Particle mass transfer characteristic prediction method and system based on machine learning algorithm

Abstract

The invention discloses a particle mass transfer characteristic prediction method and system based on a machine learning algorithm, and relates to the technical field of particle movement in a multiphase flow system. The method comprises the following steps: constructing a bidirectional coupling numerical model, simulating a particle stirring process of a stirring tank by using the bidirectional coupling numerical model, and analyzing particle mass transfer characteristics and motion characteristics; constructing a machine learning model set, determining a neural network model, training the neural network model, and predicting the behavior of particles in the stirring tank by using the trained neural network to obtain a predicted value of the mass transfer coefficient of the stirring tank; and carrying out combined analysis on the analysis result of the interaction force between the particles and the fluid and the predicted value of the mass transfer coefficient of the stirring tank to obtain an inter-phase mass transfer rule in the stirring tank. The problems that the strong nonlinearity of the stirring tank is difficult to analyze, the mixing efficiency in the solid-liquid suspension process is low, the real-time performance based on dynamic model regulation and control is poor, and industrialization cannot be completely adapted are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of particle motion in a multiphase flow system, and in particular to a method and system for predicting particle mass transfer characteristics based on a machine learning algorithm. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Solid-liquid mixing in the resource utilization of nonferrous metal smelting slag faces challenges such as balancing leaching efficiency with economic efficiency and operability, synergizing mass transfer with chemical reactions, the influence of particle size and stirring speed, and adapting kinetic models to industrialization. The coupled computational fluid dynamics-discrete element model (CFD-DEM) approach can analyze dynamic behaviors such as particle agglomeration and collision at the microscale, providing theoretical support for industrial scale-up. This technology is widely used in the chemical, pharmaceutical, and other fields to optimize mixing efficiency and reduce engineering conversion risks.

[0004] As the core equipment in the wet vanadium extraction process, the stirred tank utilizes impeller shear force to enhance the mixing and mass transfer of mineral particles and solvent, while precisely controlling reaction conditions (temperature, pressure, and velocity). Its mixing efficiency directly impacts the mass and heat transfer performance of the solid-liquid two-phase system, and has important applications in heterogeneous catalysis, biohydrogen production, and flotation separation. Due to the inherently strong nonlinearity of the stirred tank, traditional methods are unable to achieve real-time prediction of plant data. With the rapid development of artificial intelligence and computer science, the integration of CFD-DEM and machine learning (ML) techniques has become a powerful tool for real-time prediction of stirred tanks. This interdisciplinary approach demonstrates significant advantages and broad application prospects in multi-parameter optimization, fitting nonlinear relationships, improving prediction accuracy, implementing regression analysis, and reducing operating costs. Artificial neural networks (ANNs) are an effective tool widely used in the modeling, optimization, calculation, and prediction of complex systems.

[0005] However, existing technologies have the following defects when integrating CFD-DEM with machine learning (ML) technology and applying it to problems related to stirred tanks: First, due to the strong nonlinearity of the stirred tank, traditional machine learning methods do not have a deep enough analysis of the movement patterns of particles in the stirred tank, making it difficult to accurately describe the complex behavior of particles at different particle sizes and rotational speeds. This results in the inability to accurately select operating parameters in actual production, and problems such as particle sedimentation are prone to occur, resulting in a waste of resources. Secondly, traditional CFD has high computational costs and poor real-time performance, and existing fusion technologies have failed to effectively overcome this limitation, making it difficult to efficiently simulate and predict multiphase flow systems. In addition, existing technologies lack a systematic evaluation of the applicability of various machine learning algorithms in stirred tank predictions, and are unable to provide a comprehensive methodological reference for subsequent research, limiting the deep integration of multiphase flow and artificial intelligence technologies. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a method and system for predicting particle mass transfer characteristics based on a machine learning algorithm to solve the problems of the strong nonlinearity of the stirring tank itself being difficult to analyze, the low mixing efficiency of the solid-liquid suspension process, the poor real-time performance of the kinetic model-based control, and the inability to fully adapt to industrialization.

