Traffic State Estimation Method Based On Multi-source Traffic Data

The development of information technology has brought a sharp increase of multisource traffic data. And the analysis of traffic big data has brought great opportunities in a variety of transportation applications. Meanwhile, traffic data loss problem also made the application of single source traffic data faced with serious challenges. Existing traffic data processing methods often only mine the single-source data properties. Based on the making full use of multi-source traffic data, we study the fusion method of multi-source traffic data to deal with lost problem of single-source traffic data. Specifically, in this paper, we propose multi-tensor completion algorithm to deal with the missing data problem and/or sparse problem in fixed data source and/or mobile data source. The multi-tensor completion framework can mine the shared subspace of heterogenous and homogenous traffic data source to complete multi-source traffic data at the same time, combined with neural network method, the traffic data collected from multi-source can be fused to obtain more reliable traffic state information. The main content of this paper is as following.Firstly, a novel algorithm named multi-data common structure based tensor completion is proposed to deal with the missing problem of multi-source data sets. In view of the tensor based method can effectively use the multi-mode characteristics of data, a novel framework is proposed on the basis of conventional tensor completion theory, which can also make full use of multi-source data information. Three algorithms that corresponding to three shared structures: completely equal subspace, partially equal subspace and approximated equal subspace.Secondly, a traffic state estimation method based on multi-source traffic data is proposed on the basis of multi-source traffic data feature. we analyze the statistical property of multisource traffic data based on simulation, and the proposed algorithm is utilized to deal with the problem of data missing and/or data sparsity of fixed sensor and mobile sensor to obtain complete multi-source traffic data. Combined with neural network based data fusion method, our method can obtain a pretty accurate and complete traffic state.Experimental results show that the proposed method can take full advantage of multisource traffic data shared information and utilize the multi-mode characteristic of singlesource traffic data to achieve multi-source data completion, which improve the reliability of traffic state estimation.