Fiber grating dynamic temperature calibration method based on thermal response matching and deep learning and deep learning algorithm module

Abstract

This invention discloses a dynamic temperature calibration method for fiber Bragg gratings based on thermal response matching and deep learning, along with a deep learning algorithm module. The dynamic temperature calibration method includes: a thermal response matching stage between a reference temperature sensor and a fiber Bragg grating sensor; a static temperature calibration stage between the reference temperature sensor and the fiber Bragg grating sensor; and a dynamic residual compensation stage between the fiber Bragg grating sensor and a time-series residual compensation model. This dynamic temperature calibration method can simultaneously reduce both the temperature calibration error and the thermal response error of the fiber Bragg grating sensor.

Description

Technical Field

[0001] This invention relates to fiber Bragg grating sensing technology, and more particularly to a dynamic temperature calibration method for fiber Bragg gratings based on thermal response matching and deep learning, as well as a deep learning algorithm module. Background Technology

[0002] Fiber Bragg grating sensors have outstanding advantages such as small size, resistance to electromagnetic interference, high temperature resistance, stable long-distance signal transmission, and the ability to achieve multi-point distributed synchronous measurement. They have been widely used in temperature monitoring scenarios for various high-temperature dynamic working conditions, such as aero-engines, gas turbines, metallurgical blast furnaces, nuclear reactors, and heat treatment furnaces.

[0003] To accurately establish a quantitative correlation between the calibration center wavelength and the reference temperature, the industry typically uses more precise and technologically mature reference temperature sensors (such as thermocouples and platinum resistance thermometers) to calibrate fiber Bragg grating sensors. Traditional calibration methods generally place the fiber Bragg grating sensor and the reference temperature sensor within the same temperature control device, simultaneously acquiring wavelength data from the fiber Bragg grating sensor and temperature data from the reference temperature sensor. Based on this, a wavelength-temperature mapping model is constructed. Subsequently, relying on this mapping model, the calibration center wavelength of the fiber Bragg grating sensor is inverted and calculated to obtain the corresponding reference temperature.

[0004] However, existing temperature calibration methods are based on the assumption that the thermal response speeds of the fiber Bragg grating sensor and the reference temperature sensor are the same. In reality, there are significant differences in the thermal response speeds between the fiber Bragg grating sensor and the reference temperature sensor, which leads to large errors in the calibrated wavelength-temperature mapping model.

[0005] Furthermore, in most application scenarios, the reference temperature of the object being measured is usually not static, but rather in a state of rapid heating, rapid cooling, periodic fluctuation, or random disturbance. For example, during the start-up and acceleration of an aircraft engine, its temperature can rise from room temperature to hundreds of degrees Celsius within seconds; high-temperature furnaces and industrial kilns also exhibit significant temperature rise rates and marked differences in thermal inertia during start-up, shutdown, and switching of operating conditions.

[0006] However, fiber Bragg grating sensors have a slow thermal response. Before their calibrated center wavelength reaches the wavelength value corresponding to the final compensation temperature, the reference temperature of the object being measured has actually changed, which also leads to a large temperature measurement error. Summary of the Invention

[0007] To address the shortcomings of the prior art, this invention provides a dynamic temperature calibration method for fiber Bragg gratings based on thermal response matching and deep learning, which can simultaneously reduce the temperature calibration error and thermal response error of the fiber Bragg grating sensor.

[0008] The technical problem to be solved by the present invention is achieved through the following technical solution: A dynamic temperature calibration method for fiber Bragg gratings based on thermal response matching and deep learning includes the following steps: Step 1: Using the effective thermal response speed of the reference temperature sensor as the standard, optimize the packaging parameters of the fiber Bragg grating sensor using software simulation to obtain the optimal parameter combination when the thermal response times of the two sensors tend to be consistent. Step 2: Fabricate a fiber Bragg grating and encapsulate the fiber Bragg grating using an encapsulation structure corresponding to the optimal parameter combination to obtain the fiber Bragg grating sensor; Step 3: Use the reference temperature sensor to perform static temperature calibration on the fiber Bragg grating sensor to obtain the wavelength-temperature relationship function of the fiber Bragg grating sensor; Step 4: Synchronously acquire the dynamic reference temperature sequence and dynamic wavelength sequence of the reference temperature sensor and fiber optic grating sensor under various dynamic temperature change conditions; Step 5: Based on the dynamic reference temperature sequence and dynamic wavelength sequence, and combined with the wavelength-temperature relationship function, construct the dynamic thermal response feature data of the fiber optic grating sensor, and then use the dynamic thermal response feature data to train the temporal deep learning network to obtain the temporal residual compensation model. Step 6: Under actual measurement conditions, acquire the real-time wavelength sequence of the fiber Bragg grating sensor according to the preset sampling frequency; Step 7: Based on the real-time wavelength sequence and combined with the wavelength-temperature relationship function, construct the real-time thermal response characteristic data of the fiber optic grating sensor, and then input the real-time thermal response characteristic data into the time-series residual compensation model for inference, so as to obtain the residual compensation value at the current sampling time or the final compensation temperature after residual compensation as predicted by the time-series residual compensation model.

[0009] Furthermore, in step 1, using the effective thermal response speed of the reference temperature sensor as a standard, the packaging parameters of the fiber Bragg grating sensor are optimized using software simulation to obtain the optimal parameter combination when the thermal response times of the two sensors tend to be consistent. The steps are as follows: Step 11: Obtain the effective thermal response time constant of the reference temperature sensor; Step 12: Based on the packaging structure of the fiber Bragg grating sensor, construct a three-dimensional model of the packaging structure in simulation software; Step 13: In the simulation software, set the heat transfer path and multiple packaging parameters related to transient heat conduction analysis for the packaging structure, so that the simulation software can perform transient heat conduction analysis on the packaging structure according to the heat transfer path and each packaging parameter, thereby calculating the simulated thermal response time constant of the fiber Bragg grating sensor. Step 14: With the goal of minimizing the time constant difference between the effective thermal response time constant and the simulated thermal response time constant, construct an objective optimization function in the simulation software, and select several encapsulated parameters as variable parameters. Set a corresponding range of variation for each variable parameter so that the simulation software can iteratively optimize the numerical combination of each variable parameter, thereby obtaining the optimal parameter combination of each variable parameter when the time constant difference is minimized.

[0010] Furthermore, in step 11, the effective thermal response time constant of the reference temperature sensor is obtained through experimental measurement or table lookup.

