Distributed optical fiber sensing strain measurement method without pre-training

Abstract

The invention relates to the technical field of optical fiber sensing and detection, in particular to a non-pre-training distributed optical fiber sensing strain measurement method, which comprises the following steps of: acquiring a reference signal and a strain test signal, reconstructing a two-dimensional image through FFT (Fast Fourier Transform) mapping and cross-correlation calculation, skipping an independent training set construction link, and obtaining a strain measurement result; directly taking a two-dimensional image generated by the measurement as a training data source; a sliding window mechanism is adopted for cyclic prediction until the whole optical fiber is covered, and millimeter-level anomaly positioning is achieved; and meanwhile, an optimal solution algorithm is used for dynamically optimizing the number of model units and the Dropout ratio. According to the method, the dependence of a traditional scheme on a simulation training set is eliminated, the hyper-parameter optimization speed is increased, and the real-time processing requirement of the OFDR signal can be met.

Description

Technical Field

[0001] The present invention relates to the technical field of optical fiber sensing and detection, and particularly to a distributed optical fiber sensing strain measurement method without pre-training. Background Art

[0002] Distributed optical fiber sensing technology can realize long-distance continuous monitoring of external physical quantities such as temperature, strain, and vibration by utilizing the scattering effects of light waves in optical fibers (such as Rayleigh scattering, Brillouin scattering, etc.). As a representative of this technology, optical frequency domain reflectometry (OFDR) has been widely used in fields such as bridge health monitoring, oil and gas pipeline safety, and aerospace structure detection due to its advantages of millimeter-level high spatial resolution, high sensitivity, and anti-electromagnetic interference.

[0003] Patent CN117629093A discloses a method for improving the spatial resolution of an OFDR shape sensing system based on CNN, which has the following problems: 1. Strong dependence on the training set, a large amount of noise-free simulation data needs to be pre-generated to train the model, which requires additional equipment and time costs in actual operation and cannot respond in real time; 2. Lack of local anomaly detection ability. The CNN outputs an overall denoising result and cannot locate the noise interference at the local position of the optical fiber, resulting in strain measurement distortion at high resolution; 3. Poor generalization. When the optical fiber length or sampling rate changes, the training set needs to be regenerated and the network structure needs to be adjusted, with low adaptability.

[0004] Patent CN118611052A uses a CNN+GCN+LSTM composite structure to predict distributed photovoltaic power and optimizes hyperparameters through the whale optimization algorithm (WOA), which has the following problems: 1. Strong dependence on historical data. To build a model, a large amount of historical data on climate impact factors and historical data on photovoltaic power are required as inputs to predict the true value; 2. Excessive computational complexity. The composite model has a large number of parameters, and the training and inference delays are significant. It is mainly used for power prediction rather than specifically designed for denoising; 3. Low optimization efficiency. The global search convergence speed of the whale optimization algorithm is slow, and the hyperparameter tuning takes a long time.

[0005] Therefore, the present application designs a distributed optical fiber sensing strain measurement method without pre-training to solve the above problems. Summary of the Invention

[0006] The present invention provides a distributed optical fiber sensing strain measurement method without pre-training to make up for the deficiencies in the prior art.

[0007] A distributed optical fiber sensing strain measurement method without pre-training, characterized by comprising the following steps: S1, collect reference signals and test signals; S2, map the signals to the distance domain by FFT, and calculate the cross-correlation results according to window division; S3, reconstruct the cross-correlation results into a two-dimensional image, and denoise the two-dimensional image; S4, skip a large number of data training sets in the early stage during the denoising process, directly use the two-dimensional image in S3 as the training data source, and provide hyperparameters for the time series analysis model through the optimal solution algorithm. The time series analysis model performs cyclic prediction in the local information along the optical fiber to obtain the information of the entire sensing optical fiber; S5, perform hyperparameter optimization through the optimal solution algorithm, and finally output high-resolution strain measurement results.

[0008] Further, to better implement the present invention, the specific steps of the cyclic prediction in S4 are as follows: S41, select the information of the first n points along the optical fiber as the analysis window; S42, check whether each point in the analysis window is affected by noise; S43, predict the information of the (n + 1)-th point based on the analysis results; S44, slide the analysis window backward by 1 point; S45, repeat the loop of S42 - S44 until the entire optical fiber is covered.

[0009] Further, to better implement the present invention, the specific optimal solution algorithm in S5 is as follows: S51, continue to provide several groups of different model unit numbers and Dropout ratio parameters for the time series analysis model through the optimal solution algorithm; S52, the time series analysis model generates several groups of processing results through these several groups of parameters. Taking the mean square error as the evaluation criterion, select the result with the smallest mean square error, which means the best parameters and the best processing results; S53, determine whether the maximum number of iterations is reached. If it is reached, output the result; S54, if the maximum number of iterations is not reached, the optimal solution algorithm will use the current optimal parameters as the basis to generate new parameter combinations in its adjacent area to further approach the global optimum, and then evaluate these new parameters again and continue to optimize; S55, the whole process is continuously iterated until the maximum iteration condition is satisfied, and finally output the optimal parameters and their corresponding processing results as the optimal configuration and output of the model in this task.

