Artificial intelligence-based fast-expanding cloud chamber environment simulation method and system

Abstract

The application discloses a rapid expansion cloud chamber environment simulation method and system based on artificial intelligence, comprising: obtaining experimental data and experimental conditions of a to-be-simulated expansion cloud chamber, structurally classifying according to the experimental conditions according to the expansion stages based on the experimental data, obtaining feature vectors of each stage, carrying out meta-learning through a neural network according to the feature vectors, obtaining a cloud and fog simulation model and a simulation error, identifying a cloud chamber working condition interval based on the simulation error and the experimental data, obtaining an output result of the cloud and fog simulation model in a high deviation interval according to the working condition interval, correcting interval results through a CFD equation based on the output result, and obtaining cloud chamber environment simulation results. The method splits the cloud and fog expansion stages, fits the fog droplet growth process, carries out CFD local correction on the simulation high deviation interval, improves the prediction error control precision, improves the prediction and collaborative analysis efficiency of the fog formation process in a complex environment, and has good interpretability.

Description

Technical Field

[0001] This invention relates to the field of inflatable cloud chambers, and more particularly to a method and system for simulating rapid inflatable cloud chamber environments based on artificial intelligence. Background Technology

[0002] Rapidly expanding cloud chambers are crucial equipment for simulating cloud formation and precipitation environments in the atmosphere, studying aerosol particle nucleation processes, and understanding the mechanisms of weather modification catalysts. By precisely controlling abrupt changes in parameters such as pressure and temperature, the expanding cloud chamber creates a supersaturated water vapor environment, thereby analyzing the microscopic physical processes of clouds and fog. This provides important data for optimizing artificial rain enhancement operations. Currently, various theories have emerged regarding the application of expanding cloud chambers in the study of warm cloud artificial rain enhancement mechanisms, with a focus on cold cloud processes and ice nuclei research. Various isothermal cloud chambers and integrated cloud chambers have been established. To simulate the physical environment of clouds and fog, experimental conditions are set according to required environmental data, and the cloud chamber is controlled through these conditions. Due to the complex characteristics of cloud and fog physics, experiments cannot completely cover all environmental scenarios. However, with the integration of artificial intelligence technology into fields such as fluid simulation, machine learning algorithms are gradually being introduced to achieve adaptive optimization control of environmental parameters, improving simulation accuracy and experimental efficiency.

[0003] However, the integration of artificial intelligence with rapidly expanding cloud chambers is still in its early stages, and most applications focus on post-hoc analysis of single parameters. There is a lack of application for dynamic correlation of multiple parameters, boundary effect compensation, and real-time control. The lack of targeted analysis of specific physical processes in cloud chambers can easily lead to problems such as parameter mutations and nonlinear responses in the analysis results. Therefore, environmental simulation of rapidly expanding cloud chambers, intelligent prediction of fog formation processes in complex environments, and collaborative analysis of multiple physical quantities to improve simulation accuracy and experimental efficiency have become urgent problems to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for simulating rapidly expanding cloud chamber environments based on artificial intelligence.

[0005] To achieve the above objectives, the present invention is implemented according to the following technical solution: The first aspect of this invention provides a method for simulating a rapidly expanding cloud chamber environment based on artificial intelligence, comprising: The experimental data and conditions of the expanding cloud chamber to be simulated are obtained, and the experimental data are preprocessed. The experimental data includes the temperature field, pressure field, droplet spectrum and wall heat flux density inside the chamber. Based on the experimental data, the data is structured and classified according to the expansion stage according to the experimental conditions to obtain a time series dataset containing the cooling stage, expansion stage and maintenance stage, and feature vectors of each stage are extracted. Based on the feature vectors, a physical information neural network is constructed using the dissipative Hamiltonian equation. Meta-learning is then performed on the neural network to obtain a cloud and fog simulation model and simulation error. Based on the simulation error and experimental data, the cloud chamber operating condition range is identified, and the output results of the cloud and fog simulation model in the high deviation range are obtained according to the operating condition range. The identification conditions include the strong dissipation range of clouds and fog and the phase transition critical point. The output results are used as boundary conditions. The interval results are corrected by CFD equations based on the boundary conditions. The low-bias interval results and the interval correction results are then combined to obtain the cloud room environment simulation results.

[0006] Furthermore, the method for obtaining the experimental data and experimental conditions includes: Based on the expansion chamber to be simulated, temperature distribution data, pressure change data, droplet spectrum and wall heat flux density are collected in the chamber according to the expansion cycle and unified into the same spatiotemporal coordinate system. Wavelet filtering is performed on the temperature distribution data and pressure change data to obtain the temperature field and pressure field. The temperature field, pressure field, droplet spectrum and wall heat flux density are converted into a structured time series format. The wall heat flux density is the heat exchange rate between the cloud chamber wall and the gas in the chamber per unit area. Based on the structured time-series data, the temperature, humidity, and pressure of the corresponding cloud chamber control system are used as experimental conditions. The structured time-series data is correlated with the experimental conditions to obtain experimental data.

[0007] Furthermore, the method for obtaining the time-series dataset includes: Based on the pressure field, the cloud chamber expansion cycle is divided using the pressure change rate threshold method. The stage from the initial pressure, where the pressure change rate is negative and the absolute value gradually increases, is marked as the cooling stage. The stage where the pressure change rate reaches its peak and then rapidly decreases is marked as the expansion stage. The stage where the pressure change rate approaches zero and the pressure is stable is marked as the maintenance stage. The experimental data are structured and classified according to the cooling, expansion and maintenance phases to obtain time-series datasets for each phase, and feature vectors for each phase are extracted based on the time-series datasets.

