Abstract:

Accurate prediction of the fouling state of the heat exchanger can timely understand the degree of fouling, so as to implement targeted cleaning, which is of great significance to improving the economic use and production safety of the heat exchanger. The model integration of convolutional neural network (CNN) and long short-term memory network (LSTM) was used to predict the heat exchanger fouling factor. A large number of historical data of heat exchanger are used to train the CNN-LSTM model, and excellent prediction results are obtained. Compared with the single CNN and LSTM models and the multi-layer perceptron neural network (MLPNN) models in the literature, the CNN-LSTM model is more accurate and more stable. In the case shown, the coefficient of determination (R²) is 0.98167, and the average absolute percentage error (MAPE) is 3.199×10^-3. The establishment of the model not only provides a theoretical basis for solving the scale problem of the heat exchanger, but also provides a more scientific and accurate basis for the safe operation and maintenance strategy of the heat exchanger. It helps to improve the economy and production safety of the entire heat exchange section.

Key words: heat exchanger, model, prediction, neural networks, integration, algorithm

摘要：

对换热器的结垢状态进行精准预测，可以及时了解结垢程度，从而有针对性地实施清洁，对提高换热器使用经济性和生产安全性具有重要意义。利用卷积神经网络（CNN）与长短期记忆网络（LSTM）的模型集成对换热器污垢因子进行预测，采用大量换热器历史数据对所建立的CNN-LSTM模型进行训练，取得了良好的预测效果。与单一的CNN和LSTM模型及文献中的多层感知器神经网络模型（MLPNN）相比，所建立的CNN-LSTM模型具有准确性更高、稳定性更强的特点。在所展示的案例中，决定系数（R²）为0.98167，平均绝对百分误差（MAPE）为3.199×10^-3，不仅为解决换热器结垢问题提供了理论基础，而且为换热器安全运行与维护策略提供了更科学、精准的依据，有助于提升整个换热工段的经济性和生产安全性。

关键词: 换热器, 模型, 预测, 神经网络, 集成, 算法