A trace gas in-situ measuring instrument and detection method suitable for deep sea environment

Abstract

The application provides a trace gas in-situ measuring instrument suitable for deep sea environment and a detection method, and belongs to the technical field of electric digital data processing.The trace gas in-situ measuring instrument suitable for deep sea environment is carried on a remote control unmanned submersible or an autonomous underwater vehicle, pressure values and temperature values are collected, the pressure values, the temperature values and the permeation concentration sampling values in a permeation initial time sequence are input into an artificial intelligence membrane mass transfer dynamic compensation model, a predicted extreme concentration is output in advance to judge a filling state, an infrared detection module emits infrared light of each detection wave band, the infrared light is fitted by a full spectrum line matrix and is decoupled by a partial least square method, the concentration values of each target gas are calculated according to the Beer-Lambert law, and the membrane mass transfer dynamic compensation model is dynamically adjusted by a permeation state evaluation function to update step length parameters, so that the technical problems of permeation lag of a selective semi-permeable membrane caused by low temperature and high pressure in the deep sea and long response time of in-situ infrared detection are solved.

Description

Technical Field

[0001] This invention belongs to the field of electronic digital data processing technology, and specifically relates to a trace gas in-situ measuring instrument and detection method suitable for deep-sea environments. Background Technology

[0002] In-situ detection of trace gases in the deep sea is an important technological direction in the fields of marine resource exploration and ecological monitoring, with methane, Real-time in-situ acquisition of dissolved gas concentrations in the deep sea, exemplified by the infrared (IR) sensor, is of great significance for natural gas hydrate exploration and marine carbon cycle research. Currently, in-situ infrared detection technology is widely used for deep-sea dissolved gas detection due to its high selectivity response to the specific absorption fingerprints of gas molecules. Typically, a selective semi-permeable membrane is used to separate the target gas from the seawater matrix before guiding it into the infrared detection light path, and then the concentration values ​​of each target gas are calculated according to the Beer-Lambert law. However, the gas permeation process of the selective semi-permeable membrane is controlled by the diffusion coefficient within the membrane, which is strongly dependent on temperature and pressure. Under the low-temperature (1–4℃) and high-pressure (up to 20 MPa) environment of the deep sea, the diffusion coefficient of gas molecules within the membrane decreases significantly and nonlinearly compared to normal pressure and room temperature conditions. This results in a significant increase in the time required for the permeation process to reach steady state, leading to a minute-level waiting time for the sensor response, directly limiting the detection timeliness. In existing technologies, due to the lack of dynamic modeling and early prediction capabilities for the unsteady permeation mass transfer process of gas within selective semi-permeable membranes under deep-sea low-temperature and high-pressure conditions, existing in-situ detection systems can only passively wait for the permeation process to fully stabilize before triggering infrared detection. They cannot output accurate concentration values ​​in the early stages of permeation. In other words, existing technologies suffer from technical problems such as the sluggish permeation of gas within selective semi-permeable membranes caused by the deep-sea low-temperature and high-pressure environment, and the excessively long response time of in-situ infrared detection, which fails to meet the timeliness requirements of deep-sea dynamic monitoring. Summary of the Invention

[0003] In view of this, the present invention provides a trace gas in-situ measuring instrument and detection method suitable for deep-sea environments, which can solve the technical problem in the prior art that the diffusion coefficient of gas in the selective semi-permeable membrane is nonlinearly reduced and the permeation response is severely delayed due to the low temperature and high pressure environment in the deep sea, making it impossible for in-situ infrared detection to obtain accurate target gas concentration values ​​in time before the permeation process is completely stable.

[0004] The present invention is implemented as follows: The first aspect of the present invention provides a trace gas in-situ measuring instrument suitable for deep-sea environments, comprising a main unit, an infrared detection module, an integrated DICP-AF-2400 water vapor separation membrane (composite to a membrane and support), a pre-enrichment chamber, a pressure-resistant housing, and a sulfurized pressure-resistant optical fiber; the main unit is communicatively connected to the infrared detection module via the sulfurized pressure-resistant optical fiber, and all modules are coaxially integrated inside the pressure-resistant housing; the pressure-resistant housing is made of TC4 titanium alloy; the integrated DICP-AF-2400 water vapor separation membrane (composite to a membrane and support) is installed at the front end of the pressure-resistant housing and seamlessly connected to the pre-enrichment chamber using a flange sealing method; the pre-enrichment chamber adopts a microchannel structure, and a metal sintered plate is built into the chamber. The infrared detection module is located after the pre-enrichment cavity and includes a bracket, fiber optic coupler, lens, and reflector. The bracket is a one-piece stainless steel structure with universal mounting holes and heat dissipation slots at the bottom. The fiber optic coupler uses a stainless steel shell and a standard interface, is mounted on the bracket, and is equipped with a sealed protective structure and a fine-tuning mechanism. The lens uses an infrared-adaptive material with a coated transparent film, is fixedly mounted on the bracket, and is equipped with a buffer structure. The reflector is mounted on an adjustable positioning seat on the bracket, and its angle is precisely fine-tuned and then locked in place. The host receives the transmitted light intensity values ​​of each detection band and the incident light intensity values ​​of the reference band output by the infrared detection module through a sulfurized pressure-resistant optical fiber, and completes the calculation and local storage of the concentration values ​​of each target gas.

[0005] Specifically, the pressure-resistant shell is a TC4 titanium alloy shell with an overall external dimension of 55mm×500mm, a pressure resistance of 20MPa, and an applicable depth range of 0~2000m.

[0006] Specifically, the microchannel structure inside the pre-enrichment cavity is a microchannel structure with a channel diameter of 40mm and a length of 20mm.

[0007] Specifically, the metal sintered plate is a porous structure plate made of metal powder through a high-temperature sintering process, and its internal pores have the ability to physically adsorb and enrich target gas molecules.

[0008] Specifically, the flange seal refers to a sealing method that uses a flange end face and a sealing gasket to achieve an airtight connection between the integrated DICP-AF-2400 water vapor separation membrane module (composite to the membrane and support) and the pre-enrichment chamber.

[0009] A second aspect of this invention provides a method for in-situ detection of trace gases in deep-sea environments, comprising the following steps:

[0010] The above-mentioned trace gas in-situ measuring instrument suitable for deep-sea environments is mounted on a remotely operated unmanned underwater vehicle or an autonomous underwater vehicle, so that deep-sea seawater continuously flows over the outer surface of the pressure-resistant shell. Deep-sea dissolved gases selectively permeate into the pre-enrichment chamber through the integrated DICP-AF-2400 water vapor separation membrane composed of membrane and support. Moisture, salt and interfering impurities are blocked on the outside of the integrated DICP-AF-2400 water vapor separation membrane composed of membrane and support.

[0011] The deep-sea dissolved gas entering the pre-enrichment chamber is adsorbed and enriched by the built-in metal sintered plate. The depth sensor and temperature sensor respectively collect the pressure and temperature values ​​of the current deep-sea environment. The pressure and temperature values, as well as multiple permeation concentration sampling values ​​in the initial permeation time series, are input into the membrane mass transfer dynamic compensation model. The membrane mass transfer dynamic compensation model outputs the predicted extreme concentration. The predicted extreme concentration is compared with the preset filling threshold. When the predicted extreme concentration is not lower than the preset filling threshold, the deep-sea dissolved gas filling detection optical path in the pre-enrichment chamber is determined.

[0012] After the deep-sea dissolved gas in the pre-enrichment cavity fills the detection optical path, the laser source in the infrared detection module sequentially emits infrared light of each detection band and infrared light of the reference band corresponding to the absorption fingerprint of the target gas into the detection optical path. The photodetector receives the intensity value of the transmitted light of each detection band and the intensity value of the incident light of the reference band to obtain the original absorbance spectrum of each component.

[0013] The original absorbance spectra of each component are combined with the reference spectral data corresponding to the current pressure and temperature values ​​in the temperature and pressure gradient multidimensional spectral database to perform full spectral matrix fitting. The partial least squares method is used to decouple the cross-interference of overlapping spectral peaks. Based on the Beer-Lambert law, the concentration values ​​of each target gas are calculated from the transmitted light intensity values ​​of each detection band and the incident light intensity values ​​of the reference band.

