Training a Model Using Deep Learning to Predict Traffic Flow Intensity

Transport infrastructure is an important element in the development of not only a locality, but also the country as a whole. To achieve maximum effect in this direction, it is necessary to analyse the criteria for the quality of the street and road network (SRN). The most important of them is the intensity of traffic flow. This indicator is dynamic and changes under the influence of a large number of characteristics of the Driver – Car – Road – Environment (DCRE) system.
A complex system of interaction between DCRE elements makes it difficult to accurately model the intensity of traffic flow, therefore, automation of data processing processes is required.
Deep learning is one of the most effective tools in the field of machine learning and artificial intelligence, allowing creating models of high forecasting accuracy. Using neural networks, a detailed system of relationships between individual components of the influence of independent variables on the dependent one is formed. Having generalised the experience of a large number of specialists in the subject area, not only the structure of the neural network was developed, but also a program for its implementation, providing for model training and forecasting. This approach to assessing the intensity of traffic flow was used for the first time and is a scientific novelty of the project.
The objective of the work was to study the training of high-quality models using the deep learning method to predict the dependence of traffic flow intensity on weather conditions and SRN characteristics.
The paper considers the process of training models on data from the traffic flow control lines of the Integra KDD system in Khabarovsk. The principles and rules of the deep learning method are analysed to predict the dependent variable. Amodel was trained for each control line, as a result of which it became possible to comprehensively assess the intensity of traffic flow. The dependent variable is characterised not only by weather and natural conditions, but also by the geometric elements of transport structures.
The quality of the models was tested on pre-processed data that was not involved in the training. The significance of the obtained results, which can be an important contribution to the development of traffic management systems and improving the efficiency of road infrastructure, was proven graphically

1. Pugachev, I. N., Sheshera, N. G., Grigorov, D. E. Traffic Flow Intensity Research Based on Deep Learning. World of Transport and Transportation, 2024, Vol. 22, Iss. 2, pp. 12–24. DOI: https://doi.org/10.30932/1992-3252-202422-2-2.

2. Babkov, V. F. Road conditions and traffic safety [Dorozhnie usloviya i bezopasnost dvizheniya]. Moscow, Transport publ., 1993, 271 p. ISBN 5-277-01402-0.

3. Dujuan Wang, Jiacheng Zhu, Yunqiang Yin, Joshua Ignatius, Xiaowen Wei, Ajay Kumar. Dynamic travel time prediction with spatiotemporal features: using a GNN-based deep learning method. Annals of Operations Research, 2023, Vol. 340, pp. 571–591 DOI: 10.1007/s10479-023-05260-2.

4. Jiayu Liu, Xingju Wang, Yanting Li, Xuejian Kang, Lu Gao. Method of evaluating and predicting traffic state of highway network based on deep learning. Journal of Advanced Transportation, 2021, Vol. 2021, Article ID 8878494, 9 pages. DOI: 10.1155/2021/8878494.

5. Cozac, E. B. Developing a Hybrid Neural Network with Local Neural Units for Collaborative Filtering Tasks. Industrial Automatic Control Systems and Controllers, 2021, Iss. 9, pp. 19–29. DOI: 10.25791/asu.9.2021.1309.

6. Schwetz, O., Smakanov, B., Kovac, L., George, G. Neural network development and training for recognizing the state of a driver. Bulletin of KazATC, 2023, Iss. 2 (125), pp. 186–195. DOI: 10.52167/1609-1817-2023-125-2-186-195.

7. Malakhova, V. V., Malakhov, O. V. Analysis of statistical data using the mathematical apparatus of artificial intelligence [Analiz statisticheskikh dannykh s ispolzovaniem matematicheskogo apparata iskustvennogo intellekta]. Bulletin of the Vladimir Dahl Lugansk State University, 2023, Iss. 11, pp. 177–179. EDN: EADYTE.

8. Khusainov, R. M., Talipov, N. G., Katasev, A. S. Neural network convolutional model for recognizing road infrastructure objects in intelligent transport systems [Neirosetevaya svertochnaya model raspoznavaniya obektov dorozhnoi infrastruktury v intellektualnykh transportnykh sistemakh]. Bulletin of NCBZhD, 2023, Iss. 4 (58), pp. 72–79. EDN: HOHQGY.

9. Pugachev, I. N., Grigorov, D. E., Sheshera, N. G. Analysis of geometric elements of roads when assessing their accident rate by means of modern geoinformational systems. Bulletin of civil engineers, 2021, Iss. 3 (86), pp. 127–133. DOI: 10.23968/1999-5571-2021-18-3-127-133.

10. Pugachev, I. N., Skripko, P. B., Sheshera, N. G. Program approach to integrated collection and data production on vehicle intensity movements, weather conditions and natural light in hourly intervals. T-Comm: Telecommunication and transport, 2023, Vol. 17, Iss. 10, pp. 43–51. DOI: 10.36724/2072-8735-2023-17-10-43-51.

