Fiber bragg grating sensing system based on deep learning and noise reduction method

Abstract

The invention relates to the technical field of sensor noise reduction, and particularly discloses a fiber bragg grating sensing system based on deep learning and a noise reduction method.The method comprises the steps that firstly, original reflection spectrum data of a fiber bragg grating sensor and temperature data of the environment where the fiber bragg grating sensor is located are synchronously collected, and a data processing algorithm based on deep learning is introduced; the method comprises the following steps: performing empirical mode decomposition and waveform feature extraction on original reflectance spectrum data to capture intrinsic intrinsic mode waveform features of the original reflectance spectrum data, then further combining environment temperature data, and performing implicit query interactive response coding on environment temperature features and the intrinsic mode waveform features of the original reflectance spectrum to obtain the intrinsic mode waveform features of the original reflectance spectrum; therefore, the dynamic influence mode of the environment temperature on the reflectance spectrum data can be understood, and the temperature compensation and noise reduction of the reflectance spectrum data can be realized. Through the mode, the influence of temperature fluctuation on the reflection spectrum can be adaptively identified and compensated, the noise caused by environment temperature fluctuation is effectively inhibited, and the measurement precision of the fiber grating sensor is improved.

Description

Technical Field

[0001] The present application relates to the technical field of sensor noise reduction, and more specifically, to a fiber grating sensing system and noise reduction method based on deep learning. Background Art

[0002] Fiber Bragg grating sensing technology has been widely used in aerospace, bridge construction, petrochemical industry, power transmission and many other fields due to its advantages such as anti-electromagnetic interference, small size, high sensitivity and distributed measurement. However, in practical applications, ambient temperature fluctuations will produce cross-sensitivity effects with the physical quantity to be measured (such as strain and pressure), resulting in the shift of the center wavelength of the reflection spectrum and the superposition of noise, which seriously restricts the measurement accuracy of fiber Bragg grating sensors.

[0003] Specifically, a fiber Bragg grating is formed by periodically changing the refractive index within the fiber core. When a broadband light source is irradiated onto the FBG, light of a specific wavelength is reflected back, which is related to the grating period and the effective refractive index. However, temperature changes usually cause thermal expansion or contraction of the fiber material, which in turn changes the grating period and the effective refractive index, causing the reflected wavelength to drift. This wavelength shift caused by temperature is often confused with the wavelength shift caused by the target physical quantity (such as strain), which makes it difficult to measure accurately.

[0004] Traditional noise reduction methods usually rely on hardware compensation (such as reference grating method) or algorithm filtering (such as wavelet threshold denoising) to suppress temperature interference, but there are significant limitations. For example, hardware compensation methods often require additional hardware equipment, which increases the complexity and cost of the system; and wavelet threshold denoising methods often have difficulty distinguishing the nonlinear coupling characteristics of temperature drift noise and real signals when dealing with non-stationary noise or large changes in noise characteristics, resulting in limited noise reduction effects and the possibility of introducing additional signal distortion.

[0005] In recent years, with the development of deep learning technology, a new solution has been provided for the temperature cross-sensitivity problem in fiber Bragg grating sensing systems. Therefore, a fiber Bragg grating sensing system and noise reduction method based on deep learning are expected. Summary of the invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a fiber Bragg grating sensing system and noise reduction method based on deep learning, which first synchronously collects the original reflection spectrum data of the fiber Bragg grating sensor and the ambient temperature data of the fiber Bragg grating sensor, and introduces a data processing algorithm based on deep learning, and performs empirical mode decomposition and waveform feature extraction on the original reflection spectrum data to capture its inherent intrinsic modal waveform characteristics, and then further combines the ambient temperature data, and performs implicit query interactive response encoding on the ambient temperature characteristics and the original reflection spectrum intrinsic modal waveform characteristics to understand the dynamic influence pattern of the ambient temperature on the reflection spectrum data, thereby realizing temperature compensation and noise reduction of the reflection spectrum data. In this way, the influence of temperature fluctuations on the reflection spectrum can be adaptively identified and compensated, the noise caused by ambient temperature fluctuations can be effectively suppressed, and the measurement accuracy of the fiber Bragg grating sensor can be improved.

[0007] According to one aspect of the present application, a noise reduction method for a fiber Bragg grating sensing system based on deep learning is provided, which comprises: The fiber grating sensor demodulator is used to collect the original reflection spectrum data of the fiber grating sensor, and the temperature sensor is used to collect the ambient temperature data of the location where the fiber grating sensor is located; Performing waveform feature extraction based on modal decomposition on the original reflection spectrum data to obtain a set of inherent modal waveform features of the original reflection spectrum; Extracting temperature time series variation characteristics from the ambient temperature data to obtain ambient temperature time series characteristics; Performing temperature compensation deep learning based on spectrum analysis on the set of the ambient temperature time series characteristics and the original reflection spectrum inherent modal waveform characteristics to obtain a temperature compensation reflection spectrum dynamic response coding feature; Based on the temperature-compensated reflection spectrum dynamic response coding characteristics, temperature-compensated reflection spectrum data is generated.

[0008] According to another aspect of the present application, a fiber grating sensing system based on deep learning is provided, comprising: The sensor data acquisition module is used to collect the original reflection spectrum data of the fiber Bragg grating sensor by using the fiber Bragg grating sensor demodulator, and to collect the ambient temperature data of the location where the fiber Bragg grating sensor is located by using the temperature sensor; A waveform feature extraction module, used for performing waveform feature extraction based on modal decomposition on the original reflection spectrum data to obtain a set of inherent modal waveform features of the original reflection spectrum; A time series variation feature extraction module, used for extracting temperature time series variation features from the ambient temperature data to obtain ambient temperature time series features; A temperature compensation deep learning module, used for performing temperature compensation deep learning based on spectrum analysis on the set of the ambient temperature time series characteristics and the inherent modal waveform characteristics of the original reflection spectrum to obtain a temperature compensation reflection spectrum dynamic response coding feature; The compensation data generating module is used to generate the temperature compensated reflection spectrum data based on the temperature compensated reflection spectrum dynamic response coding characteristics.

