Dynamic K-Nearest Neighbors (DKNN) Approach Optimized by PSO for Daily River Inflow Prediction

Ebrahimi, Ehsan; Shourian, Mojtaba

doi:10.22048/wss.2026.575197.1021

Dynamic K-Nearest Neighbors (DKNN) Approach Optimized by PSO for Daily River Inflow Prediction

Articles in Press

Document Type : Research Article

Authors

Ehsan Ebrahimi ¹

Mojtaba shourian ²

¹ Faculty of Civil, Water and Environmental Engineering, Shahid Beheshti University, Tehran, Iran

² Faculty of Civil Engineering, Water and Environmental Engineering, Shahid Beheshti University, Tehran, Iran

10.22048/wss.2026.575197.1021

Abstract

Streamflow forecasting is one of the most important components of water resources management. Accurate long-term forecasts are essential for planning water supply and storage, while short-term predictions are crucial for anticipating extreme flows and supporting flood warning systems. Data-driven models, as relatively simple yet powerful approaches, are widely used for streamflow prediction. The K-Nearest Neighbors (KNN) method is an effective non-parametric learning approach that has been applied to solve various problems. In this study, a novel neighbor selection method called Dynamic K-Nearest Neighbors (DKNN) is introduced. Using an SVM model, optimal distance thresholds are determined, and neighbors located within this optimal range are selected for each prediction case. The performance of the proposed method was evaluated using two years of daily inflow data from the Gheshlagh Dam in western Iran. The results indicate that the proposed method improves prediction accuracy by reducing the overall error (RMSE) by 6%, with the improvement reaching 7.8% in predicting extreme events.

Keywords

KNN

Dynamic

SVM

Data Driven

Subjects