[0011] Furthermore, in step 3, the step of using the reference temperature sensor to perform static temperature calibration on the fiber Bragg grating sensor to obtain the wavelength-temperature relationship function of the fiber Bragg grating sensor is as follows: Step 31: Place the reference temperature sensor and the fiber optic grating sensor inside the same temperature control device; Step 32: Control the temperature control device to raise or lower the temperature to a certain constant temperature point, and keep it at the constant temperature point for a certain period of time so that the reference temperature sensor and the fiber optic grating sensor reach a stable state at the constant temperature point. Step 33: Collect the static reference temperature and static center wavelength corresponding to the reference temperature sensor and fiber optic grating sensor when they are in a stable state, respectively; Step 34: Repeat steps 32 to 33 to sequentially collect the static reference temperature and static center wavelength of the reference temperature sensor and fiber optic grating sensor when they are in a stable state at different isothermal points. Step 35: Take the static reference temperature and static center wavelength at the same isothermal point or the same sampling time as a set of calibration data pairs, and then fit each set of calibration data pairs to obtain the wavelength-temperature relationship function of the fiber optic grating sensor.

[0012] Furthermore, in step 4, the steps for simultaneously acquiring the dynamic reference temperature sequence and dynamic wavelength sequence corresponding to the reference temperature sensor and fiber optic grating sensor under various dynamic temperature change conditions are as follows: Step 41: Place the reference temperature sensor and the fiber optic grating sensor inside the same temperature control device; Step 42: Set multiple dynamic temperature change conditions in the temperature control device so that the temperature control device can simulate various dynamic temperature change conditions in sequence to change the temperature quickly; Step 43: When the temperature control device simulates various dynamic temperature change conditions for rapid temperature change, the dynamic reference temperature and dynamic center wavelength of the reference temperature sensor and fiber optic grating sensor are collected synchronously at each sampling time according to the preset sampling frequency, so as to obtain the dynamic reference temperature sequence and dynamic wavelength sequence of the two under various dynamic temperature change conditions.

[0013] Furthermore, the dynamic temperature change conditions include two or more of the following: step temperature change conditions, linear slope temperature change conditions, multi-segment variable speed temperature change conditions, periodic temperature change fluctuation conditions, and temperature trajectory playback conditions of the measured object.

[0014] Furthermore, the dynamic thermal response feature data includes thermal response feature vectors and their residual labels at each sampling time. In step 5, based on the dynamic reference temperature sequence and dynamic wavelength sequence, and combined with the wavelength-temperature relationship function, the dynamic thermal response feature data of the fiber optic grating sensor is constructed. Then, the dynamic thermal response feature data is used to train the temporal deep learning network to obtain the temporal residual compensation model. The steps are as follows: Step 51: Preprocess the dynamic reference temperature sequence and dynamic wavelength sequence to remove outliers, filter and denoise, and align timestamps; Step 52: Based on the wavelength-temperature relationship function, convert the dynamic wavelength sequence into the dynamic fundamental temperature sequence of the fiber Bragg grating sensor; Step 53: Subtract the dynamic reference temperature and dynamic base temperature at the same sampling time from the dynamic reference temperature sequence and the dynamic base temperature sequence to obtain the temperature residual sequence between the reference temperature sensor and the fiber Bragg grating sensor. Subtract the dynamic center wavelengths at two adjacent sampling times in the dynamic wavelength sequence to obtain the first-order difference sequence of the dynamic wavelength of the fiber Bragg grating sensor. Subtract the dynamic base temperature at two adjacent sampling times in the dynamic base temperature sequence to obtain the first-order difference sequence of the dynamic temperature of the fiber Bragg grating sensor. Step 54: For any sampling time, taking that sampling time as the endpoint, extract the local wavelength sequence, local temperature sequence, local wavelength first-order difference sequence, and local temperature first-order difference sequence within a preset time length from the dynamic wavelength sequence, dynamic base temperature sequence, dynamic wavelength first-order difference sequence, and dynamic temperature first-order difference sequence, respectively, to construct the thermal response feature vector corresponding to that sampling time, thereby obtaining the thermal response feature vectors at different sampling times; Step 55: Using the temperature residual sequence as the training label set, label the thermal response feature vector at each sampling time with the temperature residual at the corresponding sampling time as the training label; Step 56: Divide the thermal response feature vectors at each sampling time into training datasets, validation datasets, and test datasets according to the different types of dynamic temperature change conditions; Step 57: Train the temporal deep learning network using the training dataset, and use the validation dataset to optimize the hyperparameters and judge the fitting effect of the temporal deep learning network during the training process. Finally, use the test dataset to test the actual performance and generalization ability of the trained temporal deep learning network to obtain the temporal residual compensation model.

[0015] Furthermore, the real-time thermal response feature data includes the thermal response feature vector at the current sampling time; in step 7, based on the real-time wavelength sequence and combined with the wavelength-temperature relationship function, the real-time thermal response feature data of the fiber optic grating sensor is constructed, and then the real-time thermal response feature data is input into the time-series residual compensation model for inference to obtain the residual compensation value at the current sampling time predicted by the time-series residual compensation model or the final compensation temperature after residual compensation. The steps are as follows: Step 71: Preprocess the real-time wavelength sequence to remove outliers and filter for noise reduction; Step 72: Based on the wavelength-temperature relationship function, convert the real-time wavelength sequence into the real-time base temperature sequence of the fiber Bragg grating sensor; Step 73: Subtract the real-time center wavelengths of two adjacent sampling times in the real-time wavelength sequence to obtain the real-time wavelength first-order difference sequence of the fiber Bragg grating sensor; subtract the real-time base temperature of two adjacent sampling times in the real-time base temperature sequence to obtain the real-time temperature first-order difference sequence of the fiber Bragg grating sensor. Step 74: Taking the current sampling time as the endpoint, extract the historical wavelength sequence, historical temperature sequence, historical wavelength first-order difference sequence, and historical temperature first-order difference sequence within a preset time length from the real-time wavelength sequence, real-time base temperature sequence, real-time wavelength first-order difference sequence, and real-time temperature first-order difference sequence, respectively, to construct the thermal response feature vector at the current sampling time. Step 75: Input the thermal response feature vector at the current sampling time into the time-series residual compensation model for inference; Step 76: Receive the residual compensation value at the current sampling time or the final compensated temperature after residual compensation, which is predicted and output by the time-series residual compensation model.

[0016] Furthermore, the loss function used when training the temporal deep learning network is shown below. in, These are the temperature residual values ​​predicted by the model. This represents the actual measured temperature residual value. Mean square error, The mean absolute error, To smooth the constraint regularization term, , and These are the weighting coefficients.

