Simultaneous Wireless Information and Power Transfer (SWIPT) network is a promoting mode which provides continuous energy for the energy limited scenarios. Combining the relay transmitting mode and power–time-splitting mode, the optimization problem of minimizing the energy consumption is formulated. Different from the common consumption that the energy receiver can harvest energy in linear way, the nonlinear SWIPT model is taken into consideration, which is more realistic, and the causality of energy consumption and requirements of transmission rate are included as the constraints. To solve the nonlinear and coupled optimization problem, a decomposed mode in two layers is proposed. In the outer layer, the optimization of EH architecture is performed, in the inner layer, the joint optimization of beamforming vector, time division factor and power division factor are performed. Due to the coupled variables and the existence of nonlinear constraints, solving the inverse function by variable substitution is introduced, and the primary problem is transformed to a linear form. Finally, the Lagrangian function of the transformed optimization problem is solved, and the optimal solution of the problem is obtained by performing several iterations of the projection gradient method and the CVX solver alternatively. The simulations verify that the nonlinear-multi-EH circuits model is effective to improve energy conversion efficiency and reduce the energy consumption of the base station, and the results show that there is at least 26.4% improvement on energy saving compared to the benchmark schemes.