Anomaly detection addresses the problem of finding unexpected values in data sets. Often, these anomalies, also known as outliers, discordant values, or exceptions, describe patterns in the behavior of the data. Anomaly detection is important because it frequently involves significant and critical information in many application domains. In the case of nuclear fusion, there is a wide variety of anomalies that could be related to plasma behaviors, such as disruptions or low-high (L-H) transitions. In this context, there are known and unknown anomalies, where unknown anomalies represent the largest proportion of the total that can be found in nuclear fusion. This paper presents a study of the application of deep learning and architecture called Autoencoder to detect anomalies predicting (encode-decode) in a discharge.