Abstract: Due to the multi attributes, high dimensionality and big data of state data, and the fact that feature parameters cannot meet the normal distribution and variance homogeneity test, the state characterization parameters of a low-voltage switching device is difficult to select. To solve this problem, based on the rank differences between grouping states of a single nonparametric multi-state sample, as well as the calculation rules and change trend of Kruskal Wallis (KW) test statistics of the feature parameter, this paper studies and demonstrates the identification mechanism of characteristic parameters of a single sample. At the same time, based on the increment of slope change in the state sequence and the increment of subsequence variation, a sample state segmentation algorithm is proposed, and this algorithm is combined with the parameter identification mechanism to form an improved KW test theory for a single multi-state sample. Finally, a new method of selecting parameters for the state characterization of low-voltage switching devices is proposed based on the improved theory. An AC contactor is used to verify the method, which shows that this method is suitable for state characterization parameter selection of low-voltage switching devices with nonlinear, non-stationary and dynamic random distribution characteristics. It is proved that the goodness of fit of the state characterization curve obtained based on this method can reach 0.91, and the P-value of state segmentation significance can reach 8.428 48×10^–266, which improves the characterization and segmentation accuracy significantly compared with that before parameters selection and the similar method. The results lay a theoretical foundation for the further research on the accurate segmentation and identification of the state of low-voltage switching devices.