A method for predicting the screening efficiency of a vibrating screen

Abstract

The present application relates to a kind of vibration screen screening efficiency prediction method, belong to screening efficiency prediction technical field, solve the problem of low prediction accuracy, large amount of calculation, poor real-time and poor universality of vibration screen screening efficiency prediction method in prior art.The discrete element model of vibration screen and the spatial model of particle plant are constructed;Screening efficiency sample data set is constructed based on the discrete element model of vibration screen and the spatial model of particle plant;For each size of particle, based on screening efficiency sample data set, train multiple machine learning models to predict the machine learning model with the highest screening efficiency accuracy as the screening efficiency prediction model for this kind of particle;Acquire the sample to be measured, based on the size of particle selects corresponding screening efficiency prediction model, with the percentage of feed particles and the total amount of feed particles as input parameters, the screening efficiency corresponding to the size of particle can be obtained.A kind of vibration screen screening efficiency prediction method with high prediction accuracy, high computational efficiency, strong real-time and being able to adapt to various complex conditions is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of screening efficiency prediction, in particular to a vibrating screen screening efficiency prediction method. BACKGROUND

[0002] The vibrating screen is a screening device widely used in coal, mining, metallurgy, building materials, chemical industry and other industries. Its working principle is to make the screen box produce periodic vibration by using the exciting force generated by the vibrating motor or exciter, so that the material is thrown up and moves forward on the screen surface, thereby realizing the separation of materials of different particle sizes. The screening efficiency is an important indicator to measure the working effect of the vibrating screen, which is affected by many factors such as the running state of the screening equipment (amplitude, vibration frequency, vibration intensity, rotating speed, etc.), structural parameters (screen, screen opening rate, aperture size, inclination angle, etc.) and material characteristics (moisture content, feeding rate, particle size distribution, etc.).

[0003] In the prior art, there are mainly two methods to calculate the screening efficiency: (1) constructing a screening efficiency prediction model based on dynamics and probability theory, and predicting the screening efficiency of the vibrating screen through theoretical derivation and experimental data fitting. Although the established model can explain the basic phenomena in the screening process to some extent, its prediction accuracy is often unsatisfactory when facing complex working conditions and diversified materials; (2) discrete element method, which simulates the interaction between particles and between particles and screen surface by numerically solving the particle motion equation, can more realistically reproduce the screening process. For example, DEM simulation has been used to evaluate the screening efficiency of circular vibrating screen, and the results show that different vibration parameters have a significant impact on the screening efficiency. However, each simulation using the discrete element method is for specific screening parameters, and can only predict the screening efficiency under specific parameters. If the screening efficiency of the vibrating screen under multiple conditions is to be predicted, a large amount of computing resources and storage resources will be consumed, which makes large-scale industrial application or real-time simulation difficult and expensive.

[0004] Therefore, it is necessary to provide a vibrating screen screening efficiency prediction method with high prediction accuracy, high computing efficiency, strong real-time performance and adaptability to various complex working conditions. SUMMARY

[0005] In view of the above analysis, the embodiments of the present application aim to provide a vibrating screen screening efficiency prediction method to solve the problems of low prediction accuracy, large calculation amount, poor real-time performance and poor universality of the existing vibrating screen screening efficiency prediction method.

[0006] The embodiments of the present application provide a vibrating screen screening efficiency prediction method, which comprises:

[0007] constructing a discrete element model of the vibrating screen and a particle factory space model;

[0008] constructing a screening efficiency sample dataset based on the vibrating screen discrete element model and the particle plant space model, the screening efficiency sample dataset comprising: particle size, percentage of feed particles, total amount of feed particles, and particle screening efficiency;

[0009] For each size of particle, training a plurality of machine learning models based on the screening efficiency sample dataset to predict the machine learning model with the highest screening efficiency accuracy as the screening efficiency prediction model for that size of particle;

[0010] Obtaining a sample to be tested, selecting the corresponding screening efficiency prediction model based on the particle size, and taking the percentage of feed particles and the total amount of feed particles as input parameters to obtain the screening efficiency corresponding to the particle size.