[0008] Furthermore, the method for obtaining the feature vector includes: The average temperature change rate, peak wall heat flux density, pressure drop rate, and difference between initial temperature and initial pressure are extracted from the time series dataset of the cooling section. The saturated vapor pressure is calculated using the Clausius-Clapeyron equation based on the temperature field of the time series dataset in the expansion section. The ratio of the actual vapor pressure to the saturated vapor pressure in the pressure field is taken as the supersaturation. Based on the expansion segment, an ordered list is obtained from discrete measurements of the average droplet radius over time in the droplet spectrum. Central differencing is performed on the midpoints of the ordered list, and forward and backward differencing are performed on the first and last boundary points of the ordered list, respectively. The experimentally derived growth rate is calculated based on the differencing results. The residual between the predicted and experimentally derived droplet growth rate is minimized using the L-BFGS-B algorithm. The optimal interfacial energy coefficient is obtained by back-calculating the residual. The model value of the droplet growth rate is calculated based on the supersaturation and the interfacial energy coefficient. The formula for calculating the droplet growth rate is as follows: ; in The average growth rate of the droplets. Let be the diffusion coefficient of water vapor in air, fitted according to Chapman-Ninscog theory. The average thermal velocity of water vapor molecules in the temperature field was obtained from experimental data. For supersaturation, The average radius of the droplets is 1. Let be the interfacial energy coefficient of the droplets. Boltzmann's constant, The instantaneous temperature inside the cloud cabin was obtained based on the temperature field. Based on the time-series dataset of the expansion phase, peak pressure change rate, spatiotemporal distribution of supersaturation, model value of droplet growth rate, and peak droplet number concentration are extracted. Based on the time-series dataset of the maintenance phase, average pressure, steady-state value of average droplet radius, variance of temperature field distribution, and steady-state value of wall heat flux density are extracted to obtain feature vectors for each stage.

[0009] Furthermore, the method for obtaining the cloud and fog simulation model and the simulation error includes: The diffusion cycle is divided into reversible and irreversible processes based on feature vectors. A dissipative Hamiltonian equation is generated based on the system Hamiltonian of the reversible process and the dissipative Hamiltonian of the irreversible process. The residual of the dissipative Hamiltonian equation is used as the loss term of the physical information neural network to obtain the loss function. The neural network structure is a deep fully connected neural network, and the loss function is: ; in The loss function of the physical information neural network is... The total number of experimental data points. For the neural network to the first Predicted values ​​of cloud and fog state at a spatiotemporal point. For the first Experimental observations of cloud and fog conditions at a specific spatiotemporal point. The square of the L2 norm, These are the weighting coefficients. For gradient operators, For Hamiltonian, It is a Poisson matrix. The dissipation coefficient of the dissipation process; Feature vectors of different batches of cloud chambers were obtained based on the experimental conditions. These different batches included operating conditions with different initial pressures, initial temperatures, and initial aerosol concentrations. Meta-learning was performed on the neural network based on the feature vectors of these different batches of cloud chambers and the dissipative Hamiltonian equation. The weights and biases of the neural network were initialized according to the meta-learning, and the loss function for each batch of data was calculated. The network parameters were then updated through backpropagation until the change in the loss function was less than 10 over 100 consecutive iterations. -6 Furthermore, the residuals of the dissipative Hamiltonian equation are less than 10. -4 At that time, pre-training was completed; Based on the pre-trained model weights, the network parameters are fine-tuned using the feature vectors of new working conditions to obtain a cloud and fog simulation model. Based on the cloud and fog simulation model, the cloud and fog state and simulation error of the expanded cloud chamber at spatiotemporal points are output. The cloud and fog state includes droplet number concentration, supersaturation, and temperature field distribution. The simulation error includes the mean square error between experimental and predicted values ​​and the residual with the dissipative Hamiltonian equation. The fine-tuning rounds are 10 to 20 rounds.

[0010] Furthermore, the method for obtaining the output result includes: The output, simulation error, and feature vector of the cloud simulation model are aligned with the time sequence of the experimental data to obtain the working condition dataset. The working condition intervals within the cloud chamber expansion cycle are identified based on the working condition dataset. If the physical residual of the dissipation Hamiltonian equation is greater than a preset threshold, it is a strong dissipation interval. If the dissipation coefficient is more than twice the average dissipation coefficient of the cloud chamber, it is a strong dissipation interval. If the residual between the experimentally derived value and the model prediction value of the droplet growth rate is greater than 15%, it is a strong dissipation interval. The critical supersaturation is obtained from the starting point of explosive droplet generation based on the droplet spectrum. If the supersaturation reaches the critical supersaturation, it is identified as the critical point of phase transition. If the second derivative of the pressure change rate has an extreme value, it is identified as the critical point of phase transition of the cloud and fog system from non-equilibrium to steady state. Based on the strong dissipation interval and the phase transition critical point, candidate high deviation intervals are screened. If the mean square error between the experimental value and the predicted value in the candidate high deviation interval is greater than 5%, it is identified as a high deviation interval within the cloud chamber expansion cycle, and the output results within the interval are extracted. The output results include the weight change trend within the high deviation interval of the cloud simulation model.

[0011] Furthermore, the method for obtaining the simulation results of the cloud room environment includes: A three-dimensional geometric model is established based on the diameter, height, air inlet position, observation window layout, and sampling sensor position of the expanding cloud chamber to be simulated. A structured mesh is divided at the center of the chamber according to the geometric characteristics of the three-dimensional geometric model, and an unstructured mesh is divided at the boundary layer of the chamber wall, the air inlet, and the strong dissipation zone. Five to ten boundary layer meshes are generated at the chamber wall to obtain the CFD calculation mesh. The output results are used as boundary conditions of the CFD computation grid based on the high deviation interval. The velocity field of the boundary conditions is interpolated to the time step of the CFD computation grid based on the sampling timestamp of the experimental data. The spatial coordinates of the boundary conditions are mapped to the nodes of the CFD computation grid. The output results are interpolated to the center of the grid cell through bilinear interpolation. The CFD grid cell is associated with the corresponding high deviation interval label, velocity value and error value. The interval correction results are calculated for CFD mesh elements using the Navier-Stokes equations. The formula for calculating the interval correction results is as follows: ; in This represents the velocity vector of the air-water vapor mixture in the cloud chamber. This represents the velocity vector of the air-water vapor mixture in the cloud chamber. For vector outer product, The kinematic viscosity of the air-water vapor mixture is obtained based on the degree of supersaturation. For the Laplace operator, The density of the air-water vapor mixture is calculated using the ideal gas law based on the actual water vapor partial pressure, actual saturated water vapor pressure, total cloud chamber pressure, and cloud chamber temperature corresponding to the droplet spectrum. The pressure gradient vector is obtained by discretizing the pressure field corresponding to the CFD mesh using a central difference scheme. As a relaxation factor, This is an indicator function; it takes the value 1 in the high deviation range and 0 in the low deviation range. For the velocity field predicted by neurophysics, The initial velocity field calculated for CFD; Based on the CFD grid cells, the output results of the cloud and fog simulation model in the low deviation range are spliced ​​with the range correction results according to the spatiotemporal coordinates to obtain the cloud chamber environment simulation results. Based on the cloud chamber environment simulation results, the experimental data in the high deviation range are enhanced and sampled according to the error distribution, and then added to the training set of the physical information neural network.

