Research On Processing And Prediction Of Time Series Data For Traffic

With the rapid development of technology,transportation,as one of the most important components of people’s daily life,has become a key topic in the current society.Transportation is the key to the development of the country and the foundation of a strong country,and accelerating the construction of a strong transportation country is an important deployment made by the Fifth Plenary Session of the 19 th CPC Central Committee.Using Internet,cloud computing,big data,Internet of Things,and artificial intelligence technologies to develop safe,convenient,and efficient intelligent transportation is crucial to building a strong transportation country.Therefore,this paper is oriented to explore the field of processing and prediction of traffic timing data,which mainly covers various aspects of traffic data collection,cleaning,detection,processing and prediction.The focus contains the following parts:(1)Due to the complexity of traffic data and the diversity of influencing factors,the anomaly detection of traffic timing data has been a key challenge in the related research field.In Chapter 2 of this paper,an anomaly detection method for traffic data by data modeling combined with an anomaly detector is introduced,which can accurately fit the traffic data and determine the fluctuation level of the fitted value and the actual observed value(i.e.,the deviation and variance of the fitted data from the actual data are clarified).The method is paired with a variety of popular existing anomaly detection algorithms,and eventually good classification detection results are obtained.(2)With the rise of various artificial intelligence algorithms,researchers found that graph neural network algorithms have very high adaptability to traffic data,because road traffic data are often related not only to the historical data patterns of that road,but also to the data characteristics of neighboring roads.Therefore,the graph neural network algorithm has gradually become one of the best algorithms in the field of traffic data prediction in recent years.However,the original traffic data is tabular data,and it is necessary to form an effective graph network structure first in order to capture the traffic data well with graph neural network algorithm.Therefore,in Chapters 3 and 4 of this paper,two different methods for processing traffic data into network structures are presented and the accuracy is verified with good results under several completely different types of algorithms.(3)Most of the current researchers’ predictions of traffic timing data focus on the field of regression,but due to the special nature of traffic data,accurate prediction of the normal range of traffic data is often a better guide to coordinate traffic than predicting the specific values of traffic data.Therefore,this paper proposes a classification and prediction method for traffic timing data in Chapter 5,which classifies traffic data by constituting a three-dimensional network structure,and verifies the accuracy of the method by combining it with a related graph neural network classification algorithm.(4)Traffic data,as a more complex type of temporal data in Io T,contains various types of characteristics,such as passing data,speed data,traffic data,etc.At the same time,the temporal length,spatial length,and spatial length of the data can be used to predict the traffic data.At the same time,the time length and spatial span of the data are also different.Therefore,in Chapter 6 of this paper,a variety of traffic data sets of completely different types are collected and produced,and controlled experiments are conducted with several data features and model parameters that are likely to affect the prediction results.The prediction schemes applicable to different types of traffic datasets are explored.(5)Non-neural network algorithms and neural network algorithms,as the two most common types of machine learning algorithms in the field of time-series data prediction,have different advantages and disadvantages,for example,non-neural network algorithms have simple models and fast prediction speed,but the prediction effect is average in the face of complex data;neural network algorithms have high prediction accuracy,but complex models and slow iteration speed.Therefore,an improved neural network algorithm based on the importance of temporal features is proposed in Chapter 7 of this paper,which combines the advantages of the traditional non-neural network algorithm with those of the neural network algorithm.And the prediction effect is verified under various types of traffic datasets.