[0009] Compared with the prior art, the fiber grating sensing system and noise reduction method based on deep learning provided by the present application first synchronously collects the original reflection spectrum data of the fiber grating sensor and the ambient temperature data of the fiber grating sensor, and introduces a data processing algorithm based on deep learning, and performs empirical mode decomposition and waveform feature extraction on the original reflection spectrum data to capture its inherent intrinsic modal waveform characteristics. Then, it further combines the ambient temperature data, and performs implicit query interactive response encoding on the ambient temperature characteristics and the original reflection spectrum intrinsic modal waveform characteristics to understand the dynamic influence mode of the ambient temperature on the reflection spectrum data, thereby realizing temperature compensation and noise reduction of the reflection spectrum data. In this way, the influence of temperature fluctuations on the reflection spectrum can be adaptively identified and compensated, the noise caused by ambient temperature fluctuations can be effectively suppressed, and the measurement accuracy of the fiber grating sensor can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0011] Figure 1 This is a flow chart of a noise reduction method for a fiber Bragg grating sensing system based on deep learning according to an embodiment of the present application.

[0012] Figure 2 Schematic diagram of data flow of a noise reduction method for a fiber Bragg grating sensing system based on deep learning according to an embodiment of the present application.

[0013] Figure 3 This is a flowchart of sub-step S2 of the noise reduction method for a fiber Bragg grating sensing system based on deep learning according to an embodiment of the present application.

[0014] Figure 4 This is a flowchart of sub-step S4 of the noise reduction method for a fiber Bragg grating sensing system based on deep learning according to an embodiment of the present application.

[0015] Figure 5This is a flowchart of sub-step S42 of the noise reduction method for a fiber Bragg grating sensing system based on deep learning according to an embodiment of the present application.

[0016] Figure 6 This is a flowchart of sub-step S43 of the noise reduction method for a fiber Bragg grating sensing system based on deep learning according to an embodiment of the present application.

[0017] Figure 7 is a block diagram of a fiber grating sensing system based on deep learning according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0019] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0020] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0021] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0022] It is worth noting that in this application, all actions to obtain data are carried out in compliance with the relevant data protection laws and policies of the country where the data is located, and with the authorization given by the owner of the corresponding device.

[0023] In response to the technical problems described in the above background technology, this application proposes a noise reduction method for a fiber Bragg grating sensor system based on deep learning, which first synchronously collects the original reflection spectrum data of the fiber Bragg grating sensor and the ambient temperature data of the fiber Bragg grating sensor, and introduces a data processing algorithm based on deep learning, and performs empirical mode decomposition and waveform feature extraction on the original reflection spectrum data to capture its inherent intrinsic modal waveform characteristics. Then, it further combines the ambient temperature data, and performs implicit query interactive response encoding on the ambient temperature characteristics and the original reflection spectrum intrinsic modal waveform characteristics to understand the dynamic influence pattern of the ambient temperature on the reflection spectrum data, thereby achieving temperature compensation and noise reduction of the reflection spectrum data. In this way, the influence of temperature fluctuations on the reflection spectrum can be adaptively identified and compensated, the noise caused by ambient temperature fluctuations can be effectively suppressed, and the measurement accuracy of the fiber Bragg grating sensor can be improved.

[0024] Figure 1 This is a flow chart of a noise reduction method for a fiber Bragg grating sensing system based on deep learning according to an embodiment of the present application. Figure 2 FIG. 1 is a data flow diagram of a noise reduction method for a fiber Bragg grating sensing system based on deep learning according to an embodiment of the present application. Figure 1 and Figure 2 As shown, the noise reduction method of the fiber grating sensing system based on deep learning includes the following steps: S1, using a fiber grating sensor demodulator to collect the original reflection spectrum data of the fiber grating sensor, and using a temperature sensor to collect the ambient temperature data of the location of the fiber grating sensor; S2, performing waveform feature extraction based on modal decomposition on the original reflection spectrum data to obtain a set of inherent modal waveform features of the original reflection spectrum; S3, extracting temperature time series change features from the ambient temperature data to obtain ambient temperature time series features; S4, performing temperature compensation deep learning based on spectrum analysis on the set of the ambient temperature time series features and the inherent modal waveform features of the original reflection spectrum to obtain temperature compensated reflection spectrum dynamic response coding features; S5, generating temperature compensated reflection spectrum data based on the temperature compensated reflection spectrum dynamic response coding features.

[0025] In the above-mentioned noise reduction method of the fiber Bragg grating sensing system based on deep learning, the step S1 uses a fiber Bragg grating sensor demodulator to collect the original reflection spectrum data of the fiber Bragg grating sensor, and uses a temperature sensor to collect the ambient temperature data of the location of the fiber Bragg grating sensor. It should be understood that the fiber Bragg grating sensor demodulator uses the optical and electrical conversion principles to convert the reflection spectrum optical signal returned by the fiber Bragg grating sensor into an electrical signal, and uses an analog-to-digital converter to perform digital processing, thereby obtaining the original reflection spectrum data for subsequent analysis. The temperature sensor converts the ambient temperature into an electrical signal based on the change of the resistance value of the thermal resistor with temperature, or the thermoelectric effect of the thermocouple, and then amplifies and filters the electrical signal to obtain digital data that can accurately reflect the ambient temperature. In the fiber Bragg grating sensing system, temperature changes will cause the optical fiber material to expand and contract, change the grating period and effective refractive index, and cause the reflection wavelength to drift. This drift is intertwined with the wavelength offset caused by the target physical quantity, which greatly interferes with the accuracy of the measurement. Therefore, the present application helps to deeply analyze the dynamic interactive response relationship between temperature and reflection spectrum data by acquiring the original reflection spectrum data of the fiber Bragg grating sensor and the temperature data of its environment, so as to achieve effective temperature compensation and noise reduction.

[0026] Specifically, the fiber Bragg grating sensor interrogator can accurately capture the specific wavelength light signal reflected by the fiber Bragg grating sensor. Whenever a broadband light source is irradiated onto the fiber Bragg grating, due to the specific combination of the grating period and the effective refractive index, only the light of the specific wavelength that meets the Bragg condition will be reflected back. This reflection phenomenon depends not only on the physical properties of the optical fiber itself, but also on external factors such as strain and pressure. Therefore, the fiber Bragg grating sensor interrogator needs to have high sensitivity to identify tiny changes and have sufficient resolution to distinguish different reflected signals. In this way, the fiber Bragg grating sensor interrogator can effectively extract useful reflection spectrum information from complex background noise.

