Green Hydrogen Energy Storage Sizing Method Based on Adversarial Learning Technology Under Inaccurate Supervision

Weakly supervised adversarial learning represented by conditional generative adversarial network (CGAN) can account for power fluctuations and generate scenario sets to characterize wind uncertainty in energy storage stochastic planning. Therefore, this paper first designs a corrected-conditional generative adversarial network (CCGAN) and constructs a label corrector to identify and correct the label information of the input generator under the prediction misaim and wind power ramping events. The multi-objective capacity sizing optimization of the green hydrogen energy storage system (GH-ESS) is constructed. The mathematical concept of knee region based on Chebyshev distance is introduced to guide the design of a multi-objective optimization algorithm. Finally, a New England 39-bus system is taken as a case. The results show that the inaccurate supervision will lead to the deviation of planning decisions from reality, while CCGAN can generate a high confidence output scenario considering wind uncertainty, and the planning results can balance robustness and economy.