GIVP: A GRASP–ILS–VND METAHEURISTIC WITH PATH RELINKING APPLIED TO HYDROPOWER OPERATION SCHEDULING

Short-term operation of cascaded run-of-river hydropower plants is a challenging optimization problem: plants are hydraulically coupled, turbine efficiency curves are non-linear, and feasible schedules must respect reservoir levels, turbine flows, ramping rates, and unit commitment. We propose GIVP, a hybrid metaheuristic combining GRASP, ILS, VND, and Path Relinking to produce daily generation schedules. GRASP builds diverse initial solutions; ILS and VND refine them through structured neighborhood moves and perturbations; Path Relinking intensifies the search between elite solutions. The objective function penalizes constraint violations while maximizing generated power. Evaluated on real operational data from a three-plant cascade with seven generating units in southern Brazil, GIVP achieves the lowest objective-function cost in 8 of 14 test scenarios after Optuna hyperparameter tuning, outperforming Differential Evolution, PSO, and Unified PSO in solution quality at the cost of higher execution time. Results confirm hybrid local-search frameworks as a competitive approach for hydropower scheduling.