Evaluación oportuna de la vulnerabilidad dinámica de la red usando técnicas de aprendizaje automático aplicado al caso de estudio peruano

Abstract

CONDITIONING OF PROJECT PUBLICATION. work proposes the development of the vulnerability prediction of the Peruvian system through the analysis of oscillatory stability, using learning elements based on data obtained from different emergency and/or contingency conditions of the Peruvian system. Although software tools already exist to evaluate such contingencies, the idea of predicting vulnerability conditions in advance for certain conditions of a new scenario based on machine learning is a relatively new proposal. To do this, this work relies on tools such as contingency analysis to generate scenario data applied to the Peruvian network. Furthermore, using some artificial intelligence techniques and other dimensionality reduction techniques, the scenarios are classified and algorithms are trained to obtain system states that could be vulnerable with sufficient precision. State prediction is sought for probable test scenarios based on changes in network buffer levels. In addition, it is noted that the processing time and identification of conditions using a well-trained AI model is reduced, adding importance. From this work it is clear that, with good training and necessary data, vulnerable states can be predicted in the Peruvian system that give rise to early warning. Furthermore, trained algorithms help in real-time identification. Also, regardless of the nature of the network (in this case the updated network was used with renewable projects included), and with data from a certain number of chosen parameters, conditions of probable vulnerability can be identified. In other words, the rapid evaluation of the oscillatory stability of the Peruvian system by classifying the data is very feasible, and it can be applied without problems to the Scada – WAMS data of the Peruvian system. Keywords: Data mining, training, network, machine learning, etc.