[0007] In order to achieve the above object, the present invention is implemented through the following technical solutions: A first aspect of the present invention provides a method for predicting particle mass transfer characteristics based on a machine learning algorithm, comprising the following steps: A bidirectional coupling numerical model was constructed and used to simulate the particle stirring process in a stirred tank. The original data set was obtained, and the mass transfer and motion characteristics of the particles were analyzed to obtain the analysis results of the interaction force between the particles and the fluid. Construct a machine learning model set, determine the neural network model, use the original data set to train the machine learning model, use the trained neural network to predict the particle behavior in the stirred tank, and obtain the predicted value of the stirred tank mass transfer coefficient; The analysis results of the interaction force between particles and fluid are combined with the predicted value of the mass transfer coefficient of the stirred tank to obtain the mass transfer law between phases in the stirred tank.

[0008] Furthermore, the specific steps of constructing a bidirectional coupling numerical model include: Grid division is performed on the calculation area of the stirred tank, and the grid independence is verified; Determine the governing equations based on particle and fluid motion and set the corresponding parameters.

[0009] Furthermore, the specific steps of using the bidirectional coupling numerical model to simulate the particle stirring process in the stirred tank and analyze the mass transfer characteristics and motion characteristics of the particles include: The interaction between particles in the fluid domain is simulated using a two-way coupled numerical model; The relative standard deviation is introduced to quantitatively calculate the particle dispersion uniformity; Considering the movement and interaction of particles and fluid, the influence of particle motion trajectory, coupling force and process uniformity is analyzed.

[0010] Furthermore, the bidirectional coupling numerical model is a fluid mechanics discrete element coupling model, which includes a fluid mechanics model and a discrete element model. The fluid mechanics model is used to simulate the flow characteristics of the fluid, and the discrete element model is used to track the motion trajectory of solid particles under the action of the fluid.

[0011] Furthermore, in the discrete element model, the fluid flow will exert forces on the particles, and in the fluid dynamics model, the reaction of the forces acting on the particles will act on the fluid phase. Sliding mesh is enabled on one side of the fluid dynamics model to achieve sharing of model motion.

[0012] Furthermore, the specific steps for training the machine learning model using the original dataset include: Build a model set consisting of multiple machine learning models; Choose the appropriate machine learning model based on the actual task; Preprocess the original data set; The preprocessed original data set is used to train the neural network based on the genetic optimization algorithm.

[0013] Furthermore, the specific steps for using the trained neural network to predict the behavior of particles in the stirred tank are as follows: The particle position, particle velocity and corresponding flow field velocity are extracted from the CFD-DEM simulation results. The slip velocity at different positions is calculated as the input variable of the trained machine learning model, and the mass transfer coefficient is used as the output value. A nonlinear relationship fitting is constructed to achieve real-time prediction of the mass transfer process.

[0014] A second aspect of the present invention provides a particle mass transfer characteristics prediction system based on a machine learning algorithm, comprising: A bidirectional coupling model module is configured to construct a bidirectional coupling numerical model, use the bidirectional coupling numerical model to simulate the particle stirring process in the stirred tank, obtain the original data set, analyze the mass transfer characteristics and motion characteristics of the particles, and obtain the analysis results of the interaction force between the particles and the fluid; a machine learning module configured to construct a machine learning model set, determine a neural network model, train the machine learning model using the original data set, and use the trained neural network to predict the behavior of particles in the stirred tank to obtain a predicted value of the mass transfer coefficient of the stirred tank; The particle mass transfer characteristic prediction module is configured to combine the analysis results of the interaction force between the particles and the fluid with the predicted value of the mass transfer coefficient of the stirred tank to obtain the mass transfer law between the phases in the stirred tank.

[0015] The third aspect of the present invention provides a medium having a program stored thereon, which, when executed by a processor, implements the steps in the particle mass transfer characteristics prediction method based on a machine learning algorithm as described in the first aspect of the present invention.

[0016] The fourth aspect of the present invention provides a device comprising a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for predicting particle mass transfer characteristics based on a machine learning algorithm as described in the first aspect of the present invention are implemented.

[0017] One or more of the above technical solutions have the following beneficial effects: This invention discloses a method and system for predicting particle mass transfer characteristics based on a machine learning algorithm. Using a bidirectional coupling model, the system analyzes particle motion trajectories, coupling force distribution, and flow field uniformity within a stirred tank. Particle motion and mass transfer data are used to train a multi-model prediction library, and model performance is evaluated through comparative experiments. This method addresses issues such as the difficulty in analyzing the strong nonlinearity of stirred tanks, low mixing efficiency in solid-liquid suspension processes, poor real-time control based on kinetic models, and inability to fully adapt to industrialization.

