Currently, due to the competitive energy markets, more and more energy producers, concessionaires, and those who sell energy are exposed to risks caused by the volatility of electricity prices. Due to dependence on renewable sources, weather, and global financial instability, accurate electricity price forecasting is increasingly important. The main objective of this work is to evaluate the performance of forecasting methods: Deep neural networks and LSTM neural networks, using a time series of electricity prices in five regions of Brazil. The results for Deep feedforward networks show an excellent ability to predict peaks and satisfactory accuracy according to the Root Mean Squared Error of up to 9.8 R$/Mwh. Preliminary studies show that LSTM networks can predict the PLD value with a percentage Root Mean Squared Error of up to 0.13 R$/Mwh, lower than that obtained by deep feedforward networks and other methods. In 66.66% of the cases, the LSTM network presents more minor prediction errors than the Deep Feedforward network.