An artificially stimulated reservoir volume (SRV) is critical to the geothermal energy extraction from natural dry hot rocks with low porosity and low permeability. However, the characterization of SRV shape and permeability distribution is not well solved due to the geological data availability and in-situ weak control of permeability enhancement. For simplicity, most of research papers assume a homogeneous cuboid space domain. This study introduces two new improvements: First, the SRV is constructed based on micro-seismic data. Three SRVs with the same volume but different features are preset and their energy productions are qualitatively compared through numerical simulations. Second, the spatial permeability distribution is scaled into the global certainty induced by artificial permeability enhancement and the local uncertainty in natural randomness. This distribution is introduced into the stochastic simulations on geothermal extraction. Third, a thermal-hydraulic-mechanical (THM) coupling model is developed for the heat production simulation on 500 cases with possible permeability distributions. Finally, the statistical features of energy production uncertainty are evaluated through the simulation results of these 500 cases. It is found that the permeability heterogeneity in SRV domain has a significant impact on heat production. The assumption of cuboid SRV domain will overestimate the energy production from a geothermal system. Ignorance of statistical local uncertainty will make the construction of enhanced geothermal system in low permeability dry hot rock reservoirs face a high-risk adventure with current stimulation technology.