In order to reduce the signal-to-noise ratio of signal detection, a novel chaotic attractor was used as a detection tool. Lyapunov exponent method was used to analyze the dynamic characteristics. By increasing the accuracy of the critical chaotic state threshold, the number of chaotic oscillators used in covering the initial phase difference was reduced. In view of the limitations of the main phase state identification methods, a new phase state identification method was proposed and designed in circuit simulation. It can realize the automatic identification of phase states by using the characteristics of the system period. The period of intermittent chaotic state was recorded. The feasibility of the new phase state identification method was verified by designing circuit simulation. Finally, the single-exponentially weaken oscillating partial discharge signal was detected. The simulation results show that the novel chaotic attractor has rich dynamic characteristics and superior anti-noise performance. Unaffected by the transition process, the new phase state identification method is applicable to any chaotic oscillator and has the feasibility of circuit implementation. The novel chaotic attractor can achieve lower SNR detection and cooperate with ultra-high frequency sensor to detect ultra-high frequency millivolt-level partial discharge signals of transformer.