Rapid simulation method and system for aerosol diffusion
By performing high-precision computational fluid dynamics simulation and chi-square hybrid model fitting in an indoor environment, the problems of high precision and computational efficiency in aerosol diffusion simulation are solved, enabling rapid and accurate prediction of aerosol concentration. This method is suitable for emergency decision-making and training exercises in nuclear facilities and industrial production.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NAVAL UNIV OF ENG PLA
- Filing Date
- 2026-01-15
- Publication Date
- 2026-05-05
AI Technical Summary
Existing technologies cannot simultaneously meet the requirements of high accuracy and computational efficiency in aerosol diffusion simulation in indoor environments. Computational fluid dynamics simulation has low computational efficiency, while Gaussian models have insufficient accuracy in indoor environments, failing to meet the real-time and accuracy requirements of practical applications.
Aerosol concentration data is obtained through a high-precision computational fluid dynamics simulation, discretized into a cubic mesh, a baseline mesh is selected and fitted using a chi-square hybrid model, and parameters are optimized by combining a multi-agent genetic algorithm to construct a spatial scale database, thereby achieving rapid aerosol concentration prediction.
It achieves high-precision aerosol diffusion simulation in indoor environments, reduces computational complexity by several orders of magnitude, and takes less than 2 milliseconds for single-point prediction. It maintains spatial generalization accuracy consistent with high-precision fluid dynamics data and has the ability to respond quickly and adapt to changes in the scene.
Smart Images
Figure CN121980997A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of numerical simulation technology for air pollutant diffusion, specifically relating to a rapid simulation method and system for aerosol diffusion. Background Technology
[0002] In various scenarios such as nuclear facility operation and industrial production, the precise control of the diffusion behavior of radioactive materials leaking or harmful aerosols being released in indoor environments directly affects the scientific nature of emergency decision-making, the safety of personnel evacuation, and the effectiveness of training exercises. Rapidly and accurately simulating the diffusion trajectory and distribution of aerosols in indoor spaces has become a key technological support for ensuring public and environmental safety.
[0003] Currently, indoor aerosol diffusion simulation mainly relies on two technical approaches. One is computational fluid dynamics (CFD) simulation. This method, based on fundamental fluid dynamics equations, can provide high-fidelity numerical solutions to complex flow fields and particulate transport processes. Due to its excellent simulation accuracy, it has become a standard tool in related scientific research fields. However, the computational process of this technique involves complex mesh generation and iterative solution steps, resulting in low computational efficiency. A single three-dimensional transient simulation often requires a significant amount of time, leading to high computational costs. This limitation makes CFD simulation unsuitable for scenarios requiring real-time interaction or rapid response, and it cannot meet the real-time requirements of practical applications such as emergency training.
[0004] Another technical approach involves empirical or analytical models such as Gaussian plumes. These models, based on simplified assumptions and analytical solutions, offer extremely fast computation speeds and can quickly output simulation results. However, the derivation of these models typically assumes unbounded or semi-infinite spaces, making them primarily applicable to open atmospheric environments. In indoor environments, the presence of walls, various obstacles, and ventilation systems results in highly non-uniform flow fields, with complex backflow and vortex phenomena forming within the space. This renders the basic assumptions of the Gaussian model invalid in indoor environments. Directly applying such models to indoor aerosol diffusion simulations significantly increases the error in the simulation results, failing to accurately reflect the true diffusion patterns of pollutants indoors and thus failing to meet the accuracy requirements of practical applications.
[0005] In summary, existing technological approaches present a clear contradiction: high-precision computational fluid dynamics simulations cannot meet real-time requirements, while empirical or analytical models that can meet real-time requirements lack sufficient accuracy in indoor environments. Therefore, developing a rapid simulation method that can balance simulation accuracy and computational efficiency in indoor environments has become a critical technical problem that urgently needs to be solved in this field. Summary of the Invention
[0006] The present invention provides a rapid simulation method and system for aerosol diffusion, which can effectively solve the problems in the background art.
