[Objective] Currently, isolating switches have been widely used in power grids, but research on their fault diagnosis remains limited compared to equipment such as transformers and circuit breakers. The accurate identification of fault types in isolating switches through vibration signals during their operation is crucial for the normal operation of power grids and the safety of maintenance personnel. [Methods] This paper proposes an improved frilled lizard optimization (FLO) algorithm by incorporating an adaptive t-distribution perturbation strategy (tFLO). Subsequently, the algorithm is applied to parameter optimization for both sequential variational mode decomposition (SVMD) and least squares support vector machine (LSSVM) to achieve accurate fault identification in isolating switches. Firstly, an adaptive t-distribution perturbation strategy was incorporated to improve the FLO algorithm. The tFLO-SVMD then decomposed experimental data to obtain optimal modal components. Next, the refined composite multiscale fuzzy dispersion entropy (RCMFDE) of the modal components was calculated to obtain a high-dimensional feature matrix. Finally, the tFLO-LSSVM algorithm was used to classify faults in multiple sets of low dimensional feature matrices obtained by reducing the dimensionality of the high-dimensional matrices through kernel principal component analysis (KPCA). [Results] The fault diagnosis method based on tFLO-SVMD-LSSVM-RCMFDE proposed in this study for a 220 kV high voltage isolating switch achieved an accuracy of 97.92%, demonstrating effective identification of various fault types in isolating switches. [Conclusion] There are problems of slow computation speed and poor robustness of modal centers in the intrinsic mode function (IMF) components decomposed by traditional VMD methods, necessitating additional optimization of the number of modes k. The SVMD algorithm can effectively address these issues while achieving more detailed decomposition. Additionally, entropy calculation can effectively quantify the complexity and uncertainty of time series, and fuzzy dispersion entropy (FDE) has the advantages of short computation time and strong anti-interference. Compared to FDE, RCMFDE demonstrates better stability and more comprehensive feature representation. The combination of tFLO-SVMD and RCMFDE can effectively distinguish vibration signals of different fault types. In summary, this study proves that the classification method based on tFLO-SVMD-LSSVM-RCMFDE can effectively identify faults in isolating switches with high accuracy.