A Prediction Model for Short-term Traffic Flow Based on Space-time GPSO-SVM

In order to improve accuracy of short-term forecasting of traffic flow on urban roads, a model is proposed by space-time genetic-particle swarm optimization (GPSO) and support vector machine (SVM).The spatial-temporal correlation of original traffic flow of a road network is analyzed based on a principal component analysis.Instead of original traffic flow, this paper takes less principal components as predictive factors.The crossover and mutation factors of a genetic algorithm are appiled into a particle swarm optimization algorithm, which can avoid local optimization.According to the improved particle swarm optimization algorithm, parameters of SVM model are optimized, then an optimal SVM model is developed, as well as forecasting of short-term traffic flow.A practical case study is taken based on the data of a road network in Changchun City.The results show that the GPSO algorithm does not fall into local optimum when the parameters of SVM model are optimized, and the effects of optimization is better;the relative error of this proposed model is stable compared with the particle swarm SVM model and GPSO-SVM model, and the average prediction accuracy is improved by 4.96% and 3.41%, respectively.