Optical frequency domain distributed optical fiber strain / temperature sensing signal serial and parallel processing method based on graphics processor

Abstract

According to the optical frequency domain distributed optical fiber strain / temperature sensing signal serial and parallel processing method based on the graphics processor, the signal processing flow is optimized, and the signal processing speed of an optical fiber sensing system is improved by cooperatively completing signal processing through a central processing unit and the graphics processor. According to the method, signals are efficiently processed by utilizing a serial parallel computing architecture. In the serial processing part, signals of all sensing units of the optical fiber are demodulated in sequence, spectrum offset is converted into strain and temperature variation, and the number of the sensing units demodulated in parallel is dynamically adjusted according to the optical path variation of the sensing units. In the parallel processing part, signals of a plurality of sensing units are demodulated at the same time on the basis of accurately compensating the optical path. Meanwhile, the graphics processor uniformly computes a CUDA (Compute Unified Device Architecture) platform to complete serial and parallel signal processing, so that the parallel signal processing speed of massive sensing units can be remarkably accelerated, the frequency response of the sensing system is improved, and the requirements of long-distance, high-density and high-frequency response distributed optical fiber strain / temperature sensing are met.

Description

Technical Field

[0001] The present invention belongs to the technical fields of measurement and signal processing, and relates to a serial-parallel processing method for optical frequency domain distributed fiber strain / temperature sensing signals based on a graphics processing unit. Background Art

[0002] In recent years, with the rapid development of fields such as aerospace, medical surgical robots, and civil engineering structural health monitoring, the demand for high-precision and distributed measurement of structural strain and temperature has become increasingly urgent. As an advanced distributed fiber sensing technology, an optical frequency domain reflectometer (OFDR) can achieve high-precision strain and temperature measurement along the fiber and has broad application prospects. However, traditional OFDR systems face problems of low computational efficiency and insufficient real-time performance when processing long-distance and high-density sensing data, which severely limits their popularization in practical applications. Therefore, developing a computational method capable of efficiently processing OFDR signals is of great significance for promoting the development of distributed fiber sensing technology. The serial-parallel processing method for optical frequency domain distributed fiber strain / temperature sensing signals based on a graphics processing unit can significantly improve the signal processing speed and accuracy by reasonably allocating computational tasks and giving full play to the synergistic effect of the central processing unit (CPU) and the graphics processing unit (GPU), meeting the requirements for real-time performance and high-precision measurement in practical engineering. Currently, the main methods for processing optical frequency domain distributed fiber strain / temperature sensing signals include the following:

[0003] 1. In 2019, Wang Heng et al. parallelized the time-consuming fast Fourier transform (FFT) and cross-correlation operations in the OFDR processing algorithm, and also parallelized the sliding time gating algorithm and the overall OFDR demodulation algorithm.

[0004] 2. Lu Ziyi et al. from Hebei University proposed to use the Rayleigh backscattering enhanced fiber (RBEF) with the characteristics of random high backscattering points as the sensing fiber. The OFDR system acquires time-domain signals, which are converted into distance-domain signals after fast Fourier transform (FFT). By using the sliding cross-correlation method, the optical scattering spectrum offset before and after temperature change along the fiber axis is analyzed, and the wavelength deviation of the Rayleigh scattering peak with high precision is accurately measured through sliding window operation and associated with the temperature change, thus realizing distributed temperature sensing.

[0005] With the continuous improvement of the spatial resolution and sensing distance of OFDR systems, their application scenarios have been continuously enriched. However, the synchronous growth of data volume has also led to a sharp increase in signal demodulation time, thus restricting the further application of OFDR systems. Improving the real-time performance of OFDR systems has become an urgent problem to be solved. Current research mostly focuses on the upgrade of hardware platforms. In terms of software, most only accelerate some calculations and lack systematic design and research on the global parallel demodulation architecture.

[0006] In view of this, the present invention discloses a serial-parallel processing method for optical frequency domain distributed fiber strain / temperature sensing signals based on a graphics processing unit. According to the characteristics of the serial demodulation algorithm of the optical frequency domain reflectometer, this method proposes a serial-parallel demodulation processing method for sensing signals and an optical path compensation scheme, and flexibly adjusts the number of sensing units for parallel processing according to the differences in strain or temperature measurement results in different sensing fiber application scenarios. At the same time, according to the characteristics of signal processing, the CPU and GPU are used to cooperate to complete the signal processing process to achieve efficient and real-time signal demodulation. Summary of the Invention

[0007] The object of the present invention is to propose a serial-parallel processing method for optical frequency domain distributed fiber strain / temperature sensing signals based on a graphics processing unit, improve the signal processing speed and accuracy, and meet the requirements of real-time performance and high-precision measurement in actual engineering. The present invention utilizes the continuity of distributed fiber measurement in the fiber sensing system, optimizes the signal processing flow through a serial-parallel computing architecture, and improves the measurement accuracy and demodulation speed. This method uses the CPU and GPU to cooperate to complete the signal processing process, which can significantly improve the signal processing speed of the optical frequency domain distributed fiber sensing system, solve the problem of low real-time performance in traditional methods, and is especially suitable for long-distance and high-density strain and temperature monitoring scenarios.

[0008] The working principle of the present invention is to optimize the signal processing flow through the cooperation of a serial-parallel computing architecture and a central processing unit and a graphics processing unit. In the serial processing part, the signals of each sensing unit of the optical fiber are demodulated in sequence. By using the cross-correlation calculation of the reference spectrum and the measurement spectrum, combined with the theory of optical frequency domain distributed fiber, the number of spectral offset points is converted into the amount of strain and temperature change, and the number of sensing units for parallel demodulation is dynamically adjusted according to the optical path change amount caused by the sensing unit, without affecting the sensing stability and flexibly adjusting the parallelism; in the parallel processing part, the GPU is used to demodulate the signals of multiple sensing units at the same time, a parallel kernel function is constructed using the unified computing device architecture CUDA platform of the graphics processing unit, and the number of sensing units for parallel calculation is dynamically adjusted to improve the signal processing speed, enhance the measurement accuracy, and achieve high-precision and real-time monitoring of the distributed strain and temperature of the optical fiber.

