To address the technical bottlenecks of spatiotemporal feature decoupling, high hardware costs, and excessive computational complexity in ultrasonic detection of partial discharge (PD) in electrical equipment, this paper proposes a TDOA/DOA hybrid localization method based on Kernel Principal Component Analysis (KPCA) and modified noncircular FastICA (mnc-FastICA). By integrating spatiotemporal feature extraction with intelligent optimization mechanisms, this method achieves high-precision localization using a small-scale sensor array. The key innovations are as follows: First, a KPCA-assisted pseudo-whitening preprocessing framework is constructed, leveraging Polynomial kernel mapping for nonlinear signal dimensionality reduction, which preserves the correlation between time delay (TDOA) and direction-of-arrival (DOA) features while suppressing environmental noise. Second, after blind separation of ultrasonic signals via the mnc-FastICA algorithm, TDOA/DOA parameters are synchronously extracted through a combination of the Generalized Cross-Correlation (GCC) method and array manifold analysis. Finally, a maximum likelihood estimation model integrating dual parameters is established, and the African Vulture Optimization Algorithm (AVOA) is introduced to accelerate global optimal solution convergence. Experimental results demonstrate that, with a compact hardware configuration of two orthogonal arrays (8 sensors in total), the proposed method achieves a TDOA estimation error of 2.34%, DOA estimation accuracy better than 2°, and localization errors as low as 1.54 cm. This approach effectively resolves the contradictions among spatiotemporal feature coupling, hardware cost, and localization accuracy in PD detection, offering a novel solution for condition monitoring of electrical equipment.