A heavy medium separation dynamic simulation method and system

Abstract

The present application belongs to the technical field of coal washing and sorting modeling and intelligent control, and particularly relates to a dense medium separation dynamic simulation method and system. The method covers the dynamic mechanism model of the whole process of dense medium separation, and on the fast time scale, the key equipment parameters in the density control loop that cannot be directly measured are estimated online. In combination with the Monte Carlo data mining algorithm, the error between the simulated value and the actual value of the medium density is used to generate the tutor signal to supervise the learning and online correction of the online estimation. On the slow time scale, based on the error between the simulated value and the actual sampling value of the clean coal ash, the Bayesian optimization algorithm is used to optimize and update the key separation parameters of the cyclone separation process. The present application effectively solves the problems of insufficient dynamic tracking precision and adaptive ability of the traditional simulation model by using the hybrid intelligent driving and double time scale parameter updating mechanism, and provides a high-precision simulation platform and technical support for the intelligent control and operation optimization of the dense medium separation process.

Description

Technical Field

[0001] This invention belongs to the field of coal washing and beneficiation modeling and intelligent control technology, specifically involving a dynamic simulation method and system for heavy medium separation. Background Technology

[0002] Heavy media separation is one of the most widely used and precise processes in China's coal washing industry. Its basic principle is to use centrifugal force in a suspension with controllable density to achieve precise separation of coal and gangue, which is of great significance for improving the quality of clean coal, resource utilization, and reducing environmental pollution. With the deepening of the intelligent transformation of coal preparation plants, building a high-fidelity, full-process dynamic simulation system has become a key foundation for optimizing process parameters, verifying advanced control algorithms, and constructing a digital twin platform.

[0003] However, heavy media separation is a complex industrial process involving multiple technological units, including raw coal desliming, media mixing, cyclone separation, media recovery, and density control. The entire separation process exhibits strong nonlinearity and strong coupling, and the system's internal parameters possess significant multi-timescale characteristics. While parameters related to the density control loop, such as the desliming coefficients of each device, are affected by flow rate and pressure and can be obtained through synchronous instrument measurement, indicators reflecting separation quality, such as the ash content of clean coal, are limited by manual testing or instrumental analysis, resulting in long sampling cycles. To address these challenges, online simulation analysis has become a necessary means to optimize separation accuracy and ensure product quality.

[0004] Existing simulation methods mainly rely on single mechanism models or data-driven models. These models are limited by sampling time, depend on fixed parameters, and have weak physical interpretability. This results in significant limitations in simulating the entire heavy medium separation process. Their dynamic tracking accuracy and adaptive capability are insufficient, and their fit with the actual process is not high, which seriously limits their engineering guidance value in heavy medium separation applications. Summary of the Invention

[0005] To address the problems existing in the prior art, the purpose of this invention is to provide a dynamic simulation method and system for heavy medium sorting. By adopting a dual-timescale hybrid intelligent driving approach and establishing a parameter update mechanism, it effectively solves the problems of insufficient dynamic tracking accuracy and adaptive capability of traditional simulation models, and provides a high-precision simulation platform and technical support for intelligent control and operation optimization of the heavy medium sorting process.

[0006] The technical solution of this invention is: A dynamic simulation method for heavy medium sorting includes the following steps: S1. Based on the heavy media separation process, construct a dynamic mechanism model covering the entire heavy media separation process. The model includes at least a raw coal desliming screen module, a mixing tank module, a hydrocyclone module, and a density control loop formed by the flow of media from an arc screen module, a desliming screen module, a magnetic separator module, a diverter valve module, a dilute medium tank module, a combined medium tank module, and a combined medium tank water supply module. S2. On a fast timescale, the key equipment parameters that cannot be directly measured in the density control loop are estimated online, and the online estimated key equipment parameters are input into the dynamic mechanism model to obtain the simulated value of the combined density. S3. Based on the error between the simulated value and the actual sampled value of the combined density, construct a first objective function, perform global optimization on the key equipment parameters in the dynamic mechanism model, obtain the optimal combination of key equipment parameters, and feed it back to step S2 as a mentor signal to supervise and correct the online estimation process of the key equipment parameters and obtain the updated key equipment parameters. S4. Input the updated key equipment parameters back into the dynamic mechanism model to obtain the simulated value of clean coal ash content; S5. On a slow time scale, a second objective function is constructed based on the error between the simulated value and the actual sampled value of clean coal ash content. Based on the second objective function, the key separation parameters that affect the separation accuracy in the hydrocyclone module are optimized and updated. S6. Based on the dynamic mechanism model continuously updated and optimized from S2 to S5, the operating status of the heavy media sorting process is simulated in real time, and decision reference information for density control and sorting process optimization is output.

[0007] Preferably, the key equipment parameters include at least: the arc screen desliming coefficient of the arc screen module, the desliming coefficient of the clean coal desliming screen and the gangue desliming screen of the desliming screen module, and the magnetic separator recovery coefficient of the magnetic separator module.

[0008] Preferably, the desliming coefficient of the arc screen is estimated online using a case-based reasoning algorithm; the desliming coefficients of the clean coal desliming screen and the gangue desliming screen are estimated using an adaptive neural fuzzy reasoning system based on a first-order Sugeno function, and updated online using the recursive least squares method; the recovery coefficient of the magnetic separator is estimated using a random vector function linking network.

[0009] Preferably, the desliming coefficient of the arc-shaped screen is determined according to the following formula: , in, , In the formula, The desliming coefficient of the arc-shaped screen. The similarity function is based on the reciprocal of the weighted Euclidean distance.n This refers to the number of highly similar cases ultimately selected after filtering according to a preset threshold. i For this n Among the selected similar cases, the local index number used for loop traversal is... This represents the historical case value of the density of the mixture. This represents historical values ​​for the feed rate to the arc screen. Current operating value of the combined density, This represents the current operating value of the feed rate to the arc screen. and These are the weighting coefficients. This is the scaling factor.

[0010] Preferably, the recovery coefficient of the magnetic separator is determined according to the following formula: , in, , In the formula, The recovery coefficient of the magnetic separator. To output weights, L This represents the total number of hidden layer nodes in the neural network. j It is the local index number of each neuron during the traversal. ReLU For activation function, For the input vector, The input vector is given fixed weights through random generation. The input vector is a randomly generated bias. The opening degree of the water flushing valve for the magnetic separator. This refers to the opening degree of the feed gate. This refers to the feed density of the magnetic separator. The magnetic field strength of the magnetic separator.

