Research On Data Reliability In Safety Monitoring System For Construction Tunnels

Safety monitoring is of great significance in the construction of subway tunnels,and reliable data are the basis for evaluating engineering risk levels and adjusting construction strategies.The closed and complex construction environment in the tunnel has a great influence on the data reliability of the safety monitoring system.However,most of the current research focuses on hardware configuration and software development,which lacks attention to data processing.Based on the two communication methods of Zig Bee and Wi-Fi,we study the reliability guarantee methods when data flows in the system.The main contents are as follows:(1)A reliable transmission mechanism based on dual communication links with automatic networking is designed to ensure the data reliability of the acquisition process.In the acquisition phase,aiming at the two problems of co-channel interference when Zig Bee and Wi-Fi coexist and timing mismatch when Wi-Fi data is uploaded,a synchronization mechanism of dual communication links based on mutex locks is designed.Transmission reliability is ensured by controlling the sequence of channels used by Zig Bee and Wi-Fi.In the parsing phase,a sorting algorithm for the node address set based on two searches and a capture method for real-time route changes are proposed.By searching and storing the hierarchical traversal results of the data acquisition network,the automatic network structure construction is realized by comparing the old routing information with the new one.The experimental results show that the average data packet loss rate decreases from 7.47%to 2.16%after the synchronization mechanism is set,significantly improving the quality of communication.And on this basis,the system realizes the three-dimensional data monitoring combined with the topology structure.(2)A two-stage cleaning algorithm and a multivariate time series prediction method are proposed to ensure the data reliability of the analysis process.In the cleaning phase,a two-stage data cleaning algorithm(LSO-SW-RKF),which combines outlier processing and a reliable Kalman Filter,is proposed to solve the problem that the classical Kalman Filter is not suitable for nonlinear systems.In the first stage,a least square compensation algorithm based on a sliding window is proposed to deal with the outliers in the observed data.The Innovation-Hard Threshold method is designed for detection combined with data characteristics and hardware parameters.If it is an outlier,the data in the window will be used as parameters to perform the least square fitting and compensate for it.In the second stage,the system noise,which is the key parameter affecting the filtering accuracy,is modeled,and the Reliable Kalman Filter that can adapt to noise changes is constructed.Given the problem that the current systems mostly use univariate input for prediction in the prediction phase,a prediction algorithm for the representative disaster data combining similar sequences(DTW-Kmeans-LSTM)is proposed.Based on the prediction model of the Long Short-Term Memory network,the algorithm uses the method of unsupervised feature selection,Laplacian Score,to select the representative disaster data,and takes the dynamic time warping distance as the similarity metric between the sequences of the same type to select the similar sequences of the representative data together as the input of the model.The experimental results show that compared with the Kalman Filter and Extended Kalman Filter,LSO-SW-RKF improves by 79.60%,89.94%,68.31%,79.55%and 97.20%,99.35%,91.93%,97.18%on MAE,MSE,RMSE and MAPE respectively.By comparing the prediction results without using similar sequence selection and using Euclidean Distance as the similarity metric,DTW-kmeans-LSTM improves by 52.36%,51.78%,52.94%and 40.61%,44.58%,40%respectively in MAE,RMSE and MAPE;~2 is also improved to 0.75.