A grinding granularity soft measurement method, system, medium, device and terminal

Abstract

The present application belongs to the technical field of grinding granularity measurement, and discloses a grinding granularity soft measurement method, system, medium, equipment and terminal, which divides the establishment of a soft measurement model into two parts, namely linear and nonlinear, uses a LASSO algorithm to perform linear regression on time series expanded training data, establishes a linear soft measurement model, analyzes time series characteristics according to model weights, selects time series characteristics related to output as inputs of an SCN, uses residuals of real values and LASSO algorithm predicted values as outputs of the SCN and trains the SCN, establishes a nonlinear soft measurement model, synchronously optimizes linear soft measurement model weights of the LASSO algorithm and SCN model output weights through an elastic network, obtains an optimal soft measurement model, and realizes soft measurement of grinding granularity by using the optimal soft measurement model. The present application uses an elastic network to improve the prediction accuracy of a soft measurement model while reducing the complexity of the model, so that the storage cost of the model is lower and the application effect is better.

Description

Technical Field

[0001] This invention belongs to the field of grinding particle size measurement technology, and particularly relates to a soft measurement method, system, medium, equipment and terminal for grinding particle size. Background Technology

[0002] Currently, the existing copper mining process consists of three main stages: grinding, flotation, and thickening. The quality of the pulp output from the grinding stage directly affects the quality of the subsequent flotation and thickening products. Insufficient grinding will result in large mineral particles not being removed during flotation, leading to raw material waste. Overly fine grinding will cause a decrease in concentrate grade, failing to meet output standards.

[0003] Concentration and particle size are key parameters for calibrating slurry quality. While slurry concentration can be effectively measured using equipment such as concentration vessels, real-time and accurate measurement methods for slurry particle size are lacking. Currently, industrial sites primarily use manual sieving or particle size analyzers to calibrate slurry particle size. Manual sieving requires operators to dry the slurry sample before manually sieving and measuring. During calibration, the degree of drying and the sieving method vary from person to person, leading to inaccurate calibration results. Furthermore, particle size analyzers are relatively expensive, and due to process limitations in field use, slurry sampling valves cannot be kept open continuously, making real-time and accurate measurement of slurry particle size impossible.

[0004] Existing soft sensing methods estimate key parameters of production processes by constructing mathematical models between easily measurable variables and target variables. Currently, while numerous studies have been conducted on soft sensing methods both domestically and internationally, relatively few studies delve into the causal relationships between parameters during grinding, establish grinding particle size prediction models, or achieve soft sensing of grinding particle size. Data-driven modeling methods, typically represented by artificial neural networks, build soft sensing models by learning the mapping relationship between inputs and outputs in training data. However, applying these networks requires adjustments to the network structure and training parameters. The more complex the network structure, the more parameters need adjustment, and the greater the impact of different parameter configurations on network performance. Soft sensing modeling research primarily focuses on static models, neglecting the dynamic nature of industrial processes. Modeling research based on dynamic neural networks rarely provides reliable and interpretable parameter configuration methods.

[0005] To address the aforementioned issues, in order to design an effective, interpretable, and robust data-driven dynamic modeling scheme for the copper ore grinding process and to achieve modeling and prediction of grinding particle size, it is necessary to design a time-series analysis and soft measurement framework for slurry particle size from the perspective of parameter random configuration and penalized likelihood optimization.

[0006] Based on the above analysis, the problems and shortcomings of the existing technology are as follows:

[0007] (1) In the existing methods for calibrating the particle size of slurry, the degree of drying of the slurry and the operation method during screening vary from person to person when manual screening is used, which leads to inaccurate calibration results.

[0008] (2) The price of using a particle size analyzer to calibrate the particle size of slurry is relatively expensive, and due to process limitations in field use, the slurry sampling valve cannot be kept open, making it impossible to measure the particle size of slurry in real time and accurately.

[0009] (3) At present, the modeling of soft measurement mainly focuses on static models, ignoring the dynamic nature of industrial processes. Modeling techniques based on dynamic neural networks rarely provide reliable and interpretable parameter configuration methods. Summary of the Invention

[0010] To address the problems existing in the prior art, the present invention provides a soft measurement method, system, medium, device and terminal for grinding particle size, and particularly relates to a soft measurement method, system, medium, device and terminal for grinding particle size based on a random configuration network.