[0027] At the same time, the selection of temperature sensors is also crucial in collecting ambient temperature data. Temperature sensors work based on the principle of thermal resistors or thermocouples, and convert temperature information into electrical signals by detecting the resistance value of the material that changes with temperature or the thermoelectric potential generated. For fiber grating sensing systems, it is particularly important to understand and control the temperature conditions where the optical fiber is located, because temperature fluctuations will cause the optical fiber material to expand and contract, which will in turn affect the grating period and effective refractive index, and ultimately cause the reflection wavelength to drift. In order to obtain data that accurately reflects the ambient temperature, the temperature sensor must not only have high accuracy and fast response capabilities, but also be able to adapt to various complex working environments. This means that the temperature sensor should be able to operate stably under different climatic conditions and provide continuous and reliable temperature readings.

[0028] In order to further improve the accuracy of data collection, it is possible to consider using advanced signal processing technology to amplify and filter the electrical signals from the temperature sensor. For example, by amplifying the signal, the useful signal strength can be enhanced and data distortion caused by a low signal-to-noise ratio can be reduced; while appropriate filtering processing can help remove high-frequency noise interference and ensure that the obtained temperature data is as pure as possible. In addition, digital filtering algorithms can be used to post-process the collected raw data in order to more finely adjust the signal quality and make it more suitable for subsequent analysis needs.

[0029] In the specific implementation process, in order to ensure the data synchronization between the fiber Bragg grating sensor demodulator and the temperature sensor, a unified time reference is usually used to calibrate the two. This requires that the sampling frequencies of the two are matched and that their respective data can be recorded within a common time frame. Doing so not only helps to simplify the subsequent data processing process, but also ensures that all relevant information can be correctly associated when cross-dimensional data analysis is performed, thereby improving the reliability and accuracy of the overall system. In this way, combining the advantages of the fiber Bragg grating sensor demodulator and the temperature sensor, a complete data acquisition system can be constructed, laying a solid foundation for the subsequent in-depth analysis of the dynamic interactive response relationship between temperature and reflection spectrum data. This system can not only provide high-quality original reflection spectrum data, but also monitor and record the temperature conditions of the environment in which the fiber Bragg grating sensor is located in real time, allowing researchers to have a more comprehensive understanding of the specific impact of temperature changes on the performance of fiber Bragg grating sensors.

[0030] In the above-mentioned noise reduction method of the fiber Bragg grating sensing system based on deep learning, the step S2 is to extract waveform features based on modal decomposition of the original reflection spectrum data to obtain a set of inherent modal waveform features of the original reflection spectrum. Figure 3 FIG. 1 is a flowchart of sub-step S2 of the noise reduction method of the fiber Bragg grating sensing system based on deep learning according to an embodiment of the present application. Figure 3 As shown, the step S2 includes the steps of: S21, performing empirical mode decomposition on the original reflection spectrum data to obtain a set of original reflection spectrum inherent modal components; S22, using a waveform feature extractor based on a 1D-CNN model to extract waveform features of each original reflection spectrum inherent modal component in the set of original reflection spectrum inherent modal components to obtain a set of original reflection spectrum inherent modal waveform feature vectors as the set of original reflection spectrum inherent modal waveform features.

[0031] Specifically, in step S21, the original reflection spectrum data is subjected to empirical mode decomposition to obtain a set of intrinsic mode components of the original reflection spectrum. It should be understood that since the original reflection spectrum data is a complex signal with multiple frequency components superimposed, and contains multi-scale noise (such as thermal noise, environmental vibration noise), it is often difficult to extract useful feature information by directly processing it, and the computational complexity is increased. Therefore, the present application adopts empirical mode decomposition technology to decompose the original reflection spectrum data into several intrinsic mode functions according to the time scale characteristics of the original reflection spectrum data itself, so as to obtain a set of intrinsic mode components of the original reflection spectrum. Empirical mode decomposition is based on the local characteristic time scale of the signal, and decomposes the components of different frequencies in the signal step by step through an iterative screening process. Each intrinsic mode component must meet two conditions: in the entire data segment, the number of extreme points and the number of zero crossing points must be equal, or differ by at most one; at any data point, the mean of the upper and lower envelopes defined by the local maximum point and the local minimum point, respectively, is zero. In actual operation, the upper and lower envelopes of the signal are obtained through cubic spline interpolation, the mean of the envelope is calculated, and the mean is subtracted from the original signal. After multiple iterations, each original reflection spectrum intrinsic modal component is finally obtained, thereby achieving effective decomposition of the original reflection spectrum data. Here, each original reflection spectrum intrinsic modal component corresponds to the fluctuation components of different frequency scales in the original reflection spectrum data, which helps to highlight the local characteristics of the signal and reduce the complexity of the data, thereby facilitating the subsequent reflection spectrum feature extraction.

[0032] Specifically, the complexity of the original reflection spectrum data is mainly reflected in the diversity of frequency components and the variety of noise types it contains. In this case, EMD, as an adaptive data analysis tool, can decompose according to the local characteristic time scale of the signal itself without presetting any basis function or filter parameters. The core of this method lies in its iterative screening process, through which the components of different frequencies in the original reflection spectrum data can be separated step by step. In each iteration, based on the local maximum and local minimum points of the signal, the upper and lower envelopes are constructed using the cubic spline interpolation method, and the mean of these two envelopes is calculated. Subsequently, this mean is subtracted from the original signal to obtain a new signal sequence. Repeat the above steps until a specific condition is met, that is, in the entire data segment, the number of extreme points and the number of zero crossings must be equal or differ by at most one; at the same time, at any data point, the mean of the upper and lower envelopes defined by the local maximum and local minimum points is zero. When such conditions are met, an IMF component is obtained.

[0033] It is worth noting that each IMF component corresponds to a different frequency scale fluctuation component in the original reflection spectrum data. This means that through EMD decomposition, the originally complex signal can be disassembled in order according to the frequency, so that each IMF component can represent the fluctuation characteristics within a specific frequency band. Doing so not only helps to highlight the local characteristics of the signal, but also effectively reduces the complexity of the overall data, which provides great convenience for the subsequent reflection spectrum feature extraction. For example, in actual operation, for an original reflection spectrum data containing multiple frequency components, after EMD processing, multiple IMF components can be obtained, each of which focuses on a different frequency range, clearly showing the expression of the original signal at different scales.

[0034] In addition, in order to ensure the quality of each IMF component, the set criteria must be strictly followed throughout the decomposition process. Specifically, each iteration requires accurate calculation of the upper and lower envelopes and their average values ​​to ensure that the resulting new signal sequence meets the basic requirements for becoming an IMF. This includes checking whether the new sequence satisfies the relationship between the number of extreme points and the number of zero crossings, and verifying whether the mean of the upper and lower envelopes is close to zero. Only when all these conditions are met can the current sequence be confirmed as a valid IMF component. Otherwise, it is necessary to continue iterative screening until all conditions are met. Although this process may be time-consuming, it is crucial to improving the quality of the final IMF components, because high-quality IMF components are the basis for subsequent in-depth analysis.

