A gas diffusion imaging processing method considering path confidence correction

Abstract

The present application relates to the technical field of environmental monitoring, and more particularly to a gas diffusion imaging processing method considering path confidence correction, comprising: calculating link input, establishing the relationship between observation and concentration by using a nonlinear diffusion imaging model, improving inversion stability by using path length estimation and confidence correction, and obtaining reliable concentration field through energy functional optimization; based on concentration sequence and image sequence, combining brightness consistency and conservation error to estimate wind speed field, and then using physical stepping and learning residual to obtain evolution operator prediction of future concentration with physical and data constraints; embedding the prediction module into system dynamics to describe the influence of device action on concentration, configuring risk measurement design and optimization control strategy, and controlling the key area to maintain safe concentration under various disturbances. The present application solves the problems of imaging error, unstable diffusion prediction and uncontrollable intervention in the prior art.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology, and more specifically, to a gas diffusion imaging processing method that considers path confidence correction. Background Technology

[0002] With the rapid development of environmental monitoring technology, gas identification and diffusion inference methods based on spectral imaging have shown great potential in fields such as industrial safety, urban emergency response, and unattended monitoring. As a processing framework that integrates imaging mechanism modeling and data-driven prediction, this type of method, by introducing nonlinear diffusion inversion, concentration evolution modeling under physical constraints, and risk-driven control strategies, not only significantly improves the accuracy of diffusion situation estimation but also ensures reliable prediction of future concentration changes, thereby promoting the development of intelligent environmental monitoring technology. In the field of gas monitoring, ensuring the accuracy of concentration estimation and diffusion prediction is crucial for performing diffusion source tracing, hazard area assessment, and proactive intervention. Complex and variable environmental conditions often require capabilities including accurate concentration inversion, dynamic wind field estimation, and future situation prediction to meet the real-time needs of emergency response. However, limitations in existing imaging models, wind field inversion errors, and unstable diffusion predictions result in lag and uncertainty in the overall monitoring chain. Summary of the Invention

[0003] The purpose of this invention is to provide a gas diffusion imaging processing method that considers path confidence correction in order to solve the above-mentioned problems in the prior art.

[0004] This invention is achieved through the following technical solution:

[0005] A gas diffusion imaging processing method considering path confidence correction includes the following steps:

[0006] Step S1: Calculate the link input, establish the relationship between observation and concentration using a nonlinear diffusion imaging model, improve inversion stability using path length estimation and confidence correction, and then obtain a reliable concentration field through energy functional optimization. ;

[0007] Step S2: Estimate the wind speed field based on the concentration sequence and image sequence, combined with brightness consistency and conservation error. Then, by using physical steps and learning residuals, an evolution operator with both physical and data constraints is obtained to predict the concentration in the next few steps;

[0008] Step S3: Embed the prediction module into system dynamics to describe the impact of equipment actions on concentration, configure risk measurement and design and optimize control strategies to maintain safe concentrations in key areas under various disturbances.

[0009] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0010] 1. By introducing a nonlinear diffusion imaging model and path length confidence correction, the concentration inversion process can effectively reduce the uncertainties caused by topographic and incident angle variations. This method not only improves the stability of concentration estimation but also ensures the consistency of the inversion results in spatial structure, providing more accurate data input for subsequent modeling;

[0011] 2. An evolutionary operator based on mass conservation constraints and data-driven residual learning is introduced as the core component of the concentration prediction process. This allows for more reliable inferences about temporal changes in the concentration field through the combined effects of physical consistency and data compensation. This approach not only focuses on the overall diffusion trend but also accurately handles local disturbances and structural changes, significantly improving the reliability of diffusion predictions. Particularly under complex wind field conditions, this technique enhances the practical application value of diffusion simulation.