[0011] This invention is implemented as follows: a soft measurement method for grinding particle size, comprising: dividing the establishment of a soft measurement model into linear and nonlinear parts; using the Least Absolute Shrinkage and Selection Operator (LASSO) algorithm to perform linear regression on the time-series unfolded training data to establish a linear soft measurement model; analyzing time-series features based on model weights, selecting time-series features related to the output as inputs to a stochastic configuration network (SCN); using the residual between the true values ​​and the predicted values ​​of the LASSO algorithm as the output of the SCN to train the SCN to establish a nonlinear soft measurement model; simultaneously optimizing the linear soft measurement model weights of the LASSO algorithm and the model output weights of the SCN through an elastic network to obtain the optimal soft measurement model; and finally using the optimal soft measurement model to realize the soft measurement of grinding particle size.

[0012] Furthermore, the soft measurement method for grinding particle size includes the following steps:

[0013] Step 1: Define the training data X and Y, expand the time series, and use cross-validation to determine λ. LASSO ;

[0014] Step 2: Solve for the LASSO model weights and calculate the LASSO model output;

[0015] Step 3, select SCN input data X SCN And calculate the output data Y SCN Establish an SCN model;

[0016] Step 4: Generate candidate weights and biases; define a supervision mechanism to determine the weights and biases of new nodes and update the output weights of the hidden layer.

[0017] Step 5: Combine the linear and non - linear models to construct a soft - sensing model; use cross - validation to determine λ E1 , λ E2 , and optimize the elastic - net weights to obtain the optimal soft - sensing model.

[0018] Furthermore, the time - series expansion in Step 1 includes:

[0019] Based on the original data X(t)=[x1(t), x2(t),..., x c (t)] and Y(t) containing N samples, c input features, and one output feature, each feature in the input data is expanded into a time series of length τ + 1, X(t)=[x1(t), x1(t - 1),..., x1(t - τ),..., x c (t), x c (t - 1),..., x c (t - τ)].

[0020] Furthermore, in Step 2, use the LASSO algorithm to solve the linear model. Let:

[0021]

[0022] Then the linear part of the soft - sensing model

[0023] Furthermore, in Step 3, generate SCN training data. Take out the features corresponding to the non - zero elements in β LASSO from the original input data and combine them as the input data X SCN of the SCN. Take the residual between the true value and the predicted value of the LASSO algorithm as the output data

[0024] The output of the SCN where w H is the input weight of each hidden node, b H is the bias of each hidden node, σ(·) is the activation function of the hidden node, and β SCN is the output weight of each hidden node; before the training starts, given 0 < r < 1 and a non - negative monotonically increasing sequence {λ1, λ2,..., λ m} as the training parameters of the SCN, and the network is trained by randomly generating new hidden nodes each time.

[0025] Furthermore, in Step 4, by traversing the sequence {λ1, λ2,..., λ mGenerate m sets of weights and biases, each set containing N. c For [-λ j , λ j The hidden node weights w are randomly generated within the range j = 1, 2, ..., m. a and bias b a .

[0026] Determine the weights and biases of the new nodes, and define the supervision mechanism:

[0027]

[0028] In the formula, e L-1 (X SCN ) represents the prediction residual of SCN with L-1 hidden nodes, and the current process is the addition of the Lth hidden node; u L = (1-r) / (L+1). Choose ξ a >0 and maximize ξ a w a and b a The weights and biases of the new node are used as inputs, and the new node is added to the hidden layer of the SCN, updating w. H and b H .

[0029] Update the hidden layer output weights, let:

[0030] β SCN =Y SCN pinv(σ(w H X SCN +b H ));

[0031] In the formula, pinv(·) is the Moore-Penrose pseudoinverse.

[0032] If the network prediction error is less than the tolerance error or the number of hidden nodes in the network reaches the set maximum value, then training stops; otherwise, it jumps to the candidate weight and bias generation step to continue adding new hidden nodes.