[0007] This invention provides a rapid simulation method for aerosol diffusion, comprising the following steps:
[0008] S1: Obtain time-series data of aerosol concentration at various points in the target space within a preset time period through computational fluid dynamics simulation;
[0009] S2: Discretize the target space into a cubic grid of a preset scale, and calculate the average concentration of each grid at each time step;
[0010] S3: Select a grid that meets the preset conditions from all grids as the reference grid, calculate the average ratio of the concentration time series data of each of the remaining grids to the concentration time series data of the reference grid over time, and use it as the spatial scale coefficient of the grid. Then, construct a spatial scale database that does not change over time using the spatial scale coefficients of all grids.
[0011] S4: Using the concentration time-series data from the baseline grid in step S3 as the training set, a chi-square mixture model is used to fit the training set. The time-series function expression is as follows:
[0012]
[0013] in, For model components,
[0014] The first The weight parameters, degrees of freedom parameters, location parameters, and scale parameters of each component;
[0015] S5: For any target location in space, retrieve the corresponding spatial scale coefficient from the spatial scale database based on its grid location. and the obtained spatial proportion coefficient The timing function obtained in step 4 Multiplying these together yields the predicted concentration time series for the target location. .
[0016] As a further optimization of the present invention, in step S2, the simulation area is focused on the target height range, and the space within the height range is discretized into a cubic grid of a preset scale.
[0017] As a further optimization of the present invention, the preset condition in step S3 is that the concentration time series of the grid is higher than a set threshold throughout the entire simulation time period, and its cumulative concentration value is the largest within a preset search area.
[0018] As a further optimization of the present invention, it also includes the step of using an optimization algorithm to globally optimize the parameters of the chi-square mixture model in step S4 to obtain the fitted time-series function.
[0019] As a further optimization of the present invention, the optimization algorithm is a multi-agent genetic algorithm.
[0020] A rapid aerosol diffusion simulation system includes a memory and a processor; the memory stores instructions that cause the processor to execute the method.
[0021] This invention provides a rapid aerosol diffusion simulation method and system that can balance simulation accuracy and computational efficiency in an indoor environment. This invention solidifies the spatial diffusion model through a high-precision computational fluid dynamics simulation in the initial stage, and subsequent calculations are merely simple function evaluations and multiplication operations, reducing computational complexity by several orders of magnitude while maintaining spatial generalization accuracy highly consistent with computational fluid dynamics data. This fundamentally solves the technical contradiction of the trade-off between speed and accuracy. Attached Figure Description
[0022] Figure 1 This is a schematic diagram of the method flow of Example 1;
[0023] Figure 2 This is a comparison chart of aerosol curves for the grid (15, 8, 1) in Example 1;
[0024] Figure 3 This is a comparison chart of aerosol curves for the grid (11, 1, 1) in Example 1;
[0025] Figure 4 This is a comparison chart of aerosol curves for the grid (17, 9, 2) in Example 1;
[0026] Figure 5 This is a comparison chart of aerosol curves for the grid (22, 17, 2) in Example 1;
[0027] Figure 6 This is a comparison chart of aerosol curves for the grid (26, 14, 3) in Example 1;
[0028] Figure 7 This is a comparison chart of aerosol curves for the grid (12, 11, 3) in Example 1;
[0029] Figure 8 This is a schematic diagram comparing the accuracy of Example 1 and the Gaussian mixture model;
[0030] Figure 9 This is a schematic diagram comparing the accuracy of Example 1 and the fitting of a 5th-order polynomial. Detailed Implementation
[0031] Example 1
[0032] This embodiment describes a simulated training scenario of radioactive aerosol leakage applied to a typical rectangular room. The room is 21m long, 8.5m wide, and 3m high. The room has a fixed ventilation opening, and the aerosol release source is located on the ground at the center of the room.
[0033] like Figure 1 As shown, the rapid simulation method in this embodiment is implemented according to the following steps:
[0034] S1: First, a high-fidelity transient simulation of the scenario in this embodiment is performed using fluid dynamics simulation software. Specifically, the software used is ANSYS Fluent, the simulation duration is 900 seconds, and the output concentration field data is 1 second apart. The time series data of aerosol concentration at each point in the three-dimensional space within the preset time period is obtained.
[0035] S2: The simulation area focuses on the target height range. In this embodiment, only data within 1.5 meters above the ground is retained. 1.5 meters above the ground is approximately the height of the breathing zone.
[0036] The region within the target height range is divided into a 0.5m × 0.5m × 0.5m cubic grid, totaling 42 × 17 × 3 = 2142 grids. The concentration value of each grid at time t is obtained by averaging the concentrations of all the fine computational fluid dynamics grids within it, thus obtaining the average concentration of each grid at each time, denoted as .
