Traditional theoretical calculations, field measurements, and finite element methods sometimes fail to realize life cycle simulations of the temperature field and temperature effect of steel–concrete composite bridge deck systems. In this paper, a simulation method based on a back propagation–long short-term memory (BP-LSTM) network correlation model is proposed to predict the temperature field and temperature effect in a low-cost and highly efficient manner. A bridge was used as the engineering background according to a health monitoring system, and the finite element method based on the principle of heat transfer was used to expand the data sets. Data sets with structural, time, and environmental characteristics as the independent variables and temperature and the temperature effect as the dependent variables were formed. The correlation between the dependent and independent variables was verified using the maximal information coefficient. Thus, the BP-LSTM model was established, and the mean squared error loss function considering the time weight was designed. The data set was read in for training, verification, and testing, and a correlation model representing the relationship between the set’s independent and dependent variables was obtained with relatively high accuracy. Finally, combined with the bridge’s historical meteorological data, the established correlation model was used to simulate the bridge’s temperature field and temperature effect. The results indicated that the finite element calculation results of the structure’s temperature field based on the heat transfer principle were basically consistent with the measured results. The independent variables in the data set were non-linearly related to the dependent variables. The BP-LSTM’s prediction accuracy of the temperature field and temperature effect was above 98.8% and 94.5%, respectively, in good agreement with the target value. The variation law of the temperature field and temperature effect of the steel–concrete composite bridge deck system simulated by combining the historical meteorological statistics was in accordance with reality.

传统的理论计​​算、现场测量和有限元方法有时无法实现钢-混凝土组合桥面板系统的温度场和温度效应的生命周期模拟。在本文中，提出了一种基于反向传播-长短期记忆（BP-LSTM）网络相关模型的仿真方法，以低成本、高效的方式预测温度场和温度效应。根据健康监测系统，以桥梁为工程背景，并采用基于热传递原理的有限元方法来扩展数据集。形成了以结构、时间、环境特征为自变量，以温度和温度效应为因变量的数据集。使用最大信息系数验证因变量和自变量之间的相关性。由此建立BP-LSTM模型，设计考虑时间权重的均方误差损失函数。读入数据集进行训练、验证和测试，并以相对较高的精度获得表示该集自变量和因变量之间关系的相关模型。最后，结合桥梁历史气象数据，利用建立的相关模型对桥梁温度场和温度效应进行模拟。结果表明，基于传热原理的结构温度场有限元计算结果与实测结果基本一致。数据集中的自变量与因变量非线性相关。BP-LSTM对温度场和温度效应的预测精度分别在98.8%和94.5%以上，与目标值吻合较好。结合历史气象统计模拟出的钢-混凝土组合桥面系统温度场变化规律和温度效应符合实际。