Data-driven configuration optimization of an off-grid wind/PV/hydrogen system based on modified NSGA-II and CRITIC-TOPSIS

•A data-driven framework is proposed to optimize the sizing of a hybrid energy system.•A modified NSGA-II based on reinforcement learning is utilized to obtain Pareto set.•CRITIC-TOPSIS is used to decide the weight of objectives and select the best solution.•A optimal system with LCOE of 0.226 $/kWh, LPSP of 4.01% and PAR of 2.15% is obtained. This paper proposes a data-driven two-stage multi-criteria decision-making (MCDM) framework to investigate the optimal configuration of a stand-alone wind/PV/hydrogen system. In the first stage, a modified non-dominated sorting genetic algorithm (NSGA)-II based on reinforcement learning is utilized to determine a set of Pareto solutions. The objectives considered are to minimize the levelized cost of energy (LCOE), the loss of power supply possibility (LPSP) and the power abandonment rate (PAR), simultaneously. In the second stage, the Criteria Importance Though Intercrieria Correlation (CRITIC) method is utilized to determine the weight of the three objectives, while the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) approach is employed to select the unique best solution from Pareto solutions. To verify the effectiveness, the framework is applied to the wind/PV/hydrogen system located in Aksay Kazak Autonomous County, Gansu Province, China to meet an off-grid industrial park’s load demand of 1603 kWh/day and peak load of 117.17 kW. The result states that the optimal system, which consists of 83.2 kW PV panels, 160 kW wind turbines, 20 kW fuel cells, 54 kW electrolyzers and 450 m3 hydrogen storage tanks, owns the LCOE of 0.226 $/kWh, the LPSP of 4.01% and the PAR of 2.15%.