[0018] The present invention reveals the laws of particle motion through the CFD-DEM bidirectional coupling model, clarifies the motion behavior of particles of different particle sizes and rotation speeds, and provides a theoretical basis for selecting appropriate particle sizes and operating parameters in actual production, thereby avoiding resource waste caused by particle sedimentation.

[0019] The present invention provides a high-precision prediction model through machine learning, which can effectively capture the complex nonlinear relationships in the stirring tank, achieve rapid response under dynamic working conditions, and provide technical support for real-time control of industrial processes.

[0020] This paper proposes a research framework that combines coupling models with data-driven methods, breaking through the limitations of traditional CFD, which has high computational cost and poor real-time performance, and providing new ideas for efficient simulation and prediction of multiphase flow systems.

[0021] This paper systematically evaluates the applicability of multiple machine learning algorithms in stirred tank prediction, provides methodological reference for subsequent research, and promotes the deep integration of multiphase flow and artificial intelligence technology.

[0022] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0024] Figure 1 Flowchart of a method for predicting particle mass transfer characteristics based on a machine learning algorithm in Example 1 of the present invention; Figure 2 This is a flow chart of the CFD-DEM bidirectional coupling algorithm in Example 1 of the present invention; Figure 3 This is a flowchart of the genetic algorithm optimization neural network algorithm in Example 1 of the present invention; Figure 4 This is a flow chart of particle motion timing prediction in Example 1 of the present invention; Figure 5 This is a flow chart for real-time prediction of the mass transfer coefficient of a stirred tank in Example 1 of the present invention. DETAILED DESCRIPTION

[0025] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0026] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or their combinations; Example 1: The first embodiment of the present invention provides a method for predicting particle mass transfer characteristics based on a machine learning algorithm. This method uses a bidirectional coupling model to analyze particle motion trajectories, coupling force distribution, and flow field uniformity within a stirred tank. The relative standard deviation (RSD) is used to quantify particle dispersion characteristics, and the mass transfer coefficient (k) is calculated based on the Hughmark model. A multi-model prediction library is trained using particle motion and mass transfer data, and model performance is evaluated through comparative experiments.

[0027] like Figure 1 As shown, the development of the particle flow simulation technology inside the reactor of this embodiment mainly consists of two parts: theoretical research on particle stirring and particle behavior prediction by machine learning. In the theoretical research part of particle stirring, a numerical calculation model of CFD-DEM bidirectional coupling is constructed, and its accuracy is verified to be grid-independent. The numerical calculation model of CFD-DEM bidirectional coupling includes a stirring simulation part of CFD software based on fluid mechanics theory and a discrete element simulation part of DEM software based on discrete element theory. The two are combined to explore the transfer characteristics and dynamics of particles, realize CFD-DEM coupling simulation, and obtain the analysis results of the interaction force between particles and fluids. The particle behavior prediction part of machine learning determines the corresponding model structure through algorithm selection, and performs model training and optimization, and evaluates the stability and accuracy of the model through real-time time series prediction. Through CFD-DEM simulation, 1,700 sets of multiphase flow data were generated as a dataset. The performance of BP, GA-BP, XGBoost, LSTM, CNN and other algorithms in particle velocity time series prediction and mass transfer coefficient prediction were compared. Multiple indicators were used to evaluate the model accuracy and stability. Combined with computational cost and accuracy analysis, the interphase mass transfer law in the stirred tank was obtained to solve the problem of industrial incompatibility.

[0028] The specific steps include: Step 1: Construct a bidirectional coupling numerical model and use it to simulate the particle stirring process in the stirred tank to obtain the original data set. The mass transfer and motion characteristics of the particles are analyzed to obtain the analysis results of the interaction force between the particles and the fluid.

[0029] Step 1.1: Construct a bidirectional coupled numerical model.

