Comparative performance evaluation of optimization algorithms for run-of-river hydroelectric power plant production forecasting

This study investigates instantaneous Run-of-River Hydroelectric Power Plant (RoRHPP) production prediction optimization using meteorological data decomposed through Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN). Real-time meteorological and RoRHPP production data from Rize Province, Turkey, were utilized to develop and compare machine learning and optimization approaches for power forecasting. The methodology decomposed meteorological parameters (wind speed, wind direction, solar radiation, humidity, cloud cover, air temperature, and rainfall) into 82 Intrinsic Mode Function (IMF) signals using CEEMDAN. Seven machine learning algorithms were evaluated: Random Forest, XGBoost, Decision Tree, Polynomial Regression, Linear Regression, Ridge Regression, and LASSO Regression. Random Forest demonstrated superior performance with R2 values of 0.93 and 0.50 for training and test datasets, respectively. Five optimization algorithms determined optimal signal subsets from the 82 IMF signals: Artificial Bee Colony (ABC), Ant Colony Optimization (ACO), Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Firefly Algorithm (FA). Using Root Mean Square Error (RMSE) as fitness parameter, iterations stopped after 30 consecutive iterations without improvement. The CEEMDAN-ACO-RF method achieved best performance with R2 = 0.98. Compared to non-optimized CEEMDAN-RF, ACO demonstrated significant improvements: 18.18% MAE reduction, 12.04% RMSE reduction, and 0.61% R2 increase. Analysis of 1752 h of real data confirmed ACO’s superiority over other optimization algorithms for RoRHPP production prediction.