[0009] The technical solution of the present invention is: A method for serial and parallel processing of optical frequency domain distributed optical fiber strain / temperature sensing signals based on a graphics processing unit. The optical frequency domain distributed optical fiber strain / temperature sensing system includes an optical frequency domain distributed optical fiber signal acquisition device, a computer, and a sensing optical fiber. Among them, the optical frequency domain distributed optical fiber signal acquisition device is connected to the sensing optical fiber to form an optical path, and the optical frequency domain distributed optical fiber signal acquisition device is connected to the computer through a data line to form data communication. The computer can control the working state of the optical frequency domain distributed optical fiber signal acquisition device through the communication instruction of the data line and obtain the sensing data of the optical frequency domain distributed optical fiber signal acquisition device. The strain / temperature sensing information of the sensing optical fiber is obtained by the following steps:

[0010] Step 1: Set the spatial resolution of distributed optical fiber strain / temperature sensing, the starting position of optical fiber analysis for distributed optical fiber strain / temperature sensing, and the ending position of optical fiber analysis for distributed optical fiber strain / temperature sensing on the central processing unit of the computer, and calculate the number of window points and the number of windows of the short-time Fourier transform corresponding to the spatial resolution, the serial number of the starting data point for analysis in the optical frequency domain corresponding to the starting position of optical fiber analysis, and the serial number of the ending data point for analysis in the optical frequency domain corresponding to the ending position of optical fiber analysis according to the optical frequency domain distributed optical fiber sensing theory;

[0011] Step 2: Before strain or temperature acts on the sensing optical fiber, the computer controls the distributed optical fiber signal acquisition device to collect the time-domain sensing signal of the sensing optical fiber and stores it in the memory of the central processing unit of the computer. Using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit, the time-domain sensing signal in the memory of the central processing unit is transmitted to the video memory of the graphics processing unit, and the optical frequency domain sensing signal in the reference state is obtained by performing a fast forward Fourier transform of cufftPlan1d of the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit in the graphics processing unit. Using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit to construct a parallel computing kernel function to complete the following data processing. Starting from the serial number of the starting data point for analysis in the optical frequency domain, continuously select M reference data segments connected end to end in the optical frequency domain sensing signal in the reference state, corresponding to M sensing units. Each segment of data in the reference data segment is defined as Ref1, Ref2... Ref M , and the number of data points in each segment of data Ref1 to Ref M in the reference data segment is the number of window points. Add Z zeros to the end of each segment of data Ref1 to Ref M respectively, and then perform a multi-segment parallel inverse Fourier transform processing of cufftPlanMany of the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit in the graphics processing unit to obtain the first reference data segment, namely Ref_I1 to Ref_I M, and then a parallel computing kernel function is constructed using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit. For each data point in the first reference data segment, namely Ref_I1 to Ref_I M , after taking the modulus of each data point, the second reference data segment, namely Ref_IM1 to Ref_IM M is obtained. And the parallel broadcast reduction method is used to calculate the average values Mr1 to Mr of the data points within the second reference data segment, namely Ref_IM1 to Ref_IM M . Using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit, a parallel computing kernel function is constructed to subtract the average values Mr1 to Mr of the corresponding data segments from each data point within the second reference data segment, namely Ref_IM1 to Ref_IM M . The results are the reference spectral segment sets of the corresponding sensing units, namely Ref_S1 to Ref_S M . M ; M ;

[0012] Step 3: After strain or temperature acts on the sensing optical fiber, the computer controls the distributed optical fiber signal acquisition device to collect the time-domain sensing signal of the sensing optical fiber and stores it in the memory of the central processing unit of the computer. Using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit, the time-domain sensing signal in the central processing unit memory is transmitted to the video memory of the graphics processing unit, and the cufftPlan1d fast Fourier transform of the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit is used in the graphics processing unit to obtain the optical frequency-domain sensing signal in the measurement state. In the graphics processing unit, the cumulative optical path offset is initialized to 0, and the sensing parallel computing quantity N = N1;

[0013] Step 4: In the graphics processing unit, starting from the starting data point serial number in the optical frequency domain within the optical frequency-domain sensing signal in the measurement state, a parallel computing kernel function is constructed using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit to implement the following data processing. Continuously select N measurement data segments connected end to end corresponding to N sensing units. Each segment of data in the measurement data segment is respectively defined as Mea1, Mea2... Mea N . The number of data points within each segment of data Mea1 to Mea N in the measurement data segment is the window point number. The length of the measurement data segment in the optical frequency domain is N multiplied by the window point number. Z zeros are respectively filled at the end of each segment of data Mea1 to Mea N in the measurement data segment. The cufftPlanMany multi-segment parallel fast inverse Fourier transform processing of the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit is used in the graphics processing unit to obtain the first measurement data segment, namely Mea_I1 to Mea_I N, and then a parallel computing kernel function is constructed using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit (GPU) to respectively take the modulus of each data point in Measurement Data Segment 1, i.e., from Mea1 to Mea N to obtain Measurement Data Segment 2, i.e., from Mea_IM1 to Mea_IM N , and the parallel broadcast reduction method is used to calculate the average values Mm1 to Mm of the data points in Measurement Data Segment 2, i.e., from Mea_IM1 to Mea_IM N ; a parallel computing kernel function is constructed using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit (GPU) to subtract the average values Mm1 to Mm of the corresponding data segments from each data point in Measurement Data Segment 2, i.e., from Mea_IM1 to Mea_IM N ; the results are the measurement spectral segment sets of the corresponding sensing units, i.e., from Mea_S1 to Mea_S N ; N ; N ;

[0014] Step 5: In the graphics processing unit, a cross-correlation data processing function based on multi-segment parallel fast forward and inverse Fourier transforms using the cufftPlanMany is constructed using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit (GPU). The cross-correlation between the measurement spectral segment sets of N sensing units, i.e., from Mea_S1 to Mea_S N and the reference spectra of the N sensing units in the corresponding reference spectral segment set is obtained to get the spectral offset point number sequences, i.e., from Bias1 to Bias N . Furthermore, a parallel computing and judgment kernel function is constructed using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit (GPU) for the following data processing. According to the optical frequency domain distributed optical fiber theory, the spectral offset point number sequences are converted into the spectral wavelength change amount sequences of the corresponding sensing units, i.e., from Wave1 to Wave N and the strain / temperature sequences of the corresponding sensing units, i.e., from Sen1 to Sen N . In addition, the average value of the strain / temperature sequences of the corresponding sensing units, i.e., from Sen1 to Sen N is calculated The number of optical path change points caused by the strain / temperature of N sensing units is calculated where W c is the swept center wavelength of the distributed optical fiber signal acquisition device. The termination position point numbers of the N sensing units are the sum of the starting data point number for analysis in the optical frequency domain, the number of optical path change points L, and the length of the measurement data segment. If the average value of the strain / temperature sequence is greater than the threshold T, then the parallel computing quantity N = N1; if the average value of the strain / temperature sequence is less than the threshold T, then the parallel computing quantity N = N2, where N1 < N2;

