Metal mine crushing and grinding process simulation and prediction method and system

Abstract

The application discloses a kind of metal mine broken grinding process simulation prediction method and system, it is related to grinding control technical field, including, collection full particle size distribution data, ore composition data and ore grade, transmission to edge computing node carries out data cleaning, extracts key feature parameters, generates ore gene atlas vector;Actual full particle size distribution data and equipment energy consumption data are compared with operating parameters, when any data exceeds preset prediction threshold, construct time-delay causal diagram, locate fault equipment, output equipment health degree report;According to equipment health degree report, using generative adversarial network generates adversarial sample injection digital twin model, reconstructs broken grinding process parameters to complete digital twin model self-calibration, and update pre-stored typical ore sample.The application improves the working condition adaptability of broken grinding process by dynamic gene matching mechanism.

Description

Technical Field

[0001] This invention relates to the field of grinding control technology, and in particular to a method and system for simulating and predicting the crushing and grinding process in metal mines. Background Technology

[0002] In recent years, intelligent control technology for metal mine crushing and grinding processes has developed rapidly, and the combination of digital twin models and edge computing has been gradually applied to the field of process optimization. Existing technologies achieve steady-state simulation and control command generation of the crushing and grinding process by real-time acquisition of equipment current, power, and ore particle size data, combined with parameters matched to a historical ore sample database. Meanwhile, fault diagnosis methods based on causal analysis have been validated in industrial scenarios, enabling the location of equipment-level fault sources. Generative adversarial networks are applied to the self-calibration process of digital twin models, enhancing model robustness through adversarial examples.

[0003] Existing technologies still have shortcomings. Under conditions of abrupt changes in ore properties, traditional Euclidean distance matching mechanisms struggle to dynamically adapt to the nonlinear drift of ore genetic characteristics, leading to mismatches in crushing and grinding parameters. When the quartz content of the ore changes drastically, the static ore sample library cannot capture genetic vector shifts in real time, causing process malfunctions. Existing solutions exhibit response delays on the order of minutes, resulting in a surge in energy consumption and equipment wear and tear. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a simulation and prediction method for the crushing and grinding process in metal mines to solve the problem of equipment malfunction and energy consumption surge caused by mismatch in crushing and grinding process parameters due to ore gene mutations.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for simulating and predicting the crushing and grinding process in a metal mine, comprising: collecting full-size distribution data, ore composition data, and ore grade; transmitting the data to an edge computing node for data cleaning; extracting key feature parameters; generating an ore gene map vector; calculating the Euclidean distance between the ore gene map vector and a pre-stored typical ore sample; if the Euclidean distance exceeds a preset mutation threshold, searching a historical database to match the most similar ore sample and selecting the corresponding crushing and grinding process parameters; combining the crushing and grinding process parameters with real-time collected crusher current and mill power data; performing steady-state simulation calculations of the crushing and grinding process; and outputting the results. The system generates control commands based on operating parameters. According to these commands, it adjusts the crusher discharge port width, mill feed water flow rate, and hydrocyclone feed pressure in real time, while simultaneously collecting actual full-size particle distribution data and equipment energy consumption data. The system compares the actual full-size particle distribution data and equipment energy consumption data with the operating parameters. If any data exceeds a preset prediction threshold, a time-delay causal graph is constructed to locate the faulty equipment and output an equipment health report. Based on the equipment health report, a generative adversarial network is used to generate adversarial samples, which are then injected into the digital twin model. This reconstructs the crushing and grinding process parameters, completes the self-calibration of the digital twin model, and updates pre-stored typical ore samples.

[0008] As a preferred embodiment of the metal mine grinding process simulation and prediction method of the present invention, the key characteristic parameters include quartz content, pyrite content and compressive strength.

[0009] As a preferred embodiment of the metal mine grinding process simulation and prediction method of the present invention, the specific steps for transmitting the generated ore gene map vector to the digital twin model are as follows.

[0010] The algorithm of applying the mean filtering algorithm to remove noise from the full-grained distribution data, the Gaussian standardization of the ore composition data, and the integrity verification and outlier removal of the ore grade data are performed to generate a structured dataset.

[0011] Key feature parameters are extracted from the structured dataset and dynamically weighted and fused with the full-granularity distributed data in the structured dataset to generate a fused data vector.

[0012] Multi-scale frequency domain feature components are generated by scanning the full-grained distribution data in the fused data vector using a multi-scale frequency domain integral kernel. The entropy weighting factor is calculated based on the coupling relationship between the proportion of quartz and the proportion of pyrite, and then fused with the multi-scale frequency domain feature components to generate the fusion result.

[0013] The contribution of compressive strength to the fusion result is superimposed to generate the ore gene map vector.

[0014] As a preferred embodiment of the metal mine grinding process simulation and prediction method of the present invention, the specific steps for selecting the corresponding grinding process parameters are as follows:

[0015] The ore gene map vector is projected onto the tangent space of the Riemannian manifold to generate the tangent space coordinate vector;

[0016] Based on the tangent space coordinate vector and the tangent space coordinate vector of pre-stored typical mineral samples, and combined with the difference in quartz proportion, the weight of each dimension is dynamically adjusted to calculate the weighted geodesic distance.

[0017] When the weighted geodesic distance exceeds the preset mutation threshold, the mineral samples that are in the same manifold neighborhood as the current tangent space vector are retrieved from the historical database, the optimal matching sample is selected, and the crushing and grinding process parameters are extracted.

[0018] The parameter migration confidence is calculated based on the weighted geodesic distance between the current ore sample and the optimal matching sample. If the parameter migration confidence is lower than the preset confidence threshold, an alarm is triggered; otherwise, the corresponding crushing and grinding process parameters are selected.

[0019] In a preferred embodiment of the metal mine grinding process simulation and prediction method of the present invention, the specific steps for generating control commands are as follows:

[0020] The corresponding crushing and grinding process parameters are fused with the real-time collected crusher current and mill power data to construct a tensor state space model and output the state space matrix.

