Online temperature drift compensation of Fabry-Perot filter based on machine learning and linear fitting

The measurement accuracy of fiber Bragg grating (FBG) sensors is significantly impacted by the temperature drift of Fabry-Perot (F-P) filters. The current hardware-based compensating techniques are complicated and expensive. In this study, a high-accuracy, low-cost online compensating approach was proposed. The temperature drift model is built using the least square support vector regression (LSSVR), and the drift of a reference grating is skillfully used as model input to increase model accuracy. Contrarily, the conventional online compensation necessitates an enormous amount of computation time and power to build and updating model for each sensor. This work proposed a novel inter-compensation mechanism to address the computation issue. After two FBG's drifts are predicted, the predicted results are used to determine a nearby FBG' drift by linear fitting. The experimental results demonstrate that, in monotonic cooling and cooling-heating environments, respectively, the proposed approach reduces the maximal absolute error (MAXE) by 39.85 % and 82.78 % when compared to the classic moving-window method. Additionally, in two cases, the computing speeds are raised by about a third.