[0010] The beneficial effects of the present invention are: 1. The present invention uses the two-dimensional image obtained from a single measurement directly as the training data, avoiding the dependence on traditional data training sets, and can improve efficiency and reduce costs.

[0011] 2. The matrix for cyclic prediction of the present invention can accurately locate the micro-damage points of the optical fiber, and the effectiveness of predicting the next sampling point can avoid the strain misjudgment caused by multi-peaks / false peaks.

[0012] 3. The present invention uses the optimal solution algorithm to optimize the number of model units and the Dropout ratio, which can significantly improve the robustness of the model; by organizing the training data through a sliding window, it can adapt to different optical fiber lengths and sampling rates without redesigning the network structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a flowchart of the present invention; Figure 2 is a flowchart of the traditional method for processing data using a neural network; Figure 3 is a diagram of the spectral shift result of the sensing optical fiber obtained by the traditional processing method; Figure 4 is a diagram of the spectral shift result of the sensing optical fiber obtained by the processing method of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0014] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Usually, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0015] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0016] Figures 1-4 is a specific embodiment of the present invention, and this embodiment is a distributed optical fiber sensing strain measurement method without pre-training.

[0017] Such as Figure 1As shown in the figure, the traditional neural network processing method requires a large amount of data for repeated iterative training of the network. In this processing method, a large amount of cross-correlation result information is used to construct a data training set, and then the test results are further optimized by comparing with the training set. This is also a major feature of the traditional neural network, but the disadvantage is that it consumes GPU resources and the parameter tuning is complex. This embodiment can complete the training only relying on real measurement data, completely eliminating the need for a large amount of data for training, greatly reducing the data preparation complexity, shortening the system development cycle, and enhancing the feasibility and promotion of practical engineering applications. By organizing training samples in a sliding window manner, it can flexibly adapt to changes in different sampling rates, fiber lengths, and data scales, without the need to redesign or retrain the network structure, and has strong input flexibility and environmental adaptability.

[0018] The OFDR signal is high-frequency, continuous, and fine-grained physical quantity change data, and the complex models of traditional methods cannot meet the real-time processing requirements of the OFDR signal. This embodiment can, in view of the characteristics of high-frequency, continuous, fine-grained changes, and long-time series dependence of the OFDR signal, directly model the original data using a time series analysis model, fully retaining the signal evolution trend and dynamic change characteristics, accurately capturing the global evolution process, and effectively making up for the problems of weak local feature extraction and time series modeling ability of traditional methods. Combined with the optimal solution algorithm, it can automatically search and dynamically adjust the optimal parameter combination within the key hyperparameter spaces such as the number of network units and the Dropout rate, avoiding the uncertainty and performance fluctuations caused by traditional manual setting of hyperparameters. At the same time, compared with the characteristics of high computational complexity, slow convergence speed, and long training cycle of traditional models. This embodiment ensures fast convergence and improved robustness in different environments. This embodiment adopts a lightweight model design. The time series analysis model has a simple and compact structure, fast training and inference speeds, and low memory resource occupancy, and can be deployed on ordinary CPUs, small GPUs, and even edge computing devices, significantly reducing the system deployment threshold and improving the convenience and universality of practical applications.

[0019] As Figure 2 shown, the specific method of this embodiment is as follows: S1, collect the reference signal and the test signal; S2, map the signal to the distance domain through FFT, and calculate the cross-correlation result according to window division; S3, reconstruct the cross-correlation result into a two-dimensional image, and denoise the two-dimensional image; S4, skip the large amount of data training set in the early stage during the denoising process, directly use the two-dimensional image in S3 as the training data source, and provide hyperparameters for the time series analysis model through the optimal solution algorithm. The time series analysis model performs cyclic prediction in the local information along the fiber optic line to obtain the information of the entire sensing fiber optic. The specific method of cyclic prediction is: S41. Select the first n point information in the local information along the optical fiber as the analysis window; S42. Analyze whether each point in the analysis window is affected by noise; S43. Predict the (n + 1)-th point information based on the analysis results; S44. Slide the analysis window backward by 1 point; S45. Repeat steps S42 - S44 in a loop until the entire optical fiber is covered.