[0012] 3. A multi-device collaborative control strategy based on risk measurement is adopted, enabling the proactive intervention process to remain stable and safe under uncertain disturbances. When executing fan scheduling or spray equipment control, this strategy can optimize the overall concentration reduction effect and constrain the maximum concentration peak in key areas, achieving a highly efficient safety protection effect. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart of a gas diffusion imaging processing method that considers path confidence correction in an embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0016] The following is in conjunction with the appendix Figure 1 The following describes some embodiments of the present invention in detail. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0017] This invention proposes a gas diffusion imaging processing method that considers path confidence correction. In a preferred embodiment, the method is as follows: Figure 1 As shown, it includes:

[0018] Step S1: Calculate the link input, establish the relationship between observation and concentration using a nonlinear diffusion imaging model, improve inversion stability using path length estimation and confidence correction, and then obtain a reliable concentration field through energy functional optimization. ;

[0019] Step S2: Estimate the wind speed field based on the concentration sequence and image sequence, combined with brightness consistency and conservation error. Then, by using physical steps and learning residuals, an evolution operator with both physical and data constraints is obtained to predict the concentration in the next few steps;

[0020] Step S3: Embed the prediction module into system dynamics to describe the impact of equipment actions on concentration, configure risk measurement and design and optimize control strategies to maintain safe concentrations in key areas under various disturbances.

[0021] In one specific embodiment, the complete implementation path of the overall technical process of this method includes:

[0022] 1. From image to concentration: The relationship between observation and concentration is established using a nonlinear diffusion imaging model, the inversion stability is improved by path length estimation and confidence correction, and a reliable concentration field is obtained through energy functional optimization—this is the input of the entire process.

[0023] 2. From the current field to the future field: Based on the concentration sequence and image sequence, the wind speed field is estimated by combining brightness consistency and conservation error. Then, the evolution operator with both physical and data constraints is obtained by using physical stepping and learning residuals to achieve the prediction of concentration in the next few steps.

[0024] 3. From prediction to control: The prediction module is embedded in the system dynamics to describe the impact of equipment actions on concentration. Then, risk measurement is used to design and optimize control strategies so that the critical area can still maintain a safe concentration under various disturbances.

[0025] The specific implementation steps are as follows:

[0026] Several bands are provided The image, with pixel positions denoted as two-dimensional coordinates. In the band The observed values ​​are denoted as Under rarefied gas conditions, observations in each band can be decomposed into a background term. Determined by ground, environment, and equipment response. Gas diffusion / absorption term. Noise item Therefore, the basic imaging relationship exists:

[0027] ;

[0028] In the formula, It is obtained by time averaging or background modeling in a gas-free scene.

[0029] To link the diffusion term with the volume concentration, we introduce... : Gas concentration (mass per unit volume) at the ground corresponding to the pixel. The effective path length of the imaging ray through the air mass depends on the terrain height, viewing angle, and air mass thickness. A power-law form of nonlinear diffusion model is used here:

[0030] ;

[0031] An index representing the effect of path length on diffusion; Indicates the effective path length; Indicates gas concentration; where: Band The diffusion intensity coefficient is obtained through calibration. : The index of the effect of path length on diffusion. : The exponential effect of concentration on the response of this band. Substituting the above equation into the basic imaging relation and incorporating the noise term into the residual, we obtain:

[0032] .

[0033] In pixels Assuming the noise is zero-mean Gaussian, the most natural estimation method is least squares. The pixel density can be expressed as the following optimization problem:

[0034] ;

[0035] This represents the pixel-level local concentration estimate. Let be the gas concentration variable to be estimated; this formula only uses single-point information and is not stable enough when there is a lot of noise, so it is necessary to consider the spatial neighborhood.

[0036] set up For pixels The neighborhood, This represents neighboring pixels. To suppress noise while preserving edges as much as possible, a weighted smoothing term is introduced:

[0037] ;

[0038] in, For spatial smoothing regularization; Represents the concentration within the neighborhood; weight Specifically, it is set as follows: ; , represents the adjustment constant; Indicates the first Pixels in each band Observation intensity at the location; Indicates the first Pixels in each band The intensity of observation at that location.

[0039] This reduces the weight of areas with large grayscale differences in the image (often corresponding to edges), which helps to preserve locations of abrupt changes in concentration.

[0040] Therefore, by combining the data fitting term and the spatial smoothing term, we obtain the overall energy functional:

[0041] ;

[0042] in, , which are the calculation coefficients used to balance the importance of data items and smoothing terms. This represents the energy functional for estimating the overall concentration. Solving the above equation... Perform gradient descent updates:

[0043] ;

[0044] in: This represents the number of iterations. The step size. The concentration is forced to be non-negative. A preliminary concentration field is obtained after multiple iterations. , The energy functional constructed for the overall concentration field. For the first The gas concentration variable to be estimated in the next iteration. and They represent the first Subsequent The concentration value of the next iteration.

