Reliable IoT Paradigm with Ensemble Machine Learning for Faults Diagnosis of Power Transformers Considering Adversarial Attacks

Power transformer represents an important equipment in electric power systems. Transformers are not only a source of power outages for electric utilities, but they also affect customers because they interrupt power supplies. Transformers require frequent maintenance and service, which can be time-consuming and expensive. Power systems and their maintenance and servicing can be planned in advance when transformer faults are detected early with high accuracy rates. In order to avoid unplanned shutdowns, it is crucial to monitor and diagnose faults of power transformers. Based on how electrical and thermal stresses impact the insulating oil, it is possible to detect or classify transformer faults by checking for dissolved gases. In this paper, a new transformer monitoring and fault diagnosis technique are developed by utilizing an industrial Internet of Things (IoT) platform combined with an ensemble machine learning (EML) strategy. In order to prevent overlapping between diverse transformer defects, an innovative feature engineering strategy is developed based on introducing new features for gases' concentrations and their ratios. Through this, EML efficiency is enhanced due to the separation of overlapping in the datasets between diverse transformer faults. The datasets from Egyptian chemical laboratories and literature are used to train and test the proposed EML. As shown in the experiments, the proposed fault diagnosis model provides the best diagnosis accuracy, which is significantly better than the state-of-the-art machine learning approaches. Finally, the robustness of the developed EML model is further demonstrated by including white Gaussian noise that is based on the Fast Gradient Sign Method (FGSM) as well as the zeroth order optimization (Zoo) attack to validate the model against adversarial attack scenarios.