[0011] Preferably, the first objective function is determined according to the following formula: , in, , In the formula, Let the first objective function be... This is the simulated value of the combined dielectric density. The sampled value is the density of the mixture. The length of the sample window. Given the defined sequence of parameter vectors to be inverted, The desliming coefficient of the arc-shaped screen. The recovery coefficient of the magnetic separator. The desliming coefficient of the clean coal desliming screen. Desliming coefficient of gangue desliming screen.

[0012] Preferably, the optimal combination of key equipment parameters is determined according to the following formula: , in, , In the formula, This represents the optimal combination of key equipment parameters. This represents the total number of random samples performed within the parameter space. For the first m The parameter combination obtained from the second sampling; The posterior probability weights for the parameter combination. The parameter combination corresponding to the first objective function The value of , This is the attenuation factor.

[0013] Preferably, the second objective function is determined according to the following formula: , in, , In the formula, The second objective function is... This is a simulated value for the ash content of clean coal. This represents the actual sampled value of clean coal ash content. For the defined sequence of parameter vectors to be optimized, This is the overflow / underflow ratio coefficient. The overflow ash content coefficient, is the underflow ash content coefficient.

[0014] Preferably, the decision reference information includes at least the opening setting values ​​of the water supply valve and the diversion valve of the mixing tank for adjusting the density of the qualified medium, and the predicted value of the ash content of the clean coal. The dynamic relationship between the opening degree of the combined tank water supply valve and the density of the qualified medium is determined according to the following formula: , in, For a qualified medium density, Combined water supply volume and flow rate at the bottom of the tank. To set the opening degree of the water supply valve for the combined tank. For valve pressure drop, This represents the ratio of the liquid's density to the density of water. The density of water, The volumetric flow rate of the qualified medium formed after mixing. For effective mixing volume, The density of the suspension in the mixing tank. The outflow medium suspension volume flow rate is given by the combined media tank.

[0015] A dual-timescale hybrid intelligent-driven dynamic simulation system for heavy media sorting, used to implement any of the methods shown, includes: A dual-timescale hybrid intelligent-driven dynamic simulation system for heavy media sorting, used to implement the method of any one of claims 1-8, characterized in that it comprises: The model building unit is used to build a dynamic mechanism model covering the entire heavy medium separation process. The model includes at least a raw coal desliming screen module, a mixing tank module, a hydrocyclone module, and a density control loop formed by the flow of media from an arc screen module, a desliming screen module, a magnetic separator module, a diverter valve module, a dilute medium tank module, a combined medium tank module, and a combined medium tank water supply module. The fast timescale parameter estimation and optimization unit is used to estimate the key equipment parameters of the density control loop online on a fast timescale, and input the online estimated key equipment parameters into the dynamic mechanism model to obtain the simulated value of the combined density. At the same time, based on the error between the simulated value and the actual sampled value of the combined density, a first objective function is constructed to globally optimize the key equipment parameters, obtain the optimal combination of key equipment parameters, and use it as a mentor signal to supervise and correct the online estimation process. The slow timescale parameter optimization unit is used to first input the updated key equipment parameters into the dynamic mechanism model again under a slow timescale to obtain the simulated value of clean coal ash content, and then construct a second objective function based on the error between the simulated value of clean coal ash content and the actual sampled value to optimize and update the key sorting parameters that affect the sorting accuracy in the hydrocyclone module. The simulation and output unit is used to simulate the operation status of the heavy medium sorting process in real time based on the updated dynamic mechanism model, and output decision reference information for density control and sorting process optimization.

[0016] Compared with the prior art, the dynamic simulation method and system for heavy medium sorting of the present invention has the following beneficial effects: This invention first constructs a dynamic mechanism model covering the entire process as the simulation foundation, ensuring the physical interpretability of the model. Then, for several key equipment parameters in the density control loop that cannot be directly measured but change rapidly, a Monte Carlo data mining algorithm is used at a fast timescale to generate a mentor signal based on the error between the simulated and actual values ​​of the combined density. This achieves supervision and online correction of the online estimation process, fundamentally solving the dilemma of decreased accuracy over time due to fixed parameters in a single mechanism model, and the inability of data-driven models to provide online supervised learning when key parameters are missing. Simultaneously... For key separation parameters in the hydrocyclone separation process that are affected by the characteristics of raw coal and change slowly, optimization and updates are performed based on the error between the simulated value and the actual sampled value of clean coal ash content at a slow time scale. This dual-time-scale design accurately matches the multi-scale dynamic characteristics of the heavy medium separation process. Through deep integration of mechanism and data, the model can continuously and adaptively track changes in actual operating conditions, thereby significantly improving the dynamic accuracy of the simulation and its fit with the real process flow. Ultimately, it provides a high-precision and high-reliability simulation platform and decision reference for the intelligent control and operation optimization of the heavy medium separation process. Attached Figure Description

[0017] Figure 1 This is an architecture diagram of the dynamic modeling method for the heavy media sorting process driven by hybrid intelligent technology in an embodiment of the present invention.

[0018] Figure 2 This is a flow chart of the heavy medium separation process in an embodiment of the present invention.

[0019] Figure 3 The opening degree of the combined water supply valve and the diversion valve in the embodiment of the present invention.

[0020] Figure 4 The water supply valve opening degree of each device in the embodiments of the present invention.

[0021] Figure 5 This is a diagram showing the fast timescale parameter estimation results of the density loop in an embodiment of the present invention.

[0022] Figure 6 This is a simulation diagram of the combined medium density compared with the fixed-parameter heavy medium sorting model in an embodiment of the present invention.

[0023] Figure 7 This is a simulation diagram of the liquid level in the combined medium tank compared with the fixed-parameter heavy medium sorting model in an embodiment of the present invention.

[0024] Figure 8 This is a simulation diagram of the dilute medium tank level compared with the fixed-parameter heavy medium sorting model in an embodiment of the present invention.

[0025] Figure 9 This is a diagram showing the fast time-scale parameter estimation results of the sorting loop in an embodiment of the present invention.

