A method for predicting the efficiency of a sieve

Abstract

The present application relates to a kind of screening efficiency prediction method, belong to screening efficiency prediction technical field, solve the problem that accurate, effective screening efficiency cannot be obtained in time in prior art. The historical screening sample data of vibrating screen is obtained, the vibration signal is processed to obtain second characteristic component and replace the vibration signal in historical sample data, to form sample data set, and the sample data set is expanded to obtain training sample data set;Based on training sample data set, a plurality of first training sample subsets with different parameter combinations are constructed, and screening efficiency prediction model corresponding to the first training sample subset is established;The corresponding screening efficiency prediction model is trained based on each first training sample subset, the screening efficiency prediction model with the highest prediction accuracy is used as the target screening efficiency prediction model, the input parameter combination corresponding to the current to-be-measured vibrating screen is obtained, and it is input into target screening efficiency prediction model to obtain the screening efficiency of the to-be-measured vibrating screen. Realize the screening efficiency corresponding to current control parameter is quickly and efficiently obtained.

Description

TECHNICAL FIELD

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

[0002] The vibrating screen is a key equipment widely used in mining, chemical industry, pharmaceutical industry, food industry and other industries. Through the vibration force generated by the vibration motor or exciter, the material on the screen surface is screened, so as to realize the classification and screening of particles.

[0003] The screening efficiency directly affects the efficiency of the production process and the quality of the product. For example, in the mining industry, efficient screening can ensure the uniformity of the particle size of the ore, which helps subsequent crushing and enrichment operations, and improves the smelting efficiency and metal recovery rate; in the pharmaceutical industry, the screening process determines the particle size distribution of the pharmaceutical powder, thereby affecting the dissolution rate and bioavailability of the drug. By continuously optimizing the design and operating parameters of the vibrating screen, the screening efficiency can be further improved to meet the growing production demand and promote the continuous progress and development of related industries.

[0004] However, the design of operating parameters in the prior art mainly includes: (1) designed by technical personnel according to experience, but this way is difficult to obtain the screening efficiency in time in complex production environment, such as processing a plurality of different materials or in harsh working conditions; (2) simulation prediction is carried out by using a finite element model, but the model for predicting the screening efficiency usually needs to consider multiple variables, such as material properties, vibration parameters, screen structure, etc., which makes the model complexity high. For example, the discrete element method DEM simulation needs a large amount of computing resources to process the interaction and motion trajectory between particles. Such a high complexity model may be limited by computing power in practical application, especially when dealing with large-scale data sets or real-time prediction, which makes it difficult to obtain effective screening efficiency. SUMMARY

[0005] In view of the above analysis, the embodiments of the present application aim to provide a screening efficiency prediction method to solve the problem that the accurate and effective screening efficiency cannot be obtained in time in the prior art.

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

[0007] S1: obtaining historical screening sample data of the vibrating screen, each piece of historical screening sample data comprising vibration signals, vibration frequency, excitation force, screen size, screen surface inclination angle, and material particle size, material humidity, feed speed, screening efficiency of the vibrating screen;

[0008] S2: adopting an empirical mode decomposition (EMD) algorithm and a public coupling point (PCC) algorithm to process the vibration signal to obtain an intrinsic mode function (IMF) meeting a requirement, and reconstructing all IMFs meeting the requirement to obtain a first feature component;

[0009] S3: performing Fourier transform on the first feature component to obtain a second feature component, replacing vibration signals in the historical sample data with the second feature component to form a sample data set, and performing data augmentation on the sample data set to obtain a training sample data set;

[0010] S4: constructing a plurality of first training sample subsets with different parameter combinations based on the training sample data set, and establishing a screening efficiency prediction model corresponding to the first training sample subsets;

[0011] S5: training the corresponding screening efficiency prediction model based on each first training sample subset, taking the screening efficiency prediction model with the highest prediction accuracy as a target screening efficiency prediction model, and taking the parameter combination of the first training sample subset corresponding to the target screening efficiency prediction model as an input parameter combination;

[0012] S6: obtaining an input parameter combination corresponding to a current to-be-tested vibration screen, inputting the input parameter combination into the target screening efficiency prediction model to obtain the screening efficiency of the to-be-tested vibration screen.