[0035] Through this meticulous decomposition method, EMD technology can effectively reveal the structural information hidden in the original reflection spectrum data. Each IMF component not only carries the fluctuation characteristics of a specific frequency band, but also reflects the trend of the signal in this frequency band changing with time and environment. Therefore, with the help of EMD technology, purer and more representative feature information can be extracted from the original reflection spectrum data.

[0036] Specifically, in step S22, the waveform feature extractor based on the 1D-CNN model is used to extract the waveform features of each original reflection spectrum intrinsic modal component in the set of the original reflection spectrum intrinsic modal components to obtain a set of original reflection spectrum intrinsic modal waveform feature vectors as the set of original reflection spectrum intrinsic modal waveform features. It should be understood that the processing of the intrinsic modal components of the reflection spectrum by the traditional threshold denoising method relies on manually setting the threshold, which limits its adaptability in complex noise environments. When processing one-dimensional signals, the one-dimensional convolutional neural network (1D-CNN) model has the ability to automatically learn the local features and patterns of the signal. By using a one-dimensional convolution kernel to perform a sliding convolution operation on the signal, it can effectively learn the local waveform patterns (such as peaks and platform areas) of the original reflection spectrum intrinsic modal components, capture the spatial correlation between temperature drift and strain response, and thus generate the original reflection spectrum intrinsic modal waveform feature vector that can accurately characterize its waveform characteristics, providing a key feature basis for the subsequent in-depth analysis of the impact of ambient temperature on the reflection spectrum data.

[0037] In the above-mentioned noise reduction method of the fiber Bragg grating sensing system based on deep learning, the step S3 extracts the temperature time series change characteristics from the ambient temperature data to obtain the ambient temperature time series characteristics. In a specific example of the present application, the step S3 includes: extracting the time series characteristics of the ambient temperature data based on the CNN-LSTM hybrid model to obtain the ambient temperature time series feature vector as the ambient temperature time series characteristics. That is, the present application takes into account that the influence of ambient temperature on the grating has a time lag effect (such as thermal inertia of the material), and the temperature information at a single time point cannot accurately characterize its dynamic interference with the reflection spectrum data. Therefore, in order to effectively capture the time series nonlinear propagation law of ambient temperature data, the present application adopts a hybrid model combining a convolutional neural network (CNN) and a long short-term memory network (LSTM) to extract the time series characteristics of the ambient temperature data. Specifically, the CNN model can automatically extract the time series characteristics of the ambient temperature data in each local time domain through the alternating stacking of convolutional layers and pooling layers, which helps to reveal the local detail patterns of the ambient temperature changing over time, such as mutation points, heating rates, etc., while reducing the dimension of the data and improving the computational efficiency. The LSTM model is good at processing long-term dependencies in sequence data. It introduces three control gate structures, namely the forget gate, input gate and output gate, to perform sequence modeling on each local time domain feature extracted by the CNN model. It can capture the long-term change trend in the ambient temperature data, such as periodic fluctuations, overall rising and falling trends, etc., thereby achieving comprehensive modeling of the nonlinear propagation law of the ambient temperature data time series and generating an ambient temperature time series feature vector. Through the above processing, the ambient temperature time series feature vector integrates the local detail features and long-term change trends of the ambient temperature data, and can provide a more comprehensive and accurate description of the ambient temperature characteristics for subsequent temperature compensation and noise reduction processing.

[0038] In the above-mentioned noise reduction method of the fiber Bragg grating sensing system based on deep learning, the step S4 performs temperature compensation deep learning based on spectral analysis on the set of the ambient temperature time series characteristics and the original reflection spectrum inherent modal waveform characteristics to obtain the temperature compensated reflection spectrum dynamic response coding characteristics. Specifically, in order to deeply understand the dynamic influence pattern of ambient temperature on the reflection spectrum data, the present application proposes a temperature compensation deep learning method based on spectral analysis, which deeply queries and analyzes the interactive response pattern between the ambient temperature time series characteristics and the inherent modal waveform characteristics of each original reflection spectrum to separate the dual effects of temperature change and target physical quantity (such as strain) on the reflection spectrum, removes temperature-related noise components, retains the real reflection spectrum changes caused by the target physical quantity (such as strain), and dynamically adjusts the weights of each characteristic component by introducing nonlinear spectral aggregation technology to enhance the expressiveness of the target physical quantity characteristics, generate a temperature compensated reflection spectrum dynamic response coding vector, and provide an important basis for subsequent noise reduction and temperature compensation work. Among them, Figure 4 FIG. 4 is a flowchart of sub-step S4 of the noise reduction method of the fiber Bragg grating sensing system based on deep learning according to an embodiment of the present application. Figure 4 As shown, the step S4 includes the steps of: S41, performing nonlinear interactive response coding on each original reflection spectrum intrinsic modal waveform feature vector in the set of the ambient temperature time series feature vector and the original reflection spectrum intrinsic modal waveform feature vector to obtain a set of ambient temperature-original reflection spectrum intrinsic modal waveform feature implicit interactive coding vectors; S42, performing adjacency relationship graph expression on the set of ambient temperature-original reflection spectrum intrinsic modal waveform feature implicit interactive coding vectors to obtain an ambient temperature-original reflection spectrum intrinsic modal waveform implicit interactive decision point state Laplace matrix; S43, performing core component condensation based on spectral decomposition on the ambient temperature-original reflection spectrum intrinsic modal waveform implicit interactive decision point state Laplace matrix to obtain a temperature compensated reflection spectrum dynamic response coding feature vector as the temperature compensated reflection spectrum dynamic response coding feature.

[0039] Specifically, the step S41 is expressed by the formula:

[0040]

[0041]

[0042] in, represents the set of intrinsic modal waveform feature vectors of the original reflection spectrum, , , and They represent the first, second, and third in the set of original reflection spectrum intrinsic modal waveform feature vectors. and vectors, Represents the number of implicit interaction coding vectors of the inherent modal waveform characteristics of the ambient temperature-original reflection spectrum, represents the weight matrix, represents the bias term, represents the time series characteristic vector of ambient temperature, represents matrix multiplication, represents the sigmoid function, Represents the set of implicit interactive coding vectors of the inherent modal waveform characteristics of the ambient temperature-original reflection spectrum, The first one in the set of implicit interaction coding vectors representing the intrinsic modal waveform characteristics of the ambient temperature and the original reflection spectrum vectors, , and They represent the first, second and third implicit interaction coding vectors of the inherent modal waveform characteristics of the ambient temperature and the original reflection spectrum. vectors.