[0017] A deep learning algorithm module includes a processor and a memory. The memory stores a computer program and a timing residual compensation model for the processor to execute and call. When the processor executes the computer program, it calls the timing residual compensation model and performs steps 4 to 7 in the above-mentioned dynamic temperature calibration method.

[0018] The present invention has the following beneficial effects: First, the present invention uses the effective thermal response time constant of the reference temperature sensor as a standard, and reverse designs the packaging structure of the fiber optic grating sensor through software simulation or equivalent thermal model, so that the thermal response time constants of the two tend to be consistent, significantly reducing the timing mismatch under dynamic conditions from a physical perspective; on this basis, the wavelength-temperature relationship function is first established through static calibration, and then the residual dynamic error is modeled and compensated using the time-series deep neural network, and finally the residual compensation value or the final compensation temperature after residual compensation is output. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the steps of the dynamic temperature calibration method provided by the present invention.

[0020] Figure 2 This is a flowchart illustrating the steps of step 1 in the dynamic temperature calibration method provided by the present invention.

[0021] Figure 3 This is a flowchart illustrating step 3 of the dynamic temperature calibration method provided by the present invention.

[0022] Figure 4 This is a flowchart illustrating step 4 of the dynamic temperature calibration method provided by the present invention.

[0023] Figure 5 This is a flowchart illustrating step 5 of the dynamic temperature calibration method provided by the present invention.

[0024] Figure 6 This is a flowchart illustrating step 7 of the dynamic temperature calibration method provided by the present invention.

[0025] Figure 7 This is a schematic diagram of the architecture of the temporal deep learning network provided by the present invention.

[0026] Figure 8 This is a schematic diagram of the packaging structure of the fiber Bragg grating sensor provided by the present invention. Detailed Implementation

[0027] The present invention will now be described in detail with reference to the accompanying drawings and embodiments, examples of which are shown in the drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0028] In the description of this invention, it should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0029] Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first," "second," or "third" may explicitly or implicitly include one or more of that feature. In the description of this invention, "multiple" means two or more, unless otherwise explicitly specified.

[0030] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," "fixing," and "setting," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0031] Example 1 like Figure 1 As shown, a dynamic temperature calibration method for fiber optic gratings based on thermal response matching and deep learning includes the following steps: Step 1: Using the effective thermal response speed of the reference temperature sensor as the standard, optimize the packaging parameters of the fiber Bragg grating sensor using software simulation to obtain the optimal parameter combination when the thermal response times of the two sensors tend to be consistent.

[0032] In step 1, the reference temperature sensor may be, but is not limited to, a thermocouple sensor, a platinum resistance sensor, or a thermistor sensor.

[0033] The effective thermal response rate refers to the time required for the temperature output by the reference temperature sensor to change to a specific percentage of the magnitude of a temperature jump (such as suddenly placing boiling water in from room temperature) when a temperature jump occurs. Different percentages result in different defined thermal response times. For example, the thermal response time corresponding to a change to 63.2% of the jump is the effective thermal response time constant commonly used in thermodynamics.

[0034] Specifically, such as Figure 2 As shown, in step 1, using the effective thermal response speed of the reference temperature sensor as the standard, the packaging parameters of the fiber Bragg grating sensor are optimized using software simulation to obtain the optimal parameter combination when the thermal response times of the two sensors tend to be consistent. The steps are as follows: Step 11: Obtain the effective thermal response time constant of the reference temperature sensor.

[0035] In step 11, there are two methods for obtaining the effective thermal response time constant. One method is experimental measurement: Under standard dynamic temperature change conditions, a step or ramp temperature excitation is applied to the reference temperature sensor, and the output response curve of the reference temperature sensor is recorded. The effective thermal response time constant is obtained by fitting the curve with a first-order inertial model or an equivalent thermal model. Another method is the lookup table combined with correction: based on the model, packaging form, measurement medium and flow state of the reference temperature sensor, the nominal time constant is looked up in the manufacturer's product manual or relevant standard documents, and then the nominal time constant is corrected in combination with the actual installation conditions to obtain the effective thermal response time constant.

[0036] Step 12: Based on the packaging structure of the fiber Bragg grating sensor, construct a three-dimensional model of the packaging structure in simulation software.

[0037] In step 12, the simulation software can be, but is not limited to, finite element simulation software such as COMSOL Multiphysics, Ansys, or Abaqus; professional photonics simulation software such as Lumerical, RSoft, or OptiSystem / OptiFDTD; and programming tools such as MATLAB, Python, or Scilab. These finite element simulation software programs or professional photonics simulation software programs are commonly used in the field, and they integrate transient heat conduction models, allowing direct transient heat conduction analysis of the fiber Bragg grating sensor.

[0038] Step 13: In the simulation software, set the heat transfer path and multiple packaging parameters related to transient heat conduction analysis for the packaging structure, so that the simulation software can perform transient heat conduction analysis on the packaging structure according to the heat transfer path and each packaging parameter, thereby calculating the simulated thermal response time constant of the fiber Bragg grating sensor.

[0039] In step 13, the number and type of the encapsulation parameters vary depending on the encapsulation structure of the fiber Bragg grating sensor. Taking the most common encapsulation tube as an example, the encapsulation parameters related to transient thermal conductivity analysis include, but are not limited to, the inner diameter, outer diameter, wall thickness, length, and thermal conductivity of the encapsulation tube; the cavity gap and connection length between the fiber Bragg grating and the encapsulation tube; the thermal conductivity of the cavity medium; and the fiber length of the fiber Bragg grating within the encapsulation tube, etc., which will not be listed here.

[0040] After setting the heat transfer path and multiple packaging parameters related to transient heat conduction analysis for the packaging structure, the simulation software will call its built-in transient heat conduction model to perform transient heat conduction analysis on the packaging structure, and finally calculate and output the simulated thermal response time constant of the fiber Bragg grating sensor.

[0041] Step 14: With the goal of minimizing the time constant difference between the effective thermal response time constant and the simulated thermal response time constant, construct an objective optimization function in the simulation software, and select several encapsulated parameters as variable parameters. Set a corresponding range of variation for each variable parameter so that the simulation software can iteratively optimize the numerical combination of each variable parameter, thereby obtaining the optimal parameter combination of each variable parameter when the time constant difference is minimized.

[0042] In step 14, the objective optimization function is as follows: in, This is the simulated thermal response time constant of the fiber Bragg grating sensor. is the effective thermal response time constant of the reference temperature sensor.

