Short-term Traffic Flow Forecasting Based On Electronic License Plate Data

The short-term traffic flow forecasting is the key part of intelligent transportation system,and it is also an important research direction.To achieve real-time and accurate traffic flow forecasting is an important prerequisite for intelligent traffic and some other related links.It can not only optimize the urban traffic condition,but also improve the energy efficiency,reduce the pressure of the environment and accelerate the development of the society.Previous studies have been based on the single dimension of traffic flow,and most of them only consider an isolated node independently,and there is very few other relevant information that can be referenced.But the emergence of electronic license plate data can make us easer to access to relevant data.Thanks to the rich data field,the using of electronic vehicle license plate data will make the research of intelligent transportation related field enter a new stage.According to the data set of the electronic license plate,we can get the information which is difficult to grasp before,and put it into the practical research of the traffic flow forecast.Because the electronic license plate can be determined in a car,so it is very convenient for us to consider a number of nodes before and after each other through the intrinsic relationship between them,and it is difficult to do in the past.This paper takes electronic license plate data of Chongqing to obtain the corresponding data mining and research.First of all,the original data set need to make a preliminary arrangement;Then wavelet analysis is used to remove noise and extract features;After that,reasonable combination of fuzzy theory and neural network to form a suitable fuzzy neural network prediction model,analyze the input interface of the model,and explain the mechanism of it;Finally,the experiments are used to verify the rationality and validity of the theory,and the corresponding conclusions are drawn.The main contents of this paper are as follows:(1)Data de-noising and feature extraction by wavelet analysis.The arrangement of the data is only the first step in the preprocessing,and the data record is used as the unit;But the wavelet de-noising is based on the data logic as the criterion.The actual data collected will often be mixed with a variety of "impurities",by means of wavelet analysis to "purify" can make the data set closer to the actual stable state.(2)Considering the influence of the precursor collection points and the subsequent collection points on the current node traffic.Some of the vehicles of the predecessor node will be driven into the current node,it can be estimated by the conversion rate of the traffic flow in the previous period,thus,the impact on the prediction precision can be buffered when the abrupt change of the traffic flow occurs;Then the following node traffic situation may make some vehicles to change the route,so it should be appropriate reference.(3)Verify the predictability of electronic license plate data sets;Using the appropriate fuzzy neural network prediction model to improve the prediction accuracy.Neural network is a classical prediction model,which has been proved to be a good complementary combination with the fuzzy theory.(4)The experimental results are compared from different angles.Firstly,this paper analyzes the influence factors of data level;Secondly,the prediction results of neural network model and fuzzy neura l network system are compared,to illustrate the improvement of fuzzy theory on neural network;Finally,the paper compares the performance of fuzzy neural network and other classical prediction algorithms,and illustrates the scientific and rational use o f the hybrid system.At the same time,the validity of the prediction framework is verified by the prediction of the selected models.