[0043] That is, the dynamic nonlinear mapping relationship between the ambient temperature time series feature vector and the original reflection spectrum intrinsic modal waveform feature vector is constructed using the neural network architecture and nonlinear activation function, so as to explore the modulation law of temperature change on specific modal components. Specifically, the reflection spectrum components that are strongly correlated with the ambient temperature are screened out through the implicit query mechanism, and the denoised feature expression is reconstructed in the latent space to separate the temperature-sensitive area from the real signal characteristics. The generated ambient temperature-original reflection spectrum intrinsic modal waveform feature implicit interaction encoding vector can compress the global temperature-reflection spectrum interaction information, suppress noise diffusion and retain key physical characteristics.

[0044] Figure 5 FIG. 4 is a flowchart of sub-step S42 of the noise reduction method of the fiber Bragg grating sensing system based on deep learning according to an embodiment of the present application. Figure 5As shown, the step S42 includes the steps of: S421, calculating the state class neighborhood matrix of the implicit interaction decision point of the ambient temperature-original reflection spectrum inherent modal waveform based on the set of implicit interaction coding vectors of the ambient temperature-original reflection spectrum inherent modal waveform features; S422, calculating the state class degree matrix of the implicit interaction decision point of the ambient temperature-original reflection spectrum inherent modal waveform based on the set of implicit interaction coding vectors of the ambient temperature-original reflection spectrum inherent modal waveform features; S423, calculating the difference matrix between the state class degree matrix of the implicit interaction decision point of the ambient temperature-original reflection spectrum inherent modal waveform and the state class neighborhood matrix of the implicit interaction decision point of the ambient temperature-original reflection spectrum inherent modal waveform to obtain the state Laplace matrix of the implicit interaction decision point of the ambient temperature-original reflection spectrum inherent modal waveform.

[0045] More specifically, in a specific example of the present application, the step S421 includes: calculating the Poincare distance between each two implicit interaction coding vectors of the ambient temperature-original reflection spectrum intrinsic modal waveform features in the set of the ambient temperature-original reflection spectrum intrinsic modal waveform features to obtain the ambient temperature-original reflection spectrum intrinsic modal waveform implicit interaction decision point state class neighborhood matrix composed of multiple Poincare distances, which is expressed by the formula:

[0046]

[0047] in, The first one in the set of implicit interaction coding vectors representing the intrinsic modal waveform characteristics of the ambient temperature and the original reflection spectrum vectors, represents the L1 norm, represents the inverse hyperbolic cosine function, It represents the eigenvalue of the (i, j)th position in the state class neighborhood matrix of the implicit interaction decision point of the ambient temperature-original reflection spectrum intrinsic mode waveform, that is, and The Poincare norm between Represents the state class neighborhood matrix of the implicit interaction decision point of the ambient temperature-original reflection spectrum intrinsic mode waveform, , , and They respectively represent the eigenvalues ​​of the (1,1)th, (n,1)th, (1,n)th and (n,n)th positions in the state class neighborhood matrix of the implicit interaction decision point of the inherent mode waveform of the ambient temperature-original reflection spectrum.

[0048] That is, since the coupling relationship between temperature fluctuation and reflection spectrum shift presents a nonlinear topological structure in the latent space, the traditional Euclidean distance is difficult to effectively measure the local correlation of data points in high-dimensional manifolds. Therefore, the present application further uses the interactive response information between the temperature fluctuation characteristics and the intrinsic modal waveform characteristics of each reflection spectrum as the decision point, and reveals the relative position and geometric structure characteristics in the multi-dimensional space by calculating the Poincare distance, and more accurately characterizes the dynamic change law of the decision point state (that is, the influence pattern of the ambient temperature on the intrinsic modal component of the reflection spectrum), so as to more accurately capture the complex interaction pattern between the temperature-sensitive area and the reflection spectrum characteristics. In this way, the generated ambient temperature-original reflection spectrum intrinsic modal waveform implicit interaction decision point state class neighborhood matrix not only simplifies the representation complexity of high-dimensional nonlinear relationships, but also highlights the characteristic offset pattern of temperature-sensitive areas through the encoding of local structural information, so that the model can focus on the change path of key physical characteristics based on the adjacency matrix expressed in a graphical manner.

[0049] More specifically, in a specific example of the present application, the step S422 includes: calculating the characteristic deviation of each ambient temperature-original reflection spectrum intrinsic modal waveform feature implicit interaction coding vector in the set of the ambient temperature-original reflection spectrum intrinsic modal waveform feature implicit interaction coding vector relative to the set of the ambient temperature-original reflection spectrum intrinsic modal waveform feature implicit interaction coding vector, and constructing a diagonal matrix based on the characteristic deviation of each ambient temperature-original reflection spectrum intrinsic modal waveform feature implicit interaction coding vector to obtain the ambient temperature-original reflection spectrum intrinsic modal waveform implicit interaction decision point state class degree matrix, which is expressed by the formula:

[0050]

[0051] in, It represents the eigenvalue of the (i,i)th position in the state class degree matrix of the implicit interaction decision point of the ambient temperature-original reflection spectrum intrinsic modal waveform, that is, The semantic shift metric value of the set of implicit interaction coding vectors relative to the ambient temperature-original reflection spectrum intrinsic modal waveform feature, Represents the state class matrix of the implicit interaction decision point of the ambient temperature-original reflection spectrum intrinsic mode waveform, and They respectively represent the eigenvalues ​​of the (1,1)th position and the (n,n)th position in the state class degree matrix of the implicit interaction decision point of the inherent modal waveform of the ambient temperature-original reflection spectrum.

[0052] That is, the degree of deviation between the state of each decision point and the whole set is measured by the characteristic offset, so as to accurately identify the characteristic components significantly affected by temperature, thereby revealing the dynamic response pattern of the temperature-sensitive area. Specifically, the state class matrix of the decision point of the implicit interaction of the ambient temperature-original reflection spectrum intrinsic mode waveform is constructed by stacking the characteristic offset, so that the subsequent spectrum analysis can dynamically optimize the aggregation method of the global correlation information based on the node connection strength. At the same time, the state class matrix of the decision point of the implicit interaction of the ambient temperature-original reflection spectrum intrinsic mode waveform is used as the input basis of the Laplace matrix. Through the nonlinear superposition effect of the characteristic offset, the local disturbance of the temperature-sensitive area is converted into a resolvable global topological feature, thereby improving the robustness of the model to the separation of noise and real signals.

