Optimized simulation of river flow rate using regression-based models

Given the extreme values of rainfall in recent years and the increase in floods, data-driven models must also be optimized to be able to simulate the maximum and minimum values of extreme events well. Therefore, in this study, to simulate the daily river flow rate in the upstream stations of Bukan reservoir in northwestern Iran and south of Urmia Lake, an optimized nonlinear support vector regression model has been used. This model is optimized by various algorithms such as antlion optimizer (ALO), ant colony optimizer (ACO), multiverse optimizer (MVO) and salp swarm algorithm (SSA) to provide better results. Based on this, the accuracy of the four algorithms was evaluated using different criteria. The number of iterations, training and testing sets were considered fixed in all four algorithms. The results showed that according to the Nash–Sutcliffe efficiency, the performance of all four algorithms in simulating the daily flow rate is acceptable, which is also confirmed in the violin plots presented in the two phases of training and testing. The error rate calculated by the root-mean-square error statistic showed that the error rate of the antlion algorithm is less than other studied algorithms. The antlion algorithm was able to reduce the simulation error by about 11% on average in all studied stations by optimizing the parameters of the support vector regression model. Compared to ACO, MVO and SSA algorithms, ALO algorithm was able to reduce the error rate of optimized nonlinear support vector regression model in simulation of daily river flow rate by 11.5, 12.4 and 9.3%, respectively, in the test phase in studied station. Also, with the development of support vector regression model, the maximum daily discharge points in the studied stations were simulated well, and a good correlation was observed between the observed and simulated values. The developed support vector regression model has no limitations in different regions and different climates due to the use of optimization algorithms.

---

Opracowanie rekordu ze środków MNiSW, umowa nr SONP/SP/546092/2022 w ramach programu "Społeczna odpowiedzialność nauki" - moduł: Popularyzacja nauki i promocja sportu (2024).