Ensemble Learning Multiple LSSVR With Improved Harmony Search Algorithm for Short-Term Traffic Flow Forecasting

Short-term traffic flow forecasting plays an important role in current intelligent transportation system. For most models, the selection of time lag is a crucial factor affecting the forecasting performance. Instead of choosing a single time lag when constructing model, in this paper, we propose a novel approach attempting to construct multiple base forecasting models, each with different time lag and performance. Least squares support vector regression (LSSVR) with the Gaussian kernel function is adopted as the base model because of its nonlinear modeling capability, as well as empirical performance. Then, the outputs of these base models are integrated to produce final prediction through another LSSVR with the linear kernel function. This ensemble forecasting framework consists of many parameters that need to be adjusted. To address this issue, an improved harmony search algorithm tailored for our forecasting system is further developed for seeking the optimal parameters. The real-world traffic flow data are collected from several observation sites located around the intersection of Interstate 205 and Interstate 84 freeways in Portland, OR, USA. Experimental results verify that the proposed approach is able to provide better forecasting performance in comparison with other competing methods.