Real-time forecasting for short-term traffic flow based on General Regression Neural Network

Abstract

Analysis and forecasting for short-term traffic flow have become a critical problem in intelligent transportation system (ITS). This paper introduces the basic theory and features of General Regression Neural Network (GRNN) and its advantages. A forecasting model based on GRNN is built for short-term traffic flow time series at urban road section in 10-minutes interval. In order to get ideal forecasting results, the search method is used to obtain the number of input neurons and the value of smooth factor. When the number of input neuron and training samples are defined, the model can forecast the next 10-minutes traffic flow using the method of dynamic learning and single-step forecasting. Compared with the forecasting results of the traditional BP neural network (BPNN) which adopts error back-propagation learning method, this model is more accurate, and more suitable for short-term traffic flow forecasting.