This work presents a review of the main concepts of Wireless Sensor Network (WSN)s and tackles one of their main problems, which is energy consumption. This is done using the data collected by a network deployed at the Hydropower power plant in Cachoeira Dourada. First, an exploratory data analysis of the WSN using statistical and machine learning methods was performed to discover insights about the current state of the network. The analysis provided information about which nodes are more stable, correlations between the data that can be exploited to optimize transmissions, and information about the stability of the links. The work also proposes the use of a Deep Learning model, in a dual prediction scheme, to reduce the transmissions between devices in the network, reduce congestion, and save energy. To do so, a review of data prediction strategies in WSNs is performed. Different neural network based models are introduced and compared using different error metrics in prediction. Finally, a measure of the reduction in data transmission is given, considering different error thresholds. Results show that the model can save a considerable amount of data in transmission, from 70% to 90%, and still maintain a good representation of the measured data.