INTELLIGENT MODEL FOR PREDICTING THE TRAFFIC INTENSITY OF VEHICLES AT THE INTERSECTION
I. P. Bolodurina1, L. M. Antsiferova2, L. S. Grishina3
Orenburg State University, Orenburg, Russia
1 e-mail: ipbolodurina@yandex.ru
2 e-mail: antsiferova_68@mail.ru
3 e-mail: grishina_ls@inbox.ru
Abstract. The active development of the fleet of vehicles contributes to an increase in the congestion of transport networks and requires the construction of new systems for determining the places of congestion, as well as the development of ways to solve the existing problems. A special place in this area is occupied by an early assessment of the condition of transport networks, which allows timely measures to be taken to modify them. In this regard, research in the field of forecasting the main indicators of traffic flow in order to identify congestion of transport networks has become highly relevant. The use of data mining technologies, including methods of multidimensional linear regression, will make it possible to build predictive models for studying the characteristics of traffic flow. Aim. To build a model for predicting traffic intensity at an intersection using machine learning methods for effective decision-making when managing traffic flow. The scientific novelty of this study lies in the development of a multidimensional regression model for predicting traffic intensity, considering the control of retraining based on open data on observations of the number of vehicles at four different sections of the intersection. Methods. A multidimensional linear regression model was used to predict the intensity of traffic flow. Ridge, Lasso and ElasticNet regularization approaches were used as methods to improve the forecasting efficiency. Results. A model for predicting the intensity of traffic flow is constructed. As part of the experimental study, an assessment of the effectiveness of the use of regularization methods was carried out, as well as a comparative analysis of the accuracy of forecasting the model based on the initial and normalized data. Conclusion. The smallest root-mean-square error on the test data was shown by the multidimensional linear regression model with Ridge regularization. The developed model makes it possible to predict the number of vehicles passing per unit of time with an average quadratic error equal to 0.638. The obtained results of the study will allow early diagnostics of the occurrence of congestion of transport networks to optimize the movement of vehicles at various intersections of the UDS. The results of the study are of high practical importance, as they can be implemented into existing traffic management systems for effective decision-making in traffic flow management. The direction of future research includes the practical testing of the predictive model in real conditions, as well as the consideration of ensemble methods of machine learning to improve the accuracy in predicting the main indicators of the traffic flow.
Key words: traffic flow, traffic intensity, traffic management systems, machine learning methods, multidimensional linear regression, regularization.
Acknowledgements: the work was supported by a grant from the Russian Foundation for Basic Research (No. 20-07-01065 “A”), as well as a scholarship from the President of the Russian Federation for young scientists and postgraduates (No. SP-3652.2021.5).
Cite as: Bolodurina, I. P., Antsiferova, L. M., Grishina, L. S. (2022) [Intelligent model for predicting the traffic intensity of vehicles at the intersection]. Intellekt. Innovacii. Investicii [Intellect. Innovations. Investments]. Vol. 6, pp. 69–78, https://doi.org/10.25198/2077-7175-2022-6-69.