[0013] Based on the further improvement of the above method, the data augmentation of the sample data set to obtain the training sample data set comprises:

[0014] generating a generative adversarial network for sample data set augmentation; wherein the loss function of the generative adversarial network is:

[0015] L c = λ1δ1+ λ2δ2+ λ3δ3,

[0016] wherein λ1, λ2, λ3 are weight coefficients of the adversarial loss δ1, the content loss δ2 and the diversity loss δ3, respectively, and the diversity loss δ3 is:

[0017]

[0018] wherein z i , z j is a random noise vector, G c () is a generator function, and N is the number of samples in the sample data set.

[0019] Based on the further improvement of the above method, the data augmentation of the sample data set to obtain the training sample data set comprises: screening sample data with a screening efficiency greater than a preset screening threshold in the sample data set to form a first sample set;

[0020] For each sample in the first sample set, the distance between the sample and other samples is calculated, and k nearest neighbor samples of the sample are obtained according to the calculated distance; a fitting sample corresponding to the sample is generated according to the sample and the corresponding k nearest neighbor samples; the generated fitting sample is added to the sample data set to obtain a training sample data set.

[0021] Based on the further improvement of the above method, the generating of the fitting sample corresponding to the sample according to the sample and the corresponding k nearest neighbor samples comprises:

[0022]

[0023] wherein, Samnple New i is a newly generated fitting sample corresponding to the i th sample, Samnple i is the i th sample, Samnple j is the j th neighbor sample corresponding to the i th sample, δ is a random vector, and j = 1, 2, 3, …, k.

[0024] Based on the further improvement of the above method, the processing of the vibration signal by the empirical mode decomposition (EMD) algorithm and the PCC algorithm to obtain the intrinsic mode function (IMF) that meets the requirements, and the reconstruction of all the IMFs that meet the requirements to obtain the first feature component, comprises: performing the EMD algorithm on the obtained vibration signal to generate IMF components; calculating the Pearson correlation coefficient (PCC) between each IMF component and the vibration signal corresponding thereto, and screening the components in the IMF components whose PCC is greater than a first preset threshold to form an IMF component set; sorting the components in the IMF component set in descending order of the PCC, and assigning a larger weight to the IMF component at the front of the sorting and a smaller weight to the component at the back of the sorting, the sum of the weights corresponding to each component being 1, and taking the weighted sum of each component and the weight corresponding thereto as the first feature component.

[0025] Based on the further improvement of the above method, the number or type of parameters included in the sample data of each first training sample subset is different, and the parameters of the sample data in each first training sample subset include a second feature component, material granularity, and material humidity.

[0026] Based on the further improvement of the above method, the screening efficiency prediction model is a neural network model, and the loss function of the neural network model is:

[0027]

[0028] wherein, is a root mean square error, is a smoothing loss, and α, β are constants obtained based on a trial-and-error method; wherein the smoothing loss is obtained by the following way:

[0029]

[0030] wherein y i is a screening efficiency output in the i-th iteration, y i-1 is a screening efficiency output in the (i-1)-th iteration, y i-2 is a screening efficiency output in the (i-2)-th iteration, μ is a smoothing control coefficient, 0 < μ < 1, i is an iteration number, i = 3, 4, 5, …, M.

[0031] Based on the further improvement of the above method, the method further comprises: after obtaining the target screening efficiency prediction model, performing a pruning operation on the target screening efficiency prediction model.

[0032] Based on the further improvement of the above method, the pruning operation comprises: calculating an importance score of each hidden layer in the current target screening efficiency prediction model, for each hidden layer, calculating a weight threshold of each neuron based on the importance score and a custom weight, 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.

