Traffic data reconstruction based on compressive sensing with neighbor regularization

Abstract

The production and collection of the mass traffic data in the vehicle network will result in wasted bandwidth and data transmission delay. The mobile edge computing (MEC) technology is able to reduce the transmission cost and provide fast interactive response. However, the amount of data between roadside units (RSUs) and MEC servers has not decreased. Compressive sensing (CS) technology can reduce the sampling frequency and reconstruct the signal with even fewer samples than the sampling theorem requires. Therefore, in this article, we use CS technology to reduce the data in the "last mile" of traffic edge computing network. We propose two traffic data reconstruction models, which use different neighbor regularization methods in the objective function of CS reconstruction model. Different from the previous studies, we leverage the low-rank nature and the neighbor relationship between detectors to improve the reconstruction accuracy of different missing scenes. To validate the improved reconstruction models, we have carried out extensive experiments based on the actual traffic flow data of the freeway system under different loss rates and loss patterns. The missing data are caused by reduced sampling frequency and accidental transmission loss so that it is closer to the real fact. The experimental results show that even if the traffic data from the RSUs to MEC servers are decreased by 50%, the complete data can be reconstructed accurately.

Nonlinear Traffic Data Reconstruction in Large-Scale Internet of Vehicle Systems: A Neural Network Approach

“…where λ is the regularization parameter which can prevent the problem of data over-fitting. This goal is similar to the low-rank goal explained in [4]. It can also be regarded as the data estimation error obeys the Gaussian distribution, and it is assumed that the underlying features are also obeying the Gaussian distribution.…”

“…In recent years, some studies have used compressed sensing technology [21], [22] to reconstruct lost traffic flow data. The data sets used in these studies can be divided into two categories: mobile vehicle traffic data sets [23], [24], and RSU detector traffic data sets [4], [15], [25].…”

“…Fig. 4 illustrates the ideas of this article based on the compressive sensing method [4] and the connections between them. It is proved that the traffic flow data has low rank, which is a prerequisite for the use of compressed sensing to restore the data, then by adding different neighbor relationship regular terms to improve the ordinary compressed sensing reconstruction data model in the traffic scene, the SIMM1 and SIMM2 models are proposed.…”

“…The model based on compressive sensing reconstruction data uses matrix decomposition to obtain the potential features corresponding to the detector and the time point, modeling the traffic flow data. This method assumes that the target matrix X to be reconstructed can be approximated as the product PQ T of the two potential feature matrices P ∈ R m×K and Q ∈ R N ×K , where P and Q are similar to the L and R in our previous work [4]. We regard p i as a feature vector describing detector i, and q j as a feature vector describing time period j.…”

Cyber‐physical network architecture for data stream provisioning in complex ecosystems

“…Artificial intelligence (AI) adoption within cloud, fog, and edge computing paradigms has made intelligent societies a reality. Some recent contributions in cyber‐physical systems (CPS) include vehicular network compressive sensing (CS) technology, 2 IoT smart disaster management, 3 smart city vehicular edge server design, 4 and energy‐efficient workload distribution, 5 among others. Interestingly, these are all provisioned from the dedicated cloud deployments.…”