Research On Vessel Traffic Flow Prediction Using Low-rank And Sparse Decomposition Of Matrix

With the development of water transport economy in our country,the vessel traffic flow increased rapidly.It promotes the economic and social development of our country,while the risk of maritime transport is becoming more and more serious.One of the effective methods to reduce the traffic accidents is to predict the traffic flow accurately and efficiently.The prediction of vessel traffic flow provides decision-making basis for planning and management of the port.This article analyzes and summaries many kinds of vessel traffic flow prediction models.Considering the autocorrelation and low-rank property of the data,we propose to apply the low-rank and sparse decomposition theory to the prediction of vessel traffic flow.Firstly,we use the statistical theory to analyze the vessel traffic flow.The results show that the data of vessel traffic flow can meet the seasonal variation and have low-rank and sparse features.Based on this,we introduce the low-rank and sparse decomposition theory of matrix into the vessel traffic flow prediction.The traffic flow data is divided into two parts:the low-rank part and the sparse part.The low-rank part reflects the inherent and stability law of the data.The sparse part reflects the effect of the mutation factors on the traffic flow.Secondly,in order to improve the effective utilization rate of data,we add a total variation regularization to the traditional low-rank and sparse decomposition model.We establish a prediction model based on the low-rank,sparse plus total variation constraints.The model is applied to the prediction of vessel traffic flow in Tianjin port.The results show that this method can reflect the seasonal variation of vessel traffic flow when compared with the neural network and the traditional low-rank and sparse decomposition model.The method improves the prediction accuracy significantly.Finally,we extend the traditional convex optimization model by replace the original ₁l norm with the l_p norm.The forecasting model of vessel traffic flow based on the non-convex low rank plus sparse constraints is established and compared with the common methods.By solving the non-convex low rank plus sparse decomposition model,it is shown that the theory can be used to solve the non-convex model.The experimental results show that the low-rank and sparse decomposition models established in this article can effectively utilize the low-rank property of the data to improve the prediction accuracy and predict the traffic flow more accurately.