[0011] Further improvement based on the above method, the construction of the vibrating screen discrete element model and the particle plant space model comprises: constructing a vibrating screen discrete element model, the vibrating screen discrete element model comprising: a screen and a material baffle, the material baffle being located on the feed side, and the screen discrete element model being divided based on a predetermined screen length to obtain a plurality of sub-screen discrete element models with the same screen length; constructing a particle plant space model, the particle plant space model being arranged above the screen discrete element model and closely adjacent to the material baffle, and the particle generation mode and particle parameters of the particle plant space model being set based on business requirements.

[0012] Further improvement based on the above method, the construction of the vibrating screen discrete element model and the particle plant space model comprises:

[0013] A1: realizing vibrating screen screening process simulation based on the vibrating screen discrete element model and the particle plant space model, and obtaining simulation results, the simulation results comprising: the spatial position of each particle, the percentage of feed particles, and the total amount of feed particles corresponding to each time step during the simulation process;

[0014] A2: calculating the particle screening efficiency of different particle sizes corresponding to each sub-screen discrete element model based on the spatial position of each particle;

[0015] A3: taking the percentage of feed particles, the particle screening efficiency, and the total amount of feed particles of each particle at each time step corresponding to each sub-screen discrete element model as a sample in the screening efficiency sample dataset corresponding to the screen length;

[0016] A4: determining whether the current generated sample data quantity meets the predetermined quantity requirement, if yes, stopping generation; if not, adjusting the current particle generation mode and particle parameters based on business requirements, and returning to A1.

[0017] Based on the further improvement of the above method, the simulation result refers to obtaining the percentage of the feed particles, the total amount of the feed particles and the spatial position of each particle for each sub-sieve discrete element model in each time step after the screening process reaches a steady state; the steady state refers to the sum of the particle screening rate of the sieve discrete element model and the particle outflow rate at the discharge end per unit time being equal to the particle feed rate.

[0018] Based on the further improvement of the above method, the calculation of the particle screening efficiency corresponding to different particle sizes of each sub-sieve discrete element model based on the spatial position of each particle includes:

[0019] For the first sub-sieve discrete element model close to the feed end, the screening efficiency of each particle size is calculated based on the particle size, the percentage of the feed particles, the total amount of the feed particles generated in the particle factory spatial model, and the number of particles of various sizes under the first sub-sieve discrete element model;

[0020] For other sub-sieve discrete element models, the particle size, the percentage of the particles and the total amount of the particles at the discharge end of the previous sub-sieve discrete element model are taken as the particle size, the percentage of the feed particles and the total amount of the feed particles of the sub-sieve discrete element model, and the screening efficiency of each particle size is calculated based on the particle size, the percentage of the feed particles and the total amount of the feed particles of the sub-sieve discrete element model and the number of particles of various sizes under the sub-sieve discrete element model.

[0021] Based on the further improvement of the above method, the training of multiple machine learning models based on the screening efficiency sample data set includes:

[0022] Data preprocessing is performed on the screening efficiency sample data set to obtain a training sample data set;

[0023] The hyperparameters of each machine learning model are optimized based on the random grid search method, and the optimization results are taken as the initial hyperparameters of each machine learning model;

[0024] Each machine learning model is trained by K-fold cross-validation after optimization.

[0025] Based on the further improvement of the above method, the particle generation mode adopts a dynamic mode, and the particle parameters include a particle size distribution, which is 3mm, 5mm, 7mm and 10mm, respectively.

[0026] Based on the further improvement of the above method, the machine learning model with the highest predicted screening efficiency is taken as the screening efficiency prediction model for the particle, which includes:

[0027] For 3mm particles, the screening efficiency prediction model is a K-nearest neighbor regression model;

[0028] For particles of 5mm, the screening efficiency prediction model is a multilayer perceptron regression model.

[0029] Based on the further improvement of the above method, the method further comprises: after obtaining the multilayer perceptron regression model for 5mm, pruning operation is performed on it.

