Graph-data-driven co-optimization of powertrain configuration, parameter, and control

Due to the non-Euclidean nature of powertrain configuration topologies, traditional mathematic operators cannot process those non-Euclidean data. Consequently, it is difficult to model constraints and evaluate the performance of multiple configuration design schemes when topology design factors are involved in optimization. To address these challenges, this paper proposes a graph-data-driven co-optimization method for the integrated design of powertrain configuration, component parameters, and control strategies. This study engineers a graph data transform method that learns comprehensive information of powertrain topology and component features that are then converted into a latent space of real numbers for further processing. Subsequently, graph-deep-learning-based classifier and predictor are constructed to achieve constraint modeling and rapid performance evaluation, which provides a data-driven formulation of the powertrain optimization problem and significantly improves optimization efficiency. The result demonstrates a 6.37% reduction in energy consumption compared to the best-performing benchmark scheme (P2). This study pioneers a novel graph-data-driven methodology for the powertrain optimization design of hybrid Electric Vehicles (HEVs).