As an essential component of voltage shock protection device in substations, the reliability of lightning arresters is crucial for enhancing the stability of power system, while the reliability evaluation is limited by problems such as long aging test time and scarce failure data. In view of the problem of long test time, the accelerated aging test of the valve plate is designed and carried out, and the Arrhenius model is used to achieve the equivalence of the life data from the accelerated conditions to the normal conditions. Aiming at the problem of less failure data, based on the two-parameter and three-parameter Weibull distribution models, the least squares method and the grey model estimation method are used to model the reliability of small sample data, and the influence of different failure rate estimation formulas on the reliability model is analyzed. It is verified that the fitting effect of the three-parameter Weibull distribution model is better than that of the two-parameter Weibull distribution model, the correlation coefficient is increased by 0.026 8. The failure rate estimation formula affects the parameters estimated value and fitting effect of the model, and the difference between the parameter estimates can reach 1. 312 and the correlation coefficient gap can reach 0.018 3. This paper optimizes the fitting effect of the model by adjusting the model parameters and parameter estimation methods, providing a reference for improving the accuracy of reliability model parameter estimation under small sample data.