Research On Traffic State Recognition And Short-Term Traffic Flow Prediction Base On Data Mining

With the gradual development of urban traffic networks toward complexity and intelligence,reasonable traffic state recognition and accurate traffic flow prediction are important supports to realize traffic guidance and traffic planning in intelligent transportation systems.This study takes the traffic data collected from a testing station on Avenue 114 in California as the research object,and studies how to solve the problems of traffic state recognition and short-time traffic flow prediction based on data mining technology.(1)Study the problem of traffic state recognition model construction.Based on the analysis of 402 sets of data on Avenue114,the road traffic status of this section is divided into2,3,4 and 5 categories between smooth and congested respectively;based on such classification criteria,in order to improve the accuracy of traffic status recognition based on traditional K-Means model,through the optimization of its initial clustering center selection method and the distance calculation method between samples,an improved K-Means traffic state recognition model;meanwhile,a spectral clustering traffic state recognition model is established from saving the time of traffic state recognition by traditional K-Means model.(2)Study the problem of traffic flow prediction model construction.For this short-time traffic section,features are created for traffic flow data,and the original data are transformed into seven types of features,and three integrated learning models of random forest,GBDT and XGBoost are used for short-time traffic flow prediction;to solve the problem of losing historical data information due to the gradual increase of traffic flow data scale,the LSTM neural network short-time traffic flow prediction model is built.In order to solve the problem that the accuracy of the short-time traffic flow prediction model decreases due to the low data volume and low data integrity and reliability,an improved gray Markov short-time traffic flow prediction model is proposed based on the optimization of the model data accumulation method.(3)To study the problem of building a platform for short-time traffic flow forecasting system.In view of the phenomenon that the amount of traffic flow data is not easily saved and the model calculation results are easily lost,a real-time short-time traffic flow prediction system platform is built based on the combination of My SQL database and Python programming to realize the real-time incoming and saving data through My SQL,and the data and processing-model building-data prediction results saving and presentation through Python.The research results show that the improved K-Means traffic state recognition algorithm has 4.8% higher accuracy and 91% shorter computation time than the traditional K-Means recognition algorithm;meanwhile,by modifying the traffic flow data features of this road section,three kinds of random forest,GBDT and XGBoost are used to The average goodness-of-fit of the training set for predicting short-time traffic flow by using three integrated learning models of random forest,GBDT and XGBoost reached over 90%,and the validation set reached over 85%;the goodness-of-fit of LSTM and the improved gray Markov model in the training set of this road section were 81% and 90%,and the accuracy of the validation set was 81% and 84%,respectively.