Aiming at the problems of strong multivariable coupling, obvious control lag, scattered abnormal symptoms and difficult to identify by traditional methods in the operation process of environmental control equipment in vacuum rehumidification section, this paper takes K3-K4 (ZK-120) combined air conditioning unit of Baoji Cigarette Factory as the object. Relying on the temperature and humidity, pressure difference, valve opening, inverter current, frequency and power data collected by PLC, PROFINET communication and central control monitoring link, a deep learning analysis model combining timing feature extraction, attention enhancement and multi-dimensional feature fusion is constructed. The model takes operation state identification and key variable trend prediction as dual task outputs, which can describe the dynamic correlation between temperature and humidity regulation, fan load change and actuator response in a continuous time window. Experimental results show that the Accuracy, Recall and F1 of the proposed model in the state recognition task reach 95.8%, 94.6% and 94.9%, respectively. The RMSE and MAE of the model in the trend prediction task are 0.118 and 0.087, respectively, and the average amount of warning advance reaches 2.9 hours. The research shows that the deep learning method can effectively improve the accuracy and foresight of the operation data analysis of vacuum pressure equipment, and provide computational support for equipment early warning, operation and maintenance decision- making and stable operation.