[0012] A second aspect of the present invention provides an artificial intelligence-based system for simulating rapidly expanding cloud chamber environments, comprising: Data acquisition module: used to acquire experimental data and conditions of the rapid expansion cloud chamber, and to preprocess the experimental data, including the temperature field, pressure field, droplet spectrum and wall heat flux density inside the chamber; Feature classification module: used to perform structured classification of the experimental data according to the experimental conditions and expansion stages, to obtain a time series dataset containing cooling stage, expansion stage and maintenance stage, and to extract feature vectors for each stage; Data analysis module: used to construct a physical information neural network based on the feature vector using the dissipative Hamiltonian equation, perform meta-learning on the neural network, and obtain the cloud and fog simulation model and simulation error; Deviation range identification module: used to identify the cloud chamber operating condition range based on the simulation error and experimental data, and obtain the output result of the cloud and fog simulation model in the high deviation range according to the operating condition range. The identification conditions include the strong dissipation range of cloud and fog and the phase transition critical point. The environmental simulation results module is used to take the output results as boundary conditions, correct the interval results through CFD equations based on the boundary conditions, and splice the low-deviation interval results and the interval correction results to obtain the cloud room environment simulation results.

[0013] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: This invention extracts differentiated parameters such as temperature change rate and pressure fluctuation, representing different stages of cloud expansion in a rapidly expanding cloud chamber, including cooling and maintenance stages. This allows the physical state of each stage to match the feature vector. Based on experimental data, a physical information neural network embedding the dissipative Hamiltonian equation is used to fit the droplet growth process, balancing simulation efficiency and reducing the amount of simulation data. CFD dynamic correction is performed on high deviation intervals such as the phase transition critical point, reducing CFD simulation costs and improving the accuracy of velocity field and droplet size prediction error control in the corresponding high deviation intervals. By simulating the environment of the rapidly expanding cloud chamber, the invention improves the prediction and collaborative analysis efficiency of fog formation processes in complex environments. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the steps of the rapid expansion cloud chamber environment simulation method based on artificial intelligence in an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0016] Reference Figure 1 As shown, this invention provides a method for simulating a rapidly expanding cloud chamber environment based on artificial intelligence, comprising: The experimental data and conditions of the expanding cloud chamber to be simulated are obtained, and the experimental data are preprocessed. The experimental data includes the temperature field, pressure field, droplet spectrum and wall heat flux density inside the chamber. In the actual assessment, the simulated expanding cloud chamber was a two-section split cylinder, with a main cabin volume of 1.5m³. 3 It has an inner diameter of 1.2m, a height of 1.3m, and a pre-vacuum tank volume of 9.6m³. 3， The chamber is 1.8m in diameter and 3.8m in length, with a volume ratio of 6.4:1. It is constructed of 316L stainless steel with a Teflon coating on the inner walls. The observation window features a double-layered flanged glass structure. An internal low-temperature circulating fan is installed. The temperature control range is from room temperature to -50℃, with a design lower limit of -40℃. The average cooling rate is 0.26℃ / min. At -40℃, the internal temperature difference is ≤0.29℃. The pressure control range is 30~110kPa. The expansion and decompression rate is 109.02hPa / min. The operating noise is ≤60dB. The expansion fogging duration is 4min. The droplet size is 3~8μm, with an average effective diameter of 6.5μm and a maximum droplet number concentration of 6.1cm. -3 Based on the expansion chamber to be simulated, temperature distribution data, pressure change data, droplet spectrum, and wall heat flux density are collected according to the expansion cycle and unified to the same spatiotemporal coordinate system. Wavelet filtering is applied to the temperature distribution data and pressure change data to obtain the temperature field and pressure field. The temperature field, pressure field, droplet spectrum, and wall heat flux density are converted into a structured time-series format. The wall heat flux density is the heat exchange rate between the cloud chamber wall and the gas inside the chamber per unit area. Three silicon capacitive pressure sensors are installed at the upper, middle, and lower measuring points: A1 (0.6m, 0, 1.0m), A2 (0.6m, 0, 0.65m), and A3 (0.6m, 0, 0.3m). The measurement range is 0~110kPa, with an accuracy of ±0.01kPa. The system features a sampling frequency of 10Hz, a platinum resistance temperature sensor and a pressure sensor with the same coordinates, a measurement range of -50~60℃, an accuracy of ±0.1℃, a sampling frequency of 10Hz, a cloud particle spectrometer located in the dense droplet region (0.6m, 0.3m, 0.65m), a measurement range of 2~50μm, an accuracy of ±5% of particle size, a sampling frequency of 5Hz, one aerosol particle size analyzer located on the main cabin air inlet bypass, a measurement range of 180nm~10.5μm, an accuracy of ±3% of particle size, a sampling frequency of 5Hz, an LT5100 visibility meter located at (0.6m, -0.3m, 0.65m), a measurement range of 10m~50km, an accuracy of ±2% of visibility value, a sampling frequency of 1Hz, and one high-speed camera system located outside the observation window. In the actual evaluation, the temperature, humidity, and pressure of the corresponding cloud chamber control system were used as experimental conditions based on structured time-series data. The structured time-series data was correlated with the experimental conditions to obtain experimental data. The time-series data for one batch of experiments from 0 to 10 minutes were: time 0 t / min, temperature 17.27 T / ℃, total pressure 994.77 p. _total / hPa, relative humidity 65%RH / %, droplet number concentration 0N / cm -3 The average droplet size was 0 d / μm, and the aerosol number concentration was 320 (0.18~10.5μm) / cm³. -3 Time: 2t / min; Temperature: 12.5T / ℃; Total Pressure: 780.00p _total / hPa, relative humidity 92RH / %, droplet