High Precision and Speed Demodulation Algorithm for Optical Fiber Fabry–Perot Temperature Sensor

To address the contradiction between speed and accuracy in the demodulation algorithms of fiber optic Fabry-Perot temperature sensors, as well as the problem of the low accuracy of existing neural network demodulation methods, a temperature demodulation method combining principal component analysis (PCA), particle swarm optimization (PSO) and Elman neural network is proposed for the first time. The high linear correlation among input spectral features and the significant overlap in content found in neural network algorithms worsen the variance of model training parameter estimates, leading to decreased prediction accuracy—a challenge that cannot be entirely resolved through model correction alone. PCA effectively resolves the problem of the multicollinearity inherent in spectral data, and reduces the dimensionality of the original data features. The data compression ratio can reach 0.0005. PSO enhances the ability of neural network to avoid local minima in the training process, and helps neural network to search global minima more comprehensively in the parameter domain. The addition of a feedback mechanism in the Elman neural network enhances its fitting ability, enabling it to handle more complex data than traditional feedforward neural networks. Without spectral denoising, the model was trained and tested on 4100 samples with temperatures ranging from $- 5~^{\circ }$ C to $35~^{\circ }$ C. The final results achieved a root mean square error (RMSE) of $0.002~^{\circ }$ C and a demodulation speed of 0.06 ms. Consequently, the proposed PCA-PSO-Elman demodulation method can effectively capture the nonlinear mapping between temperature and spectra, and achieve higher demodulation accuracy and greater demodulation speed.