We analyzed the acoustic information within the substation scene and studied the variation patterns of key characteristic parameters of abnormal sound sources in this scenario and proposed features that can be applied to false imaging of sound field spatial distribution. For typical distribution substation scenarios, we have developed a deep learning-based joint localization method for abnormal sound sources. Firstly, we collect sound and vibration signals from typical equipment in substations using a microphone sensor array. Then, the vibration signal is tranformed to construct analytic signals and extract features at each scale. Then, we visualize the sound field by controllable beamforming. Finally, we build a deep neural network model for abnormal sound pattern recognition to predict the category and spatial distribution of abnormal sound fields. The above method we proposed can accurately predict the spatial distribution and location of abnormal sound fields within the target area under low signal-to-noise ratio and high reverberation conditions.