Between problems in the power grid expansions, distributed generation, energy market customers migrations, commercial losses and other problems, power distribution companies seek to improve both energy quality and costs reductions, enhancing profitability. A basic and crucial point for any energy company is how much energy to buy. That is, determining the amount of energy to be purchased as close as possible to that needed to serve its customers, avoiding financial losses by consuming more or less than they have. However, this is not a trivial problem, as energy consumption depends on several exogenous and endogenous factors, such as all the problems previously mentioned, in addition to economic, social, climatic, political and cultural aspects, among others. Thus, energy forecasts are realized with aid of both statistical analyzes and computational techniques. This article exposes a very short and short term energy forecast model using Neural Networks and feedback, applied in the new global context: the new coronavirus pandemic and its implications for energy consumption. The method was implemented with a real consumption dataset provided by the Brazilian energy company Equatorial from Para State and from Maranhao State. Very short term energy forecasts results reached a MAPE of around 1.2% in a 15-day window for both States, Para and Maranhao. For short term energy forecasts, results for both States were 3 possible scenarios in a window from June to December 2020, due to the unpredictability of the pandemic, especially in Brazil, which so far has shown no signs of reducing the contagion curve.