Authors

Abstract

Optimizing parameters is a crucial step in designing mechanical structures and a primary means of raising equipment efficiency. This paper proposes a multi-parameter optimization technique that combines an improved genetic algorithm(IGA) and ensemble machine learning(EML)to optimize a licorice harvester's work and structure parameters. The EML model is trained using a small sample dataset built on the coupled DEM-MBD (Multi-body Dynamics Coupled Discrete Element Method) simulation model. The impact of base learner diversity and quantity on the model's prediction accuracy is investigated. Using EML and IGA, the parameters of a licorice harvester are optimized. It is also contrasted with conventional response surface model(RSM) parameter optimization techniques. The study results show that the EML with KNN +lightGBM + catBoost as the base learner and linear as the meta-learner has an R2 of 0.959, MAE of 0.048, and RMSE of 0.06. In comparison to the RSM, EML-IGA reduces resistance by 18.16% and specific power consumption by 21.33%; in comparison to the EML and Pre-improvement genetic algorithm(PIGA), it reduces resistance by 11.36% and specific power consumption by 11.19%. It provides a reference for intelligent parameter optimization methods.

Abstract in Chinese

参数优化是机械结构设计过程中必不可少的环节，也是提高机械工作效率的主要途径之一。本研究通过集成学习与改进遗传算法结合提出一种多参数优化方法对甘草收获机的结构和工作参数进行优化。基于DEM-MBD耦合仿真模型构建小样本数据集对集成学习模型进行训练，并探究基学习器的数量与多样性对集成学习模型预测精度的影响。利用集成学习结合改进遗传算法对甘草收获机的多个参数进行优化。并与传统的响应面参数优化方法进行对比。研究结果表明，以KNN+lightGBM+catBoost为基学习器，线性拟合为元学习器的集成学习模型，其R2为0.959，MAE为0.048，RMSE为0.06。其相较于改进前的遗传算法的优化结果，阻力降低11.36%，比功耗降低11.19%，相较于传统的响应面分析法，阻力降低18.15%，比功耗降低21.33%，为智能化参数优化方法提供参考。