[0053] More specifically, the step S423 is expressed by the formula:

[0054] in, Represents the Laplace matrix of the implicit interaction decision point state of the inherent mode waveform of the ambient temperature-original reflection spectrum.

[0055] That is, by utilizing the difference operation of the degree matrix of the implicit interaction decision point state of the ambient temperature-original reflection spectrum intrinsic modal waveform and the neighborhood matrix of the implicit interaction decision point state of the ambient temperature-original reflection spectrum intrinsic modal waveform, the topological constraint relationship between the nodes in the data manifold is explicitly quantified, so that the generated Laplace matrix of the implicit interaction decision point state of the ambient temperature-original reflection spectrum intrinsic modal waveform can fuse the local connection information of the neighborhood matrix and the node importance information of the degree matrix, providing a frequency domain analysis basis that reflects the dynamic characteristics of the temperature-reflection spectrum interaction for the subsequent spectral decomposition.

[0056] Figure 6 FIG. 4 is a flowchart of sub-step S43 of the noise reduction method of the fiber Bragg grating sensing system based on deep learning according to an embodiment of the present application. Figure 6 As shown, the step S43 includes the steps of: S431, performing constraint optimization based on cut space and cyclic space on the implicit interaction decision point state Laplace matrix of the ambient temperature-original reflection spectrum intrinsic modal waveform to obtain an optimized ambient temperature-original reflection spectrum intrinsic modal waveform implicit interaction decision point state Laplace matrix; S432, performing spectral decomposition on the optimized ambient temperature-original reflection spectrum intrinsic modal waveform implicit interaction decision point state Laplace matrix to obtain a set of ambient temperature-original reflection spectrum intrinsic modal waveform implicit interaction core component coding vectors; S433, performing attention fusion on the set of ambient temperature-original reflection spectrum intrinsic modal waveform implicit interaction core component coding vectors to obtain the temperature compensated reflection spectrum dynamic response coding feature vector.

[0057] In particular, in a preferred example of the present application, the step S431 is expressed by the formula:

[0058] in, express The cut space matrix of express The circulant space matrix of represents the inverse of the matrix, represents the exponential function with base e, represents matrix addition, Represents the optimized ambient temperature-original reflection spectrum intrinsic mode waveform implicit interaction decision point state Laplace matrix.

[0059] That is, by introducing the cut space matrix and the circulant space matrix, and extracting the connected component information in the graph structure through pseudo-inverse operations, the local neighborhood relationship is combined with the global topological constraints, thereby enhancing the characterization ability of the state Laplace matrix of the implicit interaction decision point of the ambient temperature-original reflection spectrum inherent modal waveform for nonlinear dynamic interactions. Specifically, through the dual constraints of the cut space matrix and the circulant space matrix, the frequency domain analysis of the Laplace matrix of the implicit interaction decision point state of the ambient temperature-original reflection spectrum intrinsic mode waveform is extended to the topological invariant level. The generated optimized Laplace matrix of the implicit interaction decision point state of the ambient temperature-original reflection spectrum intrinsic mode waveform can accurately locate the discretized domain of temperature interference through the connected component information extracted by the cut space closure, and provide a high signal-to-noise ratio frequency domain projection space for the subsequent core component condensation. In addition, the inhibitory effect of the circulant space closure on periodic noise, combined with the algebraic expression of the global structural invariant, enables the model to have stronger generalization ability when inversely generating temperature compensated reflection spectra, and finally realizes the dynamic decoupling and high-precision denoising of cross-sensitivity effects, which significantly improves the stability and measurement reliability of the sensing system under complex working conditions.

[0060] More specifically, the step S432 is expressed by the formula:

[0061] in, represents the spectral decomposition function, Express The matrix composed of the set arrangement of the implicit interaction core component encoding vectors of the ambient temperature-original reflection spectrum intrinsic modal waveform obtained by spectrum decomposition, , and They respectively represent the first, second and kth implicit interaction core component coding vectors of the inherent modal waveform of the ambient temperature-original reflection spectrum in the set of implicit interaction core component coding vectors of the inherent modal waveform of the ambient temperature-original reflection spectrum, Express The diagonal matrix composed of the eigenvalues ​​of the implicit interaction core components of the ambient temperature-original reflection spectrum intrinsic mode waveform obtained by spectrum decomposition is , and They represent the first, second and kth implicit interaction core component eigenvalues ​​of the ambient temperature-original reflection spectrum intrinsic modal waveform in the diagonal matrix, Denotes matrix diagonalization.

[0062] That is, the optimized ambient temperature-original reflection spectrum intrinsic modal waveform implicit interaction decision point state Laplace matrix under global topology constraints is mapped to the low-dimensional spectral domain through spectral decomposition, so that the dynamic modulation law of temperature interference is explicitly expressed in the low-dimensional coordinate system, and the temperature sensitive features and real signal features are separated through the orthogonality of the eigenvectors. In addition, the set of the implicit interaction core component encoding vectors of the ambient temperature-original reflection spectrum intrinsic modal waveform as a compressed representation of the data in the spectral domain can quantify the contribution weights of different physical components to the reflection spectrum offset through the amplitude difference of the eigenvalues, providing a highly distinguishable feature basis for subsequent dynamic query responses.

[0063] More specifically, the step S433 is expressed by the formula:

[0064] in, represents the adaptive fusion network, represents the weight parameter matrix of the adaptive fusion network, represents the bias term of the adaptive fusion network, represents the core component feature significance score conversion vector of the adaptive fusion network, represents the normalization function, represents the normalized core component feature significance score factor, represents the gating threshold, Represents a mask operation, express The corresponding feature fusion weight factor, Represents the encoding feature vector of the dynamic response of the temperature compensated reflection spectrum.

[0065] That is, through the dynamic allocation of attention weights, a feature importance hierarchy is constructed in the low-dimensional spectral domain, the local feature sensitivity to temperature-sensitive areas is enhanced, and the gating mechanism is used to filter the periodic interference of cross-cycle noise, so that the generated temperature-compensated reflection spectrum dynamic response encoding feature vector can adapt to the dynamic changes of the ambient temperature fluctuation range and the reflection spectrum offset amplitude. In addition, attention fusion maps the discretized core component encoding vector to a continuous high-dimensional representation space through a context-aware feature combination strategy, providing a more physically interpretable temperature compensation benchmark for the subsequent decoder.

