Gas physical property parameter extraction method based on dynamic characteristics of micro thermal field

Abstract

The invention relates to the technical field of gas parameter extraction, in particular to a gas physical property parameter extraction method based on micro thermal field dynamic characteristics, which comprises the following steps: applying a sweep frequency excitation electric signal to a micro heater of an MEMS sensor, collecting and processing an original dynamic response signal, and generating frequency domain response data; constructing a parameterized extraction model, extracting real-time physical property parameters from the frequency domain response data, substituting the real-time physical property parameters into a signal-to-noise ratio objective function, and calculating and generating optimized excitation parameters; generating an updated sweep frequency excitation electric signal based on the optimized excitation parameter; acquiring and generating reference frequency domain response data; actually measured physical property parameters are extracted from the reference frequency domain response data, a measurement deviation signal is generated through comparison, and online calibration and correction are conducted on internal coefficients of the parameterized extraction model. According to the method, the frequency domain response of the dynamic micro thermal field is subjected to decoupling analysis, and an online self-adaption and self-correction mechanism is combined, so that accurate and reliable online extraction of the physical property parameters of the complex mixed gas is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of gas parameter extraction, in particular to a gas physical property parameter extraction method based on micro-thermal field dynamic characteristics. BACKGROUND

[0002] In the field of gas analysis and extraction, especially in specific scenarios such as city gas and industrial gas metering, there are high requirements for the precision, robustness and cost of gas physical property parameter extraction.

[0003] In the prior art, traditional physicochemical analysis methods such as gas chromatography have complex equipment and high cost, which cannot meet the needs of real-time online monitoring and low-cost large-scale deployment of city gas metering.

[0004] Another is a MEMS gas sensor based on thermal principle. This kind of sensor uses the heat dissipation characteristics of micro electric heating device to indirectly measure the thermal conductivity, specific heat capacity and other physical properties of gas by sensing the heat exchange change of gas in static or dynamic state. These methods provide a new way for high-precision gas metering.

[0005] However, in practical application of city gas and industrial metering, the existing technology still faces the following challenges: the heat conduction process of gas at micron scale is complex, the existing method generally uses a simplified physical model, resulting in significant errors in the inversion of physical property parameters; the sensor is highly sensitive to environmental temperature, pressure and humidity, which easily causes the measurement result to drift and affects the reliability of the metering data; the manufacturing fluctuations of MEMS devices cause large differences in the performance of different sensor individuals, increasing the standardization and calibration cost of batch production; real-time decoupling of physical property parameters from complex transient thermal response signals requires strong data processing capability, and traditional algorithms are difficult to meet the needs of fast-paced, online measurement and extraction.

[0006] Therefore, the present application proposes a gas physical property parameter extraction method based on micro-thermal field dynamic characteristics. SUMMARY

[0007] The purpose of the present application is to provide a gas physical property parameter extraction method based on micro-thermal field dynamic characteristics, which realizes accurate and reliable online extraction of complex mixed gas physical property parameters by decoupling and analyzing the frequency domain response of dynamic micro-thermal field and combining online adaptive and self-correcting mechanisms.

[0008] To achieve the above purpose, the present application provides the following technical solutions: The gas physical property parameter extraction method based on micro-thermal field dynamic characteristics comprises: The micro-heater of the MEMS sensor in the cavity is applied with a sweep excitation electrical signal to establish dynamic heat exchange in the mixed natural gas to be measured, and the original dynamic response signal is synchronously collected; the original dynamic response signal is subjected to signal transformation to generate frequency domain response data; A parameterized extraction model is constructed based on the physical principle of heat transfer, and real-time physical property parameters are extracted from the frequency domain response data based on different frequency domain characteristics; The real-time physical property parameters are substituted into a signal-to-noise ratio target function to calculate and generate optimized excitation parameters for frequency domain response characteristic parameter separation; and based on the optimized excitation parameters, an updated sweep excitation electrical signal is generated to replace the current sweep excitation electrical signal; The reference frequency domain response data of the reference natural gas with known standard physical property parameters are collected and generated; the measured physical property parameters are compared with the known standard physical property parameters to generate a measurement deviation signal, and the internal coefficients of the parameterized extraction model are online calibrated and corrected.

[0009] Preferably, the micro-heater of the MEMS sensor in the cavity is applied with a sweep excitation electrical signal to establish dynamic heat exchange in the mixed natural gas to be measured, and the original dynamic response signal is synchronously collected; the micro-heater is applied with a sinusoidal sweep excitation electrical signal with constant voltage amplitude, the frequency of the excitation electrical signal is continuously scanned in a logarithmic manner between a preset starting frequency and a preset ending frequency, a dynamic micro-thermal field with a heat penetration depth varying with the frequency of the excitation electrical signal is established in the mixed natural gas to be measured; and through a plurality of temperature measuring elements symmetrically arranged with the micro-heater, the original dynamic response signal containing amplitude attenuation and phase delay information generated after dynamic heat exchange between the dynamic micro-thermal field and the mixed natural gas to be measured is synchronously collected.

[0010] Preferably, the dynamic micro-thermal field includes: the dynamic micro-thermal field is an unsteady temperature field with temperature distribution and heat flux density periodically changing with time and space position, which is established in the mixed natural gas to be measured with the micro-heater as the center; the heat penetration depth of the dynamic micro-thermal field changes with the frequency of the sweep excitation electrical signal, wherein the heat penetration depth generated by the high-frequency excitation component is limited in the near-surface thermal diffusion zone of the MEMS sensor, and the heat penetration depth generated by the low-frequency excitation component is sufficient to exchange heat with the main body of the gas far away from the surface of the sensor.

[0011] Preferably, the generating of the frequency domain response data comprises: performing a synchronous cross-correlation operation on the collected original dynamic response signal and the corresponding swept excitation electrical signal; performing a fast Fourier transform on the response signal after the synchronous cross-correlation operation to generate a frequency domain complex signal containing fundamental wave and harmonic components; and separating amplitude-frequency response data and phase-frequency response data from the frequency domain complex signal, wherein the frequency domain response data is a collection of the amplitude-frequency response data and the phase-frequency response data.

[0012] Preferably, the constructing of the parameterized extraction model comprises: establishing a modified theoretical physical model based on a three-dimensional unsteady heat conduction equation describing the dynamic micro-thermal field, introducing a correction coefficient for characterizing the rarefied gas effect, and combining the geometric structure and material properties of the MEMS sensor; obtaining a theoretical frequency domain response data set under a plurality of known physical property parameter combinations through numerical analysis of the theoretical physical model; and fitting the parameterized extraction model through an identification method using the theoretical frequency domain response data set, wherein the parameterized extraction model establishes a nonlinear mapping relationship between key features of the frequency domain response data and the physical property parameters.

[0013] Preferably, the real-time physical property parameters include: a real-time comprehensive thermal diffusivity extracted based on amplitude attenuation characteristics in a high-frequency region of the frequency domain response data, and a complex thermal capacity parameter extracted based on phase delay characteristics in a low-frequency region; and the real-time density of the mixed natural gas is calculated using a gas state equation in combination with the pipeline pressure and temperature data obtained in real time from the city natural gas gate station, and the equivalent thermal conductivity and equivalent specific heat capacity of the mixed natural gas are calculated using the real-time comprehensive thermal diffusivity, the complex thermal capacity parameter, and the real-time density according to thermal physical relationships.

[0014] Preferably, the process of substituting the real-time physical property parameters into a signal-to-noise ratio objective function comprises: substituting the extracted real-time physical property parameters into a signal-to-noise ratio objective function, wherein the signal-to-noise ratio objective function represents the signal-to-noise ratio of the frequency domain response data in a key feature frequency band; and determining an optimal frequency bandwidth, sweep rate, and signal amplitude combination of the swept excitation electrical signal through an iterative optimization process with the evaluation function value as the convergence target, wherein the combination is the optimized excitation parameter.