[0043] Alternatively, in a more general case, the objective optimization function is as follows: in, This indicates the temperature deviation between the fiber Bragg grating sensor and the reference temperature sensor under typical dynamic excitation. This indicates the manufacturing difficulty, cost, or strength constraints of the fiber Bragg grating sensor. , and These are the weight parameters.

[0044] In high-precision scenarios and A higher value can be selected. A lower value can be chosen, especially in low-cost scenarios. and A lower value can be selected. A lower value can be chosen in scenarios that balance accuracy and cost. , and You can choose similar values.

[0045] Step 2: Fabricate a fiber Bragg grating and encapsulate the fiber Bragg grating using an encapsulation structure corresponding to the optimal parameter combination to obtain the fiber Bragg grating sensor.

[0046] In step 2, after obtaining the fiber Bragg grating sensor, the effective thermal response time constant of the fiber Bragg grating sensor needs to be measured to verify the simulation effect of the simulation software. If the measured value of the difference between the two time constants is the same as or close to the predicted value of the simulation software, it indicates that the simulation result is correct, and the next step is performed. If the measured value of the difference between the two time constants deviates significantly from the predicted value of the simulation software, the cause of the error needs to be manually analyzed, such as whether some packaging parameters are missing, or whether the heat transfer path is set correctly. Then, the error is corrected in the simulation software, and the optimal parameter combination is iterated again until the measured value of the difference between the two time constants is the same as or close to the predicted value of the simulation software.

[0047] In this embodiment, as Figure 8As shown, the fiber Bragg grating 1 sensor is encapsulated in the encapsulation tube 2. One end of the encapsulation tube 2 is a closed tube bottom, and the other end is an open tube opening. During encapsulation, the fiber Bragg grating 1 and the encapsulation tube 2 are first aligned along their central axis, then the fiber Bragg grating 1 is inserted into the encapsulation tube 2 from the open tube opening end. Finally, the open tube opening end of the encapsulation tube 2 is connected and fixed by means of heat fusion, glue bonding, or arc welding to form a connection area 3. Since the fiber diameter of the fiber Bragg grating 1 is smaller than the inner diameter of the encapsulation tube 2, there is a cavity structure 4 between the fiber Bragg grating 1 and the encapsulation tube 2.

[0048] Step 3: Use the reference temperature sensor to perform static temperature calibration on the fiber Bragg grating sensor to obtain the wavelength-temperature relationship function of the fiber Bragg grating sensor.

[0049] In step 3, the wavelength-temperature relationship function reflects the unique mapping relationship between the center wavelength output by the fiber optic grating sensor and the measured temperature.

[0050] Specifically, such as Figure 3 As shown, in step 3, the fiber optic grating sensor is statically calibrated using the reference temperature sensor to obtain the wavelength-temperature relationship function of the fiber optic grating sensor. The steps are as follows: Step 31: Place the reference temperature sensor and the fiber optic grating sensor inside the same temperature control device.

[0051] In step 31, the temperature control device may be, but is not limited to, a tube furnace, a constant temperature chamber, or a dry well furnace. The reference temperature sensor and the fiber optic grating sensor should be positioned as close as possible within the temperature control device to ensure that they are in the same temperature field.

[0052] Step 32: Control the temperature control device to raise or lower the temperature to a certain constant temperature point, and keep it at the constant temperature point for a certain period of time so that the reference temperature sensor and the fiber optic grating sensor reach a stable state at the constant temperature point.

[0053] In step 32, the number and temperature of the constant temperature points can be preset in the temperature control device, or a temperature range and temperature change step size can be set in the temperature control device so that the temperature control device gradually increases or decreases the temperature from the initial temperature of the temperature range according to the temperature change step size.

[0054] The holding time at each constant temperature point can be determined according to the type of temperature control device and the time constant difference between the reference temperature sensor and the fiber optic grating sensor, as long as the reference temperature sensor and the fiber optic grating sensor can reach a stable state simultaneously.

[0055] The stable state refers to the state where the reference temperature output by the reference temperature sensor no longer changes, and the center wavelength output by the fiber optic grating sensor no longer changes.

[0056] Step 33: Collect the static reference temperature and static center wavelength corresponding to the reference temperature sensor and fiber optic grating sensor when they are in a stable state.

[0057] In step 33, the static reference temperature refers to the reference temperature output by the reference temperature sensor when it is in a stable state of static temperature calibration; the static center wavelength refers to the center wavelength output by the fiber optic grating sensor when it is in a stable state of static temperature calibration.

[0058] Step 34: Repeat steps 32 to 33 to sequentially collect the static reference temperature and static center wavelength corresponding to the reference temperature sensor and fiber optic grating sensor when they are in a stable state at different isothermal points.

[0059] Step 35: Take the static reference temperature and static center wavelength at the same isothermal point or the same sampling time as a set of calibration data pairs, and then fit each set of calibration data pairs to obtain the wavelength-temperature relationship function of the fiber optic grating sensor.

[0060] In step 35, the fitting method between the static reference temperature and the static center wavelength can be, but is not limited to, polynomial fitting, piecewise interpolation fitting, fitting based on deep learning networks, etc.

[0061] Step 4: Synchronously acquire the dynamic reference temperature sequence and dynamic wavelength sequence of the reference temperature sensor and fiber optic grating sensor under various dynamic temperature change conditions.

[0062] In step 4, the dynamic reference temperature sequence includes multiple dynamic reference temperatures, with different dynamic reference temperatures corresponding to different sampling times. The dynamic reference temperature refers to the reference temperature output by the reference temperature sensor when it is in a dynamically changing temperature environment and has not yet reached a stable state. Similarly, the dynamic wavelength sequence includes multiple dynamic center wavelengths, with different dynamic center wavelengths corresponding to different sampling times. The dynamic center wavelength refers to the center wavelength output by the fiber Bragg grating sensor when it is in a dynamically changing temperature environment and has not yet reached a stable state.

[0063] The dynamic reference temperature sequence is shown below. The dynamic wavelength sequence is shown below. Where t is the sampling time.

[0064] Specifically, such as Figure 4 As shown, in step 4, the steps for simultaneously acquiring the dynamic reference temperature sequence and dynamic wavelength sequence of the reference temperature sensor and fiber optic grating sensor under various dynamic temperature change conditions are as follows: Step 41: Place the reference temperature sensor and the fiber optic grating sensor inside the same temperature control device.

[0065] In step 41, the temperature control device may be, but is not limited to, a tube furnace, a constant temperature chamber, or a dry well furnace. The reference temperature sensor and the fiber optic grating sensor should be positioned as close as possible within the temperature control device to ensure that they are in the same temperature field.