[0033] Based on the further improvement of the above method, the threshold of each neuron is evaluated based on the importance score and the custom weight, comprising:

[0034]

[0035] wherein,

[0036]

[0037] w ij is a weight of the j-th neuron in the i-th hidden layer, σ i is an importance score of the i-th hidden layer, σ is a constant, w s is a standard weight threshold of the i-th hidden layer, w avg is an average weight of all neurons in the i-th hidden layer, 0 < σ i < 1, 0 < σ < 1.

[0038] Compared with the prior art, the present application can at least achieve one of the following beneficial effects:

[0039] 1. The application provides a screening efficiency prediction method, which is realized based on a machine learning model, can quickly and efficiently obtain the screening efficiency corresponding to the input parameter combination according to the existing input parameter combination, and enables a technician to adjust the input parameters in time according to the screening efficiency to obtain a satisfactory screening efficiency.

[0040] 2. The application provides a screening efficiency prediction method, which adopts two data sample expansion methods, including generative adversarial network data expansion and K-nearest neighbor data expansion, improves the generalization ability of the screening efficiency prediction model, reduces overfitting, enhances robustness, and simplifies data acquisition cost.

[0041] 3. The application provides a screening efficiency prediction method, which performs pruning operation on the target screening efficiency prediction model after obtaining the target screening efficiency prediction model, can reduce the calculation amount and memory requirement, improve the model efficiency and generalization ability, and reduce energy consumption; further, the weight threshold of each neuron is calculated based on the importance score and the custom weight to delete the neurons, which is a dynamic process, realizes efficient model compression, and can quickly realize a compact model.

[0042] 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 be understood by implementing the application. The purposes and other advantages of the application can be realized and obtained from the contents specifically pointed out in the specification and the drawings. BRIEF DESCRIPTION OF DRAWINGS

[0043] The accompanying drawings are included to provide a further understanding of the application and are incorporated herein and constitute a part of the application.

[0044] Figure 1 FIG. 1 is a flowchart of a screening efficiency prediction method according to an embodiment of the application. DETAILED DESCRIPTION

[0045] The preferred embodiments of the application are specifically described below in combination with the drawings, wherein the drawings constitute a part of the application and are used to explain the principles of the embodiments of the application, and are not used to limit the scope of the application.

[0046] One specific embodiment of the application discloses a screening efficiency prediction method, as shown in FIG. 1, which comprises the following steps. Figure 1

[0047] ​S1: Obtain historical screening sample data of the vibrating screen, each piece of historical screening sample data comprising a vibration signal, a vibration frequency, an exciting force, a screen hole size, a screen surface inclination angle of the vibrating screen, and material granularity, material humidity, feeding speed, and screening efficiency.

[0048] In actual application scenarios, there are many parameters that affect the screening efficiency of the vibrating screen, such as the type of the vibrating screen and the model of the vibration motor, and some parameters are difficult to obtain, and some parameters have less impact on the screening efficiency. Therefore, after the parameters affecting the screening efficiency are statistically sorted, one of the parameter screening methods such as principal component analysis PCA, factor analysis, feature selection algorithm or decision tree is selected, and finally the vibration signal, the vibration frequency, the exciting force, the screen hole size, the screen surface inclination angle and the material granularity, the material humidity and the feeding speed of the vibrating screen are used as the parameters affecting the screening efficiency.

[0049] The vibration signal obtained in the present application is collected based on a sensor installed on the vibrating screen. The sensor can be installed at positions such as the screen box, screen mesh and support of the vibrating screen. The present application does not limit the number and collection positions of the sensors for collecting the vibration signal, which can meet the actual needs. It can be understood that the vibration frequency and the exciting force are directly determined by the type of the vibrating screen. The screen hole size, the screen surface inclination angle, the material granularity, the material humidity, the feeding speed, the feeding amount, the opening rate and the bulk density can be determined according to the actual scene.

[0050] In addition, the screening efficiency of the vibrating screen has a close relationship with the vibration signal, therefore, the more the collected vibration signals, the more accurate the prediction result of the final screening efficiency.

[0051] Preferably, the vibration signals collected by the present application are three, including one screen box, one screen mesh and one support.