[0026] Figure 10 This is a simulation diagram of the content of various components in refined coal in an embodiment of the present invention.

[0027] Figure 11 This is a simulation diagram of clean coal ash content compared with a fixed-parameter heavy medium separation model in an embodiment of the present invention; Figure 12 This is a flowchart of the method in an embodiment of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0029] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0030] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0031] See Figure 1 and Figure 2 As shown, to effectively address the shortcomings of traditional simulation models in terms of dynamic tracking accuracy and adaptive capability, and to provide a high-precision simulation platform and technical support for intelligent control and operational optimization of the heavy media separation process, this embodiment provides a dynamic simulation method and system for heavy media separation. The dual-timescale hybrid intelligent-driven dynamic simulation system for heavy media separation provided in this solution mainly includes a model building unit, a fast timescale parameter estimation and optimization unit, a slow timescale parameter optimization unit, and a simulation and output unit. The model building unit is used to construct a dynamic mechanism model covering the entire heavy media separation process, including: a raw coal desliming screen module, a mixing tank module, a hydrocyclone module, an arc screen module, a desliming screen module, a diverter valve module, a dilute medium tank module, a magnetic separator module, a combined medium tank module, and a combined medium tank water supply module. These modules are connected sequentially according to the heavy media separation process flow, forming a loop system for medium circulation and density control. Among them, the arc screen module, the desliming screen module, the diverter valve module, the dilute medium tank module, the magnetic separator module, the combined medium tank module, and the combined medium tank water replenishment module serve as the main components of the density control loop.

[0032] The fast-timescale parameter estimation and optimization unit is used to estimate the key equipment parameters of the density control loop online at a fast timescale, and input the online estimated key equipment parameters into the dynamic mechanism model to obtain the simulated values ​​of the combined medium density. Simultaneously, based on the error between the simulated and actual sampled values ​​of the combined medium density, a first objective function is constructed to globally optimize the key equipment parameters, obtaining the optimal combination of key equipment parameters, which is then used as a mentor signal for supervised learning and correction of the online estimation process. The slow-timescale parameter optimization unit is used at a slow timescale. First, the updated key equipment parameters are input again into the dynamic mechanism model to obtain the simulated values ​​of the clean coal ash content. Then, based on the error between the simulated and actual sampled values ​​of the clean coal ash content, a second objective function is constructed to optimize and update the key separation parameters affecting the separation accuracy in the hydrocyclone module. The simulation and output unit is used to perform real-time simulation of the operating state of the heavy medium separation process based on the updated dynamic mechanism model, and output decision reference information for density control and separation process optimization.

[0033] Specifically, the raw coal desliming screen module employs a first-principles mechanistic modeling approach, expressing the raw coal feed mass flow rate as a function of the effective feed mass flow rate after desliming, and introducing screening efficiency as a structural parameter to characterize the material mass change characteristics of the system during the raw coal pretreatment stage. The mixing tank module uses a mass conservation-based dynamic modeling approach, constructing a continuous-time state equation for the mixed slurry density, and unifying multiple input sources such as raw coal feed, media recovery, and water replenishment into the system model to describe the internal changes during the mixing process. The hydrocyclone module uses a parametric mechanistic modeling approach, establishing a dynamic model of the separation process under mass conservation constraints. It introduces overflow and underflow proportional coefficients, overflow ash coefficients, and underflow ash coefficients to characterize the nonlinear characteristics of the separation, and uses a slow-time-scale Bayesian optimization method to optimize key parameters in real time. The arc screen module combines a steady-state mechanistic model with a case-study reasoning algorithm, expressing the arc screen desliming coefficient as a function of the system's operating characteristics, and achieving online estimation and adaptive updating of this desliming parameter on a fast time scale through similar operating condition matching and weighted fusion mechanisms. The desliming screen module includes a clean coal desliming screen and a gangue desliming screen. It employs a modeling method that integrates a mechanistic model with an adaptive neuro-fuzzy inference system. The nonlinear mapping relationship between the desliming coefficient and the control input is represented as a first-order Sugeno fuzzy model, and online parameter identification is achieved through a recursive least squares algorithm. The diversion valve module uses a mechanistic modeling method based on the steady-state flow distribution assumption. The diversion ratio is modeled as a function of valve parameters and system state variables, used to describe the coupled distribution characteristics of the medium flow rate among different sub-models. The dilute medium tank module uses a dynamic modeling method based on mass conservation and geometric constraints. A state-space model of the density and level of the dilute medium suspension is established, and the change process of the medium state in the dilute medium loop is described through continuous-time differential equations. The magnetic separator module uses a stochastic vector function linked network modeling method. The magnetic separator recovery coefficient is represented as a nonlinear function of multiple input variables, and real-time updates of model parameters are achieved through an online learning strategy using gradient descent. The mixing tank module employs a dynamic mechanism modeling method based on mass conservation to construct a state equation relating the mixing density and liquid level, which describes the dynamic changes in the system state during the media collection and regulation process. The water replenishment module below the mixing tank uses an unsteady-state mixing modeling method under a fixed effective mixing volume condition to establish a dynamic model of the qualified media density. This model is then used as the controlled object in the density control loop to achieve precise modeling of the media density.

[0034] Through the above module settings, a full-process dynamic mechanism model for typical heavy media sorting processes in China was constructed, providing technical support and engineering practice guidance for the operation optimization and intelligent control of heavy media sorting processes.

[0035] Furthermore, the raw coal desliming screen module models the raw coal screening process using a physical model based on mass conservation. Specifically, the screening process of the raw coal desliming screen is determined by the raw coal feed mass flow rate. Effective feed mass flow rate after desliming The product relationship represents the screening efficiency. This is a key parameter in the process, representing the proportion of fine coal slime removed, and satisfies the following relationship:

[0036] , Furthermore, the mixing tank is described using a dynamic modeling method based on mass conservation to depict the density change process of the mixed slurry. The inputs to the mixing tank include deslimed raw coal, qualified heavy media suspension, and makeup water introduced through a water supply valve. Based on the mass conservation principle, the cumulative rate of material mass within the mixing tank is equal to the difference between the sum of the mass flow rates of the input streams and the mass flow rate of the discharge stream.

