The energy coming from the motion of the waves of seas and oceans could be an important component in the solution of the energy problem related to the pursuit of alternatives to fossil fuels. However, wave energy is still technologically immature and it has not reached the economic feasibility required for economy of scale. One of the major technological challenges for the achievement of this goal is the development of control strategies capable of maximizing the extracted energy, adapting to the conditions of the seas and oceans that surround the Wave Energy Converter (WEC) devices. To perform this task, control systems often adopt explicitly control-oriented models, that are by nature affected by uncertainties. On the contrary, to address the problem a data-driven solution is proposed here. The presented strategy applies an optimization approach based on a Gaussian Process Regression (GPR) metamodel to learn the control strategy to be applied. In order to accelerate the learning process, we present a novel method that exploits in the initial phase a previous knowledge given by simulations with the system model and based on the co-kriging concept. To test this approach the Pendulum Wave Energy Converter has been adopted as a case study. To differentiate the previous knowledge and the real system behaviour, a simplified linear model is used to obtain the prior knowledge, while a complex nonlinear one acts as the environment in which simulate the behaviour of the real system. A month-long simulation is used to validate the effectiveness of the proposed strategy, showing the ability of adapting to a real system different from the simplified model on the basis only of data, and overcoming the model-based strategy in terms of performance.