[0030] Based on the further improvement of the above method, the pruning operation comprises: calculating the importance score of each hidden layer in the current multilayer perceptron regression model, calculating the weight threshold of each neuron based on the importance score and the custom weight for each hidden layer, removing the neurons with weights lower than the weight threshold, and fine-tuning the pruned network to ensure that the prediction performance is not affected.

[0031] The threshold of each neuron is evaluated based on the importance score and the custom weight, comprising:

[0032]

[0033] Wherein,

[0034]

[0035] w ij is the weight of the jth neuron of the ith hidden layer, sigma i is the importance score of the ith hidden layer, sigma is a constant, w s is the standard weight threshold of the ith hidden layer, w avg is the average weight of all neurons in the ith hidden layer, 0 < sigma i < 1, 0 < sigma < 1.

[0036] Compared with the prior art, the present application can at least achieve one of the following beneficial effects:

[0037] 1. The present application provides a vibration screen screening efficiency prediction method, which is based on a predetermined length to divide the vibration screen discrete element model to obtain a plurality of sub-screen discrete element models with the same length. For each simulation of the particle plant space model, after steady state, each time step can generate a plurality of sets of simulation data corresponding to the working condition and the predetermined length screen. Compared with the prior art, each time step can only generate a set of simulation data corresponding to the working condition and the predetermined length. The present application has higher efficiency in constructing the screening efficiency sample data set. In addition, the various sub-screen discrete element models are arranged continuously, and the discharge of the previous sub-screen discrete element model serves as the feed of the next sub-screen discrete element model. Compared with the prior art which only obtains sample data for a fixed length screen model, the screening sample data obtained by the simulation of each sub-screen discrete element model in the present application is closer to the real vibration screen screening process.

[0038] 2. The application provides a vibrating screen screening efficiency prediction method, which is based on a machine learning model to construct a screening efficiency prediction model for different particle sizes, and can more flexibly meet business needs; in addition, compared with the way of obtaining screening efficiency in the prior art of dynamics and probability theory and discrete element method, the method proposed in the application has higher prediction accuracy and calculation efficiency, strong real-time performance and can adapt to various complex working conditions.

[0039] 3. The application provides a vibrating screen screening efficiency prediction method, which can reduce the calculation amount and memory requirement, improve the model efficiency and generalization ability, and reduce energy consumption by pruning the target screening efficiency prediction model after obtaining the screening efficiency prediction model; further, the weight threshold of each neuron is calculated based on the importance score and the custom weight to reduce the process of neuron reduction, which is a dynamic process, realizes efficient model compression, and can quickly realize a compact model.

[0040] In the application, the above technical solutions can be combined with each other to realize more preferred combination solutions. Other features and advantages of the application will be described in the subsequent specification, and some advantages will become apparent from the specification or by implementing the application. The purpose and other advantages of the application can be achieved and obtained from the contents specifically pointed out in the specification and the drawings. BRIEF DESCRIPTION OF DRAWINGS

[0041] The accompanying drawings are included to provide a better understanding of the embodiments of the application, and are not considered as limiting the application, and throughout the drawings, the same reference signs represent the same components;

[0042] Figure 1 An example diagram of a vibrating screen screening efficiency prediction method in an embodiment of the application;

[0043] Figure 2 An example diagram of a vibrating screen discrete element model in an embodiment of the application. DETAILED DESCRIPTION

[0044] The preferred embodiments of the application will be specifically described below in combination with the drawings, wherein the drawings constitute a part of the application, and are used together with the embodiments of the application to explain the principles of the application, and are not used to limit the scope of the application.

[0045] One specific embodiment of the application discloses a vibrating screen screening efficiency prediction method, as shown in Figure 1 , comprising:

[0046] S1: constructing a vibrating screen discrete element model and a particle plant space model, comprising:

[0047] S11: Construct a vibrating screen discrete element model, the vibrating screen discrete element model comprising: a screen and a material baffle, the material baffle being located at an inlet side, and the screen discrete element model being divided based on a preset screen length to obtain a plurality of screen length identical sub-screen discrete element models.

