Research on Transformer Fault Diagnosis Based on Feature Selection and Improved Frilled Lizard Optimization Algorithm

As a core power system equipment, transformers’ operational status directly affects power grid stability and safety, making transformer fault diagnosis research of great engineering application value. This paper proposes a transformer fault diagnosis method based on feature selection and optimized Support Vector Machine (SVM), enhancing diagnostic accuracy via an improved optimization algorithm and feature selection strategy.First, Recursive Feature Elimination (RFE) is used for feature selection: original data is expanded to 16 dimensions, RFE screens representative fault diagnosis features, and these are combined with original gas data into a 10-dimensional set—reducing data dimensionality and improving model generalization.Subsequently, the improved Frilled Lizard Optimization (IFLO) optimizes SVM parameters, with three key improvements: Latin Hypercube Sampling for initialization (ensuring uniform initial population distribution and anti-local optimality), Weibull flight strategy (strengthening global exploration to search the solution space effectively), and nonlinear escape energy strategy (optimizing local exploitation to boost convergence speed and accuracy).