The increased installation of photovoltaic (PV) systems raises the necessity for the development of Digital Twins (DT), in order to simulate the PV power output and ensure the normal operation of the systems. To this point, at the present paper we propose a data-driven DT based on a hybrid ensemble FFNN-LSTM model. To examine the efficiency of the proposed model, we compare its simulation accuracy to a physical based DT, i.e., Single Diode Model (SDM), and data driven DTs, i.e., Support Vector Machines (SVM), Feed Foreword Neural Networks (FFNN), Long Short-Term Memory (LSTM) model. Additionally, in terms of data-driven DTs, two different scenarios have been implemented for the selected inputs (solar irradiance and panels' temperature). The results highlight that the proposed model has the highest simulation accuracy, since it leads to lower MAE compared to SDM, while at the same time its accuracy is comparable to the simple LSTM model.