Abstract: To address issues such as the susceptibility of XLPE cable insulation to degradation, the difficulty in monitoring partial discharges, and the tendency of traditional BP neural networks to converge to local optima, an online monitoring system for XLPE cable insulation based on PSO-optimized BP neural networks was designed. Discrete wavelet transform was employed to denoise partial discharge signals, and a 17-dimensional feature vector was extracted to construct input samples. The particle swarm optimization algorithm was used to optimize the initial weights and thresholds of the BP neural network, establishing a PSO-BP discharge type identification model. The system adopts a three-tier architecture comprising high-frequency current sensors, intrinsically safe monitoring hosts, and surface-level hosts, enabling non-invasive signal acquisition, high-speed transmission, and intelligent discrimination. Experimental and field tests demonstrate that the PSO-BP model achieves an identification accuracy of 98.33%, a 3.33% improvement over the standard BP model. The system effectively filters out underground interference and stably identifies three typical defects—internal discharge, surface discharge, and tip discharge—meeting the requirements for mine explosion-proofing and uninterrupted monitoring.