[0015] Step 6: In the graphics processing unit, starting from the termination position point number in the optical frequency domain sensing signal in the measurement state, use the unified computing device architecture CUDA platform of the graphics processing unit to construct a parallel computing kernel function to implement the following data processing. Continuously select N measurement data segments connected end to end corresponding to N sensing units. Each segment of data in the measurement data segment is defined as Mea1, Mea2... Mea N , and the number of data points in each segment of data Mea1 to Mea N within the measurement data segment is the window point number. The length of the measurement data segment in the optical frequency domain is N multiplied by the window point number. For each segment of data Mea1 to Mea N in the reference measurement data segment, append Z zeros at the end respectively. In the graphics processing unit, use the cufftPlanMany multi-segment parallel fast Fourier transform processing of the unified computing device architecture CUDA platform of the graphics processing unit to obtain the first measurement data segment, namely Mea_I1 to Mea_I N , and then use the unified computing device architecture CUDA platform of the graphics processing unit to construct a parallel computing kernel function. Take the modulus of each data point in the first measurement data segment, namely Mea1 to Mea N to obtain the second measurement data segment, namely Mea_IM1 to Mea_IM N , and use the parallel broadcast reduction method to calculate the average value Mm1 to Mm of the data points within the second measurement data segment, namely Mea_IM1 to Mea_IM N , use the unified computing device architecture CUDA platform of the graphics processing unit to construct a parallel computing kernel function to subtract the average value Mm1 to Mm of the corresponding data segment from each data point within the second measurement data segment, namely Mea_IM1 to Mea_IM N . The result is the measurement spectrum segment set of the corresponding sensing unit, namely Mea_S1 to Mea_S N N ; N ;;

[0016] Step 7: In the graphics processing unit, use the unified computing device architecture CUDA platform of the graphics processing unit to construct a cross-correlation data processing function based on the cufftPlanMany multi-segment parallel fast forward and inverse Fourier transforms, and parallelly calculate the cross-correlation between the measurement spectrum segment set of N sensing units, namely Mea_S1 to Mea_S N and the reference spectra of N sensing units in its corresponding reference spectrum segment set to obtain the spectral offset point number sequence, namely Bias1 to Bias N . Furthermore, use the unified computing device architecture CUDA platform of the graphics processing unit to construct a parallel computing and judgment kernel function to perform the following data processing. According to the theory of the optical frequency domain distributed optical fiber, convert the spectral offset point number sequence into the spectral wavelength change amount sequence of the corresponding sensing unit, namely Wave1 to Wave Nand the strain / temperature sequences of the corresponding sensing units, namely Sen1 to Sen N and calculating the strain / temperature sequences of the corresponding sensing units, namely Sen1 to Sen N for their average values Calculating the number of points of the optical path change caused by N sensing units under the action of strain / temperature where W c is the swept center wavelength of the distributed optical fiber signal acquisition device. The serial number of the termination position point of the N sensing units is the sum of the termination position point serial number in the sixth step, the number of points of the optical path change L, and the length of the measured data segment in the sixth step. If the average value of the strain / temperature sequence is greater than the threshold T, the number of parallel sensing calculations N = N1. If the average value of the strain / temperature sequence is less than the threshold T, the number of parallel sensing calculations N = N2, and N1 < N2;

[0017] Step 8: Repeatedly execute Step 6 and Step 7 in sequence until the strain / temperature of all M sensing units are obtained, that is, the strain / temperature sensing information of the sensing optical fiber is obtained.

[0018] The two values of the number of parallel sensing calculations N are that the value range of N1 is 10 to 32, the value range of N2 is 32 to 128, and the value range of the threshold T is 1000 με to 9000 με or 100 °C to 900 °C.

[0019] Advantages of the present invention:

[0020] 1. The present invention can dynamically adjust the number of parallel demodulated sensing units according to the optical path change amount caused by the sensing units, optimize the signal processing flow by using a serial-parallel computing architecture, without affecting the sensing stability, and has the advantages of fast speed, good flexibility and adaptability.

[0021] 2. The method of the present invention uses the CPU and GPU to cooperate to complete the signal processing process, can dynamically compensate the optical path of the sensing unit through serial-parallel computing, and has a significant effect in improving the signal processing speed, real-time processing of large-scale data, ensuring data accuracy, and meeting the real-time monitoring requirements of long-distance and high-density optical fiber sensing. Description of the Drawings

[0022] Figure 1 is a schematic diagram of an optical frequency domain distributed optical fiber strain / temperature sensing system for a serial-parallel processing method of optical frequency domain distributed optical fiber strain / temperature sensing signals based on a graphics processing unit;

[0023] In the figure: 101. Optical frequency domain distributed optical fiber signal acquisition device, 102. Computer, 103. Sensing optical fiber.

[0024] Figure 2It is a schematic diagram of reference signal processing for the serial-parallel processing method of optical frequency domain distributed fiber strain / temperature sensing signals based on a graphics processing unit;

[0025] In the figure: 201. Spatial resolution of distributed fiber strain / temperature sensing; 202. Starting position of fiber analysis for distributed fiber strain / temperature sensing; 203. Ending position of fiber analysis for distributed fiber strain / temperature sensing; 204. Number of window points for short-time Fourier transform; 205. Number of windows; 206. Serial number of the starting data point for analysis in the optical frequency domain; 207. Serial number of the ending data point for analysis in the optical frequency domain; 208. Time-domain sensing signal; 209. Optical frequency domain sensing signal in the reference state; 210. M reference data segments connected end to end; 211. M sensing units; 212. First reference data segment; 213. Second reference data segment; 214. Set of reference spectral segments.

