为解决重介质选煤过程中分选密度识别易受厚尾噪声污染的问题，建立了ARX分选密度辨识模型，并利用学生式$ t $分布建模了密度辨识系统中的厚尾噪声，而后采用期望最大化（EM）算法将厚尾噪声识别问题公式化，最后通过仿真模拟对密度及厚尾噪声辨识模型进行了验证。结果表明：用于厚尾噪声识别的EM算法与传统极大似然估计算法（MLE）相比，可有效处理隐含变量或数据丢失问题，相应偏差范数（BN）和方差范数（VN）也均低于后者，具有更佳的鲁棒性；所估计的模型参数在有限次数迭代下即可收敛于真实值附近，算法处理厚尾噪声有效。研究结果可一定程度上提升重介质选煤过程中重悬浮液密度自动检测的准确性。

During the process of heavy-medium separation of coal, the identification of the separation density is likely to be contaminated by heavy-tailed noise. To address this issue, the ARX model and the student′s t distribution model are used to identify the separation density and the heavy-tailed noise involved in the separation density identification system. Then the identification procedure is turned formulized using the Expectation Maximum (EM) algorithm. The effectiveness of the separation density identification model developed based on the derived parameters is validated through simulation analysis. Analysis shows that compared with the traditional maximum likelihood estimation (MLE) method, the use of the EM algorithm with an indication of its higher robustness can effectively tackle problems regarding implicite variables and data loss; the bias norm (BN) and variance norm (VN) of the EM algorithm are all lower than those of the MLE method; the estimated model parameters derived using EM algorithm can be converged to approximately the true values after finite iteration operation, well demonstrating the effectiveness of the heavy-tailed noise identification method. The study made herein can help enhance to a certain degree the medium suspension density automatic detection accuracy in the process of heavy-medium separation of coal.