This dataset contains the computational results supporting the numerical analysis of the proposed QL-VNS. The experiments compare four solution approaches: QL-VNS, VNS, Simulated Annealing, and the exact solver Gurobi. The test instances vary in size and structural parameters (installations, periods, scenarios, and products) to assess scalability and robustness. For each instance and each method, 10 independent runs were executed. The dataset reports the objective value (total expected cost) and the corresponding CPU time for every run. The results presented in the associated paper are based on averages across these runs. This dataset enables full reproducibility of the computational study and provides a benchmark for future research on learning-enhanced metaheuristics and stochastic optimization under carbon regulation.