[0021] Based on the state-space matrix, with the constraint objective of minimizing the entropy increase rate of equipment operation, tensor product optimization calculation is performed in the Riemann manifold space to generate the optimal control law;

[0022] The optimal control law is penalized by the gradient penalty of the rate of change of entropy, the operating parameters are output and the energy optimization rate is calculated. If the energy optimization rate exceeds the preset optimization threshold, a control command is generated.

[0023] As a preferred embodiment of the metal mine grinding process simulation and prediction method of the present invention, the specific steps for simultaneously collecting actual full-size distribution data and equipment energy consumption data are as follows.

[0024] The control commands are decomposed into three types of operations: high-frequency pulse action, medium-frequency continuous adjustment, and low-frequency steady-state maintenance. The execution priority is dynamically allocated according to the real-time crusher current and mill power, and the layered action commands are output.

[0025] Based on hierarchical action instructions, pre-stored device physical layout parameters are invoked, and combined with the inherent response delay of the device, a collaborative instruction defining the start time and duration is generated;

[0026] Upon completion of the collaborative command, the synchronous acquisition of full-granularity distributed data and equipment energy consumption data is triggered.

[0027] As a preferred embodiment of the metal mine grinding process simulation and prediction method of the present invention, the specific steps for constructing the time-delay causality graph are as follows:

[0028] The actual full-grained distribution data and equipment energy consumption data are compared with the operating parameters. When any data exceeds the corresponding preset prediction threshold, the time window is divided based on the time of exceeding the limit.

[0029] Based on the pre-stored physical layout parameters of the equipment, analyze the disturbance propagation characteristics of the out-of-limit deviations between the equipment within each time window, and generate the dynamic correlation strength between the equipment.

[0030] Using devices as nodes and the dynamic correlation strength between devices as edge weights, combined with pre-stored signal transmission delay parameters between devices, a directed weighted graph with delay labels is generated as a delay causal graph.

[0031] As a preferred embodiment of the metal mine grinding process simulation and prediction method of the present invention, the output equipment health report includes the following specific steps.

[0032] Starting from the edge with high correlation strength in the time-delay causal graph, three layers of adjacent device nodes are dynamically expanded. The impact of neighborhood disturbances is aggregated through the time decay characteristic to generate the root cause health score of the device.

[0033] Starting from devices with low health scores, the transmission path is traced back along edges with high correlation strength to generate a path topology and output a device health report.

[0034] As a preferred embodiment of the metal mine grinding process simulation and prediction method of the present invention, the specific steps for updating the pre-stored typical ore samples are as follows:

[0035] Extract root cause health scores and transmission path scores from equipment health reports, generate health feature vectors, and generate ore gene map perturbation vectors through generative adversarial networks.

[0036] The perturbation vector of the ore gene map is injected into the digital twin model to reconstruct the crushing and grinding process parameters, complete the self-calibration of the digital twin model, and update the pre-stored typical ore samples.

[0037] Secondly, the present invention provides a simulation and prediction system for the crushing and grinding process of a metal mine, comprising a data acquisition module, a matching ore sample module, a steady-state simulation module, an execution instruction module, a fault tracing module, and a model optimization module;

[0038] The data acquisition module is used to collect full-grained distribution data, ore composition data and ore grade, transmit them to edge computing nodes for data cleaning, extract key feature parameters, and generate ore gene map vectors.

[0039] The matching mineral sample module is used to calculate the Euclidean distance between the ore gene map vector and the pre-stored typical mineral sample. If the Euclidean distance exceeds the preset mutation threshold, the most similar mineral sample is retrieved from the historical database, and the corresponding crushing and grinding process parameters are selected.

[0040] The steady-state simulation module is used to combine the crushing and grinding process parameters with the real-time collected crusher current and mill power data, perform steady-state simulation calculations of the crushing and grinding process, output operating parameters, and generate control commands.

[0041] The execution instruction module is used to adjust the width of the crusher discharge port, the flow rate of the mill water supply, and the feed pressure of the hydrocyclone in real time according to the control instructions, while collecting actual full particle size distribution data and equipment energy consumption data.

[0042] The fault tracing module is used to compare the actual full-granularity distribution data and equipment energy consumption data with the operating parameters. When any data exceeds the preset prediction threshold, a time delay cause-effect graph is constructed to locate the faulty equipment and output an equipment health report.

[0043] The model optimization module is used to generate adversarial samples using generative adversarial networks based on the equipment health report, inject them into the digital twin model, reconstruct the crushing and grinding process parameters to complete the self-calibration of the digital twin model, and update the pre-stored typical ore samples.

[0044] The beneficial effects of this invention are as follows: It enhances the adaptability of the crushing and grinding process through a dynamic gene matching mechanism. When the Euclidean distance between the ore gene map vector and a typical ore sample exceeds a preset mutation threshold, a historical database search is automatically activated to match the most similar ore sample and migrate its crushing and grinding parameters, achieving second-level parameter adaptation. Based on the weighted geodesic distance calculation using Riemannian manifold projection, combined with dynamic weight adjustment based on quartz proportion differences, it accurately captures local manifold structure changes in the ore gene. Through collaborative decision-making using parameter migration confidence and energy optimization rate thresholds, it avoids equipment overload caused by mismatches. This ensures the stability of the crushing and grinding process under conditions of ore property mutations, providing a reliable parameter basis for steady-state optimization. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0046] Figure 1 A flowchart for a simulation and prediction method for the grinding process in metal mines.

[0047] Figure 2 This is a schematic diagram of a simulation and prediction system for the crushing and grinding process in a metal mine.

[0048] Figure 3 A flowchart for parameter matching and decision-making in the grinding process.

[0049] Figure 4 A flowchart for generating ore gene map vectors. Detailed Implementation

[0050] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0051] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0052] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0053] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a method for simulating and predicting the crushing and grinding process in a metal mine, comprising the following steps:

[0054] S1. Collect full-scale distribution data, ore composition data, and ore grade, transmit them to edge computing nodes for data cleaning, extract key feature parameters, and generate ore gene map vectors.

