基于多输出回归的动力总成悬置系统刚度预测

Stiffness Prediction of Powertrain Mounting System Based on Multiple Output Regression

为能够精确、高效预测汽车动力总成悬置系统刚度从而避免通过反复调校匹配的过程,以某型横置式发动机四点悬置系统为研究对象,提出基于多输出回归算法直接预测动力总成悬置系统刚度的方法并验证了方法的有效性。首先,分别建立了MRTs,MLPR,Multi-SVR三种多输出回归刚度预测模型;随后,以数理型(RMSE、R^(2))和工程型(解耦率、隔振率)评价指标相结合进行了横向比较进而得出MRTs模型用于直接预测动力总成悬置刚度的综合性能最高的结论;最后,纵向对比了多输出回归算法和遗传算法对悬置系统刚度优化结果及工程效果,结论表明:两种算法都能较好满足动力总成悬置系统解耦率控制要求,但多输出回归算法使得悬置系统加权支反力收敛速度更快,同时运算效率显著提高。

In order to accurately and efficiently predict the stiffness of the automobile powertrain mount system and avoid the process of repeatedly adjusting and matching,the paper takes a certain type of horizontal engine four-point mount system as the research object.A method for directly predicting the stiffness of a powertrain mount system based on a multi-output regression algorithm is proposed and the effectiveness of the method is verified.First,three types of multi-output regression stiffness prediction models,MRTS,MLPR,and Multi-SVR,were established.Then,a combination of mathematical(RMSE,R^(2))and engineering(decoupling rate,vibration isolation rate)evaluation indicators was used.The horizontal comparison was carried out to conclude that the comprehensive performance of the MRTS model for directly predicting the mounting stiffness of the powertrain was the highest.Finally,vertically compare the stiffness optimization results and engineering effects of multi-output regression algorithm and genetic algorithm on mount system.The results show that both algorithms can better meet the requirements of the decoupling rate control of the powertrain mount system,but the multi-output regression algorithm makes the weighted support reaction force of the mount system converge faster and the operation efficiency is significantly improved.