A hydrogen leakage near-far field visualization method based on multi-source data fusion

Abstract

This invention relates to the field of gas leak monitoring technology and discloses a near-field and far-field visualization method for hydrogen leaks based on multi-source data fusion. The method includes simultaneously deploying a schlieren imaging system and a hydrogen concentration sensor in the near-field of the hydrogen leak, collecting data and establishing a grayscale-concentration calibration function to reconstruct a near-field two-dimensional planar concentration distribution map; constructing an initial CFD numerical model of hydrogen leak and diffusion, and correcting key parameters of the model using the near-field two-dimensional concentration distribution; substituting the optimal parameters into the corrected model to calculate and predict the far-field hydrogen diffusion path and concentration distribution in a larger spatial domain; and calculating the full-field concentration distribution by merging the near-field and far-field concentration distributions using a fusion weighting function. This addresses the technical problems of limitations in near-field monitoring, difficulties in far-field prediction, and lack of full-field correlation in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of gas leak monitoring technology, specifically to a near-field and far-field visualization method for hydrogen leaks based on multi-source data fusion. Background Technology

[0002] As a typical clean energy source, the safety of hydrogen storage, transportation, and use is a crucial prerequisite for its large-scale application. Hydrogen is colorless, odorless, and diffuses rapidly; therefore, in the event of a leak, it can easily accumulate in a localized area, forming a flammable or even explosive mixture with air, posing a serious safety hazard. Thus, achieving rapid and accurate monitoring of hydrogen leaks and visualization of concentration fields is a core technological challenge that urgently needs to be addressed in the field of hydrogen energy safety.

[0003] Traditional point-type hydrogen sensors can only acquire concentration data at a limited number of discrete points, which cannot reflect the continuous, two-dimensional concentration distribution characteristics in the vicinity of the leak source, and it is difficult to accurately characterize the dynamic process of hydrogen jet diffusion and local accumulation.

[0004] While computational fluid dynamics (CFD) numerical simulations can effectively predict gas diffusion patterns, their simulation accuracy is highly dependent on the accurate setting of near-field boundary conditions. If there are deviations in source parameters such as leakage rate and leakage direction, the far-field diffusion simulation results will differ significantly from the actual situation.

[0005] Current technologies struggle to effectively integrate and seamlessly connect high-precision near-field leakage information with wide-area far-field diffusion prediction, making it impossible to achieve full-process, visualized monitoring and early warning from the leakage source to the diffusion and accumulation area.

[0006] In summary, the field of hydrogen leak monitoring still faces prominent problems such as limited near-field monitoring methods, insufficient far-field prediction accuracy, and lack of effective correlation of information across the entire field, which urgently require breakthroughs in related technologies. Summary of the Invention

[0007] The present invention aims to provide a near-field and far-field visualization method for hydrogen leakage based on multi-source data fusion, so as to solve the technical problems of near-field monitoring limitations, far-field prediction difficulties, and lack of full-field correlation in existing technologies.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: a near-field and far-field visualization method for hydrogen leakage based on multi-source data fusion, comprising: S1. First, deploy a schlieren imaging system and a hydrogen concentration sensor simultaneously in the near-field area of ​​hydrogen leakage, and simultaneously collect schlieren image sequences and hydrogen concentration data. Samples were collected through controlled experiments, and the average gray value of the corresponding pixel area of ​​the sensor was extracted. A quantitative mapping function between the gray value of the schlieren image and the hydrogen concentration was established to complete the gray-concentration calibration. The calibration function was used to perform pixel-level concentration inversion on the actual leak schlieren image to reconstruct the near-field two-dimensional plane concentration distribution map. S2. Based on the physical parameters of the leakage scenario, an initial CFD numerical model of hydrogen leakage and diffusion that satisfies the convection-diffusion-source equation is constructed. The near-field two-dimensional concentration distribution obtained in S1 is used as the calibration data. Through data assimilation or parameter inversion algorithms, an objective function including observation data and prior parameter information is constructed. The optimal parameter combination that minimizes the function is solved. The key parameters in the model, including leakage rate and initial momentum, are adjusted so that the near-field simulation results of the model are in high agreement with the measured concentration distribution, thus completing the calibration of the CFD numerical model. S3. Substitute the optimal parameter combination obtained in S2 into the corrected CFD numerical model, update the boundary conditions and source terms, and perform transient or steady-state calculations on the convection-diffusion-source equation in a larger spatial domain to predict the diffusion path and concentration distribution of hydrogen in the far field over a large area. By using a fusion weighting function, the near-field high-precision concentration distribution and the far-field predicted concentration distribution are spliced ​​together, and then the fused full-field concentration distribution is calculated.

[0009] The principle and advantages of this scheme are as follows: First, to address the limitations of near-field monitoring, a schlieren imaging system and a hydrogen concentration sensor are deployed simultaneously in the near-field of hydrogen leakage. Schlieren image sequences and concentration data are collected simultaneously. A quantitative mapping relationship between the grayscale value of the schlieren image and the hydrogen concentration is established through controlled experiments. After grayscale-concentration calibration, pixel-level concentration inversion is performed on the actual leakage schlieren image to reconstruct the near-field two-dimensional plane concentration distribution map, thereby achieving continuous and accurate sensing of near-field concentration. Secondly, to address the issue of insufficient far-field prediction accuracy, an initial CFD numerical model is first constructed based on the physical parameters of the leakage scenario. Then, the high-precision concentration distribution reconstructed in the near field is used as calibration data. Through data assimilation or parameter inversion algorithms, the optimal combination of leakage parameters, such as leakage rate and initial momentum, is solved. The key parameters of the model are adjusted to ensure that the near-field simulation results of the CFD model are in high agreement with the measured data, thus completing the model calibration. Finally, the optimal parameters are substituted into the calibrated CFD model to predict the large-scale hydrogen diffusion path and concentration distribution in the far field. Then, by using a fusion weight function to merge the near-field measured concentration and the far-field predicted concentration, a unified concentration distribution across the entire field is obtained, achieving full-process visualization.