[0048] Exemplarily, the vibrating screen discrete element model can be constructed by using SolidWorks. Further, considering the factors of calculation efficiency, resource consumption and the like, the vibrating screen discrete element model is removed of unnecessary details and complex structures in the present application, and the vibrating screen discrete element model is finally simplified into a screen discrete element model and a material baffle discrete element model based on business requirements, as shown in FIG. 1. Figure 2 As shown in FIG. 1, the left side is the material baffle, and the lower side is the screen. In the present application, it is defined that the left side of the screen is the inlet end, and the right side is the outlet end. Understandably, the material baffle is used to limit the activity range of the particles, and the upper surface of the screen discrete element model is used to filter materials of different particle sizes.

[0049] After obtaining the screen discrete element model, the screen discrete element model can be divided based on a preset screen length specified by business requirements to obtain a plurality of screen length identical sub-screen discrete element models. For the preset screen length, the present application does not make a limitation here, and the value of the preset screen length can be set according to business requirements or actual requirements. Exemplarily, the preset screen length can be set to 250 mm.

[0050] S12: Construct a particle factory space model, the particle factory space model being arranged above the screen discrete element model and close to the material baffle, and the particle generation mode and particle parameters of the particle factory space model being set based on business requirements.

[0051] Figure 2 The position of the particle factory space model is not shown in FIG. 1, but it is generally placed at the upper left of the screen (i.e., above the inlet end) so that the particles generated thereby can fall into the screen area.

[0052] The present application does not make a limitation on the software for constructing the discrete element model, and discrete element software such as EDEM, YADE, LIGGGHTS, DEMms, SDEM and the like can be used. Exemplarily, EDEM can be used to define the particle factory space, including:

[0053] Particle definition: the user can define the physical properties of the particles, such as shape (spherical, blocky, fibrous, etc.), size, mass, density and the like;

[0054] Particle generation: Through the API source code, the generation process of particles in the simulation space can be realized, including the position, speed, rotation angle, etc. of particle generation and the particle generation method. Users can write code to generate single particles or batch generate particle groups with different attributes (for example, generate particle groups with different particle sizes according to the different particle sizes defined in the particle definition step).

[0055] Particle behavior control: The particle factory API provides methods to control the motion and interaction behavior of particles. Users can specify the collision, friction, combination between particles, and the interaction between particles and simulation devices, etc.

[0056] Customized simulation: By modifying and extending the API source file, users can create customized simulation scenarios to achieve specific particle flow patterns, particle loading and unloading processes, particle-device interactions, etc.

[0057] Integration of external data: The API supports importing initial conditions and attribute parameters of particles from external data sources, such as importing particle shapes from CAD models or obtaining particle physical properties from experimental data.

[0058] Performance optimization: By writing API code, users can optimize simulation performance, such as reducing simulation calculation time through algorithm optimization or adjusting particle management strategies during simulation.

[0059] Visual display: During simulation, the API also provides visual output of particle state. Users can control the output of particle position, speed, etc. through programming to analyze and display the results.

[0060] User interaction and automation: The API allows users to interactively control the simulation process, such as starting, pausing, and stopping the simulation, and dynamically adjusting particle parameters during simulation. At the same time, users can automate a series of simulation tasks through the API to improve work efficiency.

[0061] The above is the common operation of the particle factory space. The purpose of the particle factory space is to generate particles that meet business requirements. The specific operation of generating particles, such as the generation method of particles and the particle size, is not limited in the present application, and can be determined according to the business requirements. For example, the particle generation method can be dynamic, and the particle size can be 3mm, 5mm, 7mm, or 10mm.

[0062] In addition, periodic boundaries are used on both sides of the area between the screen inlet and outlet to limit the activity range of the particles generated by the particle factory space.

