The Prediction Method Research Of Traffic Volume At Non-detector Intersections

The Urban Traffic Flow Guidance System, which is the one of important content of the Intelligent Transportation Systems, by the organic unity of travelers, vehicles and roads and dynamic allocation of traffic flow, can effectively forestall traffic congestion and reduce traffic accidents and environmental pollution.In the Urban Traffic Flow Guidance System, the historical data of traffic flow which were acquired by detectors can only characterize the state of past traffic flow, and which directly be used for traffic guidance in short future will lead the decision-making to lagging behind. Therefore, it is obvious that the prediction of traffic flow is of importance.For forecasting traffic flow at non-detector junctions, the general ways reference the classic forecasting methods and are low prediction accuracy. In response to this situation, this paper proposes the research task of model formulation and solution algorithm about traffic flow forecasting at non-detector intersections that designs to improve forecast accuracy and reduce time for computation the model parameters. Because there are limited historical data, the paper selects the class method of causal regression analysis and combines with the contents of genetic algorithm and fuzzy theory and proposes two new integrated prediction methods.Firstly, the paper introduces the classical regression models and fuzzy regression prediction models, summarizes solution algorithms of different models, and gives program realization under the MATLAB programming environment. Secondly, through the research above-mentioned, an integrated approach is proposed which is based on ridge regression and improved fuzzy least squares model. The method is that using the historical moment and the downstream traffic flow at having-detector junctions as predictors, applying the ridge regression analysis to select the key predictors as the model input variables, which significantly affect the non-detector junctions' traffic flow, and applying the improved Diamond model which is more realistic than accurate models, because of ambiguity of the artificial observation data of traffic flow. And then through comparing with Tanaka's LP method and cumulative method in traffic flow forecasting experiments, comparison show that the new integrated approach has higher accuracy. Thirdly, through the research on genetic algorithm and fuzzy least absolute deviation model, a new GA-FLAD algorithm is proposed. A application traffic flow forecasting has been done by using the new algorithm through MATLAB genetic algorithm toolbox, the results of which show that fuzzy least absolute deviation method has better robust nature than the classical methods, and genetic algorithm solving the coefficients of fuzzy least absolute deviation model has higher accuracy and shorter computation time compared with the traditional methods, such as cumulative method, exhaustive method and so on.