Energy efficiency has a significant importance to optimize the wireless communications systems by providing high data rates. In order to develop energy efficient systems, one of the promising method is to use multiple device-to-device (D2D) underlaying multiple antenna cellular systems. The interference from cellular users to D2D pairs, the interference between D2D pairs and the interference at the base station (BS) caused by D2D pairs occur in these communications systems. In this paper, we propose energy efficient resource allocation algorithms for under-laying multi-D2D enabled multiple-antennas communications by employing different multiple antenna processing techniques at the BS. A joint method based on Dinkelbach algorithm and Message Passing Algorithm (MPA) and an approach based on deep learning with multi-layer artificial neural network are proposed to maximize the global energy efficiency (GEE) while satisfying the data rate requirements of both cellular users and D2D pairs. In MPA, the factor graph of the D2D pairs is constructed by taking into account the interference among the D2D pairs and the interference level at the BS to avoid any interruption in the cellular transmission. By relying on the training based on the proposed joint algorithm, a deep neural network approach is presented for off-line implementation. The performance results of the proposed energy efficient resource allocation algorithms are illustrated the superiority of multi-D2D communications over conventional single-D2D communications.