Aiming at the problems of complex types of noise sources, prominent low-frequency components, obvious acoustic vibration coupling and insufficient adaptability of traditional noise reduction methods in power substation facilities, a noise source localization and adaptive active noise reduction method based on multi-modal sensor fusion was proposed. The system takes microphone array, acceleration sensor, infrared thermal imager and operating state data as input, constructs the time-space synchronization and feature fusion model of acoustic, vibration, thermal imaging and working condition parameters, and combines TDOA constraints, sound pressure attenuation constraints and equipment candidate region constraints to realize noise source spatial localization. On this basis, the acoustic vibration and thermal multi-feature fusion recognition network and the location-guided adaptive FxLMS control method are introduced to realize the noise source recognition, three-dimensional coordinate estimation and active denoising of the target area. The experiment was carried out based on 9600 groups of multi-modal samples. The results show that the average positioning error of the complete model is 0.36 m, the recognition accuracy is 94.1%, and the F1 value is 93.5%. The noise reduction reaches 14.6 dB, 12.8 dB and 13.5 dB under low-frequency electromagnetic noise, broadband fan noise and structural resonance noise, respectively. The results show that the proposed method can improve the accuracy of noise source localization and the stability of active noise reduction in complex substation scenes.