[0015] Preferably, the process of online calibration and correction of internal coefficients of the parameterized extraction model comprises: the parameterized extraction model is an artificial neural network model, and the weights and bias coefficients of the artificial neural network model are adjusted through back propagation using the measurement deviation signal, so that the artificial neural network model outputs an approximation of the known standard physical property parameters when the reference frequency domain response data is input.

[0016] Preferably, the method further comprises: extracting and storing a baseline fingerprint representing the initial health state of the MEMS sensor from the baseline frequency domain response data measured for the first time under the baseline natural gas; extracting a real-time fingerprint from the real-time frequency domain response data regenerated under the baseline natural gas, comparing the real-time fingerprint with the baseline fingerprint to calculate a fingerprint drift, triggering a model correction function limit when the drift exceeds a preset threshold, and outputting an alarm signal indicating that the current extraction result has reduced reliability.

[0017] Preferably, the real-time fingerprint and the baseline fingerprint are a multi-dimensional feature vector containing normalized energy ratios and relative phase differences extracted from the fundamental wave and a plurality of detectable harmonic components of the frequency domain response data.

[0018] Compared with the prior art, the present application has the following beneficial effects: 1. The present application uses the physical separation characteristics of different physical parameters in the frequency domain response, especially the differences between the high-frequency amplitude characteristics and the low-frequency phase characteristics, to fundamentally solve the technical problem of high coupling between the thermal conductivity and the specific heat capacity in the traditional time domain analysis, which is difficult to accurately separate, thereby significantly improving the extraction accuracy of the physical parameters.

[0019] 2. The present application can adaptively adjust the excitation signal for subsequent measurement according to the real-time extracted physical parameters, so that it always works in the optimal measurement state for the current gas to be measured. This makes the present method capable of adapting to the large-scale dynamic changes of hydrogen and other components in natural gas, overcoming the "component applicability" bottleneck caused by the dependence on fixed models in the prior art.

[0020] 3. The present application can continuously compensate for the measurement drift caused by individual differences of MEMS sensors, long-term use aging or environmental changes, thereby ensuring the long-term effectiveness of the extraction model. In addition, by introducing the "fingerprint" self-diagnosis function, the reliability of the extraction process can be verified online, realizing the transition from "passive measurement" to "active state perception", and greatly improving the reliability of the measurement results. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The flowchart of the gas physical parameter extraction method based on the micro-thermal field dynamic characteristics of the present application; Figure 2 The schematic diagram of the frequency-dependent thermal penetration depth and parameter separation principle of the embodiment of the present application; Figure 3 The schematic diagram of the fingerprint self-diagnosis function of the embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical scheme and advantages of the present application clearer, further detailed description will be made to the present application in combination with the drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application. Other embodiments obtained by those skilled in the art based on the idea in the present specification without creative labor fall within the protection scope of the present application.

[0023] With reference to Figures 1 to 3 , the present application provides a gas physical property parameter extraction method based on micro-thermal field dynamic characteristics. The method of the present application is realized by a measuring device comprising a MEMS sensor, which is packaged in a static diffusion cavity. As a preferred embodiment, the static diffusion cavity is made of a material with low thermal conductivity (for example, polyether ether ketone PEEK), and the internal volume is not less than 5 cubic centimeters to ensure that the heat generated by the sensor during operation will not cause significant changes in the overall temperature of the gas in the cavity. Before measurement, the mixed natural gas to be measured is first purged and replaced, then all gas valves are closed, and after a static stabilization period of at least 30 seconds, the sweep excitation signal is applied. This process aims to eliminate macroscopic flow disturbances. In the micro-scale system of the present application, heat transfer dominated by thermal conduction is much greater than natural convection effect, so the physical model based on pure heat conduction is sufficient to ensure measurement accuracy.

[0024] As a preferred embodiment, the reference gas can be methane or methane / nitrogen mixed gas certified by the national standard, and the reference measurement can be repeated 5-10 times and the average value is taken to improve the calibration accuracy.

[0025] Embodiment 1: With reference to Figure 1 , the gas physical property parameter extraction method based on micro-thermal field dynamic characteristics comprises: In the static diffusion cavity into which the mixed natural gas to be measured from the city natural gas gate station is introduced, a sweep excitation electrical signal is applied to the micro-heater of the MEMS sensor in the cavity to establish dynamic heat exchange in the mixed natural gas to be measured, and the original dynamic response signal is synchronously collected; the original dynamic response signal is subjected to signal transformation to generate frequency domain response data; A parameterized extraction model is constructed based on the physical principles of heat transfer, and real-time physical property parameters are extracted from the frequency domain response data based on different frequency domain characteristics; The real-time physical property parameters are substituted into a signal-to-noise ratio target function to calculate and generate optimized excitation parameters for frequency domain response characteristic parameter separation; and based on the optimized excitation parameters, an updated sweep excitation electrical signal is generated to replace the current sweep excitation electrical signal; Collect and generate the benchmark frequency domain response data of the benchmark natural gas with known standard physical parameters; extract the measured physical parameters from the benchmark frequency domain response data, compare with the known standard physical parameters to generate a measurement deviation signal, and perform online calibration correction on the internal coefficients of the parameterized extraction model.

[0026] Further, the frequency-sweep excitation electrical signal is applied to the micro-heater of the MEMS sensor in the cavity, a dynamic heat exchange is established in the mixed natural gas to be measured, and the original dynamic response signal is synchronously collected, including: applying a sine wave frequency-sweep excitation electrical signal with constant voltage amplitude to the micro-heater, the frequency of the excitation electrical signal is continuously swept in a logarithmic manner between a preset starting frequency and a preset ending frequency, a dynamic micro-heat field with a heat penetration depth varying with the frequency of the excitation electrical signal is established in the mixed natural gas to be measured; and the original dynamic response signal containing amplitude attenuation and phase delay information generated after the dynamic micro-heat field and the mixed natural gas to be measured perform dynamic heat exchange is synchronously collected by the plurality of temperature measuring elements symmetrically arranged with the micro-heater.

[0027] Specifically, in a preset stationary diffusion cavity filled with the mixed natural gas to be measured, the signal processing unit applies an accurately controlled frequency-sweep excitation electrical signal to the central micro-heater of the MEMS sensor located in the cavity, and the electrical signal specifically exhibits a sine wave form with constant voltage amplitude, for example, the peak-to-peak value can be set to 1 volt. The frequency of the signal is not fixed, but is swept in a logarithmic manner within a preset frequency range, for example, from 1 Hz to 1000 Hz (preferably, 10-40 sampling points are taken at each octave, the dwell time of each point is 50-200 ms, and the sampling rate is not less than 20 kHz to ensure the frequency domain resolution and phase accuracy), and the signal is continuously excited for at least five signal periods at each frequency point to ensure that the dynamic heat field reaches a steady-state alternating current response and to ensure that there are sufficient frequency points distributed in the low frequency band and the high frequency band.

[0028] When the frequency-sweep excitation electrical signal is applied to the micro-heater, the thermal power generated by the micro-heater also periodically changes at the same frequency, thereby establishing a dynamic micro-heat field in the mixed natural gas medium to be measured around the micro-heater. A key physical property of the micro-heat field is that the heat penetration depth is negatively correlated with the frequency of the excitation signal. Specifically, when the excitation signal is in the high frequency band, the effective distance of the heat wave propagating in the gas is short, and the influence range is limited to the thin layer of gas close to the surface of the sensor chip; on the contrary, when the frequency of the excitation signal gradually decreases, the heat wave has a longer time period to diffuse outward, and the heat penetration depth significantly increases, enabling sufficient dynamic heat exchange with the main body of the gas farther away from the chip surface.

