SF₆ gas is a chemically stable and non-toxic inert gas with excellent insulation and arc extinguishing properties. It is widely used as an insulating medium in high-voltage electrical equipment. SF₆ decomposes under a strong electromagnetic field to produce a variety of low-fluorine sulfides and then chemically reacts with impurities such as trace H₂O and O₂ existing in the equipment to produce HF, SO₂F₂, CO, CF₄, H₂S, SO₂ and other gases. The type and severity of faults inside the equipment are closely related to the type and content of SF₆ decomposition products. SO₂ gas is one of the important characteristic components of SF₆ decomposition products, which can effectively reflect the severity of latent faults inside SF₆ gas insulation equipment and is highly corrosive. The increase of SO₂ content will aggravate metal surface corrosion and increase the degree of discharge. Therefore, real-time monitoring of trace SO₂ concentration is an important means to ensure the safe and stable operation of power systems. Photoacoustic Spectroscopy (PAS) based on the Beer-Lambert absorption law has been applied in the detection of SF₆ decomposition products due to its advantages of no carrier gas, high detection sensitivity, good selectivity and fast response speed. SO₂ gas was excited to a high-energy state after absorbing a broad-spectrum light of 275 nm. The life of the high-energy state was extremely short. The modulated excitation light source caused the temperature in the Photoacoustic Cell (PAC) to change periodically, and the pressure inside the non-resonant PAC with good sealing changed accordingly, generating a Photoacoustic (PA) signal. The signal processing unit inverted the concentration information of the SO₂ gas to be measured according to the amplitude of the PA signal detected by the acoustic sensor. A set of SO₂ detection device based on ultraviolet PAS was built using a miniaturized non-resonant PAC with a gas chamber volume of only 0.79 mL, which realized the high-sensitivity detection of SO₂ in SF₆ background gas. An Ultraviolet Light Emitting Diode (UV-LED) with a central wavelength of 275 nm and an output optical power of 3.233 mW was selected as the PA excitation light source of SO₂ gas, which avoided the absorption interference of ultra-high concentration SF₆ background gas and greatly reduced the system cost. The modulated excitation beam passed through the ultraviolet quartz window and vertically passed through the gas to be measured in the PAC and exited from the window at the exit end. The internal pressure of the non-resonant PAC was evenly distributed. Two acoustic sensors were symmetrically embedded in the PAC to receive the exciting sound signal in real-time and convert it into an electrical signal. Finally, the concentration information of the gas to be measured was calculated. The amplitude-frequency response of the non-resonant PA system was tested to obtain the best PA signal amplitude. In the range of 0 Hz to 500 Hz, the maximum frequency response of the system occurred at 40 Hz, and the PA signal amplitude was the highest when the operating frequency was 40 Hz. The detection sensitivity and linear response of this PA system were evaluated. Different concentrations of SO₂/SF₆ mixed gas were filled into the non-resonant PAC to analyze the PA response of the system to SO₂ gas. In the concentration range of 0 ppm to 100 ppm, there was a good linear relationship between the excitation PA signal amplitude and the SO₂/SF₆ mixed gas concentration, with a responsivity of 5.59 μV/ppm. The system's Minimum Detection Limit (MDL) was mainly affected by acoustic sensor noise and the background drift in the cell wall and window absorption. The normalized noise equivalent absorption (NNEA) coefficient normalized the absorption line intensity and the effective power of the excitation light. The NNEA coefficient was 7.2×10^-8 cm^-1W·Hz^-1/2. The detection sensitivity and stability of the system were evaluated by Allan-Werle deviation analysis. The Allan-Werle deviation analysis results showed that the Allan-Werle variance showed a continuous downward trend with the increase of the average time, indicating that the system has good stability, and the white noise in the system mainly limited the detection sensitivity. Therefore, the sensitivity of the detection system can be improved by a longer averaging time. The MDL of the system for SO₂ reached 0.1 ppm with an average time of 100 s. The proposed high-sensitivity SO₂ gas in-situ detection technology provided an effective solution for judging the failure degree of SF₆ gas insulation equipment.