Parameter optimization of ejection device based on BP neural network

In order to reduce the weight and volume occupation of ejection device and improve its buffering performance, the physical and mathematical models for ejection process and buffering process were established with a certain ejection device as the research object, and were solved and simulated by using the Runge Kutta method. The sample points are selected by using the Latin hypercube sampling method, and the solution is run in the simulation program. Then a proxy model between the input and the output is established based on BP neural network. Based on this proxy model, NSGA-Ⅱ multi-objective optimization method is used for optimization. After optimization, comparing with the initial scheme, the weight of the device is reduced by 15.52%, the final buffering speed is reduced by 54.58%, and the maximum buffering acceleration is reduced by 23.15%.