[0066] Step 42: Set multiple dynamic temperature change conditions in the temperature control device so that the temperature control device can simulate various dynamic temperature change conditions in sequence to change the temperature quickly.

[0067] In step 42, the dynamic temperature change conditions include two or more of the following: step temperature change condition (sudden temperature jump), linear slope temperature change condition (uniform speed of temperature rise and fall), multi-segment variable speed temperature change condition (fast first then slow, slow first then fast), periodic temperature change fluctuation condition (small amplitude fluctuation of sine or triangular wave), and temperature trajectory playback condition of the measured object (such as the actual temperature curve of an engine or blast furnace).

[0068] Step 43: When the temperature control device simulates various dynamic temperature change conditions for rapid temperature change, the dynamic reference temperature and dynamic center wavelength of the reference temperature sensor and fiber optic grating sensor are collected synchronously at each sampling time according to the preset sampling frequency, so as to obtain the dynamic reference temperature sequence and dynamic wavelength sequence of the two under various dynamic temperature change conditions.

[0069] In step 43, the sampling frequency is between 10Hz and 1000Hz.

[0070] Step 5: Based on the dynamic reference temperature sequence and dynamic wavelength sequence, and combined with the wavelength-temperature relationship function, construct the dynamic thermal response feature data of the fiber optic grating sensor, and then use the dynamic thermal response feature data to train the temporal deep learning network to obtain the temporal residual compensation model.

[0071] In step 5, the dynamic thermal response feature data includes thermal response feature vectors and their residual labels at each sampling time.

[0072] Specifically, such as Figure 5As shown, in step 5, based on the dynamic reference temperature sequence and dynamic wavelength sequence, and combined with the wavelength-temperature relationship function, the dynamic thermal response feature data of the fiber optic grating sensor is constructed. Then, the dynamic thermal response feature data is used to train the temporal deep learning network to obtain the temporal residual compensation model. The steps are as follows: Step 51: Preprocess the dynamic reference temperature sequence and dynamic wavelength sequence to remove outliers, filter and denoise, and align timestamps.

[0073] In step 51, the method for removing outliers may include, but is not limited to, sliding window statistical method, Laida criterion method, median absolute deviation method, interquartile range method, physical constraint method, or density-based clustering method. The filtering and denoising method may include, but is not limited to, moving average filtering method, SG filtering method, wavelet thresholding denoising method, or Kalman filtering method. The timestamp alignment method may include, but is not limited to, timestamp linear interpolation method, cross-correlation-based time delay estimation method, or FFT-based frequency domain phase method.

[0074] Step 52: Based on the wavelength-temperature relationship function, convert the dynamic wavelength sequence into the dynamic fundamental temperature sequence of the fiber optic grating sensor.

[0075] In step 52, the dynamic base temperature sequence includes multiple dynamic base temperatures, with different dynamic base temperatures corresponding to different sampling times; each dynamic base temperature is calculated from the dynamic center wavelength at the corresponding sampling time through the wavelength-temperature relationship function.

[0076] The dynamic baseline temperature sequence is shown below. Where t represents different sampling times.

[0077] Step 53: Subtract the dynamic reference temperature and dynamic base temperature at the same sampling time from the dynamic reference temperature sequence and the dynamic base temperature sequence to obtain the temperature residual sequence between the reference temperature sensor and the fiber Bragg grating sensor. Subtract the dynamic center wavelengths at two adjacent sampling times in the dynamic wavelength sequence to obtain the first-order difference sequence of the dynamic wavelength of the fiber Bragg grating sensor. Subtract the dynamic base temperature at two adjacent sampling times in the dynamic base temperature sequence to obtain the first-order difference sequence of the dynamic temperature of the fiber Bragg grating sensor.

[0078] In step 53, the temperature residual sequence is as follows: in, ; The first-order difference sequence of the dynamic wavelength is shown below. in, ; The dynamic temperature first-order difference sequence is shown below. in, .

[0079] Step 54: For any sampling time, taking that sampling time as the endpoint, extract the local wavelength sequence, local temperature sequence, local wavelength first-order difference sequence, and local temperature first-order difference sequence within a preset time length from the dynamic wavelength sequence, dynamic base temperature sequence, dynamic wavelength first-order difference sequence, and dynamic temperature first-order difference sequence, respectively, to construct the thermal response feature vector corresponding to that sampling time, thereby obtaining the thermal response feature vectors at different sampling times.

[0080] In step 54, the thermal response feature vector corresponding to sampling time t is shown below. Where N is the total number of sampling points within the preset time length. This represents the set of all dynamic center wavelengths from sampling time t-N+1 to sampling time t. This represents the set of all dynamic baseline temperatures from sampling time t-N+1 to sampling time t. Let represent the set of all first-order differences of dynamic wavelengths from sampling time t-N+1 to sampling time t. It represents the set of all dynamic temperature first-order differences from sampling time t-N+1 to sampling time t.

[0081] Step 55: Using the temperature residual sequence as the training label set, label the thermal response feature vector at each sampling time with the temperature residual at the corresponding sampling time as the training label.

[0082] In step 55, the thermal response feature vector is the sampled time t. Determine the temperature residual at sampling time t in the temperature residual sequence. and the temperature residual As a thermal response feature vector The training labels are used to label each thermal response feature vector with the corresponding temperature residual as a training label.

[0083] Step 56: Divide the thermal response feature vectors at each sampling time into training datasets, validation datasets, and test datasets according to the different types of dynamic temperature change conditions.

[0084] In step 56, the thermal response feature vectors are first grouped according to the corresponding dynamic temperature change conditions. All thermal response feature vectors under the same dynamic temperature change condition are grouped into the same group. Then, the feature vector groups of each dynamic temperature change condition are assigned to one of the training dataset, the validation dataset, and the test dataset.

[0085] For example, assuming there are five dynamic temperature change conditions, namely condition A, condition B, condition C, condition D and condition E, the feature vectors of condition A, condition B and condition C can be assigned to the training dataset, the feature vectors of condition D can be assigned to the validation dataset, and the feature vectors of condition E can be assigned to the test dataset.

[0086] Step 57: Train the temporal deep learning network using the training dataset, and use the validation dataset to optimize the hyperparameters and judge the fitting effect of the temporal deep learning network during the training process. Finally, use the test dataset to test the actual performance and generalization ability of the trained temporal deep learning network to obtain the temporal residual compensation model.