[0026] Figure 3 It is a schematic diagram of measurement signal processing for the serial-parallel processing method of optical frequency domain distributed fiber strain / temperature sensing signals based on a graphics processing unit;

[0027] In the figure: 301. Time-domain sensing signal; 302. Optical frequency domain sensing signal in the measurement state; 303. Measurement data segment; 304. N measurement data segments connected end to end; 305. Length of the measurement data segment; 306. First measurement data segment; 307. Second measurement data segment; 308. Set of measurement spectral segments; 309. Sequence of spectral offset point numbers; 310. Sequence of spectral wavelength change amounts; 311. Strain / temperature sequence. Detailed implementation manners

[0028] The following further describes the detailed implementation manners of the present invention in conjunction with the accompanying drawings:

[0029] A serial-parallel processing method for optical frequency domain distributed fiber strain / temperature sensing signals based on a graphics processing unit. The optical frequency domain distributed fiber strain / temperature sensing system includes an optical frequency domain distributed fiber signal acquisition device 101, a computer 102, and a sensing optical fiber 103. Among them, the optical frequency domain distributed fiber signal acquisition device 101 is connected to the sensing optical fiber 103 to form an optical path, and the optical frequency domain distributed fiber signal acquisition device 101 is connected to the computer 102 through a data line to form data communication. The computer 102 can use communication instructions through the data line to control the working state of the optical frequency domain distributed fiber signal acquisition device 101 and obtain the sensing data of the optical frequency domain distributed fiber signal acquisition device 101. The strain / temperature sensing information of the sensing optical fiber 103 is obtained by the following steps:

[0030] Step 1: Set the spatial resolution 201 of distributed optical fiber strain / temperature sensing, the starting position 202 of optical fiber analysis for distributed optical fiber strain / temperature sensing, and the ending position 203 of optical fiber analysis for distributed optical fiber strain / temperature sensing on the central processing unit of the computer 102. Calculate the number of window points 204 and the number of windows 205 of the short-time Fourier transform corresponding to the spatial resolution 201 according to the optical frequency domain distributed optical fiber sensing theory, the serial number 206 of the starting data point for analysis in the optical frequency domain corresponding to the starting position 202 of optical fiber analysis, and the serial number 207 of the ending data point for analysis in the optical frequency domain corresponding to the ending position 203 of optical fiber analysis;

[0031] Step 2: Before strain or temperature acts on the sensing optical fiber 103, the computer 102 controls the distributed optical fiber signal acquisition device 101 to acquire the time-domain sensing signal 208 of the sensing optical fiber 103 and store it in the memory of the central processing unit of the computer 102. Using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit, transfer the time-domain sensing signal 208 in the central processing unit memory to the video memory of the graphics processing unit and perform a fast forward Fourier transform of the cufftPlan1d of the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit in the graphics processing unit to obtain the optical frequency domain sensing signal 209 in the reference state. Use the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit to construct a parallel computing kernel function to complete the following data processing. Starting from the serial number 206 of the starting data point for analysis in the optical frequency domain, continuously select M reference data segments 210 connected end to end in the optical frequency domain sensing signal 209 in the reference state, corresponding to M sensing units 211. Each segment of data in the reference data segment 210 is defined as Ref1, Ref2... Ref M , and the number of data points in each segment of data Ref1 to Ref M in the reference data segment 210 is the number of window points 204. Add Z zeros at the end of each segment of data Ref1 to Ref M respectively, and then perform a multi-segment parallel inverse Fourier transform processing of the cufftPlanMany of the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit in the graphics processing unit to obtain the reference data segment one 212, that is, Ref_I1 to Ref_I M , and then use the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit to construct a parallel computing kernel function, and take the modulus of each data point in the reference data segment one 212, that is, Ref_I1 to Ref_I M to obtain the reference data segment two 213, that is, Ref_IM1 to Ref_IM M , and calculate the average value Mr1 to Mr of the data points in the reference data segment two 213, that is, Ref_IM1 to Ref_IM M using the parallel broadcast reduction method; M, a parallel computing kernel function is constructed using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit to subtract the average values Mr1 to Mr of the corresponding data segments from each data point in the reference data segment two 213, i.e., Ref_IM1 to Ref_IM M The result of subtracting the average values Mr1 to Mr of the corresponding data segments from each data point in the reference data segment two 213, i.e., Ref_IM1 to Ref_IM M is the reference spectral segment set 214 of the corresponding sensing unit, i.e., Ref_S1 to Ref_S M ;

[0032] Step 3: After strain or temperature acts on the sensing optical fiber 103, the computer 102 controls the distributed optical fiber signal acquisition device 101 to acquire the time-domain sensing signal 301 of the sensing optical fiber 103 and stores it in the central processing unit memory of the computer 102. Using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit, the time-domain sensing signal 301 in the central processing unit memory is transmitted to the video memory of the graphics processing unit, and the optical frequency-domain sensing signal 302 in the measurement state is obtained by performing a fast Fourier transform using the cufftPlan1d of the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit in the graphics processing unit. The cumulative optical path offset 212 is initialized to 0 in the graphics processing unit, and the sensing parallel computing quantity N = N1;

[0033] Step 4: In the graphics processing unit, starting from the starting data point serial number 206 in the optical frequency domain of the optical frequency-domain sensing signal 302 in the measurement state, a parallel computing kernel function is constructed using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit to implement the following data processing. Continuously select N measurement data segments (303) connected end to end corresponding to N sensing units 304. Each segment of data in the measurement data segment 303 is defined as Mea1, Mea2... Mea N , and the number of data points in each segment of data Mea1 to Mea N in the measurement data segment 303 is the window point number 204. The measurement data segment length 305 of the measurement data segment 303 in the optical frequency domain is N multiplied by the window point number 204. Z zeros are filled at the end of each segment of data Mea1 to Mea N in the measurement data segment 303 respectively. The measurement data segment one 306, i.e., Mea_I1 to Mea_I, is obtained by performing a multi-segment parallel fast inverse Fourier transform processing using the cufftPlanMany of the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit in the graphics processing unit. Then, a parallel computing kernel function is constructed using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit to take the modulus of each data point in the measurement data segment one 306, i.e., Mea1 to Mea N , to obtain the measurement data segment two 307, i.e., Mea_IM1 to Mea_IM N , and the measurement data segment two 307, i.e., Mea_IM1 to Mea_IM N is calculated using the parallel broadcast reduction method NThe average values Mm1 to Mm of the internal data points N , using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit to construct a parallel computing kernel function to subtract each data point in the measurement data segment two 307, namely Mea_IM1 to Mea_IM N from the corresponding average values Mm1 to Mm of the data segments N , and the result is the measurement spectral segment set 308 of the corresponding sensing units, namely Mea_S1 to Mea_S N ;

