller requires high computational performance so that it is practically impossible to be implemented into a real machine. In this paper, a deep neural network is introduced to mimic the behavior of the predesigned near-boundary control, which can work with a much-reduced computational load.