Gradually Rolling Prediction of the Use-level of Road Network Based on BP Neural Network

The use-level of road network prediction is the key for making a good plan of pavement maintenance. Traditional methods used for the use-level of road network prediction have some limitations. It is difficult for traditional prediction models to set up multi-parameter equations and predict many factors’ changes. The precision is low, particularly in dealing with massive data of road networks. A neural network has certain advantages to predict the change of road network service’s level. In this paper, a prediction model based on the Back-Propagation Neural Network (BP neural network) was established to predict use-level of a road network. China's road-related standards divide the use-level of road networks into 5 levels, so 5 state vectors were defined in the BP neural network. The status of a large number of roads over the years was turned into a 5-dimensional time series. The BP neural network was built on the MATLAB software platform. Many methods were used to improve prediction precision, such as Bayesian methods, repeat training, and so on. The prediction method was ameliorated too. Combined with factual data of road conditions in Tianjin, through the comparison of direct prediction and gradually rolling prediction, the results showed that gradually rolling prediction offered a better prediction precision and was more applicable to use in the use-level of road network prediction. Using gradually rolling prediction, this paper predicted and analyzed the changes of the use-level of road network in Tianjin.