Electro-hydrostatic actuator (EHA) systems are essential for all-electric aircraft. Due to the inherent randomness and sparse knowledge, hybrid uncertainties including aleatory and epistemic uncertainty inevitably coexists in EHA systems. Conventional control performance design cannot evaluate the entire uncertainty space while wasting time and money formulating experimental environments. Digital twin (DT) metamodeling is an efficient approach to accelerating the design of EHA under uncertainty by conducting experiments in virtual space rather than physical space. In this article, a Gaussian process (GP)-based DT metamodeling is proposed, and the probability-boxes (p-boxes) are invited for hybrid uncertainty quantification. An adaptive sampling strategy is developed for the information interaction between virtual and reality. The trained GP model is then used for obtaining optimal control performance of EHA systems.