[0037] .
[0038] S3: Select a grid that meets the preset conditions from all grids as the reference grid.
[0039] Furthermore, the preset condition can be selected as follows: the concentration time series of the grid is always higher than the set threshold throughout the entire simulation period, and its cumulative concentration value is the largest within the preset search area.
[0040] In this embodiment, within the main diffusion area near the leakage source, i.e., within ±4m in the X direction, the grid with the highest cumulative concentration value in the second and third grid layers is searched. In this embodiment, the grid with index (18,9,3) is selected as the reference grid. Its concentration sequence is .
[0041] For each grid Calculate its spatial scaling factor relative to the reference grid:
[0042]
[0043] in This refers to the time period after diffusion enters its fully developed stage, for example, the period from t>50s until its end. All A static spatial proportion database is constructed and stored in the system.
[0044] S4: Concentration sequence based on the baseline grid For training data, a chi-square mixture model is used to fit the training set. The chi-square mixture model is a linear combination of multiple chi-square distribution functions that have been translated and scaled.
[0045] When three characteristic peaks are observed in the time series, the number of components in the chi-square mixture model is set. The model is in the following form:
[0046]
[0047] in, For model components, The first The weight parameters, degrees of freedom parameters, location parameters, and scale parameters of each component.
[0048] Furthermore, this embodiment also includes a step of using an optimization algorithm to globally optimize the parameters of the chi-square mixture model to obtain a fitted time-series function, which minimizes the error between the model output and the training set data.
[0049] The optimization algorithm specifically employs a multi-agent genetic algorithm, that is, it uses a multi-agent genetic algorithm to optimize the parameter set. The optimization is performed, with the objective function being to minimize the model output. With real data
[0050] The root mean square error between them. The population size of the multi-agent genetic algorithm is set to 10×10, the crossover probability is 0.7, the mutation probability is 0.15, and the evolution is 1200 generations.
[0051] The optimal parameters obtained after optimization are shown in the table below:
[0052]
[0053] The times in the table are dimensionless times scaled by 30:1. In actual applications, they should be restored proportionally.
[0054] At this point, the root mean square error is Coefficient of determination This indicates that the fitting effect is excellent.
[0055] S5: For any target location in space, retrieve the corresponding spatial scale coefficient from the spatial scale database based on its grid location. The spatial proportion coefficient obtained from the query With the obtained timing function Multiplying these together yields the predicted concentration time series for the target location. .
[0056] Specifically, when a simulation training system needs to acquire a point in space... , Located in the grid Inside, the system performs the following operations:
[0057] A1: Based on the point The coordinates are used to locate the grid to which it belongs. .
[0058] A2: Read the scale factor corresponding to this grid from the pre-stored spatial scale database. .
[0059] A3: Call the chi-square hybrid model function with solidified parameters. Calculate the current simulation time The corresponding baseline concentration value.
[0060] A4: Multiply the two: The result is the point. exist Predicted concentration at any given time.
[0061] The calculation process involves only one database query and one function calculation. According to the test, taking 900 time points as an example, the total time consumption of single-point full-time sequence calculation on an ordinary computer is less than 2 milliseconds, achieving a true millisecond-level response.
[0062] like Figure 2-7 As shown, 30 verification points were randomly selected in the space, and predictions were made using the method of this embodiment. The predicted concentration curves showed a high degree of agreement with the original computational fluid dynamics data, with an average coefficient of determination of [missing information].
[0063] The result of 0.86 demonstrates the excellent spatial generalization ability of the method in this embodiment.
[0064] like Figure 8-9 As shown, the method of this embodiment is compared with that of Gaussian mixture model and fifth-order polynomial fitting. The root mean square error of this method is... Much lower than Gaussian mixture models and fifth-order polynomials Furthermore, it strictly maintains non-negativity at the end of the curve, demonstrating higher precision and physical rationality.
[0065] The method in this embodiment also has the ability to adapt quickly to changes in the scene.
[0066] Assuming that in actual training, the leak source containment time is reduced from 120 seconds in the baseline scenario to 60 seconds, the method in this embodiment does not require rerunning the computational fluid dynamics; the operator only needs to add the time series model to the system. The components corresponding to the later main peak, such as the weight parameters of components 2 and 3. A prediction model for a new scenario can be generated by using a scaling factor of 0.5, which simulates a scenario where the leakage is halved.