11. Koryagin, M. E., Medvedev, V. I., Shvets, Yu. V. Mathematical model of cars distribution along the road network according to the second wardrop principle based on the Greenshields model. Vestnik Rostovskogo gosudarstvennogo universiteta putei soobshcheniya, 2023, Iss. 4 (92), pp. 175–183. DOI: 10.46973/0201727X_2023_4_175.

12. Pugachev, I., Kulikov, Y., Markelov, G., Sheshera, N. Factor Analysis of Traffic Organization and Safety Systems. 12th International Conference «Organization and Traffic Safety Management in large cities», SPbOTSIC‑2016, 28–30 September 2016, St. Petersburg. St. Petersburg, 2017, pp. 529–535. DOI: 10.1016/j.trpro.2017.01.086.

13. Taghipour, H., Parsa, A. B., Chauhan, R. S., DerriblE, S., Mohammadian, A. K. A novel deep ensemble-based approach to detect crashes using sequential traffic data. IATSS Research, 2022, Vol. 46, Iss. 1, pp. 122–129. DOI: 10.1016/j.iatssr.2021.10.004.

14. Diachuk, M., Easa, S. M. Motion planning for autonomous vehicles based on sequential optimization. Vehicles, 2022, Vol. 4, Iss. 2, pp. 344–374. DOI: 10.3390/vehicles4020021.

15. Titong Jiang, Yahui Liu, Qing Dong, Tao Xu. Intention-aware interactive transformer for real-time vehicle trajectory prediction in dense traffic. Transportation Research Record, 2023, Vol. 2677, Iss. 3, pp. 946–960. DOI: 10.1177/03611981221119175.

16. Porubov, D. M., Rodin, A. A., Pinchin, A. V., Zarubin, D. N., Tumasov, A. V., Kulepov, V. F., Orlov, L. N. Experimental studies of traffic lane control systems based on neural networks. Proceedings of NSTU n.a. R. E. Alekseev, 2022, Iss. 1 (136), pp. 137–147. DOI: 10.46960/1816210X_2022_1_137.

17. Baluta, V. I., Osipov, V. P., Rykov, Yu. G., Chetverushkin, B. N. On the Concept of Influence in the Concept of Cognitive Modeling when Using the Activation Function of the ReLU Type. Journal of Information Technologies and Computing Systems, 2023, Iss. 4, pp. 59–71. DOI: 10.14357/20718632230406.

18. Boyarshinov, M. G., Vavilin, A. S. The Impact of Traffic Jams on the Duration of Vehicle Movement on a Limited Section of the Road. Modernization and Scientific Research in the Transport Complex, 2022, Vol. 1, pp. 216–220. EDN: XJGFEO.

19. Lysov, G. M., Prikhodko, F. N., Konovalova, A. A., Timoshenko, K. A. Research of the method of forecasting time series in transport using recurrent neural networks [Issledovanie metoda prognozirovaniya vremennykh ryadov na transporte s pomoshchyu rekurrentnykh neironnykh setei]. Science Diary, 2023, Iss. 1 (73). 3), Ser. No. 16. DOI: 10.51691/2541-8327_2023_1_6.

20. Alamir, H. S., Zargaryan, E. V., Zargaryan, Yu. A. Transport flow forecasting model based on neural networks for traffic prediction on roads. Izvestiya SFedU. Engineering sciences, 2021, Iss. 6 (223), pp. 124–132. DOI: 10.18522/23113103-2021-6-124-132.

21. Shlenskih, D. A., Belokopytov, M. L., Anohin, D. V. A method of data synthesis to improve the effectiveness of neural network training. Journal of Radio Electronics, 2024, Iss. 3, Ser. No. 10. DOI: 10.30898/16841719.2024.3.8.

22. Pugachev, I. N., Sheshera, N. G. The influence of the longitudinal slope on road accidents with injuries [Vliyanie velichiny prodolnogo uklona na DTP s travmatizmom]. Science and Technology in the Road Industry, 2020, Iss. 3 (93), pp. 4–7. EDN: HLQTEZ.

23. Pugachev, I. N., Sheshera, N. G., Kamenchukov, A. V. Improving methods for assessing the quality and safety of road traffic [Sovershenstvovanie metodov otsenki kachestva i bezopasnosti dorozhnogo dvizheniya]. Khabarovsk, Publishing house of Pacific State University, 2018, 160 p. ISBN: 978-5-7389-2708-9.

24. Pugachev, I. N., Sheshera, N. G. Application of the technique of coefficients of the injury in order to control the quality of future and existing roads. Quality and life, 2016, Iss. 1 (9), pp. 58–61. EDN: TKUNAG.

25. Paklin, N. Logistic Regression and ROC Analysis – Mathematical Apparatus. BaseGroup Labs. Data Analysis Technologies: website. Ryazan, 2017. [Electronic resource]: https://basegroup.ru/community/articles/logistic. Last accessed 25.04.2024.