A large model optimization method for mineral processing with mechanism model as constraint

Abstract

This invention discloses a large-scale mineral processing model optimization method with a mechanistic model as constraints, belonging to the field of mineral processing process optimization technology. The method includes: collecting full-process operation data of crushing, grinding and classification, flotation, and thickening / dewatering to construct a current operating condition baseline; generating a process intent sequence based on the current operating condition baseline and production targets, and inputting it into a unified mechanistic constraint model to form a process executable boundary; recursively generating a constraint migration chain based on the process executable boundary, and determining the migration and recovery states to form a first disturbance recovery chain and a second disturbance recovery chain; constructing a candidate control action set based on the process executable boundary, constraint migration chain, and disturbance recovery chain, and combining it with the process intent sequence to generate a unique execution action chain from the large-scale mineral processing model and execute it. This invention achieves synchronous constraints between upstream regulation and downstream acceptance, hierarchical identification of disturbance absorption levels, and collaborative optimization of full-process control actions.

Description

Technical Field

[0001] This invention relates to the field of mineral processing optimization technology, and in particular to a large-scale mineral processing model optimization method with mechanistic models as constraints. Background Technology

[0002] Gold ore beneficiation typically involves continuous processes such as crushing, grinding and classification, flotation, and thickening and dewatering. Significant material transfer, particle size evolution, and state coupling relationships exist between these processes. Current technologies often employ online monitoring, process modeling, expert rules, and optimized control methods to comprehensively adjust indicators such as grade, recovery rate, throughput, and energy consumption, thereby improving the continuous operation and production stability of the beneficiation process.

[0003] However, existing methods still have two main shortcomings when it comes to full-process collaborative optimization: First, most methods do not adequately express the constraints on the relationship between upstream and downstream processes, making it difficult to simultaneously reflect the downstream acceptable boundary when generating adjustment actions; Second, most methods do not adequately characterize the transmission and absorption mechanism of disturbances between adjacent processes, which can easily lead to inconsistencies between local optimization and overall process stability when cross-process linkage control is implemented. Summary of the Invention

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

[0005] Therefore, this invention provides a large-scale mineral processing model optimization method with mechanistic model as constraint to solve the problems of insufficient expression of upstream and downstream process bearing constraints and insufficient characterization of disturbance transmission across processes in the existing technology.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for optimizing a large-scale mineral processing model with a mechanistic model as constraints. The method includes: collecting and preprocessing operational data from the entire process of crushing, grinding, classification, flotation, and thickening / dewatering to construct a current operating condition baseline; generating a sequence of process intentions arranged in the order of process succession based on the current operating condition baseline and production targets, and inputting this sequence into a unified mechanistic constraint model to form a process executable boundary; generating a constraint migration chain by recursively extrapolating along the upstream and downstream relationships based on the process executable boundary; determining the migration and recovery states based on the current operating condition baseline to form a first disturbance recovery chain and a second disturbance recovery chain; constructing a candidate control action set based on the process executable boundary, constraint migration chain, and disturbance recovery chain; and combining this with the process intention sequence, performing cross-process linkage reasoning and screening through the large-scale mineral processing model to generate a unique execution action chain and execute it.

[0007] As a preferred embodiment of the large-scale mineral processing model optimization method with mechanistic model as constraint described in this invention, the construction of the current operating condition base includes: performing unified time-scale processing, outlier screening, and data binding on the collected full-process operation data of crushing, grinding and classification, flotation, and thickening and dewatering, and integrating the key equipment operating status data according to the process sequence to construct the current operating condition base for the current control cycle.

[0008] As a preferred embodiment of the large-scale mineral processing model optimization method with mechanistic model as constraint described in this invention, the process intent sequence includes: reading the production target corresponding to the current control cycle, and expanding the process state quantities corresponding to crushing, grinding and classification, flotation, and thickening and dewatering in the current working condition base according to the process sequence; determining the target acceptance state according to the production target corresponding to each process, and determining the allowable change direction and process intent intensity value of each process by combining the deviation of the actual state quantity of each process from the target acceptance state; and connecting the allowable change direction and process intent intensity value of each process in series according to the process sequence to form a process intent sequence.