[0087] In step 57, the temporal deep learning network is trained using the feature vector sets of operating conditions A, B, and C; the temporal deep learning network is optimized for hyperparameters and its fitting effect is judged using the feature vector set of operating condition D; and the temporal deep learning network is tested for actual performance and generalization ability using the feature vector set of operating condition E, in order to improve the model's generalization ability and enable the temporal residual compensation model to accurately identify the thermal response feature vectors of new operating conditions.

[0088] Step 6: Under actual measurement conditions, acquire the real-time wavelength sequence of the fiber Bragg grating sensor according to the preset sampling frequency.

[0089] In step 6, the sampling frequency is between 10Hz and 1000Hz.

[0090] The dynamic wavelength sequence includes multiple real-time center wavelengths, with different real-time center wavelengths corresponding to different sampling times. The real-time center wavelength refers to the center wavelength output in real time when the fiber optic grating sensor is not in a stable state under the actual measurement conditions.

[0091] The real-time wavelength sequence is shown below. Where t is the sampling time.

[0092] Step 7: Based on the real-time wavelength sequence and combined with the wavelength-temperature relationship function, construct the real-time thermal response characteristic data of the fiber optic grating sensor, and then input the real-time thermal response characteristic data into the time-series residual compensation model for inference, so as to obtain the residual compensation value at the current sampling time or the final compensation temperature after residual compensation as predicted by the time-series residual compensation model.

[0093] In step 7, the real-time thermal response feature data includes the thermal response feature vector at the current sampling time.

[0094] Specifically, such as Figure 6 As shown, in step 7, based on the real-time wavelength sequence and combined with the wavelength-temperature relationship function, the real-time thermal response characteristic data of the fiber optic grating sensor is constructed. Then, the real-time thermal response characteristic data is input into the time-series residual compensation model for inference to obtain the residual compensation value at the current sampling time predicted by the time-series residual compensation model, or the final compensated temperature after residual compensation. The steps are as follows: Step 71: Preprocess the real-time wavelength sequence to remove outliers and filter for noise reduction.

[0095] In step 71, the method for removing outliers may include, but is not limited to, sliding window statistical method, Laida criterion method, median absolute deviation method, interquartile range method, physical constraint method, or density-based clustering method. The filtering and denoising method may include, but is not limited to, moving average filtering method, SG filtering method, wavelet thresholding denoising method, or Kalman filtering method.

[0096] Step 72: Based on the wavelength-temperature relationship function, convert the real-time wavelength sequence into the real-time base temperature sequence of the fiber Bragg grating sensor.

[0097] In step 72, the real-time base temperature sequence includes multiple real-time base temperatures, with different real-time base temperatures corresponding to different sampling times; each real-time base temperature is calculated from the real-time center wavelength at the corresponding sampling time through the wavelength-temperature relationship function.

[0098] The real-time baseline temperature sequence is shown below. Where t represents different sampling times.

[0099] Step 73: Subtract the real-time center wavelengths of two adjacent sampling times in the real-time wavelength sequence to obtain the real-time wavelength first-order difference sequence of the fiber Bragg grating sensor; subtract the real-time base temperature of two adjacent sampling times in the real-time base temperature sequence to obtain the real-time temperature first-order difference sequence of the fiber Bragg grating sensor.

[0100] In step 73, the real-time wavelength first-order difference sequence is as follows: in, ; The real-time temperature first-order difference sequence is shown below. in, .

[0101] Step 74: Using the current sampling time as the endpoint, extract the historical wavelength sequence, historical temperature sequence, historical wavelength first-order difference sequence, and historical temperature first-order difference sequence within a preset time length from the real-time wavelength sequence, real-time base temperature sequence, real-time wavelength first-order difference sequence, and real-time temperature first-order difference sequence, respectively, to construct the thermal response feature vector at the current sampling time.

[0102] In step 74, corresponding to the current sampling time The thermal response feature vector is shown below. Where N is the total number of sampling points within the preset time length. Indicates from the sampling time Up to the current sampling time The set of all real-time center wavelengths, Indicates from the sampling time Up to the current sampling time The collection of all real-time baseline temperatures, Indicates from the sampling time Up to the current sampling time The set of all real-time wavelength first-order differences, Indicates from the sampling time Up to the current sampling time The set of all real-time temperature first-order differences.

[0103] Step 75: Input the thermal response feature vector at the current sampling time into the time-series residual compensation model for inference.

[0104] In step 75, the thermal response feature vectors used for inference in the time-series residual compensation model do not need to be labeled with residual tags.

[0105] Step 76: Receive the residual compensation value at the current sampling time or the final compensated temperature after residual compensation, which is predicted and output by the time-series residual compensation model.

[0106] In step 76, the current sampling time The final compensated temperature after residual compensation is shown below. in, The current sampling time Real-time base temperature, The current sampling time The residual compensation value.

[0107] Although the time-series residual compensation model actually predicts the current sampling time. residual compensation value However, due to the current sampling time thermal response feature vector The current sampling time is already included. Real-time base temperature The current sampling time can be output by setting a corresponding output function in the output layer of the time-series residual compensation model. residual compensation value Or the final compensated temperature after residual compensation .

[0108] If the time-series residual compensation model ultimately outputs the current sampling time... residual compensation value If so, the output function is as follows: If the time-series residual compensation model ultimately outputs the current sampling time... Final compensated temperature after residual compensation If so, the output function is as follows: .

[0109] The temporal deep learning network may, but is not limited to, employ a GRU gated recurrent unit network, an LSTM long short-term memory network, a TCN causal temporal convolutional network, a Transformer neural network, a CNN-LSTM hybrid neural network, a CNN-GRU neural network, or a hybrid neural network with an attention mechanism.

[0110] This embodiment uses a hybrid neural network consisting of causal convolution, gated recurrent units, and fully connected residual output as an example to further illustrate the temporal deep learning network. Figure 7 As shown, the TCN causal temporal convolutional network adopts a multi-layer architecture consisting of causal convolutional layers, recurrent layers, and fully connected layers, and its specific structure is as follows: First layer: Input layer This layer is used to input the thermal response feature vector.

[0111] Second layer: One-dimensional causal convolutional layer This layer is used to extract short-time local features from the thermal response feature vector, such as the instantaneous change patterns of wavelength, base temperature, and rate of change during sudden temperature rises or falls. It employs a one-dimensional causal convolution kernel with a size of 3–7 time steps and 16–64 output channels. Causal convolution ensures that the output at the current sampling moment depends only on the input at the current and historical sampling moments, without using any information from future sampling moments, thus satisfying the causality constraints of online real-time inference.