[0055] S1.1: Key characteristic parameters include quartz content, pyrite content, and compressive strength.

[0056] It should be noted that during the data cleaning stage, the percentage of quartz and pyrite is extracted from the ore composition data. The percentage of quartz refers to the mass percentage of quartz minerals in the total ore composition, and the percentage of pyrite refers to the mass percentage of pyrite minerals in the total ore composition. The compressive strength is extracted from the ore grade data. The compressive strength refers to the physical index of the ore's ability to resist crushing by external forces.

[0057] S1.2: Apply the mean filtering algorithm to remove noise from the full-grained distribution data, perform Gaussian standardization on the ore composition data, perform integrity verification and outlier removal on the ore grade data, and generate a structured dataset;

[0058] Specifically, a high-precision laser particle size analyzer combined with compressed sensing algorithm is used to perform non-contact rapid particle size detection on the ore and output full particle size distribution data;

[0059] By selecting a fixed window size, the median filtering algorithm removes random noise by traversing each data point in the full-grained distribution data and replacing each data point with the median of all data points within the window.

[0060] Ore composition data and ore grade data were collected using X-ray fluorescence spectrometer and X-ray diffraction sensor.

[0061] Calculate the mean and standard deviation of the ore composition data, and apply Gaussian standardization to transform the ore composition data into a distribution with a mean of 0 and a standard deviation of 1.

[0062] Check if there are any missing values ​​in the ore grade data. If there are missing values, fill them in by using the average compressive strength of adjacent data points.

[0063] Identify outliers, define reasonable boundaries based on the overall range of ore grade data distribution, delete outlier data points outside the boundaries, and generate cleaned ore grade data;

[0064] The denoised full-grained distribution data, the standardized ore composition data, and the cleaned ore grade data are integrated into a structured dataset.

[0065] It should be noted that defining a reasonable boundary refers to setting an acceptable range for compressive strength values ​​based on the historical distribution of ore grade data; data points exceeding the acceptable range are considered outliers.

[0066] For example, data points are deleted when the compressive strength value is below 10 MPa or above 200 MPa.

[0067] S1.3: Extract key feature parameters from the structured dataset, perform dynamic weighted fusion with the full-granularity distributed data in the structured dataset, and generate a fused data vector;

[0068] Specifically, the proportion of quartz, pyrite, and compressive strength are directly read from the structured dataset as key feature parameters. Based on the magnitude of the difference between the proportions of quartz and pyrite, the combination tendency of the full-grained data vector and the key feature parameter vector is determined. Data vectors with a high combination tendency are taken as the dominant content, and data vectors with a low combination tendency are taken as the secondary content. The dominant content and the secondary content are directly spliced ​​together to generate the fused data vector.

[0069] S1.4: Scan the full-grained distribution data in the fused data vector using a multi-scale frequency domain integral kernel to generate multi-scale frequency domain feature components. Calculate the entropy weighting factor based on the coupling relationship between the quartz and pyrite proportions, and fuse it with the multi-scale frequency domain feature components to generate the fusion result.

[0070] Specifically, the multi-scale Fourier transform calculates the integral values ​​at each scale in the full-grained distributed data as multi-scale frequency domain feature components by setting integration kernels of different scales.

[0071] The entropy weighting factor is calculated based on the coupling relationship between the proportion of quartz and the proportion of pyrite. The entropy weighting factor is then fused with the multi-scale frequency domain feature components to generate a fusion result.

[0072] The entropy weighting factor is calculated based on the coupling relationship between the proportion of quartz and the proportion of pyrite, and the expression is as follows:

[0073] ;

[0074] In the formula, Represents the entropy weighting factor. Indicates the real-time quartz content. This indicates the historical standard deviation of the quartz content. This indicates the real-time percentage of pyrite. This indicates the historical standard deviation of the pyrite content. Indicates mineral interaction inhibition factor, This represents the coupling coefficient.

[0075] It should be noted that, , , and The dimension of is %. It is dimensionless, and the final output is... Since it is dimensionless, we maintain dimensional consistency.

[0076] The coupling coefficient was determined by fitting the interaction inhibition effect of quartz and pyrite in historical data, with an example value of 0.05.

[0077] S1.5: The contribution of compressive strength to the fusion result is superimposed to generate the ore gene map vector.

[0078] Specifically, when superimposing the contribution value of compressive strength to the fusion result, the adjustment range of each element in the fusion result is determined according to the ratio of compressive strength to the historical maximum compressive strength, and the adjustment range is uniformly superimposed on the corresponding element of the fusion result vector to generate the ore gene map vector.

[0079] S2. Calculate the Euclidean distance between the ore gene map vector and the pre-stored typical ore sample. If the Euclidean distance exceeds the preset mutation threshold, search the historical database to match the most similar ore sample and select the corresponding crushing and grinding process parameters.

[0080] S2.1: Project the ore gene map vector onto the tangent space of the Riemannian manifold to generate the tangent space coordinate vector;

[0081] Specifically, based on the geometric properties of Riemannian manifolds, the ore gene map vector is used as a point in a high-dimensional space;

[0082] The average position of the ore gene map vector of typical mineral samples pre-stored in the historical database in the Riemannian manifold is used as a fixed reference point.

[0083] By using a local linear transformation at a fixed reference point, the ore gene map vector is mapped to a linear coordinate representation in the tangent space of the Riemannian manifold, generating a tangent space coordinate vector.