[0037] Density of the mixed slurry in the mixing tank The time-varying dynamic model is represented by the following equation: , in, The volumetric flow rate of the mixing tank. The effective volume of the mixing tank. The density of the mixed slurry, The density change rate of the mixed slurry. For combined medium volumetric flow rate, For the density of the medium, To replenish water volume flow rate, For the density of the water to be replenished, This represents the effective feed mass flow rate after desliming.

[0038] Furthermore, the hydrocyclone module performs the following steps: Step 1: The separation process of the mixed slurry in the hydrocyclone is described using a modeling method based on the principle of mass conservation. This separation process is determined by the relationship between the mass flow rate and the volumetric flow rate of the material. The core function of the hydrocyclone is to separate the mixed slurry into two products, overflow and underflow, through centrifugal force. Based on the mass conservation relationship of the material, the mass flow rate of each material in the hydrocyclone can be described by the following equation:

[0039] , in, This refers to the mass flow rate of the hydrocyclone feed. ( ), ( )and ( ), ( ), representing the volumetric flow rate and density of the overflow and underflow, respectively. and These represent the overflow and underflow volumes, respectively. This is the overflow / underflow ratio coefficient. The mass fraction of each component in the feed is... The mass fractions of the corresponding components in the overflow and underflow are respectively and subscript j Indicates ash content ( ), sulfur content ( ), moisture ( ), volatile matter ( ),medium( ) and carbon ( ) components.

[0040] Subsequently, the hydrocyclone overflow enters the arc screen and clean coal desliming screen through the overflow port to complete the desliming process. The resulting clean coal product is then sent to a centrifuge for dewatering. Assume that the centrifuge only changes the moisture content of the material, without affecting the mass of other components besides moisture, and that the dewatering efficiency remains constant. Based on this, the mathematical description of the mass flow rates of each component in the final clean coal can be constructed as follows:

[0041] , in , , , and This represents the mass fraction of each component in the final clean coal.

[0042] Step 2, regarding the overflow-underflow ratio coefficient Overflow ash coefficient ash content coefficient of bottom flow We designed a slow-time-scale parameter update algorithm based on Bayesian optimization for three key variables.

[0043] The objective function is constructed as follows: , in, In the formula, The second objective function is... This is a simulated value for the ash content of clean coal. This represents the actual sampled value of clean coal ash content. For the defined sequence of parameter vectors to be optimized, This is the overflow / underflow ratio coefficient. The overflow ash content coefficient, is the underflow ash content coefficient.

[0044] The objective function is modeled using a Gaussian process to obtain its posterior distribution, enabling prediction and optimization of key parameters in unsampled regions.

[0045] , in, It is a mean function. The covariance function is represented using the RBF kernel function: , in, Let the length scale be 1. The acquisition function uses the expected boost function: , in, To standardize the improvement quantity, The cumulative distribution function of the standard normal distribution. Let be the probability density function of the standard normal distribution. In each iteration, find the function that maximizes . The points of the function are used as candidate parameters, and the model is run to obtain new simulated ash values ​​and a new objective function until the convergence condition is met.

[0046] Furthermore, the arc screen module performs the following steps: Step 1: Using a mass conservation-based modeling method, the separation process of the overflow slurry in the arc screen is described. The total mass flow rate of the suspension entering the arc screen varies between the underflow and overflow streams, depending on the arc screen desliming coefficient. The mass is redistributed while the total mass of the system remains conserved.

[0047] The mathematical model of the arc screen process can be represented by the following formula: , in, , and These represent the volumetric flow rate, density, and mass flow rate of the hydrocyclone overflow, respectively. , and These represent the volumetric flow rate, density, and mass flow rate under the arc screen, respectively.

[0048] Step 2, regarding the desizing coefficient of the arc screen Online estimation is performed using a case-based reasoning algorithm: First, a historical case library is built, with each case containing a feature vector. and the corresponding desizing coefficient In the reasoning process, in order to quantify the current working condition... The matching degree of historical cases is defined by a similarity function based on the reciprocal of the weighted Euclidean distance as follows:

[0049] , in, This represents the historical case value of the density of the mixture. This represents historical values ​​for the feed rate to the arc screen. Current operating value of the combined density, This represents the current operating value of the feed rate to the arc screen. and These are weighting coefficients used to adjust the importance of the two factors in similarity calculation. This is a scaling factor used to adjust the rate at which similarity decays with changes in error.

[0050] When new data is collected, select all cases from the case library whose similarity exceeds a preset threshold. The similar cases are used to calculate the similarity-weighted average of the case solutions using the following formula, which yields the arc-shaped screen disintegration coefficient: , In the formula, The desliming coefficient of the arc-shaped screen. The similarity function is based on the reciprocal of the weighted Euclidean distance. n This refers to the number of highly similar cases ultimately selected after filtering according to a preset threshold. i For this n The local index number used for looping through the selected similar cases.

[0051] Furthermore, the desliming and screening module performs the following steps: Step 1: Based on the law of conservation of mass, establish mathematical models for the clean coal desliming screen and the gangue desliming screen respectively: The clean coal desliming screen is used to deslim the product from the hydrocyclone overflow after it has been treated by an arc screen. Its model is as follows: , in, , , These represent the mass flow rate, density, and volumetric flow rate of the material flowing through the arc-shaped screen. and These are the volumetric flow rate of the flushing water to the clean coal desliming screen and the opening degree of the flushing valve, respectively. The desliming coefficient of the clean coal desliming screen. Mass flow rate of underflow from clean coal desliming screen

[0052] Gangue desliming screens are used to deslim the underflow products of hydrocyclones. Their model is as follows: , in, The volumetric flow rate of the flushing water for the gangue desliming screen. For valve opening, and The volumetric flow rate and density of the underflow from the hydrocyclone. This represents the mass flow rate under the arc-shaped screen.

[0053] Step 2, considering the desliming coefficient of the gangue desliming screen. Demediation coefficient of clean coal demediation screen Within a certain range, it is only related to the opening degree of the flushing valve and shows a positive correlation. The larger the opening degree of the flushing valve, the larger the desliming coefficient. Therefore, an adaptive neural fuzzy inference system based on the first-order Sugeno function can be used to estimate the desliming coefficient of clean coal and gangue desliming screen.