[0112] Third layer: Normalization layer and nonlinear activation layer This layer is used to perform batch normalization or layer normalization on the feature maps output by the second layer to accelerate model convergence and improve training stability. Subsequently, nonlinear expressive power is introduced through nonlinear activation functions (such as ReLU or LeakyReLU), enabling the network to fit complex dynamic error patterns.

[0113] Fourth layer: The second one-dimensional causal convolutional layer This layer is used to further extract higher-order, more abstract local temporal relationships from short-term features, enhancing the model's robustness to noise and sensitivity to subtle changes in dynamic signals. Its kernel size and number of channels can be the same as or appropriately increased as the first layer, so as to stack one-dimensional causal convolutions again on the basis of the first set of "convolution + activation".

[0114] Fifth layer: One or more gated recurrent units or long short-term memory networks This layer is used to input the local feature sequences extracted from the fourth layer into the recurrent neural network layer. It employs a gated recurrent unit (GRU) or a long short-term memory (LSTM) network to effectively capture thermal inertia dependence, hysteresis effects, and nonlinear dynamic evolution patterns over longer time scales through its internal memory units and gating mechanisms. This layer can be configured as a single layer or multiple stacked layers; the number of layers and hidden units can be adjusted according to the scale of the training data and the complexity of the dynamic operating conditions.

[0115] Sixth layer: Fully connected layer This layer maps the temporal feature vectors output from the fifth layer to a high-dimensional space, performing nonlinear combination and dimensionality reduction. This layer contains several neurons (e.g., 32-128) and is used with dropout or L2 regularization to prevent overfitting and improve the model's generalization ability.

[0116] Seventh layer: Output layer This layer is a fully connected layer with no activation function (linear activation) and outputs the residual compensation value. Depending on the application requirements, this layer can be configured with an appropriate output function to output the residual compensation amount at the current sampling time or the final compensated temperature after residual compensation.

[0117] In this embodiment, the loss function used when training the temporal deep learning network is as follows: in, These are the temperature residual values ​​predicted by the model. This represents the actual measured temperature residual value. Mean square error, The mean absolute error, To smooth the constraint regularization term, , and These are the weighting coefficients.

[0118] Weighting coefficient , and Initial values ​​can be set by technicians based on experience, and then gradually optimized based on the training results during model training.

[0119] Example 2 A deep learning algorithm module includes a processor and a memory. The memory stores a computer program and a timing residual compensation model for the processor to execute and call. When the processor executes the computer program, it calls the timing residual compensation model and performs steps 4 to 7 in the dynamic temperature calibration method described in Embodiment 1.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention and not to limit them. Although the embodiments of the present invention have been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the embodiments of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic temperature calibration of fiber grating based on thermal response matching and deep learning, characterized in that, Includes the following steps: Step 1: Using the effective thermal response speed of the reference temperature sensor as the standard, optimize the packaging parameters of the fiber Bragg grating sensor using software simulation to obtain the optimal parameter combination when the thermal response times of the two sensors tend to be consistent. Step 2: Fabricate a fiber Bragg grating and encapsulate the fiber Bragg grating using an encapsulation structure corresponding to the optimal parameter combination to obtain the fiber Bragg grating sensor; Step 3: Use the reference temperature sensor to perform static temperature calibration on the fiber Bragg grating sensor to obtain the wavelength-temperature relationship function of the fiber Bragg grating sensor; Step 4: Synchronously acquire the dynamic reference temperature sequence and dynamic wavelength sequence of the reference temperature sensor and fiber optic grating sensor under various dynamic temperature change conditions; Step 5: Based on the dynamic reference temperature sequence and dynamic wavelength sequence, and combined with the wavelength-temperature relationship function, construct the dynamic thermal response feature data of the fiber optic grating sensor, and then use the dynamic thermal response feature data to train the temporal deep learning network to obtain the temporal residual compensation model. Step 6: Under actual measurement conditions, acquire the real-time wavelength sequence of the fiber Bragg grating sensor according to the preset sampling frequency; Step 7: Based on the real-time wavelength sequence and combined with the wavelength-temperature relationship function, construct the real-time thermal response characteristic data of the fiber optic grating sensor, and then input the real-time thermal response characteristic data into the time-series residual compensation model for inference, so as to obtain the residual compensation value at the current sampling time or the final compensation temperature after residual compensation as predicted by the time-series residual compensation model.

2. The fiber grating dynamic temperature calibration method of claim 1, wherein, In step 1, using the effective thermal response speed of the reference temperature sensor as a standard, the packaging parameters of the fiber Bragg grating sensor are optimized using software simulation to obtain the optimal parameter combination when the thermal response times of the two sensors tend to be consistent. The steps are as follows: Step 11: Obtain the effective thermal response time constant of the reference temperature sensor; Step 12: Based on the packaging structure of the fiber Bragg grating sensor, construct a three-dimensional model of the packaging structure in simulation software; Step 13: In the simulation software, set the heat transfer path and multiple packaging parameters related to transient heat conduction analysis for the packaging structure, so that the simulation software can perform transient heat conduction analysis on the packaging structure according to the heat transfer path and each packaging parameter, thereby calculating the simulated thermal response time constant of the fiber Bragg grating sensor. Step 14: With the goal of minimizing the time constant difference between the effective thermal response time constant and the simulated thermal response time constant, construct an objective optimization function in the simulation software, and select several encapsulated parameters as variable parameters. Set a corresponding range of variation for each variable parameter so that the simulation software can iteratively optimize the numerical combination of each variable parameter, thereby obtaining the optimal parameter combination of each variable parameter when the time constant difference is minimized.

3. The fiber grating dynamic temperature calibration method of claim 2, wherein, In step 11, the effective thermal response time constant of the reference temperature sensor is obtained through experimental measurement or table lookup.

4. The fiber grating dynamic temperature calibration method of claim 1, wherein, In step 3, the fiber Bragg grating sensor is statically calibrated using the reference temperature sensor to obtain the wavelength-temperature relationship function of the fiber Bragg grating sensor. The steps are as follows: Step 31: Place the reference temperature sensor and the fiber optic grating sensor inside the same temperature control device; Step 32: Control the temperature control device to raise or lower the temperature to a certain constant temperature point, and keep it at the constant temperature point for a certain period of time so that the reference temperature sensor and the fiber optic grating sensor reach a stable state at the constant temperature point. Step 33: Collect the static reference temperature and static center wavelength corresponding to the reference temperature sensor and fiber optic grating sensor when they are in a stable state, respectively; Step 34: Repeat steps 32 to 33 to sequentially collect the static reference temperature and static center wavelength of the reference temperature sensor and fiber optic grating sensor when they are in a stable state at different isothermal points. Step 35: Take the static reference temperature and static center wavelength at the same isothermal point or the same sampling time as a set of calibration data pairs, and then fit each set of calibration data pairs to obtain the wavelength-temperature relationship function of the fiber optic grating sensor.

