In TDLAS detection technology, the second harmonic signal carrying gas concentration information is susceptible to the amplitude and frequency of the scanning and modulation signals. Based on TDLAS, a hardware system for CO concentration detection was built and compared with the corresponding simulation model to study the influence of modulation parameters on the peak value, signal-to-noise ratio, symmetry, and peak width of the second harmonic signal. The optimal modulation parameters for the system were determined through experiments, which improved the detection accuracy without changing the hardware. Detection of CO absorption spectroscopy at 1567.7nm revealed that the measured concentration decreases with increasing temperature, with a maximum relative error exceeding 15%. To reduce the temperature effect on measurements, compensation techniques using RBF and BP neural network, PSO-BP neural network, and WOA-BP neural network algorithm was applied to the system. The results showed that the WOA-BP neural network method provided the best compensation effect, reducing the relative error of the corrected concentration to below 1%, effectively enhancing the accuracy and stability of the system in variable temperature environments. This research provides reference for setting modulation parameters and achieving precise detection, and offers valuable guidance for subsequent experiments.