[0034] Step 5: Use the CUDA platform of the graphics processing unit to construct a cross-correlation data processing function based on cufftPlanMany for multi-segment parallel fast forward and inverse Fourier transforms in the graphics processing unit, and perform parallel computing on the measurement spectral segment sets 308 of N sensing units 304, namely Mea_S1 to Mea_S N and the cross-correlation with the reference spectra of the N sensing units 304 in the corresponding reference spectral segment set 214 to obtain the spectral offset point number sequence 309, namely Bias1 to Bias N , and then use the CUDA platform of the graphics processing unit to construct a parallel computing and judgment kernel function to perform the following data processing. According to the optical frequency domain distributed optical fiber theory, convert the spectral offset point number sequence 309 into the spectral wavelength change amount sequence 310 of the corresponding sensing units, namely Wave1 to Wave N and the strain / temperature sequence 311 of the corresponding sensing units, namely Sen1 to Sen N , and calculate the average value of the strain / temperature sequence 311 of the corresponding sensing units, namely Sen1 to Sen N ; Calculate the number of optical path change points caused by the N sensing units 304 under the action of strain / temperature where W c is the swept center wavelength of the distributed optical fiber signal acquisition device 101. The termination position point numbers 312 of the N sensing units 304 are the sum of the starting data point number 206 in the optical frequency domain analysis, the number of optical path change points L, and the measurement data segment length 305. If the average value of the strain / temperature sequence 310 is greater than the threshold T, then the sensing parallel computing quantity N = N1. If the average value of the strain / temperature sequence 310 is less than the threshold T, then the sensing parallel computing quantity N = N2, and N1 < N2;

[0035] Step 6: In the graphics processing unit, starting from the termination position point number 312 in the optical frequency domain sensing signal 302 in the measurement state, use the unified computing device architecture CUDA platform of the graphics processing unit to construct a parallel computing kernel function to implement the following data processing. Continuously select N measurement data segments 303 connected end to end corresponding to N sensing units 304. Each segment of data in the measurement data segment 303 is defined as Mea1, Mea2... Mea N , and for each segment of data Mea1 to Mea N in the measurement data segment 303, the number of data points is the window point number 204. The measurement data segment length 305 of the measurement data segment 303 in the optical frequency domain is N multiplied by the window point number 204. For each segment of data Mea1 to Mea N in the reference measurement data segment 303, append Z zeros at the end. In the graphics processing unit, use the cufftPlanMany multi-segment parallel fast Fourier transform processing of the unified computing device architecture CUDA platform of the graphics processing unit to obtain the measurement data segment one 306, that is, Mea_I1 to Mea_I N , and then use the unified computing device architecture CUDA platform of the graphics processing unit to construct a parallel computing kernel function. Take the modulus of each data point in the measurement data segment one 306, that is, Mea1 to Mea N to obtain the measurement data segment two 307, that is, Mea_IM1 to Mea_IM N , and use the parallel broadcast reduction method to calculate the average value Mm1 to Mm N of the data points in the measurement data segment two 307, that is, Mea_IM1 to Mea_IM N . Use the unified computing device architecture CUDA platform of the graphics processing unit to construct a parallel computing kernel function to subtract the average value Mm1 to Mm N corresponding to each data point in the measurement data segment two 307, that is, Mea_IM1 to Mea_IM N . The result is the measurement spectrum segment set 308 of the corresponding sensing unit, that is, Mea_S1 to Mea_S N ;

[0036] Step 7: In the graphics processing unit, use the unified computing device architecture CUDA platform of the graphics processing unit to construct a cross-correlation data processing function based on the cufftPlanMany multi-segment parallel fast forward and inverse Fourier transforms, and parallelly calculate the cross-correlation between the measurement spectrum segment set 308 of N sensing units 304, that is, Mea_S1 to Mea_S N and the reference spectra of the N sensing units 304 in the corresponding reference spectrum segment set 213 to obtain the spectral offset point number sequence 309, that is, Bias1 to Bias N, and then use the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit to build parallel computing and judgment kernel functions for the following data processing. According to the theory of optical frequency domain distributed optical fiber, convert the spectral offset point sequence 309 into the spectral wavelength change amount sequence 310 of the corresponding sensing units, namely Wave1 to Wave N and the strain / temperature sequence 311 of the corresponding sensing units, namely Sen1 to Sen N , and calculate the average value of the strain / temperature sequence 311 of the corresponding sensing units, namely Sen1 to Sen N Calculate the number of optical path change points caused by the strain / temperature of N sensing units 304 where W c is the swept center wavelength of the distributed optical fiber signal acquisition device 101. The termination position point number 312 of N sensing units 304 is the sum of the termination position point number 312 in the sixth step, the number of optical path change points L, and the measurement data segment length 305 in the sixth step. If the average value of the strain / temperature sequence 310 is greater than the threshold T, then the sensing parallel computing quantity N = N1. If the average value of the strain / temperature sequence 310 is less than the threshold T, then the sensing parallel computing quantity N = N2, where N1 < N2;

[0037] Step 8: Repeat steps 6 and 7 sequentially until the strain / temperature of all M sensing units is obtained, that is, the strain / temperature sensing information of the sensing optical fiber 103 is obtained.

[0038] The two values of the sensing parallel computing quantity N are that the value range of N1 is 10 to 32, the value range of N2 is 32 to 128, and the value range of the threshold T is 1000 με to 9000 με or 100 °C to 900 °C.

[0039] The working process of the present invention is as follows:

[0040] The optical frequency domain distributed optical fiber signal acquisition device 101 is connected to the sensing optical fiber 103 to form an optical path. The optical frequency domain distributed optical fiber signal acquisition device 101 and the computer 102 are connected through a data line to form data communication. The computer 102 can control the working state of the optical frequency domain distributed optical fiber signal acquisition device 101 through the data line using communication instructions and obtain the sensing data of the optical frequency domain distributed optical fiber signal acquisition device 101. The strain / temperature sensing information of the sensing optical fiber 103 is obtained by the following steps:

[0041] ​Step 1: Set the spatial resolution 201 of distributed optical fiber strain / temperature sensing, the starting position 202 of optical fiber analysis for distributed optical fiber strain / temperature sensing, and the ending position 203 of optical fiber analysis for distributed optical fiber strain / temperature sensing on the central processor of the computer 102. Calculate the number of window points 204 and the number of windows 205 of the short-time Fourier transform corresponding to the spatial resolution 201, the serial number 206 of the starting data point for analysis in the optical frequency domain corresponding to the starting position 202 of optical fiber analysis, and the serial number 207 of the ending data point for analysis in the optical frequency domain corresponding to the ending position 203 of optical fiber analysis according to the theory of optical frequency domain distributed optical fiber sensing;

[0042] Step 2: The computer 102 processes the time-domain sensing signal 208 through the fast forward Fourier transform of cufftPlan1d on the CUDA platform to obtain the optical frequency domain sensing signal 209 in the reference state. Use the CUDA platform to construct a parallel computing kernel function to complete the following data processing: Starting from the serial number 206 of the starting data point for analysis in the optical frequency domain, continuously select M reference data segments 210 connected end to end in the optical frequency domain sensing signal 209 in the reference state, corresponding to M sensing units 211. Each segment of data is defined as Ref1, Ref2... Ref M . For each segment of data Ref1 to Ref in the reference data segment 210 M , respectively append Z zeros at the end, and then use the multi-segment parallel inverse Fourier transform of cufftPlanMany on the CUDA platform to obtain the reference data segment one 212. Construct a kernel function to take the modulus of each data point in the reference data segment one 212 to obtain the reference data segment two 213, and use the parallel broadcast reduction method to calculate the average values Mr1 to Mr of each segment of data points M . Each data point in the reference data segment two 213 is respectively subtracted from the corresponding data segment average values Mr1 to Mr M to obtain the set of reference spectral segments 214 corresponding to the sensing units.