[0063] S2: constructing a screening efficiency sample dataset based on the vibration screen finite element model and the particle plant space model, the screening efficiency sample dataset comprising: particle size, percentage of feed particles, total amount of feed particles, and particle screening efficiency.

[0064] The constructing a screening efficiency sample dataset based on the vibration screen discrete element model and the particle plant space model comprises:

[0065] A1: realizing vibration screen screening process simulation based on the vibration screen discrete element model and the particle plant space model, and obtaining simulation results, the simulation results comprising: spatial position of each particle, percentage of feed particles, total amount of feed particles corresponding to each time step in the simulation process, wherein the total amount of feed particles in the present application refers to the total number of particles with different sizes generated by the particle plant space model according to business requirements.

[0066] The simulation results refer to the percentage of feed particles, the total amount of feed particles, and the spatial position of each particle corresponding to each sub-screen discrete element model in each time step after the screening process reaches a steady state.

[0067] It can be understood that before the simulation process, simulation parameters need to be set, including screen surface movement mode, screen surface properties, screen surface movement state, screening material shape and properties, feed amount, feed speed, and each simulation parameter can be set according to business requirements. Exemplarily, the screen surface movement mode can be simple harmonic vibration.

[0068] After setting the simulation parameters, iterative simulation can be performed until the sum of the particle mass passing through the screen per unit time and the particle discharge rate at the discharge end is equal to the particle feed rate, that is, the screening process enters a steady state, and the discrete element iterative solution is continued, and the discrete element simulation data is saved for at least 10s above the steady state.

[0069] The discrete element simulation data mainly includes position information (three-dimensional coordinates), movement information (force, speed, acceleration, angular velocity, torque, etc.), and attribute information (mass, density, Young's modulus, sliding friction coefficient, rolling friction coefficient, etc.) of particles and geometric bodies (screen mesh). Taking EDEM as an example, the information of particles and geometric models at each time step is stored in an h5 format file, and the naming method is "time step +.h5".

[0070] In the present application, according to business requirements, the data required by the present application is the spatial position of each particle, the percentage of feed particles, and the total amount of feed particles corresponding to each time step.

[0071] The present application divides the vibration screen discrete element model into a plurality of sub-screen discrete element models of the same length according to the preset length, each sub-screen discrete element model is arranged continuously, and the discharge of a previous sub-screen discrete element model is used as the feed of a next sub-screen discrete element model.

[0072] A2: calculating the particle screening efficiency corresponding to each sub-screen discrete element model according to the spatial position of each particle, including:

[0073] For the first sub-screen discrete element model close to the feed end, the screening efficiency of each particle size is calculated according to the particle size generated in the particle factory spatial model, the percentage of the feed particles, the total amount of the feed particles, and the number of particles of each size under the first sub-screen discrete element model.

[0074] For other sub-screen discrete element models, the particle size, the percentage of the particles, and the total amount of the particles at the discharge end of a previous sub-screen discrete element model are used as the particle size, the percentage of the feed particles, and the total amount of the feed particles of the sub-screen discrete element model, and the screening efficiency of each particle size is calculated according to the particle size, the percentage of the feed particles, and the total amount of the feed particles of the sub-screen discrete element model and the number of particles of each size under the sub-screen discrete element model.

[0075] For the screening efficiency of each particle size in each sub-screen discrete element model, the screening efficiency is calculated by the following method:

[0076] B1: counting the number Q of the particles of the size under the screen;

[0077] B2: calculating the product of the percentage of the feed particles and the total amount of the feed particles of the size as the total amount M of the particles of the size;

[0078] B3: the screening efficiency S = Q / M.

[0079] A3: using the percentage of the feed particles, the screening efficiency, and the total amount of the feed particles of each particle of each time step corresponding to each sub-screen discrete element model as a sample in the screening efficiency sample data set corresponding to the screen length.

[0080] For each working condition simulation of the particle factory spatial model, after the steady state, a plurality of groups of simulation data corresponding to the working condition and the preset length screen can be generated at each time step. For example, if the current screen length is 1000 mm and the preset screen length is 250 mm, the current screen can be divided into four 250 mm sub-screens, and therefore, for the same working condition parameters, four sample data corresponding to the 250 mm screen length can be obtained simultaneously at each time step by using the method proposed in the present application.

