The analysis and monitoring of Power Quality (PQ) are a research topic that has concerned with the scientific community in recent years. Several studies have been realized, in which, the object of study regard different approaches; the impact of deficient power supply, the effects caused by the charges in the system or methodologies for the detection, identification and classification of the phenomena that are referred as PQ disturbances, these problematics must be faced to counter the negative effects generated. In this paper is presented two approaches for the characterization and classification of various PQ disturbances, the techniques are involved in the application of subjects related to artificial intelligence; machine learning and deep learning, which in recent years have been used in different areas of study with a good performance for the applications developed. A comparative from these two techniques is performed; the first, machine learning technique, linear discriminant analysis, and the second technique proposed is a deep learning tool called autoencoder. The methodology is tested with a case study with real signals that contain variations in the voltage signal, the results from each technique are presented and the conclusions indicate the comparative realized.