Due to the high-order model of photovoltaic and wind power generation systems, it is complicated to accurately establish the detailed state-space model of the multi-source renewable energy microgrid (MG) system. In addition, when the MGs are interconnected into microgrid clusters (MGCs), the difficulties of the control and stability analysis are greatly increased. However, the fast and slow dynamics of power electronic interface-based units may not be sufficiently separated on time-scale, which cannot directly follow the assumptions of the traditional model reduction method. This paper selects a typical grid-forming hybrid renewable energy MGC, where the multi-timescale characteristics of the system considering detailed electromagnetic and electromechanical transient modes are analyzed. An identification method without trial-and-error searching for coupling dynamics is proposed under the non-classical singular perturbation characteristics, which is unreported in previous studies and is different from the traditional power systems. Moreover, the reduced-order model can characterize multi-timescale while guaranteeing computational efficiency, which is able to further perform the key parameter optimization and stability analysis for the larger-scale MGCs. The theoretical analysis and the time-domain simulations verify the feasibility and the accuracy of the reduced-order model.