number concentration 2.8N / cm -3 The average droplet size was 3.2 d / μm, and the aerosol number concentration was 410 (0.18~10.5μm) / cm³. -3 Time 4t / min, temperature 12.5T / ℃, total pressure 558.67p _total / hPa, relative humidity 100RH / %, droplet number concentration 6.1N / cm -3 The average droplet size was 5.3 d / μm, and the aerosol number concentration was 580 (0.18~10.5μm) / cm. -3 ; Based on the experimental data, the data is structured and classified according to the expansion stage according to the experimental conditions to obtain a time series dataset containing the cooling stage, expansion stage and maintenance stage, and feature vectors of each stage are extracted. In the actual evaluation, during the expansion phase (0-4 min), the pressure decreased from 994.77 hPa to 558.67 hPa, with a cooling rate of 0.174 °C / min (17.27 °C → 10.3 °C), and the water vapor reached 100% saturation. During the maintenance phase (4-8 min), the pressure was maintained at 558.67 ± 0.5 hPa, and the temperature at 10.1 ± 0.1 °C. The droplets were in a stable growth stage, and the droplet number concentration was maintained at 5.2-6.1 cm⁻¹. -3During the dissipation phase (8-10 min), the pressure rises to 620 hPa, the temperature rises by 1.3℃, and the supersaturation drops below 1.05. The droplets gradually evaporate. The temperature change rate during the expansion phase is calculated segmentally as follows: 0.68℃ / min from room temperature to 10℃, 0.43℃ / min from 10℃ to 0℃, and 0.41℃ / min from 0℃ to -10℃. The pressure drop rate is 109.02 hPa / min. At 4 min, the supersaturation, calculated from the humidity sensor and saturated vapor pressure, is 1.05, corresponding to a temperature of 10.3℃. The saturated vapor pressure, calculated using the Clausius-Clapeyron equation, is 1.29 kPa, and the partial pressure of water vapor is 1.35 kPa. Therefore, the water vapor volume fraction is 2.42%. Based on the water vapor volume fraction and the ideal gas law, the density of the air-water vapor mixture is calculated to be 0.70 kg / m³. 3 kinematic viscosity ; of which 1.5×10 -5 m 2 / s represents the standard kinematic viscosity of pure air at room temperature; 0.0242 represents the water vapor volume fraction; the average temperature change rate during the cooling section is -0.174℃ / min (the difference between the final and initial temperatures of the cooling section, divided by the stage duration); the average pressure decrease rate is -109.02 hPa / min; the final temperature during the cooling section is 10.3℃; the final pressure during the cooling section is 558.67 hPa; the final supersaturation during the cooling section is 1.05; and the change in aerosol number concentration is 580-320 cm⁻¹. -3 The droplet number concentration at the end of the cooling phase was 6.1 cm⁻¹. -3 The average supersaturation during the maintenance period was 1.03, the average droplet growth rate was 0.45 μm / min, and the average droplet number concentration was 5.5 cm⁻¹. -3 , to obtain the feature vectors of each stage; Based on the feature vectors, a physical information neural network is constructed using the dissipative Hamiltonian equation. Meta-learning is then performed on the neural network to obtain a cloud and fog simulation model and simulation error. In practical evaluation, the diffusion cycle is divided into reversible and irreversible processes based on feature vectors. A dissipative Hamiltonian equation is generated based on the system Hamiltonian of the reversible process and the dissipative Hamiltonian of the irreversible process. The residual of the dissipative Hamiltonian equation is used as the loss term of the physical information neural network to obtain the loss function. The neural network structure is a deep fully connected neural network, where the input layer has 10 neurons and 10 dimensions, including temperature, pressure, relative humidity, supersaturation, water vapor volume fraction, density of the air-water vapor mixture, droplet number concentration, average droplet size, aerosol number concentration, and visibility. The number of neurons in hidden layers 1 to 3 changes from 256 to 128 to 64, and the activation function for hidden layers 1 to 3 is ReLU. The input and hidden layers are initialized using He normal distribution. The activation function for output layer 5 is... The output layer is Sigmoid, with 5 dimensions including flow velocity, pressure, droplet number concentration, and average droplet size. Output layer 5 is initialized normally. Feature vectors for different batches of cloud chambers are obtained based on the experimental conditions, including different initial pressures, temperatures, and aerosol concentrations. Meta-learning is performed on the neural network based on the feature vectors and dissipative Hamiltonian equations of these different batches of cloud chambers. The training data consists of 150 sets of cloud chamber experimental data: 120 sets for training and 30 sets for validation. Each set contains 80 spatiotemporal sampling points, for a total of 9600 training samples. The regularization coefficient of the loss function is 0.008. The optimizer uses Adam, with an initial learning rate of 0.001 and 8000 iterations. The early stopping condition is that the validation set loss changes by ≤10 for 50 consecutive iterations. -6 The MSE loss was 0.009 at the 3000th round, and training stopped when the MSE loss reached 0.002, thus obtaining the cloud and fog simulation model. At t=4min, the prediction accuracy of the cloud and fog simulation model was PINN, which predicted a fog droplet number concentration of 6.0 cm⁻¹. -3 Experimental value: 6.1 cm -3 The error was 1.6%; the predicted particle size d = 5.2 μm, the experimental value was 5.3 μm, and the error was 1.9%. Based on the simulation error and experimental data, the cloud chamber operating condition range is identified, and the output results of the cloud and fog simulation model in the high deviation range are obtained according to the operating condition range. The identification conditions include the strong dissipation range of clouds and fog and the phase transition critical point. In actual evaluations, the dissipative Hamiltonian equation corresponding to the physical residuals has a dissatisfaction rate > 8 × 100. -4 If the deviation between the PINN-predicted droplet growth rate and the experimentally derived value in the growth rate residual is >12%, and the local velocity gradient of the flow field gradient is >0.4 m / (s・m), then the high deviation range is the core region in the middle of the main cabin (coordinates: radial 0.4~0.8 m, height 0.4~0.9 m, time 3~5 min). The output results are used as boundary conditions. The interval results are corrected by CFD equations based on the boundary conditions. The low-bias interval results and the interval correction results are then combined to obtain the cloud room environment simulation results.