[0033] Furthermore, in step five, the soft measurement model outputs... but:

[0034]

[0035] The soft sensor model is optimized using an elastic network. Optimization vectors c1 and c2 are defined, with lengths equal to or less than β. LASSO and β SCN Same, let:

[0036]

[0037] In the formula, Let the Hadamard product be represented. Then the optimized soft measurement model is:

[0038]

[0039] Another object of the present invention is to provide a grinding particle size soft measurement system applying the aforementioned grinding particle size soft measurement method, the grinding particle size soft measurement system comprising:

[0040] The time series expansion module is used to set the training data X and Y, expand the time series, and determine λ through cross-validation. LASSO ;

[0041] The model weight calculation module is used to solve for the LASSO model weights and calculate the LASSO model output;

[0042] The SCN model building module is used to select SCN input and output data and build an SCN model.

[0043] The node weight update module is used to generate candidate weights and biases; it defines a supervision mechanism to determine new node weights and biases and update the hidden layer output weights.

[0044] The network weight optimization module is used to combine linear and nonlinear models to construct a soft sensor model; cross-validation determines λ. E1 ,λ E2 Optimize the weights of the elastic network to obtain the optimal soft measurement model.

[0045] Another object of the present invention is to provide a computer device including a memory and a processor, the memory storing a computer program, which, when executed by the processor, causes the processor to perform the steps of the grinding particle size soft measurement method.

[0046] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the described grinding particle size soft measurement method.

[0047] Another object of the present invention is to provide an information data processing terminal for implementing the aforementioned grinding particle size soft measurement system.

[0048] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0049] First, the soft measurement method for copper ore grinding particle size based on stochastic configuration network (SCN) provided by this invention divides the establishment of the soft measurement model into two parts: linear and nonlinear. First, the Least Absolute Shrinkage and Selection Operator (LASSO) algorithm is used to perform linear regression on the time-series unfolded training data, and the time series is analyzed according to the model weights to select time-series features related to the output as inputs to the SCN. Then, the residual between the true value and the predicted value of the LASSO algorithm is used as the output of the SCN to train the SCN. Finally, the model weights of the LASSO algorithm and the output weights of the SCN are optimized through an elastic network, which further improves the prediction accuracy of the soft measurement model.

[0050] This invention uses the LASSO algorithm to establish a linear model and analyze time-series characteristics, and the SCN algorithm to establish a nonlinear model. An elastic network is used to simultaneously optimize the linear and nonlinear models. A time-series expansion method is employed to process the original data, enabling the established soft-sensor model to fit the time-series characteristics of industrial processes. Based on the modeling results of the LASSO algorithm, this invention analyzes the time-series characteristics in the training data, removing invalid time-series information from the original data and reducing the computational cost of SCN modeling. The application of SCN ensures the model's approximation capability and also eliminates the influence of training parameter settings on the modeling effect.

[0051] Second, this invention utilizes elastic networks for global optimization of linear and nonlinear models, which improves the prediction accuracy of soft measurement models while reducing model complexity, resulting in lower model storage costs and better application effects. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart of the grinding particle size soft measurement method provided in the embodiments of the present invention;

[0054] Figure 2 This is a schematic diagram of the grinding particle size soft measurement method provided in the embodiments of the present invention;

[0055] Figure 3 This is a schematic diagram of the random configuration network training process provided in an embodiment of the present invention;

[0056] Figure 4This is a structural diagram of a soft measurement model based on a random configuration network provided in an embodiment of the present invention;

[0057] Figure 5 This is a schematic diagram of the soft measurement modeling process based on a random configuration network provided in an embodiment of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0059] To address the problems existing in the prior art, the present invention provides a soft measurement method, system, medium, equipment and terminal for grinding particle size. The present invention will be described in detail below with reference to the accompanying drawings.

[0060] like Figure 1 As shown, the grinding particle size soft measurement method provided in this embodiment of the invention includes the following steps:

[0061] S101, Set up training data X and Y and expand the time series, then use cross-validation to determine λ. LASSO ;

[0062] S102, Solve for the LASSO model weights and calculate the LASSO model output;

[0063] S103, Select SCN input data X SCN And calculate the output data Y SCN Establish an SCN model;

[0064] S104, Generate candidate weights and biases; Define a supervision mechanism to determine the weights and biases of new nodes and update the output weights of the hidden layer;

[0065] S105, Combine linear and nonlinear models to construct a soft-sensor model; determine λ through cross-validation. E1 ,λ E2 Optimize the weights of the elastic network to obtain the optimal soft measurement model.