[0030] A bidirectionally coupled numerical model is a coupled fluid dynamics discrete element model, consisting of a fluid dynamics model and a discrete element model. The fluid dynamics model simulates the flow characteristics of the fluid, while the discrete element model tracks the trajectories of solid particles under the influence of the fluid. In bidirectionally coupled CFD-DEM simulations, the fluid flow characteristics are simulated using traditional continuum mechanics methods, which can calculate the hydrodynamic forces acting on individual particles. The trajectories and motion of the particles are tracked using a discrete particle model, which treats each particle as an independent entity and accounts for factors such as inter-particle interactions, fluid forces, and gravity. By combining a continuum model of the fluid with a discrete model of the particles, the CFD-DEM model comprehensively simulates the interaction between the fluid and particles, providing a powerful tool for understanding and optimizing multiphase flow systems. In CFD-DEM, particles are part of the fluid flow and affect the fluid in a bidirectional manner. This means that the particle's motion is affected by interactions with other particles and the surrounding fluid, and the flow is also affected by the presence of the particles. In the discrete element model, the fluid flow exerts forces on the particles, including pressure gradient forces, drag forces, and virtual mass forces. In the fluid dynamics model, the reaction of forces acting on particles will act on the fluid phase. Sliding meshing is enabled on the fluid dynamics model side to share model motion. These settings are automatically converted to the DEM motion frame when importing the CAS file, and the motion settings in CFD are automatically converted to the new motion frame in the DEM. In this way, consistent motion of shared geometry can be achieved between the two programs.

[0031] Step 1.1.1: Mesh the stirred tank calculation area and verify the mesh independence.

[0032] The computational domain of a stirred tank refers to the three-dimensional spatial extent of the fluid and particles within the stirred tank that is of interest when simulating fluid flow and particle motion within the tank. This domain encompasses the motion of both the fluid and solid phases (particles) within the stirred tank and serves as the fundamental physical space for meshing, solving governing equations, and setting parameters.

[0033] Step 1.1.2: Determine the governing equations for particle and fluid motion and set the corresponding parameters.

[0034] In this embodiment, the fluid in the stirred tank is liquid solvent water, which is the continuous phase in the solid-liquid two-phase flow system and forms a complex relationship with the particles (discrete phase) through bidirectional coupling. In the CFD-DEM coupling model, the control equations are determined, including the NS equations for fluid flow and the Newtonian equations of motion for particle motion. CFD and DEM parameters are set, specifically, Figure 2 As shown, the CFD initial fluid values include physical parameters, velocity, pressure, temperature and CFD time step The physical parameters are the logistic properties of the fluid, which in this embodiment include density, viscosity, specific heat capacity, etc.

[0035] The average values of the DEM unit include interaction forces, heat transfer rate, solid phase velocity, solid phase volume fraction, and the time step of the updated DEM ( ). Carry out CFD-DEM bidirectional coupling simulation.

[0036] Step 1.2: Use a bidirectional coupling numerical model to simulate the particle stirring process in the stirred tank and analyze the mass transfer and motion characteristics of the particles.

[0037] Step 1.2.1: Use a bidirectionally coupled numerical model to simulate the interaction between particles in the fluid domain.

[0038] In a specific embodiment, the present invention uses the Eulerian-Lagrangian solid-liquid two-phase model to simulate the particle flow in the stirred tank. In this model, CFD is responsible for simulating the flow characteristics of the fluid, while DEM tracks the motion trajectory of the solid particles under the action of the fluid. The fluid phase is regarded as the continuous phase. In order to accurately capture the liquid-solid mixing behavior, the turbulence model adopts the standard k-epsilon (Standard k-epsilon) of the Reynolds time-averaged model (RANS). k-ε ) model, which is based on the Boussinesq eddy viscosity hypothesis, which links the Reynolds stress to the average velocity gradient through eddy viscosity and can effectively predict turbulent characteristics. This model is favored for its economy, stability and high computational accuracy. In order to reduce the amount of calculation and simplify the particle model, rock particles are regarded as spherical particles in the DEM. The solid phase is regarded as a discrete phase, and the motion trajectory is tracked by Newton's second law. The collisions between particles and between particles and boundaries are taken into account. Therefore, the parameters between particles and between particles and boundaries are set according to actual conditions. The drag model between the solid-liquid phases adopts the Huilin-Gidaspow model, and the lift model adopts the Saffman model. The Huilin-Gidaspow model and the Saffman model are existing models in this field and will not be described here.

