基于BP-LSTM网络的列车脱轨系数预测

Prediction of Train Derailment Coefficient Based on BP-LSTM Network

研究了基于误差逆传播网络(BPNN)和长短时记忆神经网络(LSTM)的组合式预测模型及其在脱轨系数预测中的应用。通过在SIMPACK多体力学仿真软件上建立的车-轨系统仿真场景,形成深度学习训练数据集。使用Python语言在TensorFlow框架上开发单项网络,基于仿真数据集,对单项预测模型BPNN和LSTM展开训练。评估单项模型的预测精度,使用加权平均法对单项模型进行组合,建立组合式脱轨系数预测模型并分析模型预测性能。结果表明,相较于单项模型,组合式预测模型能更准确地预测出脱轨系数的变化趋势。

This paper studies the combined prediction model based on error back propagation network(BPNN)and long short-term memory neural network(LSTM)and its application in the prediction of derailment coefficients.A deep learning training data set is formed through the vehicle-track system simulation scene established on the SIMPACK multi-body mechanics simulation software.Use the Python language to develop a single-item network on the TensorFlow framework,and train the single-item prediction models BPNN and LSTM based on the simulation data set.Evaluate the prediction accuracy of the single model,use the weighted average method to combine the single models,establish the combined derailment coefficient prediction model and analyze the model prediction performance.The results show that the combined prediction model can predict the change trend of derailment coefficients more accurately than the single model.