Prospects of application of artificial neural networks for forecasting of cargo transportation volume in transport systems

The purpose of research – to identify the prospects for the use of neural network approach in relation to the tasks of economic forecasting of logistics performance, in particular of volume freight traffic in the transport system promiscuous regional freight traffic, as well as to substantiate the effectiveness of the use of artificial neural networks (ANN), as compared with the efficiency of traditional extrapolative methods of forecasting. The authors consider the possibility of forecasting to use ANN for these economic indicators not as an alternative to the traditional methods of statistical forecasting, but as one of the available simple means for solving complex problems.

Materials and methods. When predicting the ANN, three methods of learning were used: 1) the Levenberg-Marquardt algorithm-network training stops when the generalization ceases to improve, which is shown by the increase in the mean square error of the output value; 2) Bayes regularization method - network training is stopped in accordance with the minimization of adaptive weights; 3) the method of scaled conjugate gradients, which is used to find the local extremum of a function on the basis of information about its values and gradient. The Neural Network Toolbox package is used for forecasting. The neural network model consists of a hidden layer of neurons with a sigmoidal activation function and an output neuron with a linear activation function, the input values of the dynamic time series, and the predicted value is removed from the output. For a more objective assessment of the prospects of the ANN application, the results of the forecast are presented in comparison with the results obtained in predicting the method of exponential smoothing.

Results. When predicting the volumes of freight transportation by rail, satisfactory indicators of the verification of forecasting by both the method of exponential smoothing and ANN had been obtained, although the neural network showed the best result (the average relative forecast error was 8.97% for ANN and 11.21% for the method of exponential smoothing, respectively). This can explained by the fact that the temporal dynamic range of the values of the volumes of cargo transportation by this type of transport, for the period under review, has a nonlinear but uniformly changing character. In the case of forecasting the volumes of cargo transportation by road, the time series of initial values for the reporting period is simultaneously non-linear and unevenly changing. This explains the large values of forecasting errors by the method of exponential smoothing (the average relative forecast error of 47.47% for methods of exponential smoothing ); the forecast error with ANN was 13.97%, therefore the results of the prediction obtained by the method of exponential smoothing are considered unsatisfactory, and for ANN – satisfactory.

The conclusion. The results of the study confirm the feasibility of using trained artificial neural networks in forecasting the volumes of freight traffic with different cargo flows that have the initial statistical data of which have an uneven nonlinearly changing character in the time dynamic series. A sufficiently high verification in the application of ANN for difficult-to-forecast indicators of the transport process confirms the practical significance of the application of this method in the modeling of the logistics network. 

1. Osovskiy, S. Neyronnye seti dlya obrabotki informatsii. Per. s pol’skogo I.D. Rudinskogo. Moscow: Finansy i statistika, 2004. 344 p. (In Russ.)

2. Koskela T. Neural network methods in analyzing and modelling time varying processes Espoo, 2003. pp. 1–72. URL: http://citeseerx.ist.psu.edu/ viewdoc/summary?doi=10.1.1.77.5993 (paid access to article).

3. Tadeusevich R., Borovik B., Gonchazh T., Lepper B. Elementarnoe vvedenie v tekhnologiyu neyronnykh setey s primerami programm. Per. s pol’sk. I.D. Rudinskogo. Moscow: Goryachaya liniya Telekom, 2011. 408 p. (In Russ.)

4. Fedorov E.E. Iskusstvennye neyronnye seti. Krasnoarmeysk: DVNZ «DonNTU», 2016. 338 p. (In Russ.)

5. Merkulova Yu.V. Situatsionno-strategicheskoe planirovanie v ekonomike: monografiya v 2-kh tomakh. Vol. 2. Modelirovanie optimal’nykh strategiy i programm. Moscow: Ekonomi-ka, 2015. 464 p. (In Russ.)

6. Vladimirova L.P. Prognozirovanie i planirovanie v usloviyakh rynka. Moscow: Dashkov i K, 2012. 308 p. (In Russ.)

7. Pamula T. Neural networks in transportation research recent applications. Transport problems Vol. 11, Issue 2, pp. 27–36. 2016. URL: http:// transportproblems.polsl.pl/pl/Archiwum/2016/ zeszyt2/2016t11z2_03.pdf.

8. Gosasang V., Chandraprakaikul W., Kiattisin S. An Application of Neural Networks for Forecasting Container Throughput at Bangkok Port. Proceedings of the World Congress on En-gineering. Vol 1, WCE 2010, June 30 July 2, 2010, London, U.K. URL: http://www.iaeng.org/publication/ WCE2010/WCE2010_pp137-141.pdf

9. Schmidhuber, J. Deep learning in neural networks: An overview. Neural Networks. Vol. 61. P. 85–117. 2015. URL: http://www.sciencedirect. com/science/article/pii/S0893608014002135 (paid access to article).

10. Wasserman P.D. Advanced methods in neural computing. New York: Van Nostran Reinhold, 1993. 240 p.

11. Kostina L.N., Gareeva G.A. Neyronnye seti v zadachakh prognozirovaniya vremennykh ryadov. Mezhdunarodnyy zhurnal «Innovatsionnaya nauka». 6. 2015. P. 70–73. (In Russ.)