[0010] Overcoming the limitations of traditional point sensors that can only acquire discrete point concentration data, this method combines schlieren imaging with a concentration sensor to achieve accurate reconstruction of the near-field two-dimensional continuous concentration distribution. It can clearly depict the dynamic process of hydrogen jet diffusion and local accumulation, making up for the inability of discrete monitoring to reflect the full-field near-field concentration characteristics, improving the comprehensiveness and accuracy of near-field monitoring, and solving the problem of the limitations of traditional point sensors in near-field monitoring.

[0011] Using high-precision concentration data measured in the near field as the basis for model calibration, key parameters of the CFD model, such as leakage rate and initial momentum, are optimized through parameter inversion. This corrects the far-field prediction bias caused by inaccurate near-field boundary conditions in traditional CFD models, significantly improves the prediction accuracy of far-field hydrogen diffusion paths and concentration distribution, and makes the far-field simulation results more consistent with actual leakage scenarios. This solves the problems of difficulty and insufficient accuracy in far-field prediction in CFD numerical simulation.

[0012] By integrating weighting functions to achieve seamless splicing of near-field high-precision concentration distribution and far-field predicted concentration distribution, a full-process concentration visualization system from the leak source to the far-field diffusion and accumulation area is constructed. This breaks the current situation of the separation between near-field monitoring and far-field prediction, realizes the integrated presentation of full-field concentration information, provides comprehensive and coherent data support for hydrogen leak safety early warning, and solves the problem of the lack of near-field and far-field full-field correlation in existing technologies.

[0013] No complex hardware modifications are required. It can be implemented based on existing schlieren imaging systems, hydrogen concentration sensors, and CFD simulation tools. Through standardized calibration, correction, and fusion processes, it can be adapted to different leakage scenarios, such as different leakage rates and different spatial scales. It is highly practical and easy to promote and apply in engineering.

[0014] Preferably, as an improvement, S1 specifically includes: S101. Synchronous data acquisition: Deploy a Z-type schlieren imaging system and a matrix-distributed hydrogen concentration sensor in the near-field area of ​​hydrogen leakage, so that the field of view of the schlieren system covers the sensor area, and synchronously acquire schlieren image sequences and sensor hydrogen concentration data. S102, Gray-concentration calibration: Samples are collected through controlled experiments, gray values ​​of corresponding pixel regions are extracted, and a quantitative mapping function between gray values ​​of schlieren images and hydrogen concentration is established by combining Gaussian process regression. S103. Near-field two-dimensional concentration distribution calculation: The pixel grayscale values ​​of the actual leakage schlieren image are converted into hydrogen concentration values ​​using the quantitative mapping function to reconstruct the near-field two-dimensional planar concentration distribution map, and the relative concentration field can be calculated to enhance contrast or remove background.

[0015] The beneficial effects of this improvement are: the spatiotemporal synchronous acquisition of schlieren image sequences and sensor concentration data completely eliminates subsequent grayscale-concentration calibration deviations caused by data asynchrony, providing a precisely matched data source for subsequent quantitative mapping. The use of a calibration function to achieve pixel-level concentration inversion from actual leak schlieren images completely breaks the limitation of traditional point sensors that can only acquire discrete point concentrations. Calculating the relative concentration field effectively enhances the contrast of the concentration field, removes background interference, and makes the concentration change characteristics of the leak area clearer.

[0016] Preferably, as an improvement, the formula for calculating the average grayscale value of the pixel region corresponding to the point sensor position in the schlieren image extracted in step S102 is as follows: ; in, is the average grayscale value corresponding to the sensor position in the i-th frame image; The image pixel region corresponding to the position of the point sensor; For the region Pixels in; For a moment The grayscale value of the pixel with coordinates (x, y) in the acquired image.

[0017] The beneficial effect of this improvement is that the formula accurately correlates the physical location of the point sensor with the pixel area of ​​the schlieren image, ensuring that the extracted average gray value corresponds precisely to the hydrogen concentration value measured by the sensor, thus avoiding mapping deviations caused by misalignment between the pixel area and the sensor position.

[0018] The formula incorporates a time parameter, which can calculate the average gray value of the corresponding sensor position for different frames of schlieren images. This adapts to the dynamic acquisition scenario of schlieren image sequences, enabling real-time and accurate extraction of gray data at different times. This data is then dynamically matched with the synchronously acquired sensor concentration data, meeting the needs of near-field dynamic monitoring of hydrogen leakage and ensuring that the gray-concentration calibration can be dynamically adjusted along with the leakage process.

[0019] Preferably, as an improvement, in step S102, when establishing the mapping function based on Gaussian process regression, it is assumed that the gray level is a function of hydrogen concentration. The kernel function uses a radial basis function kernel: ; Where m(G) is the mean function of the Gaussian process, Let covariance function be the function of the Gaussian process. Let l be the variance of the function, and l be the correlation length scale; By calibrating the experimental sample set, the parameters of the Gaussian process are determined, and arbitrary gray values ​​are obtained. The corresponding predicted concentration mean and variance are used to construct a complete gray-level-concentration calibration function f(G).

[0020] The beneficial effects of this improvement are: the radial basis function kernel (RBF kernel) can accurately capture the complex nonlinear relationship between gray value and hydrogen concentration. Compared with traditional methods such as linear regression and polynomial fitting, it has higher fitting accuracy, adapts to the nonlinear correlation characteristics between gray value and concentration under different leakage scenarios, and avoids the concentration range deviation that cannot be covered by linear fitting.

[0021] Preferably, as an improvement, the pixel-level density inversion formula in S103 is:

[0022] The formula for calculating the relative concentration field is:

[0023] in, For a moment The hydrogen concentration value at pixel (x, y). The background image is grayscale without leakage. This represents the relative concentration value.

[0024] The beneficial effects of this improvement are: the pixel-level concentration inversion formula directly converts the gray value of each pixel in the actual leakage schlieren image into the corresponding hydrogen concentration value through the calibration function f(G), completely breaking the limitation of traditional point sensors that can only obtain discrete point concentrations, realizing continuous, two-dimensional pixel-level concentration distribution reconstruction in the near field region, which can clearly depict the dynamic process of hydrogen jet diffusion and local accumulation, and fully present the spatial distribution characteristics of near field concentration.