[0081] A4: judging whether the current generated sample data quantity meets the preset quantity requirement, if yes, stopping generating; if not, adjusting the current particle generation mode and particle parameters based on the business requirement, and returning to A1.

[0082] The machine learning model is used as the screening efficiency prediction model in the application, and the number of training samples is an important indicator affecting the prediction accuracy of the machine learning model, therefore, the number of training sample data needs to be set to make the prediction effect of the machine learning model meet the requirement.

[0083] If it is determined that the current generated sample data quantity does not meet the requirement, the particle generation mode and particle parameters in the construction of the particle factory space model can be adjusted to generate training samples under a new working condition. Exemplarily, the sample data quantity can be set to 5000-10000.

[0084] S3: training multiple machine learning models based on the screening efficiency sample data set for each size of particle to predict the machine learning model with the highest screening efficiency accuracy as the screening efficiency prediction model of the particle.

[0085] Understandably, the application constructs the sample data set corresponding to each size of particle when constructing the screening efficiency sample data set, which can be referred to as step A3. Therefore, when obtaining the screening efficiency prediction model of the particle, the sample data set corresponding to the size of the particle is used to train the machine learning model, so that the screening efficiency prediction model of the particle size can be obtained.

[0086] The training of multiple machine learning models based on the screening efficiency sample data set comprises:

[0087] The screening efficiency sample data set is subjected to data preprocessing to obtain a training sample data set, the hyperparameters of each machine learning model are optimized based on a random grid search method, the optimization result is used as the initial hyperparameters of each machine learning model, and each machine learning model after optimization is trained by K-fold cross validation.

[0088] The machine learning model can be a support vector machine, a random forest, a Gaussian process regression, a BP neural network regression, a gradient boosting regression, etc., which is not limited in the application. Moreover, the random grid search method and the K-fold cross validation algorithm used in the application are public algorithms in the field, and the specific implementation is not limited, which only improves the prediction accuracy of each machine learning model by the two methods.

[0089] Furthermore, a multi-objective optimization algorithm (for example, a particle swarm algorithm, a genetic algorithm, a Harris hawk algorithm, etc.) can also be used to optimize the hyperparameters of each machine learning model to construct a machine learning prediction model with high prediction accuracy.

[0090] It can be understood that for different vibration screen discrete element models, the corresponding screen efficiency prediction model of each particle size is different.

[0091] For example, for a 250mm long vibration screen in the present application, for a particle of 3mm, the screen efficiency prediction model is a K-neighbor regression model; for a particle of 5mm, the screen efficiency prediction model is a multilayer perception regression model.

[0092] For example, if the screen efficiency prediction model of the final particle is a neural network model, a pruning algorithm can be further used to optimize the model to reduce the amount of calculation and memory requirement, and improve the efficiency and generalization ability of the model. Specifically, the importance score of each hidden layer in the current neural network model is calculated, and for each hidden layer, the weight threshold of each neuron is calculated based on the importance score and the custom weight, the neurons with weights lower than the weight threshold are removed, and the pruned network is fine-tuned to ensure that the prediction performance is not affected; the threshold of each neuron is evaluated based on the importance score and the custom weight, including:

[0093]

[0094] wherein,

[0095]

[0096] w ij is the weight of the jth neuron in the ith hidden layer, σ i is the importance score of the ith hidden layer, σ is a constant, w s is the standard weight threshold of the ith hidden layer, w avg is the average weight of all neurons in the ith hidden layer, 0 < σ i < 1, 0 < σ < 1.

[0097] In addition, using the method provided in the present application, the screen efficiency of particles smaller than a fixed value can also be predicted, such as the screen efficiency of particles smaller than 6mm. In essence, when constructing the training sample, the corresponding percentage of particles smaller than 6mm, the total amount of particles, and the screen efficiency of particles are used as the training sample, and then different machine learning models are used for screen efficiency prediction. The machine learning model with the highest prediction screen efficiency accuracy is used as the screen efficiency prediction model for particles smaller than 6mm.

