To achieve efficient classification and three-dimensional （3D） localization of partial discharge （PD） in switchgear，an integrated circular array sensor combining three types of signals，namely transient earth voltage （TEV），ultra-high frequency （UHF），and acoustic array （AA），is designed.A multimodal fusion classification model and an improved nonlinear crested porcupine optimizer （CPO） algorithm are proposed.First，a sensor system with multi-channel acquisition capability is constructed through structural integration and performance testing.Second，a neural network model is built based on nine-dimensional time-frequency domain features to achieve accurate classification of multiple types of discharge defects.Finally，a time delay extraction method combining ∣Δ∣-value time difference screening and a cross-correlation algorithm is proposed，and a consistency objective function with physical constraints is constructed.The CPO algorithm is improved to enhance the robustness and stability of localization.Experimental results indicate that the defect classification accuracy reaches 91.38%，and the localization error is controlled at the 10 mm level，validating the effectiveness of the method in terms of classification capability and localization performance.This study provides a new technical solution for the detection and localization of partial discharge in electrical equipment.