[0054] The adaptive neural fuzzy inference system based on the first-order Sugeno function comprises five layers: input fuzzification, rule activation, rule normalization, rule conclusion, and output computation. The input layer membership function uses the following Gaussian function to transform the input into fuzzy values:

[0055] , in, For membership degree, For valve opening, It is the center of the fuzzy set. It is the width of the fuzzy set.

[0056] The membership relationship between the input and the fuzzy set is represented by calculating the activation strength of each rule. A larger activation strength indicates a greater contribution of the rule to the final output. Then, the activation values ​​of all rules are normalized, and a first-order Sugeno model is used. Calculate the conclusion for each rule. To adapt to time-varying operating conditions, the model's consequent parameters... Online estimation of the desizing coefficients of clean coal and gangue desizing screens can be achieved by using the recursive least squares method for online updating. The covariance matrix is ​​defined. P and forgetting factor , t This represents the current discrete time step. This represents the latest data regression vector acquired at the current moment. The parameter update function is:

[0057] , This method can dynamically track the drift of the performance of the desliming sieve by using a forgetting factor while ensuring the convergence speed.

[0058] Furthermore, the diversion valve module proportionally distributes the medium flow rate into the combined medium tank and the dilute medium tank, thereby achieving efficient medium circulation and density control. Its mathematical model can be expressed by the following formula: , in, and These represent the densities of the suspensions entering the combined medium tank and the dilute medium tank after the diversion, respectively. The density of the material passing through the arc screen. For the opening degree of the diverter valve, and These are the equivalent pipe diameters leading to the combined medium tank and the dilute medium tank, respectively. and These are the volumetric flow rates flowing into the combined medium tank and the dilute medium tank, respectively.

[0059] Furthermore, the rarefied medium module describes the density change during the rarefied process of the medium through a dynamic modeling method based on mass conservation.

[0060] , in, This represents the total undersize mass flow rate of the desliming screen. ( ), ( ), ( )and (m) represents the density, volume, cross-sectional area, and liquid level of the dilute medium tank, respectively. ( ), ( ), ( ), ( )and ( The numbers () represent the volumetric flow rate and density from the diversion valve, the desliming screen, and the water supply valve, respectively. ( (This refers to the opening degree of the water supply valve for the dilute medium tank.)

[0061] Furthermore, the magnetic separator module performs the following steps: Step 1: Based on the law of conservation of mass, establish a mathematical model for the media recovery process of the magnetic separator: , in, and The inputs are mass flow rate and density, respectively. The flushing water volume flow rate, The recovery coefficient of the magnetic separator. This refers to the volumetric flow rate of the suspension at the outlet of the magnetic separator. Adjust the opening of the water supply valve for the magnetic separator.

[0062] Step 2, the recovery coefficient of the magnetic separator is affected by the opening degree of the magnetic separator's flushing valve. Magnetic field strength of magnetic separator B Feed gate opening Magnetic separator feed density Due to the influence of multiple variables, the RVFLN network is used for description.

[0063] Constructing input vectors The input vector is processed with randomly generated fixed weights. and bias The output is mapped to the hidden layer and processed by the ReLU activation function to obtain the hidden layer output, which is then processed through the output weights. Linear combination yields an estimate of the magnetic separator recovery coefficient. Its mathematical expression is as follows:

[0064] , Unlike the pseudo-inverse solution of traditional RVFLN, this paper proposes an output layer weight method to adapt to the characteristics of online streaming data. Iterative updates are performed using gradient descent: , in For learning rate, This serves as the mentor signal. This update method avoids the computational overhead of large-scale matrix inversion, making it more suitable for real-time control systems.

[0065] Step 3 addresses the lack of direct measurement methods for the desliming coefficients of the arc screen, clean coal desliming screen, gangue desliming screen, and magnetic separator in actual production, leading to a training dilemma for the intelligent model without supervised signals. A parameter inversion strategy based on Monte Carlo data mining is adopted to mine supervisory signals that conform to the current production process, thereby achieving continuous online updates of the four key parameters of the density loop.

[0066] Define the sequence of parameter vectors to be inverted as follows: Then the objective function is constructed as follows: , In the formula, Let the first objective function be... This is the simulated value of the combined dielectric density. The sampled value is the density of the mixture. The length of the sample window. Given the defined sequence of parameter vectors to be inverted, The desliming coefficient of the arc-shaped screen. The recovery coefficient of the magnetic separator. The desliming coefficient of the clean coal desliming screen. Desliming coefficient of gangue desliming screen.

[0067] Perform in the preset parameter space After random sampling, one Parameter combination and objective function value The posterior probability weights for each parameter combination are calculated as follows: , in, As a decay factor, it allows for a wider exploration range in the early stages and a greater focus on the optimal solution in the later stages.

[0068] Finally, the parameters can be estimated by weighted averaging: , This represents the optimal combination of key equipment parameters. This represents the total number of random samples performed within the parameter space. For the first m The parameter combination obtained from the second sampling; The posterior probability weights for the parameter combination. The parameter combination corresponding to the first objective function The value of , This is the attenuation factor.

[0069] This estimated value is regarded as the optimal parameter solution under the current working conditions. It is used as a mentor signal to feed back to the front-end case reasoning, ANFIS and RVFLN models in real time for supervised learning, thereby realizing data mining to guide model updates.

[0070] Furthermore, the media recovery module, used to describe the dynamic changes in the media recovery process, can be represented by the following mathematical model: , in, , , , and , These refer to the volumetric flow rate and density of the medium suspension flowing into the diversion valve, the medium suspension recovered by the magnetic separator, and the supplementary medium introduced during the media addition process. The outflow medium suspension volumetric flow rate of the combined tank. The cross-sectional area of ​​the combined tank is... The density of the suspension in the mixing tank. and These represent the volume and liquid level of the mixing tank, respectively.

[0071] Furthermore, the bottom water replenishment module of the combined medium tank is used to describe the dynamic characteristics of the final change in the combined medium density after bottom water replenishment, and its mathematical model can be expressed as: , in, The volumetric flow rate of water replenishment at the bottom of the mixing tank. To adjust the opening of the water supply valve, The volumetric flow rate of the qualified medium formed after mixing. For effective mixing volume, The final qualified medium density.