[0014] Based on the current pressure value, current temperature value, and the predicted extreme concentration and target gas concentration values ​​of the membrane mass transfer dynamic compensation model, the permeation state assessment function value is calculated, and the update step size parameter of the membrane mass transfer dynamic compensation model is adjusted according to the interval in which the permeation state assessment function value is located.

[0015] The concentration values ​​of each target gas are transmitted to the host via a sulfurized pressure-resistant optical fiber. The host then stores the data locally to complete a single in-situ detection. After the remotely controlled unmanned underwater vehicle or autonomous underwater vehicle moves to the next detection location, the aforementioned steps are repeated.

[0016] Specifically, the membrane mass transfer dynamic compensation model is a model that integrates the physical constraints of Fick's second law. The input layer receives the current temperature value, the current pressure value, and multiple permeation concentration sampling values ​​within the time series during the initial permeation stage. The output layer outputs the predicted extreme concentration value. The total loss function is composed of the weighted sum of the mean square error term of the data fitting and the residual term of the partial differential equation of Fick's second law.

[0017] The hidden layer of the membrane mass transfer dynamic compensation model consists of three fully connected layers, with 128, 64 and 32 neurons in each layer, respectively. The activation function is a hyperbolic tangent function, and each backpropagation forces the network weights to satisfy the physical boundary conditions described by Fick's second law.

[0018] The step of adjusting the update step size parameter based on the range of the permeability state evaluation function value is as follows: when the permeability state evaluation function value is not greater than 0.05, the update step size parameter remains unchanged; when the permeability state evaluation function value belongs to the range (0.05, 0.15), the update step size parameter is increased to 1.5 times the original update step size parameter; when the permeability state evaluation function value is greater than 0.15, the update step size parameter is increased to 3 times the original update step size parameter, and a supplementary acquisition process for training data under the current temperature and pressure values ​​is triggered.

[0019] Specifically, the temperature and pressure gradient multidimensional spectral database is formed by pre-collecting high-resolution infrared absorption spectral data of each target gas under various temperature and pressure gradient conditions covering the deep sea temperature range of 1 to 4℃ and the pressure range of 0.1 to 20MPa, and storing the spectral database according to the dual-dimensional index of temperature and pressure values.

[0020] Specifically, the joint fitting of the full spectrum matrix involves forming an absorption matrix from the reference spectrum data of multiple target gas components under the current temperature and pressure conditions, and using partial least squares method to perform overall regression on the original absorbance spectra of each component, while obtaining the concentration contribution of each component, thereby achieving the decoupling and separation of overlapping absorption peaks.

[0021] Specifically, the infiltration initial time series is a time series composed of multiple infiltration concentration sampling values ​​collected at continuous time points from the time the deep-sea dissolved gas begins to infiltrate into the pre-enrichment chamber until the predicted extreme concentration is output. The infiltration concentration sampling values ​​are calculated by converting the transmitted light intensity value of the detection band received in real time by the photodetector using the Beer-Lambert law.

[0022] Specifically, the training dataset for the membrane mass transfer dynamic compensation model is established by setting up multiple temperature and pressure gradient conditions covering the deep-sea temperature range of 1–4℃ and the pressure range of 0.1–20MPa in a laboratory pressure vessel. The changes in permeate concentration over time and the corresponding steady-state extreme concentrations under each temperature and pressure gradient condition are recorded. The temperature value, pressure value, and permeate concentration sampling values ​​in the initial permeation time series are used as input features, and the corresponding steady-state extreme concentrations are used as labels to form the training dataset.

[0023] Specifically, the preset filling threshold is the minimum concentration of deep-sea dissolved gas in the pre-enrichment cavity required for the infrared detection module to reliably trigger detection. This value is determined by laboratory calibration experiments, and the unit is [unit missing]. .

[0024] Specifically, the target gas absorption fingerprint refers to the spectral characteristics of different target gas molecules having unique absorption responses to infrared light in each detection band in the infrared band. Each target gas absorption fingerprint corresponds to a unique detection band, which is used to distinguish different target gas components in the joint fitting of the full spectrum matrix.

[0025] Specifically, the sulfurized pressure-resistant optical fiber is a high-pressure-resistant optical fiber whose outer sheath has been treated with a sulfurization process, enabling stable wired transmission of optical and data signals between the infrared detection module and the host in a deep-sea high-pressure environment.

[0026] This invention constructs a membrane mass transfer dynamic compensation model that incorporates the physical constraints of Fick's second law. It uses temperature, pressure, and sampled permeate concentration values ​​from the initial permeation time series as inputs, and outputs the predicted extreme concentration before the permeation process reaches steady state. This solves the technical problem of gas permeation delay within selective semi-permeable membranes caused by the low-temperature, high-pressure environment of the deep sea. The membrane mass transfer dynamic compensation model incorporates the residual term of the partial differential equation of Fick's second law into the total loss function, forcing the network weights to satisfy the physical boundary conditions of gas diffusion within the membrane during training. This enables the model to extrapolate the steady-state extreme concentration based on the permeation slope, thus compensating for the physical permeation delay with algorithmic prediction. Infrared detection can be triggered without waiting for the permeation process to completely end, significantly reducing the single detection cycle. In summary, this invention solves the technical problems mentioned in the background art, such as the gas permeation delay within selective semi-permeable membranes caused by the low-temperature, high-pressure environment of the deep sea, and the excessively long response time of in-situ infrared detection, which fails to meet the timeliness requirements of deep-sea dynamic monitoring. Attached Figure Description

[0027] Figure 1 This is a flowchart of the method of the present invention.

[0028] Figure 2 This is a schematic diagram of the overall structure of the in-situ measuring instrument involved in the present invention.

[0029] Figure 3 This is a schematic diagram of the infrared detection module structure of the in-situ measuring instrument involved in the present invention.

[0030] Figure 4 This is a distribution diagram of the diffusion coefficient under different temperature and pressure conditions.

[0031] Figure 5 This is a comparison chart of the time series of osmotic concentration and the predicted extreme concentration.

[0032] Figure 6 Comparison of collisional broadening of infrared absorbance lines in multi-component infrared spectra.

[0033] Figure 7 The diagram shows the permeability state assessment function value and the response to step size adjustment.

[0034] In the attached figures, the reference numerals are explained as follows: 1. Probe housing; 2. Front cover of probe housing; 3. Rear cover of probe housing; 4. Water vapor separation membrane; 5. Porous metal sintered plate; 6. Pre-enrichment cavity; 71. Support; 72. Fiber optic coupler; 73. Lens; 74. Reflector. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below.

[0036] like Figure 1 The diagram shown is a flowchart of a trace gas in-situ detection method suitable for deep-sea environments provided by the present invention. This method includes the following steps:

[0037] S01. The trace gas in-situ measuring instrument suitable for deep-sea environment is mounted on a remotely operated unmanned underwater vehicle or an autonomous underwater vehicle, so that deep-sea seawater continuously flows over the outer surface of the pressure-resistant shell. Deep-sea dissolved gas selectively permeates into the pre-enrichment chamber through the integrated DICP-AF-2400 water vapor separation membrane composed of membrane and support. Moisture, salt and interfering impurities are blocked on the outside of the integrated DICP-AF-2400 water vapor separation membrane composed of membrane and support.

[0038] S02. The deep-sea dissolved gas entering the pre-enrichment chamber is adsorbed and enriched by the built-in metal sintered plate. The depth sensor and temperature sensor respectively collect the pressure and temperature values ​​of the current deep-sea environment. The pressure and temperature values, as well as multiple permeation concentration sampling values ​​in the initial permeation time series, are input into the membrane mass transfer dynamic compensation model. The membrane mass transfer dynamic compensation model outputs the predicted extreme concentration. The predicted extreme concentration is compared with the preset filling threshold. When the predicted extreme concentration is not lower than the preset filling threshold, the deep-sea dissolved gas filling detection optical path in the pre-enrichment chamber is determined, and S03 is entered.

