Research On Intelligent State Recognition Technology Of Expressway Bridge Electromechanical System

Fault detection based on electrical data of Supervisory Control and Data Acquisition(SCADA)is an essential technology in the intelligent transportation system,which is significant to the state perception,operation,and maintenance of highway electromechanical system.For the purpose of solving effective recognition methods of highway electromechanical systems,in view of the changes reflected by the electromechanical equipment in monitoring system,detecting trip faults,and classifying specific fault types are the main research purpose of this paper.Firstly,based on the composition and structure of Taizhou Bridge electromechanical system,the data acquisition plan of the electrical data is studied.The parameter structure of the SCADA power system which reflects the state of the electromechanical equipment of the Taizhou Bridge is analyzed,and the electrical information matrix based on Taizhou Bridge power data structure is constructed.Secondly,the Stack Sparse Auto Encoder model(SSAE)with classifiers for trip fault detection are studied.The experimental results show that the feature extraction performance of the two-layer Stack Autoencoder model is better than that of the single-layer and three-layer.Stack Autoencoder,whose dimensionality reduction data under the Gaussian kernel SVM classifier is better than linear kernel,polynomial kernel,and Sigmoid kernel;the sparsely represented Autoencoder network can effectively improve the feature extraction performance,and its classification accuracy reaches 89.8 %.Thirdly,some conventional recurrent neural network modesl for fault type classification are studied,which is Long Short Term Memory(LSTM),classical Recurrent Neural Network,and Gate Recurrent Unit.The experimental results show that the Long Short-Term MemoryFault Classification Model(LSTM-FCM)is better than the Recurrent Neural Network-Fault Classification Model(RNN-FCM)and the Gate Recurrent Unit-Fault Classification Model(GRU-FCM).The best model has a training accuracy of 94.77% and a verification accuracy of88.84%.Finally,a Fused Stack Sparse Long Short-Term Memory(FSS-LSTM)model based on a deep sparse structure is proposed.The network adds sparsity constraints to the multi-layer LSTM network structure to improve classification efficiency.The unit uses the input gate,forget gate and output gate of the LSTM network to control the information of the memory cell.Besides,The fusion rule is to fuse the output of the shallow network in time dimension,and fuse the high-dimensional vector to retrain in the deep network.The experimental results show that the performance of the proposed fused long-short-term memory network model based on the deep sparse structure is better than the electrical fault classification method of the conventional recurrent neural network,with a training accuracy of 97.15% and a test accuracy of 93.34%.