[0025] Preferably, as an improvement, S2 specifically includes: S201. Construct an initial CFD numerical model. Based on the physical parameters of the leakage scenario, establish a CFD numerical simulation model of hydrogen leakage and diffusion. The model satisfies the convection-diffusion-source equation:

[0026] The source term is defined as a finite volume source.

[0027] Where r is the computational domain position vector, t is the time variable, u(r,t) is the velocity field, and D(r,t) is the turbulent diffusion coefficient. Let c be the source term and c be the concentration. Location of the leak. For the strength of the mass source, For the initial momentum flux, Divergence operator; S202. Model calibration: Using the near-field two-dimensional concentration distribution obtained in S103 as calibration data, key parameters in the model, including leakage rate and initial momentum, are adjusted through data assimilation or parameter inversion algorithms to ensure that the model's near-field simulation results closely match the measured concentration distribution, thus obtaining the optimal parameter combination. .

[0028] The beneficial effects of this improvement are: it clearly constructs a CFD numerical model based on the physical parameters of the leakage scenario, and the model satisfies the convection-diffusion-source equation. At the same time, it treats the leakage source as a finite volume source and defines the source term S(r,t;p). Compared with the traditional point source assumption, it more realistically reflects the physical process of hydrogen being ejected from the leak, can more accurately describe the initial momentum and mass distribution of the near-field jet, reduces the deviation between the initial model setting and the actual scenario, and provides a more reasonable physical basis for subsequent model correction.

[0029] By using the near-field two-dimensional concentration distribution as calibration data, and through data assimilation or parameter inversion algorithms, key parameters such as leakage rate and initial momentum in the model are adjusted in a targeted manner, so that the near-field simulation results of the model are in high agreement with the measured concentration distribution. This effectively corrects the deviation caused by inaccurate near-field boundary conditions in traditional CFD models, fundamentally improves the accuracy of the model, and ensures that the subsequent far-field prediction results are more consistent with the actual leakage situation.

[0030] Preferably, as an improvement, in step S202, an objective function is constructed as follows:

[0031] Find the parameter combination that minimizes the objective function.

[0032] in, Here, d is the observation operator, R is the near-field concentration observation data vector, and p is the observation error covariance. Let B be the vector of prior parameter values, and let B be the prior parameter covariance.

[0033] The beneficial effect of this improvement is that the objective function J(p) integrates the observation data error term and the parameter prior information error term, which not only makes full use of the near-field concentration observation data d obtained from S103, but also takes into account the parameter prior value. The rationality of this approach avoids correction bias caused by relying solely on observed data or prior parameter values; the optimal parameter combination is obtained by solving for the minimum value of the objective function. This enabled the accurate inversion of key parameters such as leakage rate and initial momentum, ensuring that the near-field simulation results of the model are in high agreement with the measured data.

[0034] Preferably, as an improvement, S3 specifically includes: S301, High-precision far-field simulation, combining the optimal parameters obtained from S202 Substitute the initial CFD numerical model, update the boundary conditions and source terms, and run the model in a larger spatial domain to perform transient or steady-state calculations to predict the far-field large-scale hydrogen diffusion path and concentration distribution. S302, Full-field concentration distribution visualization: By seamlessly stitching and fusing the near-field high-precision concentration distribution map with the far-field predicted concentration distribution map through a fusion weighting function, a complete hydrogen concentration distribution visualization map is generated. It can also calculate the concentration isosurface and the volume parameters of the lower explosive limit concentration area to achieve risk assessment.

[0035] The beneficial effects of this improvement are: performing transient or steady-state calculations over a larger spatial domain enables accurate prediction of the diffusion path and concentration distribution of hydrogen over a large far-field area. Compared to the far-field predictions of traditional uncorrected models, the accuracy is significantly improved, effectively predicting hazardous areas of hydrogen diffusion. By fusing weighting functions, the high-precision near-field concentration distribution map is seamlessly stitched with the far-field predicted concentration distribution map to generate a complete full-field concentration distribution visualization map. This completely breaks down the current disconnect between near-field monitoring and far-field prediction, achieving full-process concentration visualization from the leak source to the far-field diffusion area, facilitating a comprehensive understanding of the overall leak diffusion situation by personnel.

[0036] By calculating the concentration isosurface and the volume parameters of the lower explosive limit concentration region, the abstract concentration field information is transformed into specific and quantifiable indicators of the degree of danger, avoiding the limitation of traditional visualization which can only be observed intuitively but cannot quantify the risk.

[0037] Preferably, as an improvement, the convection-diffusion-source equation for the far-field diffusion simulation in S301 is:

[0038] in, (r,t) represents the far-field hydrogen concentration. (r,t) represents the large-scale wind field velocity distribution. For the effective diffusion coefficient, the source term S uses the actual leakage rate and momentum inverted by S202; The model boundary conditions are set as follows: near-field and far-field splicing boundary. The far-field concentration equals the near-field concentration at the exit boundary. The concentration is 0 at the wall boundary. The normal gradient at the concentration is 0.

[0039] The beneficial effects of this improvement are: it clearly defines three types of boundary conditions, including splicing boundaries. The far-field concentration equals the near-field concentration, ensuring a seamless transition between near-field and far-field simulations and avoiding numerical jumps at the splicing point; exit boundary The concentration is 0 at the wall boundary. The concentration normal gradient is 0, which conforms to the actual physical laws of hydrogen diffusion. This effectively avoids numerical oscillations and non-convergence during the calculation process, and improves the stability and computational efficiency of the far-field simulation.

[0040] Preferably, as an improvement, the fusion weight function w(r) is defined in S302 in the near-field region. Weight 1, far field region Weight 0, overlapping region Gaussian smoothing weights based on distance are used:

[0041] The comprehensive field concentration distribution is defined as:

[0042] Where r is the position vector in the computational domain, From position r to the splicing boundary The shortest distance, To control the parameters of the smooth width, Near-field concentration, For far-field concentration prediction, The overall concentration after fusion; The fused concentration field was converted into a two-dimensional / three-dimensional visualization image using a color mapping function, and the volume of the hydrogen explosion lower limit concentration region was calculated. Assess the level of leakage risk;

[0043] in, This refers to the lower explosive limit concentration of hydrogen. At time t, the hydrogen concentration exceeds the lower explosive limit of hydrogen. The volume of the region; For the entire computational domain; For indicator functions, when If the condition is met, D=1; otherwise, D=0.