[0039] S03. After the deep-sea dissolved gas in the pre-enrichment cavity fills the detection optical path, the laser source in the infrared detection module sequentially emits infrared light in each detection band and infrared light in the reference band corresponding to the absorption fingerprint of the target gas into the detection optical path. The photodetector receives the intensity value of transmitted light in each detection band and the intensity value of incident light in the reference band to obtain the original absorbance spectrum of each component.

[0040] S04. The original absorbance spectra of each component are combined with the reference spectral data corresponding to the current pressure and temperature values ​​in the multidimensional spectral database of temperature and pressure gradient. The partial least squares method is used to decouple the cross-interference of overlapping spectral peaks. The concentration of each target gas is calculated from the transmitted light intensity value of each detection band and the incident light intensity value of the reference band according to the Beer-Lambert law.

[0041] S05. Based on the current pressure value, current temperature value, and the predicted extreme concentration and target gas concentration values ​​of the membrane mass transfer dynamic compensation model, calculate the permeation state assessment function value, and adjust the update step size parameter of the membrane mass transfer dynamic compensation model according to the interval in which the permeation state assessment function value is located.

[0042] S06. The concentration values ​​of each target gas are transmitted to the host via a sulfurized pressure-resistant optical fiber. The host synchronously stores the data locally to complete a single in-situ detection. After the remotely controlled unmanned underwater vehicle or autonomous underwater vehicle moves to the next detection position, S01 to S05 are repeated.

[0043] The membrane mass transfer dynamic compensation model is constructed based on the unsteady permeation mass transfer process of deep-sea dissolved gas within the integrated DICP-AF-2400 water vapor separation membrane (composite to the membrane and support) under deep-sea low-temperature and high-pressure conditions. The model uses Fick's second law to physically constrain the evolution of deep-sea dissolved gas concentration within the membrane over time and space. The physical constraint formula is as follows:

[0044] ;

[0045] in The concentration of deep-sea dissolved gases within the membrane, in units of ; For reference concentration, units ; For time, in units ; For reference time, unit ; Thermobaric diffusion coefficient, in units of ; For reference diffusion coefficient, units ; Intra-membrane coordinates, units ; For film thickness, in units ;

[0046] The temperature-pressure related diffusion coefficient With temperature value (unit ) and pressure value (unit The relationship formula is expressed as follows:

[0047] ;

[0048] in For activation energy, unit ; Gas constant, unit ; For reference temperature, unit ; The pressure-induced diffusion inhibition coefficient is dimensionless. For reference pressure, unit ;

[0049] After obtaining multiple permeation concentration sampling values ​​and corresponding permeation slopes in the initial permeation time series, the membrane mass transfer dynamic compensation model can output the predicted extreme concentration in advance without waiting for the permeation process to reach a complete steady state, thereby compensating for the physical permeation delay caused by low temperature and high pressure.

[0050] The specific structure of the membrane mass transfer dynamic compensation model is as follows: The input layer receives the current temperature value, the current pressure value, and multiple permeation concentration samples within the initial permeation time series, which together constitute the input feature vector; the hidden layer consists of three fully connected layers, with the number of neurons in each layer being 128, 64, and 32 respectively, and the activation function is the hyperbolic tangent function; the loss function of the hidden layer fuses the data error term and the residual term of the partial differential equation of Fick's second law; each backpropagation forces the network weights to satisfy the physical boundary conditions described by Fick's second law; the output layer outputs one node, corresponding to the predicted extreme concentration value.

[0051] The steps for establishing the training dataset for the membrane mass transfer dynamic compensation model specifically include: setting up multiple temperature and pressure gradient conditions covering a deep-sea temperature range of 1–4℃ and a pressure range of 0.1–20 MPa in a laboratory pressure vessel; and using an integrated DICP-AF-2400 water vapor separation membrane (composite to a membrane and support) to separate methane of known concentrations under each temperature and pressure gradient condition. A standard gas permeation experiment was conducted, and the permeation concentration change sequence over time and the corresponding steady-state extreme concentration were recorded under various temperature and pressure gradient conditions to form training sample pairs. Temperature, pressure and permeation concentration sampling values ​​in the initial time series of permeation were used as input features, and the corresponding steady-state extreme concentrations were used as labels to form a training dataset.

[0052] The specific steps for training the membrane mass transfer dynamic compensation model include: iteratively updating the network weights using the Adam optimization algorithm; the total loss function is composed of a weighted sum of the mean square error term of the data fitting and the residual term of the partial differential equation based on Fick's second law, and the formula for the total loss function is as follows:

[0053] ;

[0054] in The total loss function value, in units. ; To predict the mean square error between the extreme concentration and the labeled extreme concentration, the unit is... ; Let Fick's second law partial differential equation be the mean square value of the residual at the sampling point inside the membrane, in units of... ; For reference loss amount, unit ; , The weight coefficient is dimensionless. During training, the total loss function value is monitored on the validation set. Training is terminated when the total loss function value on the validation set no longer decreases for 20 consecutive training rounds.

[0055] The membrane mass transfer dynamic compensation model brings the following technical benefits to the entire detection method: The low temperature and high pressure environment of the deep sea causes a nonlinear decrease in the temperature-pressure related diffusion coefficient of dissolved gases in the integrated DICP-AF-2400 water vapor separation membrane (composite to the membrane and support), resulting in a significant extension of the time required for permeation to reach steady state. If infrared detection is triggered only after the permeation has fully stabilized, the sensing response will face a delay of minutes, which cannot meet the timeliness requirements of deep-sea dynamic monitoring. The membrane mass transfer dynamic compensation model incorporates the physical constraints of Fick's second law partial differential equation into the total loss function, forcing the model weights to satisfy the diffusion physics of dissolved gases in the integrated DICP-AF-2400 water vapor separation membrane during training. This allows for accurate extrapolation of the steady-state extreme concentration after obtaining the permeation slope corresponding to the permeation concentration sampling value in the initial permeation time series, without waiting for the physical permeation process to completely end. This achieves the effect of algorithmically compensating for physical delays, significantly compressing the entire detection cycle, which is particularly important in deep-sea dynamic monitoring scenarios where temperature and pressure values ​​change drastically.

[0056] The permeability state assessment function is calculated based on the current pressure value, current temperature value, and the relative deviation between the predicted extreme concentration and the concentration values ​​of each target gas. The formula for the permeability state assessment function is as follows:

[0057] ;

[0058] in The value of the permeability state assessment function is dimensionless. Dimensionless, for reference evaluation purposes; To predict extreme concentrations, units ; The values ​​represent the measured concentrations of the corresponding components in each target gas concentration value, in units of... ; For reference concentration, units ; Current temperature value, unit ; Current pressure value, unit ; , , These are weighting coefficients, dimensionless.

[0059] When the permeability state assessment function value satisfy When, the current update step size parameter of the membrane mass transfer dynamic compensation model remains unchanged; when When the update step size parameter is increased to 1.5 times the original update step size parameter, the convergence speed of the membrane mass transfer dynamic compensation model is accelerated; when When the update step size parameter is increased to three times the original update step size parameter, a supplementary data acquisition process for the training data corresponding to the current temperature and pressure values ​​is triggered, so that the membrane mass transfer dynamic compensation model can be retrained locally under the current conditions.

[0060] The aforementioned temperature and pressure gradient multidimensional spectral database refers to a database formed by pre-collecting high-resolution infrared absorption spectral data of each target gas under various temperature and pressure gradient conditions covering the deep-sea temperature range of 1–4℃ and pressure range of 0.1–20MPa, and storing the data in a dual-dimensional index based on temperature and pressure values. Under ultra-high pressure conditions in the deep sea, the collision frequency of gas molecules increases, the infrared spectral lines broaden due to collisions, and the absorption peaks of multiple components overlap. The temperature and pressure gradient multidimensional spectral database records the measured patterns of spectral line broadening under different temperature and pressure values, providing an accurate reference spectral data benchmark for the joint fitting of the entire spectral matrix.

