This letter highlights the importance of channel prediction during channel estimation (CE) in an energy efficient wireless communication system. Specifically, an energy beamforming-assisted multi-antenna system is considered in which energy receiver sends pilot signal to energy transmitter for CE. A long-short term memory based neural network is then utilized for predicting channel characteristics with respect to past CEs. Performance of learning strategy is evaluated, and efficacy of proposed approach is verified by varying number of estimations against predictions and observing maximum energy harvested at receiving antenna. Proposed scheme is also compared and found to outperform two existing benchmarks.