[0066] As a preferred embodiment, such as Figure 2 As shown, the grinding particle size soft measurement method provided in this embodiment of the invention specifically includes the following steps:

[0067] Step 1, Time Series Unfolding: Given the original data X(t) = [x1(t), x2(t), ..., x...] containing N samples, c input features, and one output feature... c(t)] and Y(t), each feature in the input data is expanded into a time series of length τ + 1, that is, X(t) = [x1(t), x1(t - 1),..., x1(t - τ),..., x c (t), x c (t - 1),..., x c (t - τ)].

[0068] Step 2, Solve the linear model using the LASSO algorithm. Let:

[0069]

[0070] Then the linear part of the soft sensor model

[0071] Step 3, Generate SCN training data: Extract the features corresponding to the non-zero elements in β LASSO from the original input data and combine them into the input data X SCN of the SCN. The residual between the true value and the predicted value of the LASSO algorithm is used as the output data of the SCN, that is

[0072] Step 4, The output of the SCN where w H is the input weight of each hidden node, b H is the bias of each hidden node, σ(·) is the activation function of the hidden node, and β SCN is the output weight of each hidden node. This network is trained by randomly generating a new hidden node each time. Before the training starts, 0 < r < 1 and a non-negative monotonically increasing sequence {λ1, λ2,..., λ m} need to be given as the training parameters of the SCN. Among them, the training process of the random configuration network is as Figure 3 shown.

[0073] Step 5, Generate candidate weights and biases: Generate m groups of weights and biases by traversing the sequence {λ1, λ2,…, λ m}. Each group contains N c pairs of hidden node weights w j and biases b j randomly generated within the range of [-λ a and λ a .

[0074] Step 6, Determine the new node weights and biases, and define the supervision mechanism:

[0075]

[0076] In the formula, eL-1 (X SCN ) represents the prediction residual of SCN with L-1 hidden nodes (currently, the process of adding the Lth hidden node), u L = (1-r) / (L+1). Choose the option that makes ξ = (1-r) / (L+1). a >0 and maximize ξ a w a and b a The weights and biases of the new node are used as inputs, and the new node is added to the hidden layer of the SCN, updating w. H and b H .

[0077] Step 7, update the hidden layer output weights, let:

[0078] β SCN =Y SCN pinv(σ(w H X SCN +b H (3)

[0079] In the formula, pinv(·) is the Moore-Penrose pseudoinverse. If the network prediction error is less than the tolerance error or the number of hidden nodes reaches the set maximum value, training stops. Otherwise, it jumps to step 5 to add new hidden nodes.

[0080] Step 8, Build the soft measurement model: Output the soft measurement model... Right now:

[0081]

[0082] Among them, the soft measurement model structure based on random configuration networks is as follows: Figure 4 As shown.

[0083] Step 9, optimize the soft sensor model using an elastic network: set optimization vectors c1 and c2, with lengths equal to β respectively. LASSO and β SCN Same, let:

[0084]

[0085] In the formula, Let the Hadamard product be represented. Then the optimized soft measurement model is:

[0086]

[0087] Among them, the soft measurement modeling process based on random configuration networks is as follows: Figure 5 As shown.

[0088] The grinding particle size soft measurement system provided in this embodiment of the invention includes:

[0089] The time series expansion module is used to set the training data X and Y, expand the time series, and determine λ through cross-validation. LASSO ;

[0090] The model weight calculation module is used to solve for the LASSO model weights and calculate the LASSO model output;

[0091] The SCN model building module is used to select SCN input and output data and build an SCN model.

[0092] The node weight update module is used to generate candidate weights and biases; it defines a supervision mechanism to determine new node weights and biases and update the hidden layer output weights.

[0093] The network weight optimization module is used to combine linear and nonlinear models to construct a soft sensor model; cross-validation determines λ. E1 ,λ E2 Optimize the weights of the elastic network to obtain the optimal soft measurement model.