[0020] S5. Perform hyperparameter optimization through the optimal solution algorithm, and finally output the high-resolution strain measurement result; The specific optimal solution algorithm is as follows: S51. Continuously provide several groups of different model unit numbers and Dropout ratio parameters for the time series analysis model through the optimal solution algorithm; S52. The time series analysis model generates several groups of processing results through these several groups of parameters. Using the mean squared error as the evaluation criterion, select the result with the smallest mean squared error, which means the best parameters and the best processing results; S53. Determine whether the maximum number of iterations is reached. If so, output the result; S54. If the maximum number of iterations is not reached, the optimal solution algorithm will generate a new parameter combination in the vicinity based on the current optimal parameters to further approach the global optimum, and then evaluate these new parameters again to continue the optimization; S55. The entire process is iterated continuously until the maximum iteration condition is met, and finally output the optimal parameters and their corresponding processing results as the optimal configuration and output of the model in this task.

[0021] From Figure 2 It can be seen that compared with the traditional data processing method, this embodiment abandons the disadvantages brought by constructing the training set with a large amount of training data. By relying on the obtained cross-correlation result information, a strategy of combining global and local search is adopted. For example, select the first 20 point information in the local information along the optical fiber to analyze whether the 0 - 20th points are affected by noise, and then predict whether the information of the 21st point is correct, and repeat this process in turn to obtain the information of the entire sensing optical fiber. This processing method greatly improves the processing efficiency, and at the same time, higher spatial resolution strain information measurement can also be obtained using this embodiment.

[0022] In addition, from Figure 1 Figure 2From the flowchart, the traditional neural network processing method focuses more on the overall network structure, relying only on the neural network to process and denoise signals, without reasonably using the data evaluation mechanism to reduce the difficulty of model construction. Therefore, the model construction is complex, while this embodiment can effectively reduce the complexity of the model. In addition, this embodiment can adaptively adjust the parameters of the time series analysis model through the optimal solution algorithm, avoiding manual parameter adjustment, with better prediction accuracy, and can be applied to different noise environments. The traditional neural network processing method, on the other hand, uses fixed parameters, has no self-adaptability, and requires manual parameter adjustment.

[0023] As Figure 3 and Figure 4 From the comparison in, through experiments, at a spatial resolution of 1.5 mm, it is difficult for the traditional processing method to distinguish different strains, which is extremely likely to cause incorrect analysis, while the method of this embodiment can well analyze the strain results.

[0024] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Any other modifications or equivalent replacements made by those of ordinary skill in the art to the technical solutions of the present invention should be covered within the scope of the claims of the present invention as long as they do not depart from the spirit and scope of the technical solutions of the present invention.

Claims

1. A distributed optical fiber sensing strain measurement method without pre-training, characterized in that Including the following steps: S1. Collect reference signals and test signals; S2. The signals are mapped to the range domain by FFT, and the cross-correlation results are calculated by dividing into windows; S3. The cross-correlation results are reconstructed into a two-dimensional image, and the two-dimensional image is denoised; S4. In the denoising process, a large amount of previous data training sets are skipped, and the two-dimensional image of S3 is directly used as the training data source. The optimal solution algorithm provides hyperparameters for the time series analysis model. The time series analysis model performs cyclic prediction in the local information along the optical fiber to obtain the information of the entire sensing optical fiber; S5. Hyperparameter optimization is performed by the optimal solution algorithm, and finally a high-resolution strain measurement result is output.

2. The pre-training-free distributed optical fiber sensing strain measurement method according to claim 1, wherein: The specific steps of the cyclic prediction in S4 are: S41. Select the first n point position information as the analysis window in the local information along the optical fiber; S42. Analyze whether each point in the analysis window is affected by noise; S43. Predict the (n + 1)th point position information based on the analysis result; S44. The analysis window slides backward by 1 point; S45. Repeat S42 - S44 in a loop until the entire optical fiber is covered.

3. The pre-training-free distributed optical fiber sensing strain measurement method according to claim 2, wherein: The specific method of S42 is: The model automatically models its internal pattern by learning the change trend of the n point position information window. Even if several points are jumpy noises, the overall trend can still be captured, so that the generated (n + 1)th point is more "smooth".

4. The pre-training-free distributed optical fiber sensing strain measurement method according to claim 1, wherein: The optimal solution algorithm in S5 is specifically: S51. Continue to provide several groups of different model unit numbers and Dropout ratio parameters for the time series analysis model by the optimal solution algorithm; S52. The time series analysis model generates several groups of processing results through these several groups of parameters. Using the mean square error as the evaluation criterion, select the result with the smallest mean square error, which means the best parameters and the best processing results; S53. Judge whether the maximum number of iterations is reached. If so, output the result; S54. If the maximum number of iterations is not reached, the optimal solution algorithm will generate a new parameter combination in the adjacent area based on the current optimal parameters to further approximate the global optimum. Then, evaluate these new parameters again and continue to optimize; S55. The whole process is continuously iterated until the maximum iteration condition is met, and finally the optimal parameters and their corresponding processing results are output as the optimal configuration and output of the model in this task.

Images