Accuracy Evaluation Of Highway Tunnel Monitoring Data Based On Machine Learning

In recent years,China’s highway tunnel construction has made remarkable achievements,realizing the transformation from "tunnel power" to "tunnel power",with a large number of highway tunnels completed and put into use,tunnel operation and management is facing new challenges.At present,most highway tunnels in China adopt regional control and switch ring network control transmission technology,and have realized the digitalization and partial intelligence of equipment control and data monitoring management in the tunnel.However,the general operating environment in the tunnel is harsh,and the operation data monitoring equipment such as visibility,CO,wind speed and direction,brightness,and vehicle detector have frequent failures,and cannot operate stably for a long time.Moreover,the integration of various operational data is insufficient,which is difficult to provide an effective data basis for the accurate control of ventilation control,lighting control,traffic control and guidance.Therefore,based on machine learning,this thesis studies the accuracy of operation monitoring data collected by electromechanical systems of highway tunnels.It aims to improve the accuracy of tunnel operation monitoring data through the selection,processing,feature extraction and data research and judgment of tunnel operation monitoring data,so as to provide data support for the accurate management and control of tunnel operation and effectively improve the level of tunnel operation control and control.The main research content includes the following aspects:(1)The types,characteristics and collection methods of highway tunnel operation monitoring data are introduced,and tunnel operation monitoring data selection based on validity,multi-modality,reliability and operability is proposed according to data types,so as to lay a data foundation for subsequent research.(2)By analyzing the three mainstream missing data processing methods of KNN,RF and DNN,and comparing the interpolation accuracy and interpolation efficiency,a missing preprocessing method for highway tunnel operation monitoring data based on random forest algorithm is proposed,which realizes efficient and accurate completion of missing data.(3)By integrating the efficiency of the isolated forest algorithm and the flexibility of the Prophet algorithm,the isolated forest and Prophet algorithm are integrated,and an anomaly detection algorithm for highway tunnel operation monitoring data based on the combination of coarse and fine granularity is proposed,which realizes the rapid identification and labeling of abnormal data.(4)Aiming at the problem that the accuracy of the data collected by a single detector is insufficient and cannot fully and accurately reflect the actual operation status of highway tunnels,a multimodal information fusion model based on CNN-LSTMAttention is proposed.The CNN-LSTM deep learning model and self-attention mechanism are applied to the tunnel operation monitoring data research and judgment through multimodal information fusion,the CNN-LSTM model extracts features from non-intrusive multimodal data,and the self-attention mechanism integrates visibility,CO,traffic volume and other data to obtain the optimal information fusion model,and the research and judgment output can accurately reflect the real operation status of highway tunnels.Finally,this model is integrated into the intelligent management and control platform of Jiulingshan Tunnel to ensure the high reliability and high reliability of tunnel operation monitoring data,support the calculation and analysis of air volume required under normal working conditions,and carry out scientific and accurate automatic ventilation control.