基于GBDT算法的弓网动态匹配特性预测模型

A study on prediction model of dynamic matching characteristics of pantograph-catenarysystem based on the GBDT algorithm

高速铁路通过弓网系统的滑动电接触获取电能驱动列车运行,弓网动态匹配特性是保障良好滑动电接触的基础。首先,建立了弓网动态匹配的有限元分析模型,并通过与文献结果对比验证了模型的正确性。采用拉丁超立方抽样方法,对接触网的关键结构参数和运行速度参数进行样本抽样,获得输入参数集;然后,利用有限元模型对输入参数集开展大量计算分析并进行结果的特征提取,获得弓网动态匹配关键特征参量的输出结果,结合输入和输出结果,构成了样本数据集;最后,采用梯度提升决策树(gradient lifting decision tree, GBDT)算法对数据集进行学习训练和验证测试,建立弓网动态匹配特性预测模型,并将其与基于决策树、随机森林、极端随机树和极端梯度提升树算法的4个模型进行对比分析。结果表明,基于GBDT算法的模型预测精度更高、稳定性更好,在测试集上的R~2达到了0.929,能够准确快速地评估弓网匹配特性。通过对GBDT模型进行参数重要性分析可知,运行速度对弓网匹配特性的影响程度最大,达61%,其次是接触线的张力、承力索张力和档距。该研究初步探索了采用机器学习方法建立预测模型来替代有限元模型的可能性,所建立的模型可用于弓网动态匹配特性的快速预测与评价。

High-speed railway acquires electric energy to drive the train through the sliding electric contact of pantograph-catenary system.The dynamic matching characteristics of pantograph-catenary system is the basis of ensuring good sliding electric contact.In this study,firstly,a finite element analysis model to simulate dynamic interaction of pantograph-catenary system was established,and its validity was verified by comparing with the results in the literature.Then,the Latin hypercube sampling method was used to sample the main structural parameters of catenary as well as operating speed parameter,and the input parameters were obtained.Using the finite element model,a large number of calculation and analysis of the input parameter set were carried out and the results were extracted,and the output results of the key evaluation indexes of the dynamic matching characteristics were obtained.The input and output results are combined to form the sample data set.Finally,the Gradient Lifting Decision Tree(GBDT)algorithm was used to learn and train the dataset,based on which the prediction model of dynamic matching characteristics of pantograph-catenary system was obtained.The model was compared with the four other models,i.e.random forest,extreme random tree and extreme gradient lifting tree algorithm.The results show that the prediction accuracy of the GBDT-based model is higher and the stability is better.The R2 on the test set reaches 0.929,which can accurately and quickly evaluate the dynamic matching characteristics.The parameter importance analysis of GBDT model shows that the operation speed has the greatest influence on the contact quality,which is 61%,followed by the tension in contact wire,the tension in messenger wire and the span length.This study explores the possibility of using machine learning methods to establish prediction models instead of finite element models,and the established models can be used for rapid prediction and evaluation of the dynamic matching characteristics of pantograph-catenary system.