A Data Cleaning Method for Power Transformer Situational Awareness Based on RANSAC and Improved IForest Algorithm

Interference from operational environments and data acquisition devices can introduce noise into transformer monitoring systems, compromising the accuracy of equipment status analysis. To address this, enhancing data quality through effective cleaning techniques is crucial. This paper proposes an innovative data cleaning approach for power transformers, combining the Random Sample Consensus (RANSAC) algorithm with an enhanced Isolation Forest (IForest) model. By leveraging unsupervised learning strategies, the method focuses on analyzing dissolved gas data in transformer oil. RANSAC is employed to establish a mathematical model reflecting gas component relationships, based on the gas content ratio principle. Meanwhile, the improved IForest learns from historical data to construct a data cleaning model, which is then used for automated dataset cleaning. Experimental results confirm the method's effectiveness in mitigating the impact of noisy and unreliable data on detection accuracy.