[0017] In the actual evaluation, the high-deviation zone exhibited the highest droplet concentration and fastest growth rate, with a predicted velocity field of 0.52 m / s and an initial CFD calculation of 0.48 m / s, resulting in an error of 8.3%. The cloud simulation model showed significant prediction errors. A 1:1 scale geometric model was constructed using SolidWorks based on the diameter, height, inlet location, observation window layout, and sampling sensor locations of the simulated expanding cloud chamber. The main chamber wall thickness was 15 mm. The pre-vacuum tank was horizontally positioned and connected to the main chamber via a controllable solenoid valve with an 80 mm diameter. The observation window diameter was 200 mm, located at (0.6 m, 0.6 m, 0.65 m). Structured meshes with a size of 6 mm were used in the non-droplet-dense areas of the main chamber and the pre-vacuum tank. The unstructured mesh size in the high-deviation zone was [not specified]. The 2mm CFD mesh has 2800 unstructured mesh elements, accounting for 0.93% of the total mesh. It needs CFD correction. The wall boundary layer has 7 layers, with the first layer having a thickness of 0.15mm and a growth rate of 1.3, adapted to the viscous characteristics of the stainless steel wall. The mesh orthogonality is >0.92 and the twist rate is <0.28. The CFD computational mesh is obtained. The output results are used as the boundary conditions of the CFD computational mesh based on the high deviation interval. The velocity field of the boundary conditions is interpolated to the time step of the CFD computational mesh based on the sampling timestamp of the experimental data. The spatial coordinates of the boundary conditions are mapped to the nodes of the CFD computational mesh. The output results are interpolated to the center of the mesh element through bilinear interpolation. The CFD mesh element is associated with the corresponding high deviation interval label, velocity value, and error value. In the actual evaluation, initial CFD calculations were performed using the OpenFOAM solver based on the CFD mesh elements. The calculation time step was 0.002 s, with 3000 iterations. At t=4 min, the velocity field in the high-deviation interval was 0.48 m / s, while the predicted value from the cloud simulation model was 0.52 m / s. Gradient descent minimization optimization was then performed based on the error difference between the velocity field in the high-deviation interval and the predicted value, resulting in a relaxation factor of 0.65 and an error difference of 0.04 m / s. Based on the relaxation factor, the interval correction results were calculated for the CFD mesh elements using the Navier-Stokes equations. The velocity field in the high-deviation range was 0.506 m / s, with an error of 4.5% compared to the experimentally measured value of 0.53 m / s, representing a 45.8% reduction compared to the original value. Based on CFD grid cells, the cloud and fog simulation model output in the low-deviation range was stitched together with the range correction results according to spatiotemporal coordinates to obtain the cloud chamber environment simulation results. The spatiotemporal resolution of the cloud chamber environment simulation results was 0.002 s × 2 mm, and the flow field data included the x / y direction velocity field (0.45~0.55 m / s), pressure field (558.67±1.2 kPa), droplet spectrum (3~8 μm), and droplet number concentration (5.2~6.1 cm⁻¹). -3 The average effective diameter (6.3~6.7μm), visibility (80~120m), aerosol activation rate (65%~72%), and simulated droplet number concentration are specifically 6.0cm. -3 =6.1cm -3 The simulated average effective diameter is 6.4 μm with an error of 1.6%, compared to the experimental value of 6.5 μm, with an error of 1.5%. The simulated pressure field error is 0.8%, and the simulated temperature field error is 0.3℃.

[0018] In this embodiment, the method for obtaining the experimental data and experimental conditions includes: Based on the expansion chamber to be simulated, temperature distribution data, pressure change data, droplet spectrum and wall heat flux density are collected in the chamber according to the expansion cycle and unified into the same spatiotemporal coordinate system. Wavelet filtering is performed on the temperature distribution data and pressure change data to obtain the temperature field and pressure field. The temperature field, pressure field, droplet spectrum and wall heat flux density are converted into a structured time series format. The wall heat flux density is the heat exchange rate between the cloud chamber wall and the gas in the chamber per unit area. Based on the structured time-series data, the temperature, humidity, and pressure of the corresponding cloud chamber control system are used as experimental conditions. The structured time-series data is correlated with the experimental conditions to obtain experimental data.

[0019] In this embodiment, the method for obtaining the time-series dataset includes: Based on the pressure field, the cloud chamber expansion cycle is divided using the pressure change rate threshold method. The stage from the initial pressure, where the pressure change rate is negative and the absolute value gradually increases, is marked as the cooling stage. The stage where the pressure change rate reaches its peak and then rapidly decreases is marked as the expansion stage. The stage where the pressure change rate approaches zero and the pressure is stable is marked as the maintenance stage. The experimental data are structured and classified according to the cooling, expansion and maintenance phases to obtain time-series datasets for each phase, and feature vectors for each phase are extracted based on the time-series datasets.

[0020] In this embodiment, the method for obtaining the feature vector includes: The average temperature change rate, peak wall heat flux density, pressure drop rate, and difference between initial temperature and initial pressure are extracted from the time series dataset of the cooling section. The saturated vapor pressure is calculated using the Clausius-Clapeyron equation based on the temperature field of the time series dataset in the expansion section. The ratio of the actual vapor pressure to the saturated vapor pressure in the pressure field is taken as the supersaturation. Based on the expansion segment, an ordered list is obtained from discrete measurements of the average droplet radius over time in the droplet spectrum. Central differencing is performed on the midpoints of the ordered list, and forward and backward differencing are performed on the first and last boundary points of the ordered list, respectively. The experimentally derived growth rate is calculated based on the differencing results. The residual between the predicted and experimentally derived droplet growth rate is minimized using the L-BFGS-B algorithm. The optimal interfacial energy coefficient is obtained by back-calculating the residual. The model value of the droplet growth rate is calculated based on the supersaturation and the interfacial energy coefficient. The formula for calculating the droplet growth rate is as follows: ; in The average growth rate of the droplets. Let be the diffusion coefficient of water vapor in air, fitted according to Chapman-Ninscog theory. The average thermal velocity of water vapor molecules in the temperature field was obtained from experimental data. For supersaturation, The average radius of the droplets is 1. Let be the interfacial energy coefficient of the droplets. Boltzmann's constant, The instantaneous temperature inside the cloud cabin was obtained based on the temperature field. Based on the time-series dataset of the expansion phase, peak pressure change rate, spatiotemporal distribution of supersaturation, model value of droplet growth rate, and peak droplet number concentration are extracted. Based on the time-series dataset of the maintenance phase, average pressure, steady-state value of average droplet radius, variance of temperature field distribution, and steady-state value of wall heat flux density are extracted to obtain feature vectors for each stage.

