Green communication with sustainable energy is being considered for 5G cellular network and Internet of Things (IoT) mainly with focus on energy harvesting (EH) to prolong network lifetime. Moreover, device-to-device (D2D) communication on shared channels is also considered as a promising technology to achieve high data rates, ultralow latency communication, and high spectral efficiency. In this paper, we investigated resource allocation in EH-aided D2D communication underlying 5G cellular along with enabling IoT services. The objective is to maximize throughput of the network subject to the joint constraints on user performance, number of admitted users for equity and fair usage, mode assignment (cellular or D2D) as per available energy, and transmit power allocation along with EH techniques, which results in a mixed integer nonlinear programming problem. We have proposed a low complexity and efficient algorithm, adaptive resource allocation and energy sentient network (ARA-ESN) using branch-cut, branch-bound, and mesh adaptive direct search (MADs) solutions, where cellular or D2D communication is based on available energy and user performance criteria along with EH through ambient energy and radio frequency (RF) energy transfer techniques. We have applied the outer approximation based linearization technique that guarantees the convergence to the optimal solution. The results show that ARA-ESN branch-cut outperforms ARA-ESN branch-bound and ARA-ESN MADs. Moreover, we have also observed that ambient harvesting increases performance of network due to better acquisition of energy as compared to RF energy transfer.