[0072] See Figure 12 As shown, based on the above system, this solution further provides a dynamic simulation method for heavy medium sorting, including the following steps: S1: Based on the heavy media separation process, construct a dynamic mechanism model covering the entire heavy media separation process. The dynamic mechanism model includes at least a system model consisting of a raw coal desliming screen module, a mixing tank module, a hydrocyclone module, an arc screen module, a desliming screen module, a diverter valve module, a dilute medium tank module, a magnetic separator module, a combined medium tank module, and a combined medium tank water supply module, which are formed by the flow of media.

[0073] S2: On a fast timescale, online estimation is performed for several key parameters in the density control loop that lack direct measurement methods. These key parameters include: the desliming coefficient of the arc screen module, the desliming coefficients of the clean coal desliming screen and the gangue desliming screen of the desliming screen module, and the magnetic separator recovery coefficient of the magnetic separator module.

[0074] S3: Input the real-time collected system process data into the dynamic mechanism model, and run the dynamic mechanism model based on the key parameters estimated online in step S2 to calculate and output the simulated value of the combined medium density.

[0075] S4: On the fast timescale, a first objective function is constructed based on the error between the simulated density value and the actual sampled value output in step S3. The Monte Carlo data mining algorithm is used to globally optimize and solve the multiple key parameters mentioned in step S2. The optimal parameter combination obtained is fed back to step S2 as a mentor signal to supervise and correct the online estimation process of the key parameters.

[0076] S5: On a slow time scale, based on the updated key parameters, run the dynamic mechanism model to calculate and output the simulated value of clean coal ash content; construct a second objective function based on the error between the simulated value of clean coal ash content output in step S5 and the actual sampled value, and use a Bayesian optimization algorithm to optimize and update the key sorting parameters that affect the sorting accuracy in the hydrocyclone module. The key sorting parameters include the overflow-underflow ratio coefficient, the overflow ash content coefficient, and the underflow ash content coefficient.

[0077] S6: Based on the dynamic mechanism model continuously updated and optimized through steps S2 to S5, the operating status of the heavy medium separation process is simulated in real time, and decision reference information for density control and separation process optimization is output. The decision reference information includes at least the opening setting values ​​of the water supply valve and diversion valve of the medium tank for adjusting the density of qualified medium, and the predicted value of clean coal ash content.

[0078] Furthermore, this invention deeply integrates the mechanistic model with a data-driven algorithm. To verify the effectiveness of the dynamic model proposed in this invention, production data from a coal preparation plant in Yulin, Shaanxi Province, on the evening of April 26, 2025, was used in a MATLAB environment for verification. The coal preparation plant's process is as follows: Figure 2 As shown, hydrocyclones are used for separation, and density is controlled by adjusting the water supply valve and diversion valve of the combined medium tank. The coal preparation plant currently has three sensors: a level gauge in the dilute medium tank, a density gauge in the combined medium pipeline, and a level gauge in the combined medium tank, used to detect relevant data in the medium recovery loop, with a sampling cycle of 5 seconds. Due to time constraints for manual sampling and testing during heavy medium separation, ash content data is tested every 5 minutes. All valve openings use data from real coal preparation plants. The differential equations mentioned in this paper are solved using the Runge-Kutta method. Based on this, to verify the effectiveness of the proposed multi-timescale parameter update algorithm, a comparative experiment is conducted with a fixed-parameter heavy medium separation model. The fixed parameters are selected offline by optimizing production data to achieve the optimal condition for the model. Specifically, the desliming coefficient of the arc screen is 95%, the desliming coefficient of the clean coal and gangue desliming screen is 99%, and the recovery coefficient of the magnetic separator is 20%.

[0079] First, simulation verification was performed on the density control loop. The simulation results are as follows: Figures 3-8 As shown. Figure 3The diagram illustrates the motion of two key actuators in the density control loop: the diversion valve and the water supply valve for the combined medium tank. It can be seen that during the density control process in heavy medium separation, the diversion valve operates more significantly and changes more slowly, while the water supply valve for the combined medium tank fluctuates frequently and continuously. This is because the diversion valve's influence on the combined medium density is relatively delayed and significant, and it is typically used as a "coarse adjustment" method. The density meter is usually installed at the rear end of the water supply valve in the combined medium tank; therefore, the combined medium density is more sensitive to adjustments in the water supply valve. When the density is roughly within the target range, the water supply valve is usually adjusted frequently to achieve precise control of the combined medium density through "fine adjustments."

[0080] Figure 4 The opening degrees of the water supply valves for the gangue desliming screen, the clean coal desliming screen, and the dilute medium tank were demonstrated. All valve data were obtained from the field. It is worth mentioning that, because the plant has achieved full automation and intelligence, the valve opening degrees are calculated and implemented by the automatic control program, resulting in frequent fluctuations in the opening degrees of each valve.

[0081] Regarding the estimation of the desliming coefficient of the arc screen, in the case-based reasoning method, the weighting coefficients for the bulk density and the raw coal feed amount are respectively taken as... and The scaling factor was set to 0.1, the similarity threshold to 0.7, and cases with similarity exceeding this threshold were selected for weighted fusion. The descaling coefficients of clean coal and gangue were estimated using the ANFIS method, with the conclusion function parameters updated online via recursive least squares. The membership function was a Gaussian function, the number of fuzzy sets was 3, and the forgetting factor was set to 0.99. The recovery coefficient of the magnetic separator was estimated using an RVFLN network with 20 hidden nodes. The input weights were initialized using Xavier, the output weights were initially zero, the activation function was ReLU, and the learning rate was 0.001 during training. The global parameter optimization of the density control loop employed a Monte Carlo data mining strategy, with 100 sampling times, an optimization window length of 10, and a decay factor of 0.95.

[0082] In the density control loop, the online estimation results of four parameters—the desliming coefficient of the arc screen, the desliming coefficient of the clean coal desliming screen, the desliming coefficient of the gangue desliming screen, and the recovery coefficient of the magnetic separator—are as follows: Figure 5 As shown, because online estimation methods tailored to production characteristics were designed for each of the four parameters, the parameters are updated in real time based on the simulation results at each sampling moment (fast time scale) during online operation, continuously learning from the mentor signal, thereby improving the overall model accuracy. Among them, the magnetic separation recovery coefficient has the greatest impact on density simulation; therefore, the variation range of the magnetic separator recovery coefficient is relatively large when adjusting the simulated density of the composite medium, while the variation range of the other three parameters is relatively small.