[0045] In this embodiment, the above inversion process relies on ,and It is difficult to measure directly, therefore a learnable estimation model is needed. (Given a topographic elevation map.) and a set of local features extracted from the image (For example, texture intensity, etc.). Construct a function:

[0046] ;

[0047] in, This represents a learnable path length estimation model. This is a topographic elevation map. For local features, This is the calibration loss function for the path length estimation model. It is a set of parameters that needs to be fitted using calibration data. During the calibration phase, a small number of calibration values ​​containing the actual path length are used. In the scenario where the loss is minimized:

[0048] .

[0049] To reflect the impact of terrain complexity on the reliability of path estimation, a path confidence field is introduced. Constructed based on height gradient:

[0050] ;

[0051] in, This is the confidence decay coefficient. Controlling the rate at which confidence level changes with terrain undulation is used to control the impact of height gradient on the rate of confidence level decay; a larger gradient will cause the confidence level in areas with terrain undulation to decrease rapidly. Indicates the corrected effective path length; Indicates the reference path length. Indicates the magnitude of height variation; in the fundamental overall energy functional Replace with The final energy functional is obtained. Finally, the path length is corrected using the confidence level.

[0052] ;

[0053] in, It is an empirical reference value, which may vary depending on the terrain. When the value is very small, the path length is closer to the reference value. In actual inversion, it is only necessary to include the path length in the overall energy functional. Replace with This can significantly improve the stability of concentration estimation.

[0054] After completing the single-frame concentration inversion, consider continuous time. ,time The concentration field is denoted as The corresponding wind speed field is denoted as . for Always The horizontal component at the location. for Always Vertical component at location.

[0055] Small time step Neglecting chemical reactions, the approximate conservation equation for gas transport is as follows:

[0056] ;

[0057] In the formula: This is the spatially dependent diffusion coefficient. This refers to the source term per unit time and unit volume (which can be generated by contributions from leaks, etc.). This indicates the next time step predicted by the model. Gas concentration at location This represents the spatial gradient operator, used to describe the rate of change of a physical quantity in space;

[0058] The above formula requires known information. However, wind speed cannot be directly measured from images. Therefore, a cost function is constructed by combining image brightness consistency and concentration conservation error. First, it is approximated that in the short term, air mass movement is mainly driven by... Therefore, the pixel grayscale should satisfy:

[0059] ;

[0060] in, for Time pixel position The brightness value of the observed image at that location, for The brightness value of the observed image at that time. The step size.

[0061] Based on this definition, brightness consistency error is:

[0062] ;

[0063] Secondly, construct the conserved error term:

[0064] .

[0065] To ensure that the wind speed field is spatially smooth while still allowing for structures such as vortices, its gradient is canonically constrained using the following formula:

[0066] ;

[0067] in, To prevent the gradient from becoming numerically unstable at zero, a small constant is used. Combining the above three factors, the total energy is defined as:

[0068] ;

[0069] in, For brightness consistency error, For the conserved error term, For regular expression constraints, For total energy, , representing the coefficients of the conservation error term and the regularization constraint, controlling the importance of the conservation error and regularization term. Estimating the wind speed field involves solving:

[0070] ;

[0071] Iterative updates are used during implementation:

[0072] ;

[0073] in, For wind speed field estimation, For wind speed field variables, To update the step size. Obtained after iteration convergence. , This indicates the final optimized wind speed field. The value of the position, Indicates the first iteration during the iteration process Step wind speed field, This is for updating the step size.

[0074] Yes Then, based on the conservation equation, an explicit prediction step is constructed to obtain a rough estimate of the concentration at the next time step:

[0075] ;

[0076] This formula represents the predicted concentration at the next time step after residual correction; the formula depends on the... and The rough assumptions are difficult to fully reflect reality. Therefore, a residual operator is introduced. It learns the shortcomings of the physical model:

[0077] ;

[0078] These are the parameters of the residual operator (e.g., convolution kernel, weights, etc.). It represents environmental structural information, such as building masks, equipment locations, etc. This represents the current concentration field. (For training purposes...) Assuming there is a set of historical or simulated "real" concentration sequences First, define the data error loss:

[0079] ; Indicates data error loss. For discrete time step index;

[0080] Meanwhile, to ensure that the learned evolutionary operator still roughly satisfies the conservation relationship, a physical consistency loss is introduced:

[0081] ;

[0082] This represents the loss of physical consistency.

