Forecasting Method For Key Operation Indicators Of Urban Road Based On Massive Data

Operation indicators to evaluate a city road are very important for urban traffic conditions. Accurately, many researches are focus on the efficient, fast, and reasonable prediction of these indicators at present. It can be used to analyze whether the urban road planning is reasonable, and can provide reasonable travel tips to avoid continuously road congestion. In all related operation indicators for the road running, travel time and traffic flow are two of the most important basic indicators, whose role in the whole metric system is particularly critical.With rich traffic data collection ways, it provides sufficient data foundation to monitor and predict for urban road running indicators. How to use these new real-time monitoring data for forecasting the road travel time and traffic flow indicators becomes a crucial problem. In this paper, we forecast these two indicators through analyzing the changes in time and space through the vehicle timing based on urban roads camera license plate recognition data collected. Since the license plate recognition data sets are large scale, how to efficiently handle huge amounts of data, how to accurately calculate the travel time and traffic flow were measured, and how to achieve higher accuracy prediction on the basis of measured results become the key issues for travel time and vehicle traffic prediction.To solve the above problems, the main work and contributions are as follows:We propose a travel time prediction method based on Kalman filter method. This method is based on license plate recognition data set. And we completed short time prediction the path travel time using the Kalman filter method. Compare the predicted results with the historical average method linear regression method, the prediction accuracy of the results shows a significant extent.We propose a method for improving traffic flow forecasting model based on Gray Model. The license plate recognition method based on the original data set, based on the definition given on traffic, measured by pre-computing statistics on traffic using the measured result of applying improved residual Gray Model in a short time traffic predictions. Gray model results with experimental data sets used for verification, compared to the classic Grey Model, the accuracy of residual Grey Model algorithm can improve in some degree.The Design and implementation of the two predictor methods are based on Hadoop distributed computing environment, and we verify the effectiveness of the method through sampling and prediction experiments based on a mega-based real license plate recognition data set. Comparing to the traditional travel time prediction method, the prediction accuracy of our method has significantly improved, and the relative error is reduced by about 30%; and comparing to traffic flow prediction methods of classical gray model, the prediction accuracy of our method has a certain improvement, and the relative error decreases about 8%.