TimeGAN based distributionally robust optimization for biomass-photovoltaic-hydrogen scheduling under source-load-market uncertainties

Abstract

The Biomass-Photovoltaic-Hydrogen Integrated Energy System (BPH-IES) is receiving increasing attentions due to the utilization of hydrogen storage for renewables accommodation. However, owing to its internal components and capability for participating in electricity and carbon markets, source-load-market uncertainties such as the random photovoltaic output, loads, as well as electricity and carbon prices are non-negligible, which seriously hinders the stable operation and realization of economic and environmental benefits for BPH-IES. In this paper, in view of the insufficient samples and the time-series attribute for the uncertainties, a novel Time-series Generative Adversarial Network (TimeGAN) based Distributionally Robust Optimization (DRO) model is proposed for BPH-IES scheduling. Based on TimeGAN learning the dynamics of the limited samples of uncertainties, sufficient generated samples are obtained with approximating the real distributions. Then, the temporal covariance conditions are introduced in the ambiguity set construction, aiming at getting rid of those distributions unmatching the temporal correlations of the enlarged samples. Finally, DRO model for scheduling is established, which jointly optimizes multi-energy operation and electricity-carbon trading strategy for BPH-IES considering the “worst-case” distributions within the narrowed ambiguity set. The solution procedure is developed according to the duality and semi-definite programming theories. Simulations have verified that for BPH-IES, the proposed model: 1) improves the economic and environmental benefits by 10.04% and 7.79% respectively, due to the TimeGAN and temporal covariance based modifications; 2) maintains operation stability and 100% renewables accommodation under the source-load uncertainties; 3) reduces the fluctuation of benefits via considering the market uncertainties.