Assessment of Power Transformer Technical State Using Explainable Artificial Intelligence

The functioning of deeply integrated technological systems in the power industry depends on the quality of power supply systems. Reliable power supply requires minimizing emergencies at power plants and substations. Reducing the accident rate is possible by switching to the maintenance and repair of high-voltage equipment based on its actual technical state. Machine learning methods make it possible to assess the actual technical state of various equipment types based on an aggregated analysis of many parameters, in contrast to methods based on comparing parameters with boundary values without con-sidering their mutual influence. Therefore, machine learning can be an effective tool for decision support systems for equipment diagnostics. The opacity of the models and the lack of interpretability of their recommendations significantly complicate industry implementation. The purpose of the study is to de-velop a method for increasing the interpretability of machine learning models for assessing the technical state of equipment. A power transformers’ oil dataset, decision tree ensembles and the Shapley additive explanations were used. The novelty of this study is the modification of the additive Shapley additive explanations, aimed at increasing the information content of the visualization of machine learning model output interpretation. The most significant results are the substantiation of the applicability of explain-able artificial intelligence for the assessment of the technical state of high-voltage equipment. In addition, for the first time in this problem, a Light gradient boosting method was applied. The proposed method allows for increasing the validity of the technical state assessment of high-voltage equipment using ma-chine learning.