Dissertation > Industrial Technology > Automation technology,computer technology > Automation technology and equipment > Automation components,parts > Transmitter ( converter),the sensor

A Pattern Recognition Method in M-Z Fiber Distributed Disturbance Sensing System

Author WangSiYuan
Tutor LouShuQin
School Beijing Jiaotong University
Course Communication and Information System
Keywords fiber distributed disturbance sensing system (FDDSS) Mach-Zehnderinterferometer frequency-time feature ANN pattern recognition
Type Master's thesis
Year 2014
Downloads 18
Quotes 0
Download Dissertation

ABSTRACT:M-Z fiber distributed disturbance sensing system (FDDSS)with its advantages like:simple structure, wide detection range, no external field power as well asno electromagnetic interference, is wildly used in areas such as perimeter security or pipelines. In this paper, a novel pattern recognition algorithm is proposed with the help of a short time frequency-time characteristic which extracted from the calculation of short time average level cross rate, and used to feature element modeling. The paper also fully summarizes the current mainstream signal feature extraction algorithms and pattern recognition algorithms, and a novel algorithm is proposed and fully verified by experiment.This paper conducted the following research areas:1, the principle of M-Z fiber distributed disturbance sensing system. Starting from waveguide theory, combined with the theory of elasticity of the fiber, the paper derived the relationship between the output signal of the sensor and the external disturbance signal. According to the characteristics of the output signal,the frequency-timecharacteristic of the sensor can be used for pattern recognition.2, the sensor output signal’s frequency-time feature extraction algorithm. From the theoretical and experimental in-depth analysis of several mainstream frequency-time feature extraction algorithm, combined with the advantages and disadvantages of each method,it is proposed to extract that the average short-term level-cross rate of the signal can greatly reduce the computational complexity and increase the speed of operation.3, the signal characteristics of dimensionality reduction algorithms and eliminate time method to the alignment error. Artificial neural network pattern recognition algorithm is the core algorithm and analyzing the advantages and disadvantages, which leads to the need ofdimensionality reduction problems and eliminate time-aligned problems. Getting inspired by the idea of mathematics graphics, cutting the data using the "overlap " technology into segments, creating multiple models for each segments, using a dynamic programming algorithm to select the best model for each segments,is the main procedure of the algorithm. The experimental results show that this method can effectively distinguish between the role of a variety of transient disturbance events, with an average speed of recognition within0.26seconds, the average recognition accuracy above97.4%.

Related Dissertations
More Dissertations