Nowadays, Energy Efficiency (EE) is a very significant indicator for MIMO system from green communication perspective. We initially formulate an optimization problem regarding the EE. However, running such a traditional algorithm will spend much time, which can't suit the frequently changing channel state information. In this paper, we implement a deep learning model to fit a traditional algorithm regarding maximizing EE. In detail, a deep learning model, which is deep neural network (DNN) structure in our paper, has been trained to learn the map between the channel state information (CSI) and optimal power allocation for each data stream of each user. We find that applying the deep learning model can remarkably reduce the running time to solve the optimization problem compared to the traditional algorithm. The numerical results also show that our deep learning model can guarantee very high accuracy to find the optimal power allocation compared to the traditional algorithm.