准确识别索力具有重要现实意义，但频率法在实际应用中存在计算公式复杂多样的问题。基于此，本文将物理信息神经网络（Physics-informed Neural Network, PINN）引入索力识别领域，提出了基于物理信息神经网络的斜拉索索力识别方法。该方法在神经网络损失函数中引入表征斜拉索索力的弦振动方程损失项，并利用斜拉索的动位移响应进行网络训练，通过最小化损失函数即可实现索力识别。基于数值仿真算例和室内试验模型的有限测试数据，利用本文所提方法实现了索力的准确识别。研究结果表明，在本文所设定的所有斜拉索索力识别问题中，本文方法的索力识别误差均不超过3%。这一结果证明了该方法具有较高的识别精度、较强的可靠性以及良好的可扩展性，适用于实际工程中斜拉索索力的识别。

Accurate identification of cable tension has important practical significance, but the vibration-based tension estimation methods has the problem of complex and diverse calculation formulas in application. Based on this, this paper introduces the physics-informed neural network (PINN) into the field of cable tension identification, and proposes a method of cable tension identification based on physics-informed neural network. In this method, the loss term of string vibration equation representing the cable tension is introduced into the loss function of neural network, and the network is trained by using the dynamic displacement response of the cable. The cable tension can be identified by minimizing the loss function. Based on the finite test data of numerical simulation examples and indoor test model, the proposed method is used to realize the accurate identification of cable tension. The results show that the identification error of the proposed method is less than 3% in all the cable tension identification problems set in this paper. This result proves that this method has high identification accuracy, strong reliability and good scalability, and is suitable for the identification of cable tension in practical engineering.

国家自然科学基金(51978577)