[0083] To facilitate the analysis of the accuracy of the established model, three metrics are selected: mean squared error (MSE), mean relative error (MRE), and maximum error (Max Error) to measure the model accuracy.

[0084] The three indicators are calculated as follows: , , , in, The simulated values ​​of the proposed model are... These are the actual values ​​from on-site sampling.

[0085] The simulation results of the combined medium density, combined medium tank level, and dilute medium tank level using the hybrid intelligent drive method are as follows: Figures 6 to 8 As shown, the model established using the hybrid intelligent hybrid-driven modeling method more closely reflects the dynamic characteristics of the heavy media sorting site compared to the fixed-parameter modeling method. Specifically, Figure 6 In the simulation, the hybrid modeling method can accurately simulate the fluctuations in the density of the composite medium, with an average relative error of only 0.064% within a simulation time of 1750 s. Since density is a crucial variable requiring high precision and of utmost concern in the field, this method demonstrates significant advantages in density prediction due to its small error, highlighting its practical application advantages. Figure 7 In the simulation of liquid level in the combined tank, the hybrid intelligent modeling method is significantly better than the fixed parameter modeling method. Compared with the analysis in Table 1, the average relative error decreased from 3.35% to 2.20%, indicating the effectiveness of the proposed hybrid intelligent modeling method. Figure 8 While mechanistic modeling methods can describe the overall change pattern to some extent, they have certain deviations in most periods and cannot accurately describe the changes in the liquid level of the dilute medium tank.

[0086] Table 1 Key Variable Analysis Table For the hydrocyclone sorting loop, a Bayesian optimization method is used to update key parameters on a slow timescale, running once every 300 seconds, with a maximum of 30 evaluations. The slow timescale parameter estimation results are shown in the figure below. Figure 9 As shown. Overflow / underflow ratio coefficient,

[0087] The three key variables—overflow ash coefficient, underflow ash coefficient, and total ash coefficient—are updated only after each ash sampling, maintaining a relatively slow time scale. During each parameter update, the performance index defined by the following formula is used as the optimization objective function:

[0088] , in, In the formula, The second objective function is... This is a simulated value for the ash content of clean coal. This represents the actual sampled value of clean coal ash content. For the defined sequence of parameter vectors to be optimized, This is the overflow / underflow ratio coefficient. The overflow ash content coefficient, is the underflow ash content coefficient.

[0089] By minimizing the error between the ash content verification value and the model simulation value, key parameters can be corrected online. This method can adaptively adjust model parameters according to changes in system operating status, effectively improving the simulation accuracy of overflow ash and the dynamic tracking capability of the model.

[0090] Simulation results of the hydrocyclone sorting circuit are as follows: Figures 10 to 11 As shown. Figure 10 This is a simulation diagram showing the proportions of various components in the clean coal. The carbon content is approximately 74%, ash approximately 12%, volatile matter approximately 12%, while sulfur and moisture are relatively low, each around 0.8%. The simulation results are consistent with the actual operating characteristics of the heavy media separation process in terms of component distribution, indicating that the model has good rationality at the material composition level.

[0091] To further verify the effectiveness of the proposed model, it was compared with the actual ash content samples taken on-site. The sampling period of the on-site ash analyzer was 5 minutes, as detailed below. Figure 11 As shown in the figure, during the simulation period, the model established using the hybrid intelligent drive algorithm achieved a MSE of 0.013 between the simulated and sampled ash values, significantly lower than the 0.37 of the constant parameter modeling method; a MRE of 0.81%, compared to 4.1% for the constant parameter modeling method; and a Max Error of 0.19, also significantly better than the 0.86 of the constant parameter modeling method. All three indicators are far lower than those of the constant parameter modeling method. This indicates that the hybrid intelligent drive modeling method can effectively adjust model parameters and accurately simulate the actual operating state of the hydrocyclone, providing important guidance for production in heavy medium separation sites.

[0092] In summary, the present invention has the following technical advantages: 1. This invention overcomes the limitation of fixed-parameter mechanistic models being prone to distortion under complex working conditions. By deeply integrating the mechanistic model with data-driven processes, and using real-time errors to generate mentor signals, it corrects key equipment parameters that are difficult to measure online, achieving high-precision dynamic tracking and accurate simulation of the entire heavy medium sorting process.

[0093] 2. This invention overcomes the problem that a single time scale cannot simultaneously account for the dynamic characteristics of multiple variables. It adopts a dual time scale mechanism: the fast scale tracks the high-frequency changing density control parameters online, while the slow scale optimizes the low-frequency updated ash parameters, accurately matching the actual physical process and significantly improving the model's adaptability.

[0094] 3. This invention overcomes the shortcomings of traditional open-loop simulation in guiding actual production. Based on intelligent algorithms, it continuously reduces operational errors, outputs real-time information on the opening degrees of the water supply valve and diversion valve, and predicts the ash content of the clean coal, forming dynamic feedback and closed-loop optimization. This provides highly reliable decision support for on-site sorting process control.

[0095] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A dynamic simulation method for heavy medium sorting, characterized in that, Includes the following steps: S1. Based on the heavy media separation process, construct a dynamic mechanism model covering the entire heavy media separation process. The model includes at least a raw coal desliming screen module, a mixing tank module, a hydrocyclone module, and a density control loop formed by the flow of media from an arc screen module, a desliming screen module, a magnetic separator module, a diverter valve module, a dilute medium tank module, a combined medium tank module, and a combined medium tank water supply module. S2. On a fast timescale, the key equipment parameters that cannot be directly measured in the density control loop are estimated online, and the online estimated key equipment parameters are input into the dynamic mechanism model to obtain the simulated value of the combined density. S3. Based on the error between the simulated value and the actual sampled value of the combined density, construct a first objective function, perform global optimization on the key equipment parameters in the dynamic mechanism model, obtain the optimal combination of key equipment parameters, and feed it back to step S2 as a mentor signal to supervise and correct the online estimation process of the key equipment parameters and obtain the updated key equipment parameters. S4. Input the updated key equipment parameters back into the dynamic mechanism model to obtain the simulated value of clean coal ash content; S5. On a slow time scale, a second objective function is constructed based on the error between the simulated value and the actual sampled value of clean coal ash content. Based on the second objective function, the key separation parameters that affect the separation accuracy in the hydrocyclone module are optimized and updated. S6. Based on the dynamic mechanism model continuously updated and optimized from S2 to S5, the operating status of the heavy media sorting process is simulated in real time, and decision reference information for density control and sorting process optimization is output.

