Short-term traffic flow forecasting based on the fusion of RBF neural networks and fuzzy Markov chains

RBF neural networks and fuzzy Markov chains were studied to forecast traffic flow, which has great randomness and fluctuation. A short-time rolling forecasting method for traffic flow was proposed. This method used an RBF neural network to establish a forecasting baseline and the residual errors were achieved. The fuzzy c-means algorithm was applied to divide the residual errors into different fuzzy states for fuzzy Markov chains and state centers were obtained. Then fuzzy state transition probability matrices were calculated. Subsequently, the state transition was analyzed to determine the most possible state of the forecasting value, and then the corresponding state center was used to revise the forecasting value to achieve a more accurate one. Meanwhile, the real-time rolling forecasting mechanism for traffic flow was realized. Using this method to conduct a simulation experiment on real data, the results demonstrate that this method is superior to the general RBF neural networks in precision and has good adaptability to the dynamic traffic flow environment.