[0083] The ultimate training goal is:

[0084] ;

[0085] in, This represents the overall objective function for training the evolution operator. Weights that control physical consistency. By minimizing the above equation, the resulting evolution operator has two advantages: on the one hand, it explicitly embeds the physical structure of transport-diffusion; on the other hand, it compensates for modeling errors through residual learning.

[0086] With predictive capabilities in place, the next step is to actively intervene in the diffusion process using multiple controllable devices (fans, spray systems, etc.) to suppress the concentration in key areas to a safe level. (Time steps should be considered.) , Let the system state be a natural number, defined at each moment:

[0087] ;

[0088] in: For time steps The system status, Current concentration field Environmental parameters, such as overall wind direction, wind speed, and temperature, are available on-site. The controllable device has its motion vector set as follows:

[0089] ;

[0090] Each component Indicates the first The power, opening and closing angle, and other control parameters of the equipment. To ensure feasibility, the actions must meet the following constraints:

[0091] ;

[0092] No. The minimum permissible action of an execution unit represents the minimum control input allowed for the device under safety and functional constraints. This lower limit is typically determined by device structural limitations, minimum operating thresholds, or safety specifications.

[0093] No. The maximum permissible action of an execution unit represents the maximum control input allowed by the device without causing overload, damage, or safety risks. This upper limit is used to prevent the control strategy from issuing unexecutable or dangerous instructions.

[0094] Construct a parameterized decision function Mapping states to actions:

[0095] ;in Let these be the parameters to be optimized (e.g., linear weights or neural network parameters). Under controlled conditions, the concentration evolution operator needs to consider device effects. The source term can be written as:

[0096] ;

[0097] in: This refers to natural leakage or background source items. Description of the The device is positioned The effect of gas transport (obtained experimentally). Substitute the above equation into the previous explicit step equation and continue using the residual operator. ,get:

[0098] .

[0099] Based on the previous definitions of state and action, given the decision parameters After that, the entire system was determined from Evolved to In this way, For time steps The system state. To measure the control effect, first define a region of interest. (e.g., areas with high concentration of people), and set spatial weights. Different positions are assigned different levels of importance.

[0100] At any moment Given an action back, It can be calculated from the above formula. Based on this, the instantaneous loss function is defined as follows:

[0101] ;

[0102] in, For a moment The immediate loss; Safety concern areas A spatial weighting function for the safety concern area is used to measure the importance of different spatial locations to the safety objective; This is weighted by the cost of actions, reflecting the costs such as energy consumption and equipment wear. In real-world environments, model errors and random disturbances (such as sudden wind direction changes) exist. These uncertainties are represented by a disturbance variable, and the system operates under these disturbances. The total loss generated is:

[0103] ;

[0104] To emphasize safety in the worst-case scenario, a simple risk metric is introduced: [the risk metric is defined within a set of possible perturbations]. Internally, focus on the penalized "near-worst-case" scenario:

[0105] ;

[0106] In the formula: It measures disturbance The degree of "deviation from the normal state", such as the distance from a certain reference distribution. Tolerance to extreme disturbances is controlled. The above formula shows that even in rare but dangerous situations, the optimized control strategy will tend to reduce losses in the worst-case scenario.

[0107] In this embodiment, to ensure that the concentration in the critical area does not exceed the specified threshold... Add constraints at each time step :

[0108] ;

[0109] Adding the above equation as a soft constraint to the objective function yields a penalized overall index:

[0110] ;

[0111] in, To constrain the penalty coefficient, the policy parameters are finally updated using gradient descent or a similar numerical optimization method. This represents the cumulative loss under disturbance conditions; This represents the objective function for risk perception. This represents the disturbance deviation function. For the expectation of the perturbation variable, Indicates the optimization goal of security perception:

[0112] ;

[0113] in, , This represents the number of outer iterations. The resulting decision function... It can balance average control performance and safety under extreme conditions in multiple disturbance scenarios. , The first Second and third The strategy parameters for the next iteration. Update the step size for the strategy parameters; This represents the gradient of the objective function with respect to the policy parameters.