5. The fiber grating dynamic temperature calibration method of claim 1, wherein, In step 4, the steps for simultaneously acquiring the dynamic reference temperature sequence and dynamic wavelength sequence of the reference temperature sensor and fiber optic grating sensor under various dynamic temperature change conditions are as follows: Step 41: Place the reference temperature sensor and the fiber optic grating sensor inside the same temperature control device; Step 42: Set multiple dynamic temperature change conditions in the temperature control device so that the temperature control device can simulate various dynamic temperature change conditions in sequence to change the temperature quickly; Step 43: When the temperature control device simulates various dynamic temperature change conditions for rapid temperature change, the dynamic reference temperature and dynamic center wavelength of the reference temperature sensor and fiber optic grating sensor are collected synchronously at each sampling time according to the preset sampling frequency, so as to obtain the dynamic reference temperature sequence and dynamic wavelength sequence of the two under various dynamic temperature change conditions.

6. The fiber optic grating dynamic temperature calibration method according to claim 1 or 4, characterized in that, The dynamic temperature change conditions include two or more of the following: step temperature change condition, linear slope temperature change condition, multi-segment variable speed temperature change condition, periodic temperature change fluctuation condition, and temperature trajectory playback condition of the measured object.

7. The fiber grating dynamic temperature calibration method of claim 1, wherein, The dynamic thermal response feature data includes thermal response feature vectors and their residual labels at each sampling time. In step 5, based on the dynamic reference temperature sequence and dynamic wavelength sequence, and combined with the wavelength-temperature relationship function, the dynamic thermal response feature data of the fiber optic grating sensor is constructed. The following steps are taken to train the temporal deep learning network using the dynamic thermal response feature data to obtain the temporal residual compensation model: Step 51: Preprocess the dynamic reference temperature sequence and dynamic wavelength sequence to remove outliers, filter and denoise, and align timestamps; Step 52: Based on the wavelength-temperature relationship function, convert the dynamic wavelength sequence into the dynamic fundamental temperature sequence of the fiber Bragg grating sensor; Step 53: Subtract the dynamic reference temperature and dynamic base temperature at the same sampling time from the dynamic reference temperature sequence and the dynamic base temperature sequence to obtain the temperature residual sequence between the reference temperature sensor and the fiber Bragg grating sensor. Subtract the dynamic center wavelengths at two adjacent sampling times in the dynamic wavelength sequence to obtain the first-order difference sequence of the dynamic wavelength of the fiber Bragg grating sensor. Subtract the dynamic base temperature at two adjacent sampling times in the dynamic base temperature sequence to obtain the first-order difference sequence of the dynamic temperature of the fiber Bragg grating sensor. Step 54: For any sampling time, taking that sampling time as the endpoint, extract the local wavelength sequence, local temperature sequence, local wavelength first-order difference sequence, and local temperature first-order difference sequence within a preset time length from the dynamic wavelength sequence, dynamic base temperature sequence, dynamic wavelength first-order difference sequence, and dynamic temperature first-order difference sequence, respectively, to construct the thermal response feature vector corresponding to that sampling time, thereby obtaining the thermal response feature vectors at different sampling times; Step 55: Using the temperature residual sequence as the training label set, label the thermal response feature vector at each sampling time with the temperature residual at the corresponding sampling time as the training label; Step 56: Divide the thermal response feature vectors at each sampling time into training datasets, validation datasets, and test datasets according to the different types of dynamic temperature change conditions; Step 57: Train the temporal deep learning network using the training dataset, and use the validation dataset to optimize the hyperparameters and judge the fitting effect of the temporal deep learning network during the training process. Finally, use the test dataset to test the actual performance and generalization ability of the trained temporal deep learning network to obtain the temporal residual compensation model.

8. The fiber grating dynamic temperature calibration method of claim 1, wherein, The real-time thermal response feature data includes the thermal response feature vector at the current sampling time. In step 7, based on the real-time wavelength sequence and combined with the wavelength-temperature relationship function, the real-time thermal response feature data of the fiber optic grating sensor is constructed. Then, the real-time thermal response feature data is input into the time-series residual compensation model for inference to obtain the residual compensation value at the current sampling time predicted by the time-series residual compensation model or the final compensation temperature after residual compensation. The steps are as follows: Step 71: Preprocess the real-time wavelength sequence to remove outliers and filter for noise reduction; Step 72: Based on the wavelength-temperature relationship function, convert the real-time wavelength sequence into the real-time base temperature sequence of the fiber Bragg grating sensor; Step 73: Subtract the real-time center wavelengths of two adjacent sampling times in the real-time wavelength sequence to obtain the real-time wavelength first-order difference sequence of the fiber Bragg grating sensor; subtract the real-time base temperature of two adjacent sampling times in the real-time base temperature sequence to obtain the real-time temperature first-order difference sequence of the fiber Bragg grating sensor. Step 74: Taking the current sampling time as the endpoint, extract the historical wavelength sequence, historical temperature sequence, historical wavelength first-order difference sequence, and historical temperature first-order difference sequence within a preset time length from the real-time wavelength sequence, real-time base temperature sequence, real-time wavelength first-order difference sequence, and real-time temperature first-order difference sequence, respectively, to construct the thermal response feature vector at the current sampling time. Step 75: Input the thermal response feature vector at the current sampling time into the time-series residual compensation model for inference; Step 76: Receive the residual compensation value at the current sampling time or the final compensated temperature after residual compensation, which is predicted and output by the time-series residual compensation model.

9. The fiber grating dynamic temperature calibration method of claim 1, wherein, The loss function used when training the temporal deep learning network is shown below. Among them, among them, These are the temperature residual values ​​predicted by the model. This represents the actual measured temperature residual value. Mean square error, The mean absolute error, To smooth the constraint regularization term, , and These are the weighting coefficients.

10. A deep learning algorithm module comprising a processor and a memory, characterized in that, The memory stores a computer program and a timing residual compensation model for the processor to execute and call. When the processor executes the computer program, it calls the timing residual compensation model and performs steps 4 to 7 in the dynamic temperature calibration method of claim 1.

Images