[0009] As a preferred embodiment of the large-scale mineral processing model optimization method based on the mechanism model as a constraint described in this invention, the unified mechanism constraint model includes: constructing a unified mechanism constraint model using a piecewise linear transfer mapping model; the piecewise linear transfer mapping model includes: establishing a transfer mapping relationship from the output state quantity of the upstream process to the input state quantity of the downstream process based on historical transfer samples, and correcting the transfer mapping relationship through boundary correction parameters; the boundary correction parameters include the reversible margin, the degree of downstream transfer compression, the degree of downstream sensitivity amplification, and the equipment adjustability margin.

[0010] As a preferred embodiment of the large-scale mineral processing model optimization method based on the mechanistic model as a constraint described in this invention, the formation of the process executable boundary includes: determining the allowable change direction and allowable change magnitude of each process within the current control cycle based on the actual state quantity, target acceptance state quantity, stable operating range, reversible margin, and equipment adjustability margin of each process; determining the allowable duration of each process within the current control cycle based on the state change rate of adjacent time periods, downstream acceptance compression degree, and downstream sensitivity amplification degree; and combining the allowable change direction, allowable change magnitude, allowable duration, and downstream acceptance constraints of each process to form the process executable boundary.

[0011] As a preferred embodiment of the large-scale mineral processing model optimization method based on the mechanism model as a constraint described in this invention, the generation of the constraint migration chain includes: determining the direction of output change based on the allowable change direction; calculating the executable change amount within the current control cycle based on the state change rate and allowable duration; and taking the allowable change range as the final change amount when the executable change amount is greater than the allowable change range; superimposing the final change amount along the allowable change direction with the current value of the output amount to obtain the executable output value at the end of the control cycle; inputting the executable output value into a unified mechanism constraint model to obtain the corresponding downstream predicted acceptance state; and connecting the downstream predicted acceptance states corresponding to each process in series according to the process sequence to form a constraint migration chain.

[0012] As a preferred embodiment of the large-scale mineral processing model optimization method with mechanistic model as constraint as described in this invention, the determination of migration state includes: when the predicted acceptance state in the constraint migration chain is between the upper limit of the corresponding target acceptance state and the lower limit of the target acceptance state, and is within the stable operating range, the corresponding constraint migration element is marked as a stable migration state; when the predicted acceptance state is between the upper limit of the corresponding target acceptance state and the lower limit of the target acceptance state, but exceeds the stable operating range, the corresponding constraint migration element is marked as a compressed migration state; when the predicted acceptance state exceeds the upper limit of the target acceptance state or is lower than the lower limit of the target acceptance state, the corresponding constraint migration element is marked as a exceeded migration state.

[0013] As a preferred embodiment of the large-scale mineral processing model optimization method based on the mechanism model as a constraint according to the present invention, the formation of the first disturbance recovery chain and the second disturbance recovery chain includes: for constrained migration elements in the migration stable state and migration compression state, calculating the deviation of the predicted acceptance state from the target acceptance state, and calculating the current remaining absorbable interval width of the downstream process; using the ratio of the current remaining absorbable interval width to the deviation as the disturbance recovery criterion value; when the disturbance recovery criterion value is greater than or equal to 1, marking the corresponding constrained migration element as fully recovered; when the disturbance recovery criterion value is greater than 0 and less than 1, marking the corresponding constrained migration element as limited recovery; when the predicted acceptance state is in the migration over-limit state, or the current remaining absorbable interval width of the downstream process is zero, marking the corresponding constrained migration element as unrecoverable; connecting the constrained migration elements in the fully recovered state in process order to form the first disturbance recovery chain; and connecting the constrained migration elements in the limited recovery state in process order to form the second disturbance recovery chain.

