Traffic Flow Status Recognition And Prediction System Based On Multi-source Data

In the field of intelligent transportation,urban expressway traffic flow state identification and prediction is a very important research hotspot.By reasonably judging traffic flow state and correctly predicting traffic flow state,real-time road traffic condition assessment can be provided.Some existing traffic flow state identification and prediction systems in China still have shortcomings in real-time and intuitive,and the system identification and prediction caused by using a data source is not accurate enough.This paper designs and implements a traffic flow state recognition and prediction system based on multi-source data,which makes the system more improved in real-time,intuitive and accuracy.The main work includes the following aspects:Firstly,by analyzing and calculating the original traffic flow data(license plate identification data,floating car GPS data and Internet traffic condition data)from various sources,the average speed of vehicles with multiple sources at the same time interval corresponding to the urban expressway is obtained.According to the existing research,the appropriate preprocessing method is selected to identify the abnormal data and repair the missing data to ensure the correctness and integrity of the multi-source traffic flow velocity data,and effectively improve the actual application effect of the data.Secondly,according to the ambiguity and uncertainty characteristics of traffic flow state,the FCM algorithm and process are studied and analyzed.Since the traditional FCM clustering algorithm needs to specify the number of clusters in advance,when there is no specific number of classifications,this method has Certain limitations,and the traditional FCM clustering algorithm is also sensitive to the initial clustering center.For this reason,before performing FCM clustering,the optimal number of clusters and the clustering center at this time are obtained by DBSCAN density clustering.then substituting the obtained number of clusters and the initial clustering center into the FCM clustering algorithm to obtain the final clustering center,and calculating the membership degree of each state of the data points to be processed,according to the division of these clustering states,the state with the largest degree of membership is selected as the state recognition result.Experiments show that the improved FCM algorithm results are more reasonable.Then,based on the characteristics of short-term traffic flow state prediction,the principles and different characteristics of the algorithm of recurrent neural network(RNN)and long short term memory(LSTM)are studied and analyzed,and the Encoder-Decoder model is combined to propose the traffic flow.The seq2 seq neural network prediction model of the state sequence is adjusted by the model training,and the traffic flow state data of the current time and the previous period can be used to predict the trend of the traffic flow state for a certain period of time in the future.Experiments show that the prediction based on the Encoder-Decoder LSTM model is more accurate.Finally,through the web development framework Django,the design and implementation of the various functional modules of the B/S system are carried out using the Python language.The system can calculate and preprocess the original data to obtain multi-source traffic flow velocity data.At the same time,according to the model deployed by the system,it can identify the real-time traffic flow state,and predict the traffic flow state for a period of time,state recognition and prediction results are shown in a visual way.