[0039] The specific simulation process is as follows Figure 2As shown, the Eulerian-Lagrangian model is used. First, the Standard k-ε turbulence model is used in CFD to set the initial fluid parameters. Then, the initial fluid values are derived from CFD and transferred to DEM for data import. The Huilin-Gidaspow drag model and Saffman lift model are used in DEM to calculate the initial unit average and correction , and export the calculated data as the data for the next stage of CFD. After CFD imports the data exported by DEM, update the initial value of the solid phase. Run DEM until the CFD calculation time ( ), run CFD until the calculation time ( Record the data changes of DEM and CFD respectively until the calculation time of DEM and CFD reaches the set time ( t final, t final), end the simulation calculation and output the simulation results. is the CFD calculation time, is the calculation time of DEM, and t final is the setting time.

[0040] Step 1.2.2: Introduce relative standard deviation to quantify the particle dispersion uniformity.

[0041] To quantify particle dispersion in a solid-liquid mixer, the relative standard deviation (RSD) of the particle count within the sample volume is defined. As a statistical measure of data dispersion, RSD, when used to quantify particle dispersion in solid-liquid mixing, reflects the uniformity of particle distribution within the sample volume. The calculation process involves dividing the mixing tank into multiple volume units, counting the number of particles in each unit, and calculating the average particle count and standard deviation at each time point. The RSD is the ratio of the standard deviation to the average particle count. The resulting RSD is then plotted to analyze the uniformity of particle dispersion.

[0042] Step 1.2.3: Consider the motion and interaction of particles and fluid, and analyze the effects of particle motion trajectory, coupling forces, and process uniformity.

[0043] Analyzing particle dynamics plays a vital role in industry and scientific research. Particle motion analysis involves analyzing the distribution of particles within a stirred tank at varying particle sizes and impeller speeds. This analysis can optimize processes such as stirring, mixing, separation, and conveying, improving production efficiency and product quality. The goal of solid-liquid mixing is to create a liquid suspension, which requires overcoming sedimentation effects. The coupling forces between fluids and particles are calculated to investigate the degree of interaction. Particle trajectories are tracked to obtain information such as particle velocity and displacement. Flow field cloud images are used to analyze the impact of flow field characteristics on particle motion.

[0044] Step 1.3: The PBT blades are tilted at a certain angle relative to the shaft to generate significant radial flow, which facilitates particle suspension. To analyze the particle dynamics, this example uses a CFD-DEM bidirectional coupled simulation method based on the PBT blades to analyze the particle distribution within the stirred tank and the degree of interaction with the continuous phase.

[0045] Step 1.3.1: Analyze the effect of varying the particle size in the stirred tank on the particle dynamics of the two-phase system.

[0046] Using CFD-DEM to simulate the motion of particles with 1 mm, 500 μm, and 100 μm size parameters, while keeping all other parameters constant, the researchers simulated their motion in a stirred tank and visualized the results. The results showed that the 1 mm and 500 μm particles largely settled at the bottom of the tank, while the 100 μm particles remained suspended in the fluid. These results indicate that particle motion is significantly affected by particle size; smaller particles settle more slowly and tend to remain suspended in the liquid.

[0047] Step 1.3.2: Analyze the effect of changing the impeller speed on the particle distribution and interaction degree of the two-phase system.

[0048] The impeller speed was set between 300 and 1000 rpm, with all other parameters remaining constant. The motion trajectory of 100 μm particles in the stirred tank was simulated using the CFD-DEM method. The results were visualized, and the dispersion uniformity of the particles at different speeds was calculated using the aforementioned RSD. The results show that as the speed increases, the drag effect of the flow field on the particles significantly increases, the uniformity of the axial particle distribution in the stirred tank is improved, the particle motion cycle is accelerated, and the fluid phase mixing is more uniform.

[0049] Step 2: Construct a machine learning model set, determine the neural network model, use the original data set to train the machine learning model, use the trained neural network to predict the particle behavior in the stirred tank, and obtain the predicted value of the mass transfer coefficient of the stirred tank.

