Shang Qiufeng, Qin Wenjie
Department of Electronic and Communication Engineering, North China Electric Power University, No. 619 Yong Hua Street, Baoding 071003, China.
Sensors (Basel). 2020 Mar 26;20(7):1840. doi: 10.3390/s20071840.
The fiber Bragg grating (FBG) sensor calibration process is critical for optimizing performance. Real-time dynamic calibration is essential to improve the measured accuracy of the sensor. In this paper, we present a dynamic calibration method for FBG sensor temperature measurement, utilizing the online sequential extreme learning machine (OS-ELM). During the measurement process, the calibration model is continuously updated instead of retrained, which can reduce tedious calculations and improve the predictive speed. Polynomial fitting, a back propagation (BP) network, and a radial basis function (RBF) network were compared, and the results showed the dynamic method not only had a better generalization performance but also had a faster learning process. The dynamic calibration enabled the real-time measured data of the FBG sensor to input calibration models as online learning samples continuously, and could solve the insufficient coverage problem of static calibration training samples, so as to improve the long-term stability, accuracy of prediction, and generalization ability of the FBG sensor.
光纤布拉格光栅（FBG）传感器的校准过程对于优化性能至关重要。实时动态校准对于提高传感器的测量精度至关重要。在本文中，我们提出了一种利用在线序贯极限学习机（OS-ELM）对FBG传感器温度测量进行动态校准的方法。在测量过程中，校准模型是不断更新而不是重新训练，这可以减少繁琐的计算并提高预测速度。将多项式拟合、反向传播（BP）网络和径向基函数（RBF）网络进行了比较，结果表明动态方法不仅具有更好的泛化性能，而且学习过程更快。动态校准使FBG传感器的实时测量数据能够作为在线学习样本不断输入校准模型，并能解决静态校准训练样本覆盖不足的问题，从而提高FBG传感器的长期稳定性、预测精度和泛化能力。
