Thermometry based on nitrogen-vacancy (NV) center in diamond realizes temperature sensing by measuring the zero-field splitting parameter (D) between the sublevels of its ground state. Since diamond has high stability, the resistance to interference, and can be processed into different sizes, thermometry is regarded as a feasible solution to high-precision temperature measurement on the micro-nano scale, which has the potential to be developed into practical application. In the measurement of D, the laser is applied to excite spin polarization and a microwave is then used to manipulate the electron spin. The optically detected magnetic resonance (ODMR) spectrum is obtained by detecting fluorescence released in the transition of the electrons, and the spectral lines are fitted to determine the values of D. Fluctuation of laser power is one of the important sources of experimental noise. To realize a high probability of spin polarization applying laser with high power is necessary. However, the fluctuation at a relatively high laser power will influence the probability of spin polarization, decrease the stability of the fluorescence intensity detected, and increase the error of data. The traditional method, which outputs the photon count rate directly without any reference for normalization, will reflect the fluctuation of laser power in the spectrum. This paper proposes a method to reduce the laser-induced error, in which a reference value of photon count without effective manipulation of microwave can be measured at each count using a specifically-encoded pulse sequence, and normalized fluorescence intensity will be obtained in the spectrum after normalizing the photon count under the manipulation of microwave to the reference. In this way, the effect of the laser power fluctuation on the spectrum will be weakened. The comparative experiments were carried out at 300 K on the ODMR measurement system built in the laboratory. A count time of 0.8 s and a preparation time of 0.001 s were determined considering the optimization of the two time-related parameters. Under different laser powers, three groups of D values were measured by the direct fitting, mathematical, and pulse code normalization methods. The experimental results demonstrate that using pulse code to obtain reference for normalization can effectively suppress laser-induced error to 62.5% of that without any reference, and improve the data accuracy from 179 kHz with no reference and 165 kHz with mathematical reference to 56.9 kHz by comparison. The method can effectively suppress the laser-induced error, improve the resolution of the determination of D, and lay a foundation for the practical development of thermometry based on NV centers.