Aiming at the problem of degradation of inversion accuracy due to noise entrainment in the second harmonic signal, which is the detection signal of gas concentration in the tunable semiconductor laser absorption spectroscopy (TDLAS) technique. Proposing a denoising algorithm based on the combination of the Improved Northern Goshawk Optimization (INGO) algorithm, Variational Mode Decomposition (VMD) and Improved Wavelet Thresholding Method. The INGO algorithm, which minimizes the value of the envelope entropy as an objective function, is used to find the number of modes k and the penalty factor α, which are important parameters in the VMD. Filtering the effective signal IMF components by the fact that the product of the energy density and the average period of each intrinsic mode function (IMF) after decomposition is a constant. The joint wavelet improvement thresholding method is used to reduce the effective signal and combined to obtain the denoised signal. Comparative analysis of simulation experiments shows that the INGO-VMD-improved wavelet thresholding method can effectively filter out the noise signal, reduce the amplitude error, improve the inversion accuracy, and the noise reduction effect is superior. The waveform similarity coefficient after noise reduction is 99.86%, the signal-to-noise ratio reaches 25.230 5 dB, and the root mean square error reaches 0.015 23%. The peak error decreases by 0.157 compared to the noise-containing signal.