Abstract: Magneto-elastic inductance sensor is the main equipment to monitor the stress of prestressed steel strand by magneto-elastic inductance method, but the temperature will cause monitoring error. In order to improve the monitoring accuracy, firstly, the principle of stress monitored by magneto-elastic inductance method and the influence mechanism of temperature were analyzed. Secondly, four groups of prestressed steel strands with different tension control forces were tested for temperature compensation, and the temperature sensitivity of the sensor was studied to find that the sensor inductance drift was related to the temperature variation amplitude and path, and the relative error caused by temperature was up to 11.1%. Finally, BP neural network and optimized GA-BP neural network were respectively used to compensate temperature by K-fold cross verification method. The research results show that: compared with BP neural network, GA-BP neural network can effectively correct the temperature error, improve the generalization ability, and reduce the relative error to 3.3% after optimization.

Key words: bridge engineering, magneto-elastic inductance method, prestress monitoring, temperature compensation, K-fold cross validation, optimization neural network

摘要： 磁弹电感传感器是磁弹电感法监测预应力钢绞线应力的主要设备。温度会引起监测误差，为提升监测精度，首先，分析了磁弹电感法监测应力的原理以及温度影响机理；其次，对4组不同张拉控制力的预应力钢绞线进行温度补偿试验，研究了传感器的温度敏感性，发现传感器电感漂移与变温幅度和路径相关，温度引起的相对误差最大达11.1%；最后，利用K折交叉验证方法，分别采用BP神经网络与优化后的GA-BP神经网络进行温度补偿。研究结果表明：相较于BP神经网络，GA-BP神经网络可有效修正温度误差，提升泛化能力，优化后相对误差降至3.3%。

关键词: 桥梁工程, 磁弹电感法, 预应力监测, 温度补偿, K折交叉验证, 优化神经网络