Tunable Diode Laser Absorption Spectroscopy (TDLAS) is widely used in the combustion diagnosis, measuring trace gases and industrial process control and so on due to its high sensitivity, high resolution, non-intrusive and real-time detection. The second-harmonic (2f) detection technology of wavelength modulation spectroscopy (WMS) is one of the main methods for TDLAS gas sensing. The laser is one of the core components of TDLAS-WMS online detection system and its output wavelength and baseline of 2f background signal will drift due to changes of the laser’s operating temperature and other factors during long-term operation, resulting in reduced accuracy of gas concentration inversion and stability of TDLAS-WMS online detection system. Aiming at the above problems, according to the fundamental absorption properties of NO gas molecules in the mid-infrared band of 5.176~5.189 μm, a distributed feedback, continuous-wave quantum cascade laser (DFB-CW QCL) with the peak wavelength of 5.184 μm will be chosen to analyze the drift variation of the laser center output wavelength corresponding to the sampling points and the characteristics of the 2f absorption and background signal drift. Based on the above analysis, the average peak-to-peak value of 2f signal is used to replace the peak value of 2f signal to establish the gas concentration inversion model to correct baseline drift of 2f background signal as well as combining with the signal-to-noise ratio is optimized as the principle of wavelength drift correction, and the 2f background signal drift correction method is finally established to eliminate the adverse effect of the 2f background signal drift on the gas concentration inversion results in the long-term continuous detection of TDLAS-WMS online detection system. The results showed that the average peak-to-peak value of the 2f signal increased with the concentration of NO gas. The average peak-to-peak value of the 2f signal and the concentration of the NO sample showed a good linear relationship and linear fit R² of the fitted curve reached 0.999 9. In the 60 min continuous monitoring experiment which was conducted by using NO sample with a volume concentration of 20×10^-6, the standard deviation, the maximum relative error and the relative error mean square value of the inversion concentration decreased from 0.19×10^-6 to 0.07×10^-6, decreased from 6.30% to 3.85% and decreased from 24.39% to 9.99% after the drift correction, respectively. The method can effectively suppress the influence of 2f background signal drift on the gas concentration inversion result, and significantly improve the sensitivity, accuracy and stability of the TDLAS-WMS online detection system continuous monitoring.