[0061] The joint fitting of the full spectrum matrix refers to forming an absorption matrix by combining reference spectrum data of multiple target gas components under the current temperature and pressure conditions, and using partial least squares method to perform overall regression solution on the original absorbance spectrum of each component, while obtaining the concentration contribution of each component, and realizing the decoupling and separation of overlapping absorption peaks. The partial least squares method has the advantage of anti-collinearity compared with the single-peak inversion algorithm when the absorption peak overlap is severe, and maintains the independence and accuracy of the concentration values ​​of each target gas under the condition of cross interference.

[0062] The Beer-Lambert law is used to convert the transmitted light intensity values ​​of each detection band and the incident light intensity values ​​of the reference band into the concentration values ​​of each target gas. The formula is expressed as follows:

[0063] ;

[0064] in The intensity value of incident light in the reference band, in units ; This is a reference value for the incident light intensity in the reference band, in units of... ; The transmitted light intensity values ​​for each detection band are in units of... ; To measure the reference value of transmitted light intensity in the specified band, the unit is... ; The target gas absorption cross section, unit ; The number density of gas molecules, in units of ; To detect the effective optical path length, the unit ; , , These are the corresponding reference values, and the units are respectively. , , .

[0065] The target gas absorption fingerprint refers to the spectral characteristics of different target gas molecules having a unique absorption response to a certain wavelength of infrared light in the infrared band. Each target gas absorption fingerprint corresponds to a unique detection band, which is used to distinguish different target gas components in the joint fitting of the full spectrum matrix.

[0066] The initial infiltration time series refers to the time series composed of multiple infiltration concentration sampling values ​​collected at continuous time points from the time the deep-sea dissolved gas begins to infiltrate into the pre-enrichment chamber until the predicted extreme concentration is output; the infiltration concentration sampling values ​​are calculated by converting the transmitted light intensity value of the detection band received in real time by the photodetector using the Beer-Lamber law.

[0067] The permeation slope refers to the rate of change of the permeation concentration sampling value with time during the initial permeation time series. It is calculated as the ratio of the difference between adjacent permeation concentration sampling values ​​to the corresponding time interval during the initial permeation time series, and the unit is 1. The membrane mass transfer dynamic compensation model uses the permeation slope extrapolation to predict extreme concentrations.

[0068] The preset filling threshold refers to the minimum concentration of deep-sea dissolved gas in the pre-enrichment cavity required for the infrared detection module to reliably trigger detection. This value is determined by laboratory calibration experiments and is expressed in units of... .

[0069] The metal sintered plate is a porous structure plate made of metal powder through a high-temperature sintering process. Its internal pores have the ability to physically adsorb and enrich target gas molecules. It is used to enrich low-concentration deep-sea dissolved gas to a concentration level not lower than a preset filling threshold in the pre-enrichment chamber.

[0070] The flange seal refers to a sealing method that uses the flange end face and sealing gasket to achieve an airtight connection between two components, used for seamless docking between the integrated DICP-AF-2400 water vapor separation membrane module (composite to the membrane and support) and the pre-enrichment chamber.

[0071] The sulfurized pressure-resistant optical fiber refers to a high-pressure-resistant optical fiber whose outer sheath has been treated with a sulfurization process. It is used to achieve stable wired transmission of optical and data signals between the infrared detection module and the host in a deep-sea high-pressure environment.

[0072] like Figure 2 and Figure 3As shown, the structure of the trace gas in-situ measuring instrument for deep-sea environments provided by this invention is described in detail below: The trace gas in-situ measuring instrument for deep-sea environments includes a main unit, an infrared detection module, an integrated DICP-AF-2400 water vapor separation membrane (composite to a membrane and support), a pre-enrichment chamber, a pressure-resistant shell, and a sulfurized pressure-resistant optical fiber. The main unit is communicatively connected to the infrared detection module via the sulfurized pressure-resistant optical fiber. All modules are coaxially integrated inside the pressure-resistant shell. The overall dimensions are 55mm × 500mm, the pressure resistance is 20MPa, and the applicable depth range is 0–2000m. The pressure-resistant shell is suitable for deep-sea high-pressure environments and... The external protective structure for high-salt corrosion environments is made of TC4 titanium alloy. The integrated DICP-AF-2400 water vapor separation membrane, a composite of the membrane and support, is installed at the front end of the pressure-resistant shell and seamlessly connected to the pre-enrichment chamber using a flange seal. This membrane is used for selective permeation and separation of deep-sea dissolved gases, blocking water, salt, and interfering impurities. The pre-enrichment chamber is immediately following the integrated DICP-AF-2400 water vapor separation membrane and employs a microchannel structure with a channel diameter of 40mm and a length of 20mm. A sintered metal plate is built into the chamber to filter the permeated deep-sea dissolved gases. Adsorption and enrichment are performed to fill the subsequent detection optical path with dissolved deep-sea gases. The infrared detection module, located after the pre-enrichment cavity, includes a bracket, fiber optic coupler, lens, and reflector. The bracket is a one-piece stainless steel structure with universal mounting holes and heat dissipation grooves at the bottom, providing stable mounting and positioning for the fiber optic coupler, lens, and reflector. The fiber optic coupler uses a stainless steel shell and a standard interface, is mounted on the bracket, and features a sealed protective structure. It includes a fine-tuning mechanism to achieve efficient coupling and transmission of infrared light between the sulfurized pressure-resistant fiber and the detection optical path. The lens uses an infrared-adaptive material with a coated transparent film and is fixedly mounted on the bracket. A buffer structure is used to collimate and focus the infrared light. The reflector is mounted on an adjustable positioning seat of the bracket. After the angle is precisely fine-tuned, it is locked and fixed to change the direction of infrared light propagation and realize the optical path reversal, ensuring that the infrared light covers the detection optical path in the pre-enrichment cavity according to the preset path. The host receives the transmitted light intensity values ​​of each detection band and the incident light intensity values ​​of the reference band output by the infrared detection module through a sulfurized pressure-resistant optical fiber, completes the calculation and local storage of each target gas concentration value, and transmits each target gas concentration value to the host system of the remotely operated unmanned underwater vehicle or autonomous underwater vehicle equipped with the trace gas in-situ measuring instrument suitable for deep-sea environment.

[0073] It should be noted that the first key technical idea of ​​this invention is the design of a membrane mass transfer dynamic compensation model that embeds physical law constraints into the artificial intelligence model. Traditional data-driven models have weak generalization ability to extraterrestrial temperature and pressure conditions when the training sample coverage is limited. However, this invention incorporates the residual term of the partial differential equation of Fick's second law as a physical constraint term into the total loss function. This ensures that the network weights are continuously constrained by the physical laws of diffusion during training iterations. Even under extreme temperature and pressure gradient conditions with sparse training data, the model still has physical consistency in extrapolating the predicted extreme concentration, thus providing prediction results that conform to the diffusion laws even when data coverage is insufficient.

[0074] The second key technical approach is a multi-group decoupling method based on joint fitting of a multi-dimensional spectral database of temperature and pressure gradients and the full spectral matrix. Deep-sea ultra-high pressure causes collisional broadening of infrared spectral lines and overlapping absorption peaks of multiple components. Traditional single-peak inversion algorithms fail due to the linear correlation of absorption contributions from each component. This invention pre-establishes a high-resolution reference spectral database covering the temperature and pressure range of the deep sea, and uses partial least squares to perform a global regression on the multi-component absorption matrix, simultaneously solving for the concentration contributions of each component. By utilizing the differences in response characteristics of different components across the entire spectral band, it can effectively separate the contributions of each component even under strong collinearity conditions, ensuring the accuracy of multi-component concentration values.

[0075] The third key technical approach is an adaptive parameter adjustment mechanism driven by the permeation state assessment function. Simply relying on the fixed parameters of the membrane mass transfer dynamic compensation model may lead to accumulated prediction biases when temperature and pressure conditions change abruptly. This invention designs a permeation state assessment function that weights and synthesizes the relative deviation between the predicted extreme concentration and the measured concentrations of each target gas, the current temperature offset, and the pressure offset into a permeation state assessment function value. The step size parameter is dynamically adjusted and updated based on the interval in which the permeation state assessment function value falls. When the deviation is large, local fine-tuning training is also triggered, forming a continuous online correction closed loop for the membrane mass transfer dynamic compensation model.