[0014] As a preferred embodiment of the large-scale mineral processing model optimization method with mechanistic model as constraint described in this invention, the construction of the candidate control action set includes: generating candidate actions based on the adjustable parameters of each process, and screening the candidate actions in combination with the process executable boundary, constraint migration chain and disturbance recovery chain of the corresponding process; retaining the candidate actions that meet the boundary constraints and do not lead to migration exceeding the limit or unrecoverable behavior, and dividing them into priority candidate actions and limited candidate actions according to the recovery status to form a candidate control action set.

[0015] As a preferred embodiment of the large-scale mineral processing model optimization method based on the mechanism model as a constraint according to the present invention, the step of generating and executing a unique action chain includes: inputting the process intent sequence and the candidate control action set into the large-scale mineral processing model; the large-scale mineral processing model performing cross-process linkage combination reasoning on the candidate actions in the candidate control action set according to the order and intensity of the process intent elements of each process in the process intent sequence to generate a full-process control action chain; performing constraint migration consistency verification and disturbance recovery closure verification on the full-process control action chain; and determining the full-process control action chain that has passed the constraint migration consistency verification and disturbance recovery closure verification as the unique action chain and executing it.

[0016] The beneficial effects of this invention are as follows: by using a unified mechanism constraint model and constraint migration chain, the effect of synchronous constraint between upstream regulation and downstream acceptance is achieved; by using migration state, recovery state and disturbance recovery chain, the effect of hierarchical identification of the degree of disturbance absorption is achieved; by using the cross-process linkage reasoning of candidate control action set and mineral processing large model, the effect of coordinated optimization of control actions throughout the entire process is achieved. Attached Figure Description

[0017] 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.

[0018] Figure 1 This is a flowchart of a large-scale mineral processing model optimization method with mechanistic models as constraints.

[0019] Figure 2 This is a flowchart showing the process intent sequence and the process executable boundary.

[0020] Figure 3 A flowchart for constraining the formation of migration chains and disturbance recovery chains.

[0021] Figure 4 A flowchart for constructing a candidate set of control actions and generating a unique execution action chain.

[0022] Figure 5 A comparative data chart for predicting the acceptance status. Detailed Implementation

[0023] 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.

[0024] 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.

[0025] 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.

[0026] Reference Figures 1-5 This is one embodiment of the present invention, which provides a method for optimizing a large-scale mineral processing model with a mechanistic model as a constraint, comprising the following steps: S1. Collect and preprocess the entire process operation data of crushing, grinding and classification, flotation and thickening and dewatering to construct the current working condition basis.

[0027] Sampling points are set up at the following locations: coarse crushing feed belt, crushing discharge belt, mill feed inlet, mill discharge inlet, hydrocyclone feed pipeline, hydrocyclone overflow pipeline, flotation roughing cell, cleaning cell, reagent addition point, thickener feed well, thickener underflow pipeline, and concentrate and tailings sampling locations.

[0028] The online data acquisition includes at least the following: raw ore feed rate, crushed ore discharge particle size, mill current, mill power, mill feed rate, hydrocyclone feed pressure, slurry concentration, hydrocyclone overflow particle size, flotation slurry pH, aeration rate, foam image characteristics, reagent addition rate, thickener mud layer interface height, underflow concentration, and underflow rate. Simultaneously, key equipment operating status data is acquired, including at least the crushing equipment start / stop status, mill load status, slurry pump operating status, flotation agitation status, and thickener drive status. Grade analysis data, particle size analysis data, and moisture analysis data are also obtained as testing data.

[0029] It should be noted that, in this embodiment, the foam image features preferably include the average gray level of the foam, the rate of change of foam texture, the proportion of connected area of ​​the foam, and the frequency of foam bursting.