[0052] S2: The vibration signal is processed by using an empirical mode decomposition EMD algorithm and a public coupling point PCC algorithm to obtain a required intrinsic mode function IMF, and a first feature component is obtained by reconstructing all the required intrinsic mode functions IMFs.

[0053] Specifically, the empirical mode decomposition EMD algorithm is performed on the obtained vibration signal to generate an IMF component; the Pearson correlation coefficient PCC between each IMF component and the corresponding vibration signal is calculated, and the components with the Pearson correlation coefficient PCC greater than a first preset threshold value in the IMF components are selected to form an IMF component set; each component in the IMF component set is sorted in descending order of the Pearson correlation coefficient PCC, and the IMF components at the front of the sorting are assigned a larger weight, and the components at the back are assigned a smaller weight, the sum of the weights corresponding to each component is 1, and the weighted sum result of each component and the corresponding weight is taken as the first feature component.

[0054] Exemplarily, the first preset threshold value can be 0.7, and the first feature component IMf total is:

[0055] IMf total = W1*IMf1+ W2*IMf2+ W3*IMf3+ … + W N *LMf N ,

[0056] wherein,

[0057] W1+ W2+ W3+ … + W N = 1,

[0058] IMf1> IMf2> IMf3> … > IMf N ,

[0059] W1> W2> W3> … > W N ,

[0060] wherein, W N is a weight corresponding to IMf N , IMf N is a component in the IMF component set, and N is the number of components in the IMF component set.

[0061] S3: performing Fourier transform on the first feature component to obtain a second feature component, replacing the vibration signal in the historical sample data with the second feature component to form a sample data set, and performing data augmentation on the sample data set to obtain a training sample data set.

[0062] wherein, performing Fourier transform on the first feature component to obtain a second feature component comprises: performing Fourier transform on the first feature component to obtain a frequency domain signal, extracting the mean value, peak value and skewness of the frequency domain signal, and taking the mean value, peak value and skewness as the second feature component.

[0063] Optionally, the historical screening sample data can be subjected to data cleaning, data conversion, data dimension reduction, data reconstruction and other preprocessing operations, so as to improve the data quality and improve the accuracy of the screening efficiency prediction. It should be noted that the data preprocessing operation can be determined according to the business requirements, and the present application does not limit it here. It is subject to the ability to meet business requirements. Preferably, the material particle size and material humidity can be normalized to obtain a first material particle size and a first material humidity. At this time, the vibration signal, the material particle size and the material humidity in the historical sample data can be replaced with the second feature component, the first material particle size and the first material humidity. By normalizing part of the data, the accuracy of the screening efficiency prediction model can be further improved.

[0064] In addition, the application proposes two data sample expansion methods, including:

[0065] Method one, using a generative adversarial network for data expansion, the loss function of the generative adversarial network is:

[0066] L c =λ1δ1+λ2δ2+λ3δ3,

[0067] Wherein, λ1, λ2, λ3 are weight coefficients of the adversarial loss δ1, the content loss δ2 and the diversity loss δ3 respectively. The adversarial loss is used to improve the quality of the generated sample, and the calculation method is represented as:

[0068] δ1=E x～pdata [logD c (x)]+E z～pz [log(1-D c (G c (z)))],

[0069] Wherein, pdata is the distribution of real data, z is random noise, pz is the distribution of random noise, D c () is the discriminator function, G c () is the generator function, E represents the expectation, and x represents the real sample. The calculation method of the content loss is represented as:

[0070]

[0071] Wherein, |||2 is the L2 norm, Nc is the sample number, x i is the i-th real sample, and z i is the i-th random noise vector. The diversity loss is:

[0072]

[0073] Wherein, z i , z j are random noise vectors, G c () is the generator function, and N is the number of samples in the sample data set.

[0074] Method two, selecting sample data in the sample data set whose screening efficiency is greater than a preset screening threshold to form a first sample set, and for any sample in the first sample set, the following operations are performed:

[0075] A1: for each sample, calculate the distance between the sample and other samples, and obtain k nearest neighbor samples of the sample according to the calculated distance.

