Seawater Salinity Sensor Based on Dual Resonance Peaks Long-Period Fiber Grating and Back Propagation Neural Network

In this work, a seawater salinity sensor based on dual resonance peaks long-period fiber grating (LPFG), and back propagation neural network (BPNN) has been proposed. An LPFG working near a dispersion turning point (DTP) has been theoretically designed and actually fabricated, which has a refractive index (RI) sensitivity of 1304 nm/RIU in the RI range of 1.33156-1.39947. Then, the sensor was investigated under different salinity and temperature conditions, in which the salinity changed from 5 parts per thousand to 40 parts per thousand and the temperature changed from 0 degrees C to 30 degrees C. The highest seawater salinity sensitivity of 0.2662 nm/parts per thousand was achieved when the temperature was 30 degrees C, while the highest average temperature sensitivity was 0.482 nm/degrees C. The BPNN was employed for temperature compensation and salinity prediction because of the nonlinear response to the variations of seawater salinity and temperature, whose mean absolute error and maximum absolute error are 2.1774 and 6.8901, respectively. Because the prediction accuracy is poor, the genetic algorithm has been employed for BPNN optimization. Then, BPNN-based genetic algorithm method achieved a mean absolute error of 0.10139 and a maximum absolute error of 0.19016, whose prediction accuracy was increased by an order of magnitude. This novel sensor based on optimized BPNN could be applied in the field of seawater salinity measurement.

Automated summary

In this study, a seawater salinity sensor based on dual resonance peaks long-period fiber grating (LPFG) and back propagation neural network (BPNN) was proposed. The LPFG was designed and fabricated to have a refractive index sensitivity in the range of 1.33156-1.39947. The sensor was tested under different salinity and temperature conditions, and BPNN was used for temperature compensation and salinity prediction. BPNN-based genetic algorithm optimization significantly improved the prediction accuracy.