[0044] The beneficial effects of this improvement are: the fusion weight function w(r) sets reasonable weights for different regions, effectively avoiding numerical jumps at the near-field and far-field splicing points, making the full-field concentration distribution after fusion continuous and smooth, and fully presenting the concentration change law from the leakage source to the far-field diffusion.

[0045] By using a color mapping function, the fused concentration field is converted into a two-dimensional / three-dimensional visualization image, making the overall concentration distribution more intuitive. This allows staff to quickly identify high-concentration danger zones and grasp the diffusion trend.

[0046] The formula for calculating the volume of the lower explosive limit concentration region. It transforms abstract concentration field information into specific hazardous volume indicators, which can accurately quantify the area where hydrogen concentration exceeds the lower explosive limit at different times, intuitively reflecting the degree of leakage hazard, and solving the limitation of traditional visualization that can only observe but not quantify risks. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the structure of an embodiment of the present invention. Detailed Implementation

[0048] The following detailed description illustrates the specific implementation method: Example The basics are as follows: Figure 1 As shown, a near-field and far-field visualization method for hydrogen leakage based on multi-source data fusion includes: S1. Multi-source synchronous data acquisition and near-field concentration field reconstruction transform qualitative schlieren images into quantitative near-field two-dimensional concentration distributions.

[0049] S101. Synchronous data acquisition: In the near-field area of ​​a potential hydrogen leak, deploy a Z-shaped schlieren imaging system and one or more point-distributed hydrogen concentration sensors. The hydrogen concentration sensors are arranged in a matrix to ensure that the field of view of the schlieren system covers the area where the sensors are located. Simultaneously acquire schlieren image sequences and sensor hydrogen concentration data during the leak process.

[0050] Schlieren imaging systems are based on the principle of light refraction. When light passes through a transparent medium with a density gradient, the different densities in different regions of the medium cause the light to travel at different speeds, resulting in refraction and changes in the direction and intensity of the refracted light. Schlieren systems, through specific optical arrangements, transform these minute changes in light intensity into observable image changes, thus visually displaying the distribution of the density gradient within the medium.

[0051] Schlieren imaging systems include: The light source provides uniform incident light, typically using a high-intensity point light source or parallel light source to ensure that the directionality and intensity of the light meet the measurement requirements.

[0052] A collimating lens converts the light emitted from a light source into a parallel beam, allowing it to be incident perpendicularly onto the area being measured.

[0053] The measured region, i.e. the transparent medium region to be observed, such as the region where hydrogen flows, is the object of interest for the change in density gradient.

[0054] Schlieren mirror: It is generally a large-aperture mirror or lens used to collect light after passing through the area being measured and focus it onto the blade or detector.

[0055] Blade: Located at the focal plane of the schlieren lens, it is used to block part of the light. By adjusting the position and degree of cutting of the blade, the contrast of the image can be enhanced, and the changes in density gradient can be made more obvious.

[0056] Imaging system: including camera or other image sensor, used to record processed light information to form schlieren images.

[0057] In this technical solution, the schlieren system is mainly used to acquire image information corresponding to the position of the point sensor. Through the schlieren system, a series of schlieren images reflecting the density gradient changes in the measured area can be obtained, with each image corresponding to a specific time point. Then, combined with subsequent image processing and analysis steps, the gray values ​​of the pixel regions corresponding to the point sensor positions are extracted from these schlieren images, and the average gray value of the region is calculated.

[0058] S102. Gray-scale-concentration calibration: Establish a quantitative relationship between gray-scale values ​​of schlieren images and hydrogen concentration.

[0059] Controlled experimental acquisition: Under laboratory conditions with known leakage rate, ambient temperature, and air pressure, standard schlieren images of different concentration gradients were acquired. Grayscale value extraction: Extracting the grayscale values ​​of the pixel region corresponding to the point sensor location in the schlieren image, as follows: Select the pixel region corresponding to the sensor from the image. Calculate the average gray value of this area: ; in, The average grayscale value at the corresponding sensor location in the i-th frame image is calculated based on the average grayscale values ​​of all pixels within that region, and is used to reflect the acquisition time of the i-th frame image. The average brightness of the area corresponding to the sensor; It is a two-dimensional set of pixel regions, representing the pixel region in the image corresponding to the point sensor position, where each element (x,y) in the region is the pixel coordinate in the image.

[0060] For the region The number of pixels in a region is the total number of pixels contained within that region. This number is used for normalization when averaging the grayscale values ​​of the pixels within the region.

[0061] Indicates at time The grayscale value of the pixel at coordinates (x, y) in the acquired image. It is a three-dimensional function, where x and y determine the pixel's position in the image plane. This indicates the time point at which the image was acquired.

[0062] A mapping function is established to correlate the grayscale values ​​with the synchronous concentration values ​​measured by the sensor. Through regression analysis, a grayscale-concentration calibration curve or mapping function for the schlieren imaging system is established. The regression analysis includes linear and polynomial fitting, etc., as detailed below: Assume the quantitative relationship between gray level and hydrogen concentration can be expressed as: C = f(G); Where C is the hydrogen concentration and G is the gray value.

[0063] To improve the model's fitting accuracy and generalization ability, the function is established using a Gaussian process regression method. This method does not presuppose a fixed function form, but instead assumes that it follows a Gaussian process distribution:

[0064] Where f(G) is the functional relationship between gray value G and hydrogen concentration; G is the input gray value variable, corresponding to the gray value of a specific region in the schlieren image; It is another grayscale variable used for calculating the covariance function in a Gaussian process (GP).

[0065] Let f(G) be the mean function of a Gaussian process, which gives the expected value of the function f(G) when the gray value is G. The covariance function of the Gaussian process, also known as the kernel function, is used to describe the correlation between samples of different gray levels.

[0066] The kernel function is in the form of a radial basis function kernel: ; in, is the variance of the function, reflecting the output amplitude; l is the correlation length scale, reflecting the sensitivity of grayscale changes to concentration changes. The smaller l is, the more sensitive the grayscale is to concentration.