[0076] A2: generating a fitting sample corresponding to the sample according to the sample and the corresponding k nearest neighbor samples, including:

[0077]

[0078] wherein SamnpleNew i is a newly generated fitting sample corresponding to the i th sample, Sample i is the i th sample, Sample j is the j th adjacent sample corresponding to the i th sample, δ is a random vector, j = 1, 2, 3,..., k.

[0079] A3: adding the generated fitting sample to the sample data set to obtain a training sample data set.

[0080] The present application is not limited to using any way to expand the sample, and one way can be used only, or two ways can be used simultaneously, so as to meet the actual sample quantity required by the user.

[0081] After the sample is expanded by the method one or the method two, the expanded sample is merged with the sample data set to obtain a training sample data set.

[0082] S4: constructing a plurality of first training sample subsets with different parameter combinations based on the training sample data set, and establishing a screening efficiency prediction model corresponding to the first training sample subset.

[0083] wherein the number or type of parameters contained in the sample data of each first training sample subset is different, and the parameters of the sample data in each first training sample subset include the second characteristic component, the material particle size and the material humidity. The establishment of the screening efficiency prediction model corresponding to the first training sample subset means that the input parameters of the screening efficiency prediction model correspond to the number of parameters in the first training sample subset.

[0084] S5: training the corresponding screening efficiency prediction model based on each first training sample subset, taking the screening efficiency prediction model with the highest prediction accuracy as the target screening efficiency prediction model, and taking the parameter combination of the first training sample subset corresponding to the target screening efficiency prediction model as the input parameter combination.

[0085] The screening efficiency in the first training sample subset is taken as the true value, and the remaining parameters are taken as the input data to train the screening efficiency prediction model. When the change value of the loss function value of the screening efficiency prediction model in continuous multiple iteration processes is less than a preset threshold value or reaches an iteration upper limit, the training of the screening efficiency prediction model is completed. Specifically, after obtaining the first training sample subset, the first training sample subset is divided into a training set, a test set and a validation set according to a certain ratio, and then the screening efficiency prediction model is trained. Preferably, the training set, the test set and the validation set are divided in a ratio of 8:1:1; the preset threshold value is 0.001, and the iteration upper limit is 1000.

[0086] The loss function of the screening efficiency prediction model is:

[0087]

[0088] wherein, is a root mean square error, is a smoothing loss, and α and β are constants obtained based on a trial and error method; wherein the root mean square error is:

[0089]

[0090] The smoothing loss is obtained in the following manner:

[0091]

[0092] wherein y i is the screening efficiency output in the i-th iteration, is the screening efficiency recorded in the sample data set, Y i-1 is the screening efficiency output in the (i-1)-th iteration, y i-2 is the screening efficiency output in the (i-2)-th iteration, μ is a smoothing control coefficient, 0<μ<1, i is an iteration number, i=3, 4, 5,..., M.

[0093] After obtaining each screening efficiency prediction model, the screening efficiency prediction model with the highest prediction accuracy is selected as the target screening efficiency prediction model.

[0094] The screening efficiency prediction model can be one of a CNN, a DNN or an RNN model, and the specific form of the screening efficiency model is not limited in the present application, such as a common BP neural network, an SVM model and the like, so as to realize screening efficiency prediction as a screening standard.

[0095] After obtaining the target screening efficiency prediction model, in order to further reduce the complexity of the model, improve the generalization ability of the model and reduce the risk of overfitting, a pruning operation can be performed on the target screening efficiency prediction model. The pruning operation includes:

[0096] The importance score of each hidden layer in the current target screening efficiency prediction model is calculated, for each hidden layer, the weight threshold of each neuron is calculated based on the importance score and the custom weight, neurons with weight lower than the weight threshold are removed, and the pruned network is fine-tuned to ensure that the prediction performance is not affected.

