Abstract
Gas-insulated switchgear (GIS) is an important component of power transmission systems. Accurate partial discharge (PD) pattern recognition is a key requirement for identifying internal insulation defects within the equipment. However, ultra-high-frequency (UHF) PD pulse sequences produced by different insulation defects usually contain complex nonlinear temporal structures and multi-scale periodic variations. These coupled characteristics are difficult to describe adequately via a single feature-mapping strategy. Thus, this paper proposes a temporal-frequency dual-branch stochastic configuration network (TF-SCN), which consists of two heterogeneous hidden-layer branches, for GIS PD pattern recognition. Specifically, in the temporal branch, the model uses a non-periodic, nonlinear activation function similar to that used in a conventional SCN to capture the nonlinear temporal characteristics. The frequency-sensitive branch introduces paired sine–cosine harmonic nodes with shared random projection parameters to capture frequency-sensitive features. The hidden outputs of the two branches are concatenated into a joint temporal-harmonic feature space, and the output weights are solved under the residual inequality constraints for GIS PD classification. To verify the superiority of the proposed model, comparative experiments are conducted on a dataset containing four PD patterns collected from the GIS PD experimental platform. Several baseline models, including 1DCNN, BPNN, SVM, KELM, RVFL, and SCN, are selected for performance comparison. The results show that, compared to 1DCNN, BPNN, SVM, KELM, RVFL, and SCN, TF-SCN effectively extracts distinguishable features in both the time and frequency domains, thereby achieving the best overall performance. Furthermore, its recognition performance remains consistently superior even on noisy data with signal-to-noise ratios ranging from 50 dB to 20 dB. By integrating highly sensitive UHF sensors with the proposed TF-SCN, this study presents a robust, AI-enhanced intelligent sensing and fault diagnosis system for continuous condition monitoring of power equipment.