Conventional plasma absorbers are challenging to obtain high electron density and sizeable spatial scale for effective absorption while meeting the applied requirements of low profile and low power consumption. Although the frequency selective surface (FSS) has proved to realize a lower profile of plasma absorber with some empirical patterns adequately, the issue of the FSS design matching the dispersion distribution of complicated plasmas is still in suspense. A reverse prediction method referenced as the forecast and design Conditional Generative Adversarial Network (FD-CGAN) is proposed to generate a pixelated FSS between double-layer plasma periodic arrays. The reflection attenuation characteristics examined by experiments show that the addition of the FSS makes the coupling absorption effect surpass that of either pixelated FSS or plasma solely. Measurements in reconfigurable working modes and array arrangements demonstrate that the proposed configuration maximizes absorption effectiveness in the same profile, accompanied by the simulation. An interfacial void model is proposed to assist the design of the composite absorbing structure, together with an equivalent circuit for the hybrid absorber including periodic patterns with stochastic distribution characteristics, which analyze the absorption effect of the composite structure. The study provides a new approach for various microwave applications, including multilayer radarabsorbing structures, plasma-based stealth technology, and reconfigurable filters.