[0097] The importance score is calculated based on the importance of the current hidden layer, the more important the layer, the higher the importance score obtained, but the sum of all importance scores needs to be 1. The importance score can be calculated by using Garson method, gradient-based sensitivity analysis, deep Taylor decomposition (DTD), random permutation feature importance (PFI), SmoothGrad method singular value decomposition (SVD), etc. The present application does not limit this, and after calculating the importance of each hidden layer, the importance of the layer and the number of hidden layers in the target screening efficiency prediction model are assigned weights.

[0098] For example, the number of hidden layers in the current target screening efficiency prediction model is 5, and the importance is 34251. The importance score of the 3rd layer is 0.4, the importance score of the 4th layer is 0.2, the importance score of the 2nd layer is 0.2, the importance score of the 5th layer is 0.1, and the importance score of the 1st layer is 0.1. According to the importance score, the 1st layer is pruned first, then the 5th layer is pruned, and the layers are pruned in order according to the importance score until the preset pruning rate is met. The pruning rate is set according to actual needs.

[0099] The threshold of each neuron is evaluated based on the importance score and the custom weight, including:

[0100]

[0101] wherein,

[0102]

[0103] w ij is the weight of the jth neuron of 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.

[0104] The weight of the current neuron is set based on the threshold of each neuron, neurons with weight value of 0 are deleted, and the pruned network is fine-tuned to ensure that the prediction performance is not affected. The pruned model is used as the final target screening efficiency prediction model.

[0105] Compared with the prior art, the screening efficiency prediction method provided by the embodiment is realized based on a machine learning model, the screening efficiency corresponding to an existing input parameter combination can be quickly and efficiently obtained according to the input parameter combination, so that a technical personnel can adjust the input parameter in time according to the screening efficiency to obtain a satisfactory screening efficiency; two data sample expansion methods, including a generative adversarial network data expansion and K-neighbor data expansion, are adopted, the generalization ability of the screening efficiency prediction model is improved, overfitting is reduced, the robustness is enhanced, and the data acquisition cost is simplified; after the target screening efficiency prediction model is obtained, a pruning operation is performed on the target screening efficiency prediction model, the calculation amount and the memory requirement can be reduced, the model efficiency and the generalization ability are improved, and the energy consumption is reduced; furthermore, the weight threshold of each neuron is calculated based on the importance score and the custom weight to be used in the process of deleting neurons, which is a dynamic process, efficient model compression is realized, and a compact model can be quickly realized.

[0106] Those skilled in the art can understand that all or part of the processes of the above-mentioned embodiments can be completed by a computer program instructing related hardware, and the program can be stored in a computer readable storage medium. The computer readable storage medium includes a magnetic disk, an optical disk, a read-only memory, a random access memory, etc.

[0107] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method of predicting the efficiency of a sieve, characterized in that, The method comprises the following steps: S1: obtaining historical screening sample data of a vibrating screen, each piece of historical screening sample data comprising a vibration signal, a vibration frequency, an exciting force, a screen aperture size, a screen surface inclination angle, and material particle size, material humidity, feed speed, and screening efficiency of the vibrating screen; S2: processing the vibration signal by using an empirical mode decomposition (EMD) algorithm and a principal component correlation (PCC) algorithm to obtain intrinsic mode functions (IMFs) meeting requirements, and reconstructing all IMFs meeting requirements to obtain a first feature component; S3: performing Fourier transform on the first feature component to obtain a second feature component, replacing the vibration signal in the historical screening sample data with the second feature component to form a sample data set, and performing data augmentation on the sample data set to obtain a training sample data set; S4: constructing a plurality of first training sample subsets with different parameter combinations based on the training sample data set, and establishing a screening efficiency prediction model corresponding to each first training sample subset; S5: training the screening efficiency prediction model based on each first training sample subset, selecting a screening efficiency prediction model with the highest prediction accuracy as a target screening efficiency prediction model, and selecting a parameter combination of a first training sample subset corresponding to the target screening efficiency prediction model as an input parameter combination; S6: obtaining an input parameter combination corresponding to a current vibrating screen to be measured, inputting the input parameter combination into the target screening efficiency prediction model, and obtaining the screening efficiency of the vibrating screen to be measured.

