Abstract: This paper proposes a power allocation algorithm that can maximize the average throughput of the system on the basis of a cellular heterogeneous network with energy harvesting. Users and relay harvest RF energy from the source, then users use these energy to transmit information to the source via the relay through the way of TDMA. The optimal model is built on the condition that the causality of energy harvested by the users and the relay is satisfied and the total power of them is restricted. We allocate the power of relay and users so as to maximize the average throughput of the system by making use of the Lagrange multiplier method and the KKT optimal conditions. Moreover, we simplify co-optimization of relay power and user power to optimization of user power by way of equivalent channel gains, then we get the optimal power allocation through the gradient algorithm. Simulation results show that, comparing with the average power allocation algorithm subject to the causality of energy harvested by the relay and users, the algorithm stated in this paper can effectively improve the average throughput of the system.