Optimal Operation of Distribution Network With Hydrogen Storage and Fuel Cells

Hydrogen energy promotes sustainable and clean energy and facilitates better utilization of renewable energy sources. The advance in power to hydrogen technology, such as electrolyzers, causes the future shift toward hydrogen storage as a promising technology to store the excess generation in the distribution network. However, the operating decisions of power distribution networks with renewable energy and power to hydrogen technology are affected by the stochastic nature of generation and load, which calls for sophisticated, more advanced optimization techniques to determine the optimal operating mechanism under these uncertain factors. Moreover, the centralized approach to solving such a problem requires an extensive communication infrastructure and processing of large system data. This paper proposes an advanced optimization methodology to optimize the operation of distribution networks with hydrogen storage and fuel cells, aiming to reduce the communication burden and address generation and load uncertainties. The problem is formulated as a distributionally robust optimization in which the ambiguity set is constructed based on Wasserstein distance to address the uncertainties in the electricity/hydrogen demands, solar photovoltaic (PV) generation, and hydrogen reserve. The proposed formulation coordinates the optimal schedule of the generation assets in the integrated power distribution-hydrogen systems using Benders decomposition under the worst probability distribution of uncertain variables. The modified IEEE 34-bus distribution network with two hydrogen storage systems is used to evaluate the performance of the proposed algorithm. The numerical results show that the total expected operation cost in the proposed DRO algorithm is 1.98% higher than those obtained using a deterministic approach due to considering the generation and demand uncertainties. Moreover, the out-of-sample performance shows that the proposed DRO algorithm obtained 0.48% lower expected operation cost compared to the stochastic programming counterpart when different scenarios are tested.