[0029] In this excitation process, multiple groups of temperature measuring elements symmetrically arranged around the micro-heater are used to monitor the temperature fluctuations in real time at their spatial positions; due to the inherent energy attenuation and time lag in heat conduction in the gas, the temperature response signals collected by these temperature measuring elements will exhibit clear amplitude attenuation and phase delay characteristics when compared with the original excitation electrical signal. The original electrical signal set collected by all temperature measuring elements, which contains amplitude and phase information, collectively constitutes the original dynamic response signal and is transmitted synchronously to the signal processing unit for subsequent processing.

[0030] By applying a sweep excitation to build a frequency-dependent controllable dynamic micro-thermal field in the gas, amplitude and phase composite information across different thermal penetration depths can be obtained from a single excitation process. This method greatly enriches the data dimension for physical property parameter identification, laying a solid data foundation for subsequent high-precision separation and extraction of multiple coupled thermal physical parameters such as thermal conductivity and specific heat capacity.

[0031] Further, with reference to Figure 2 , the dynamic micro-thermal field includes: the dynamic micro-thermal field is a non-steady temperature field centered on the micro-heater, established in the mixed natural gas to be measured, with periodic changes in temperature distribution and heat flux density over time and spatial position; the thermal penetration depth of the dynamic micro-thermal field changes with the frequency of the sweep excitation electrical signal, wherein the thermal penetration depth generated by high-frequency excitation components (such as Figure 2 shown on the left) is limited to the near-surface thermal diffusion region of the MEMS sensor, and the response characteristics in this region are mainly determined by the integrated thermal diffusivity; while the thermal penetration depth generated by low-frequency excitation components (such as Figure 2 shown on the right) is sufficient to exchange heat with the main body of the gas away from the sensor surface, and the response characteristics in this region are mainly determined by the composite heat capacity parameter.

[0032] Specifically, the micro-thermal field established in the mixed natural gas to be measured by the sweep excitation electrical signal is essentially a three-dimensional non-steady temperature field with the micro-heater as the geometric center and radiating outward. The "non-steady" is reflected in the fact that the temperature value and heat flux density at any point in the field domain are not constant but periodically fluctuate over time, and the fluctuation frequency is strictly synchronized with the frequency of the applied excitation electrical signal.

[0033] The most important feature of the dynamic micro-thermal field is its frequency dependence of thermal penetration depth, which defines the spatial range over which the thermal wave can effectively propagate and cause significant temperature fluctuations. In the present embodiment, when the frequency of the swept-frequency excitation electrical signal is in the high frequency range, for example, above several hundred hertz, the thermal wave decays before it has time to propagate far in the gas medium due to the extremely short cycle of heating and cooling, and thus the thermal penetration depth generated is very shallow. At this time, the heat exchange process is mainly limited to a thin layer of region near the solid surface of the MEMS sensor chip, i.e. the near-surface thermal diffusion zone.

[0034] On the contrary, when the frequency of the excitation electrical signal is reduced to the low frequency range, for example, several or tens of hertz, the duration of a complete heating and cooling cycle becomes longer, which allows sufficient time for heat to diffuse from the micro-heater to a more distant spatial region. Therefore, the thermal penetration depth generated by the low-frequency excitation component is sufficient to penetrate the near-surface region and reach the gas bulk portion which is not affected by the sensor solid boundary, and to perform sufficient dynamic heat exchange with it. In this way, a complete frequency scanning process realizes the layer-by-layer detection of the thermal physical properties of different spatial levels from the sensor surface to the deep gas bulk.

[0035] The present embodiment realizes the layer-by-layer detection of the thermal physical processes of the near-field and far-field of the gas by precisely controlling the thermal penetration depth with frequency, which provides a key physical basis for subsequent separation of the boundary effect interference of the sensor structure and accurate extraction of the intrinsic property parameters of the gas bulk, thereby effectively improving the accuracy and reliability of the final measurement results.

[0036] Further, the generation of the frequency domain response data includes: performing synchronous cross-correlation operation on the collected original dynamic response signal and the corresponding swept-frequency excitation electrical signal; performing fast Fourier transform on the response signal after the synchronous cross-correlation operation to generate a frequency domain complex signal containing fundamental wave and harmonic components; separating the amplitude frequency response data and the phase frequency response data from the frequency domain complex signal, and the frequency domain response data is the set of the amplitude frequency response data and the phase frequency response data.

[0037] Specifically, after receiving the original dynamic response signal from the temperature measuring element and the swept-frequency excitation electrical signal synchronized therewith, the signal processing unit immediately starts a series of digital signal processing steps. The processing unit performs synchronous cross-correlation operation on the two groups of time domain signals, and the operation takes the excitation signal as a reference to enhance and extract the response signal component homologous to the excitation from the background noise.

[0038] After the cross-correlation operation is completed, a new time-domain response signal after noise reduction is obtained, and the signal processing unit performs fast Fourier transform (FFT) on the optimized time-domain signal. The transform converts the signal from the time domain to the frequency domain to generate a frequency-domain complex signal composed of a series of complex numbers. The frequency-domain signal not only contains the fundamental wave component with the same frequency as the excitation frequency, but also contains multiple harmonic components generated by the nonlinear characteristics of the thermal physical process. These harmonic components also carry information about the physical properties of the gas to be measured.

[0039] For each frequency point in the frequency domain, the processing unit obtains the amplitude response and phase response at the frequency point by calculating the modulus and argument of the corresponding complex number. The amplitude response data corresponding to all scanning frequency points are collected to form the amplitude-frequency response data. Similarly, the phase response data corresponding to all frequency points are collected to form the phase-frequency response data. The frequency-domain response data is composed of the amplitude-frequency response data and the phase-frequency response data, and is stored for subsequent parameterized model extraction.

[0040] As a more preferred embodiment, to further improve the accuracy of the extracted physical property parameters, the generated frequency-domain response data can be subjected to self-consistency verification and noise suppression processing before parameterized model extraction. Based on the Kramer-Kronig relationship, the amplitude-frequency response data is used to calculate the theoretical phase-frequency response data. The difference between the theoretical phase-frequency response data and the measured phase-frequency response data is compared to evaluate the noise level in the measurement data. The Kramer-Kronig relationship is used to suppress noise in the frequency-domain response data.

[0041] Specifically, using this physical constraint, a theoretical phase-frequency response data corresponding to the measured amplitude-frequency response data is calculated by numerical integration (for example, by performing fast Fourier transform on the amplitude-frequency response data, multiplying by a step function in the frequency domain to realize Hilbert transform, and then performing inverse fast Fourier transform to obtain the theoretical phase-frequency response). Then, the root mean square error between the theoretical phase-frequency data and the measured phase-frequency data is calculated. The error value can be used as an indicator to evaluate the current signal-to-noise ratio. Further, the measured amplitude-frequency and phase-frequency data can be iteratively corrected. In each iteration, the theoretical value calculated by the Kramer-Kronig relationship is weighted and averaged until the two values converge to a stable value to the extent that the Kramer-Kronig relationship is satisfied, thereby obtaining a frequency-domain response data with deep noise suppression and stronger self-consistency. The "purified" data is then used for subsequent physical property parameter extraction.

[0042] The embodiment introduces the basic physical principle of the Kramers-Kronig relationship to construct a data self-consistency verification and deep noise suppression method. The method uses physical laws as prior knowledge to "purify" the measured data, which not only can evaluate the data quality, but also can achieve more essential noise suppression than traditional filtering, greatly improving the purity and reliability of the frequency domain response data, thereby laying a solid foundation for subsequent higher precision physical parameter extraction.

[0043] The embodiment greatly suppresses random noise in the original signal by performing a synchronous cross-correlation operation first, significantly improves the signal-to-noise ratio, and can accurately analyze the amplitude and phase characteristics of the signal by combining with the fast Fourier transform, ensuring that the data used for subsequent physical parameter extraction has extremely high purity and reliability, especially enhancing the effective identification ability of the high-frequency weak response signal.