[0043] Step 3: After strain or temperature acts on the sensing optical fiber 103, the computer 102 controls the distributed optical fiber signal acquisition device 101 to collect the time-domain sensing signal 301 of the sensing optical fiber 103 and upload it to the computer 102. The computer 102 processes the time-domain sensing signal 301 through the fast forward Fourier transform of cufftPlan1d on the CUDA platform to obtain the optical frequency domain sensing signal 302 in the measurement state, initialize the cumulative optical path offset 212 to 0, and the parallel computing quantity N of sensing = N1;

[0044] Step 4: Starting from the starting data point serial number 206 in the optical frequency domain within the optical frequency domain sensing signal 302 in the measurement state, construct a parallel computing kernel function on the CUDA platform to perform the following data processing. Continuously select N measurement data segments 303 connected end to end corresponding to N sensing units 304. Each segment of data in the measurement data segment 303 is defined as Mea1, Mea2... Mea N , and the number of data points in each segment of data Mea1 to Mea N within the measurement data segment 303 is the window point number 204. The measurement data segment length 305 of the measurement data segment 303 in the optical frequency domain is N multiplied by the window point number 204. For each segment of data Mea1 to Mea N in the measurement data segment 303, append Z zeros at the end respectively, and then use the cufftPlanMany multi-segment parallel fast Fourier transform processing on the CUDA platform to obtain the measurement data segment one 306, namely Mea_I1 to Mea_I N . Construct a parallel computing kernel function to calculate the modulus of each data point in the measurement data segment one 306 to obtain the measurement data segment two 307. Use the parallel broadcast reduction method to calculate the average values Mm1 to Mm N of the data points within the measurement data segment two 307. Then, the result of subtracting the average value of the corresponding data segment from each data point within the measurement data segment two 307 is the measurement spectrum segment set 308 of the corresponding sensing unit;

[0045] Step 5: Utilize the CUDA platform of the unified computing device architecture of the graphics processing unit in the graphics processing unit to construct the cross-correlation based on the cufftPlanMany multi-segment parallel fast forward and inverse Fourier transforms, and parallel compute the cross-correlation between the measurement spectrum segment sets 308 of N sensing units 304 and the reference spectra of the N sensing units 304 in the corresponding reference spectrum segment set 214 to obtain the spectral offset point number sequence 309. Furthermore, according to the optical frequency domain distributed optical fiber theory, convert the spectral offset point number sequence 309 into the spectral wavelength change amount sequence 310 of the corresponding sensing unit and the strain / temperature sequence 311 of the corresponding sensing unit, namely Sen1 to Sen N , and calculate the average value of the strain / temperature sequence 311 of the corresponding sensing unit, namely Sen1 to Sen N ; Calculate the number of optical path change points caused by the strain / temperature of N sensing units 304 where W c is the swept center wavelength of the distributed optical fiber signal acquisition device 101. The termination position point serial number 312 of N sensing units 304 is the sum of the starting data point serial number 206 in the optical frequency domain analysis, the optical path change point number L, and the measurement data segment length 305. If the average value of the strain / temperature sequence 310 If it is greater than the threshold value T, the number of sensing parallel computations N = N1. If the average value of the strain / temperature sequence 310 is less than the threshold value T, the number of sensing parallel computations N = N2, where N1 < N2;

[0046] Step 6: Starting from the starting data point serial number 206 in the optical frequency domain within the optical frequency domain sensing signal 302 in the measurement state, construct a parallel computing kernel function on the CUDA platform to perform the following data processing. Continuously select N measurement data segments 303 connected end to end corresponding to N sensing units 304. Each segment of data in the measurement data segment 303 is respectively defined as Mea1, Mea2... Mea N , and the number of data points in each segment of data Mea1 to Mea N in the measurement data segment 303 is the window point number 204. The length 305 of the measurement data segment in the optical frequency domain of the measurement data segment 303 is N multiplied by the window point number 204. For each segment of data Mea1 to Mea N in the measurement data segment 303, Z zeros are respectively filled at the end, and then the CUDA large segment data fast Fourier transform is performed to obtain the measurement data segment one 306, that is, Mea_I1 to Mea_I N , and for each data point in the measurement data segment one 306, the modulo calculation is obtained by using thread broadcast reduction to obtain the measurement data segment two 307. The average values Mm1 to Mm N of the data points in the measurement data segment two 307 are respectively calculated by using thread broadcast reduction. Then, the result of subtracting the average value of the corresponding data segment from each data point in the measurement data segment two 307 is the measurement spectrum segment set 308 of the corresponding sensing unit;

[0047] Step 7: The cross-correlation between the measurement spectrum segment set 308 of N sensing units 304 and the reference spectra of the N sensing units 304 in the corresponding reference spectrum segment set 213 obtains the spectral offset point number sequence 309. Furthermore, according to the theory of the optical frequency domain distributed optical fiber, the spectral offset point number sequence 309 is converted into the spectral wavelength change amount sequence 310 of the corresponding sensing unit and the strain / temperature sequence 311 of the corresponding sensing unit, that is, Sen1 to Sen N , and the average value of the strain / temperature sequence 311 of the corresponding sensing unit, that is, Sen1 to Sen N ; Calculate the number of optical path change points caused by the strain / temperature action of N sensing units 304 where W c is the swept center wavelength of the distributed optical fiber signal acquisition device 101. The termination position point serial number 312 of N sensing units 304 is the sum of the termination position point serial number 312 in the previous step, the optical path change point number L, and the measurement data segment length 305 in the previous step. If the average value of the strain / temperature sequence 310 If it is greater than the threshold T, the number of parallel sensing computations N = N1. If the average value of the strain / temperature sequence 310 is less than the threshold T, the number of parallel sensing computations N = N2, where N1 < N2;

[0048] Step 8: Repeat Steps 6 and 7 sequentially until the strain / temperature of all M sensing units are obtained, i.e., the strain / temperature sensing information of the sensing optical fiber 103 is obtained.

