To address the issues of partial discharge in eco-friendly gas-insulated switchgear (GIS) during operation, and to overcome the technical limitations of traditional detection methods, this study aims to achieve high-precision perception and effective feature extraction of weak discharge signals, thereby providing technical support for early fault diagnosis of equipment and the intelligent operation and maintenance of power systems.

A distributed multi-point sampling photoacoustic spectral signal acquisition system was constructed, utilizing a tunable quantum cascade laser and a high-sensitivity photoacoustic sensor. The signal intensity was enhanced through the resonance effect, enabling efficient collection of photoacoustic spectral signals generated by discharge. In response to the non-stationary, non-linear, and high-dimensional characteristics of the collected signals, the ensemble empirical mode decomposition method was employed to decompose the signals into the intrinsic mode function components of different frequency components, clearly describing the intrinsic dynamic characteristics of the signal. Subsequently, an adaptive window width transform algorithm based on Kaiser window function dynamically adjusted the window width according to the local signal features, enabling precise extraction of discharge signal spectral features.

The experimental results indicated that the repeatability error of signal acquisition was below 0.0019, and the Renyi entropy value exceeded 0.907, reliably extracting the spectral features at different frequencies. The signal resolution reached 90 dB, the feature extraction time was 5.2 s, and the signal-to-noise ratio improved by 25 dB. All performance indicators outperformed results from traditional methods. The suppression effects on laser intensity noise, gas flow noise, and electromagnetic interference reached 82%, 76%, and 91%, respectively. The method effectively distinguished the discharge characteristics of three typical eco-friendly gases, achieving a minimum detection limit as low as 1×10⁻⁶ and reducing the response time by over 60%. It also enabled synchronous identification of five components, with temperature stability of ±0.20 ℃.

This method successfully integrates the high sensitivity of photoacoustic spectroscopy with the precise analytical capabilities of advanced signal processing algorithms, overcoming traditional bottlenecks in noise suppression and weak signal extraction. It offers advantages such as excellent collection stability, strong anti-interference ability, high feature extraction accuracy, and rapid response speed, achieving high-precision, real-time extraction of weak discharge features. This offers a novel technical pathway for the early fault diagnosis of eco-friendly GIS and fully meets the practical needs of intelligent operation and maintenance of power systems.