[0044] Further, the construction of the parameterized extraction model comprises: based on the three-dimensional unsteady heat conduction equation describing the dynamic micro-thermal field, and introducing a correction coefficient for characterizing the rare gas effect, combining the geometric structure and material properties of the MEMS sensor, a modified theoretical physical model is established; through numerical analysis of the theoretical physical model, a theoretical frequency domain response data set under a plurality of known physical parameter combinations is obtained; using the theoretical frequency domain response data set, a parameterized extraction model is generated by identification method, and the parameterized extraction model establishes a nonlinear mapping relationship between the key features of the frequency domain response data and the physical parameters.

[0045] Specifically, a theoretical physical model describing the behavior of a dynamic micro-thermal field is established, which is based on a three-dimensional unsteady heat conduction partial differential equation. To ensure the accuracy of the model, the model incorporates key geometric structure information of the MEMS sensor, such as the size of the micro-heater (e.g., the heating resistor, with a size of 50 microns x 50 microns and a resistance value of about 200Ω, operating power range 1-10mW), the precise spatial layout of the temperature measuring element (in one preferred embodiment, a pair of temperature measuring elements, with the core being a temperature measuring resistor, symmetrically arranged along a single axis on both sides of the micro-heater, with a center-to-edge distance of 30-200µm from the heating area, and a typical resistance value of about 1kΩ, with a temperature resolution better than 1mK), and the thermal physical properties of the materials that make up the chip (such as silicon nitride film and silicon-based substrate, with a thickness of 0.5-2µm for the silicon nitride film, a thermal conductivity of about 2W·m-1·K-1, and a specific heat capacity of about 700J·kg-1·K-1; the silicon-based substrate has a thickness of 200-500µm, a thermal conductivity of about 150W·m-1·K-1, and a specific heat capacity of about 700J·kg-1·K-1).

[0046] To extend the model's applicability to low pressure environments, a rarefied gas correction is introduced in the model. When the working pressure is lower than about 10 kPa, the mean free path of gas molecules approaches or exceeds the characteristic dimension of the MEMS device, at which point the continuum assumption fails. By introducing a correction factor related to the Knudsen number, the heat transfer boundary condition between the gas and the solid interface is adjusted. Specifically, the adjustment is implemented using a temperature jump model: at the interface where the sensor solid surface is in contact with the gas, the heat flux density is set as the product of the effective heat transfer coefficient and the temperature difference between the solid and the gas, where the effective heat transfer coefficient is calculated by dividing the gas thermal conductivity by a correction term, which is equal to the device characteristic length plus twice the product of the thermal accommodation coefficient and the Knudsen number. This enables the model to accurately characterize the heat transfer physics in the slip flow regime. Thus, a modified theoretical physics model that integrates the device structure, material properties, and gas physics is established.

[0047] Due to the complexity of the theoretical physics model, an analytical solution cannot be directly obtained, so a numerical analysis method is used for solving. Through finite element analysis, a theoretical frequency domain response dataset is systematically generated. When performing the simulation, the outer surface of the silicon substrate in the model is set to a constant ambient temperature, and only heat conduction effects are considered for all solid surfaces. This process includes: setting a series of discrete parameter combinations covering the typical properties of the natural gas to be tested, such as thermal conductivity from 0.02 W / m·K to 0.05 W / m·K, and volumetric heat capacity from 1000 J / m3·K to 3000 J / m3·K; for each parameter combination, a frequency sweep process simulation consistent with the actual excitation conditions is performed, thereby calculating the corresponding theoretical frequency domain response.

[0048] The last step is to generate a parameterized extraction model using the theoretical dataset. Algorithms in the field of system identification, such as nonlinear least squares fitting or constructing a multilayer feedforward neural network, are used to train the dataset. Specifically, the neural network can be set to include an input layer, a hidden layer with 16 neurons, and an output layer, where the hidden layer uses a rectified linear unit as the activation function. The goal of training is to establish a direct mapping from specific features of the frequency domain response data (such as amplitude and phase values at 5 Hz, 50 Hz, and 500 Hz, constituting 6 input nodes) to gas property parameters (thermal conductivity, volumetric heat capacity, constituting 2 output nodes); after training, the parameterized extraction model can quickly calculate the corresponding property parameter values according to the input measured frequency domain response data; the training process can use the backpropagation algorithm driven by the Adam optimizer, with an initial learning rate of 0.001.

[0049] As a preferred embodiment, the specific characteristic frequencies constituting the input nodes can be determined by performing a sensitivity analysis on the aforementioned theoretical physical model. Specifically, the analysis process includes: performing simulation for a reference gas physical property parameter combination to obtain a reference frequency domain response curve; independently perturbing a single physical property parameter (for example, increasing the thermal conductivity by 1%) and performing simulation again; by comparing the response curves before and after perturbation, the sensitivity of the response signal (amplitude and phase) to the physical property parameter at each frequency point is calculated; by repeating this process, the sensitivity-frequency distribution of different physical property parameters can be obtained; finally, according to the distribution, select those frequency points with significant differentiated sensitivity to different physical property parameters as specific characteristics, for example, select a frequency point that is most sensitive to volumetric heat capacity at low frequency and a frequency point that is most sensitive to thermal diffusivity at high frequency, which together constitute part of the input nodes.

[0050] The model ensures high fidelity of the model in a wide working range (for example, from normal pressure to near vacuum) by combining device-specific parameters and rare gas correction. The generated parameterized model can quickly and accurately invert the electrical measurement signal into the thermal physical property parameters of the target gas core.

[0051] Further, the real-time physical property parameters include: a real-time comprehensive thermal diffusivity extracted based on the amplitude attenuation characteristic in the high frequency region of the frequency domain response data, and a complex heat capacity parameter extracted based on the phase delay characteristic in the low frequency region; and in combination with the pipeline pressure and temperature data obtained in real time from the city natural gas gate station, the real-time density of the mixed natural gas is calculated using the gas state equation, and according to the thermal physical relationship, the equivalent thermal conductivity and equivalent specific heat capacity of the mixed natural gas are calculated using the real-time comprehensive thermal diffusivity, complex heat capacity parameter and real-time density.

[0052] Specifically, to extract real-time physical property parameters from the obtained frequency domain response data, the data is input into a pre-established parameterized extraction model for calculation. The calculation process first partitions the frequency domain response data: in the high frequency region, for example, the part with a frequency higher than 200 Hz, the amplitude attenuation characteristic of the thermal wave is most sensitive to the thermal diffusion process of the gas, and the model calculates and outputs a real-time comprehensive thermal diffusivity accordingly; at the same time, in the low frequency region, for example, the part with a frequency lower than 20 Hz, the phase delay characteristic of the thermal wave is closely related to the volumetric heat capacity (i.e. the product of density and specific heat capacity) of the gas, and the model calculates and outputs a real-time comprehensive thermal diffusivity accordingly. is the equivalent thermal conductivity, ​​(This refers to volumetric heat capacity); meanwhile, in the low-frequency region, such as the part with frequencies below 20 Hz, since the phase delay characteristics of the heat wave are closely related to the volumetric heat capacity of the gas (i.e., the product of density and specific heat capacity), the model calculates and outputs a composite heat capacity parameter accordingly. This parameter is the gas volumetric heat capacity output by the parameterized extraction model. ,in For density, (Equivalent specific heat capacity). These two physical quantities extracted from the model ( and The final equivalent thermal conductivity and equivalent specific heat capacity will be calculated and separated by combining real-time density data in the future.