[0067] Let the sample set obtained from the calibration experiment be... ; Indicates at time The grayscale value of a specific region in the acquired schlieren image; Indicates at time The corresponding hydrogen concentration value measured by the sensor; N is the number of samples.

[0068] Its corresponding covariance matrix is:

[0069] in, The main body of the N×N covariance matrix is ​​formed by the covariance values ​​between sample i and sample j calculated based on the kernel function. To observe the noise variance, I is the identity matrix, plus... This is to account for the noise effects present during the measurement process.

[0070] Then any grayscale value The corresponding predicted concentration mean and variance can be expressed as follows:

[0071]

[0072] in:

[0073]

[0074] For any given grayscale value The corresponding average concentration was predicted using a Gaussian process regression model. It is a column vector, that is ;in, It's a kernel function. It is the gray value of the i-th sample in the calibration experiment. The new grayscale value was measured. Compared with the gray values ​​of the training samples The similarity between them. It is a column vector, that is This indicates the grayscale sample in the calibration experiment. The hydrogen concentration value obtained by synchronous measurement.

[0075] This indicates the grayscale value. Predicted concentration value The variance is used to measure the uncertainty of the predicted concentration value. It is the gray value of the new sample. Its own covariance reflects The inherent uncertainty. It relates to the relationship between training data and new samples, through the covariance matrix K and its inverse matrix. To measure the impact of training data on the prediction of new samples.

[0076] The parameters of the Gaussian process were determined using sample data obtained through calibration experiments. Within the Gaussian process regression framework, the mean concentration was predicted. That is, the function f( The estimated value of the concentration variance. This quantifies the function f(G) in The variance indicates the uncertainty of the prediction at that gray level. The larger the variance, the more uncertain the prediction at that gray level; the smaller the variance, the more reliable the prediction. By calculating the mean and variance of the predicted concentration corresponding to different gray levels, the mapping relationship between gray level and concentration, as well as the uncertainty of this mapping, can be comprehensively described, thus obtaining a complete description of the gray level-concentration calibration function f(G), i.e., obtaining the gray level-concentration calibration function f(G).

[0077] S103. Near-field two-dimensional concentration distribution calculation: For the schlieren image of the actual leak, using the calibration relationship established in S102, the gray value of each pixel in the image is converted into a hydrogen concentration value, thereby reconstructing the two-dimensional planar concentration distribution map of the near-field of the leak, as follows: For the actual leakage schlieren image at any given time , Pixel-level density inversion using the calibration function f(G):

[0078] Using a pre-established gray-to-density calibration function f(G), the gray-to-density calibration function f(G) is used to measure the gray-to-density of each pixel (x, y) in the actual leaky schlieren image at time t. grayscale value Convert to the corresponding hydrogen concentration value .

[0079] Thus, a two-dimensional hydrogen concentration distribution field is obtained:

[0080] The matrix clearly represents the time at time... The two-dimensional hydrogen concentration distribution across the entire image area, with each element corresponding to the concentration value of a specific pixel in the image.

[0081] To enhance contrast or remove background, calculate the relative concentration field:

[0082] in, The grayscale value is the background grayscale value of the leak-free image. The relative concentration field is obtained by subtracting the concentration value of the leak-free background image (which is obtained by converting the grayscale value of the leak-free image) from the concentration value of the leak-free background image (which is obtained by converting the grayscale value of the leak-free background image).

[0083] S2. Near-field data input and numerical model calibration: The model parameters are optimized in reverse using measured data to solve the problem of large far-field prediction deviation caused by uncertainty in near-field boundary conditions in CFD numerical simulation.

[0084] S201. Constructing the initial CFD numerical model: Based on the physical parameters of the leakage scenario, such as leak outlet size, pressure, and ambient wind speed and direction, establish a CFD numerical simulation model for hydrogen leakage and diffusion, as detailed below: A CFD model for hydrogen leakage-jet flow, turbulent transport and concentration diffusion is established as the forward operator, with concentration (mass fraction or volume fraction) c(r,t) as the main controlled parameter. c(r,t) is a function of position r and time t, representing the concentration of hydrogen at a specific position and time.

[0085] The model satisfies the convection-diffusion-source equations (turbulence model terms can be added):

[0086] Here, r is the computational domain position vector, used to represent a position in space. In two-dimensional or three-dimensional space, it contains coordinate information of the corresponding dimension. t is the time variable, used to describe how the physical process changes over time. For the velocity field, its expression is u(r,t), which is also a function of position r and time t, describing the magnitude and direction of the fluid velocity at each point in space.

[0087] D(r,t) is the turbulent diffusion coefficient, which reflects the magnitude of the diffusion capacity of matter due to turbulent motion. is the source term, used to describe the source strength distribution of hydrogen leakage, and p is a set of unknown parameters used to adjust and characterize the characteristics of the leakage source.

[0088] This represents the rate of change of concentration c over time; It is a convection term that describes concentration transport caused by the velocity field u. It is the diffusion term, reflecting the material diffusion process caused by the concentration gradient and the turbulent diffusion coefficient D; This is the source term, representing the contribution of the hydrogen leakage source to the concentration.

[0089] Treat the leakage source as a finite volume source:

[0090] in, This indicates the location of the leak. The mass source intensity is a function of time t and parameter p, describing the mass of hydrogen leaked from the leak source per unit time. The initial momentum flux is related to the initial momentum of the leakage source and is also a function of time t and parameter p. The divergence operator is used to describe the degree of divergence of a vector field.

[0091] For Dirac delta function, exist It takes a value of infinity at one location and a value of 0 at other locations, and its integral over the entire space equals 1. It is used to precisely locate the source term at the leak location. Place.

[0092] This indicates an increase in concentration due to mass leakage, and a decrease in the mass source intensity. Multiply by the Dirac delta function to ensure that the mass is only at the leak location. Add it to the computation domain.

[0093] The effect of the initial momentum of the leakage source on the concentration distribution was considered. Initial momentum flux. Combined with the Dirac delta function, it describes the distribution of momentum at the leak location and its effect on the motion and concentration diffusion of the surrounding fluid.

