基于GRU神经网络的桥梁变形预测方法研究

Research on bridge deformation prediction method based on GRU neural network

桥梁是我国道路交通网中的重要组成部分,在服役过程中由于过载、疲劳效应、外来冲击以及材料老化等原因,不可避免会产生结构的损伤,因此桥梁的变形预测对灾害的防治具有重要意义。近些年来随着深度学习的兴起和迅速发展,长短期记忆网络(long short-term memory, LSTM)作为一种特殊的循环神经网络(recurrent neural network)常被应用于桥梁变形预测。而门控循环单元网络(gated recurrent unit, GRU)作为LSTM的一种改进方法,具有参数量少、过拟合风险小等优点,因此提出了一种基于GRU神经网络来预测桥梁变形的方法。在工程试验数据分析中,分别利用灰色预测模型GM、遗传算法优化BP神经网络、长短期记忆网络LSTM和门控循环单元网络GRU对某桥梁变形数据进行预报分析。结果表明:GRU模型相较于其他预测模型具有更强的学习能力,预测效果明显优于其他时间序列分析方法。

Bridges are an important part of China’s road traffic network. Structural damage will inevitably occur due to overload, fatigue effect, external impact and material aging during service. Therefore, the prediction of bridge deformation is of great significance to the prevention and control of disasters. In recent years, with the rise and rapid development of deep learning, Long-Term and Short-Term Memory(LSTM), as a special Recurrent Neural Network(RNN), is often used in bridge deformation prediction. As an improved method of LSTM, Gated Recurrent Unit(GRU)network has the advantages of less parameters and less risk of over fitting. Therefore, a method based on GRU neural network to predict bridge deformation is proposed in this paper. In this experiment, the deformation prediction of a bridge is verified by Grey Model(GM), Genetic Algorithm Optimized BP Neural Network, LSTM and GRU. The results demonstrate that GRU model has stronger learning ability than other prediction models, and the prediction effect is obviously better than other time series analysis methods.