[0049] The technical innovation points and beneficial technical effects of the serial-parallel processing method for optical frequency domain distributed fiber optic strain / temperature sensing signals based on a graphics processing unit are as follows: The signal processing method of the present invention is based on a serial-parallel computing architecture, making full use of the multi-core and multi-thread characteristics of the central processing unit and the graphics processing unit, and significantly improving the processing speed compared with traditional methods. During the signal processing, the conventional processing method is to demodulate the signals of each sensing unit one by one, which is slow and time-consuming and resource-consuming. However, the present invention realizes the efficient utilization of computing resources by dynamically adjusting the number of sensing units for parallel computing, and realizes the parallel acceleration of demodulation on the basis of ensuring accurate optical path compensation. In addition, the present invention uses the CPU and GPU to cooperate to complete the signal processing process, constructs a parallel kernel function using the unified computing device architecture CUDA platform of the graphics processing unit, and reasonably allocates the computing power resources of the GPU, which can significantly accelerate the speed of parallel signal processing of a large number of sensing units and improve the frequency response of the sensing system. In practical applications, this method can achieve high-precision and real-time monitoring of the strain and temperature along the optical fiber, meeting the requirements of long-distance and high-density distributed fiber optic sensing.

Claims

1. A serial-parallel processing method for optical frequency domain distributed fiber strain / temperature sensing signals based on a graphics processing unit. The optical frequency domain distributed fiber strain / temperature sensing system includes an optical frequency domain distributed fiber signal acquisition device (101), a computer (102), and a sensing optical fiber (103), wherein, The optical frequency domain distributed optical fiber signal acquisition device (101) is connected to the sensing optical fiber (103) to form an optical path. The optical frequency domain distributed optical fiber signal acquisition device (101) is connected to the computer (102) through a data line to form data communication. The computer (102) can use communication instructions through the data line to control the working state of the optical frequency domain distributed optical fiber signal acquisition device (101) and obtain the sensing data of the optical frequency domain distributed optical fiber signal acquisition device (101). The strain / temperature sensing information of the sensing optical fiber (103) is obtained by the following steps: Step 1: Set the spatial resolution (201) of distributed optical fiber strain / temperature sensing, the starting position (202) of optical fiber analysis for distributed optical fiber strain / temperature sensing, and the ending position (203) of optical fiber analysis for distributed optical fiber strain / temperature sensing on the central processing unit of the computer (102). Then, according to the optical frequency domain distributed optical fiber sensing theory, calculate the number of window points (204) and the number of windows (205) of the short-time Fourier transform corresponding to the spatial resolution (201), the serial number (206) of the starting data point for analysis in the optical frequency domain corresponding to the starting position (202) of optical fiber analysis, and the serial number (207) of the ending data point for analysis in the optical frequency domain corresponding to the ending position (203) of optical fiber analysis; Step 2: Before strain or temperature acts on the sensing optical fiber (103), the computer (102) controls the distributed optical fiber signal acquisition device (101) to acquire the time-domain sensing signal (208) of the sensing optical fiber (103) and stores it in the central processing unit memory of the computer (102). Using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit, the time-domain sensing signal (208) in the central processing unit memory is transmitted to the video memory of the graphics processing unit, and the optical frequency-domain sensing signal (209) in the reference state is obtained by performing a fast Fourier transform (FFT) using the cufftPlan1d function of the CUDA platform in the graphics processing unit. Using the CUDA platform to build a parallel computing kernel function to complete the following data processing. Starting from the starting data point sequence number (206) in the optical frequency domain analysis, M consecutive reference data segments (210) are selected in the optical frequency-domain sensing signal (209) in the reference state, corresponding to M sensing units (211). Each segment of data in the reference data segment (210) is defined as Ref1, Ref2... Ref M , and each segment of data Ref1 to Ref M in the reference data segment (210) has the number of data points as the window points (204). For each segment of data Ref1 to Ref M in the reference data segment (210), Z zeros are padded at the end, and then the multi-segment parallel inverse FFT processing using the cufftPlanMany function of the CUDA platform in the graphics processing unit is performed to obtain the reference data segment one (212), namely Ref_I1 to Ref_I M . Then, using the CUDA platform to build a parallel computing kernel function, for each data point in the reference data segment one (212), namely Ref_I1 to Ref_I M , after taking the modulus, the reference data segment two (213) is obtained, namely Ref_IM1 to Ref_IM M . And the parallel broadcast reduction method is used to calculate the average values Mr1 to Mr M of the data points in the reference data segment two (213), namely Ref_IM1 to Ref_IM M . Using the CUDA platform to build a parallel computing kernel function, for each data point in the reference data segment two (213), namely Ref_IM1 to Ref_IM M , subtracting the corresponding data segment average values Mr1 to Mr M results in the reference spectral segment set (214) of the corresponding sensing unit, namely Ref_S1 to Ref_S M ; Step 3: After strain or temperature acts on the sensing optical fiber (103), the computer (102) controls the distributed optical fiber signal acquisition device (101) to acquire the time-domain sensing signal (301) of the sensing optical fiber (103) and stores it in the memory of the central processing unit of the computer (102). Using the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit, the time-domain sensing signal (301) in the central processing unit memory is transmitted to the video memory of the graphics processing unit, and the optical frequency-domain sensing signal (302) in the measurement state is obtained by performing a fast Fourier transform of the cufftPlan1d using the CUDA platform of the graphics processing unit in the graphics processing unit. Initialize the cumulative optical path offset (212) to 0 in the graphics processing unit, and the number of sensing parallel computations N = N1; Step 4. In the graphics processing unit, starting from the starting data point number (206) analyzed in the optical frequency domain of the optical frequency domain sensing signal (302) in the measurement state, use the unified computing device architecture CUDA platform of the graphics processing unit to construct a parallel computing kernel function to implement the following data processing. Continuously select N measurement data segments (303) connected end to end corresponding to N sensing units (304). Each segment of data in the measurement data segment (303) is defined as Mea1, Mea2... Mea N , and the number of data points in each segment of data Mea1 to Mea N in the measurement data segment (303) is the window point number (204). The length (305) of the measurement data segment in the optical frequency domain of the measurement data segment (303) is N multiplied by the window point number (204). For each segment of data Mea1 to Mea N in the