[0114] This invention proposes a gas diffusion imaging processing method that considers path confidence correction. By introducing a diffusion inversion model, a concentration-wind field joint evolution operator, and a risk-driven control strategy, it significantly improves the reliability and application value of gas monitoring and diffusion control technologies, providing an innovative system solution for environmental monitoring and emergency response. By addressing the problems of imaging errors, unstable diffusion prediction, and uncontrollable intervention in traditional technologies, this method offers a new technical path for gas diffusion suppression and risk management, with particularly broad application prospects in industrial monitoring and urban public safety.

[0115] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0116] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. This computer software product, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0117] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A gas diffusion imaging processing method considering path confidence correction, characterized in that, Includes the following steps: Step S1: Calculate the link input, establish the relationship between observation and concentration using a nonlinear diffusion imaging model, improve inversion stability using path length estimation and confidence correction, and obtain the concentration field through energy functional optimization; Step S2: Based on the concentration sequence and image sequence, the wind speed field is estimated by combining brightness consistency and conservation error. Then, the physical step and learning residual are used to obtain an evolution operator that combines physical and data constraints to predict the concentration in the next few steps. Step S3: Embed the prediction module into system dynamics to describe the impact of equipment actions on concentration, configure risk measurement and design and optimize control strategies to maintain a safe concentration in key areas under various disturbances; Step S1, establishing the relationship between observation and concentration, includes: Step S11: Several bands are set. Image, pixels Position is denoted as two-dimensional coordinates , , These are two-dimensional coordinate values; in the band The observed values ​​are denoted as Under rarefied gas conditions, observations in each band are decomposed into a background term. Gas diffusion / absorption terms and noise items The basic imaging relationship is expressed as follows: ； Step S12: Using a power-law nonlinear diffusion model, the diffusion term is related to the volume concentration. By introducing the gas concentration at the ground corresponding to the pixel and the effective path length of the imaging ray through the gas cloud, we obtain: ； in, An index representing the effect of path length on diffusion; Indicates the effective path length; Indicates gas concentration. , indicating band The diffusion intensity coefficient; , which represents the exponential response of concentration to this band; Improving inversion stability in step S1 includes: Step S13: Introduce the exponential effect of path length on diffusion, substitute the nonlinear diffusion model formula into the basic imaging relationship, and incorporate the noise term into the residual, to obtain: ； Step S14: In pixels Assuming the noise is a zero-mean Gaussian, and using the least squares estimation method, the pixel density is transformed into an optimization problem, expressed as: ； This represents the pixel-level local concentration estimate. The gas concentration variable to be estimated; Step S15: Combine spatial neighborhood, set For pixels The neighborhood, Representing neighboring pixels, edges are preserved and a weighted smoothing term is introduced: ； in, For spatial smoothing regularization; Represents the concentration within the neighborhood; weight Specifically, it is set as follows: ; , represents the adjustment constant; Indicates the first Pixels in each band Observation intensity at the location; Indicates the first Pixels in each band Observation intensity at the location; Step S1, including energy functional optimization and concentration field calculation, includes: Step S16: Combining the data fitting term and the spatial smoothing term, the basic overall energy functional is obtained, expressed as: ； in, , which are the calculation coefficients used to balance the importance of data items and smoothing items; This represents the energy functional for estimating the overall concentration. Step S17: Solve the overall energy functional and... After performing gradient descent updates, we get: ； and They represent the first Subsequent The concentration value of the next iteration. The energy functional constructed for the overall concentration field; For the number of iterations, Step size, The forced concentration is non-negative. For the first The gas concentration variable to be estimated in the next iteration; Step S18: Establish a learnable estimation model and construct a function. The calibration phase utilizes calibration values ​​with actual path lengths. In the scenario where the loss is minimized, we obtain: ； This represents a learnable path length estimation model. This is the calibration loss function for the path length estimation model. This is a topographic elevation map. Local features; Step S19: Introduce a path confidence field Constructed based on height gradient The confidence level is used to correct the path length, resulting in: ； in, , This is the confidence decay coefficient; Indicates the corrected effective path length; Indicates the reference path length. This indicates the magnitude of the change in height.