[0076] The synergistic effect of the three technical approaches is reflected in the following aspects: the membrane mass transfer dynamic compensation model provides predicted extreme concentrations under physical constraints to determine the filling state; the temperature and pressure gradient multidimensional spectral database and the full spectrum matrix are jointly fitted to convert the light signal after filling into accurate concentration values ​​of each target gas; and the permeation state assessment function feeds back the concentration values ​​of each target gas to the parameter update process of the membrane mass transfer dynamic compensation model. The three form a closed-loop link of sensing, detection, and correction, enabling the system to maintain a synergistic unity of prediction accuracy and detection timeliness in the dynamic environment of drastic temperature and pressure changes in the deep sea.

[0077] It should be noted that the present invention also solves the following technical problems:

[0078] This invention also solves the technical problems of collisional broadening of infrared absorption lines in multi-component gases, overlapping absorption peaks, and the inability of traditional single-peak inversion algorithms to accurately separate the concentration contributions of each component due to cross-interference caused by the ultra-high pressure environment in the deep sea. Under deep-sea high pressure conditions, the collision frequency of gas molecules increases, and the infrared absorption lines of each component undergo significant broadening, including methane, etc. The absorption peaks of multiple components intertwine in the spectral domain. Single-peak fitting methods suffer from severe systematic errors because they fail to consider the response relationships of each component across the entire spectral range. This invention addresses this by pre-establishing a multidimensional spectral database of temperature and pressure gradients, recording the measured patterns of spectral line broadening for each component under various combinations of temperature and pressure values. During detection, a full-spectrum matrix joint fitting method is used to perform overall partial least squares regression between the multi-component absorption matrix and the measured original absorbance spectra of each component. By utilizing the differences in the response modes of each component across the entire spectral range, cross-interference is decoupled, thereby accurately outputting the concentration values ​​of each target gas even under conditions of severe spectral line overlap at high pressure.

[0079] This invention also solves the technical problem of the membrane mass transfer dynamic compensation model accumulating prediction bias and failing to maintain the accuracy of predicted extreme concentrations under sudden changes in deep-sea temperature and pressure conditions. During the movement of deep-sea remotely operated vehicles (ROVs) or autonomous underwater vehicles (AUVs), the water depth and temperature continuously change, and the abrupt changes in pressure and temperature values ​​can cause the prediction results of the membrane mass transfer dynamic compensation model to deviate under the current parameters. This invention integrates the relative deviation between the predicted extreme concentration and the measured concentrations of each target gas with the deviations of the current temperature and pressure values ​​into a permeability state evaluation function value. Based on the interval of the permeability state evaluation function value, the update step size parameter is adjusted in stages, and local fine-tuning training is triggered when the deviation exceeds the upper limit. This enables the membrane mass transfer dynamic compensation model to continuously track changes in operating conditions online and maintain the accuracy of predicted extreme concentrations.

[0080] Specifically, the principle of this invention is as follows: The core reason why this invention can solve the above-mentioned technical problems is that the unsteady-state permeation mass transfer process of gas within a selective semi-permeable membrane essentially follows the diffusion partial differential equation described by Fick's second law. A physical correlation exists between the initial concentration time series and the final steady-state extreme concentration, determined by this partial differential equation. Therefore, extrapolating the steady-state extreme concentration based on the initial permeation slope is physically valid. The membrane mass transfer dynamic compensation model embeds the residual term of the Fick's second law partial differential equation into the total loss function, continuously constraining the network weights to satisfy this physical law during backpropagation. This enables the model to infer the steady-state extreme concentration from limited initial permeation data. Simultaneously, the explicit modeling of the temperature-pressure related diffusion coefficient allows the model to perceive the differences in diffusion rates under different temperature and pressure gradients in the deep sea, ensuring the accuracy of the predicted extreme concentration even when temperature and pressure values ​​change drastically. This logically forms a closed loop: physical laws ensure the theoretical feasibility of the prediction, and data-driven approaches ensure the engineering accuracy of the prediction. The two work together to effectively compensate for the permeation delay.

[0081] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0082] The specific implementation of step S01 is as follows: the trace gas in-situ measuring instrument is mounted on a remotely operated unmanned underwater vehicle or an autonomous underwater vehicle, so that deep seawater continuously flows over the outer surface of the pressure-resistant shell, and the deep sea dissolved gas selectively permeates into the pre-enrichment chamber through the integrated DICP-AF-2400 water vapor separation membrane composed of membrane and support, while water, salt and interfering impurities are blocked on the outside of the membrane.

[0083] The specific implementation of step S02 is as follows: the deep-sea dissolved gas entering the pre-enrichment chamber is adsorbed and enriched by the built-in metal sintered plate, and the depth sensor and temperature sensor respectively collect the current pressure value. (unit ) and temperature value (unit The membrane mass transfer dynamic compensation model uses Fick's second law to physically constrain the evolution of deep-sea dissolved gas concentration within the membrane over time and space. The physical constraint formula is as follows:

[0084] ;

[0085] In the formula, The concentration of deep-sea dissolved gases within the membrane, in units of ; For reference concentration, units The laboratory calibration experiment was conducted at a temperature... ,pressure Measured under the specified conditions; For time, in units ; For reference time, unit The time when the osmotic concentration can be stably detected by the photodetector for the first time in the osmosis experiment is usually taken; Thermobaric diffusion coefficient, in units of ; For reference diffusion coefficient, units , for temperature ,pressure Measured diffusion coefficient under the given conditions; Intra-membrane coordinates, units The range of values ​​is to ; For film thickness, in units The parameters are given by the instrument manufacturing process. The left side of the formula is the partial derivative of dimensionless concentration with respect to dimensionless time, and the right side... This is the ratio of dimensionless diffusion coefficients. This represents the second partial derivative of the dimensionless concentration with respect to a dimensionless coordinate system; both sides are dimensionless quantities. Temperature-barotropic diffusion coefficient. With temperature value and pressure value The relational formula is expressed as follows:

[0086] ;

[0087] In the formula, For activation energy, unit The results were obtained by fitting the membrane permeation experimental data with the Arrhenius equation under multiple temperature gradient conditions in the laboratory. This is the gas constant, with a value of 8.314. ; For reference temperature, unit Experience value: 277 (Corresponding to a typical deep-sea temperature of 4℃); The pressure-induced diffusion inhibition coefficient is dimensionless and was obtained by regression fitting of membrane permeation experimental data under multiple pressure gradient conditions in the laboratory. For reference pressure, unit The empirical value is 0.1 The first term on the right-hand side of the formula is the temperature correction factor, and the second term is the pressure correction factor. The product of these two terms and the ratio of the diffusion coefficient on the left-hand side are both dimensionless. Multiple permeation concentration samples are presented within the initial permeation time series. ( ,unit ) by photodetector in the first The intensity values ​​of transmitted light in the detection band received at each sampling time are converted from the intensity values ​​of incident light in the reference band using Beer-Lambert's law. The conversion relationship is consistent with the formula given in step S04. (Permeation slope) (unit It is calculated by the ratio of the difference between adjacent sample values ​​to the corresponding time interval, and the formula is expressed as follows:

[0088] ;

[0089] In the formula, For the first The permeation slope corresponding to each sampling time, in units ; For the first The osmotic concentration sampled at each sampling time, in units of... ; For the first The osmotic concentration sampled at each sampling time, in units of... ; For the first With the The time interval between sampling moments, in units The dimensions on both sides of this formula are... Input feature vector of the membrane mass transfer dynamic compensation model It is composed of temperature, pressure, and osmotic concentration sampling values, and the formula is expressed as follows:

[0090] ;

[0091] In the formula, The input feature vector has a dimension of Dimensionless; superscript Indicates vector transpose; This represents the total number of infiltration concentration samples during the initial infiltration time series, dimensionless, with an empirical value of 10 to 30. The model output predicts the extreme concentration. (unit ),Will With preset filling threshold (unit (Determined by laboratory calibration experiments) for comparison, when When the pre-enrichment cavity is filled with deep-sea dissolved gas, the detection optical path is determined, and the process proceeds to step S03.

