基于APSO_ELM算法的地铁列车碳滑板磨耗的预测

Prediction of Carbon Slide Wear of Subway Train Based on APSO_ELM Algorithm

为了准确预测地铁列车碳滑板磨耗量,本文选择训练速度快、参数设置少、准确度高的极限学习机(ELM)模型。针对ELM模型在训练过程中随机产生权值和阈值,导致模型泛化能力不足、稳定性差等缺点,引入基于收缩因子改进的自适应粒子群算法(APSO)对其进行优化,提出了一种基于自适应粒子群优化极限学习机(APSO-ELM)的碳滑板磨耗预测模型。将该模型运用到碳滑板磨耗实例预测中,在选取的270组样本数据中,前235组作为训练样本,后35组作为测试样本,以影响碳滑板磨耗的主要因素——地铁运行公里数作为输入参数,以碳滑板厚度为输出参数,将预测结果与ELM模型预测进行对比。结果表明,APSO-ELM模型有较高的预测精度,预测值更逼近于实际值,验证了APSO-ELM模型在碳滑板磨耗预测中的可靠性和有效性。

In order to accurately predict the wear of subway train carbon slide,extreme learning machine(ELM)model with fast training speed,less parameter setting and high accuracy was selected.In view of the ELM model randomly generated in the pro⁃cess of training the weights and thresholds,leading to poor stability of model generalization ability insufficiency,the weaknesses,in⁃troduced based on contraction factor improved adaptive particle swarm optimization(APSO)for its optimization,put forward a kind of extreme learning machine based on adaptive particle swarm optimization(APSO-ELM)carbon slide abrasion prediction model.To apply the model to predict carbon slide abrasion instance,in the selection of 270 groups of sample data,the first 235 group as the training sample,after 35 group as the test sample,in the subway running mileage as the main factors influencing the carbon slide abrasion for input parameters,output parameters by carbon board thickness,comparing the prediction results and the ELM model prediction,the results show that the APSO-ELM model has higher prediction accuracy,and the predicted value is closer to the actual value,which verifies the reliability and validity of APSO-ELM model in carbon slide wear prediction.