Research On The Key Technologies Of Information Extraction Of Brillouin Optical Time Domain Analyzer

Fiber optic sensing technology is moving forward day by day,and distributed fiber optic sensing system based on Brillouin scattering is gaining more and more attention in the areas such as: pipeline monitoring,national defense and structural-health monitoring,and it has been applied in national economy and defense industry.As the sensing distance and spatial resolution increase,the data that needs to be processed in the sensing system is increasing exponentially,traditional data processing methods fail to meet the real-time requirements of the sensing systems.To efficiently and accurately obtain distributed fiber optic sensing information is the key to the problem,and recently machine learning and neural networks have become research hotspots.The research in this thesis is based on this,and the main contents of this thesis are as follows:(1)In order to reduce the influence of noise on the later extraction of Brillouin sensor information,the methods of wavelet transformation,gaussian filtering and,wavelet transformation followed by gaussian filtering were used to denoise the experimental data.The experimental results show that,the third method can improve the signal-to-noise ratio of the experimental data by about 2.6d B while ensuring an unchanged spatial resolution.(2)To handle with the problem that a traditional neural network is usually randomly initialized and tend to fall into local optimum resulting in a slow convergence speed.Whale optimization algorithm(WOA)is proposed to optimize the initialization weights and biases of the neural network to prevent the network falling into the local optimum with a slower convergence speed,so that the performance of the neural network can be improved.The experimental results show that,the root mean square error(RMSE)and standard deviation(SD)of the sensor information extracted by the neural network optimized by WOA algorithm are reduced by about 35% and 29% respectively,and the network training time is reduced by about half.(3)To target the problems that the uneven population distribution and that the linear convergence factor in the WOA algorithm cannot balance the global exploration and local exploitation,mix whale optimization algorithm(MWOA)is proposed with selectionbased population initialization strategy and a nonlinear convergence factor.MWOA algorithm is proved to be able to further improve the performance of neural network with experimental data of different signal-to-noise ratios.Comparing with WOA and particle swarm optimization alogrithm,and traditional neural network,the MWOA algorithm can further optimize the neural network,and the RMSE and SD of extracting sensor information are reduced by about 46% and 43% than traditional neural network while reducing the network training time by about 63%.