基于LSTM-VAE的电梯异常检测

Elevator anomaly detection based on LSTM-VAE model

为保障电梯的正常运行,传统的定期维保方式浪费了大量的人力物力,目前业界对于电梯的按需维保有迫切的需求。按需维保的核心技术是异常状态检测,为了有效解决电梯状态的时序数据异常检测问题,提出了LSTM(Long Short-Term Memory Networks,长短期记忆网络)与VAE(Variational Auto-Encoder,变分自编码器)结合的模型LSTM-VAE。基于电梯门机开合与垂梯升降的加速度时序数据,使用长短期记忆网络对加速度数据进行编码并提取加速度数据的时序特征,再通过变分自编码器重构模型,用重构的加速度数据与输入数据计算重构误差,找出异常数据的阈值。通过采集的真实垂梯和门机的加速度数据,实验发现LSTM-VAE在检测率上要优于LSTM和VAE方法。

In order to ensure the normal operation of elevators,traditional regular maintenance methods waste a lot of manpower and material resources.At present,the industry has an urgent need for elevator maintenance on-demand.The core technology of on-demand maintenance is abnormal state detection.In order to effectively solve the problem of abnormal detection of time series data of elevator status,a model LSTM-VAE,which is a combination of Long Short-Term Memory Networks(LSTM) and Variational Auto-Encoder(VAE),is proposed.Based on the acceleration time series data of the opening and closing of the elevator door machine and the vertical elevator lifting,a long and short-term memory network is used to encode the acceleration data and extract the time series characteristics of the acceleration data.Through the reconstruction model of the variational encoder,the reconstruction error is calculated from the reconstructed acceleration data and the input data,and the threshold of abnormal data is found.Through the collected acceleration data of the real vertical elevator and door machine,the experiment found that LSTM-VAE is superior to LSTM and VAE methods in detection rate.