In order to improve the detection performance and integration of TDLAS detection system， A TDLAS system based on a portable adjustable optical path length cylindrical mirror multi-pass cell was designed， which achieved the simulated optical path length of 14 m， 10.6 m， 9.2 m， 7.1 m and 5.8 m. To reduce the impact of noise generated during gas measurement in TDLAS system on detection accuracy and sensitivity， the Variational Mode Decomposition （VMD） wavelet denoising algorithm optimized by the Northern Goshawk Optimization （NGO） was proposed. The simulation results showed that compared with other algorithms， the NGO-VMD wavelet denoising algorithm could effectively reduce noise without causing signal distortion. The performance of the system was evaluated by testing CH₄ using a DFB laser with the central wavelength of 1.653 μm based on direct absorption detection technology. The results showed that the signal-to-noise ratio of the detection signal was increased from 66 to 109， and the lower detection limit of the system was improved from 12.1 ppm to 7.28 ppm. Allan analysis of variance showed that the optimal detection sensitivity of the system was improved from 641 ppb to 526 ppb at the integration time of 263 s after optimization with the NGO-VMD wavelet algorithm. The finding provided a reference for improving the performance of TDLAS trace gas detection system.