[0030] Using the plant control clock as a unified time reference, the online data collected at each location of crushing, grinding and classification, flotation and thickening dewatering are rearranged onto the same time axis according to the sampling period (e.g., 10 seconds). For data with a sampling frequency higher than the sampling period, the median value within the time window is taken as the representative value of the current period. For data with a sampling frequency lower than the sampling period, time interpolation is performed based on the two most recent valid sampling points to obtain a unified time-stamped data sequence.

[0031] Furthermore, validity screening was performed on each variable.

[0032] Specifically, for continuous variables such as raw ore feed rate, mill power, hydrocyclone feed pressure, slurry concentration, flotation pH, and underflow concentration, several adjacent (e.g., 3) sample values ​​are read before and after the current moment, and the local variation amplitude is calculated. When the current sample value jumps abruptly compared to the median level of the adjacent sampling interval, and the jump does not correspond to the equipment start / stop status, valve switching status, or manual sampling operation record at the same location, the sample value is marked as an outlier, and the outlier is replaced by an adjacent valid sample value. For foam image features and equipment operating status data, the online image recognition results and equipment status return values ​​at the corresponding moment are used as the verification basis. When the image recognition results are inconsistent with the on-site return status, the moment data is marked as pending verification and is not used as the main judgment data for the current cycle.

[0033] Read the sampling time, sampling location, and sampling batch number corresponding to each piece of test data, and backfill them to the same time axis and the same location identifier; when the return time of the test data is later than the online acquisition time, use the sampling time as the belonging time and write the test data into the operation record of the corresponding process location; for multiple pieces of test data returned from the same sampling location within a test cycle, form a test sequence according to the order of sampling time, and bind the test sequence with the online operation record of the corresponding time period to obtain the full-process fusion record of the online operation data and the test data.

[0034] Online monitoring data, key equipment operating status data, monitoring data, and full-process integrated records are integrated in the order of processes to construct the current operating condition base for the current control cycle.

[0035] The current operating condition base includes at least the effective variable values, anomaly markers, verification markers, laboratory test binding results, and location connection relationships for each process location within the current control cycle.

[0036] It should be noted that the preferred detection characteristics include the grade of the ore fed into the mill, the particle size of the slurry fed into the flotation mill, the grade of the concentrate, the grade of the tailings, and the solids content of the underflow; the positional relationship refers to the correspondence between the crusher discharge position and the mill feed position, the mill discharge position and the hydrocyclone feed position, the hydrocyclone overflow position and the flotation feed position, and the flotation tailings position and the thickener feed position.

[0037] S2. Generate a sequence of process intentions arranged in the order of process succession based on the current working conditions and production targets, and input them into a unified mechanism constraint model to form the process executable boundary.

[0038] Furthermore, the current operating condition base corresponding to the current control cycle is read, and the production target corresponding to the control cycle is read.

[0039] Among these, production targets include at least the concentrate grade target, recovery rate target, processing volume target, unit energy consumption target, and equipment stable operation target.

[0040] The process state quantities corresponding to crushing, grinding and classification, flotation and thickening and dewatering in the current working condition base are expanded in the process sequence to form the process state sequence of the current control cycle.

[0041] Among them, the process parameters include: crushing discharge particle size and crushing capacity corresponding to the crushing process; mill power, mill feed rate, hydrocyclone feed pressure, slurry concentration and hydrocyclone overflow particle size corresponding to the grinding and classification process; slurry pH, aeration rate, reagent addition rate, concentrate grade and tailings grade corresponding to the flotation process; and mud interface height, underflow concentration and underflow flow rate corresponding to the thickening and dewatering process.

[0042] Furthermore, based on the process state sequence, the target acceptance state is determined for each process.