[0066] In the above-mentioned noise reduction method of the fiber Bragg grating sensing system based on deep learning, the step S5 generates the temperature-compensated reflection spectrum data based on the dynamic response coding feature of the temperature-compensated reflection spectrum. In a specific example of the present application, the step S5 includes: inputting the temperature-compensated reflection spectrum dynamic response coding vector into a noise reduction generation module based on a decoder to obtain the temperature-compensated reflection spectrum data. Specifically, the decoder is based on a multi-layer neural network architecture and has a powerful data reconstruction capability. It can parse the reflection spectrum features closely related to the target physical quantity from the temperature-compensated reflection spectrum dynamic response coding vector, while suppressing the interference of temperature-related noise, and gradually restore the reflection spectrum data reflecting the change of the target physical quantity. In the decoding stage, the decoder performs a spatial reverse mapping of the temperature-compensated reflection spectrum dynamic response coding vector according to the input temperature-compensated reflection spectrum dynamic response coding vector through a series of deconvolution operations, so as to gradually map it into a smooth reflection spectrum, thereby realizing the conversion from the high-dimensional coding feature space to the data space, and obtaining the reflection spectrum data after temperature compensation.

[0067] In summary, the noise reduction method of the fiber Bragg grating sensing system based on deep learning according to the embodiment of the present application is explained, which first synchronously collects the original reflection spectrum data of the fiber Bragg grating sensor and the ambient temperature data of the fiber Bragg grating sensor, and introduces a data processing algorithm based on deep learning, and performs empirical mode decomposition and waveform feature extraction on the original reflection spectrum data to capture its inherent intrinsic modal waveform characteristics, and then further combines the ambient temperature data, and performs implicit query interactive response encoding on the ambient temperature characteristics and the original reflection spectrum intrinsic modal waveform characteristics to understand the dynamic influence pattern of the ambient temperature on the reflection spectrum data, thereby realizing temperature compensation and noise reduction of the reflection spectrum data. In this way, the influence of temperature fluctuations on the reflection spectrum can be adaptively identified and compensated, the noise caused by ambient temperature fluctuations can be effectively suppressed, and the measurement accuracy of the fiber Bragg grating sensor can be improved.

[0068] Furthermore, a fiber grating sensing system based on deep learning is also provided.

[0069] Figure 7FIG. 1 is a block diagram of a fiber Bragg grating sensing system based on deep learning according to an embodiment of the present application. Figure 7 As shown, according to the deep learning-based fiber grating sensing system 100 of the embodiment of the present application, it includes: a sensor data acquisition module 110, which is used to use a fiber grating sensor demodulator to collect the original reflection spectrum data of the fiber grating sensor, and use a temperature sensor to collect the ambient temperature data of the location of the fiber grating sensor; a waveform feature extraction module 120, which is used to perform waveform feature extraction based on modal decomposition on the original reflection spectrum data to obtain a set of inherent modal waveform features of the original reflection spectrum; a time series change feature extraction module 130, which is used to extract temperature time series change features from the ambient temperature data to obtain ambient temperature time series features; a temperature compensation deep learning module 140, which is used to perform temperature compensation deep learning based on spectral analysis on the set of the ambient temperature time series features and the original reflection spectrum inherent modal waveform features to obtain temperature compensated reflection spectrum dynamic response coding features; a compensation data generation module 150, which is used to generate temperature compensated reflection spectrum data based on the temperature compensated reflection spectrum dynamic response coding features.

[0070] Here, those skilled in the art can understand that the specific operations of each module in the above-mentioned fiber grating sensing system based on deep learning have been referred to above. Figures 1 to 6 The noise reduction method of the fiber Bragg grating sensing system based on deep learning has been introduced in detail, and therefore, its repeated description will be omitted.

[0071] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

[0072] In the above embodiments, the description of each embodiment has its own emphasis. For the parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiment described above is only schematic. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0073] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference to a figure in a claim should not be considered as limiting the claim to which it relates.

[0074] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units stated in the system claims can also be implemented by one unit through software or hardware.

[0075] Finally, it should be noted that the above description has been given for the purpose of illustration and description. In addition, the above embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A noise reduction method for a fiber Bragg grating sensing system based on deep learning, characterized in that: include: The fiber grating sensor demodulator is used to collect the original reflection spectrum data of the fiber grating sensor, and the temperature sensor is used to collect the ambient temperature data of the location where the fiber grating sensor is located; Performing waveform feature extraction based on modal decomposition on the original reflection spectrum data to obtain a set of inherent modal waveform features of the original reflection spectrum; Extracting temperature time series variation characteristics from the ambient temperature data to obtain ambient temperature time series characteristics; Performing temperature compensation deep learning based on spectrum analysis on the set of the ambient temperature time series characteristics and the original reflection spectrum inherent modal waveform characteristics to obtain a temperature compensation reflection spectrum dynamic response coding feature; Based on the temperature-compensated reflection spectrum dynamic response coding characteristics, temperature-compensated reflection spectrum data is generated.

2. The noise reduction method of the fiber Bragg grating sensing system based on deep learning according to claim 1, characterized in that: The original reflection spectrum data is subjected to waveform feature extraction based on modal decomposition to obtain a set of inherent modal waveform features of the original reflection spectrum, including: Performing empirical mode decomposition on the original reflection spectrum data to obtain a set of original reflection spectrum intrinsic mode components; A waveform feature extractor based on a 1D-CNN model is used to extract waveform features of each original reflection spectrum intrinsic modal component in the set of original reflection spectrum intrinsic modal components to obtain a set of original reflection spectrum intrinsic modal waveform feature vectors as the set of original reflection spectrum intrinsic modal waveform features.

3. The noise reduction method of the fiber Bragg grating sensing system based on deep learning according to claim 2, characterized in that: Extracting the temperature time series variation characteristics from the ambient temperature data to obtain the ambient temperature time series characteristics includes: The ambient temperature data is subjected to time series feature extraction based on a CNN-LSTM hybrid model to obtain an ambient temperature time series feature vector as the ambient temperature time series feature.