[0067] The prediction results of the adjusted model for the new scenario of a 60-second lockdown were compared with the results of a computational fluid dynamics simulation specifically rerun for this scenario. The two results showed a high degree of consistency in trends. This demonstrates that the simulation system can quickly respond to changes in the scene by simply adjusting the parameters, which greatly improves its practicality and flexibility.
[0068] This embodiment solidifies the spatial diffusion model through a high-precision computational fluid dynamics simulation in the early stage. Subsequent calculations are merely simple function evaluations and multiplication operations, reducing computational complexity by several orders of magnitude. Experimental results show that this method achieves a single-point, full-time-series computation time of less than 2 milliseconds, representing an improvement of approximately [missing information - likely a number] hours compared to traditional simulations of 5-10 hours.
[0069] It achieves a multiple of [number] times higher accuracy while maintaining spatial generalization accuracy highly consistent with computational fluid dynamics data, validating the average coefficient of determination at verification points. This fundamentally solves the technical contradiction that speed and accuracy cannot be achieved simultaneously.
[0070] This embodiment boasts a solid physical foundation and excellent model adaptability. The spatiotemporal decoupling strategy based on quasi-steady-state physical observations gives the method clear physical meaning, rather than being a black-box model. This embodiment employs a chi-square mixture model as the time-series fitting tool; its non-negative, right-skewed mathematical properties closely match the physical processes of aerosol concentration diffusion and decay, ensuring the physical rationality of the simulation results and a high goodness of fit. The baseline point... .
[0071] This embodiment exhibits good adjustability and practicality. The parameters of the chi-square mixture model in this embodiment, such as weights... It has intuitive physical meaning and can be quickly adapted to changes in boundary conditions such as leakage intensity and containment time through simple adjustments, without the need to rerun computational fluid dynamics, which greatly enhances the practicality and flexibility of the method in dynamic training scenarios.
[0072] Example 2
[0073] This embodiment specifically includes a memory and a processor. The memory stores instructions that cause the processor to execute the method in embodiment 1.
[0074] It should be noted that memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.
Claims
1. A rapid simulation method for aerosol diffusion, characterized in that, Includes the following steps: S1: Obtain time-series data of aerosol concentration at various points in the target space within a preset time period through computational fluid dynamics simulation; S2: Discretize the target space into a cubic grid of a preset scale, and calculate the average concentration of each grid at each time step; S3: Select a grid that meets the preset conditions from all grids as the reference grid, calculate the average ratio of the concentration time series data of each of the remaining grids to the concentration time series data of the reference grid over time, and use it as the spatial scale coefficient of the grid. Then, construct a spatial scale database that does not change over time using the spatial scale coefficients of all grids. S4: Using the concentration time-series data from the baseline grid in step S3 as the training set, a chi-square mixture model is used to fit the training set. The time-series function expression is as follows: in, For model components, The first The weight parameters, degrees of freedom parameters, location parameters, and scale parameters of each component; S5: For any target location in space, retrieve the corresponding spatial scale coefficient from the spatial scale database based on its grid location. and the obtained spatial proportion coefficient The timing function obtained in step S4 Multiplying these together yields the predicted concentration time series for the target location. .
2. The rapid simulation method for aerosol diffusion according to claim 1, characterized in that, In step S2, the simulation area is focused on the target height range, and the space within the height range is discretized into a cubic mesh of a preset scale.
3. The rapid simulation method for aerosol diffusion according to claim 1, characterized in that, In step S3, the preset condition is that the concentration time series of the grid is higher than a set threshold throughout the entire simulation time period, and its cumulative concentration value is the largest within the preset search area.
4. The rapid simulation method for aerosol diffusion according to claim 1, characterized in that, It also includes the step of using an optimization algorithm to globally optimize the parameters of the chi-square mixture model in step S4 to obtain the fitted time-series function.
5. The rapid aerosol diffusion simulation method according to claim 4, characterized in that, The optimization algorithm is a multi-agent genetic algorithm.
6. A rapid aerosol diffusion simulation system, characterized in that, It includes a memory and a processor; the memory stores instructions that cause the processor to perform the method according to any one of claims 1-5.