[0043] Specifically, for the crushing process, the target particle size entering the mill is used as the target acceptance state for crushing output; for the grinding and classification process, the target particle size entering the flotation process and the target power utilization level of the mill are used together as the target acceptance state for grinding and classification output; for the flotation process, the target concentrate grade and tailings loss control requirements are used together as the target acceptance state for flotation output; for the thickening and dewatering process, the lower limit of the allowable underflow concentration of the next process and the emission stability requirements are used together as the target acceptance state for thickening and dewatering output.

[0044] The adjustment directions of each process are connected in series according to the process sequence of crushing, grinding and classification, flotation and thickening and dewatering to obtain the process intention sequence.

[0045] The process intent sequence consists of multiple process intent elements arranged in process order; each process intent element includes the allowed direction of change for the corresponding process and the process intent intensity value.

[0046] To characterize the strength of the technological intent of each process, in this embodiment, the first... The process intention intensity value is calculated for each process step, expressed as follows: ; in, For the first The intensity value of the process intention of each step, For process numbering, This is the start time of the current control cycle. This is the end time of the current control cycle. For continuous time variables within the current control cycle, For the first Each process is at a certain time The corresponding actual state quantities, For the first The target acceptance status quantity corresponding to each process. For the first The fluctuation scale of the actual state quantity of each process in the historical stable operation sample.

[0047] It should be noted that, These correspond sequentially to the crushing process, grinding and classifying process, flotation process, and thickening and dewatering process. For the crushing process, the actual state quantity is the crushed discharge particle size; for the grinding and classification process, the actual state quantity is the hydrocyclone overflow particle size; for the flotation process, the actual state quantity is the tailings grade; and for the thickening and dewatering process, the actual state quantity is the underflow concentration. The target acceptance state quantity is obtained by reading the stable running segment data that matches the current production target from the historical production database, and extracting the center value of the corresponding process output state quantity from the stable running segment based on the production target corresponding to the current control cycle. By filtering from the historical production database, the first A stable operating sample of each process that continuously meets production targets without equipment alarms or manual intervention is obtained by extracting the actual state quantity sequence of the process and calculating the standard deviation of the actual state quantity sequence relative to the center value.

[0048] Furthermore, based on the process intent intensity value and the positional relationship between the actual state quantity of each process and the upper limit and lower limit of the target acceptance state, the permissible change direction of each process is determined.

[0049] Specifically, when the actual state quantity is higher than the upper limit of the target acceptance state, the allowed change direction is determined as the convergence adjustment direction; when the actual state quantity is lower than the lower limit of the target acceptance state, the allowed change direction is determined as the compensation adjustment direction; when the actual state quantity is between the upper limit of the target acceptance state and the lower limit of the target acceptance state, the process intention intensity value of the corresponding process and the stability judgment threshold are compared.

[0050] Furthermore, when the process intention intensity value is not higher than the stability judgment threshold, the allowed change direction is determined as the stability maintenance direction; when the process intention intensity value is higher than the stability judgment threshold, the judgment is made according to the deviation direction of the actual state quantity relative to the target acceptance state quantity. When it is higher than the target acceptance state quantity, it is determined as the convergence adjustment direction, and when it is lower than the target acceptance state quantity, it is determined as the compensation adjustment direction.

[0051] It should be noted that the stability judgment threshold is obtained by screening out stable operation samples in the historical production database that continuously meet the production target for the corresponding process and have not experienced equipment alarms or manual intervention, extracting the process intent intensity value sequence corresponding to the stable operation sample, and using the average value in the process intent intensity value sequence as the stability judgment threshold. The preferred value range is [0.8, 1.2].

[0052] It should be noted that the upper limit and lower limit of the target acceptance status are determined by first screening out stable operating periods in the historical production database that have continuously achieved the target concentrate grade, recovery rate, and processing volume without equipment alarms or manual intervention, and then calculating the maximum and minimum values ​​of the corresponding process status quantities within the stable operating periods, and then adjusting them in conjunction with the production target of the current control cycle.

