Research And Implementation Of Intelligent Traffic Flow Prediction System

With the rapid development of China’s economy and the improvement of people’s income level,the volume of urban vehicles has become more and more high,which makes urban road traffic congestion,traffic accidents and environmental pollution become more and more serious,among these problems,traffic congestion in urban roads is the most prominent.The governance of urban road traffic congestion cannot only rely on transportation policy and urban road expansion and transformation,but also rely on scientific and data supported traffic flow prediction methods to analyze traffic flow,and based on the analysis to take traffic drainage measures to achieve the purpose of alleviating traffic congestion.This thesis focuses on prediction of short-term traffic flow and medium-long term traffic flow,traffic flow data acquisition based on the three-frame difference frame-area traffic flow collection method,final realization of intelligent traffic flow prediction system with high reliability and high precision.The main contents of this thesis are as follows:(1)The study of traffic flow collection method based on video image processing.The traditional method of traffic flow collection based on video and image processing mostly uses the way of inter frame difference or background difference combined with virtual coil counting,and the speed of collection is still to be improved.This thesis puts forward the three-frame difference frame area traffic flow collection method,through setting the effective calculation area,reducing the amount of calculation of differential calculations,for the same video monitoring,through the experimental comparison,the speed of this method is 30% faster than that of the traditional method.(2)The study of the prediction model of short-term traffic flow.In this thesis,BP neural network prediction model is used to predict short-term traffic flow.The experimental results show that the average absolute error of BP neural network prediction model is 6.39%,and the accuracy is still to be improved.In this thesis,the genetic algorithm is used to optimize the weights and thresholds of the BP neural network,which can easily fall into the local minimum value,combined with the periodicity of traffic flow,this thesis puts forward BP neural network based on genetic algorithm optimization and weighted calculation of historical data prediction model(GA-BP-H).The experiment shows that the average absolute error of GA-BP-H is 5.04%(3)The study of the prediction model of short-term traffic flow.This thesis first carried out an experiment on using BP neural network and GA-BP-H to predict traffic flow in medium-long term.The average absolute error of BP neural network is 6.74%,the average absolute error of GA-BP-H is 6.72%,and the accuracy is still to be improved.Due to the periodic characteristics of traffic flow,and there is a great difference between the working day and the weekend,combined with characteristics of short-term traffic flow prediction,this thesis puts forward a self-regression integral sliding average model with weekly cycle(ARIMA-W).The experiment shows that the average absolute error of ARIMA-W is 6.11%.Based on the content above,this thesis finally designs and realizes intelligent traffic flow prediction system,and verifies the stability of the system through system test.