[0094] Current grinding processes primarily utilize semi-autogenous mills and ball mills to crush mineral raw materials, followed by hydrocyclones to select appropriately sized mineral particles as output. The entire process involves numerous time-delay stages. The mineral raw materials supplied by the feeding equipment are first fed into the semi-autogenous mill via a belt conveyor. After preliminary crushing, they enter the slurry pump tank and are then pumped into the hydrocyclone. Lighter particles in the hydrocyclone are carried by the water flow from the top to the flotation section, while heavier particles are discharged from the bottom to the ball mill for further crushing before re-entering the slurry pump tank and hydrocyclone. The conveyor transport, mill crushing, and hydrocyclone screening all involve significant time delays, and the complex production process results in a strong nonlinear relationship between various parameters and slurry particle size. Although particle size analyzers can quickly detect slurry particle size, the sampler cannot be continuously operated to prevent insufficient slurry capacity, as the tested slurry needs to be stored in a recovery tank. Therefore, real-time particle size measurement is not possible.

[0095] In the specific implementation of this invention, it is first necessary to obtain the approximate lag time of the impact of changes in easily measurable sensor input variables on slurry particle size based on field process experience. Assuming the number of easily measurable sensor input variables is 10, the longest lag time affecting slurry particle size is approximately 49, i.e., c = 10, τ = 49. According to the technical process, the implementation steps of this invention are as follows:

[0096] Step 1, Time Series Expansion: Given the original input data X(t) = [x1(t), x2(t), ..., x...] containing 10 field variables... 10Based on the grinding particle size Y(t), each feature in the input data is expanded into a time series of length 50, i.e., X(t) = [x1(t), x1(t-1), ..., x1(t-49), ..., x1(t-49), ..., x1(t-49)]. 10 (t), x 10 (t-1), ..., x 10 (t-49)].

[0097] Step 2, LASSO algorithm to solve the linear model: First, determine λ using cross-validation. LASSO The optimal value.

[0098] Step 3, let λ LASSO =0.

[0099] Step 4: Let i = 1, divide the training data X and Y into 5 equal parts according to time, and select the i-th sample as the validation set X. V and Y V The remaining samples are used as the training set X. T and Y T ,make:

[0100]

[0101] Calculate the validation set error of the soft measurement model

[0102] Step 5, if i = 5, then let Otherwise, let i = i + 1 and jump back to step S22.

[0103] Step 6, let λ LASSO =λ LASSO +0.01, until ||β LASSO ||1=0.

[0104] Step 7, select the λ that minimizes E. LASSO ,make:

[0105]

[0106] The linear part of the soft measurement model

[0107] Step 8, Generate SCN training data: β LASSO The features corresponding to the non-zero elements in the original input data are extracted and combined to form the input data X of the SCN. SCN The residual between the actual value and the LASSO algorithm's predicted value is used as the output data of SCN, i.e.

[0108] Step 9, SCN output Where wH is the input weight of each hidden node, b H is the bias of each hidden node, σ(·) is the activation function of the hidden node, β SCN is the output weight of each hidden node. The network is trained by randomly generating a new hidden node each time. Before the start of training, 0 < r < 1 and a non-negative monotonically increasing sequence {λ1, λ2,..., λ m} need to be given as the training parameters of the SCN.

[0109] Step 10, generation of candidate weights and biases: By traversing the sequence {λ1, λ2,..., λ m}, m groups of weights and biases are generated, and each group contains N c For the hidden node weights w j randomly generated from [-λ j , λ a , j = 1, 2,..., m, and biases b a .

[0110] Step 11, determination of new node weights and biases, definition of supervision mechanism:

[0111]

[0112] In the formula, e L-1 (X SCN ) is the prediction residual of the SCN when there are L - 1 hidden nodes (currently the addition process of the Lth hidden node), u L = (1 - r) / (L + 1). Select w a and b a that can make ξ a > 0 and maximize ξ a as the weights and biases of the new node, and add the new node to the hidden layer of the SCN, and update w H and b H .

[0113] Step 12, update of the hidden layer output weights, let:

[0114] β SCN = Y SCN pinv(σ(w H X SCN + b H )) (10)

[0115] In the formula, pinv(·) is the Moore - Penrose pseudoinverse. At this time, if the network prediction error is less than the tolerance error or the number of network hidden nodes reaches the set maximum value, the training stops. Otherwise, it jumps to step S42 to continue adding new hidden nodes.