[0053] In this embodiment, the unified mechanism constraint model is constructed using a piecewise linear transfer mapping model. It is based on the transfer mapping relationship between the output state variables of the upstream process and the input state variables of the downstream process, established according to historical transfer samples, and the mapping relationship is corrected by boundary correction parameters.

[0054] Specifically, a piecewise linear connection mapping model from single input to single output is established for each pair of adjacent processes. When there are multiple input state variables downstream, multiple independent piecewise linear connection mapping models from single input to single output are established. Each connection mapping model is fixedly divided into three segments: low connection segment, stable connection segment, and high connection segment. The segmentation is based on the value distribution of the output state variables of the corresponding upstream processes in historical stable operation samples. First, the output state variables of the upstream processes are sorted from smallest to largest and divided into three groups according to the number of variables, corresponding to the low connection segment, stable connection segment, and high connection segment, respectively. A linear connection mapping relationship is established independently for each segment.

[0055] Among them, the boundary correction parameters include the reversible margin, the degree of compression of downstream acceptance, the degree of amplification of downstream sensitivity, and the adjustable margin of the equipment.

[0056] It should be noted that the adjustable margin refers to the remaining range that can be adjusted between the actual state quantity of the current process and the upper limit or lower limit of the target acceptance state; the downstream acceptance compression degree refers to the proportion by which the original allowable adjustment range of the current process is compressed due to the current tight operation of the downstream process; the downstream sensitivity amplification degree refers to the degree to which the state quantity of the downstream process is amplified when the state quantity of the upstream process changes by a unit; the equipment adjustable margin refers to the remaining proportion that the current equipment can still be safely adjusted within the current control cycle. It is calculated by reading the rated operating upper limit, safety protection lower limit, current actual operating value and currently occupied adjustment amount of the corresponding equipment within the current control cycle, calculating the remaining adjustable space of the equipment from the safety boundary along the allowable change direction, and taking the ratio of the remaining adjustable space to the full adjustable range of the equipment as the equipment adjustable margin, with a value range of [0,1].

[0057] During training, the unified mechanism constraint model first inputs the upstream process output state variables from historical acceptance sample pairs into the unified mechanism constraint model to calculate the corresponding downstream predicted acceptance state. The downstream predicted acceptance state is then compared with the actual downstream acceptance state in the historical samples to obtain the acceptance deviation. When the acceptance deviation exceeds the error threshold, the model is corrected item by item in the following order: acceptance mapping coefficient, downstream acceptance compression degree, downstream sensitivity amplification degree, target acceptance state variable, stable operating range, callback margin, and equipment adjustable margin. The step size of each correction is set to a fixed correction ratio (e.g., 5%) of the current value of the corresponding parameter. At the same time, the target acceptance state variable is always between the upper limit and the lower limit of the target acceptance state, the downstream acceptance compression degree is always between 0 and 1, and the downstream sensitivity amplification degree and equipment adjustable margin are always not less than 0. Correction stops when the acceptance deviation is not higher than the error threshold.

[0058] Specifically, when the acceptance deviation is positive, if the output state quantity of the upstream process is higher than its target acceptance state quantity, the acceptance slope coefficient is reduced; if the output state quantity of the upstream process is not higher than its target acceptance state quantity, the acceptance intercept coefficient is reduced; at the same time, the downstream acceptance compression degree is increased, and the downstream sensitivity amplification degree is reduced. When the acceptance deviation is negative, if the output state quantity of the upstream process is higher than its target acceptance state quantity, the acceptance slope coefficient is increased; if the output state quantity of the upstream process is not higher than its target acceptance state quantity, the acceptance intercept coefficient is increased; at the same time, the downstream acceptance compression degree is reduced, and the downstream sensitivity amplification degree is increased.