[0050] like Figure 4As shown, the prediction process of the particle motion time series prediction model of the embodiment of the present invention mainly includes a CFD-DEM coupling model, particle motion characteristic analysis, a model algorithm set and a prediction verification tool. Specifically, after the CFD-DEM coupling model and particle motion characteristic analysis of the above steps, the original data set obtained by simulation is used as a training set to train the selected model, wherein the selected model comes from the model algorithm set, and a variety of ML algorithms such as regression model, neural network, random forest, decision tree, support vector machine, etc. are selected through the model algorithm set, and the trained model is used to predict the particle velocity in the stirring tank. The prediction result visualization tool is used to show the comparison between the predicted value and the actual value. The prediction accuracy of each model is evaluated using evaluation indicators such as mean square error and mean absolute error.

[0051] Step 2.1: Build a model ensemble consisting of multiple machine learning models.

[0052] In a specific embodiment, various machine learning models have their own strengths and weaknesses in machine learning algorithm predictions, and selecting the appropriate model is crucial for accurately predicting particle motion. In practical applications, a comprehensive comparison of multiple machine learning models is necessary, with performance evaluated from multiple dimensions to identify the most suitable prediction model, providing strong support for the optimized design and efficient operation of stirred tanks. Different machine learning models exhibit distinct characteristics when processing complex multiphase flow data within stirred tanks. In practical engineering applications, many other excellent machine learning models exist. Among ANN models, backpropagation (BP) minimizes error by adjusting weights and biases, making it suitable for handling nonlinear problems. However, training speed can be slow, prone to overfitting, and requires a large amount of training data. Convolutional neural networks (CNNs), with their convolutional and pooling layer structures, effectively extract local features, reduce the number of parameters, and are suitable for processing image data. However, they require high computational resources and may require GPU acceleration. Long short-term memory networks (LSTMs) can address the vanishing gradient problem through a gating mechanism and capture long-term dependencies. Extreme learning machines (ELMs), as single-hidden-layer feedforward neural networks, are fast to train and can directly compute the output layer after randomly initializing weights. Due to its limited generalization capabilities, it's suitable for predictions on small data samples. The Extreme Gradient Boosting (XGBoost) ensemble algorithm has strong overfitting resistance and efficient feature analysis, making it suitable for rapid deployment of medium-sized data, but requires fine-tuning of parameters. Random Forest (RF) can process high-dimensional data in parallel and quickly select features, but it carries the risk of overfitting. Support Vector Machine (SVM) balances accuracy and efficiency in time series prediction, making it suitable for dynamic responses on small and medium-sized data.

[0053] Step 2.2: Select an appropriate machine learning model based on the actual task.

[0054] Step 2.3: Preprocess the original dataset.

[0055] Step 2.4: Use the preprocessed original data set to train the neural network based on the genetic optimization algorithm.

[0056] like Figure 3 As shown, the genetic algorithm optimization neural network of this embodiment can improve performance and global search capabilities. Genetic algorithm is a global optimization algorithm based on natural selection and genetic mechanisms that simulates the biological evolution process. The algorithm randomly generates an initial population by encoding the weights and threshold parameters of the neural network into chromosomes, and each individual represents a set of possible network weights and thresholds. After that, the population of individuals is iteratively optimized mainly by genetic selection, crossover and mutation. The optimal chromosome is finally extracted from the optimized individuals and decoded into the initial weights and thresholds of the neural network. Using these optimized initial values, the neural network further performs traditional backpropagation training to adjust the weights and thresholds to minimize errors. Since the initial weights and thresholds have been optimized by the genetic algorithm, the neural network will converge to the global optimal solution faster during training, thereby improving prediction accuracy and generalization ability.

[0057] Step 2.5: Use the trained neural network to predict the behavior of particles in the stirred tank and obtain the predicted value of the mass transfer coefficient of the stirred tank.

[0058] Step 3: Combine the analysis results of the interaction force between particles and fluid with the predicted value of the mass transfer coefficient of the stirred tank to obtain the mass transfer law between phases in the stirred tank.

[0059] Specifically, the particle position, particle velocity and corresponding flow field velocity are extracted from the CFD-DEM simulation results, and the slip velocity at different positions is calculated as the input variable of the trained machine learning model. The mass transfer coefficient is used as the output value to construct a nonlinear relationship fitting and realize real-time prediction of the mass transfer process.

