Research On Traffic Data Synthesis And Prediction Method Based On Parallel Learning

Existing intelligent transportation researches usually use real data from physical space or cyberspace directly,but the quality of such data is uneven,and most of them have problems including data ambiguity,data errors,and data missing.In addition,data barriers,which widely exist between different data sources,limit the in-depth development of related research.In recent years,the continuous innovation and rapid development of technical methods in the fields of data mining,deep learning,along with the introduction of new theoretical frameworks such as parallel systems,parallel data,and parallel learning,have inspired new research ideas for artificially generating large-scale available data,making artificial data in the research and application of intelligent transportation become possible.Therefore,this paper proposes a study on traffic flow data synthesis and prediction methods based on parallel learning.The main research contents are as follows:In view of the difficulty in expressing the data structure in previous studies,as well as the complex interrelationships and impacts on the spatial and temporal dimensions between the states of the urban roads,a network model based on Convolutional Neural Networks is designed.It can extract the spatiotemporal characteristics implied in the real traffic flow data,which will be applied in subsequent related studies.Considering the problems of scarcity,sparseness or lack of available effective traffic data,the data generation model based on Generative Adversarial Networks is constructed,according to the parallel learning theory.The proposed models realize the effective synthesis of traffic flow data,and the reasonable validity of the model design and the generated data are analyzed.To deal with the complex non-stationary temporal dynamic characteristics of traffic flow data and the spatial dependence of road networks,the hidden spatiotemporal features of real traffic flow data are used.And traffic data prediction models based on Traffic-Condition-Aware Ensemble Learning are designed accordingly.Meanwhile,the synthesized parallel data and the real data are used for comparative experiment and analysis.The prediction results demonstrate the necessity of research on traffic flow data feature,the rationality of traffic flow data synthesis methods based on parallel learning,the effectiveness of the synthesized parallel data in the application,and the help of the traffic flow data prediction method based on state awareness ensemble learning to improve the effect of traffic flow prediction.In summary,this paper combines the cutting-edge theories and methods in related fields within recent years to conduct research on traffic data synthesis and prediction methods,which provides a reasonable and feasible new research idea for intelligent transportation research.