[0094] S202, Model Correction: The near-field two-dimensional concentration distribution information obtained in S103, i.e. the relative concentration of each pixel, is used as key correction data and input into the initial CFD numerical model for comparison with the model simulation results.

[0095] By using data assimilation or parameter inversion algorithms, key unknown or uncertain parameters in the model, mainly leakage rate and initial momentum, are adjusted to ensure that the simulation results in the near-field region are in high agreement with the measured concentration distribution obtained from S103, as detailed below: In S103, near-field two-dimensional measurements provide concentration observations on one or more planes / arrays. The observation operator is defined as... This is an operator that maps the model concentration field to the observation space. Its input is a parameter vector p, and its output H(p) is the model prediction corresponding to the observed data. It takes into account factors such as sampling, field of view, integration time, and sensor response, enabling the model prediction to be compared with the actual observation under the same conditions.

[0096] The parameter vector p is a set of unknown parameters optimized by the inversion algorithm, which includes at least parameters related to leakage rate and initial momentum; by adjusting the parameter vector p, the model predictions are made consistent with the observations.

[0097] The expression for the objective function J(p) is:

[0098] Where d is the observation data vector, i.e., the concentration observation data obtained from the near-field two-dimensional measurement in S103, which is the true value obtained from the actual experiment and is used to compare with the model prediction value. R is the observation error covariance, including sensor noise and representativeness error. It reflects the unreliability of the observation data and is used in the objective function to weight the difference between the model prediction and the observation. Observation data with larger errors have relatively smaller weights in the objective function.

[0099] B represents the prior covariance of the parameters, which is the regularization term in the inversion process. It is used to constrain the range of values ​​for the parameter vector p and prevent overfitting during parameter inversion. Based on prior knowledge of the parameters, it reflects the degree of uncertainty of the parameters.

[0100] Let p be the parameter prior value vector, which is the initial estimate of the parameter vector p given based on prior knowledge. In the objective function, it is used to measure the degree of deviation from the prior values ​​during the parameter optimization process.

[0101] Then, find the parameter combination that minimizes the objective function:

[0102] By continuously adjusting the parameter vector p and calculating the objective function J(p), an optimization algorithm is used to find the parameter combination that minimizes the objective function. This optimal parameter combination represents the leakage rate and initial momentum values ​​that best reflect the actual situation after comprehensively considering observational data and prior information.

[0103] The obtained optimal parameters Substituting these parameters into the initial CFD numerical model allows for the correction of key parameters. The corrected model can more accurately simulate the hydrogen leakage and diffusion process in the near-field region, and its simulation results are in high agreement with the measured concentration distribution in S103. This improves the model's accuracy and reliability in simulating actual conditions, providing a more accurate basis for subsequent predictions and analyses.

[0104] S3, Far-field wide-area concentration distribution prediction and visualization, stitches together the high-precision concentration measured in the near field with the wide-area concentration predicted in the far field, to achieve full-chain visualization "from source to far field".

[0105] S301, High-precision far-field simulation: Run the CFD numerical model after S202 correction to perform transient or steady-state calculations.

[0106] Since the near-field boundary conditions of the model have been accurately corrected, its simulation results can accurately predict the diffusion path and concentration distribution of hydrogen over a large area in the far field, as follows: Based on the parameter vector corrected by S202

[0107] The updated boundary conditions and source terms are constructed and used as new inputs to the CFD numerical model.

[0108] Source Item Definition

[0109] Far-field diffusion simulations are also based on the convection-diffusion equations, but operate in a larger spatial domain: Convection-diffusion-source equations for far-field diffusion simulations:

[0110] in, (r,t) represents the far-field hydrogen concentration. (r,t) represents the large-scale wind field velocity distribution. For the effective diffusion coefficient, the source term S uses the actual leakage rate and momentum inverted by S202.

[0111] Boundary conditions are applied:

[0112] in, This forms the boundary between the near and far fields. For the export boundary, This is the wall boundary. This boundary condition specifies the values ​​of hydrogen concentration or concentration gradient for the CFD numerical model at different boundaries to ensure the uniqueness and rationality of the model solution.

[0113] The near-field and far-field splicing boundary, at which the far-field hydrogen concentration is... Take the concentration value calculated in the near field This enables seamless integration of near-field and far-field simulations.

[0114] The outlet boundary is where the hydrogen concentration is set to 0, indicating that hydrogen flows out of the computational domain and its concentration gradually approaches the ambient concentration.

[0115] Wall boundary, This indicates that the normal gradient of hydrogen concentration at the wall is 0, meaning there is no normal concentration change of hydrogen at the wall, reflecting the wall's obstruction of hydrogen diffusion. n represents the wall surface. The direction of the normal.

[0116] The far field can be solved using transient CFD:

[0117] The global spatiotemporal concentration distribution is obtained through multi-step time integration. (r,t). Let be the far-field hydrogen concentration at the next moment (t+Δt). Let be the far-field hydrogen concentration at the current time (t). Δt is the time step, i.e., the time interval from the current time to the next time step. It is a divergence operator. This is a large-scale wind field velocity vector, describing the magnitude and direction of wind velocity in the far-field region. The effective diffusion coefficient reflects the diffusion capacity of hydrogen in a far-field environment due to factors such as molecular motion and turbulence. The source term represents the contribution of the hydrogen leakage source to the far-field hydrogen concentration, and includes information such as leakage rate and initial momentum.

[0118] S302. Visualization of Full-Field Concentration Distribution: The near-field high-precision concentration distribution map obtained in S103 is seamlessly stitched and fused with the far-field predicted concentration distribution map obtained in S301 to generate a complete hydrogen concentration distribution visualization map covering a wide area from the vicinity of the leak source to the far-field space, as shown below: make (r,t) represents the near-field concentration. (r,t) represents the far-field predicted concentration.

[0119] Define a fusion weight function w(r) to smooth the near-far field region:

[0120] The fusion weighting function w(r) is used to smooth the near-field and far-field regions. In the near-field region... A weight of 1 indicates that near-field concentration data is primarily used in this region; in the far-field region... The weight is 0, meaning that far-field predicted concentration data is mainly used; in the overlapping region The Gaussian smoothing function is used to achieve a smooth transition of weights, avoiding abrupt changes in concentration at the splicing point.