[0053] As a more preferred implementation, in order to achieve higher precision parameter decoupling, harmonic information can also be used for extraction. At least the phases of the second and third harmonic components can be extracted from the frequency domain complex signal. Based on the evolution relationship of the phases of the fundamental wave, the second harmonic, and the third harmonic at different excitation frequencies, a nonlinear phase spectrum is constructed. And according to the specific distortion characteristics generated by the nonlinear coupling of gas thermophysical parameters in the nonlinear phase spectrum, the equivalent thermal conductivity and the equivalent specific heat capacity are directly calculated and separated.

[0054] Specifically, at least the phases of the second and third harmonic components are extracted from the frequency domain complex signal. Based on the evolution relationship of the phases of the fundamental wave, the second harmonic, and the third harmonic at different excitation frequencies, a nonlinear phase spectrum is constructed. One specific construction method is to calculate a combined phase spectrum, the expression of which is: spectral line (f) = 2 × fundamental wave phase (f) - second harmonic phase (f), where f is the excitation frequency.

[0055] Subsequently, specific distortion features resulting from the nonlinear coupling of gas thermophysical parameters are extracted from this nonlinear phase spectrum. A key distortion feature is the frequency value of the intersection point between the combined phase spectrum curve and the zero phase axis. When constructing the parameterized extraction model, a nonlinear mapping relationship between this intersection frequency value and the gas's equivalent thermal conductivity and equivalent specific heat capacity has been pre-established through numerical simulation (e.g., a two-dimensional lookup table mapping the intersection frequency value to the corresponding thermal conductivity and specific heat capacity, which can be filled by numerical simulation in 1 Hz steps within the frequency range of 10 Hz to 300 Hz). Therefore, in actual measurements, once the intersection frequency value is determined, the equivalent thermal conductivity and equivalent specific heat capacity can be directly calculated and separated by looking up the table and performing linear interpolation or fitting functions. The results can be fused or mutually verified with the aforementioned results based on the fundamental amplitude and phase frequency feature extraction to improve the accuracy of the final output.

[0056] The embodiment improves the analysis dimension from the linear response domain to the nonlinear response domain by constructing and utilizing the phase spectrum of the nonlinear harmonic, deeply excavates the "fingerprint" about the nonlinear coupling of the thermal physical parameters contained in the harmonic phase, and thus can more directly and robustly decouple the thermal conductivity and the specific heat capacity, fundamentally overcomes the precision bottleneck of linear approximation decoupling, and significantly improves the accuracy of parameter extraction.

[0057] To complete the final physical parameter decoupling, the pressure and temperature data of the pipeline are also needed to be obtained in real time from the pipe network monitoring point of the city natural gas gate station through a standard industrial bus (for example, Modbus). According to the obtained real-time pressure (for example, 1.0 MPa) and temperature (for example, 288.15 K), and by applying the AGA8 gas state equation conforming to the industry standard for calculation, the real-time density of the mixed natural gas under the working condition can be accurately obtained.

[0058] Finally, the three obtained parameters, the real-time comprehensive thermal diffusivity ( , i.e., thermal diffusivity ), the composite heat capacity parameter ( , i.e., volumetric heat capacity ), and the real-time density ( ), are subjected to final decoupling calculation. The specific calculation logic is based on the basic thermophysical principles: Calculation of equivalent specific heat capacity ( ): According to the physical definition of volumetric heat capacity ( ), by dividing the model-extracted composite heat capacity parameter ( ) by the real-time density ( ), the equivalent specific heat capacity ( ) is obtained.

[0059] Calculation of equivalent thermal conductivity ( ): According to the physical definition of thermal diffusivity ( ), by multiplying the model-extracted real-time comprehensive thermal diffusivity ( ) by the composite heat capacity parameter ( ), the equivalent thermal conductivity ( ) is obtained.

[0060] These two finally calculated parameters are the core indicators representing the current gas composition and quality.

[0061] The embodiment preliminarily separates the highly coupled thermal diffusivity and heat capacity effect by utilizing the response characteristic differences of the dynamic micro-thermal field at different frequency bands, accurately calculates the density in combination with the pipeline pressure and temperature data, and finally realizes the output of the absolute and mutually independent equivalent thermal conductivity and equivalent specific heat capacity from a single dynamic thermal response measurement, which provides comprehensive and reliable physical basis for online and real-time accurate characterization of the mixed natural gas with continuously changing composition.

[0062] Further, the process of substituting the real-time physical property parameter into the signal-to-noise ratio objective function comprises: substituting the extracted real-time physical property parameter into the signal-to-noise ratio objective function, the signal-to-noise ratio objective function representing the signal-to-noise ratio of the frequency domain response data in the key characteristic frequency band; through an iterative optimization process, taking the evaluation function value as the convergence target, determining the optimal frequency bandwidth, sweep rate and signal amplitude combination of the sweep excitation signal, and the combination is the optimized excitation parameter.

[0063] The calculated real-time physical property parameter, such as the equivalent thermal conductivity and the equivalent specific heat capacity, is immediately used for online optimization and calibration of the subsequent sweep excitation condition, thereby forming an adaptive measurement closed loop. This process is realized through a signal-to-noise ratio objective function, which mathematically represents the signal quality of the frequency domain response data in the key characteristic frequency band, i.e., the high frequency and low frequency regions used to extract the real-time comprehensive thermal diffusivity and composite heat capacity parameters. Specifically, the value of the objective function can be set as the weighted sum of the predicted signal-to-noise ratios in the high and low frequency bands (for example, to focus on the high frequency signal which is more sensitive to signal-to-noise ratio, the high frequency band weight can be set to 0.6 and the low frequency band weight to 0.4). The high frequency band can be taken as 200 Hz to 800 Hz, and the low frequency band can be taken as 5 Hz to 20 Hz. The predicted signal-to-noise ratio of each frequency band is determined by the built-in heat transfer physical model, taking the current physical property parameter and the sweep excitation signal parameters to be optimized (frequency bandwidth, sweep rate and signal amplitude) as input, and calculating the ratio of the predicted theoretical response amplitude to the estimated background noise level. The background noise level can be measured once during device calibration, or dynamically estimated by analyzing the signal in the non-excitation frequency band during each measurement. Specifically, the signal-to-noise ratio objective function can be specifically represented as: ; Wherein: is the final evaluation value of the signal-to-noise ratio objective function (i.e., the evaluation function), is the weight coefficient of the high frequency band signal-to-noise ratio (for example, it can be set to 0.6), is the weight coefficient of the low frequency band signal-to-noise ratio (for example, it can be set to 0.4), is the predicted signal-to-noise ratio of the high frequency key characteristic frequency band, is the predicted signal-to-noise ratio of the low frequency key characteristic frequency band.

[0064] The above formula shows that the total objective function value ( ) is the weighted sum of the predicted signal-to-noise ratios of the high and low frequency bands, and the optimization purpose of the present application is to maximize the value of the through iterative optimization.

[0065] The predicted signal-to-noise ratio of each frequency band ( ) can be defined as: ; wherein represents or ; is the theoretical response amplitude, which is calculated by the heat transfer physical model according to the current extracted real-time physical property parameters and the to-be-optimized swept-frequency excitation signal parameters (i.e. frequency bandwidth, swept-frequency rate and signal amplitude); is the estimated background noise level, which is the background noise level that can be measured or dynamically estimated during device calibration as described above.

[0066] wherein, the calculation process of the theoretical response amplitude is: using the modified theoretical physical model, taking the calculated real-time equivalent thermal conductivity and equivalent specific heat capacity, and the to-be-optimized swept-frequency excitation signal parameters (signal amplitude, frequency) as inputs, performing forward heat conduction calculation through finite element simulation, and extracting the response amplitude corresponding to the frequency point, which is the theoretical response amplitude.

[0067] The dynamic estimation process of the background noise level is: for the generated frequency domain complex signal, in the non-excitation frequency band (e.g. 1100 Hz to 1200 Hz) outside the swept-frequency range (e.g. 1 Hz to 1000 Hz), the root mean square value of the amplitude data is calculated, and it is taken as the estimated background noise level of the current measurement.

