Search

CN-122020383-A - Ore grinding granularity robust soft measurement method based on double Gaussian mixture distribution

CN122020383ACN 122020383 ACN122020383 ACN 122020383ACN-122020383-A

Abstract

The invention discloses a grinding granularity robust soft measurement method based on double Gaussian mixture distribution, which comprises the steps of constructing a random configuration network as a basic prediction model, then explicitly establishing a noise model formed by weighted mixing of small-variance Gaussian components and large-variance Gaussian components, respectively fitting conventional measurement noise and process abnormal noise, and adopting a desired maximization algorithm to jointly iterate and optimize network output weight and noise model parameters under a Bayesian framework to realize self-adaptive learning of parameters and intelligent weighting of samples. The method can output noise decomposition parameters including normal working condition weights and abnormal variance ratios, and provides interpretable decision support for ore grinding operation. Experiments show that the prediction accuracy, robustness and interpretability of the model in a mixed noise environment are effectively improved, and reliable technical support is provided for optimal control of the ore grinding process.

Inventors

  • LIU XIN
  • LI QIQI
  • DAI WEI
  • YAN PEI
  • LIU XIAOQING
  • NAN JING

Assignees

  • 中国矿业大学

Dates

Publication Date
20260512
Application Date
20260130

Claims (10)

  1. 1. The grinding particle size robust soft measurement method based on double Gaussian mixture distribution is characterized by comprising the following steps of: step 1, collecting historical data of a grinding process, and constructing a training data set Wherein, the vector is input Respectively represent the ore feeding amount Water supply to inlet of mill Overflow concentration of classification equipment Output scalar Represents the grain size of the ground ore, Is the total data volume in the dataset; Step 2, based on the training data set, constructing a random configuration network SCN as a basic prediction model, wherein the random configuration network sequentially adds hidden nodes through an incremental supervision mechanism, and the first node of the random configuration network SCN is a first node of the random configuration network The network output of the individual nodes is expressed as: ; Wherein, the Is provided with The random configuration network of hidden nodes predicts the output, In order to input the vector(s), In order to activate the function, Respectively the first The input weights, bias and output weights of the hidden nodes, Configuring the output weights of the network for random, with Output matrix of hidden layers of individual hidden nodes The definition is as follows: ; If it has If the random configuration network of hidden nodes does not meet the termination condition, a new hidden node is generated, the hidden layer outputs Expressed as: ; step 3, calculating the prediction error of the random configuration network as Wherein, the method comprises the steps of, Establishing a double Gaussian mixed noise model for describing the prediction error of the ore grinding process for the prediction error of the random configuration network, wherein the probability density function is as follows: ; Wherein, the Representation of Obeying mean value of Variance is Is used for the distribution of the gaussian distribution of (c), Is the mixed weight of the noise under normal working conditions, For the normal noise variance to be the same, Is the amplification factor of the variance of abnormal noise relative to normal noise, and ; Step 4, under the Bayesian framework, the output weight of the network is configured randomly Introducing Gaussian prior distribution; ; Wherein, the Is a matrix of units which is a matrix of units, Outputting weights for a network Is a function of the variance of (a), Representing output weights Is a covariance matrix of (a); step 5, adopting a expectation maximization algorithm to jointly estimate parameter sets Optimum parameters of (a) For new grinding working condition input The grinding granularity predicted value is 。
  2. 2. The soft measurement method of grinding particle size robustness based on double-gaussian mixture distribution according to claim 1, wherein in the step 3, the process of constructing the double-gaussian mixture noise model is as follows: Introducing binary implicit variables Satisfies the following conditions And is also provided with Defining its a priori distribution: , At a given point Under the condition of (1) observation data The conditional probability of (2) obeys a gaussian distribution when , Obeying mean value of Variance is When the Gaussian distribution of (1) , Obeying mean value of Variance is Gaussian distribution of (c), namely: , ; By implying variable Marginalizing to obtain observed data The edge probability distribution of (2) is the double Gaussian mixture model: ; taking complete data sets Logarithm of posterior distribution, calculating logarithmic posterior probability of complete data The conditions of (2) are as follows: ; Wherein, the Is a constant value, and is used for the treatment of the skin, Represent the first Parameter set for multiple iterations, superscript Represent the first And iterating for a plurality of times.
  3. 3. The method for soft measurement of grinding grain size robustness based on double gaussian mixture distribution according to claim 2, wherein in step 5, the iterative optimization step of the expectation maximization algorithm is as follows: Step 4-1, parameter set to be estimated An initialization is performed such that the data of the data storage device, , =0; Step 4-2, estimating based on the current parameters Calculating implicit variables Characterization of the first embodiment The samples belong to Posterior expectation of individual gaussian components , ; Step 4-3, updating parameters by maximizing the expectation of the complete data log posterior, wherein the updating of the output weight is realized by solving a weighted least squares problem: In which, in the process, A diagonal weight matrix assigned to each training sample based on the current noise model; step 4-4, iteratively executing the step 4-2 and the step 4-3 until the following parameter convergence conditions are met, thereby obtaining the optimal parameters After model training, optimal parameters Robust prediction for mill granularity; ; Wherein, the Is a preset positive number.
  4. 4. The method for soft measurement of grinding grain size robustness based on double Gaussian mixture distribution according to claim 3, wherein in the step 4-2, an implicit variable is calculated Posterior expectation of (2) The formula of (2) is as follows: ; ; Wherein, the , Represent the first The error of the individual samples results from the probability of a normal noise component, Represent the first The error of the individual samples results from the probability of an abnormal noise component.
  5. 5. The method for soft measurement of fine grain size based on double Gaussian mixture distribution according to claim 4, wherein in the step 4-3, the first step is Sub-update noise model parameters By maximizing the formula of (1) for expectation And letting the partial derivative be zero gives: ; ; 。
  6. 6. The soft measurement method of grinding particle size robustness based on double Gaussian mixture distribution according to claim 5, wherein in the step 4-3, the weight matrix is Is the nth diagonal element of (2) The calculation formula of (2) is as follows: ; the weight is used to adaptively reduce the effect on the model of samples identified as anomalies when updating the network weights And the sample corresponding to the threshold value is larger than the preset threshold value.
  7. 7. The soft measurement method of grinding grain size robustness based on double Gaussian mixture distribution according to claim 4, wherein in the step 4-3, the prior variance is updated By maximizing the expectation Middle AND Related items And let the partial derivative be zero gives: 。
  8. 8. the method for soft measurement of grinding granularity robustness based on double Gaussian mixture distribution as set forth in claim 1, wherein in said step 2, an incremental supervision mechanism is used to add hidden nodes when constructing a random configuration network When the hidden nodes are involved, multiple groups of candidate parameters are randomly generated from a preset interval Parameters (parameters) The following inequality constraint is satisfied: ; Wherein, the And In order to be able to set a predetermined amount of the quantity, , , , The representation has The random configuration network SCN residual vector of each node stops generating new nodes until the maximum hidden node number is reached or the residual meets the termination condition.
  9. 9. A dual gaussian mixture noise robust neural network modeling system for an ore grinding process, characterized by comprising a processor and a memory, wherein the memory stores a computer program, and the processor implements the steps of the ore grinding granularity robust soft measurement method based on dual gaussian mixture distribution according to any one of claims 1-8 when executing the program.
  10. 10. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method for robust soft measurement of grinding particle size based on a dual gaussian mixture distribution according to any of claims 1 to 8.

