在不确定度评定时，准确判断数据的分布类型至关重要，提出一种利用反向传播（BP）神经网络训练模型来判别数据分布类型的方法，并将其应用于平面水平度测量不确定度评定的过程中。为了从多个维度全面分析数据的分布特征，所提方法将原始数据转化为7个特征指标，再引入BP神经网络自动学习和提取这些指标之间的复杂非线性关系。通过训练，获得了一个能够准确判别数据分布类型的网络模型，经参数配置优化，该模型展现出优越性能，不仅训练速度快，而且判别准确率高达99.52%。与传统方法相比，该模型在各类分布判别中均表现优异。最后，通过对3路电容传感器测量平面水平度的测量评定实例，通过该模型判别并评定得到了该平面倾角的最佳估计值为7.041″，标准差为0.055″，比传统方法判别并评定的结果具有更小的标准差和更小的置信区间，进一步验证了该模型在蒙特卡洛测量不确定度评定过程中的实用性。

In the field of uncertainty evaluation， how to accurately judge the distribution type of data is very important. Therefore， a method of using BP neural network training model to distinguish data distribution type is proposed， and it is applied to the uncertainty evaluation process of plane levelness measurement. In order to comprehensively analyze the distribution characteristics of data from multiple dimensions， the original data is transformed into 7 feature indicators by the proposed method， and then the backpropagation （BP） neural network is introduced to automatically learn and extract the complex nonlinear relationship between these indicators. Through training， a network model that can accurately distinguish data distribution types is obtained. After parameter configuration optimization， the model shows excellent performance， not only the training speed is fast， but also the discrimination accuracy is as high as 99.52%. Compared with traditional methods， this model performs well in all kinds of distribution discrimination. Finally， through a measurement and evaluation example of three-channel capacitive sensors for measuring the flatness of a plane， the model discriminates and evaluates that the best estimated value of the plane inclination is 7.041″， with a standard deviation of 0.055″. Compared with the results discriminated and evaluated by traditional methods， it has a smaller standard deviation and a narrower confidence interval， further verifying the practicability of the model in the Monte Carlo measurement uncertainty evaluation process.