In view of thermal conductive network (TCN) conversion from high thermal dissipation to low thermal dissipation, a new theory of thermal conductivity (TC) enhancement was proposed in this paper. The TC enhancement effect was greatly limited due to the formation of an uncompacted self-assembly network in the composites prepared by traditional compounding methods. In order to achieve the goal of TCN conversion from high thermal dissipation to low thermal dissipation, we combined the TCN densification through space restriction forced network assembly (SCFNA) with adding rigid particles. A composite system of polydimethylsiloxane/short carbon fiber/glass bubble (PDMS/SCF/GB) was selected as the modelling object and used to perform verification experiments. On the basis of Nan model, the TCN conversion mathematical model of TC prediction was established. The effects of volume fraction, thermal boundary resistance, thermal contact resistance, aspect ratio and distribution state of fillers on the effective thermal conductivity (ETC) of this PDMS composite were all systematically studied and discussed. In order to verify the accuracy of the model established to predict the TC of PDMS/SCF/GB composite during TCN conversion, a series of experiments were carried out to compare with the model results.