[0121] To avoid mutations, overlapping regions Gaussian smoothing weights based on distance are used:

[0122] Where r is the position vector in the computational domain, From position r to the splicing boundary The shortest distance, The larger the parameter σ is, the wider the smooth transition area.

[0123] The comprehensive field concentration distribution is defined as:

[0124] This formula is used to calculate the overall field concentration distribution, using the fusion weighting function w(r) to determine the near-field concentration. and far-field predicted concentration By performing a weighted summation, a complete hydrogen concentration distribution covering the entire space from the vicinity of the leak source to the far field is obtained.

[0125] The fused concentration field Convert to color or grayscale images for 2D / 3D visualization.

[0126] Color mapping function:

[0127]

[0128] The output can be a directly displayable 2D image or a 3D rendering field. . The color or grayscale value at position r and time t. For the concentration field after fusion . This is a color mapping function that maps the input concentration value c to the corresponding color or grayscale value. It is a predefined color map or function that returns the corresponding color or grayscale value based on the input normalized value (ranging from 0 to 1).

[0129] c is the concentration value input into the color mapping function F. The minimum concentration value in the concentration field, The maximum concentration value in the concentration field is used to normalize the concentration value to the range of 0-1.

[0130] For transient results, concentration isosurfaces are generated to visually represent the distribution of areas that reach or exceed a certain concentration threshold.

[0131] To reach or exceed the concentration threshold at time t The set of regions, i.e., the spatial regions corresponding to the concentration isosurfaces. The output includes the full-field concentration distribution function. Visualized map Key isosurfaces or concentration threshold ranges, maximum concentration, average concentration, leakage volume integral, etc.

[0132] Calculate the leakage volume integral to obtain the volume of the area where the hydrogen concentration reaches the lower explosive limit, which is used to assess the degree of danger of the hydrogen leak.

[0133]

[0134] in, This refers to the lower explosive limit concentration of hydrogen. At time t, the hydrogen concentration exceeds the lower explosive limit of hydrogen. The volume of the region. For the entire computational domain. For indicator functions, when If the condition is met, D=1; otherwise, D=0.

[0135] Specifically, the invention is implemented in a potential leak area of ​​an indoor hydrogen refueling station.

[0136] A Z-type schlieren spectrometer is deployed around the hydrogen storage cylinder assembly and valves, ensuring its field of view covers an area of ​​approximately 2m × 2m. Within this field of view, three high-precision electrochemical hydrogen sensors are strategically placed to effectively monitor potential leak areas.

[0137] A controlled hydrogen leak experiment with a known flow rate was conducted in a laboratory environment, simultaneously acquiring schlieren video and sensor data. Through in-depth processing and analysis of the acquired data, a quadratic polynomial calibration curve between the schlieren image grayscale value and hydrogen concentration was established, applicable to this specific environment, providing a precise basis for subsequent concentration conversion.

[0138] During normal operation of the hydrogen refueling station, ensure the continuous and stable operation of the schlieren system. Once a trace hydrogen leak is detected, the system immediately triggers an alarm mechanism and records relevant data. Using established calibration curves, the schlieren image at the initial stage of the leak is converted into a two-dimensional concentration cloud map in real time, thereby accurately showing the location of the leak source and the near-field hydrogen diffusion range.

[0139] A CFD (Computational Fluid Dynamics) model incorporating the detailed geometry of the room was constructed. The acquired near-field concentration distribution map was input into the model as "real data." A genetic algorithm was used to inversely optimize the leakage rate parameters in the model to minimize the error between the simulated concentration field and the measured concentration field at the same location, thereby improving the model's accuracy.

[0140] Using the corrected model, we predict and analyze the concentration accumulation of hydrogen in the entire room, including far-field areas such as corners, ceilings, and vents, within the next 5 minutes.

[0141] The measured near-field concentration cloud map and the predicted far-field concentration cloud map are overlaid onto the 3D layout map of the hydrogen refueling station to generate a full-site concentration risk map. Based on preset concentration thresholds, such as 10%, 25%, and 50% of the lower flammability limit, different risk areas are color-coded and clearly displayed on the monitoring center's large screen. Simultaneously, corresponding early warning information is sent to staff according to the risk level to enable timely response measures and ensure the safe operation of the hydrogen refueling station.

[0142] The above descriptions are merely embodiments of the present invention, and common knowledge such as specific technical solutions and / or characteristics are not described in detail here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the technical solutions of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A near-field and far-field visualization method for hydrogen leakage based on multi-source data fusion, characterized in that, include: S1. First, deploy a schlieren imaging system and a hydrogen concentration sensor simultaneously in the near-field area of ​​hydrogen leakage, and simultaneously collect schlieren image sequences and hydrogen concentration data. Samples were collected through controlled experiments, and the average gray value of the corresponding pixel area of ​​the sensor was extracted. A quantitative mapping function between the gray value of the schlieren image and the hydrogen concentration was established to complete the gray-concentration calibration. The calibration function was used to perform pixel-level concentration inversion on the actual leak schlieren image to reconstruct the near-field two-dimensional plane concentration distribution map. S2. Based on the physical parameters of the leakage scenario, an initial CFD numerical model of hydrogen leakage and diffusion that satisfies the convection-diffusion-source equation is constructed. The near-field two-dimensional concentration distribution obtained in S1 is used as the calibration data. Through data assimilation or parameter inversion algorithms, an objective function including observation data and prior parameter information is constructed. The optimal parameter combination that minimizes the function is solved. The key parameters in the model, including leakage rate and initial momentum, are adjusted so that the near-field simulation results of the model are in high agreement with the measured concentration distribution, thus completing the calibration of the CFD numerical model. S3. Substitute the optimal parameter combination obtained in S2 into the corrected CFD numerical model, update the boundary conditions and source terms, and perform transient or steady-state calculations on the convection-diffusion-source equation in a larger spatial domain to predict the diffusion path and concentration distribution of hydrogen in the far field over a large area. By using a fusion weighting function, the near-field high-precision concentration distribution and the far-field predicted concentration distribution are spliced ​​together, and then the fused full-field concentration distribution is calculated.

