错位光纤干涉激光谱结合BP神经网络的温度传感研究

在分析不同温度时单模错位光纤干涉光谱对应波长的条件下，搭建三层BP神经网络模型对温度传感进行研究，解决了常规光纤测温系统复杂和精度不高的问题。对建立的网络模型参数进行探讨，将采集的激光波长与对应的温度数据，经BP神经网络训练，对比得到最佳网络结构，达到在训练完成的网络输入层输入激光波长值时，便可在输出层得到对应的温度预测值。结果证明，实验输出的预测温度值与实际温度值之间表现出明显的相关性，即预测值能够逼近实测值。温度校正和预测相关系数分别达到0.999 61和0.979 27，校正标准误差与预测标准误差分别为0.017 5和0.144 0，得到预测集的平均相对误差为0.17%，剩余预测误差RPD可达到5.258 3，RPD大于3.0，说明定标效果良好，所建模型可用于实际的检测。另外，将该算法用于了带校正的双耦合结构单模错位光纤测温系统中，结果表明BP神经网络方法能够较好的处理错位光纤测温系统中激光光谱数据和温度之间的非线性关系，预测温度值与实测温度值之间的相关度为0.996 58，得到预测温度值与实际温度值之间平均相对误差为0.63%，从而提高了光纤测温传感器的精度和稳定性，同时也验证了该算法在光纤传感上的可行性，也为错位光纤的压力、曲率等其他物理量传感的精确测量提供了新思路。

Combined Transmission Laser Spectrum of Core-Offset Fiber and BP Neural Network for Temperature Sensing Research

When studying the wavelength response to the temperature of the single mode fiber interference laser spectrum,a three layer BP neural network model is built to solve the problem of high complexity and low accuracy of temperature measure-ment system.With the Discussion of the parameters of network model,we obtain the optimal network structure by comparing the data acquisition which is the laser wavelength corresponding to its temperature trained by BP neural network.With network training completed and the wavelength of input laser reached the specified value,the corresponding temperature prediction can be obtained from the output layer.In conclusion,it shows a clear correlation between the predictive value and the actual value,i.e. the former is approximately equal to the latter.The correlation coefficients of the calibration and prediction are 0.9 9 9 6 1 and 0.979 27,respectively;while the standard errors of the calibration and prediction are 0.017 5 and 0.144 0,respectively,and the average relative error of prediction set is 0.17%.The residual predictive deviation (RPD),obtained theoretically,is 5.258 3. RPD>3.It indicates that the calibration effect is good,and the model can be used for practical testing.In addition,the algo-rithm is also applied to the system of double coupled structure with single-mode core-offset fiber and correction for the tempera-ture measurement.The results show that BP neural network method can deal with the nonlinear relationship between the laser spectral data and the temperature in the optical fiber temperature measurement system.The correlation and the average relative error between the predicted temperature and the true temperature are 0.996 58 and 0.63%,respectively.The precision and sta-bility of the fiber optic temperature sensor are significantly improved.At the same time,the feasibility of the proposed algorithm is verified in the fiber optical sensor system.It also provides a new way for the accurate measurement of pressure,curvature and other physical quantities of the core-offset fiber.