The invention discloses a sleeper service life prediction method based on time series data feature fusion, which comprises the following steps: constructing a sleeper service life prediction model to realize sleeper service life value prediction and sleeper state monitoring, and establishing an intelligent sleeper platform system based on the model; the model comprises the following steps: constructing an LSTM for each sensor, combining features output by all intermediate layers, and outputting comprehensive features of the sensor through a full-connection layer; inputting an intermediate layer result output by the LSTM into Transform to obtain weight distribution corresponding to the features among the sensors; and inputting data obtained by the LSTM and the Transformer as a comprehensive time sequence data feature set into the random forest model for training to obtain a final sleeper life prediction model. The method can be applied to the intelligent sleeper platform to automatically analyze and predict the change of the service life of the sleeper so as to monitor the state of the sleeper; the platform comprises a field machine module, a data transmission module, a data storage and analysis module and an operation and maintenance service module.
本发明公开了一种基于时序数据特征融合的轨枕寿命预测方法,构建轨枕寿命预测模型实现对轨枕寿命值预测及轨枕状态监测,基于该模型建立了智慧轨枕平台系统;模型包括:为每种传感器均构建LSTM,将所有中间层输出的特征组合后经全连接层输出传感器综合特征;将LSTM输出的中间层结果输入Transformer中得到传感器间特征对应的权重分配;由LSTM、Transformer得到的数据作为综合时序数据特征集输入随机森林模型中训练得到最后的轨枕寿命预测模型。本方法能够应用于智慧轨枕平台的实现自动分析预测的轨枕寿命的变化来进行轨枕状态监测;其中平台包括现场机模块、数据传输模块、数据储存和分析模块、运维服务模块。
Sleeper life prediction method based on time series data feature fusion
一种基于时序数据特征融合的轨枕寿命预测方法
2024-01-09
Patent
Electronic Resource
Chinese
Train positioning system based on sleeper defect feature detection
European Patent Office | 2020
|European Patent Office | 2023
|Frame-sleeper track promises a longer life
British Library Online Contents | 2002