2. The near-field and far-field visualization method for hydrogen leakage based on multi-source data fusion according to claim 1, characterized in that, S1 specifically includes: S101. Synchronous data acquisition: Deploy a Z-type schlieren imaging system and a matrix-distributed hydrogen concentration sensor in the near-field area of ​​hydrogen leakage, so that the field of view of the schlieren system covers the sensor area, and synchronously acquire schlieren image sequences and sensor hydrogen concentration data. S102, Gray-concentration calibration: Samples are collected through controlled experiments, gray values ​​of corresponding pixel regions are extracted, and a quantitative mapping function between gray values ​​of schlieren images and hydrogen concentration is established by combining Gaussian process regression. S103. Near-field two-dimensional concentration distribution calculation: The pixel grayscale values ​​of the actual leakage schlieren image are converted into hydrogen concentration values ​​using the quantitative mapping function to reconstruct the near-field two-dimensional planar concentration distribution map, and the relative concentration field can be calculated to enhance contrast or remove background.

3. The near-field and far-field visualization method for hydrogen leakage based on multi-source data fusion according to claim 2, characterized in that: The formula for calculating the average grayscale value of the pixel region corresponding to the point sensor position in the schlieren image extracted in S102 is as follows: ; in, is the average grayscale value corresponding to the sensor position in the i-th frame image; The image pixel region corresponding to the position of the point sensor; For the region Pixels in; For a moment The grayscale value of the pixel with coordinates (x, y) in the acquired image.

4. The near-field and far-field visualization method for hydrogen leakage based on multi-source data fusion according to claim 3, characterized in that: In step S102, when establishing the mapping function based on Gaussian process regression, it is assumed that the functional relationship between gray level and hydrogen concentration is... The kernel function uses a radial basis function kernel: ; Where m(G) is the mean function of the Gaussian process, Let covariance function be the function of the Gaussian process. Let l be the variance of the function, and l be the correlation length scale; By calibrating the experimental sample set, the parameters of the Gaussian process are determined, and arbitrary gray values ​​are obtained. The corresponding predicted concentration mean and variance are used to construct a complete gray-level-concentration calibration function f(G).

5. A near-field and far-field visualization method for hydrogen leakage based on multi-source data fusion according to claim 4, characterized in that: The pixel-level density inversion formula in S103 is as follows: The formula for calculating the relative concentration field is: in, For a moment The hydrogen concentration value at pixel (x, y). The background image is grayscale without leakage. This represents the relative concentration value.

6. The near-field and far-field visualization method for hydrogen leakage based on multi-source data fusion according to claim 5, characterized in that, S2 specifically includes: S201. Construct an initial CFD numerical model. Based on the physical parameters of the leakage scenario, establish a CFD numerical simulation model of hydrogen leakage and diffusion. The model satisfies the convection-diffusion-source equation: The source term is defined as a finite volume source. Where r is the computational domain position vector, t is the time variable, u(r,t) is the velocity field, and D(r,t) is the turbulent diffusion coefficient. Let c be the source term and c be the concentration. Location of the leak. For the strength of the mass source, For the initial momentum flux, Divergence operator; S202. Model calibration: Using the near-field two-dimensional concentration distribution obtained in S103 as calibration data, key parameters in the model, including leakage rate and initial momentum, are adjusted through data assimilation or parameter inversion algorithms to ensure that the model's near-field simulation results closely match the measured concentration distribution, thus obtaining the optimal parameter combination. .

7. A near-field and far-field visualization method for hydrogen leakage based on multi-source data fusion according to claim 6, characterized in that: In step S202, an objective function is constructed as follows: Find the parameter combination that minimizes the objective function. in, Here, d is the observation operator, R is the near-field concentration observation data vector, and p is the observation error covariance. Let B be the vector of prior parameter values, and let B be the prior parameter covariance.

8. The near-field and far-field visualization method for hydrogen leakage based on multi-source data fusion according to claim 7, characterized in that: S3 specifically includes: S301, High-precision far-field simulation, combining the optimal parameters obtained from S202 Substitute the initial CFD numerical model, update the boundary conditions and source terms, and run the model in a larger spatial domain to perform transient or steady-state calculations to predict the far-field large-scale hydrogen diffusion path and concentration distribution. S302, Full-field concentration distribution visualization: By seamlessly stitching and fusing the near-field high-precision concentration distribution map with the far-field predicted concentration distribution map through a fusion weighting function, a complete hydrogen concentration distribution visualization map is generated. It can also calculate the concentration isosurface and the volume parameters of the lower explosive limit concentration area to achieve risk assessment.

9. A near-field and far-field visualization method for hydrogen leakage based on multi-source data fusion according to claim 8, characterized in that: The convection-diffusion-source equations for the far-field diffusion simulation in S301 are as follows: in, (r,t) represents the far-field hydrogen concentration. (r,t) represents the large-scale wind field velocity distribution. For the effective diffusion coefficient, the source term S uses the actual leakage rate and momentum inverted by S202; The model boundary conditions are set as follows: near-field and far-field splicing boundary. The far-field concentration equals the near-field concentration at the exit boundary. The concentration is 0 at the wall boundary. The normal gradient at the concentration is 0.

10. A near-field and far-field visualization method for hydrogen leakage based on multi-source data fusion according to claim 9, characterized in that: In step S302, a fusion weighting function w(r) is defined in the near-field region. Weight 1, far field region Weight 0, overlapping region Gaussian smoothing weights based on distance are used: The comprehensive field concentration distribution is defined as: Where r is the position vector in the computational domain, From position r to the splicing boundary The shortest distance, To control the parameters of the smooth width, Near-field concentration, For far-field concentration prediction, The overall concentration after fusion; The fused concentration field was converted into a two-dimensional / three-dimensional visualization image using a color mapping function, and the volume of the hydrogen explosion lower limit concentration region was calculated. Assess the level of leakage risk; in, This refers to the lower explosive limit concentration of hydrogen. At time t, the hydrogen concentration exceeds the lower explosive limit of hydrogen. The volume of the region; For the entire computational domain; For indicator functions, when If the condition is met, D=1; otherwise, D=0.

Images