2. The gas diffusion imaging processing method considering path confidence correction as described in claim 1, characterized in that, The calculation of the wind speed field in step S2 includes: Step S21: Time The concentration field is denoted as The corresponding wind speed field is denoted as In step length Neglecting chemical reactions, we obtain an approximate conservation equation for gas transport: ； in, The diffusion coefficient is spatially dependent. For source terms per unit time and per unit volume, for Always The horizontal component at the location. for Always Vertical component at location This indicates the next time step predicted by the model. Gas concentration at location This represents the spatial gradient operator, used to describe the rate of change of a physical quantity in space. For wind speed field; Step S22: Construct a cost function by combining image brightness consistency and density conservation error, and express the pixel grayscale as... The brightness consistency error is defined as: ； in, for Time pixel position The brightness value of the observed image at that location, for Observe the image brightness value at all times; Step S23: Construct the conserved error term, expressed as: ； Simultaneously, its gradient is regularized using the following formula, resulting in: ;in, This is a constant used to prevent the gradient from becoming numerically unstable at zero. Step S24: The total energy is defined as follows: ； in, For brightness consistency error, For the conserved error term, For regular expression constraints, For total energy, , representing the coefficients of the conservation error term and the regularization constraint; The estimated wind speed field obtained by solving is expressed as: Iterative update ;in, To update the step size, the iteration converges to obtain... ; For wind speed field estimation, For wind speed field variables, Indicates the first iteration during the iteration process The wind speed field of the step; This indicates the final optimized wind speed field. The value of the position.

3. The gas diffusion imaging processing method considering path confidence correction as described in claim 2, characterized in that, The prediction of concentrations for future steps in step S2 includes: Step S25: Based on Based on the conservation formula, an explicit prediction step is constructed to obtain the concentration at the next time step, expressed as: ； This indicates the predicted concentration at the next time step after residual correction. Step S26: Introduce the residual operator ,get: ； in, For the parameters of the residual operator, Indicates environmental structure information. This represents the current concentration field. The parameters of the residual operator are trained on historical or simulated real concentration sequences. Define data error loss: ; Indicates data error loss. Index for discrete time steps; Step S27: Introduce physical consistency loss: ； The physical consistency loss is represented; the final training objective is obtained and minimized to obtain the evolution operator; where the final training objective is: ; Let represent the overall objective function for training the evolution operator; where Weights that control physical consistency.

4. The gas diffusion imaging processing method considering path confidence correction as described in claim 3, characterized in that, In step S3, the prediction module is embedded in system dynamics to describe the impact of device actions on concentration, including: Step S31: Consider the time step , Let be a natural number, and define the system state at each time step as . ,in, Indicates the current concentration field; Represents environmental scalars; Step S32: Definition The motion vector of the controllable device is: , where each component Indicates the first The control quantities and actions of the equipment all satisfy the box constraints. No. The minimum allowed actions for each execution unit; No. The maximum allowed actions per execution unit; Step S33: Construct a parameterized decision function Mapping states to actions is represented as follows: ;in, These are the parameters to be optimized. Step S34: Considering the device effect and concentration evolution operator, continue using the residual operator. ,get: ； Among them, source terms , For natural leakage or background source terms; Description of the The device is positioned The impact of gas transport.

5. A gas diffusion imaging processing method considering path confidence correction as described in claim 4, characterized in that, The risk measurement design and optimization control strategy in step S3 specifically includes: Define the instantaneous loss function as: ;in, , which is the weight of the action cost; For a moment The immediate loss; Safety concern areas A spatial weighting function for the safety concern area is used to measure the importance of different spatial locations to the safety objective; The computing system under disturbance The total loss generated is: ; In a set of possible perturbations Internally, risk measurement is introduced: ； in, This represents the disturbance deviation function; Tolerance for extreme disturbances should be controlled. This represents the cumulative loss under disturbance conditions; This represents the objective function for risk perception. Add constraints at each time step , To define a threshold; By adding constraints to the objective function, we obtain a penalized overall index, expressed as: ； This indicates the goal of optimizing security perception. , is the constraint penalty coefficient. Let be the expectation of the perturbation variable; Updating policy parameters using gradient descent or similar numerical optimization methods is represented as: ； in, The outer iteration number, , The first Second and third The strategy parameters for the next iteration. Update the step size for the policy parameters. This represents the gradient of the objective function with respect to the policy parameters. For the first The value of the security awareness optimization objective function is calculated in the next iteration.

Images