[0068] The above formula shows that the predicted signal-to-noise ratio (SNR) of each frequency band is the standard definition of signal-to-noise ratio, i.e. the ratio of the theoretically measurable signal amplitude (A) to the estimated background noise level (N).

[0069] Specifically, the newly obtained real-time physical property parameters are substituted into the signal-to-noise ratio objective function. The heat transfer model inside the function is consistent with the aforementioned theoretical physical model, which can predict the theoretical signal amplitude that should be generated in the current excitation condition in the key characteristic frequency band according to the input physical property parameters. The value of the objective function is the ratio of the theoretical signal amplitude to a preset or real-time evaluated background noise level.

[0070] ​​​To maximize the objective function value, an iterative optimization process, such as a gradient boosting algorithm, will be performed. In each iteration, the algorithm will calculate the gradient of the objective function value with respect to the three key parameters of the current swept-frequency excitation signal, i.e. the frequency bandwidth, the sweep rate, and the signal amplitude. Then, the algorithm will make a small adjustment to these three parameters in the direction of the gradient to find the direction that will make the objective function value grow the fastest. This iterative process continues until the increment of the objective function value is smaller than a pre-set convergence threshold (e.g. the increment of the last two iterations is smaller than 0.1%) or the pre-set maximum number of iterations is reached.

[0071] When the iterative process converges, a set of optimal swept-frequency excitation signal parameter combinations is determined. For example, the initial excitation condition can be 1 volt amplitude, 1-1000 Hz bandwidth, and the optimized combination can be adjusted to 1.1 volt amplitude, 5-800 Hz bandwidth, and a slower sweep rate. This optimized combination is the optimized excitation parameter, which will be used to generate the next and all subsequent swept-frequency excitation signals to replace the previous excitation scheme.

[0072] This embodiment can continuously adjust the excitation condition to the optimal state according to the real-time changes of the measured gas properties by establishing a dynamic optimization feedback mechanism. This ensures that the measurement process is always carried out under conditions that can produce the highest signal-to-noise ratio regardless of the drift of the gas components, thereby fundamentally ensuring the accuracy and stability of the extracted property parameters in long-term and continuous measurement processes.

[0073] Further, the process of online calibration and correction of the internal coefficients of the parameterized extraction model includes: the parameterized extraction model is an artificial neural network model, and the weights and bias coefficients of the artificial neural network model are adjusted through back propagation using the measurement deviation signal, so that the artificial neural network model outputs can approach the known standard property parameters when the reference frequency domain response data is input.

[0074] Specifically, the parameterized extraction model is specifically an artificial neural network model, and the correction process needs to use a reference natural gas with known standard property parameters, for example, a standard mixed gas with accurate metering and stable components.

[0075] The correction process is started when a pre-set maintenance period (e.g. every 24 hours) or an external instruction is triggered, the reference natural gas is introduced into the stationary diffusion chamber, and a complete swept-frequency excitation and response collection is performed to generate a set of reference frequency domain response data. The reference frequency domain response data is input into the artificial neural network model currently in use, and the model outputs a set of measured property parameters.

[0076] Then, the set of measured physical property parameters is compared with the known standard physical property parameters of the reference natural gas, and the difference between them is calculated, which is the measurement bias signal, quantifying the measurement error of the current model at the specific calibration point.

[0077] Once a non-zero measurement bias signal is obtained, it is used to adjust the internal coefficients of the artificial neural network model online through a backpropagation algorithm. During backpropagation, the measurement bias signal is taken as the target error, and the error gradient is calculated layer by layer from the output layer to the input layer. According to the calculated gradient, a small adjustment controlled by the preset learning rate is applied to the weight and bias coefficients of each layer in the network. The direction of this adjustment aims to make the model's output closer to the known standard physical property parameters when it receives the same reference frequency domain response data next time.

[0078] As a more preferred embodiment, in order to improve the wide-range adaptability of online calibration correction, when online calibration correcting the internal coefficients of the parameterized extraction model, it further includes: when generating the reference frequency domain response data, applying at least two different temperature or pressure working conditions to the reference natural gas to be measured to obtain multiple sets of working condition related measurement bias signals; based on the multiple sets of measurement bias signals, establishing a state-dependent correction model that maps measurement bias to working conditions; when extracting real-time physical property parameters, compensating the output of the parameterized extraction model according to the real-time acquired pipeline pressure and temperature data through the state-dependent correction model.

[0079] Specifically, in the first calibration or specific maintenance process, at least two different temperature or pressure working conditions are applied to the reference natural gas to be measured, for example, measurements are taken at two temperature points of 280 K and 300 K respectively, to obtain two or more sets of working condition related measurement bias signals. Based on these multiple sets of data, a state-dependent correction model that maps measurement bias to working conditions is established, which can be a two-dimensional lookup table that stores the correspondence between bias and (temperature, pressure). Specifically, the lookup table can be filled by calibration at multiple working condition points (for example, nine calibration points formed by combining temperatures of 280 K, 300 K, and 320 K, and pressures of 0.8 MPa, 1.0 MPa, and 1.2 MPa) covering the working range. In real-time correction, the bias compensation value at any working condition is calculated by linear interpolation. When extracting real-time physical property parameters, the final output of equivalent thermal conductivity and equivalent specific heat capacity is added to the bias compensation value corresponding to the current real-time working condition obtained by lookup table interpolation, to obtain the final corrected result.

[0080] The embodiment can effectively compensate for the nonlinear measurement deviation caused by the change of working conditions by establishing and applying a state-dependent correction model, upgrade the traditional "single-point correction" to "curved surface correction" covering the entire working interval, greatly improve the accuracy and universality of online calibration correction, and ensure that the measurement results remain highly accurate within a wide temperature and pressure range.

[0081] The embodiment can periodically calibrate itself by performing an online calibration correction process. This method can effectively compensate for the impact of various slow-changing factors on measurement accuracy without disassembling the device for inspection, ensuring the accuracy and consistency of measurement results throughout the life cycle, and greatly improving the reliability and maintainability of the method.

[0082] Further, the method further comprises: extracting and storing a baseline fingerprint representing the initial health state of the MEMS sensor from the baseline frequency domain response data measured for the first time under the baseline natural gas; extracting a real-time fingerprint from the real-time frequency domain response data regenerated under the baseline natural gas, comparing the real-time fingerprint with the baseline fingerprint to calculate a fingerprint drift, and triggering a model correction function limit and outputting an alarm signal indicating that the current extraction result has reduced reliability when the drift exceeds a preset threshold.

[0083] Reference Figure 3 The function runs in a periodic diagnostic process. First, the real-time fingerprint of the current measurement is extracted, and the preset baseline fingerprint is called from the storage. The drift between the two is calculated and compared with the health threshold; if the threshold is not exceeded, the sensor state is determined to be normal and the model correction is allowed to be executed; if the threshold is exceeded, the sensor may be deteriorated, in which case the model correction function is limited to prevent false calibration, and an alarm signal is output at the same time.

[0084] Specifically, when the device is first debugged or factory calibrated, a pre-aging process is first performed on the MEMS sensor, for example, the sensor is operated in a state of being powered on and continuously exciting the micro-heater for at least 24 hours, so that its thermal response characteristics reach stability. After the pre-aging process is completed, the baseline natural gas is introduced into the stationary diffusion chamber and a complete sweep measurement is performed, thereby obtaining an initial baseline frequency domain response data. From this data representing the initial state of the sensor in long-term stable operation, a set of key parameters that can fully represent the nonlinear characteristics of the dynamic thermal response of the sensor is extracted and stored, and the set of parameters collectively constitutes a baseline fingerprint representing the initial health state of the MEMS sensor.

