A Prediction Method Of Short-Term Traffic Flow Based On BNDs

Traffic congestion has become an important problem which restricts the positive operation of modern economic society. However, as a hot research of contemporary intelligent transportation systems, the real-time, accurate and efficient short-time traffic flow forecasting technique is the pressing needs of strong measures to easing traffic congestion and improve social benefits. It could make maximum use of urban road network capacity, reduce environmental pollution and energy consumption, and reduce the probability of traffic accidents. Therefore, a further research on short-term traffic flow forecasting methods has both theoretical and practical significance.Firstly, this paper recalled and summarized the past of research results, and then compared the forecast effects and the strengths and weaknesses of various short-term traffic flow forecasting methods. Secondly, accounting for the nonstationarity, non-linearity, and high-uncertainty of the short-term traffic flow sequence, a combined forecasting model based on wavelet theory and depth faith network had been introduced and boiled. Thirdly, to demonstrate the validity of the combined forecasting model, it had been applied to predict the short-term traffic flow about A38 road in England. Moreover, after comparing the combined forecasting model with other traditional models, this paper summarized the advantages and disadvantages and scopes of different models.Main work and conclusions of this study are as follows:(1)Based on the analysis of A38 road’s traffic flow data, we demonstrated that there are four main characteristics of traffic flow sequence: a pronounced cyclical variation, sequence correlations, aggregation effects and strong-uncertainties;(2)In order to improve the data quality, we used history substitution method to repair traffic data outliers and missing values, applied the wavelet analysis to eliminate the sequence noise, and took normalization method to eliminate scale differences.(3)Deep belief networks was applied to forecast short-term traffic flow by using the pretreatment data. More specifically, we used a restricted Boltzmann machine to be responsible for unsupervised learning, and BP back propagation neural network to supervise learning. In order to achieve the forecast goals, we adjusted the parameters of the model many times to make the prediction error converges. The prediction showed that,(4)We used neural networks to predict the same data. Results showed that deep belief networks better than other models in the forecast.