DOI – https://doi.org/10.46793/Vodoprivreda57.5-6.201PG
Key words – machine learning, reservoir scheme management, Python code, simulation models, water level fluctuation
SUMMARY
Management of water resources has come to a point where its complexity often cannot be resolved with use of existing empirical or theoretical methods [1] [2]. The complexity has arised with the ever increasing demands from water resources, primarely needs for clean drinking water, water for irrigation of agricultural fields, energy production, and flood protection on the other end [3]. Most reservoir schemes have become multi-user, complying with needs opposing eachother in quantity, quality and timing. In the past and at some extend at the present also, scheme operators use simple charts of data correlation on which decisions are made for release of water to a certain water user. However, many schemes are not single-reservoir but multi-reservoir, and decisions made on upper reservoirs impact the impoundment on the downstream reservoir by default. Therefore, scheme operators are facing incresed pressure on optimal operation with their schemes and release of water for downstream users. As a result, research has been conducted on possible use of machine learning algorithms to find a target parameter of interest, in the following case chosen as water level of the most downstream reservoir in a complex, multi-reservoir scheme.
Through the process of train and test, prediction on water level of a reservoir is conducted for the chosen scheme of three existing reservoirs in the lower valley of ‘Treska’ river located in Republic of North Macedonia. The scheme consists of ‘Kozjak’, ‘Sv. Petka’ and ‘Matka’ reservoirs, all existing and in function. ‘Kozjak’ is the most upstream reservoir, whereas ‘Matka’ is the most downstream reservoir of all, and the oldest dam in operation in North Macedonia, ever since 1938 [4].
Six machine learning methods are applied for analyses of the water level fluctuations in ‘Matka’ reservoir: Artificial Neural Networks (ANNs), Support Vector Machines/ Support Vector Regression (SVM/SVP), Random Forest (RF), Decision Trees (DT), Gaussian Process Regression (GPR) and Boosted regression trees (BT). The results are focused on combination of different parameters that impact the water level of the most downstream reservoir of the system as target parameter. Based on the analyses, the RF, DT, and BT models are recommended for further analyses of the system’s operations, as they demonstrated the best performance during both the training and testing phases, as well as in the prediction period.
Autori: Frosina PANOVSKA GEORGIEVSKA, Stevcho MITOVSKI, Ljupcho PETKOVSKI
REFERENCES
[1] Welsh, W. D., Vaze, J., Dutta, D., Rassam, D., Rahman, J. M., et al.: An integrated modelling framework for regulated river systems. Environmental Modelling & Software, 39, 81–102, 2013. Available at: https://doi.org/10.1016/j.envsoft.2012.02.022
[2] Lins, H. F.: The Imperative of Water Resources Assessment. WMO Bulletin, 57(3), World Meteorological Organization, 2008. Available at: https://wmo.int/media/magazine-article/imperative-of-water-resources-assessment
[3] World Bank: The New Economics of Water Scarcity and Variability. Washington, DC: World Bank, 2017
[4] Tanchev L.: Dams and appurtenant structures, 2nd edition, CRC Press, 2014
[5] Mitchell M. T., Machine Learning, McGraw-Hill Science/Engineering/Math, March 1997
[6] Ibrahim, D. An Overview of Soft Computing, Procedia Computer Science, Vol. 102, Pages 34-38, 2016. Available at: https://doi.org/10.1016/j.procs.2016.09.366
[7] USGS (2019) How Much Water is There on Earth? Available at: https://www.usgs.gov/special-topics/water-science-school/science/how-much-water-there-earth (Accessed: November 25, 2025)
[8] Hariri-Ardebili M., Pourkamali-Anaraki F., An Automated Machine Learning Engine with Inverse Analysis for Seismic Design of Dams, Water, Volume 14, 2002. Available at: https://doi.org/10.3390/w14233898
[9] Chen W., Wang X., Tong D., Cai Z., Zhu Y., Liu C.: Dynamic early-warning model of dam deformation based on deep learning and fusion of spatiotemporal features, Knowledge – Based Systems, Volume 233, December 2021. Available at: https://doi.org/10.1016/j.knosys.2021.10753
[10] Saavedra Valeriano, O. C., T. Koike, K. Yang, T. Graf, X. Li, L. Wang, and X. Han, Decision support for dam release during floods using a distributed biosphere hydrological model driven by quantitative precipitation forecasts, Water Resources Research, 46, W10544, 2010. Available at: https://doi.org/10.1029/2010WR009502
[11] Gunes Sen, S., Machine Learning-Based Water Level Forecast in a Dam Reservoir: A Case Study of Karaçomak Dam in the Kızılırmak Basin, Türkiye, Sustainability 2025, 17(18), 8378; Available at: https://doi.org/10.3390/su17188378
[12] HEC ResSim 3.3 Manual, USACE, 2021. Available at: https://www.hec.usace.army.mil/software/hec-ressim/documentation/HEC-ResSim_33_UsersManual.pdf
[13] James G., Witten D., Hastie T., Tibshirani R., An Introduction to Statistical Learning, 2021, Springer