[0085] In the subsequent operation, whenever an online recalibration is performed on the parameterized extraction model, a real-time frequency-domain response data is regenerated under the same reference natural gas environment. A real-time fingerprint is extracted from the real-time data using the same algorithm as that used to generate the reference fingerprint.

[0086] Subsequently, the change in the state of the sensor is quantified by calculating the difference between the real-time fingerprint and the stored reference fingerprint. One specific way of calculation is to regard the two fingerprints as vectors in a multi-dimensional feature space and to calculate the Euclidean distance between them. The distance value is the fingerprint drift.

[0087] The fingerprint drift is compared with a threshold value that is pre-set based on sensor aging experiments and statistical data. For example, one preferred method of setting the threshold value is to perform accelerated aging experiments on a batch of sensors and to set the threshold value as the average fingerprint drift when the measurement error of the sensors reaches a pre-set warning line (e.g., 2% of the initial value). In a typical embodiment, the threshold value can be set to 0.8 for the normalized fingerprint vectors. If the calculated drift does not exceed the threshold value, it indicates that the state of the sensor is stable and the online recalibration process can proceed normally. Otherwise, if the drift exceeds the pre-set threshold value, it indicates that the sensor may have undergone irreversible physical aging or surface contamination and its response characteristics have changed significantly. In this case, the model recalibration function is limited, for example, the online recalibration of the internal coefficients of the artificial neural network model is automatically suspended or prohibited to prevent the model from being adjusted based on unreliable sensor responses. At the same time, an alarm signal indicating that the reliability of the current extraction result has decreased is output, for example, a specific status code or maintenance alarm is used to notify the host computer or the operator that the health status of the sensor is abnormal.

[0088] By introducing the fingerprint monitoring mechanism, the method can effectively distinguish between changes in gas composition and physical state degradation of the sensor, which not only prevents incorrect correction of the algorithm model due to problems with the sensor itself, but also provides a key basis for predictive maintenance of the equipment, significantly enhancing the robustness and intelligence level of the entire measurement scheme.

[0089] Further, the real-time fingerprint and the reference fingerprint are multi-dimensional feature vectors containing normalized energy ratios and relative phase differences extracted from the fundamental wave and multiple detectable harmonic components of the frequency-domain response data.

[0090] Specifically, the real-time fingerprint and the reference fingerprint used to represent the health status of the MEMS sensor are multi-dimensional feature vectors containing normalized energy ratios and relative phase differences extracted from the fundamental wave and multiple detectable harmonic components of the frequency-domain response data.

[0091] The detailed generation process is as follows: analyze the frequency domain response data at each excitation frequency point, identify the fundamental component (whose frequency is the same as the excitation frequency) and the detectable harmonic components (such as the second harmonic and the third harmonic) whose energy is significantly higher than the background noise level.

[0092] Calculate the normalized energy ratio. For each detectable harmonic component, extract its energy (proportional to the square of its amplitude) and calculate the ratio of the energy to the fundamental component energy. For example, if the second harmonic and the third harmonic are detectable, calculate the ratio of the second harmonic energy to the fundamental energy and the ratio of the third harmonic energy to the fundamental energy. These two ratios constitute the first part of the feature vector.

[0093] Calculate the relative phase difference. For each detectable harmonic component, extract its phase angle and calculate the difference between the phase angle and the phase angle of the fundamental component. For example, calculate the difference between the second harmonic phase and the fundamental phase, and the difference between the third harmonic phase and the fundamental phase. These two phase differences constitute the second part of the feature vector.

[0094] Arrange all the calculated normalized energy ratios and relative phase differences in a predetermined order to form a single multi-dimensional feature vector. For example, a four-dimensional feature vector using the second and third harmonics can be represented as: [second harmonic energy ratio, third harmonic energy ratio, second harmonic relative phase difference, third harmonic relative phase difference]. This vector is the fingerprint. The final fingerprint vector is a twelve-dimensional vector composed of the above four-dimensional feature vectors calculated at multiple predetermined fixed excitation frequencies (such as 20 Hz, 150 Hz, and 600 Hz) in sequence.

[0095] Using this fingerprint composed of normalized energy ratios and relative phase differences, rather than using absolute amplitudes or phases, can greatly improve the robustness of diagnosis. Energy ratios are not sensitive to small fluctuations in excitation signal amplitude, while phase differences can effectively offset possible common phase drift in the measurement link. At the same time, the characteristics of harmonic components are highly dependent on the microphysical state of the sensor surface, so this fingerprint has extremely high sensitivity to changes in the health status of the sensor, such as physical aging or contamination.

[0096] The application provides a high-precision, self-adaptive physical property parameter extraction method for complex mixed gas, which can obtain rich frequency domain response information from single measurement by constructing a dynamic micro-thermal field by applying a sweep excitation, and realizes effective decoupling and accurate extraction of multiple coupled parameters such as thermal conductivity and specific heat capacity; the method further introduces two closed-loop regulation mechanisms of online optimization of excitation conditions and online calibration and correction of parameterized models, the online optimization ensures that the measurement process always has the highest signal-to-noise ratio to adapt to the change of gas components, and the online correction can effectively compensate the drift generated by long-term operation of the sensor, so as to realize high-precision, high-stability real-time online measurement of mixed gas with dynamic changes of components, and significantly improve the accuracy and long-term reliability of measurement.

[0097] Embodiment 2 This embodiment takes the application scene of real-time physical property parameter extraction of natural gas with dynamically changing hydrogen mixing ratio in the pipeline at the city natural gas gate station to illustrate the complete process of the application, in this scene, with the change of power grid load, the hydrogen ratio injected into the natural gas pipeline by electrolyzing water to produce hydrogen may fluctuate between 0% and 20%, resulting in significant and rapid changes in key physical property parameters such as thermal conductivity and specific heat capacity of the mixed gas.

[0098] In the initial stage of device deployment, initialization and baseline state establishment are performed. A baseline natural gas with accurately known components and standard physical property parameters (for example, methane with a purity of 99.9%) is passed into the stationary diffusion chamber; then a complete sweep excitation process is performed, for example, an excitation electrical signal with a peak-to-peak value of 1 volt and a frequency swept from 1 hertz to 5000 hertz in a logarithmic step manner is applied, and the original dynamic response signal is synchronously collected; after the signal is transformed, an initial baseline frequency domain response data is generated, from which a baseline fingerprint representing the initial health state of the MEMS sensor is extracted and stored, the fingerprint is specifically a multi-dimensional feature vector composed of the relative values of harmonic components with respect to the fundamental wave, to ensure the robustness of diagnosis, as a specific example, the components of the fingerprint are extracted and combined at one or more preset fixed excitation frequency points, for example, a four-dimensional feature vector can be set, which is composed as follows: normalized energy ratio of second harmonic; normalized energy ratio of third harmonic; relative phase difference of second harmonic; relative phase difference of third harmonic.

[0099] After initialization, the device switches to the normal online extraction mode, and the mixed natural gas to be measured from the main pipeline is introduced into the static diffusion chamber. At this time, the closed-loop operation of measurement-extraction-optimization is periodically performed. Based on the currently stored optimized excitation parameters (initially, the default parameters can be used), the sweep excitation signal is generated, and the response is collected. After transformation, the real-time frequency domain response data is obtained. The data is input into a pre-set parameterized extraction model (a pre-trained artificial neural network model) based on the physical principles of heat transfer. According to the amplitude attenuation characteristics of the frequency domain response data in the high frequency band (e.g., above 200 Hz), the real-time comprehensive thermal diffusivity is extracted. According to the phase delay characteristics in the low frequency band (e.g., below 20 Hz), the complex heat capacity parameter is extracted.

