Short-term Traffic Flow Prediction Based On Stacking Model Fusion

Real-time and accurate prediction of traffic flow in a short period of time is the core component of an intelligent transportation system,and it is also an important technical basis for the successful completion of the construction of an intelligent transportation city.The spatiotemporal correlation,randomness and mutation of traffic flow are very prominent in a short period of time.Once abnormal traffic occurs in a section of the road network,the traffic of both upstream and downstream sections will be affected.The traditional short-term traffic flow prediction methods can’t predict accurately when the traffic flow fluctuates greatly due to too little consideration of the influencing factors of traffic flow or the limitation of the algorithm used,which cannot fully meet the growing demand of real-time traffic prediction.In this paper,based on the research of the existing short-term traffic flow prediction model,the popular strategy of machine learning Stacking is used to build a short-term traffic flow prediction model based on the fusion of Stacking model,to achieve a more accurate and stable prediction of short-term traffic flow.The main contents of this paper are as follows:(1)The common collection methods of cross-section traffic data are summarized,the processing flow and related methods of cross-section traffic data are elaborated in detail,and the isolated forest algorithm is introduced to identify the abnormal traffic detector data.The quality analysis of the collected traffic data of SCATS section is carried out,the necessity of processing the original data is pointed out,and the corresponding example is analyzed.(2)Combined with the urban traffic data collected from April 6,2020 to June 28,2020 in Yuxi City,Yunnan Province,the main characteristics and spatio-temporal correlation of urban road traffic flow were analyzed.The input features of the prediction model are constructed from four aspects of spatio-temporal correlation,time,weather and road information,and the correlation of the input features is analyzed and selected by using the maximum information coefficient method(MIC).(3)On the basis of previous studies,SVM,Random forest,BP neural network,XGBoost,GBDT and Light GBM with better prediction performance were selected to build a single prediction model,and the results were compared and analyzed.In the single model predicted results were analyzed,based on the principle of choosing "good but different",choose the suitable model as primary study,secondary learning machine adopt LASSO regression,construct three different Stacking two layers structure prediction model to forecast the short-term traffic flow,and the experimental results were analyzed.The experimental results show that the three Stacking models proposed in this paper have good prediction performance and are superior to the traditional single model in each result evaluation index.Among them,BP neural network,Random forest,XGBoost and Light GBM are used as primary learners,and LASSO regression is used as secondary learners to build the two-layer structure Stacking prediction model with the best performance,which can provide stable and reliable technical support for the development of intelligent transportation system.