4. The noise reduction method of the fiber Bragg grating sensing system based on deep learning according to claim 3, characterized in that: The temperature compensation deep learning based on spectrum analysis is performed on the set of the ambient temperature time series characteristics and the original reflection spectrum inherent modal waveform characteristics to obtain the temperature compensation reflection spectrum dynamic response coding characteristics, including: Performing nonlinear interactive response coding on each original reflection spectrum intrinsic modal waveform feature vector in the set of the ambient temperature time series feature vector and the original reflection spectrum intrinsic modal waveform feature vector to obtain a set of ambient temperature-original reflection spectrum intrinsic modal waveform feature implicit interactive coding vectors; The set of implicit interaction coding vectors of the ambient temperature-original reflection spectrum intrinsic modal waveform characteristics is expressed in an adjacency graph to obtain an implicit interaction decision point state Laplace matrix of the ambient temperature-original reflection spectrum intrinsic modal waveform; The Laplace matrix of the implicit interaction decision point state of the ambient temperature-original reflection spectrum intrinsic mode waveform is subjected to core component aggregation based on spectrum decomposition to obtain a temperature-compensated reflection spectrum dynamic response coding feature vector as the temperature-compensated reflection spectrum dynamic response coding feature.

5. The noise reduction method of the fiber Bragg grating sensing system based on deep learning according to claim 4, characterized in that: The set of the implicit interaction coding vectors of the ambient temperature-original reflection spectrum intrinsic modal waveform features is expressed in an adjacency graph to obtain the implicit interaction decision point state Laplace matrix of the ambient temperature-original reflection spectrum intrinsic modal waveform, including: Based on the set of implicit interaction coding vectors of the ambient temperature-original reflection spectrum intrinsic modal waveform features, a state-like neighborhood matrix of the ambient temperature-original reflection spectrum intrinsic modal waveform implicit interaction decision point is calculated; Based on the set of implicit interaction coding vectors of the characteristic of the ambient temperature-original reflection spectrum intrinsic modal waveform, calculating the state class degree matrix of the implicit interaction decision point of the ambient temperature-original reflection spectrum intrinsic modal waveform; The difference matrix between the class degree matrix of the implicit interaction decision point state of the ambient temperature-original reflection spectrum intrinsic modal waveform and the class neighborhood matrix of the implicit interaction decision point state of the ambient temperature-original reflection spectrum intrinsic modal waveform is calculated to obtain the implicit interaction decision point state Laplace matrix of the ambient temperature-original reflection spectrum intrinsic modal waveform.

6. The noise reduction method of the fiber Bragg grating sensing system based on deep learning according to claim 5, characterized in that: Based on the set of implicit interaction coding vectors of the ambient temperature-original reflection spectrum intrinsic modal waveform features, the state class neighborhood matrix of the ambient temperature-original reflection spectrum intrinsic modal waveform implicit interaction decision point is calculated, including: The Poincare distance between every two implicit interaction coding vectors of the ambient temperature-original reflection spectrum intrinsic modal waveform features in the set of implicit interaction coding vectors of the ambient temperature-original reflection spectrum intrinsic modal waveform features is calculated to obtain a state-class neighborhood matrix of the ambient temperature-original reflection spectrum intrinsic modal waveform implicit interaction decision points composed of multiple Poincare distances.

7. The noise reduction method of the fiber Bragg grating sensing system based on deep learning according to claim 6, characterized in that: Based on the set of implicit interaction coding vectors of the characteristic of the ambient temperature-original reflection spectrum intrinsic modal waveform, the state class degree matrix of the implicit interaction decision point of the ambient temperature-original reflection spectrum intrinsic modal waveform is calculated, including: Calculate the characteristic deviation of each implicit interaction coding vector of the ambient temperature-original reflection spectrum inherent modal waveform characteristics in the set of implicit interaction coding vectors of the ambient temperature-original reflection spectrum inherent modal waveform characteristics relative to the set of implicit interaction coding vectors of the ambient temperature-original reflection spectrum inherent modal waveform characteristics, and construct a diagonal matrix based on the characteristic deviation of each implicit interaction coding vector of the ambient temperature-original reflection spectrum inherent modal waveform characteristics to obtain the state class degree matrix of the ambient temperature-original reflection spectrum inherent modal waveform implicit interaction decision point.

8. The noise reduction method of the fiber Bragg grating sensing system based on deep learning according to claim 7, characterized in that: The Laplace matrix of the implicit interaction decision point state of the ambient temperature-original reflection spectrum intrinsic modal waveform is subjected to core component aggregation based on spectrum decomposition to obtain a temperature-compensated reflection spectrum dynamic response coding feature vector as the temperature-compensated reflection spectrum dynamic response coding feature, including: Performing constraint optimization based on cut space and cyclic space on the implicit interaction decision point state Laplace matrix of the ambient temperature-original reflection spectrum intrinsic mode waveform to obtain an optimized implicit interaction decision point state Laplace matrix of the ambient temperature-original reflection spectrum intrinsic mode waveform; Performing spectral decomposition on the optimized ambient temperature-original reflection spectrum intrinsic mode waveform implicit interaction decision point state Laplace matrix to obtain a set of ambient temperature-original reflection spectrum intrinsic mode waveform implicit interaction core component encoding vectors; Attention fusion is performed on the set of implicit interaction core component encoding vectors of the ambient temperature-original reflection spectrum intrinsic modal waveform to obtain the temperature compensated reflection spectrum dynamic response encoding feature vector.

9. The noise reduction method of the fiber Bragg grating sensing system based on deep learning according to claim 8, characterized in that: Generating temperature-compensated reflection spectrum data based on the temperature-compensated reflection spectrum dynamic response coding feature includes: The temperature-compensated reflection spectrum dynamic response encoding vector is input into a decoder-based noise reduction generation module to obtain the temperature-compensated reflection spectrum data.

10. A fiber Bragg grating sensing system based on deep learning, characterized in that: include: The sensor data acquisition module is used to collect the original reflection spectrum data of the fiber Bragg grating sensor by using the fiber Bragg grating sensor demodulator, and to collect the ambient temperature data of the location where the fiber Bragg grating sensor is located by using the temperature sensor; A waveform feature extraction module, used for performing waveform feature extraction based on modal decomposition on the original reflection spectrum data to obtain a set of inherent modal waveform features of the original reflection spectrum; A time series variation feature extraction module, used for extracting temperature time series variation features from the ambient temperature data to obtain ambient temperature time series features; A temperature compensation deep learning module, used for performing temperature compensation deep learning based on spectrum analysis on the set of the ambient temperature time series characteristics and the inherent modal waveform characteristics of the original reflection spectrum to obtain a temperature compensation reflection spectrum dynamic response coding feature; The compensation data generating module is used to generate the temperature compensated reflection spectrum data based on the temperature compensated reflection spectrum dynamic response coding characteristics.

Images