[0092] The specific implementation of step S03 is as follows: After the deep-sea dissolved gas in the pre-enrichment cavity fills the detection optical path, the laser source in the infrared detection module sequentially emits infrared light of each detection band and reference band corresponding to the target gas absorption fingerprint into the detection optical path, and the photodetector receives the intensity value of the transmitted light of each detection band. (unit , (Identify the target gas component number) and the incident light intensity value in the reference band. (unit ), thus obtaining the original absorbance spectra of each component.

[0093] The specific implementation of step S04 is as follows: The original absorbance spectra of each component are compared with the corresponding current pressure values ​​in the temperature and pressure gradient multidimensional spectral database. With temperature value The reference spectral data were subjected to full-spectrum matrix joint fitting, and partial least squares method was used to decouple overlapping spectral peaks from cross-interference. Based on the Beer-Lambert law, the concentration values ​​of each target gas were calculated from the transmitted light intensity values ​​of each detection band and the incident light intensity values ​​of the reference band, as expressed in the following formula:

[0094] ;

[0095] In the formula, The intensity value of incident light in the reference band, in units ; This is a reference value for the incident light intensity in the reference band, in units of... The values ​​were calibrated in the laboratory under standard operating conditions. For the first The transmitted light intensity value of the component corresponding to the detection band, in units of ; For the first Reference values ​​for transmitted light intensity in the component detection band, in units The values ​​were calibrated in the laboratory under standard operating conditions. For the first Component target gas absorption cross section, unit The data is obtained by looking up the corresponding temperature and pressure values ​​in the temperature and pressure gradient multidimensional spectral database. For the first Number density of component gas molecules, unit ; To detect the effective optical path length, the unit The parameters are given by the instrument manufacturing process; Reference value for the absorption cross section of the reference component, unit ; Reference value for the number density of gas molecules of the reference component, unit ; This is the optical path reference value, in units of... The value is determined by laboratory calibration experiments. The left side of the formula represents the logarithm of the light intensity ratio, and the right side represents the numerator. Dimensions are denominator Dimensionless Both sides are dimensionless quantities. An absorption matrix is ​​constructed by combining reference spectral data of multiple target gas components under current temperature and pressure conditions. The formula is expressed as follows:

[0096] ;

[0097] In the formula, The absorption matrix has dimensions of . Dimensionless; The number of wavenumber points is dimensionless and determined by the spectral resolution of the temperature and pressure gradient multidimensional spectral database. The number of target gas components, dimensionless, is determined by the detection task; For the first Component in Reference absorbance values ​​at each wavenumber point, dimensionless, derived from the current pressure value in the multidimensional spectral database of temperature and pressure gradients. With temperature value Reference spectral data, , Measured original absorbance spectral vector Each component By the The measured light intensity values ​​at each wavenumber point are calculated using the Beer-Lambert law, and expressed as a vector as follows:

[0098] ;

[0099] In the formula, The measured original absorbance spectral line vector has a dimension of . Dimensionless; For the first The intensity of incident light at a reference wavelength at each wavenumber point, in units of ; For the first The transmitted light intensity value of the detection band at each wavenumber point, in units of ; For the first Reference values ​​of transmitted light intensity in the detection band corresponding to each wavenumber point, in units of The values ​​were calibrated in the laboratory under standard operating conditions. In step S03 For the same physical quantity in the first... Discrete values ​​at each wavenumber point In step S03 For the same physical quantity in the first... The discrete values ​​at each wavenumber point have the same meaning. Concentration contribution vector of each component. The definition is as follows:

[0100] ;

[0101] In the formula, The contribution vector for the concentration of each component, with dimension . ,unit ; For the first Measured concentration values ​​of the target gas components, in units. , Partial least squares method uses the absorption matrix... Compared with the measured original absorbance spectral line vector Simultaneously projecting onto the latent variable space effectively eliminates collinearity interference caused by overlapping absorption peaks, thus... Perform an overall regression solution, and finally output the concentration contribution vector of each component. The concentration values ​​of each target gas were obtained. .

[0102] The specific implementation method of step S05 is as follows: based on the current pressure value Current temperature value and predicted extreme concentration Calculate the permeation state assessment function value based on the concentration values ​​of each target gas. The formula is expressed as follows:

[0103] ;

[0104] In the formula, The value of the permeability state assessment function is dimensionless. For reference evaluation purposes, this is a dimensionless quantity with an empirical value of 1. The predicted extreme concentration output by the membrane mass transfer dynamic compensation model, in units of ; The first of the target gas concentration values Measured concentration values ​​of the components, in units ; For reference concentration, units In step S02 The meaning is consistent; Current temperature value, unit ; For reference temperature, unit In step S02 The meaning is consistent; Current pressure value, unit ; For reference pressure, unit In step S02 The meaning is consistent; , , Let be the weighting coefficients, dimensionless, with empirical values ​​of 0.6, 0.2, and 0.2, respectively, satisfying the following conditions: The formula's terms represent the combined contributions of concentration prediction bias, temperature bias, and pressure bias to the permeability state. All three terms are dimensionless. The sum of the three terms on the right side and the sum of the terms on the left side... Dimensions are consistent. Based on... The update step size parameter of the membrane mass transfer dynamic compensation model in the given interval (Dimensionless, the learning rate parameter during online model updates) is adjusted as follows: When At that time, maintain Unchanged; when At that time, Updated to ;when At that time, Updated to And trigger the current temperature value With pressure value The process of supplementing training data under corresponding operating conditions enables the membrane mass transfer dynamic compensation model to be retrained with local fine-tuning under the current operating conditions.

[0105] The specific implementation of step S06 is as follows: The concentration values ​​of each target gas... The data is transmitted to the host computer via a sulfurized pressure-resistant optical fiber. The host computer then performs local storage synchronously to complete a single in-situ detection. After the remotely controlled unmanned underwater vehicle or autonomous underwater vehicle moves to the next detection location, steps S01 to S05 are repeated.

[0106] Furthermore, a total loss function considering the reference value was designed, and the formula is expressed as follows:

[0107] ;

[0108] In the formula, The total loss function value, in units. ; To predict the mean square error between the extreme concentration and the labeled extreme concentration, the unit is... The formula is expressed as follows:

[0109] ;

[0110] In the formula, The total number of training samples is dimensionless. For the first The predicted extreme value concentration corresponding to each training sample, in units ; For the first The label extreme value concentration corresponding to each training sample, in units It is derived from the steady-state extreme concentration measured in laboratory permeation experiments. Let Fick's second law partial differential equation be the mean square value of the residual at the sampling point inside the membrane, in units of... The formula is expressed as follows:

[0111] ;

[0112] In the formula, The total number of sampling points within the membrane is dimensionless, representing the total number of discrete sampling points within the membrane for the partial differential equation of Fick's second law. For the first The model-predicted intramembrane dissolved gas concentration at each configuration point, in units of... ; For the first The time corresponding to each configuration point, in units ; For the first The intramembrane coordinates of each configuration point, in units of ; For the first Temperature value corresponding to each configuration point, in units ; For the first Pressure value corresponding to each configuration point, in units The residual term within parentheses is a dimensionless quantity, multiplied by... back Dimensions restored to ; For reference loss amount, unit It is derived from the variance estimation of the concentration labels in the training dataset; , These are weighting coefficients, dimensionless, with empirical values ​​of 0.7 and 0.3, used to balance the contributions of the data fitting error term and the physical constraint residual term to the total loss function. The left side of the formula... The total loss is dimensionless. The two terms on the right are the normalized data error term and the normalized physical constraint residual term, respectively. Both sides are dimensionless quantities. By incorporating the physical constraints of the partial differential equation of Fick's second law into the total loss function, the model weights are forced to satisfy the physical laws of diffusion of dissolved gases in the membrane during the training process.

