Research On Intelligent Monitoring And Recognition Methods Of Railway Safety Based On Optical Fiber Sensing

With the rapid development of the global railway industry,"heavy-duty freight" and "high-speed passenger" together constitute the two major trends of railway in China.Heavy-haul trains are usually formed by large special freight cars,which are characterized by many cars,heavy load and long haul distance.Therefore,the requirements of operation safety measures are more stringent.On the other hand,China's railway operating mileage is long and the geology,landform and climatic conditions are very complex in the surrounding areas.Environment along the railway and natural disaster are always key issues affecting railway safety,such as illegal pedestrians on the road,foreign object penetration limit(color steel tile and dustproof screen),etc.After introducing the basic principle of distributed optical fiber vibration sensing system and the research status of railway safety monitoring technology in detail,this paper carried out the research on the theory and application of train running condition monitoring and railway perimeter intrusion protection based on distributed acoustic sensor(DAS)technology,and achieved some important progress.The main work and conclusions of the thesis are as follows:(1)The denoising algorithm of optical fiber sensing signal based on EMD-CIIT is proposed for the first timeThe pearson correlation coefficient(PCC)between intrinsic mode functions(IMFs)and measured signal are calculated so that can determine the trip point between noise dominant mode and signal dominant mode in IMFs.Then,the IMFs with PCC value greater than 0.1 are denoised based on the EMD-CIIT algorithm.Finally,the IMFs and residual are reconstructed to obtain denoised signals.As seen from the knocking and damaging test results,it proves the Signal Noise Ratio(SNR)of signal for the proposed method has been obviously enhanced compared with the wavelet threshold,Wiener filter and spectral subtraction denoising method,which can be raised up to 30.9 d B and32.82 d B respectively.(2)The SNR enhancement algorithm for ?-OTDR system based on CEEMDANIT is proposedThere will be "mode mixing" in EMD,and the modes decomposed by CEEMDAN contain some residual noise and "spurious" modes.To solve these problems,the SNR enhancement algorithm for ?-OTDR system based on CEEMDAN-IT is proposed.The trip point is determined according to the fact that the product of energy density of the IMF and its corresponding averaged period is a constant.Then,the IMFs of noise dominant mode are treated with interval thresholding,and finally the IMFs and residual are reconstructed to obtain denoised signals.CEEMDAN-IT can achieve better effect that the SNR of knocking and damaging signal at the disturbance location increases to51.21 d B and 52.11 d B respectively when compared with the wavelet thresholding and CEEMDAN hard thresholding.(3)Research on the application of DAS technology in the monitoring of heavy-haul railway train running conditionThe novel method for monitoring the train running condition of the heavy-haul railway is put forward by using DAS technology.The method can pick up the vibration signal generated by the surrounding environment through optical fiber for real-time monitoring and positioning.Firstly,the wavelet decomposition algorithm is used to denoise the original signal.The experiment selects three wavelet basis functions,db12,sym8 and coif5 with a scale range of 1-4.Then,the vibration signal is preprocessed,including feature extraction in time domain,peak detection and secondary filtering.Finally,the connected components labeling algorithm is used to detect connected region in the image,and the train movement trajectory is obtained.Due to the uncertainty of wavelet basis function and decomposition level,a novel denoising method is verified by analyzing the time domain waveform characteristics of vibration signal,that is,the cubical smoothing algorithm with five-point approximation("5-3").It is found that the effect of "5-3" is basically the same as the wavelet decomposition,but the calculation time is greatly reduced.In addition,the improved Canny algorithm is used to extract the edge of the target in the image,so as to realize online train monitoring.Through on-site test of Shuohuang Railway,the fact that the application can accurately acquire the running condition information such as the length,speed,position and direction of trains in real-time.The positioning error is about 10 m.(4)Research on recognition method of railway perimeter intrusions based on?-OTDR optical fiber sensing technologyTo reduce the nuisance alarm rate(NAR)and improve the reliability of railway intrusion monitoring system,the XGBoost model based on data driven is built to identify different intrusions.The method is implemented from the sensor to the final output,including data collection,signal processing(framing,denoising),feature extraction,model designing and evaluation.The results of field experiments show that the average recognition accuracy of the model is as high as 98.5% for the common five intrusions.All the related performance metrics of confusion matrix are better than other popular methods,such as Random Forest,Support Vector Machine and Multilayer Perceptron.