Transformer Oil Chromatographic Anomaly Data Cleansing Based on Correlation Analysis and PSO-BPNN

The efficacy of online transformer monitoring systems is severely compromised by anomalous data arising from sensor malfunctions and other aberrations, particularly in certain monitored characteristics of oil chromatography data. To effectively discern and cleanse these irregularities, an anomaly data cleansing method predicated on correlation analysis and neural networks is proposed. Initially, a clustering algorithm is employed to segment the sequences, which is then used in conjunction with the quartile method to identify outliers in the transformer oil chromatographic data. A strong correlation feature set is then selected through correlation analysis and integrated with a neural network algorithm to construct an anomaly data cleansing model. Finally, the weights and thresholds of the neural network model are optimized using a particle swarm algorithm, thereby enhancing the accuracy of anomaly data cleansing. The efficacy of the model is validated by its application to real transformer operational data. Experimental analysis demonstrates that this optimized model can effectively improve both the accuracy and stability of anomaly data cleansing in transformer oil chromatography.