[0084] S2.2: Specifically, based on the tangent space coordinate vector and the tangent space coordinate vector of pre-stored typical mineral samples, and combined with the difference in quartz proportion, the weights of each dimension are dynamically adjusted to calculate the weighted geodesic distance, the expression of which is:

[0085] ;

[0086] In the formula, Indicates the weighted geodesic distance. Indicates a dimension index. Indicates the total number of dimensions. Indicates to arrive Perform a traversal and summation. The first tangent space coordinate vector of the current ore represents the first tangent space coordinate vector. dimensional components, The first tangent space coordinate vector of a pre-stored typical mineral sample represents the first tangent space coordinate vector. dimensional components, The first tangent space coordinate vector of the current ore represents the first tangent space coordinate vector. The first component of the tangent space coordinate vector of the pre-stored typical mineral sample is the tangent space coordinate vector of the dimensional component. The squared Euclidean distance of the dimensional component, This indicates that all mineral samples in the historical database were in the [number]th [period]. The variance of the dimension, This represents the adjustment coefficient. The exponential decay term representing the difference in the proportion of quartz. Indicates the attenuation coefficient. This indicates the current percentage of quartz in the ore. This indicates the percentage of quartz in the pre-stored mineral sample.

[0087] It should be noted that, , , , and All are dimensionless, and the final output is... The output is dimensionless, while maintaining dimensional consistency.

[0088] The adjustment coefficient is calibrated through historical data regression analysis to balance the weighted effects of characteristic variance and quartz difference; the example value is 0.1. The attenuation coefficient is determined based on the sensitivity fit of the quartz ratio difference to the crushing efficiency; the example value is 1.0.

[0089] S2.3: When the weighted geodesic distance exceeds the preset mutation threshold, retrieve mineral samples in the historical database that are in the same manifold neighborhood as the current tangent space vector, select the best matching sample, and extract the crushing and grinding process parameters;

[0090] Specifically, when the weighted geodesic distance exceeds the preset mutation threshold, the historical database is called, and all mineral samples whose geometric distance from the current tangent space vector in the Riemannian manifold is less than the neighborhood radius are selected from the historical database to generate neighborhood mineral samples.

[0091] The mineral sample with the smallest weighted geodetic distance from the neighboring mineral samples is selected as the optimal matching sample, and the crushing and grinding process parameters of the optimal matching sample are directly extracted from the historical database.

[0092] When the weighted geodetic distance does not exceed the preset mutation threshold, the historical database retrieval process is skipped, and the currently valid crushing and grinding process parameters in the digital twin model are directly used.

[0093] It should be noted that the preset mutation threshold is based on a fixed judgment standard set by the similarity analysis of historical mineral samples, and the example value is 10.0; mineral samples smaller than the neighborhood radius refer to the pre-stored typical mineral samples in the historical database whose geometric distance from the Riemann manifold of the current tangent space vector is smaller than the neighborhood radius (e.g., 5.0); the pre-stored crushing and grinding process parameters refer to the combination of parameters pre-stored in the historical database for typical mineral samples, including the setting value of the crusher discharge port width, the setting value of the mill feed water flow rate, and the setting value of the hydrocyclone feed pressure.

[0094] S2.4: Calculate the parameter migration confidence based on the weighted geodesic distance between the current ore sample and the optimal matching sample. If the parameter migration confidence is lower than the preset confidence threshold, an alarm is triggered; otherwise, the corresponding crushing and grinding process parameters are selected.

[0095] It should be noted that the preset reliability threshold is set based on the statistics of historical parameter migration failure cases, and the example value is 0.3;

[0096] If the confidence level is lower than the preset confidence threshold, an alarm will be triggered. If the confidence level is higher than the preset confidence threshold, the corresponding crushing and grinding process parameters will be selected directly. The corresponding crushing and grinding process parameters refer to the combination of parameters of the crusher discharge port width setting, mill water flow rate setting and hydrocyclone feed pressure setting pre-stored in the historical database of the optimal matching sample.

[0097] Specifically, the parameter migration confidence is calculated based on the weighted geodesic distance between the current mineral sample and the optimal matching sample, expressed as follows:

[0098] ;

[0099] In the formula, Indicates the confidence level of parameter transition. This represents the confidence decay coefficient. This represents the exponentially decaying term.

[0100] It should be noted that, It is dimensionless, and the final output is... Since it is dimensionless, we maintain dimensional consistency.

[0101] The confidence decay coefficient is calibrated by fitting historical data, controlling the rate at which the confidence decays with distance. The example value is 0.8.

[0102] A superior approach addresses the technical bottleneck of traditional DCS real-time matching in metal mine grinding processes by employing a collaborative mechanism of Euclidean distance mutation threshold determination and Riemannian manifold spatial orientation retrieval. Traditional methods require traversing the entire ore sample database for linear similarity calculations and suffer from a high misjudgment rate for fluctuations in key parameters such as quartz content. Furthermore, the lack of fault tolerance mechanisms leads to a high rate of equipment misadjustment. By using an Euclidean distance exceeding a preset mutation threshold as a trigger condition, matching of non-mutated ore samples is avoided. Nonlinear interference is then removed through Riemannian manifold projection, and a new parameter migration confidence circuit breaker mechanism is added to achieve millisecond-level accurate matching, enhanced anti-interference capabilities, and fault risk interception, forming a closed-loop decision chain of "mutation detection - manifold dimensionality reduction - confidence circuit breaker".

[0103] S3. Combine the crushing and grinding process parameters with the real-time collected crusher current and mill power data, perform steady-state simulation calculations of the crushing and grinding process, output operating parameters, and generate control commands.

[0104] S3.1: The corresponding crushing and grinding process parameters are fused with the real-time collected crusher current and mill power data to construct a tensor state space model and output the state space matrix;

[0105] It should be noted that the specific steps for constructing the tensor state space model are as follows: The corresponding crushing and grinding process parameters, including the crusher discharge port width setting, the mill feed water flow setting, and the hydrocyclone feed pressure setting, are aligned and merged with the real-time collected crusher current value and mill power value using the same timestamp to form a five-dimensional data vector; the five-dimensional data vectors from multiple consecutive time steps are stacked into a three-dimensional structure, where the first dimension is the time step sequence, the second dimension is the parameter type, and the third dimension is the equipment type; the dynamic coupling relationship between different equipment parameters in the three-dimensional structure is extracted through tensor decomposition, and this dynamic coupling relationship is mapped to a linear state transition relationship to obtain the tensor state space model.