[0098] S4: Obtain a sample to be tested, select the corresponding screen efficiency prediction model based on the particle size, and use the percentage of particles and the total amount of particles as input parameters to obtain the screen efficiency corresponding to the particle size.

[0099] In practical applications, in order to obtain the screening efficiency of the current material, the types of particles contained in the current material (e.g., size) can be identified, and the percentage of each particle size can be calculated. The total amount of material and the percentage of that type of particle are used as input parameters and input into the screening efficiency prediction model for that particle size, thereby obtaining the screening efficiency of that type of particle in the current material.

[0100] Compared with existing technologies, the vibrating screen screening efficiency prediction method provided in this embodiment divides the vibrating screen discrete element model into multiple sub-screen discrete element models of the same length based on a preset length. For each working condition simulation of the particle plant spatial model, after steady state, each time step can generate multiple sets of simulation data corresponding to the working condition and the preset length of the screen. Compared with the existing technology, which can only generate one set of simulation data corresponding to the working condition and the preset length at each time step, the present invention has a higher efficiency in constructing screening efficiency sample datasets. In addition, the sub-screen discrete element models are arranged continuously, and the output of the previous sub-screen discrete element model is used as the input of the next sub-screen discrete element model. Compared with the existing technology, which only obtains sample data for a screen model of a fixed length, the screening sample data obtained by simulating each sub-screen discrete element model in the present invention is closer to the actual vibrating screen screening process. A screening efficiency prediction model for different particle sizes is constructed based on a machine learning model, which can more flexibly meet business needs. Furthermore, compared to existing methods using dynamics, probability theory, and discrete element methods, the proposed method for obtaining screening efficiency using a machine learning model offers higher prediction accuracy and computational efficiency, stronger real-time performance, and adaptability to various complex working conditions. After obtaining the screening efficiency prediction model, pruning is performed on the target screening efficiency prediction model, which reduces computational load and memory requirements, improves model efficiency and generalization ability, and reduces energy consumption. Moreover, the process of calculating the weight threshold for each neuron based on importance scoring and custom weights for neuron reduction is a dynamic process, achieving efficient model compression and enabling rapid implementation of a compact model.

[0101] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0102] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the screening efficiency of a vibrating screen, characterized in that, include: Construct a discrete element model of a vibrating screen and a spatial model of a particle factory; Based on the discrete element model of the vibrating screen and the spatial model of the particle factory, a screening efficiency sample dataset is constructed. The screening efficiency sample dataset includes: particle size, percentage of feed particles, total amount of feed particles, and particle screening efficiency. For each particle size, multiple machine learning models are trained based on the screening efficiency sample dataset, and the machine learning model with the highest accuracy in predicting screening efficiency is used as the screening efficiency prediction model for that particle size. The sample to be tested is obtained, the particle size types in the sample are identified, and the percentage and total amount of each particle size are identified. Based on the particle size type, the corresponding screening efficiency prediction model is selected. The percentage and total amount of each particle size are used as input parameters to obtain the screening efficiency for each particle size, thus obtaining the screening efficiency of the sample to be tested. The construction of the discrete element model of the vibrating screen and the spatial model of the particle factory includes: A discrete element model of a vibrating screen is constructed. The discrete element model of the vibrating screen includes a screen and a material baffle. The material baffle is located on the feed side. The screen discrete element model is divided into multiple sub-screen discrete element models with the same screen length based on a preset screen length. Construct a particle factory spatial model, which is set above the screen discrete element model and close to the material baffle, and set the particle generation method and particle parameters of the particle factory spatial model based on business requirements; The construction of the screening efficiency sample dataset based on the discrete element model of the vibrating screen and the spatial model of the particle factory includes: A1: Based on the discrete element model of the vibrating screen and the spatial model of the particle factory, the vibrating screen screening process is simulated, and the simulation results are obtained. The simulation results include the spatial position of each particle at each time step in the simulation process, the percentage of feed particles, and the total amount of feed particles. A2: Calculate the particle sieving efficiency for different particle sizes corresponding to each sub-sieve discrete element model based on the spatial position of each particle; A3: Take the percentage of each type of particle in the feed, the particle screening efficiency, and the total amount of feed particles at each time step corresponding to the discrete element model of each sub-sieve as a sample in the screening efficiency sample dataset corresponding to that sieve length. A4: Determine whether the amount of sample data generated meets the preset quantity requirement. If it does, stop generating. If it does not, adjust the current particle generation method and particle parameters based on business needs and return to A1.

