Prediction of Freeway Travel Time in Incident Management Evaluation Based on Genetic Neural Network

摘要

In order to evaluate the freeway incident management systems, the normal travel time for calculating the incident delays and benefits usually needs to be predicted. The evaluation result is very sensitive to incident delay, so it is inaccurate to use month, week, or day average value for describing the normal travel time. The artificial neural network with strong adaptability and high accuracy was proposed to solve the problem. Firstly the structures of a predicting model and data acquisition plan were proposed. Then a genetic algorithm was introduced to improve the traditional BP(Back propagation) neural network considering that BP network easily falls into the local convergence, and the genetic algorithm has a good global search capability. GA-BP(Genetic algorithm improved back propagation) network showed better convergence speed and predicting accuracy than BP in the study of Ningtong freeway travel time. The final error of GA-BP can be accepted, and the forecast results meet the actual need.