measurement data segment (303), append Z zeros at the end. In the graphics processing unit, use the cufftPlanMany multi-segment parallel fast Fourier transform processing of the unified computing device architecture CUDA platform of the graphics processing unit to obtain the first measurement data segment (306), that is, Mea_I1 to Mea_I N , and then use the unified computing device architecture CUDA platform of the graphics processing unit to construct a parallel computing kernel function. Take the modulus of each data point in the first measurement data segment (306), that is, Mea1 to Mea N to obtain the second measurement data segment (307), that is, Mea_IM1 to Mea_IM N , and use the parallel broadcast reduction method to calculate the average values Mm1 to Mm of the data points in the second measurement data segment (307), that is, Mea_IM1 to Mea_IM N . Use the unified computing device architecture CUDA platform of the graphics processing unit to construct a parallel computing kernel function to subtract the average values Mm1 to Mm of the corresponding data segments from each data point in the second measurement data segment (307), that is, Mea_IM1 to Mea_IM N . The result is the set of measurement spectral segments (308) of the corresponding sensing units, that is, Mea_S1 to Mea_S N ; N ; N ; Step 5: In the graphics processing unit, use the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit to construct a cross-correlation data processing function based on cufftPlanMany multi-segment parallel fast forward and inverse Fourier transforms, and parallelly calculate the measurement spectral segment sets (308) of N sensing units (304), i.e., Mea_S1 to Mea_S N The cross-correlation of the reference spectra of N sensing units (304) in the corresponding reference spectral segment sets (214) to obtain the spectral offset point number sequence (309), i.e., Bias1 to Bias N , and then use the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit to construct a parallel computing and judgment kernel function for the following data processing. According to the optical frequency domain distributed optical fiber theory, convert the spectral offset point number sequence (309) into the spectral wavelength change amount sequence (310) of the corresponding sensing units, i.e., Wave1 to Wave N and the strain / temperature sequence (311) of the corresponding sensing units, i.e., Sen1 to Sen N , and calculate the average value of the strain / temperature sequence (311) of the corresponding sensing units, i.e., Sen1 to Sen N Calculate the number of optical path change points caused by N sensing units (304) under the action of strain / temperature where W c is the swept center wavelength of the distributed optical fiber signal acquisition device (101). The termination position point numbers (312) of N sensing units (304) are the sum of the starting data point number (206) for analysis in the optical frequency domain, the number of optical path change points L, and the measurement data segment length (305). If the average value of the strain / temperature sequence (310) is greater than the threshold T, then the sensing parallel computing quantity N = N1. If the average value of the strain / temperature sequence (310) is less than the threshold T, then the sensing parallel computing quantity N = N2, where N1 < N2;​ Step 6: In the graphics processing unit, starting from the termination position point serial number (312) within the optical frequency domain sensing signal (302) in the measurement state, use the unified computing device architecture CUDA platform of the graphics processing unit to construct a parallel computing kernel function to implement the following data processing. Continuously select N measurement data segments (303) connected end to end corresponding to N sensing units (304). Each segment of data in the measurement data segment (303) is defined as Mea1, Mea2... Mea N , and each segment of data Mea1 to Mea N in the measurement data segment (303) has the number of data points as the window points (204). The length (305) of the measurement data segment (303) in the optical frequency domain is N multiplied by the window points (204). For each segment of data Mea1 to Mea N in the reference measurement data segment (303), append Z zeros at the end. In the graphics processing unit, use the cufftPlanMany multi-segment parallel fast Fourier transform processing of the unified computing device architecture CUDA platform of the graphics processing unit to obtain the first measurement data segment (306), namely Mea_I1 to Mea_I N , and then use the unified computing device architecture CUDA platform of the graphics processing unit to construct a parallel computing kernel function. Take the modulus of each data point in the first measurement data segment (306), namely Mea1 to Mea N to obtain the second measurement data segment (307), namely Mea_IM1 to Mea_IM N , and use the parallel broadcast reduction method to calculate the average values Mm1 to Mm N of the data points within the second measurement data segment (307), namely Mea_IM1 to Mea_IM N . Use the unified computing device architecture CUDA platform of the graphics processing unit to construct a parallel computing kernel function to subtract the average values Mm1 to Mm N corresponding to each data point within the second measurement data segment (307), namely Mea_IM1 to Mea_IM N . The result is the set of measurement spectral segments (308) of the corresponding sensing units, namely Mea_S1 to Mea_S N ; Step 7: In the graphics processing unit, use the Compute Unified Device Architecture (CUDA) platform of the graphics processing unit to construct a cross-correlation data processing function based on cufftPlanMany multi-segment parallel fast forward and inverse Fourier transforms, and perform parallel computing on the measurement spectral segment sets (308) of N sensing units (304), i.e., Mea_S1 to Mea_S N to obtain the spectral offset point sequence (309), i.e., Bias1 to Bias, of the cross-correlation between the reference spectra of the N sensing units (304) in the corresponding reference spectral segment set (213). N Furthermore, use the CUDA platform of the graphics processing unit to construct a parallel computing and judgment kernel function for the following data processing. According to the theory of optically frequency-domain distributed optical fiber, convert the spectral offset point sequence (309) into the spectral wavelength change amount sequence (310) of the corresponding sensing units, i.e., Wave1 to Wave N and the strain / temperature sequence (311) of the corresponding sensing units, i.e., Sen1 to Sen N , and calculate the average value Sen of the strain / temperature sequence (311) of the corresponding sensing units, i.e., Sen1 to Sen N , and calculate the number of optical path change points caused by the strain / temperature of the N sensing units (304). Where W c is the swept center wavelength of the distributed optical fiber signal acquisition device (101). The termination position point numbers (312) of the N sensing units (304) are the sum of the termination position point numbers (312) in Step 6, the optical path change point number L, and the measurement data segment length (305) in Step 6. If the average value of the strain / temperature sequence (310) is greater than the threshold T, then the sensing parallel computing quantity N = N1. If the average value of the strain / temperature sequence (310) is less than the threshold T, then the sensing parallel computing quantity N = N2, where N1 < N2; Step 8: Repeat steps 6 and 7 sequentially until the strain / temperature of all M sensing units are obtained, that is, the strain / temperature sensing information of the sensing optical fiber (103) is obtained.

2. The method for serial-parallel processing of optical frequency domain distributed optical fiber strain / temperature sensing signals based on a graphics processing unit according to claim 1, characterized in that: The two values of the number N of parallel sensing calculations are as follows: the value range of N1 is 10 to 32, the value range of N2 is 32 to 128, and the value range of the threshold T is 1000 με to 9000 με or 100 °C to 900 °C.

Images