The utilization of sensors in water management facilities offers new possibilities. The data coming from these sensors can expedite the whole water management procedure by enabling the adaptation of several Machine Learning techniques to extract useful knowledge from the said data and generate several predictions. A very important process that takes place in water management is the wastewater treatment, where several contaminants are removed from wastewater so that this water can return to the water cycle. Wastewater treatment is a complicated process, consisting of several steps, in order to ensure the safety of the water prior to reentering the water cycle. As a result, the wastewater treatment plants consume vast amounts of energy to support the whole procedure. Given the fact that this energy is not infinite, since the energy crisis is continuous, it is imperative that the necessary and appropriate technical tools are implemented, in order to ensure that the wastewater treatment plants are as energy efficient as possible. This manuscript proposes a neural network architecture for forecasting the energy consumption of a wastewater treatment plant. It compares different architectures and optimizers, focusing on operational parameters, selecting the combination that best minimizes prediction bias. According to the results of this study, neural network architectures with less complexity, namely linear and dense neural networks that utilize the Stochastic Gradient Descend and Nadam optimizers respectively, achieved higher forecasts’ accuracy and required less training time than more complex approaches such as Recurrent Neural Networks.