Abstract: Partial discharge (PD) spectrum contains the discharge information e.g., discharge type and discharge energy, which provides a new possibility for accurate judgment of discharge state through synchronously monitoring of multispectral light pulses. In this paper, a single-photon-level multispectral PD synchronous detection system with seven independent detection bands is designed by combining with the working characteristics of the miniature solid state sensor and the PD spectral distribution. On this basis, three multispectral ratio characteristics and their evolution laws with the discharge development under three typical discharges are introduced. Finally, k-means clustering algorithm and deep neural networks (DNN) are applied to establish PD pattern recognition models based on the multispectral ratio characteristics respectively. The results show that the multispectral ratio characteristics of each PD have fingerprint characteristics, and they perform over 90% accuracy in PD pattern recognition without relying on traditional phase-resolved partial discharge (PRPD) characteristics. Multispectral ratio characteristics analysis performs a new PD diagnosis method independent of power frequency voltage phase, which provides the reference for PD diagnosis of DC system as well as AC system with voltage phase loss.