Research On Traffic Flow Forecasting Method Based On Incremental Tensor Model

This paper investigates the problem of predicting traffic flow in intelligent transportation systems.Considering the high latitude and high complexity of spatiotemporal data,an incremental tensor model is constructed by using the tensor structure that achieves excellent performance in the field of knowledge graph.Combined with the multi-dimensionality,immediacy and high-precision characteristics of traffic flow data,a traffic flow forecasting method based on incremental tensor model is designed.The research includes the following aspects:First,an incremental tensor model is constructed.From the perspective of high-dimensional data representation,an incremental tensor structure for simulating the tensor flow generation process is proposed.In order to avoid the potential interference of early historical data while ensuring data integrity,a weighting mechanism that assigns different weights to data in different periods according to data characteristics is introduced,which makes the model more explanatory when dealing with spatio-temporal data.Secondly,a traffic flow prediction method based on incremental tensor model is designed.Considering the multiple characteristics of the traffic flow prediction field,the gravitational search method stemming from the gravity rule is used to optimize the specific selection of data weights in different periods during the increment;regarding the selection of tensor completion method,the tensor filling method is used,among which fast low-rank tensor complement method is utilized for analysis and prediction,so as to balance high precision with low time consumption;Thirdly,in order to verify the validity of the model,different types of methods are evaluated by real-world data sets,and long-term prediction problems are also evaluated to test the extensibility of the proposed method.The results show that the research method proposed in this paper achieves superior results in comparison with all other methods,and also has a good performance in long-term traffic flow prediction task.