[0021] In this embodiment, the method for obtaining the cloud and fog simulation model and simulation error includes: The diffusion cycle is divided into reversible and irreversible processes based on feature vectors. A dissipative Hamiltonian equation is generated based on the system Hamiltonian of the reversible process and the dissipative Hamiltonian of the irreversible process. The residual of the dissipative Hamiltonian equation is used as the loss term of the physical information neural network to obtain the loss function. The neural network structure is a deep fully connected neural network, and the loss function is: ; in The loss function of the physical information neural network is... The total number of experimental data points. For the neural network to the first Predicted values ​​of cloud and fog state at a spatiotemporal point. For the first Experimental observations of cloud and fog conditions at a specific spatiotemporal point. The square of the L2 norm, These are the weighting coefficients. For gradient operators, For Hamiltonian, It is a Poisson matrix. The dissipation coefficient is the dissipation factor of the dissipation process. Feature vectors of different batches of cloud chambers were obtained based on the experimental conditions. These different batches included operating conditions with different initial pressures, initial temperatures, and initial aerosol concentrations. Meta-learning was performed on the neural network based on the feature vectors of these different batches of cloud chambers and the dissipative Hamiltonian equation. The weights and biases of the neural network were initialized according to the meta-learning, and the loss function for each batch of data was calculated. The network parameters were then updated through backpropagation until the change in the loss function was less than 10 over 100 consecutive iterations. -6 Furthermore, the residuals of the dissipative Hamiltonian equation are less than 10. -4 At that time, pre-training was completed; Based on the pre-trained model weights, the network parameters are fine-tuned using the feature vectors of new working conditions to obtain a cloud and fog simulation model. Based on the cloud and fog simulation model, the cloud and fog state and simulation error of the expanded cloud chamber at spatiotemporal points are output. The cloud and fog state includes droplet number concentration, supersaturation, and temperature field distribution. The simulation error includes the mean square error between experimental and predicted values ​​and the residual with the dissipative Hamiltonian equation. The fine-tuning rounds are 10 to 20 rounds.

[0022] In this embodiment, the method for obtaining the output result includes: The output, simulation error, and feature vector of the cloud simulation model are aligned with the time sequence of the experimental data to obtain the working condition dataset. The working condition intervals within the cloud chamber expansion cycle are identified based on the working condition dataset. If the physical residual of the dissipation Hamiltonian equation is greater than a preset threshold, it is a strong dissipation interval. If the dissipation coefficient is more than twice the average dissipation coefficient of the cloud chamber, it is a strong dissipation interval. If the residual between the experimentally derived value and the model prediction value of the droplet growth rate is greater than 15%, it is a strong dissipation interval. The critical supersaturation is obtained from the starting point of explosive droplet generation based on the droplet spectrum. If the supersaturation reaches the critical supersaturation, it is identified as the critical point of phase transition. If the second derivative of the pressure change rate has an extreme value, it is identified as the critical point of phase transition of the cloud and fog system from non-equilibrium to steady state. Based on the strong dissipation interval and the phase transition critical point, candidate high deviation intervals are screened. If the mean square error between the experimental value and the predicted value in the candidate high deviation interval is greater than 5%, it is identified as a high deviation interval within the cloud chamber expansion cycle, and the output results within the interval are extracted. The output results include the weight change trend within the high deviation interval of the cloud simulation model.

[0023] In this embodiment, the method for obtaining the cloud room environment simulation results includes: A three-dimensional geometric model is established based on the diameter, height, air inlet position, observation window layout, and sampling sensor position of the expanding cloud chamber to be simulated. A structured mesh is divided at the center of the chamber according to the geometric characteristics of the three-dimensional geometric model, and an unstructured mesh is divided at the boundary layer of the chamber wall, the air inlet, and the strong dissipation zone. Five to ten boundary layer meshes are generated at the chamber wall to obtain the CFD calculation mesh. The output results are used as boundary conditions of the CFD computation grid based on the high deviation interval. The velocity field of the boundary conditions is interpolated to the time step of the CFD computation grid based on the sampling timestamp of the experimental data. The spatial coordinates of the boundary conditions are mapped to the nodes of the CFD computation grid. The output results are interpolated to the center of the grid cell through bilinear interpolation. The CFD grid cell is associated with the corresponding high deviation interval label, velocity value and error value. The interval correction results are calculated for CFD mesh elements using the Navier-Stokes equations. The formula for calculating the interval correction results is as follows: ; in This represents the velocity vector of the air-water vapor mixture in the cloud chamber. This represents the velocity vector of the air-water vapor mixture in the cloud chamber. For vector outer product, The kinematic viscosity of the air-water vapor mixture is obtained based on the degree of supersaturation. For the Laplace operator, The density of the air-water vapor mixture is calculated using the ideal gas law based on the actual water vapor partial pressure, actual saturated water vapor pressure, total cloud chamber pressure, and cloud chamber temperature corresponding to the droplet spectrum. The pressure gradient vector is obtained by discretizing the pressure field corresponding to the CFD mesh using a central difference scheme. As a relaxation factor, This is an indicator function; it takes the value 1 in the high deviation range and 0 in the low deviation range. For the velocity field predicted by neurophysics, The initial velocity field calculated for CFD; Based on the CFD grid cells, the output results of the cloud and fog simulation model in the low deviation range are spliced ​​with the range correction results according to the spatiotemporal coordinates to obtain the cloud chamber environment simulation results. Based on the cloud chamber environment simulation results, the experimental data in the high deviation range are enhanced and sampled according to the error distribution, and then added to the training set of the physical information neural network.