[0106] Specifically, the corresponding crushing and grinding process parameters are aligned with the real-time collected crusher current value and mill power value according to the timestamp to form a five-dimensional data vector;

[0107] Five-dimensional data vectors from multiple consecutive time steps are stacked into a three-dimensional tensor. The three-dimensional tensor contains time steps, parameter dimensions, and device dimensions. The dynamic coupling relationship between devices is extracted through tensor decomposition, and a state space matrix is ​​generated.

[0108] S3.2: Specifically, based on the state-space matrix, with the constraint objective of minimizing the entropy increase rate of equipment operation, tensor product optimization is performed in the Riemannian manifold space to generate the optimal control law, the expression of which is:

[0109] ;

[0110] In the formula, This represents the optimal control law vector. This represents the vector of control variables to be optimized. To find the minimum value of the constrained objective in the value space. This represents the vector of control variables to be optimized. Represents the state transition matrix. Represents the state vector. Represents the free response component. Represents the control input matrix. This represents the component of the influence of the control input on the state. This represents the L2 norm squaring operation. This represents the entropy increase penalty weight coefficient. Represents the device's operating entropy. Indicates time, Represents the rate of entropy increase. This represents the maximum permissible rate of entropy increase. This indicates a penalty term for violating the entropy increase constraint.

[0111] It should be noted that, Dimensionless The dimension is kW / K. The dimension is kW / K. The dimension of is (kW / K)². The dimension is K² / kW². It is dimensionless, and the final output is... Since it is dimensionless, we maintain dimensional consistency.

[0112] The entropy increase penalty weight coefficient is calibrated through regression analysis of historical fault data; the example value is 10.0.

[0113] S3.3: Apply an entropy increase rate gradient penalty to the optimal control law, output the operating parameters and calculate the energy optimization rate. If the energy optimization rate exceeds the preset optimization threshold, then generate control commands.

[0114] Specifically, calculate the entropy increase rate of the optimal control law. If the entropy increase rate exceeds the allowable fluctuation range, scale the element values ​​of the optimal control law proportionally.

[0115] The adjustment values ​​of the crusher discharge port width, mill feed water flow rate, and hydrocyclone feed pressure are extracted from the optimal control law after penalty and used as operating parameters.

[0116] The energy optimization rate is calculated based on the operating parameters. If the energy optimization rate exceeds the preset optimization threshold, the operating parameters are converted into control commands that the equipment can execute. If the energy optimization rate does not exceed the preset optimization threshold, the parameter matching rollback mechanism is triggered, and the historical database is searched again to match the most similar mineral sample.

[0117] It should be noted that the allowable fluctuation range refers to the empirical value of the maximum allowable change in the rate of entropy increase between adjacent iteration steps; the example value is 0-0.05.

[0118] The preset optimization threshold is the minimum effective optimization ratio set based on historical energy consumption optimization effect statistics. The example value is 0.1.

[0119] The energy optimization rate is calculated using the following expression:

[0120] ;

[0121] In the formula, Indicates the energy optimization rate. Represents the gradient operator, Represents the optimal control law vector The gradient vector of the device's operating entropy. This represents the original control parameter vector before optimization. This indicates the amount of change in the control parameter.

[0122] It should be noted that, Dimensionless Dimensionless, ultimately It is dimensionless, but we maintain dimensional consistency.

[0123] S4. According to the control command, adjust the width of the crusher discharge port, the flow rate of the mill water supply and the feed pressure of the hydrocyclone in real time, and collect the actual full particle size distribution data and equipment energy consumption data at the same time.

[0124] S4.1: Decompose the control commands into three types of operations: high-frequency pulse action, medium-frequency continuous adjustment, and low-frequency steady-state maintenance. Dynamically allocate execution priorities based on real-time crusher current and mill power, and output layered action commands.

[0125] Specifically, high-frequency pulse actions: identify brief and drastic adjustments in the control command that need to be completed in milliseconds, such as the instantaneous opening and closing of the crusher discharge port width;

[0126] Medium-frequency continuous adjustment: Extracts operations that require continuous adjustment on the order of seconds, such as gradually increasing or decreasing the feed water flow of a mill.

[0127] Low-frequency steady-state maintenance: Screening parameters that need to remain stable on the minute level, such as maintaining a constant value of hydrocyclone feed pressure;

[0128] The crusher current fluctuation amplitude and mill power change rate are monitored in real time. If the crusher current fluctuation amplitude exceeds the limit, high-frequency pulse action is executed first; if the mill power change rate exceeds the limit, medium-frequency continuous adjustment is executed first; if neither the crusher current fluctuation amplitude nor the mill power change rate exceeds the limit, the operation is executed in the order of low-frequency steady-state maintenance, medium-frequency continuous adjustment and high-frequency pulse action, and the layered action command is output.

[0129] It should be noted that exceeding the limit refers to exceeding the safe range; the judgment criterion for the crusher current fluctuation amplitude exceeding the limit is whether the difference between the peak and trough of the crusher current in real time exceeds the pre-stored experience threshold, with an example value of 50 amperes; the judgment criterion for the mill power change rate exceeding the limit is whether the absolute value of the mill power change at adjacent acquisition times exceeds the pre-stored experience threshold, with an example value of 100 kilowatts per second.

[0130] S4.2: Based on hierarchical action instructions, call the pre-stored physical layout parameters of the device, and combine the inherent response delay of the device to generate a collaborative instruction that defines the start time and duration;

[0131] Specifically, read the pre-stored physical layout parameters of the equipment, including the spatial distance between the crusher, mill, and hydrocyclone, as well as the transmission path length of the composite physical transmission signal of the equipment control command transmission signal and the status feedback signal;

[0132] The transmission delay of equipment control command transmission signals to the crusher, mill and hydrocyclone is determined based on the spatial distance in the equipment physical layout parameters.

[0133] The response hysteresis of the crusher performing mechanical actions, the response hysteresis of the mill performing hydraulic adjustments, and the response hysteresis of the hydrocyclone performing pressure adjustments are determined based on the inherent response delay of the equipment.

