Research Of The Urban Road Short-term Traffic Flow Prediction Based On WiFi Probe Data

With the continuous improvement of urban modernization,the problem of traffic congestion is also becoming increasingly prominent.The upgrading of software and hardware in information technology provides an important guarantee for the development of intelligent transportation system,and the short-time traffic flow prediction in intelligent transportation system is one of the important means to effectively alleviate traffic congestion.With the emergence of Internet of Vehicles(Io V)technology,traffic data has moved from the original period of single scarcity to the current period of multi-source mass,and the traditional traffic detection data has gradually lost its advantages in terms of construction cost and real-time data transmission.The assembly of vehicle-mounted WiFi has greatly increased the proportion of vehicle-connected networks,and also generated a large amount of wireless communication data,which has the characteristics of large sample size,good real-time performance and low acquisition cost.Therefore,it is also of important research value to use it as traffic detection data.Based on WiFi wireless detection data,this paper studies the characteristics of short-term traffic flow in the road section,and designs the corresponding prediction model based on deep learning theory.The main research work is as follows:(1)Firstly,the acquisition principle of WiFi wireless detection data is analyzed,and the acquired WiFi wireless detection data is matched with the roadside probe device,and the traffic flow time series of the road section is constructed.Then we use Pearson coefficient to analyze the correlation of traffic flow data in time and space.Finally,the data are preprocessed,including the screening and elimination of abnormal data,and the mean replacement of missing data,etc.(2)According to the characteristics of traffic flow time series data,it is transformed into the input sequence of LSTM network by means of sliding window,and then the parameters and structure of LSTM network are adjusted,and the DSLSTM prediction model is established,finally,the prediction performance of the model with different feature inputs is analyzed respectively for single and multi-link sequence inputs.The results show that the prediction accuracy of DSLSTM model with the characteristics of upstream and downstream sections is improved.(3)In order to further improve the prediction performance of the model under the condition of multi-link sequence input,a convolutional long and short term memory prediction model combined with attention mechanism(Conv-LSAM)was designed.The convolution module is added before the LSTM layer to achieve the effect of spatial feature extraction,and the attention mechanism is introduced to enhance the effectiveness of LSTM in time feature extraction.The experimental results show that the Conv-LSAM model can effectively capture the spatial-temporal characteristics of traffic flow,and has better prediction results compared with other benchmark models.