Optimization of Vehicle Structures based on Improved Deep Forest and Grey Wolf Optimizer

Optimization design of the vehicle thin-walled structures (TWS) is one of the critical ways to reduce structural mass and enhance the mechanical properties of the whole body. Traditional numerical simulations are effective and have been widely used to solve this problem. However, carrying out a high-fidelity simulation is really time-consuming and the optimization design heavily relies on engineer's experience. In this work, a deep regression forest model (DRF) is used to approximately simulate the TWS physical system, and through an improved grey wolf optimizer (AIGWO), the optimization design is adaptively realized. Specifically, a multi-grained circular scanning and pooling-based cascade forest are used to build the DRF model. The mathematical expression of the DRF is generated. Compared with five state-of-the-art machine learning-based models, the DRF shows the best accuracy. Subsequently, a reward strategy is proposed to enable AIGWO to adaptively determine its optimization actions. Three classic TWS vehicle parts are optimized. The results demonstrate that the AIGWO outperforms other four advanced swarm intelligence algorithms. The structural mass of three TWS cases are reduced and the mechanical properties are enhanced. The whole computational time based on the proposed method is highly shortened compared with numerical simulation. In conclusion, the proposed method can efficiently realize the optimization design of vehicle TWS problem.