Abordagens baseadas em processamento de sinais e inteligência artificial para identificar a inserção de contato principal e resistivo em disjuntores SF6

Abstract

High-voltage circuit breakers play a vital role in a substation, facilitating the connection and disconnection of loads, as well as isolating the system during fault conditions. Among the problems related to circuit breakers operating with capacitor banks, one of the main issues is related to the closing operation, which can generate high inrush current transients associated with high frequencies. These transients may cause damage and premature wear to the circuit breakers and the grid equipment connected to them. Thus, monitoring the main parameters of circuit breakers becomes essential to anticipate failures and estimate their useful life, resulting in economic, operational, and strategic gains. In this context, this work presents an approach based on signal processing and Artificial Intelligence (AI) to identify the instants of insertion of the pre-insertion resistor and the main contact during the operation of a high-voltage SF6 circuit breaker. To this end, current and voltage signals from a real Brazilian substation are used as inputs to the AI-based models, resulting in a feature vector with 600 rows and 82 columns, considering the noise and interference common in this type of environment. Thus, the proposed modeling considers signal preprocessing steps with techniques such as discrete derivative, integral, frequency-domain transforms, and wavelets, to reduce noise and highlight relevant features for feature extraction, the generation of the dataset for model training, the use of different machine learning techniques, and the use of two machine learning models for evaluation: the Multilayer Perceptron (MLP) and the AdaBoost classifier, both trained to automatically identify the critical points in the signal and, finally, the identification of the best moments for controlled switching of the circuit breakers. The results showed that the MLP achieved 94% accuracy, while the AdaBoost reached 95% accuracy. The conclusions indicate that the proposed methodologies are robust for use in real-world environments, allowing for more reliable fault prediction, optimized maintenance, and reduced operational costs. Thus, the work directly contributes to improving the closing synchronism in high-voltage circuit breakers and mitigating critical transients in the electrical system.