[0134] The transmission delay and response hysteresis are superimposed on the time nodes required by the hierarchical action instructions to generate a coordinated instruction that defines the start time and duration.

[0135] S4.3: When the collaborative instruction is completed, the synchronous acquisition of full-granularity distributed data and equipment energy consumption data is triggered.

[0136] Specifically, when the end time of all stratification action commands is reached, a full-size distribution data acquisition command is sent to the vibrating screen analyzer, and an energy consumption data acquisition command is sent to the electricity meter and power sensor at the same time.

[0137] Full-granularity distributed data at the same timestamp is bound and stored with device energy consumption data.

[0138] S5. Compare the actual full-granularity distribution data and equipment energy consumption data with the operating parameters. When any data exceeds the preset prediction threshold, construct a time delay cause-effect graph, locate the faulty equipment, and output an equipment health report.

[0139] S5.1: Compare the actual full-grained distribution data and equipment energy consumption data with the operating parameters. When any data exceeds the corresponding preset prediction threshold, the time window is divided based on the time of exceeding the limit.

[0140] Specifically, the prediction deviation between the actual full-scale distribution data and the operating parameters is calculated item by item, and the absolute prediction deviation between the equipment energy consumption data and the target value in the operating parameters is also calculated.

[0141] When any deviation value exceeds the corresponding preset prediction threshold, it is marked as an over-limit moment. Based on the over-limit moment, the time window is divided by tracing back a preset time and extending forward a preset time.

[0142] It should be noted that the corresponding preset prediction thresholds are set based on the allowable range of process fluctuations and the safe operation boundary of equipment in historical production data. The example values ​​are 5% for the actual full-scale distribution data deviation threshold and 10% for the equipment energy consumption data deviation threshold. The example values ​​for the forward tracing preset duration and the backward extension preset duration are 5 minutes.

[0143] S5.2: Based on the pre-stored physical layout parameters of the equipment, analyze the disturbance propagation characteristics of the excessive deviations between the equipment within each time window, and generate the dynamic correlation strength between the equipment;

[0144] Specifically, the upstream and downstream relationships between equipment are determined based on the spatial distance and material flow direction in the pre-stored physical layout parameters of the equipment.

[0145] Within the time window, if the time point when the upstream equipment exhibits an excessive deviation is earlier than the time point when the downstream equipment exhibits an excessive deviation, the direction of disturbance propagation is determined to be from upstream to downstream.

[0146] Based on the proportional relationship between the severity of the deviations of downstream equipment and the severity of the deviations of upstream equipment, and combined with the effect of spatial distance, the dynamic correlation strength between equipment is generated.

[0147] S5.3: Using devices as nodes and the dynamic correlation strength between devices as edge weights, combined with pre-stored signal transmission delay parameters between devices, a directed weighted graph with delay labels is generated as a delay causal graph.

[0148] Specifically, the crusher, mill, and hydrocyclone are treated as independent nodes. Based on the dynamic correlation strength between the devices, directed edges are drawn between upstream and downstream device nodes where there is a disturbance propagation direction, and the dynamic correlation strength between the devices is marked as the weight value of the corresponding directed edge.

[0149] The pre-stored signal transmission delay parameters between devices are called, and a transmission delay mark is added to each directed edge to form a complete directed weighted graph structure containing nodes, weighted directed edges, and delay marks, which serves as a delay causal graph.

[0150] It should be noted that the pre-stored signal transmission delay parameter between devices refers to the fixed time delay required for the composite physical transmission signal of device control command transmission signal and status feedback signal, which is pre-stored in the physical layout of the devices, to be transmitted from the upstream device to the downstream device. It includes the transmission time of the composite physical transmission signal of device control command transmission signal and status feedback signal in the physical line and the inherent processing delay of the device interface.

[0151] S5.4: Starting from the edge with high correlation strength in the time delay causal graph, dynamically expand the three-layer adjacent device nodes, aggregate the impact of neighborhood disturbances through time decay characteristics, and generate the device root cause health score.

[0152] Specifically, edges whose edge weights in the time-delay causal graph exceed a preset correlation strength threshold are selected as the starting point for expansion;

[0153] The first layer consists of devices that directly connect the two ends of an edge that exceeds a preset association strength threshold.

[0154] The second layer consists of devices that are directly connected to the device nodes in the first layer.

[0155] The third layer consists of devices that are directly connected to the device nodes in the second layer.

[0156] A time decay effect is applied based on the time delay distance between each device node and the starting point within the three layers. The closer the time delay distance, the higher the influence weight of the node, and the farther the time delay distance, the lower the influence weight of the node.

[0157] The current deviation values ​​of all device nodes within the three layers are superimposed to form the comprehensive disturbance impact.

[0158] Based on the current deviation value of the equipment and the overall disturbance impact, a root cause health score of the equipment is generated.

[0159] It should be noted that the preset correlation strength threshold is set based on historical disturbance propagation data analysis, and the example value is 0.6.

[0160] S5.5: Starting from a device with a low health score, trace back the transmission path along the edge with high correlation strength, generate the path topology, and output a device health report.

[0161] Specifically, the device node with a root cause health score lower than the preset confidence threshold is used as the backtracking starting point. In the time delay causal graph, the backtracking is traced back to the source device node along the incoming edge with the edge weight greater than the preset correlation strength threshold, forming a transmission path chain.

[0162] The device nodes in the transmission path chain are sorted according to the transmission direction and connected with directed arrows to generate the transmission path topology.

[0163] Output a structured table containing three columns: device name, device root cause health score, and transmission path topology, as a device health report.

[0164] S6. Based on the equipment health report, use generative adversarial networks to generate adversarial samples and inject them into the digital twin model. Reconstruct the crushing and grinding process parameters to complete the self-calibration of the digital twin model and update the pre-stored typical ore samples.