Description

Ore grinding granularity robust soft measurement method based on double Gaussian mixture distribution Technical Field The invention belongs to the technical field of mineral processing intellectualization, and particularly relates to a grinding particle size robust soft measurement method based on double Gaussian mixture distribution. Background Grinding is a key link in the mineral processing process, and the grinding granularity, namely the particle ratio of the particle size smaller than 0.074mm, directly influences the subsequent separation efficiency, concentrate grade and metal recovery rate. Accurate and real-time prediction of ore grinding granularity is important for stable production, energy consumption reduction and economic benefit improvement. However, in actual production, the mechanism of the ore grinding process is complex, and the characteristics of strong nonlinearity, multivariable coupling, large time lag and the like are provided, so that accurate modeling based on a physical mechanism is extremely difficult. Data-driven soft measurement techniques provide an effective approach for this, where random configuration networks, by virtue of their randomized learning, incremental construction, and pervasive approximation capabilities, present significant advantages in complex industrial process modeling. However, in the specific scene of ore grinding, the data quality is severely challenged by the fact that on one hand, the sensor measurement noise, the environmental electromagnetic interference and the like cause that the data contains Gaussian background noise, and on the other hand, the ore hardness mutation, equipment sudden faults (such as falling of a lining plate and damage of a steel ball), valve clamping stagnation and even artificial misoperation can introduce outliers with large amplitude and abnormal distribution to form mixed noise interference of coexistence of impulse noise and heavy tail noise. The traditional random configuration network relies on a least square criterion to perform parameter estimation, is highly sensitive to abnormal values, and seriously impairs model precision and generalization capability. The existing robust improvement method, such as a model based on single Laplace distribution or kernel density estimation, can inhibit the influence of abnormal values to a certain extent, but has obvious limitations that the noise is generally assumed to be subjected to single heavy tail distribution, conventional measurement noise with different properties and sudden abnormal interference in the grinding process cannot be distinguished and quantified explicitly, the model has insufficient interpretation, the process state is difficult to be clearly revealed to process personnel, and meanwhile, part of the method has high calculation complexity, so that the requirements of real-time monitoring and quick response of the grinding process are difficult to be met. Therefore, developing a soft measurement method which can precisely match the mixed noise characteristics of the grinding process, has excellent robustness, good interpretability and high-efficiency computing capability, and becomes a technical problem to be solved in order to realize the fine and intelligent operation of the grinding process. Disclosure of Invention The invention aims at providing a grinding particle size robust soft measurement method based on double Gaussian mixture distribution aiming at the grinding process data characteristics, and the method is characterized in that a double-component Gaussian mixture model is constructed, normal measurement noise and abnormal process noise in the grinding process are explicitly separated and quantized, on the basis, the combination Bayesian learning of network parameters and noise parameters is realized by adopting an Expectation Maximization (EM) algorithm, and reliability weights are distributed for each training sample in a self-adaptive mode, so that the robustness and the interpretability of the model to abnormal working conditions are greatly improved while the prediction precision is ensured. In order to solve the technical problems and achieve the aim of the invention, the invention provides a grinding particle size robust soft measurement method based on double Gaussian mixture distribution, which comprises the following steps: step 1, collecting historical data of a grinding process, and constructing a training data set Wherein, the vector is inputRespectively represent the ore feeding amountWater supply to inlet of millOverflow concentration of classification equipmentOutput scalarRepresents the grain size of the ground ore,Is the total data volume in the dataset; step 2, based on the training data set, constructing a random configuration network as a basic prediction model, wherein the random configuration network sequentially adds hidden nodes through an incremental supervision mechanism, and the first node of the random configuration netwo