[0100] At the same time, the real-time pipeline pressure (e.g., 1.2 MPa) and temperature (e.g., 290 K) data are obtained from the control system of the gate station through the Modbus bus. The real-time density of the mixed natural gas under the current working condition is accurately calculated using the AGA8 gas state equation. According to the thermal physical relationship, the real-time comprehensive thermal diffusivity, complex heat capacity parameter, and real-time density are calculated to obtain the final real-time physical property parameters, such as equivalent thermal conductivity and equivalent specific heat capacity. The extracted parameter results are output to the control system of the gate station.

[0101] After obtaining the latest real-time physical property parameters, they are immediately substituted into a signal-to-noise ratio objective function. Through gradient boosting algorithm, iterative optimization is performed to improve the signal-to-noise ratio of the key feature band as the convergence target. A new set of optimized excitation parameters is calculated and generated, which may include adjusted frequency bandwidth, sweep rate, and signal amplitude. This set of optimized excitation parameters will automatically replace the current excitation parameters for the next extraction cycle.

[0102] To ensure the accuracy of long-term extraction, the device automatically performs online calibration correction and health diagnosis every 24 hours. At this time, the valve will automatically switch, and the reference natural gas will be introduced again. The device performs a measurement to generate a new real-time frequency domain response data. First, a real-time fingerprint is extracted from the real-time data and compared with the reference fingerprint stored during initialization to calculate the fingerprint drift.

[0103] If the drift is below the preset threshold (e.g., 5%), the sensor is determined to be in a normal health state, and the subsequent calibration correction process is continued. The real-time frequency domain response data is input into the current artificial neural network model to obtain a set of measured physical property parameters. The measured values are compared with the known standard physical property parameters of the reference natural gas to generate a measurement deviation signal. Using the measurement deviation signal, the internal weights and bias coefficients of the artificial neural network model are fine-tuned and corrected through the back propagation algorithm.

[0104] If the fingerprint drift exceeds the preset threshold, the system determines that the sensor health state is abnormal, suspends the current model correction process to prevent false calibration, and outputs a corresponding alarm signal.

[0105] After the correction is completed, the valve switches back to the main pipeline gas source, and the device continues to perform the regular physical property parameter extraction task with an online optimized excitation scheme and an online calibrated extraction model.

[0106] The above merely describes a specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application, therefore, the protection scope of the present application should be limited by the protection scope defined in the claims.

Claims

1. A method for extracting gas property parameters based on microthermal field dynamic characteristics, characterized in that, The method comprises the following steps: A sweep excitation electrical signal is applied to the micro-heater of the MEMS sensor in the static diffusion cavity into which the mixed natural gas to be tested from the city natural gas gate station is introduced, a dynamic heat exchange is established in the mixed natural gas to be tested, and an original dynamic response signal is synchronously collected; the original dynamic response signal is subjected to signal transformation to generate frequency domain response data; A parameterized extraction model is constructed based on the physical principle of heat transfer, and real-time physical property parameters are extracted from the frequency domain response data based on different frequency domain characteristics; The real-time physical property parameters are substituted into a signal-to-noise ratio target function to calculate and generate optimized excitation parameters for frequency domain response characteristic parameter separation; and based on the optimized excitation parameters, an updated sweep excitation electrical signal is generated to replace the current sweep excitation electrical signal; Reference natural gas with known standard physical property parameters is collected and reference frequency domain response data is generated; measured physical property parameters are extracted from the reference frequency domain response data, compared with the known standard physical property parameters to generate a measurement deviation signal, and the internal coefficients of the parameterized extraction model are subjected to online calibration and correction. 2.The method according to claim 1, wherein, The sweep excitation electrical signal is applied to the micro-heater of the MEMS sensor in the static diffusion cavity into which the mixed natural gas to be tested from the city natural gas gate station is introduced, a dynamic heat exchange is established in the mixed natural gas to be tested, and an original dynamic response signal is synchronously collected; the original dynamic response signal is subjected to signal transformation to generate frequency domain response data; 3.The method according to claim 2, wherein, The dynamic micro-thermal field includes that the dynamic micro-thermal field is a non-steady temperature field with the micro-heater as the center, which is established in the mixed natural gas to be tested, and the temperature distribution and the heat flux density both change periodically with time and space position; the heat penetration depth of the dynamic micro-thermal field changes with the frequency of the sweep excitation electrical signal, wherein the heat penetration depth generated by the high-frequency excitation component is limited in the near-surface thermal diffusion zone of the MEMS sensor, and the heat penetration depth generated by the low-frequency excitation component is sufficient to exchange heat with the gas body part far away from the sensor surface. 4.The method according to claim 1, wherein, The generation of the frequency domain response data comprises: performing synchronous cross-correlation operation on the collected original dynamic response signal and the corresponding sweep excitation electrical signal; performing fast Fourier transform on the response signal after the synchronous cross-correlation operation to generate a frequency domain complex signal containing fundamental wave and harmonic components; separating the amplitude frequency response data and the phase frequency response data from the frequency domain complex signal, and the frequency domain response data is the collection of the amplitude frequency response data and the phase frequency response data.

5. The method according to claim 2, wherein The construction of the parameterized extraction model comprises: based on a three-dimensional unsteady heat conduction equation describing the dynamic micro-thermal field, introducing a correction coefficient for characterizing the rarefied gas effect, combining the geometric structure and material properties of the MEMS sensor, establishing a modified theoretical physical model; through numerical analysis of the theoretical physical model, a theoretical frequency domain response data set under a plurality of known physical property parameter combinations is obtained; using the theoretical frequency domain response data set, a parameterized extraction model is fitted and generated through an identification method, which establishes a nonlinear mapping relationship between the key features of the frequency domain response data and the physical property parameters. 6.The method according to claim 1, wherein The real-time physical property parameters include: a real-time comprehensive thermal diffusivity extracted based on the amplitude attenuation characteristics in the high-frequency region of the frequency domain response data, and a complex thermal capacity parameter extracted based on the phase delay characteristics in the low-frequency region; and combining the pipeline pressure and temperature data obtained in real time from the city natural gas gate station, the real-time density of the mixed natural gas is calculated using the gas state equation, and according to the thermal physical relationship, the equivalent thermal conductivity and equivalent specific heat capacity of the mixed natural gas are calculated using the real-time comprehensive thermal diffusivity, complex thermal capacity parameter and real-time density. 7.The method according to claim 1, wherein, The process of substituting the real-time physical property parameters into the signal-to-noise ratio objective function comprises: substituting the extracted real-time physical property parameters into the signal-to-noise ratio objective function, which characterizes the signal-to-noise ratio of the frequency domain response data in the key feature frequency band; through an iterative optimization process, the frequency bandwidth, sweep rate and signal amplitude combination of the optimal sweep excitation signal are determined as the convergence target to improve the evaluation function value, and the combination is the optimized excitation parameter. 8.The method according to claim 1, wherein, The process of online calibration and correction of the internal coefficients of the parameterized extraction model comprises: the parameterized extraction model is an artificial neural network model, and the weights and bias coefficients of the artificial neural network model are adjusted through back propagation using the measurement deviation signal, so that the artificial neural network model outputs an approximation of the known standard physical property parameters when the reference frequency domain response data is input. 9.The method according to claim 1, wherein, The method further comprises: extracting and storing a reference fingerprint representing the initial health status of the MEMS sensor from the reference frequency domain response data first measured under the reference natural gas; extracting a real-time fingerprint from the real-time frequency domain response data regenerated under the reference natural gas, comparing the real-time fingerprint with the reference fingerprint to calculate a fingerprint drift, and when the drift exceeds a preset threshold, triggering a model correction function limit and outputting an alarm signal indicating that the credibility of the current extraction result is reduced.

10. The method according to claim 9, wherein The real-time fingerprint and the reference fingerprint are multi-dimensional feature vectors containing normalized energy ratios and relative phase differences extracted from the fundamental wave and a plurality of detectable harmonic components of the frequency domain response data.

Images