[0165] S6.1: Extract root cause health score and transmission path score from equipment health report, generate health feature vector, and generate ore gene map perturbation vector through generative adversarial network;

[0166] It should be noted that the pre-training process of the generative adversarial network is as follows: A training dataset is formed by collecting historical ore gene map vectors and corresponding device health report data; a generator network and a discriminator network are constructed. The generator network takes the health feature vectors from the device health report as input and outputs a fake ore gene map perturbation vector; the discriminator network takes the real ore gene map vector or the fake perturbation vector as input and outputs the probability of true / false discrimination; the generator network and discriminator network are trained alternately until the discriminator network can no longer distinguish between real and fake vectors; the parameters of the generator network are fixed to complete the pre-training, thus obtaining the generative adversarial network.

[0167] Specifically, the device root cause health score value recorded in each line of the device health report is read, and the number of arrow connectors in the transmission path topology string of the same line is parsed as the transmission path score value.

[0168] The device root cause health score and transmission path score of a single device are combined into a two-dimensional vector as a health feature vector;

[0169] The health feature vector is input into the generative adversarial network generator to generate a perturbation vector with the same dimension as the ore gene map vector, which is then used as the ore gene map perturbation vector.

[0170] S6.2: Inject the ore gene map perturbation vector into the digital twin model to reconstruct the crushing and grinding process parameters, complete the self-calibration of the digital twin model, and update the pre-stored typical ore samples.

[0171] Specifically, the perturbation vector of the ore gene map is superimposed with the current ore gene map vector to generate the perturbation vector of the ore gene map;

[0172] The disturbed ore gene map vector is input into the digital twin model to refit the physical constraints of the crushing and grinding process, and the reconstructed crushing and grinding process parameters are output to complete the self-calibration of the digital twin model.

[0173] The reconstructed crushing and grinding process parameters are bound to the corresponding ore gene map vectors, covering the original parameter records of pre-stored typical ore samples in the historical database, thereby realizing the update of pre-stored typical ore samples.

[0174] This embodiment also provides a simulation and prediction system for the crushing and grinding process in a metal mine, including: a data acquisition module, a ore sample matching module, a steady-state simulation module, an execution instruction module, a fault tracing module, and a model optimization module; the data acquisition module is used to collect full-size distribution data, ore composition data, and ore grade, transmit them to edge computing nodes for data cleaning, extract key feature parameters, and generate ore gene map vectors; the ore sample matching module is used to calculate the Euclidean distance between the ore gene map vector and pre-stored typical ore samples. If the Euclidean distance exceeds a preset mutation threshold, the system searches the historical database to match the most similar ore sample and selects the corresponding crushing and grinding process parameters; the steady-state simulation module is used to compare the crushing and grinding process parameters with real-time collected crusher current and mill power data. The system combines the following modules: a steady-state simulation calculation of the crushing and grinding process, outputting operating parameters and generating control commands; an execution command module, used to adjust the crusher discharge port width, mill feed water flow rate, and hydrocyclone feed pressure in real time according to the control commands, while simultaneously collecting actual full-size distribution data and equipment energy consumption data; a fault tracing module, used to compare the actual full-size distribution data and equipment energy consumption data with the operating parameters, constructing a time-delay causal graph when any data exceeds a preset prediction threshold, locating the faulty equipment, and outputting an equipment health report; and a model optimization module, used to generate adversarial samples using a generative adversarial network based on the equipment health report, injecting them into the digital twin model, reconstructing the crushing and grinding process parameters to complete the self-calibration of the digital twin model, and updating pre-stored typical ore samples.

[0175] This embodiment also provides a computer device applicable to the simulation and prediction method for metal mine grinding process, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the simulation and prediction method for metal mine grinding process proposed in the above embodiment.

[0176] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0177] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the simulation and prediction method for the metal mine grinding process as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0178] In summary, this invention enhances the adaptability of the crushing and grinding process through a dynamic gene matching mechanism. When the Euclidean distance between the ore gene map vector and a typical ore sample exceeds a preset mutation threshold, a historical database search is automatically activated to match the most similar ore sample and migrate its crushing and grinding parameters, achieving second-level parameter adaptation. Based on the weighted geodesic distance calculation using Riemannian manifold projection, and combined with dynamic weight adjustment based on quartz proportion differences, local manifold structure changes in the ore gene are accurately captured. Through collaborative decision-making using parameter migration confidence and energy optimization rate thresholds, equipment overload caused by mismatches is avoided. This ensures the stability of the crushing and grinding process under conditions of ore property mutations, providing a reliable parameter basis for steady-state optimization.

[0179] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method of simulating and predicting a metal mine comminution circuit, the method comprising: Comprising, ​ Collecting full particle size distribution data, ore composition data and ore grade, transmitting to edge computing node for data cleaning, extracting key feature parameters, generating ore gene atlas vector, the specific steps are as follows, Apply median filter algorithm to full particle size distribution data to remove noise, perform Gaussian standardization on ore composition data, perform integrity verification and outlier rejection on ore grade data, and generate structured data set; From the structured data set, extract key feature parameters, and dynamically weight the full particle size distribution data in the structured data set to generate a fused data vector; Integrate the full particle size distribution data in the fused data vector through a multi-scale frequency domain integral kernel, generate multi-scale frequency domain feature components, calculate the entropy weight factor according to the coupling relationship between quartz content and pyrite content, and fuse it with the multi-scale frequency domain feature components to generate a fusion result; Superimpose the contribution value of the fusion result on the crushing strength to generate the ore gene atlas vector; Calculate the Euclidean distance between the ore gene atlas vector and the pre-stored typical ore sample, if the Euclidean distance exceeds the pre-set mutation threshold, retrieve the most similar ore sample in the historical database, and select the corresponding crushing and grinding process parameters; Combine the corresponding crushing and grinding process parameters with the real-time collected crusher current and mill power data to perform steady-state simulation calculation of the crushing and grinding process, output operation parameters, and generate control instructions, the specific steps are as follows, Fuse the corresponding crushing and grinding process parameters with the real-time collected crusher current and mill power data to construct a tensor state space model, and output a state space matrix; Based on the state space matrix, minimize the device running entropy increase rate as the constraint target, perform tensor product optimization calculation in the Riemannian manifold space to generate the optimal control law; Perform entropy increase rate gradient penalty on the optimal control law, output the operation parameters and calculate the energy optimization rate, if the energy optimization rate exceeds the pre-set optimization threshold, generate the control instruction; According to the control instruction, real-time adjust the crusher discharge port width, mill water flow and cyclone feed pressure, and collect actual full particle size distribution data and equipment energy consumption data; Compare the actual full particle size distribution data and equipment energy consumption data with the operation parameters, if any data exceeds the pre-set prediction threshold, construct a time delay causal diagram, locate the fault equipment, and output the equipment health report; According to the equipment health report, generate an adversarial sample using a generative adversarial network to inject into a digital twin model, reconstruct the crushing and grinding process parameters to complete the self-calibration of the digital twin model, and update the pre-stored typical ore sample.