[0116] Step 13, Construct the soft measurement model: Output the soft measurement model... Right now:

[0117]

[0118] Step 14, optimize the soft sensor model using an elastic network: set optimization vectors c1 and c2, with lengths equal to β respectively. LASSO and β SCB Same, let:

[0119]

[0120] In the formula, Denotes the Hadamard product, λ E1 and λ E2 Optimization and λ LASSO If the optimization method is consistent, then the optimized soft measurement model is:

[0121]

[0122] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0123] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A soft measurement method for grinding particle size, characterized in that, include: Linear regression was performed on the time-series unfolded training data using the LASSO algorithm to establish a linear soft measurement model. Time series features were analyzed based on the model weights, and time series features related to the output were selected as inputs to the SCN. The residual between the true value and the predicted value of the LASSO algorithm was used as the output of the SCN and the SCN was trained to establish a nonlinear soft measurement model. The weights of the linear soft measurement model of the LASSO algorithm and the output weights of the SCN model were simultaneously optimized through an elastic network to obtain the optimal soft measurement model. Finally, the optimal soft measurement model was used to realize the soft measurement of grinding particle size. The soft measurement method for grinding particle size includes the following steps: Step 1: Define the training data X and Y, expand the time series, and use cross-validation to determine λ. LASSO ; Step 2: Solve for the LASSO model weights and calculate the LASSO model output; Step 3, select SCN input data X SCN And calculate the output data Y SCN Establish an SCN model; Step 4: Generate candidate weights and biases; define a supervision mechanism to determine the weights and biases of new nodes and update the output weights of the hidden layer; Step 5: Combine linear and nonlinear models to construct a soft measurement model; perform cross-validation to determine λ. E1 , λ E2 Optimize the weights of the elastic network to obtain the optimal soft measurement model; The time series expansion in step one includes: In containing One sample The original data of one input feature and one output feature and Based on this, each feature in the input data is expanded to a length of time series ; In step two, the LASSO algorithm is used to solve the linear model, let: ； The linear part of the soft measurement model ; In step three, SCN training data is generated. The features corresponding to the non-zero elements are extracted from the original input data and combined to form the input data for SCN. The residual between the actual value and the LASSO algorithm prediction is used as the output data of SCN. ; SCN output ,in The input weights for each hidden node are... This represents the bias of each hidden node. is the activation function for hidden nodes. The output weights of each hidden node are given before training begins. and a non-negative monotonically increasing sequence As training parameters for SCN, the network is trained by randomly generating new hidden nodes each time; In step four, the sequence is traversed. generate Group weights and biases, each group containing From , Hidden node weights randomly generated within the range and bias ; Determine the weights and biases of the new nodes, and define the supervision mechanism: ； In the formula, For SCN in having The prediction residual at the nth hidden node, currently at the th... The process of adding hidden nodes; ; Choose to make And maximize of and Assigning weights and biases to new nodes, and adding the new nodes to the hidden layer of the SCN, then updating... and ; Update the hidden layer output weights, let: ； In the formula, This is a Moore-Penrose pseudo-inverse; If the network prediction error is less than the tolerance error or the number of hidden nodes in the network reaches the set maximum value, then training stops; otherwise, it jumps to the candidate weight and bias generation step to continue adding new hidden nodes. In step five, the soft measurement model outputs... ,but: ； Optimize the soft sensor model using an elastic network and define the optimization vector. and The lengths are respectively with and Same, let: ； In the formula, Let the Hadamard product be represented. Then the optimized soft measurement model is: 。 2. A grinding particle size soft measurement system applying the grinding particle size soft measurement method as described in claim 1, characterized in that, The grinding particle size soft measurement system includes: The time series expansion module is used to set the training data X and Y, expand the time series, and determine λ through cross-validation. LASSO ; The model weight calculation module is used to solve for the LASSO model weights and calculate the LASSO model output; The SCN model building module is used to select SCN input and output data and build an SCN model. The node weight update module is used to generate candidate weights and biases; it defines a supervision mechanism to determine new node weights and biases and update the hidden layer output weights. The network weight optimization module is used to combine linear and nonlinear models to construct a soft sensor model; cross-validation determines λ. E1 ,λ E2 Optimize the weights of the elastic network to obtain the optimal soft measurement model.

3. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the grinding particle size soft measurement method as described in claim 1.

4. A computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the grinding particle size soft measurement method as described in claim 1.

5. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the grinding particle size soft measurement system as described in claim 2.

Images