[0113] To better understand and implement this invention, the following is a specific application scenario of embodiment 2: To illustrate the application effect of this invention in a practical deep-sea trace gas detection task, technicians mounted a trace gas in-situ measuring instrument suitable for deep-sea environments onto an autonomous underwater vehicle and conducted a deep-sea dissolved gas in-situ detection operation in a certain sea area. The target gas was... and The instrument has a detection depth range of 200–1800 m, corresponding to a pressure range of approximately 2–18 MPa and a water temperature range of approximately 1.2–3.8℃. The overall dimensions of the instrument are 55 mm × 500 mm. The pressure-resistant outer shell is made of TC4 titanium alloy, with a rated pressure resistance of 20 MPa, meeting the depth requirements of this operation.

[0114] The autonomous underwater vehicle (AUV) descends along a pre-set path, with deep-sea water continuously flowing over the pressure-resistant outer surface of its shell. The integrated DICP-AF-2400 water vapor separation membrane, a composite of membrane and support structure, remains in direct contact with the deep-sea water. (Deep-sea dissolution...) and Driven by the concentration gradient across the membrane, the gas permeates towards the pre-enrichment chamber. Moisture, salts, and other interfering impurities are blocked on the outside of the membrane and do not enter the detection chamber. The pre-enrichment chamber has a channel diameter of 40 mm and a length of 20 mm. Inside the chamber, a sintered metal plate continuously performs physical adsorption and enrichment on the low-concentration dissolved gases that permeate in.

[0115] At a detection depth of 1200m, the depth sensor recorded a current pressure of 12.1 MPa, and the temperature sensor recorded a current temperature of 2.3℃, or 275.45 K. Under these temperature and pressure conditions, the membrane... Thermo-pressure related diffusion coefficient Compared to the reference operating conditions at normal temperature and pressure, the detection time is significantly reduced. If the detection is triggered only after the permeation has fully reached a steady state, the waiting time will be extended to several minutes, which cannot meet the timeliness requirements for detection during dynamic cruise of autonomous underwater vehicles. To solve this problem, the membrane mass transfer dynamic compensation model starts collecting permeate concentration samples during the initial permeation time series immediately after the permeation starts. It continuously records permeate concentration samples at six time points: 10s, 20s, 30s, 40s, 50s, and 60s after the permeation starts, and simultaneously calculates the permeation slope between adjacent sampling points. The current temperature, pressure, and the aforementioned permeate concentration samples are then input into the membrane mass transfer dynamic compensation model. The model's hidden layers consist of three fully connected layers with 128, 64, and 32 neurons respectively. The activation function is a hyperbolic tangent function, and the total loss function integrates the mean square error term of the data fitting and the residual term of the partial differential equation based on Fick's second law, forcing the network weights to satisfy the physical laws of dissolved gas diffusion within the membrane during training. The model outputs data even with only 60 seconds of initial permeation sampling data. The predicted extreme concentration is 8.52. , The predicted extreme concentration is 3.17. The predicted extreme concentrations are all no lower than the preset filling threshold of 3.0. The system determines that the dissolved gas in the pre-enrichment cavity has filled the detection optical path, triggering the infrared detection module to start. Figure 5The graph comparing the time series of infiltration concentration with the predicted extreme concentration clearly shows the relationship between the initial sampling value, the model-predicted extreme concentration, and the physical steady-state concentration, as well as the lead time of the predicted trigger time relative to the physical steady-state time.

[0116] The laser source in the infrared detection module emits laser light sequentially into the detection optical path and... and The infrared light in each detection band and the reference band corresponding to the fingerprint is absorbed. Under the current high pressure condition of 12.1 MPa, the collision frequency of gas molecules increases significantly, and the infrared spectral lines are broadened due to collisions. and The absorption peak is at wavenumber 2800~ There is significant overlap within the area. For example... Figure 6 As shown, Figure 6 This is a comparison of collisional broadening of multi-component infrared absorbance lines, comparing pressures of 0.1 MPa and 20 MPa. and The broadening of superimposed spectral lines clearly shows the significant increase in the full width at half maximum (FWHM) and the overlap of adjacent absorption peak edges under high pressure. The photodetector receives the transmitted light intensity values ​​of each detection band and the incident light intensity values ​​of the reference band to obtain the original absorbance spectra of each component. These are then combined with reference spectral data corresponding to 275.45 K and 12.1 MPa conditions in the temperature and pressure gradient multidimensional spectral database for full-spectrum matrix fitting. Partial least squares decoupling is used to remove cross-interference from overlapping peaks. The concentration values ​​of each target gas are calculated based on the Beer-Lambert law. The detection location in this study... The measured concentration was 8.41. , The measured concentration was 3.09. .

[0117] After completing the concentration calculation, the permeability state assessment function value is calculated based on the current pressure value of 12.1 MPa, the current temperature value of 275.45 K, the predicted extreme concentration, and the concentration values ​​of each target gas. Location of this test The calculated value is 0.038, which satisfies the requirement. Under the given conditions, the update step size parameter of the membrane mass transfer dynamic compensation model remains unchanged. The autonomous underwater vehicle completed 40 in-situ detections throughout the entire dive and cruise process, and the permeability state evaluation function values ​​corresponding to each detection are shown in Table 1.

[0118] Table 1. Permeability status assessment function values ​​and step size adjustment strategies for each number of tests.

[0119]

[0120] like Figure 7 As shown, Figure 7 The permeation state assessment function value and step size adjustment response plot show the results of 40 tests. The values ​​changed successively and the corresponding three-step adjustment response distributions were analyzed. During the 11th to 15th tests, the autonomous underwater vehicle experienced significant changes in pressure and temperature gradients. When the value enters the middle range, the step size automatically increases to accelerate model convergence; the pressure changes are more drastic during the 21st to 23rd tests. If the value exceeds 0.15, the step size is increased to three times the original value, and supplementary collection of training data and local fine-tuning training under the current operating condition are automatically triggered. This allows the model to realign with the physical diffusion law under this operating condition, and subsequent detection... The value then dropped below 0.05, and the step size returned to a stable state.

[0121] Membrane interior under different temperature and pressure conditions Temperature-pressure related diffusion coefficient The distribution pattern is as follows Figure 4 As shown, Figure 4 The diffusion coefficient distribution under different temperature and pressure conditions is shown in the figure, covering multiple gradient conditions with temperatures ranging from 1 to 4℃ and pressures from 0.1 to 20 MPa. A clear pattern is observed: the diffusion coefficient monotonically decreases with increasing pressure and slightly increases with increasing temperature. This verifies the strong nonlinear response characteristics of the temperature-pressure correlated diffusion coefficient to the low-temperature and high-pressure environment of the deep sea, and also confirms the time-sensitivity dilemma faced by directly waiting for steady-state permeation without introducing a membrane mass transfer dynamic compensation model. (The data obtained from various tests are presented.) and The concentration value is transmitted to the host via a sulfurized pressure-resistant optical fiber, and the host synchronously completes local storage. The single detection cycle is about 75 seconds, which is significantly reduced compared to the time required for physical penetration to reach a complete steady state.

[0122] Compared to traditional deep-sea gas sampling and analysis methods, this invention represents a significant advancement in technical principles. Traditional methods rely on underwater samplers to bring water or gas samples back to the laboratory for analysis. During this process, dissolved gases are susceptible to escape and redissolution due to changes in temperature and pressure, leading to distorted concentration results and failing to capture the instantaneous concentration distribution under true deep-sea conditions. This invention achieves selective gas-liquid separation in situ using an integrated DICP-AF-2400 water vapor separation membrane combining a membrane and a support. Infrared absorption measurement is directly performed in the detection optical path, fundamentally eliminating phase transition interference and concentration loss during sample transport. Traditional deep-sea spectral detection schemes, without considering the impact of deep-sea low temperature and high pressure on the membrane permeation process, consume considerable time waiting for the permeation to reach steady state. The membrane mass transfer dynamic compensation model, by embedding Fick's second law partial differential equation as a physical constraint into the neural network loss function, allows the model to accurately extrapolate the steady-state extreme concentration in the early stages of permeation. This algorithmic compensation compensates for the physical permeation delay caused by deep-sea low temperature and high pressure—a capability that purely data-driven models, lacking physical constraints, cannot achieve. To address the issues of infrared spectral line collisional broadening and multi-component absorption peak overlap caused by deep-sea high pressure, a multi-dimensional spectral database of temperature and pressure gradients pre-stores measured spectral line broadening patterns under different temperature and pressure conditions. Combined with partial least squares full-spectral matrix joint fitting, this method offers a significant advantage in anti-interference under collinearity conditions compared to traditional single-peak inversion algorithms, ensuring the accuracy of independent decoupling of multi-component concentrations. The permeability state assessment function, through real-time monitoring of the deviation between predicted extreme concentrations and measured concentrations, tightly couples the dynamic adjustment of the model update step size with the actual changes in the deep-sea temperature and pressure environment. This enables the entire detection system to continuously adaptively calibrate in complex and dynamic deep-sea environments, rather than relying on a static model with fixed parameters.