[0024] A second aspect of the present invention also provides an artificial intelligence-based rapid expansion cloud chamber environment simulation system, comprising: Data acquisition module: used to acquire experimental data and conditions of the rapid expansion cloud chamber, and to preprocess the experimental data, including the temperature field, pressure field, droplet spectrum and wall heat flux density inside the chamber; Feature classification module: used to perform structured classification of the experimental data according to the experimental conditions and expansion stages, to obtain a time series dataset containing cooling stage, expansion stage and maintenance stage, and to extract feature vectors for each stage; Data analysis module: used to construct a physical information neural network based on the feature vector using the dissipative Hamiltonian equation, perform meta-learning on the neural network, and obtain the cloud and fog simulation model and simulation error; Deviation range identification module: used to identify the cloud chamber operating condition range based on the simulation error and experimental data, and obtain the output result of the cloud and fog simulation model in the high deviation range according to the operating condition range. The identification conditions include the strong dissipation range of cloud and fog and the phase transition critical point. The environmental simulation results module is used to take the output results as boundary conditions, correct the interval results through CFD equations based on the boundary conditions, and splice the low-deviation interval results and the interval correction results to obtain the cloud room environment simulation results.

[0025] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. A method for simulating rapidly expanding cloud chamber environments based on artificial intelligence, characterized in that, Includes the following steps: The experimental data and conditions of the expanding cloud chamber to be simulated are obtained, and the experimental data are preprocessed. The experimental data includes the temperature field, pressure field, droplet spectrum and wall heat flux density inside the chamber. Based on the pressure field, the cloud chamber expansion cycle is divided using the pressure change rate threshold method. The stage from the initial pressure, where the pressure change rate is negative and the absolute value gradually increases, is marked as the cooling stage. The stage where the pressure change rate reaches its peak and then rapidly decreases is marked as the expansion stage. The stage where the pressure change rate approaches zero and the pressure is stable is marked as the maintenance stage. The experimental data are structured and classified according to the cooling, expansion, and maintenance phases to obtain time-series datasets for each phase. Feature vectors for each phase are then extracted from the time-series datasets, including: Based on the time series dataset of the cooling section, the average temperature change rate, peak wall heat flux density, pressure drop rate, difference between the initial temperature and the end temperature of the cooling section, and difference between the initial pressure and the end pressure of the cooling section are extracted. Based on the temperature field of the time series dataset in the expansion section, the saturated vapor pressure is calculated using the Clausius-Clapeyron equation. The ratio of the actual vapor pressure of the pressure field to the saturated vapor pressure is taken as the supersaturation. Based on the expansion segment, an ordered list is obtained from discrete measurements of the average droplet radius over time in the droplet spectrum. Central differencing is performed on the midpoints of the ordered list, and forward and backward differencing are performed on the first and last boundary points of the ordered list, respectively. The experimentally derived growth rate is calculated based on the differencing results. The residual between the predicted and experimentally derived droplet growth rate is minimized using the L-BFGS-B algorithm. The optimal interfacial energy coefficient is obtained by back-calculating the residual. The model value of the droplet growth rate is calculated based on the supersaturation and the interfacial energy coefficient. The formula for calculating the droplet growth rate is as follows: ； in The average growth rate of the droplets. Let be the diffusion coefficient of water vapor in air, fitted according to Chapman-Ninscog theory. The average thermal velocity of water vapor molecules in the temperature field was obtained from experimental data. For supersaturation, The average radius of the droplets is 1. Let be the interfacial energy coefficient of the droplets. Boltzmann's constant, The instantaneous temperature inside the cloud cabin was obtained based on the temperature field. Based on the time series dataset of the expansion phase, the peak pressure change rate, the spatiotemporal distribution of supersaturation, the model value of droplet growth rate, and the peak droplet number concentration are extracted. Based on the time series dataset of the maintenance phase, the average pressure, the steady-state value of the average droplet radius, the variance of the temperature field distribution, and the steady-state value of the wall heat flux density are extracted to obtain the feature vectors of each stage. Based on the feature vectors, a physical information neural network is constructed using the dissipative Hamiltonian equation. Meta-learning is then performed on the neural network to obtain a cloud and fog simulation model and simulation error. Based on the simulation error and experimental data, the cloud chamber operating condition range is identified, and the output results of the cloud and fog simulation model in the high deviation range are obtained according to the operating condition range. The identification conditions include the strong dissipation range of clouds and fog and the phase transition critical point. The output results are used as boundary conditions. The interval results are corrected by CFD equations based on the boundary conditions. The low-bias interval results and the interval correction results are then combined to obtain the cloud room environment simulation results.

2. The method for simulating a rapidly expanding cloud chamber environment based on artificial intelligence according to claim 1, characterized in that, The methods for obtaining the experimental data and experimental conditions include: Based on the expansion chamber to be simulated, temperature distribution data, pressure change data, droplet spectrum and wall heat flux density are collected in the chamber according to the expansion cycle and unified into the same spatiotemporal coordinate system. Wavelet filtering is performed on the temperature distribution data and pressure change data to obtain the temperature field and pressure field. The temperature field, pressure field, droplet spectrum and wall heat flux density are converted into a structured time series format. The wall heat flux density is the heat exchange rate between the cloud chamber wall and the gas in the chamber per unit area. Based on the structured time-series data, the temperature, humidity, and pressure of the corresponding cloud chamber control system are used as experimental conditions. The structured time-series data is correlated with the experimental conditions to obtain experimental data.