2. The method of predicting the efficiency of a screen according to claim 1, wherein, The data augmentation on the sample data set to obtain the training sample data set comprises: generating the sample data set by using a generative adversarial network (GAN), wherein a loss function of the GAN is: L c = λ1δ1+ λ2δ2+ λ3δ3, wherein λ1, λ2, and λ3 are weight coefficients of an adversarial loss δ1, a content loss δ2, and a diversity loss δ3, respectively, and the diversity loss δ3 is: where z i , z j are random noise vectors, G c () is a generator function, and N is the number of samples in the sample dataset.

3. The method of predicting the efficiency of a screen according to claim 1, wherein The data augmentation on the sample data set to obtain the training sample data set comprises: selecting sample data with a screening efficiency greater than a preset screening threshold in the sample data set to form a first sample set; for each sample in the first sample set, performing the following operations: for each sample, calculating a distance between the sample and other samples, obtaining k nearest neighbors of the sample according to the calculated distance, generating a fitting sample corresponding to the sample according to the sample and the k nearest neighbors, and adding the fitting sample to the sample data set to obtain the training sample data set.

4. The method of predicting the efficiency of a screen according to claim 3, wherein, The generation of the fitting sample corresponding to the sample according to the sample and the k nearest neighbors comprises: where SampleNew i is the newly generated fitted sample corresponding to the ith sample, Sample i is the ith sample, Sample j is the jth neighboring sample corresponding to the ith sample, δ is a random vector, and j = 1, 2, 3,..., k.

5. A method of predicting the efficiency of a sieve according to claim 2 or 4, characterised in that, performing an empirical mode decomposition (EMD) algorithm on the obtained vibration signal to generate IMF components; calculating a Pearson correlation coefficient (PCC) between each IMF component and the vibration signal corresponding to the IMF component, and selecting components with a Pearson correlation coefficient greater than a first preset threshold in the IMF components to form an IMF component set; and ​ The IMFs in the IMF component set are sorted in descending order according to Pearson correlation coefficients PCC, and the IMFs with higher ranking are assigned with larger weights, and the IMFs with lower ranking are assigned with smaller weights, the sum of the weights corresponding to the IMFs is 1, and the weighted sum of each IMF and the weight corresponding to the IMF is taken as a first feature component.

6. The method of predicting the efficiency of a screen according to claim 5, wherein, The number or type of parameters included in the sample data of each first training sample subset is different, and the parameters of the sample data in each first training sample subset include a second feature component, material granularity and material humidity.

7. A method of predicting the efficiency of a sieve according to claim 6, characterised in that, The screening efficiency prediction model is a neural network model, and a loss function of the neural network model is: wherein, is the root mean square error, a smoothing loss, a and β are constants obtained based on a trial and error method; wherein the smoothing loss is obtained by the following way: where y i is the output of the screening efficiency of the i-th iteration, y i-1 is the output of the screening efficiency of the i-1-th iteration, y i-2 is the output of the screening efficiency of the i-2-th iteration, μ is a smoothing control coefficient, 0 < μ < 1, i is the iteration number, i = 3, 4, 5, …, M.

8. The method of predicting the efficiency of a screen according to claim 7, wherein, The method further includes: after obtaining the target screening efficiency prediction model, performing a pruning operation on the target screening efficiency prediction model.

9. The method of predicting the efficiency of a screen according to claim 8, wherein, The pruning operation includes: Importance scores of each hidden layer in the current target screening efficiency prediction model are calculated, for each hidden layer, a weight threshold of each neuron is calculated based on the importance score and a custom weight, 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.

10. The method of predicting the efficiency of a screen according to claim 9, wherein, The threshold of each neuron is evaluated based on the importance score and the custom weight, and the threshold of each neuron is evaluated based on the importance score and the custom weight. In the formula, w represents the custom weight, and s represents the importance score. w ij wi 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.

Images