RBFNN-Bagging-Model-Based Study on Bus Speed Predication

To establish intelligent bus information systems for the purpose of providing information support for the “smart city” construction, the speed of buses running in the urban road network must be accurately predicted. Common prediction models on bus speed by adopting neural network or supporting technologies like Support Vector Regression (SVR) can well predict vehicle speed on uni-structural sections, but when the prediction scope is extended to the general urban road network (with coexistence of various complex section structures), these models can hardly achieve satisfactory generalization effect, and may generate significant differences in prediction accuracy on different section structures. Therefore, this paper puts forward a RBFNN (Radial Basis Function Neural Network)-based Bagging integrated learning prediction model which can effectively deal with issues concerning the accurate predication of bus speed in the context of general road network. Major research contributions of this paper include: (1) Introducing speed of taxi with sufficient data and a high road coverage rate as the secondary data source so as to make up for sparseness of bus positioning data; (2) Selecting RBFNN as the base model and based on integrated learning philosophy, improving it to RBFNN-Bagging model, which can overcome the shortcomings of uni-structural model and better adapt to differences in section structures. The model raised in this paper, through verification of measured data, has realized an over-90% prediction accuracy rate of bus speed in different sections within the general urban road network, and has witnessed an over-10% promotion in prediction accuracy when compared with that of the neural network and SVR model.