Speed estimation based on multiple kernel learning

Traffic state identification is one of the core missions of Intelligent Transportation Systems. In order to correctly identify the state of traffic flow, the traffic speed must be obtained accurately. In Shanghai, most of the industrial loop detectors (ILDs) are installed in a single loop way. These ILDs can only detect the parameters of flow, saturation, etc., but the speed can not be detected. If the relationship between the traffic flow and speed can be mined accurately, we can obtain the speed using the flow data directly. The purpose of this study is to use multiple kernel support vector regression (MKL-SVR) algorithm to model the relationship between the traffic speed and flow, then estimate the speed accurately. Extensive experiments have been performed to evaluate the performances of the four algorithms: polynomial fitting algorithm, BP neural networks, SVR and MKL-SVR. The experimental results show that MKL-SVR has the best and most robust performances.