改进LSTM的脱轨系数预测方法  

Derailment Coefficient Prediction Method Based on Improved LSTM

作　　者：张卜 刘怡伶 张文静 刘学文 ZHANG Bu;LIU Yi-lin;ZHANG Wen-jing;LIU Xue-wen(School of Mechanical and Automobile Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)

机构地区：[1]上海工程技术大学机械与汽车工程学院,上海201620

出　　处：《软件导刊》2022年第2期27-31,共5页Software Guide

基　　金：国家自然科学基金项目(51675324);上海市高校教师资助计划项目(ZZGCD15039);上海新能源汽车振动噪声测试与控制专业技术服务平台项目(18DZ2295900)。

摘　　要：研究基于长短时记忆神经网络模型的优化方法及其在脱轨系数预测中的应用,通过SIMPACK建立列车—轨道仿真场景得到网络训练所需数据集,构建基于长短时记忆神经网络的脱轨系数预测模型,借助动态学习率和Dropout方法针对学习率及网络结构进行优化,并使用优化后的长短时记忆神经网络对脱轨系数进行预测。脱轨系数预测结果表明,经过优化的长短时记忆神经网络预测模型在测试集上的预测误差相较优化前的模型减小23.9%,动态学习率和Dropout方法能高效地优化预测模型,可使模型较准确地预测出脱轨系数变化趋势,可为进一步研究提供支持。Research the optimization method based on long and short-term memory neural network model and its application in the prediction of derailment coefficient. Establish train-track simulation scenarios through SIMPACK to obtain data sets required for network training,build a prediction model of derailment coefficients based on long and short-term memory neural networks,use dynamic learning rate and dropout methods to optimize learning rate and network structure,and use the optimized The long and short-term memory neural network predicts the derailment coefficient. The prediction results of the derailment coefficient show that the prediction error of the optimized long-and short-term memory neural network prediction model on the test set is reduced by 23.9% compared with the model before optimization. The dynamic learning rate and dropout method can efficiently optimize the prediction model. The model accurately predicts the change trend of the derailment coefficient,which can provide support for further research.

关 键 词：脱轨系数预测 深度学习 动态学习率 DROPOUT 

分 类 号：TP389.1[自动化与计算机技术—计算机系统结构]