[0060] In a specific embodiment, the analysis and prediction of the mass transfer process in the stirred tank. The mass transfer process is a direct factor affecting the dynamic characteristics of the stirred tank, which usually involves the dissolution of substances on the surface of solid particles into the liquid, or the adsorption of substances in the liquid onto the surface of solid particles. The mass transfer rate is affected by factors such as the fluid flow characteristics, the contact area between the solid and liquid phases, and the concentration difference between the phases. When conducting in-depth research on the mass transfer process, the mass transfer coefficient k As a key parameter to measure the mass transfer rate, it is of great significance. k ) is a key parameter to characterize the rate of mass transfer process, which can directly and accurately reflect the speed of material transfer between solid-liquid systems. Sh ) and the Schmidt number ( Sc ) as two dimensionless numbers, both for kTo accurately obtain this parameter, this example uses the Hughmark model to calculate the important parameters in the mass transfer process. Sh .at the same time, Sc According to Standard k-ε The model value is set to 0.7 to ensure the accuracy and reliability of the calculation of the entire mass transfer process. In order to achieve accurate real-time prediction of the mass transfer coefficient, different ML algorithms can be used to predict and compare them.

[0061] like Figure 5 As shown, in the real-time prediction process of the mass transfer coefficient of a stirred tank according to an embodiment of the present invention, raw data is obtained through CFD-DEM coupled simulation, including information such as particle velocity, position, and coupling force. This data reflects the motion state and force conditions of the particles in the stirred tank. The flow field data includes fluid velocity, turbulent kinetic energy, etc., reflecting the flow characteristics of the fluid in the stirred tank. Turbulent kinetic energy affects the fluid mixing and mass transfer process. Based on the raw data set, the particle spatial position and slip velocity are calculated, and the mass transfer coefficient is calculated at the same time. This data is then normalized to make data of different dimensions comparable and improve the model training effect. A suitable model is selected from a variety of ML models, and different models are suitable for different data characteristics and problem scenarios. Hyperparameters of the model are determined. Hyperparameters, such as the number of layers and neurons in the neural network, have a significant impact on model performance. The model is trained with the processed data, and the model performance is evaluated using evaluation indicators. When the model accuracy meets the requirements, the predicted value of the mass transfer coefficient of the stirred tank is output, realizing real-time prediction of the mass transfer coefficient, providing support for the research and process optimization of the mass transfer process in the stirred tank.

[0062] Example 2: A second embodiment of the present invention provides a particle mass transfer characteristics prediction system based on a machine learning algorithm, comprising: A bidirectional coupling model module is configured to construct a bidirectional coupling numerical model, use the bidirectional coupling numerical model to simulate the particle stirring process in the stirred tank, obtain the original data set, analyze the mass transfer characteristics and motion characteristics of the particles, and obtain the analysis results of the interaction force between the particles and the fluid; a machine learning module configured to construct a machine learning model set, determine a neural network model, train the machine learning model using the original data set, and use the trained neural network to predict the behavior of particles in the stirred tank to obtain a predicted value of the mass transfer coefficient of the stirred tank; The particle mass transfer characteristic prediction module is configured to combine the analysis results of the interaction force between the particles and the fluid with the predicted value of the mass transfer coefficient of the stirred tank to obtain the mass transfer law between the phases in the stirred tank.

[0063] Example 3: A third embodiment of the present invention provides a medium having a program stored thereon, which, when executed by a processor, implements the steps of the particle mass transfer characteristics prediction method based on a machine learning algorithm as described in the first embodiment of the present invention.

[0064] Example 4: Embodiment 4 of the present invention provides a device, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for predicting particle mass transfer characteristics based on a machine learning algorithm as described in Embodiment 1 of the present invention are implemented.

[0065] The steps involved in the above embodiments 2, 3 and 4 correspond to those in the method embodiment 1. For the specific implementation methods, please refer to the relevant description part of the embodiment 1.

