Research On Temperature Demodulation Method Of Optical Fiber Brillouin Gain Spectrum Based On AioT Framework

Under the era background when Internet of Internet of Things(Io T)is moving towards Artificial Intelligence of Things(AIo T),novel sensing,monitoring,and demodulation systems are continuously emerging and staying up-to-date with the latest advancements.With the development of artificial intelligence technology to various application fields,which also driving innovation across the optical fiber sensing system based on Io T.The fiber sensing has been injected with new intelligent attributes.Sensing and demodulation schemes based on optical fiber Brillouin scattering theory have been widely used in the fields of large-scale infrastructure health monitoring and public environment safety applications.The primary research object of this thesis is the temperature demodulation system of the fiber Brillouin gain spectrum.Specifically,focused on the demodulation and processing methods of the Brillouin scattering spectrum data.A method of selecting characteristic data and filtering redundancy data by means of sampling is proposed,so as to find a breakthrough point for solving the problem of low processing efficiency of Brillouin gain spectrum data.The research method combining the advantage of sampling processing method and the Artificial Neural Network model,which is determined to optimize the performance of the Brillouin gain spectrum temperature demodulation system.Based on the AIo T framework,the end-cloud collaboration and distributed demodulation scheme design is realized.By analyzing the effect of sampling method on the performance of the Brillouin gain spectrum demodulation system,a prototype of the Brillouin gain spectrum temperature demodulation scheme is proposed.The sampling strategy and the neural network model to support multi-criteria decisionmaking,so as to provide a better solution for the matching application demand.The main work and innovations of this paper are as follows:1.To address the problem analyzes the impact and bottleneck of data dimension on temperature demodulation accuracy when utilizing a neural network model for temperature demodulation.A variance-weighted sampling method for demodulating fiber temperature through stimulated Brillouin gain spectrum was proposed in this thesis.Through the study of the relationship between statistical characteristics of Brillouin gain spectrum and temperature information.The method selected by variance characteristics to sampling the original data to low dimension is proposed.The variance weight sampling vector is designed,and the data with low dimension is further simplified.Using the neural network model to demodulate the temperature of the Brillouin gain spectrum,it is verified that the proposed method significantly improves temperature demodulation accuracy,especially when sampling efficiency are taken into account.2.To address the problem of data redundancy in the raw data of Brillouin gain spectrum.A scheme to optimize the performance of the resampling filter method for temperature demodulation of fiber Brillouin gain spectrum in this thesis.The influence of the Brillouin gain spectrum curve feature on temperature demodulation is studied,the optimized frequency range including this feature is clarified on the temperature demodulation model,and the original data is filtered.The compression of the traditional uniform sampling interval is realized by using the second-order differential operator resampling filtering method.Experimental results verified that the resampling filtering method can improve the temperature demodulation accuracy and greatly compress the original sampling data.3.To address the problem of noise reducing the stability of the Brillouin gain spectrum temperature demodulation system.A Brillouin gain spectrum temperature balance demodulation system is proposed in this thesis.Combining the distributed computing capabilities of AIo T devices,the resampling filtering method and the variance weight coding sampling method are simultaneously deployed in the system.Combining the advantage of the above two methods,a balanced sampling method is proposed.The simulated noise with different types and strength is used to synthesize the Brillouin gain spectrum dataset.Under different noise conditions,it is verified that this balance demodulation system has the characteristics of strong stability and high demodulation accuracy.4.To address the problem of matching different sampling methods with the performance of the Brillouin gain spectrum temperature demodulation system,a segmented verification evaluation mode and decision-making method are proposed.Evaluation datasets suitable for different conditions were constructed.A comprehensive evaluation model is designed using the objective weighting method to assess the statistical characteristics of performance indicators.The decision-making workflow of the sampling method under the AIo T framework is designed,and the decision method of sampling matching with system performance is proposed.Moreover,the research results have the potential to provide useful insights for other data processing methods based on Artificial Intelligence of Things.