A Dynamic Configuration Design Method Based on Cloud Model and Deep Reinforcement Learning

Rapidly developing technology and changing market make customer requirement (CR) change rapidly. In the early stage of product development, timely access to CRs is crucial. However, the previous design methods cannot grasp the changes of customers requirement preferences in time and give a better design scheme. Therefore, a dynamic configuration design method based on cloud model dealing with requirement uncertainty and obtaining CR preference is proposed to quickly solve the design scheme. Firstly, the reverse cloud conversion algorithm named Multiple Backward Cloud Transformation based on Sampling with Replacement (MBCT-SR) is used to convert the CR data into a multi-level evaluation cloud model, and the CR preference is calculated. Secondly, the design structure matrix is constructed to map the CR preference to the configuration instance. Then, the Deep Q Network (DQN) algorithm model is established and trained, and the updated CR preference is input to quickly solve the product configuration scheme. Finally, the effectiveness of the proposed method is verified by an example analysis of high-speed train bogies.