中文核心期刊
CSCD来源期刊
中国科技核心期刊
RCCSE中国核心学术期刊
重庆交通大学学报（自然科学版）

重庆交通大学学报（自然科学版） ›› 2025, Vol. 44 ›› Issue (6): 131-138.DOI: 10.3969/j.issn.1674-0696.2025.06.14

• 桥梁与隧道工程 • 上一篇    

基于ANN-KF-BiLSTM的桥梁温度多步预测

闫文佳1，江鸥1，李鸿先2，徐嘉璐3   

  1. (1.云南云岭高速公路工程咨询有限公司, 云南 昆明 650000；2.云南省交通投资建设集团有限公司昆明东管理处， 云南 昆明 650000；3.东南大学 经济管理学院，江苏 南京 211189)
  • 收稿日期:2024-07-01 修回日期:2024-12-19 发布日期:2025-06-30
  • 作者简介:闫文佳（1985—），男，河南郑州人，高级工程师，主要从事公路工程科技创新项目研究及管理方面的工作。E-mail：5065857061@qq.com 通信作者：徐嘉璐（2001—），女，重庆人，硕士研究生，主要从事大数据挖掘方面的研究。E-mail：katemac1996@hotmail.com
  • 基金资助:
    云南省科技厅数字交通研究项目（202205AG070008）；云南省交通投资建设集团有限公司科技创新项目（YCIC-YF-2022-25）

Multi-step Bridge Temperature Prediction Based on ANN-KF-BiLSTM

YAN Wenjia1，JIANG Ou1，LI Hongxian2，XU Jialu3   

  1. (1. Yunnan Yunling Highway Engineering Consulting Co., Ltd., Kunming 650000, Yunnan, China; 2. Kunming East Management Office of Yunnan Communications Investment and Construction Group Co., Ltd., Kunming 650000, Yunnan, China; 3. School of Economics and Management, Southeast University, Nanjing 211189, Jiangsu, China）
  • Received:2024-07-01 Revised:2024-12-19 Published:2025-06-30

PDF

摘要： 利用深度学习中学习特征能力较强的人工神经网络（ANN）模型和学习时间序列能力较强的双向长短期记忆网络(BiLSTM)模型，辅以卡尔曼滤波(KF)对人工神经网络模型的结果进行动态调整，基于stacking集成策略融合ANN 和BiLSTM模型，构建了一个既能利用气象温度又能记忆桥梁自身温度时间序列的ANN-KF-BiLSTM模型。以云南省某连续刚构桥的温度预测为例，验证了该模型的有效性。研究结果表明：ANN-KF-BiLSTM模型在桥梁温度多步预测中表现出明显优势，在预测时间步数小于96时，拟合程度超过0.89，在预测步数达到168时，平均拟合程度仍可达到约0.76；相较于基准模型，ANN-KF-BiLSTM模型拟合程度更高，预测稳定性更好。研究结果改善了当前利用深度学习模型预测桥梁温度集中于单步预测的状况，为桥梁温度的多步预测提供了一种有效的方法。

关键词: 桥梁工程；桥梁温度；双向长短期记忆网络；卡尔曼滤波；ANN-KF-BiLSTM模型

Abstract: An ANN-KF-BiLSTM model that could both utilize meteorological temperature and memorize the time series of the bridge itself temperature was developed by utilizing the artificial neural network (ANN) with great feature learning ability and the bidirectional long short-term memory network (Bi-LSTM) with strong time series learning ability in deep learning, while the Kalman filter (KF) was supplemented to adjust the output of ANN dynamically, which was based on the stacking ensemble strategy, combining ANN and BiLSTM neural networks. The effectiveness of the proposed method was verified by taking the temperature prediction of a continuous rigid bridge in Yunnan Province as an example. The research results show that the ANN-KF-BiLSTM model has obvious advantages in the multi-step prediction of bridge temperature, with a fitting degree of more than 0.89 when the number of prediction time steps is less than 96, and when the number of prediction steps reaches 168, the average fitting degree can still reach nearly 0.76. Compared with the benchmark model, the fitting degree of the proposed model is higher, and the model prediction stability is better. The proposed model improves the current situation that using deep learning models to predict bridge temperature concentrates on a single step prediction, providing an effective method for multi-step prediction of bridge temperature.

Key words: bridge engineering; bridge temperature; bidirectional long short-term memory network; Kalman filter; ANN-KF-BiLSTM model

中图分类号: