基于深度学习的路面状况监控与预测物联网设计

Road Condition Monitoring and Prediction IoT Design Leveraging Deep Learning

由于路面结冰严重威胁行车安全,因而公路结冰预警技术成为保障安全行驶的关键性技术。基于大连地区五年历史数据集,利用神经网络特征提取与压缩的能力,提出了基于深度学习的短时间道路结冰预测系统,并且构建了相应的硬件平台。具体地说,首先,构建了LSTM、CNN、TEXTCNN、ConvLSTM、Transformer五种神经网络从时域、空域等多维度挖掘数据集的内在结构,并对下一时刻道路结冰状况进行预测。其次,将训练好的五种神经网络调试并下载到树莓派4b平台,实现了低成本的硬件系统。最后,实验结果表明,TEXTCNN模型的预测准确率可以达到97%,同时鲁棒性最高,有效地解决高速公路高危路段的路面凝冰预测的技术难题。

Due to the severe threat of icy roads to driving safety,road ice early warning technology for highways has become a critical technology for ensuring safe driving.Based on the five-year historical dataset in Dalian,a short-term road icing prediction system based on deep learning is proposed via utilizing the feature extraction and compression capabilities of neural networks,and a corresponding hardware platform is constructed.Specifically,this paper first builds five neural networks of LSTM,CNN,TEXTCNN,ConvLSTM,and Transformer to mine the intrinsic structure of the dataset from multiple dimensions such as time and space domains,and predicts the road icing situation at the next moment.Secondly,this paper fine-tunes and downloads the five trained neural networks to the Raspberry Pi 4b platform,realizing a low-cost hardware system.Finally,experimental results show that the prediction accuracy of the TEXTCNN model can reach 97%with the highest robustness,effectively solving the technical problem of road icing prediction in high-risk sections of highways.