The dairy processing industry is a major field within the food processing industry, in which homogenization technology has become a standardized dairy processing method. With the dairy processing system becoming more and more complex, people put forward higher requirements for the stability of the dairy homogenization process. Nowadays, industrial sensors such as pressure and temperature of various monitoring systems are becoming more and more common, and the data collected are generally massive and unstable, so forecasting these data and taking preventive measures can effectively improve the stability of the processing process. In this paper, we selected LSTM, BiLSTM and CNNLSTM-Attention models to predict the pressure of dairy homogenization process. BiLSTM obtained the bidirectional dependence of time series by traversing the data twice in positive and negative directions. And the prediction model with CNN-LSTM-Attention can add weight to the input features to further improve the prediction performance of the model. We selected a real data set related to the dairy homogenization process as the research object, analyzed and compared the predictive performance of the three models, and the results showed that CNN-LSTM-Attention can achieve the optimal prediction effect by adjusting the hyperparameters reasonably, but its model training time is relatively high. Compared with the other two prediction models, BiLSTM has better robustness and shorter training time. In this paper, multiple comparative experiments were conducted to provide references for the selection of prediction models.