A new population initialization method of genetic algorithm applied in FBG inhomogeneous strain demodulation

摘要

FBG is a kind of promising high precision strain sensor, and it can not only detect the homogeneous strain, but also identify non-uniform strain distribution. In the application of FBG in inhomogeneous strain sensing, genetic algorithm is an important method to reconstruct the non-uniform strain from reflection spectra of FBG. However, the practical reconstruction of genetic algorithm demonstrates its shortcomings such as low computational efficiency, easily falling into local optimal solution, etc , and it is well known that there is a great relationship between computational efficiency and population initialization of genetic algorithm. In general genetic algorithm employed in FBG strain reconstruction, the initialized population is randomly distributed strain along FBG axial direction, which ignores the continuity between neighbor strains. To reduce the number of population parameters and make the original population more close to the real strain distribution, a new method of population initialization is proposed here, that is using polynomial function parameters to be the initialized population instead of the randomly distributed strain, supposed that the FBG axial strains can be described as a polynomial function with independent variable of axial position. In simulation experiments, the reflection spectrums of a 10mm-long FBG are obtained from T-Matrix method in four cases of strain-free, linear-distributed strain, parabola-distributed strain and exponential -distributed strain, and then the general genetic algorithm and the new genetic algorithm with simplified population initialization were applied to reconstruct the strain distribution from the reflection spectrums respectively. The experiment results verify the supposition of the polynomial function of the FBG, and show clearly that the new method can improve the computational efficiency of genetic algorithm in FBG inhomogeneous strain demodulation greatly. From the results, it is found that with the same calculation accuracy, the computing time of the new population initialization method is reduced to about 1/5 of the general on average.