针对实际应用中光纤光栅传感器服役时间长和工作环境恶劣等原因导致的标定曲线缓慢漂移问题，提出了一种改进在线顺序极限学习机用于光纤光栅传感系统的动态标定。在初始训练阶段引入正则化避免产生奇异矩阵，提高泛化能力。在线学习阶段引入自适应遗忘因子对新旧样本比重进行调整，提高预测精度。通过试验进行对比分析，验证了该方法的优越性。结果表明，与传统标定方法相比，所提方法的均方根误差(root mean square error, RMSE)指标始终最低，R2指标始终最高，具有较高的精度和较好的泛化性能，解决了标定曲线缓慢漂移问题，满足光纤光栅传感器的要求，可以应用于实际工程。
英文摘要:
Aiming at the problem of slow drift of calibration curve caused by long service time and bad working environment of fiber bragg grating (FBG) sensor in practical application, an improved online sequential limit learning machine is proposed for dynamic calibration of FBG sensor system. In the initial training stage, regularization is introduced to avoid singular matrix and improve generalization ability. In the online learning stage, an adaptive forgetting factor is introduced to adjust the proportion of new and old samples to improve the prediction accuracy. The superiority of this method is verified through the comparative analysis of experiments. The results show that, compared with the traditional calibration methods, the root mean square error (RMSE) index of the proposed method is always the lowest and R2 index is always the highest, which has higher accuracy and better generalization performance. It solves the problem of slow drift of the calibration curve, meets the requirements of FBG sensors, and can be applied to practical projects.
