Hazardous discharge is the primary threat to power apparatus. Utilizing the fact that discharge is companied by intrinsic light radiation in specific spectral bands, a micro ternary spectral hypersensitive sensor array is developed for condition monitoring and fault prediction. With this new concept, a multispectral diagnosis strategy is proposed and implemented on a gas-insulated system, demonstrating significant potential for fault classification as well as quantitative risk evaluation, circumventing the necessities of phase-resolved statistics and complex denoising algorithms which are the most challenging but necessary aspects for traditional discharge diagnosis. Thus, it is an intrinsic and stable approach for practical fault prediction, especially for HVDC gas-insulated system. © 2020 Elsevier Ltd