基于数据建模的铁路弓网接触力辨识

Identification of Railway Pantograph-Catenary Contact Force Based on Data Modeling

在对载荷辨识技术应用情况进行分析的基础上,基于铁路弓网综合检测数据,先结合数据特征进行时域分析、频域分析、趋势项消除及区段融合等预处理,后分别采用机器学习中的BP,ELM和LSTM神经网络3种数据建模方法,分别以4个受电弓振动加速度和2个硬点振动加速度为输入,对弓网接触力进行数据建模载荷辨识,并与试验结果进行对比。结果表明:采用3种神经网络建立的数值模型均可通过受电弓和硬点振动加速度数据辨识出弓网接触力的区段大值;采用4个振动加速度建立的数值模型较采用2个硬点加速度有更好的辨识效果;采用LSTM神经网络建立的数值模型避免了训练过程中的梯度爆炸和梯度消失等问题,具有较高的辨识相关性。

On the basis of analyzing the application condition of load identification technology and the comprehensive inspection data of the railway pantograph-catenary,firstly,time domain analysis,frequency domain analysis,trend item elimination and segment fusion and other pretreatments are carried out combined with data characteristics.Then,three data modeling methods of BP,ELM and LSTM neural network in machine learning are used.Taking the 4 pantograph vibration accelerations and 2 hard spot vibration accelerations as inputs respectively,the pantograph-catenary contact force is identified by data modeling and compared with field test results.The results show that the numerical models established by the three kinds of neural networks can identify the large value of pantograph-catenary contact force through the data of pantograph and hard spot vibration accelerations.The numerical model established with 4 vibration accelerations has better identification effect than that with 2 hard spot accelerations.The numerical model established with LSTM neural network avoids the problems of gradient explosion and gradient disappearance during the training process,and has higher identification correlation.