A learning approach to low carbon autonomous vehicles and its applications on real time energy management of downsized hybrid diesel engines 2019-01-1056
Autonomous vehicles have attracted global interest because of the great potentials in reducing traffic accidents and improving passengers comfort. Meanwhile, more stringent legislation on vehicle fuel economy and exhaust emissions are being carried out. Developing low carbon autonomous vehicles is a critical theme for future automotive industry but also faces strong technical challenges. This paper tries to address these challenges from a machine learning based optimisation approach. Employing vehicle onboard sensors, the future driving conditions can be predicted. The predicted information is utilized by machine learning methods to model the behaviour of propulsion systems and thereafter, to optimise the fuel economy. In this paper, we applied this approach on an autonomous vehicle driven by a diesel engine which is downsized by an electrically assisted turbocharger (EAT).
Engine downsizing using advanced turbocharging techniques is a promising solution to improve vehicles energy efficiency. An EAT can improve the transient performance at low speeds and recover surplus exhaust gas energy at high speeds. A high efficient control solution of the electrified turbocharged diesel engine (ETDE) is the key to maximise its benefits in engine downsizing and energy recovery. However, the model-based controller design of the ETDE is challenging because the high nonlinearity of engine model. In this paper, we employed the machine learning method, in terms of the local linear model tree (LOLIMOT) to build a virtual engine. The future engine operating points are predicted by onboard sensors, while the corresponding engine models are rapidly linearized and explicit model predictive controllers (EMPC) are built. As a result, the optimal control laws can be generated in a time-rolling sequence. A higher level controller is built to manage the battery state-of-charge (SOC) onboard in real time. Therefore, the engine can always run at its highest efficiency with proper maintenance on the battery SOC level.