12. Jiang C. and Song F. Sunspot Forecasting by Using Chaotic Time-series Analysis and NARX Network. JCP. 2011. 6(7). P. 1424–1429. URL: http://journaldatabase.info/articles/sunspot_forecasting_by_using_chaotic.html (paid article).

13. Zhang X. and Frey R. Improving ARMAGARCH forecasts for high frequency data with regime-switching ARMA-GARCH. Journal of Computational Analysis & Applications. 2015. 18(1). URL: https://www.researchgate.net/publication/228142366

14. Kambouroudis D.S., McMillan D.G. and Tsakou K. Forecasting Stock Return Volatili-ty: A Comparison of GARCH, Implied Volatility, and Realized Volatility Models. Journal of Fu-tures Markets. 2016. 36(12). P. 1127–1163. URL: https://cronfa.swan.ac.uk/Record/cronfa34904 DOI: 10.1002/fut.21783

15. Domashchenko D.V., Nikulin E.E. Prognozirovanie ryadov dinamiki rynochnykh indikatorov na osnove nelineynoy avtoregressionnoy neyronnoy seti. Statistika i Ekonomika. 2017. 3. P. 4-9. DOI: 10.21686/2500-3925-2017-3-4-9 (In Russ.)

16. Grachev A. V. Kiseleva T. V., Dobrynin A.S., Koynov R.S. O metode vybora promezhutochnykh uzlov peredachi dannykh dlya marshrutizatsii v ierarkhicheskikh setyakh raznoy topologii. Ekonomika, statistika i informatika. 2014. No. 5. P. 161–164 URL: https://cyberleninka.ru/article/n/o-metodevybora-promezhutochnyh-uzlov-peredachi-dannyh-dlya-marshrutizatsii-v-ierarhicheskih-setyahraznoy-topologii-1(In Russ.)

17. Bragina E.I., Dekatov D.E., Egorova I.E. Razrabotka metodologii podderzhki prinyatiya resheniy v strategicheskom upravlenii predpriyatiem v usloviyakh neopredelennosti. Internet-vestnik VolgGASU. Seriya «Stroitel’naya informatika». 2013. No. 9 (26). P. 1–5. URL: http://vestnik.vgasu.ru/ attachments/DekatovEgorovaBragina1-2013_9(26). pdf (In Russ.)

18. Komleva N.V., Khlopkova O.A. Obrabotka kontenta v informatsionnykh sredakh na osnove neyronechetkoy modeli prinyatiya resheniy. Statistika i Ekonomika. 2013. No. 5. P. 188–192. URL: DOI: 10.21686/2500-3925-2013-5-188-192 (In Russ.)

19. Gunasekaran, M. and Ramaswami, K.S. A Fusion Model Integrating ANFIS and Artificial Immune Algorithm for Forecasting Indian Stock Market (June 22, 2011). Journal of Ap-plied Sciences, 11(16): pp. 3028-3033. URL: https://papers. ssrn.com/sol3/papers.cfm?abstract_id=2335390 (In Russ.)

20. Al’-Bareda Ali Yakh’ya Senan, Pupkov K.A. Algoritm resheniya zadachi sinteza upravleniya metodom iskusstvennykh neyronnykh setey. Vestnik RUDN, seriya Inzhenernye issledovaniya. 2016. No. 2. P. 7–16. URL: https://cyberleninka.ru/ article/n/algoritm-resheniya-zadachi-sinteza-upravleniya-metodom-iskusstvennyh-neyronnyh-setey (In Russ.)

21. Al’-Bareda A.Ya.S., Pupkov K.A. Reshenie zadachi sinteza sistemy upravleniya metodom iskusstvennykh neyronnykh setey. Fundamental’nye issledovaniya. 2015. No. 11 (Part 5). P. 853—857. URL: https://fundamental-research.ru/ru/article/ view?id=39521 (In Russ.)

22. Long T.B., Thai L.H., Hanh T. Face Recognition Using Circularly Orthogonal Moments and Radial Basis Function Neural Network & Genetic Algorithm. 2012 7th IEEE Confer-ence on Industrial Electronics and Applications (ICIEA). P. 523—527. URL: http://conference.researchbib. com/view/event/

23. Huang Jeng-Tze, Tseng Ming-Lei. Global Adaptive Neural Tracking Control of Strict-Feedback Systems with Bounded Uncertainty. 2012 7th IEEE Conference on Industrial Electron-ics and Applications(ICIEA). P. 175—180. URL: http:// conference.researchbib.com/view/event/

24. Diveev A.I., Shmal’ko E.Yu. Variatsionnyy geneticheskiy algoritm dlya resheniya zadachi optimal’nogo upravleniya. Sovremennye problemy nauki i obrazovaniya. 2014. No. 1. URL: http:// www.science-education.ru/115-11474 (In Russ.)

25. Averkin A.N. Gibridnye modulyarnye neyronnye seti. Statistika i Ekonomika. No. 4. 2016. P. 8–11. DOI:10.21686/2500-3925-2016-4-8-11 (In Russ.)