[0123] It should be noted that the variables involved in this invention are explained in detail in Tables 2 and 3.

[0124] Table 2. Variable Explanation Table (Part 1)

[0125]

[0126] Table 3. Variable Explanation Table (Part Two)

[0127]

[0128] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A trace gas in-situ measuring instrument suitable for deep-sea environments, characterized in that, The system includes a main unit, an infrared detection module, an integrated DICP-AF-2400 water vapor separation membrane (composite to a membrane and support), a pre-enrichment chamber, a pressure-resistant housing, and a sulfurized pressure-resistant optical fiber. The main unit communicates with the infrared detection module via the sulfurized pressure-resistant optical fiber, and all modules are coaxially integrated inside the pressure-resistant housing. The pressure-resistant housing is made of TC4 titanium alloy. The integrated DICP-AF-2400 water vapor separation membrane (composite to a membrane and support) is installed at the front end of the pressure-resistant housing and seamlessly connected to the pre-enrichment chamber using a flange-type sealing method. The pre-enrichment chamber employs a microchannel structure with a built-in sintered metal plate. The infrared detection module, located downstream of the pre-enrichment chamber, includes a support, fiber optic coupler, lens, and reflector. The support is a one-piece stainless steel structure with universal mounting holes and heat dissipation grooves at the bottom. The fiber optic coupler, with a stainless steel shell and standard interface, is mounted on the support and features a sealed protective structure, along with a fine-tuning mechanism. The lens, made of infrared-adaptive material with a coated transparent film, is fixed to the support and features a buffer structure. The reflector is mounted on an adjustable positioning seat on the support, and its angle is precisely fine-tuned before being locked in place. The main unit receives the transmitted light intensity values ​​of each detection band and the incident light intensity values ​​of the reference band from the infrared detection module via a sulfurized pressure-resistant optical fiber, completing the calculation and local storage of the target gas concentration values.

2. The in-situ measuring instrument according to claim 1, characterized in that, The pressure-resistant shell is specifically a TC4 titanium alloy shell with an overall external dimension of 55mm×500mm, a pressure resistance of 20MPa, and an applicable depth range of 0~2000m.

3. The in-situ measuring instrument according to claim 2, characterized in that, The microchannel structure inside the pre-enrichment cavity is specifically a microchannel structure with a channel diameter of 40mm and a length of 20mm.

4. The in-situ measuring instrument according to claim 3, characterized in that, The metal sintered plate is specifically a porous structure plate made of metal powder through a high-temperature sintering process, and its internal pores have the ability to physically adsorb and enrich target gas molecules.

5. The in-situ measuring instrument according to claim 4, characterized in that, The flange seal is specifically a sealing method that uses a flange end face and a sealing gasket to achieve an airtight connection between the integrated DICP-AF-2400 water vapor separation membrane module (composite to the membrane and support) and the pre-enrichment chamber.

6. A method for in-situ detection of trace gases suitable for deep-sea environments, characterized in that, Includes the following steps: The trace gas in-situ measuring instrument for deep-sea environments as described in any one of claims 1 to 5 is mounted on a remotely operated unmanned underwater vehicle or an autonomous underwater vehicle, so that deep-sea seawater continuously flows over the outer surface of the pressure-resistant shell, and deep-sea dissolved gases selectively permeate into the pre-enrichment chamber through the integrated DICP-AF-2400 water vapor separation membrane composed of membrane and support. Moisture, salt and interfering impurities are blocked on the outside of the integrated DICP-AF-2400 water vapor separation membrane composed of membrane and support. The deep-sea dissolved gas entering the pre-enrichment chamber is adsorbed and enriched by the built-in metal sintered plate. The depth sensor and temperature sensor respectively collect the pressure and temperature values ​​of the current deep-sea environment. The pressure and temperature values, as well as multiple permeation concentration sampling values ​​in the initial permeation time series, are input into the membrane mass transfer dynamic compensation model. The membrane mass transfer dynamic compensation model outputs the predicted extreme concentration. The predicted extreme concentration is compared with the preset filling threshold. When the predicted extreme concentration is not lower than the preset filling threshold, the deep-sea dissolved gas filling detection optical path in the pre-enrichment chamber is determined. After the deep-sea dissolved gas in the pre-enrichment cavity fills the detection optical path, the laser source in the infrared detection module sequentially emits infrared light of each detection band and infrared light of the reference band corresponding to the absorption fingerprint of the target gas into the detection optical path. The photodetector receives the intensity value of the transmitted light of each detection band and the intensity value of the incident light of the reference band to obtain the original absorbance spectrum of each component. The original absorbance spectra of each component are combined with the reference spectral data corresponding to the current pressure and temperature values ​​in the temperature and pressure gradient multidimensional spectral database to perform full spectral matrix fitting. The partial least squares method is used to decouple the cross-interference of overlapping spectral peaks. Based on the Beer-Lambert law, the concentration values ​​of each target gas are calculated from the transmitted light intensity values ​​of each detection band and the incident light intensity values ​​of the reference band. Based on the current pressure value, current temperature value, and the predicted extreme concentration and target gas concentration values ​​of the membrane mass transfer dynamic compensation model, the permeation state assessment function value is calculated, and the update step size parameter of the membrane mass transfer dynamic compensation model is adjusted according to the interval in which the permeation state assessment function value is located. The concentration values ​​of each target gas are transmitted to the host via a sulfurized pressure-resistant optical fiber. The host then stores the data locally to complete a single in-situ detection. After the remotely controlled unmanned underwater vehicle or autonomous underwater vehicle moves to the next detection location, the aforementioned steps are repeated.

7. The in-situ detection method according to claim 6, characterized in that, The membrane mass transfer dynamic compensation model is specifically a model that integrates the physical constraints of Fick's second law. The input layer receives the current temperature value, the current pressure value, and multiple permeation concentration sampling values ​​within the time series during the initial permeation stage. The output layer outputs the predicted extreme concentration value. The total loss function is composed of the weighted sum of the mean square error term of the data fitting and the residual term of the partial differential equation of Fick's second law.

8. The in-situ detection method according to claim 7, characterized in that, The hidden layer of the membrane mass transfer dynamic compensation model consists of three fully connected layers, with 128, 64 and 32 neurons in each layer, respectively. The activation function is a hyperbolic tangent function, and each backpropagation forces the network weights to satisfy the physical boundary conditions described by Fick's second law.

9. The in-situ detection method according to claim 8, characterized in that, The step of adjusting the update step size parameter based on the range of the permeability state evaluation function value is as follows: when the permeability state evaluation function value is not greater than 0.05, the update step size parameter remains unchanged; when the permeability state evaluation function value belongs to the range (0.05, 0.15), the update step size parameter is increased to 1.5 times the original update step size parameter; when the permeability state evaluation function value is greater than 0.15, the update step size parameter is increased to 3 times the original update step size parameter, and a supplementary acquisition process for training data under the current temperature and pressure values ​​is triggered.

10. The in-situ detection method according to claim 9, characterized in that, The aforementioned temperature and pressure gradient multidimensional spectral database is specifically formed by pre-collecting high-resolution infrared absorption spectral data of each target gas under various temperature and pressure gradient conditions covering the deep-sea temperature range of 1–4℃ and pressure range of 0.1–20MPa, and storing the spectral database according to a dual-dimensional index of temperature and pressure values.

Images