2. The dynamic simulation method for heavy medium sorting according to claim 1, characterized in that, The key equipment parameters include at least: the arc screen desliming coefficient of the arc screen module, the desliming coefficient of the clean coal desliming screen and the desliming coefficient of the gangue desliming screen of the desliming screen module, and the magnetic separator recovery coefficient of the magnetic separator module; the key separation parameters include at least: the overflow underflow ratio coefficient, the overflow ash coefficient, and the underflow ash coefficient.

3. The dynamic simulation method for heavy medium sorting according to claim 2, characterized in that, The desliming coefficient of the arc screen is estimated online using a case-based reasoning algorithm; the desliming coefficients of the clean coal desliming screen and the gangue desliming screen are estimated using an adaptive neural fuzzy reasoning system based on a first-order Sugeno function, and updated online using the recursive least squares method; the recovery coefficient of the magnetic separator is estimated using a random vector function linking network.

4. The dynamic simulation method for heavy medium sorting according to claim 3, characterized in that, The desliming coefficient of the arc-shaped screen is determined according to the following formula: ， in, , In the formula, The desliming coefficient of the arc-shaped screen. The similarity function is based on the reciprocal of the weighted Euclidean distance. n This refers to the number of highly similar cases ultimately selected after filtering according to a preset threshold. i For this n Among the selected similar cases, the local index number used for loop traversal is... This represents the historical case value of the density of the mixture. This represents historical values ​​for the feed rate to the arc screen. Current operating value of the combined density, This represents the current operating value of the feed rate to the arc screen. and These are the weighting coefficients. This is the scaling factor.

5. The dynamic simulation method for heavy medium sorting according to claim 3, characterized in that, The recovery coefficient of the magnetic separator is determined according to the following formula: ， in, , In the formula, The recovery coefficient of the magnetic separator. To output weights, L This represents the total number of hidden layer nodes in the neural network. j It is the local index number of each neuron during the traversal. ReLU For activation function, For the input vector, The input vector is given fixed weights through random generation. The input vector is a randomly generated bias. The opening degree of the water flushing valve for the magnetic separator. This refers to the opening degree of the feed gate. This refers to the feed density of the magnetic separator. This represents the magnetic field strength of the magnetic separator.

6. The dynamic simulation method for heavy medium sorting according to claim 1, characterized in that, The first objective function is determined according to the following formula: ， in, , In the formula, Let the first objective function be... This is the simulated value of the combined dielectric density. The sampled value is the density of the mixture. The length of the sample window. Given the defined sequence of parameter vectors to be inverted, The desliming coefficient of the arc-shaped screen. The recovery coefficient of the magnetic separator. The desliming coefficient of the clean coal desliming screen. Desliming coefficient of gangue desliming screen.

7. The dynamic simulation method for heavy medium sorting according to claim 6, characterized in that, The optimal combination of key equipment parameters is determined according to the following formula: ， in, , In the formula, This represents the optimal combination of key equipment parameters. This represents the total number of random samples performed within the parameter space. For the first m The parameter combination obtained from the second sampling; The posterior probability weights for the parameter combination. Let the first objective function be... This is the attenuation factor.

8. The dynamic simulation method for heavy medium sorting according to claim 1, characterized in that, The second objective function is determined according to the following formula: ， in, , In the formula, The second objective function is... This is a simulated value for the ash content of clean coal. This represents the actual sampled value of clean coal ash content. Given the defined sequence of parameter vectors to be optimized, This is the overflow / underflow ratio coefficient. The overflow ash content coefficient, is the underflow ash content coefficient.

9. The dynamic simulation method for heavy medium sorting according to claim 1, characterized in that, The decision reference information includes at least the opening settings of the combined water supply valve and the diversion valve for adjusting the density of the qualified medium, as well as the predicted value of the ash content of the clean coal. The dynamic relationship between the opening degree of the combined tank water supply valve and the density of the qualified medium is determined according to the following formula: ， in, For a qualified medium density, Combined water supply volume flow rate at the bottom of the tank To set the opening degree of the water supply valve for the combined tank. For valve pressure drop, This represents the ratio of the liquid's density to the density of water. The density of water, The volumetric flow rate of the qualified medium formed after mixing. For effective mixing volume, The density of the suspension in the mixing tank. The outflow medium suspension volume flow rate is given by the combined media tank.

10. A dual-timescale hybrid intelligent driven dynamic simulation system for heavy media sorting, used to implement the method shown in any one of claims 1-9, characterized in that, include: The model building unit is used to build a dynamic mechanism model covering the entire heavy medium separation process. The model includes at least a raw coal desliming screen module, a mixing tank module, a hydrocyclone module, and a density control loop formed by the flow of media from an arc screen module, a desliming screen module, a magnetic separator module, a diverter valve module, a dilute medium tank module, a combined medium tank module, and a combined medium tank water supply module. The fast timescale parameter estimation and optimization unit is used to estimate the key equipment parameters of the density control loop online on a fast timescale, and input the online estimated key equipment parameters into the dynamic mechanism model to obtain the simulated value of the combined density. At the same time, based on the error between the simulated value and the actual sampled value of the combined density, a first objective function is constructed to globally optimize the key equipment parameters, obtain the optimal combination of key equipment parameters, and use it as a mentor signal to supervise and correct the online estimation process. The slow timescale parameter optimization unit is used to first input the updated key equipment parameters into the dynamic mechanism model again under a slow timescale to obtain the simulated value of clean coal ash content, and then construct a second objective function based on the error between the simulated value of clean coal ash content and the actual sampled value to optimize and update the key sorting parameters that affect the sorting accuracy in the hydrocyclone module. The simulation and output unit is used to simulate the operation status of the heavy medium sorting process in real time based on the updated dynamic mechanism model, and output decision reference information for density control and sorting process optimization.

Images