[0066] Those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using a general-purpose computer device. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0067] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A method for predicting particle mass transfer characteristics based on a machine learning algorithm, characterized in that: The following steps are involved: A bidirectional coupling numerical model was constructed and used to simulate the particle stirring process in a stirred tank. The original data set was obtained, and the mass transfer and motion characteristics of the particles were analyzed to obtain the analysis results of the interaction force between the particles and the fluid. Construct a machine learning model set, determine the neural network model, use the original data set to train the machine learning model, use the trained neural network to predict the particle behavior in the stirred tank, and obtain the predicted value of the stirred tank mass transfer coefficient; The analysis results of the interaction force between particles and fluid are combined with the predicted value of the mass transfer coefficient of the stirred tank to obtain the mass transfer law between phases in the stirred tank.

2. The particle mass transfer characteristics prediction method based on a machine learning algorithm according to claim 1, characterized in that: The specific steps of building a bidirectional coupled numerical model include: Grid division is performed on the calculation area of the stirred tank, and the grid independence is verified; Determine the governing equations based on particle and fluid motion and set the corresponding parameters.

3. The particle mass transfer characteristics prediction method based on machine learning algorithm according to claim 1, characterized in that: The specific steps for simulating the particle stirring process in a stirred tank using a bidirectional coupling numerical model and analyzing the mass transfer and motion characteristics of the particles include: The interaction between particles in the fluid domain is simulated using a two-way coupled numerical model; The relative standard deviation is introduced to quantitatively calculate the particle dispersion uniformity; Considering the movement and interaction of particles and fluid, the influence of particle motion trajectory, coupling force and process uniformity is analyzed.

4. The particle mass transfer characteristics prediction method based on machine learning algorithm according to claim 3, characterized in that: The bidirectional coupling numerical model is a fluid mechanics discrete element coupling model, which includes a fluid mechanics model and a discrete element model. The fluid mechanics model is used to simulate the flow characteristics of the fluid, and the discrete element model is used to track the motion trajectory of solid particles under the action of the fluid.

5. The particle mass transfer characteristics prediction method based on machine learning algorithm according to claim 4, characterized in that: In the discrete element model, the fluid flow will exert forces on the particles. In the fluid dynamics model, the reaction of the forces on the particles will act on the fluid phase. Sliding mesh is enabled on one side of the fluid dynamics model to achieve shared model motion.

6. The particle mass transfer characteristics prediction method based on machine learning algorithm according to claim 1, characterized in that: The specific steps for training a machine learning model using the original dataset include: Build a model set consisting of multiple machine learning models; Choose the appropriate machine learning model based on the actual task; Preprocess the original data set; The preprocessed original data set is used to train the neural network based on the genetic optimization algorithm.

7. The particle mass transfer characteristics prediction method based on machine learning algorithm according to claim 6, characterized in that: The specific steps for using the trained neural network to predict the behavior of particles in the stirred tank are: The particle position, particle velocity and corresponding flow field velocity are extracted from the CFD-DEM simulation results. The slip velocity at different positions is calculated as the input variable of the trained machine learning model, and the mass transfer coefficient is used as the output value. A nonlinear relationship fitting is constructed to achieve real-time prediction of the mass transfer process.

8. A particle mass transfer characteristics prediction system based on machine learning algorithm, characterized in that: include: A bidirectional coupling model module is configured to construct a bidirectional coupling numerical model, use the bidirectional coupling numerical model to simulate the particle stirring process in the stirred tank, obtain the original data set, analyze the mass transfer characteristics and motion characteristics of the particles, and obtain the analysis results of the interaction force between the particles and the fluid; a machine learning module configured to construct a machine learning model set, determine a neural network model, train the machine learning model using the original data set, and use the trained neural network to predict the behavior of particles in the stirred tank to obtain a predicted value of the mass transfer coefficient of the stirred tank; The particle mass transfer characteristic prediction module is configured to combine the analysis results of the interaction force between the particles and the fluid with the predicted value of the mass transfer coefficient of the stirred tank to obtain the mass transfer law between the phases in the stirred tank.

9. A computer-readable storage medium, characterized in that A plurality of instructions are stored therein, and the instructions are suitable for being loaded by a processor of a terminal device and executing the particle mass transfer characteristic prediction method based on a machine learning algorithm according to any one of claims 1 to 7.

10. A terminal device, characterized in that: It includes a processor and a computer-readable storage medium, the processor is used to implement each instruction; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded by the processor and executed by the particle mass transfer characteristics prediction method based on the machine learning algorithm according to any one of claims 1 to 7.

Images