Research On Tower State Monitoring Technology Based On Time Series

Tower safety monitoring is very important for the safety of communication tower and pylon.During the working process,it is constantly affected by extreme climate such as wind,frost,rain,snow,electromagnetic field and even earthquake,and it is very easy to have major accidents such as bending deformation,tilting and even collapse.Therefore,many researchers have implemented the tower online monitoring system based on wireless sensors to supervise and warn the safety of the tower online,so as to improve the efficiency of tower operation and maintenance.In the monitoring system,how to reasonably use the collected state data for anomaly detection and early warning is an important research topic in the industry.The status data of the tower returned by the sensor is the standard time series data.From the perspective of data timing,this paper realizes the anomaly detection algorithm based on the status data,aiming at the characteristics of short period,strong randomness,high complexity,more normal data than abnormal ones.The main research contents are as follows:Firstly,an anomaly detection model based on the prediction of long-term and short-term memory network model is proposed.The algorithm uses the long-term and short-term memory network model to learn the trend of the normal state data of the iron tower,predict the state in the future and compare it with the real state,and determine the state of the iron tower according to the experience threshold.Secondly,an integrated learning based anomaly detection method,LSTM bagging and LSTM boosting,is proposed.This algorithm is based on the idea of the integrated learning method Bagging and Ada Boost,respectively to realize the improved long-short term memory network prediction model.By comparing the prediction error with the empirical threshold,the abnormal state of the tower can be identified.Finally,a dynamic threshold method based on sliding window is proposed to detect the anomaly.The algorithm improves the anomaly detection results on the basis of previous research results.Through the sliding window model and the exponential moving average method,the dynamic threshold which adapts to the changing trend of the tower state is obtained.Based on the long-short term memory network model and the dynamic threshold,the detection algorithm of tower time series anomaly effectively reduce the missed rate and false alarm rate of abnormal detection.A complete anomaly detection algorithm based on prediction is implemented in the process of tower condition monitoring,and the prediction model and threshold are optimized respectively.Experiment with data collected from the actual installation of the tower equipment,the results show that the anomaly detection algorithm has good performance and can achieve the monitoring of tower.