This paper addresses the problem of data prediction and abnormality warning for ammonia critical equipment by building a mathematical model based on the idea of Prognostics and Health Management (PHM) to predict data trends and identify abnormalities. The sub-series type classification and simple identification of real-time data are effectively achieved by using sequence segmentation algorithm, the effective prediction of data by using long and short-term memory networks, and the real-time monitoring and early warning of abnormal data by using the Gaussian distribution model. The core functions include a highly reducible digital twin model, real-time data monitoring based on Gaussian distribution, equipment data prediction based on long and short-term memory networks, and real-time visualization of equipment status and data trends.