2. The metal mine comminution circuit simulation and prediction method of claim 1, wherein: The key feature parameters include quartz content, pyrite content and crushing strength.

3. The metal mine comminution circuit simulation and prediction method of claim 1, wherein: The specific steps of selecting the corresponding crushing and grinding process parameters are as follows, Project the ore gene atlas vector to the Riemannian manifold tangent space to generate a tangent space coordinate vector; Based on the tangent space coordinate vector and the tangent space coordinate vector of the pre-stored typical ore sample, adjust the weight of each dimension dynamically according to the difference in quartz content, calculate the weighted geodesic distance; When the weighted geodesic distance exceeds the pre-set mutation threshold, retrieve the ore sample in the same manifold neighborhood as the current tangent space vector in the historical database, select the optimal matching sample, and extract the crushing and grinding process parameters; According to the current ore sample and the optimal matching sample, the parameter migration confidence is calculated by the weighted geodesic distance, if the parameter migration confidence is lower than the pre-set confidence threshold, an alarm is triggered, otherwise the corresponding crushing and grinding process parameters are selected.

4. The metal mine comminution circuit simulation and prediction method of claim 1, wherein: The actual full particle size distribution data and the equipment energy consumption data are collected simultaneously, and the specific steps are as follows, The control instruction is decomposed into three types of operations, i.e., high-frequency pulse action, medium-frequency continuous adjustment and low-frequency steady-state maintenance, the real-time crusher current and mill power are dynamically allocated to execute priority, and layered action instructions are output; Based on the layered action instructions, the pre-stored device physical layout parameters are called, the inherent response delay of the device is combined, the collaborative instructions defining the starting time and duration are generated; When the collaborative instructions are completed, the synchronous collection of the full particle size distribution data and the equipment energy consumption data is triggered.

5. The metal mine comminution circuit simulation and prediction method of claim 1, wherein: The time-delay causal diagram is constructed, and the specific steps are as follows, The actual full particle size distribution data and the equipment energy consumption data are compared with the operation parameters, and when any data exceeds the corresponding pre-set prediction threshold, the time window is divided based on the overrun time; According to the pre-stored device physical layout parameters, the disturbance propagation characteristics of the overrun deviation in each time window among the devices are analyzed, and the dynamic correlation strength among the devices is generated; Taking the devices as nodes and the dynamic correlation strength among the devices as edge weights, and combining the pre-stored signal transmission delay parameters among the devices, a directed weighted graph with time delay markers is generated as the time-delay causal diagram.

6. The metal mine comminution circuit simulation and prediction method of claim 1, wherein: The device health degree report is output, and the specific steps are as follows, Taking the high correlation strength edges in the time-delay causal diagram as the starting point, the three-layer adjacent device nodes are dynamically expanded, the neighborhood disturbance influence is aggregated through the time decay characteristics, and the device root cause health degree score is generated; Starting from the low health degree score device, the conduction path is backtracked along the high correlation strength edges, the path topology is generated, and the device health degree report is output.

7. The metal mine comminution circuit simulation and prediction method of claim 1, wherein: The pre-stored typical ore sample is updated, and the specific steps are as follows, The root cause health degree score and the conduction path score are extracted from the device health degree report, the health feature vector is generated, the ore gene map disturbance vector is generated through the generative adversarial network, the crushing and grinding process parameters are reconstructed by injecting the ore gene map disturbance vector into the digital twin model, the self-calibration of the digital twin model is completed, and the pre-stored typical ore sample is updated. It comprises a data collection module, a matching ore sample module, a steady-state simulation module, an execution instruction module, a fault tracing module and a model optimization module; 8. A metal mine crushing and grinding circuit simulation and prediction system based on the metal mine crushing and grinding circuit simulation and prediction method according to any one of claims 1 to 7, characterized in that: The data collection module is used for collecting full particle size distribution data, ore composition data and ore grade, transmitting to an edge computing node for data cleaning, extracting key feature parameters, and generating an ore gene map vector; The matching ore sample module is used for calculating the Euclidean distance between the ore gene map vector and the pre-stored typical ore sample, if the Euclidean distance exceeds the pre-set mutation threshold, the most similar ore sample is searched in the historical database, and the corresponding crushing and grinding process parameters are selected; The steady-state simulation module is used for combining the crushing and grinding process parameters with the real-time collected crusher current and mill power data, executing steady-state simulation calculation of the crushing and grinding process, outputting operation parameters, and generating control instructions; The execution instruction module is used for adjusting the crusher discharge port width, the mill water flow and the cyclone feed pressure in real time according to the control instructions, and simultaneously collecting actual full particle size distribution data and equipment energy consumption data. ​ The fault tracing module is configured to compare the actual whole grain level distribution data and the equipment energy consumption data with the operation parameters, construct a time delay causal diagram when any data exceeds a preset prediction threshold, locate a fault equipment, and output an equipment health degree report. The model optimization module is configured to inject an adversarial sample into a digital twin model by using a generative adversarial network according to the equipment health degree report, reconstruct a crushing process parameter to complete self-calibration of the digital twin model, and update a pre-stored typical ore sample.

Images