2. The method for predicting the screening efficiency of a vibrating screen according to claim 1, characterized in that, include: The simulation results refer to the percentage of feed particles, the total number of feed particles, and the spatial position of each particle in the discrete element model of each sub-screen at each time step after the screening process reaches a steady state. The steady state refers to the sum of the particle screening rate and the particle outflow rate at the discharge end of the screen discrete element model equaling the particle feeding rate per unit time.

3. The method for predicting the screening efficiency of a vibrating screen according to claim 2, characterized in that, The calculation of particle sieving efficiency for different particle sizes corresponding to each sub-sieve discrete element model based on the spatial position of each particle includes: For the first sub-screen discrete element model near the feed end, the screening efficiency for each particle size is calculated based on the particle size, feed particle percentage, total feed particle quantity generated in the particle factory spatial model, and the number of particles of various sizes below the first sub-screen discrete element model. For other sub-screen discrete element models, the particle size, particle percentage, and total number of particles at the discharge end of the previous sub-screen discrete element model are used as the particle size, feed particle percentage, and total feed particle number of this sub-screen discrete element model. Based on the particle size, feed particle percentage, and total feed particle number of this sub-screen discrete element model, as well as the number of particles of various sizes below this sub-screen discrete element model, the screening efficiency for each particle size is calculated.

4. The method for predicting the screening efficiency of a vibrating screen according to claim 3, characterized in that, The training of multiple machine learning models based on the sieving efficiency sample dataset includes: The screening efficiency sample dataset is preprocessed to obtain the training sample dataset; The hyperparameters of each machine learning model are optimized using a random grid search method, and the optimization results are used as the initial hyperparameters of each machine learning model. K-fold cross-validation was used to train the optimized machine learning models.

5. The method for predicting the screening efficiency of a vibrating screen according to claim 4, characterized in that, include: The particle generation method is dynamic, and the particle parameters include particle size distribution, which are 3mm, 5mm, 7mm and 10mm respectively.

6. The method for predicting the screening efficiency of a vibrating screen according to claim 5, characterized in that, The method of using the machine learning model with the highest predicted screening efficiency as the screening efficiency prediction model for this type of particle includes: For 3mm particles, the screening efficiency prediction model is the K-nearest neighbor regression model; For 5mm particles, the screening efficiency prediction model is a multilayer perceptron regression model.

7. The method for predicting the screening efficiency of a vibrating screen according to claim 6, characterized in that, The method further includes: After obtaining the multilayer perceptron regression model corresponding to 5mm, pruning is performed on it.

8. The method for predicting the screening efficiency of a vibrating screen according to claim 7, characterized in that, The pruning operation includes: Calculate the importance score of each hidden layer in the current multilayer perceptron regression model. For each hidden layer, calculate the weight threshold of each neuron based on the importance score and custom weights. Subtract neurons with weights lower than the weight threshold and fine-tune the pruned network to ensure that the prediction performance is not affected. The threshold for evaluating each neuron based on importance scores and custom weights includes: ， in, ， Let be the weight of the j-th neuron in the i-th hidden layer. Score the importance of the i-th hidden layer. It is a constant. The standard weight threshold for the i-th hidden layer. Let be the average weight of all neurons in the i-th hidden layer. , .

Images