Research On Prediction Of Road State And Environmental Change Based On Deep Learning

In recent years,with the rapid development of China’s smart cities,Io T infrastructure has been widely deployed.Various sensors and monitoring systems are constantly monitoring and collecting in real time,generating increasingly rich multidimensional data,especially road data and environmental data.The current use of the collected data is most just in monitoring concerns about the current dynamics.With the approach of emerging technologies,these data that have been collected can be intelligently analyzed and processed through AI technology to significantly improve the service level of smart cities.Therefore,this paper uses deep learning models based on historical data collected by sensors to address the problem of road icing prediction,and air quality prediction in the context of environmental changes.The aim is to predict the future changes based on the current road icing and air quality trends to predict the changes in the future period,so as to be able to make advance predictions for prevention and control in order to accelerate the construction of new smart cities.The main research work of this topic is as follows.(1)Multi-source spatio-temporal data collection and pre-processing.Based on the complex factors of traffic road environment changes,it is difficult to be informed in time.Road icing monitoring system built through icing monitoring sensors,real-time collection of road state data.Air quality data is collected in real time through multiple air monitoring stations.After acquiring the data,the collected data are pre-processed such as missing value filling and normalization,followed by multi-source data fusion by time stamp and station location.The data are aligned temporally and spatially to increase the feature dimensionality and reliability of the data.(2)Fusion of self-attention mechanism and GAN for road icing prediction.After studying the problem solution and combining the data characteristics,this paper decomposes the road icing prediction problem into a time series prediction and an unbalanced data set classification problem.A road icing prediction algorithm that incorporates the self-attention mechanism and GAN is designed.Firstly,the self-attention mechanism is used to predict the data in the future period,then a part of icing sample data is generated by GAN according to the distribution of icing sample data,and finally all the data are put into the classifier for classification.The experiments show that the algorithm can effectively and accurately predict the road icing in the future period,and also has better performance compared with the existing methods.(3)Air quality prediction based on GC-SEQ model.The atmospheric pollutant components are complex and diverse in origin,and the time dependence between data is strong.In addition,due to the dynamic movement of the atmosphere,the pollutant concentrations between monitoring stations can affect each other.For this reason,this paper proposes a GC-SEQ-based air quality prediction model that can predict the air quality of multiple stations simultaneously.First,GRU is used to obtain the temporal features in the data,and then the matrix multiplication operation in GRU is replaced by the convolution operator in GCN to form a GC-GRU unit,so that the model can better capture the temporal and spatial features.Finally,the combined GC-GRU units are constructed into a Seq2 Seq model so as to predict the air quality for a future period.The experiments show that the prediction performance is greatly improved compared with SVM,GCN,LSTM and Seq2 Seq models by combining the characteristics of spatio-temporal data.