3. The method for simulating a rapidly expanding cloud chamber environment based on artificial intelligence according to claim 1, characterized in that, The method for obtaining the cloud and fog simulation model and simulation error includes: The diffusion cycle is divided into reversible and irreversible processes based on feature vectors. A dissipative Hamiltonian equation is generated based on the system Hamiltonian of the reversible process and the dissipative Hamiltonian of the irreversible process. The residual of the dissipative Hamiltonian equation is used as the loss term of the physical information neural network to obtain the loss function. The neural network structure is a deep fully connected neural network, and the loss function is: ； in The loss function of the physical information neural network is... The total number of experimental data points. For the neural network to the first Predicted values ​​of cloud and fog state at a spatiotemporal point. For the first Experimental observations of cloud and fog conditions at a specific spatiotemporal point. The square of the L2 norm. These are the weighting coefficients. For gradient operators, For Hamiltonian, It is a Poisson matrix. The dissipation coefficient is the dissipation factor of the dissipation process. Feature vectors of different batches of cloud chambers were obtained based on the experimental conditions. These different batches included operating conditions with different initial pressures, initial temperatures, and initial aerosol concentrations. Meta-learning was performed on the neural network based on the feature vectors of these different batches of cloud chambers and the dissipative Hamiltonian equation. The weights and biases of the neural network were initialized according to the meta-learning, and the loss function for each batch of data was calculated. The network parameters were updated through backpropagation. If the change in the loss function over 100 consecutive iterations was less than 10... −6 Furthermore, the residuals of the dissipative Hamiltonian equation are less than 10. −4 If so, then pre-training is complete; Based on the pre-trained model weights, the network parameters are fine-tuned using the feature vectors of new working conditions to obtain a cloud and fog simulation model. Based on the cloud and fog simulation model, the cloud and fog state and simulation error of the expanded cloud chamber at spatiotemporal points are output. The cloud and fog state includes droplet number concentration, supersaturation, and temperature field distribution. The simulation error includes the mean square error between experimental and predicted values ​​and the residual with the dissipative Hamiltonian equation. The fine-tuning rounds are 10 to 20 rounds.

4. The method for simulating a rapidly expanding cloud chamber environment based on artificial intelligence according to claim 1, characterized in that, The method for obtaining the output result includes: The output, simulation error, and feature vector of the cloud simulation model are aligned according to the time sequence of the experimental data to obtain the working condition dataset. The working condition intervals within the cloud chamber expansion cycle are identified based on the working condition dataset. If the physical residual of the dissipation Hamiltonian equation is greater than a preset threshold, it is a strong dissipation interval. If the dissipation coefficient is more than twice the average dissipation coefficient of the cloud chamber, it is a strong dissipation interval. If the residual between the experimentally derived value and the model prediction value of the droplet growth rate is greater than 15%, it is a strong dissipation interval. The critical supersaturation is obtained from the starting point of explosive droplet generation based on the droplet spectrum. If the supersaturation reaches the critical supersaturation, it is identified as the critical point of phase transition. If the second derivative of the pressure change rate has an extreme value, it is identified as the critical point of phase transition of the cloud and fog system from non-equilibrium to steady state. Based on the strong dissipation interval and the phase transition critical point, candidate high deviation intervals are screened. If the mean square error between the experimental value and the predicted value in the candidate high deviation interval is greater than 5%, it is identified as a high deviation interval within the cloud chamber expansion cycle, and the output results within the interval are extracted. The output results include the weight change trend within the high deviation interval of the cloud simulation model.

5. The method for simulating a rapidly expanding cloud chamber environment based on artificial intelligence according to claim 1, characterized in that, The method for obtaining the simulation results of the cloud room environment includes: A three-dimensional geometric model is established based on the diameter, height, air inlet position, observation window layout, and sampling sensor position of the expanding cloud chamber to be simulated. A structured mesh is divided at the center of the chamber according to the geometric characteristics of the three-dimensional geometric model, and an unstructured mesh is divided at the boundary layer of the chamber wall, the air inlet, and the strong dissipation zone. Five to ten boundary layer meshes are generated at the chamber wall to obtain the CFD calculation mesh. The output results are used as boundary conditions of the CFD computation grid based on the high deviation interval. The velocity field of the boundary conditions is interpolated to the time step of the CFD computation grid based on the sampling timestamp of the experimental data. The spatial coordinates of the boundary conditions are mapped to the nodes of the CFD computation grid. The output results are interpolated to the center of the grid cell through bilinear interpolation. The CFD grid cell is associated with the corresponding high deviation interval label, velocity value and error value. The interval correction results are calculated for CFD mesh elements using the Navier-Stokes equations. The formula for calculating the interval correction results is as follows: ； in This represents the velocity vector of the air-water vapor mixture in the cloud chamber. For vector outer product, The kinematic viscosity of the air-water vapor mixture is obtained based on the degree of supersaturation. For the Laplace operator, The density of the air-water vapor mixture is calculated using the ideal gas law based on the actual water vapor partial pressure, actual saturated water vapor pressure, total cloud chamber pressure, and cloud chamber temperature corresponding to the droplet spectrum. The pressure gradient vector is obtained by discretizing the pressure field corresponding to the CFD mesh using a central difference scheme. As a relaxation factor, This is an indicator function; it takes the value 1 in the high deviation range and 0 in the low deviation range. For the velocity field predicted by neurophysics, The initial velocity field calculated for CFD; Based on the CFD grid cells, the output results of the cloud and fog simulation model in the low deviation range are spliced ​​with the range correction results according to the spatiotemporal coordinates to obtain the cloud chamber environment simulation results. Based on the cloud chamber environment simulation results, the experimental data in the high deviation range are enhanced and sampled according to the error distribution, and then added to the training set of the physical information neural network.

6. An AI-based rapid expansion cloud chamber environment simulation system, used to execute the AI-based rapid expansion cloud chamber environment simulation method according to any one of claims 1 to 5, characterized in that, The system includes: Data acquisition module: used to acquire experimental data and conditions of the rapid expansion cloud chamber, and to preprocess the experimental data, including the temperature field, pressure field, droplet spectrum and wall heat flux density inside the chamber; Feature classification module: used to perform structured classification of the experimental data according to the experimental conditions and expansion stages, to obtain a time series dataset containing cooling stage, expansion stage and maintenance stage, and to extract feature vectors for each stage; Data analysis module: used to construct a physical information neural network based on the feature vector using the dissipative Hamiltonian equation, perform meta-learning on the neural network, and obtain the cloud and fog simulation model and simulation error; Deviation range identification module: used to identify the cloud chamber operating condition range based on the simulation error and experimental data, and obtain the output result of the cloud and fog simulation model in the high deviation range according to the operating condition range. The identification conditions include the strong dissipation range of cloud and fog and the phase transition critical point. The environmental simulation results module is used to take the output results as boundary conditions, correct the interval results through CFD equations based on the boundary conditions, and splice the low-deviation interval results and the interval correction results to obtain the cloud room environment simulation results.

Images