Abstract:

Since there are many factors that influence the heat transfer coefficient in the horizontal tube falling film evaporation process, it is difficult to describe its complex nonlinear relationship by relying on traditional correlations. Hence, a BP neural network is proposed to construct a predictive model due to its exceptional capability in fitting functions. Additionally, a GA-BP model is developed to further refine the neural network through genetic algorithms. The parameters fed into the neural network include spray density, evaporation temperature, tube diameter, tube spacing, and salinity. To assess the models' efficacy, predicted values are compared with actual values while also introducing heat transfer correlations as a reference. The results indicate that the GA-BP model exhibits a mean absolute percentage error (MAPE) of only 8.10%, marking a 31.93% reduction compared to the original BP neural network and leading to an approximate 70% improvement in predictive accuracy over traditional heat transfer correlations. It appears that neural network methods hold a distinct advantage in modeling nonlinear systems, effectively predicting the heat transfer coefficient of horizontal tube falling film evaporation. This lays a reliable foundation for further optimizing operational parameters and enhancing the evaporator's heat transfer performance.

Key words: BP neural network, genetic algorithm, heat transfer, prediction, horizontal tube falling film, evaporation

摘要：

由于横管降膜蒸发过程中传热系数的影响因素众多，依靠传统的关联式难以描述其复杂的非线性关系，因此提出利用BP神经网络强大的函数拟合能力构建模型以实现对传热系数的预测，并采用遗传算法进一步优化，建立了GA-BP预测模型，其中输入神经网络的参数包括喷淋密度、蒸发温度、管径、管间距和工质盐度。将模型的预测结果与真实值进行对比，同时引入传热关联式作为对照。结果表明，GA-BP模型的平均绝对百分比误差（MAPE）仅为8.10%，相比原始BP神经网络降低了31.93%，预测精度较传统的传热关联式提升70%左右。可见神经网络方法在非线性系统建模方面具有较大优势，能够有效预测横管降膜蒸发传热系数，为进一步优化工况参数、提高蒸发器的传热性能提供了可靠依据。

关键词: BP神经网络, 遗传算法, 传热, 预测, 横管降膜, 蒸发