Speaker
Description
The hazards of overloaded vehicles are extremely severe, causing significant losses to the nation and its people. The weigh-in-motion (WIM) system, as an essential technology for monitoring vehicle loads, is recognized as an effective solution for identifying overloaded vehicles. However, the vast amounts of data accumulated by current WIM systems merely preserve and record conventional vehicle load information, with insufficient exploration and utilization of data concerning overloaded vehicles, preventing its ultimate application in managing such issues. The advancement of deep learning has greatly optimized the overall processes of data analysis and modeling, allowing for better capture of abstract features and underlying patterns within the data. To this end, the paper proposes an end-to-end deep learning method based on CNN-LSTM, aiming to effectively identify and predict overloaded vehicles by utilizing historical monitoring data from the WIM system, particularly from the perspective of the time series characteristics of overloaded vehicles. The convolutional neural network (CNN) offers excellent feature extraction capabilities, while the long short-term memory network (LSTM) can learn the dependencies within time series data. By combining these two approaches, the prediction accuracy for overloaded vehicles is improved. Results indicate that, compared to traditional single models, the proposed method demonstrates better accuracy and robustness, enhancing the information acquisition capabilities of traffic management departments, assisting in the coordinated governance of resources, and improving decision-making capabilities and overall efficiency in managing overload issues.