[0059] It should be noted that the acceptance mapping coefficient is obtained by substituting the upstream process output state quantity and the corresponding downstream process input state quantity in the historical acceptance sample pair into the piecewise linear acceptance mapping model and performing least squares fitting. The acceptance slope coefficient is preferably in the range of [0,3], and the acceptance intercept coefficient is preferably in the range between the minimum and maximum values ​​of the historical sample of the corresponding downstream process input state quantity. The acceptance mapping coefficient is used to characterize the transmission relationship of the change of the upstream process output state quantity to the change of the downstream process input state quantity. The acceptance slope coefficient is used to represent the response amplitude of the downstream process input state quantity when the upstream process output state quantity changes by one unit. The acceptance intercept coefficient is used to represent the benchmark acceptance level of the downstream process input state quantity when the upstream process output state quantity is at the benchmark level.

[0060] It should be noted that the error threshold is obtained by calculating the absolute error sequence between the predicted downstream acceptance status and the actual downstream acceptance status in historical stable operation samples, removing abnormal jump errors, and taking the maximum value of the remaining absolute errors. The preferred value range is 2% to 8% of the historical sample average of the input status quantity of the corresponding downstream process.

[0061] Specifically, when the upstream process is a crushing process, the downstream predicted acceptance status is the mill feed particle size status; when the upstream process is a grinding and classification process, the downstream predicted acceptance status is the flotation particle size status and the flotation slurry concentration status; and when the upstream process is a flotation process, the downstream predicted acceptance status is the thickener feed concentration status and the feed flow rate status.

[0062] Furthermore, the current operating condition base, process intention sequence, and process intention intensity values ​​of each process are jointly input into the unified mechanism constraint model.

[0063] The unified mechanism constraint model first determines the permissible change direction of each process within the current control cycle based on the actual state quantities, process intention sequence, and process intention intensity value of each process. Then, based on the target acceptance state quantity, stable operating range, reversible margin, and equipment adjustability margin corresponding to each process, it determines the permissible change magnitude of each process within the current control cycle. Finally, based on the state change rate of adjacent time periods within the current control cycle, the degree of downstream acceptance compression, and the degree of downstream sensitivity amplification, it determines the permissible duration of each process within the current control cycle.

[0064] Specifically, the allowable duration is determined based on the current permissible change direction of the process, establishing the corresponding acceptance boundary; the actual state quantities of the process in the previous and next time periods within the current control cycle are read, and the state change per unit time is calculated as the basic state change rate; the basic state distance is determined based on the state distance between the current actual state quantity and the corresponding acceptance boundary; the basic state distance is compressed and corrected according to the downstream acceptance compression degree, and the value obtained by multiplying the basic state distance by one and subtracting the downstream acceptance compression degree is used as the corrected state distance; the basic state change rate is amplified and corrected according to the downstream sensitivity amplification degree, and the data obtained by multiplying the basic state change rate by the downstream sensitivity amplification degree is used as the corrected state change rate; when the corrected state change rate is not zero, the ratio of the corrected state distance to the corrected state change rate is used as the allowable duration of the process within the current control cycle; when the corrected state change rate is zero, the upper limit of the current control cycle is determined as the allowable duration.

[0065] It should be noted that when the same process corresponds to only one downstream acceptance constraint, the allowable change direction, allowable change magnitude, and allowable duration output by the corresponding piecewise linear acceptance mapping model are directly used as the process executable boundary of the process; when the same process corresponds to multiple downstream acceptance constraints, the allowable change direction, allowable change magnitude, and allowable duration corresponding to each downstream acceptance constraint are calculated separately, and the result with the smallest allowable change magnitude, the shortest allowable duration, and consistent with the current allowable change direction is determined as the final process executable boundary of the process.

[0066] The permissible direction of change, permissible range of change, permissible duration, and downstream acceptance restrictions of this process are combined to form the process executable boundary.

[0067] S3. Based on the process executable boundary, the upstream and downstream connection relationship is recursively deduced to generate a constraint migration chain. Combined with the current operating condition, the migration status and recovery status are determined to form the first disturbance recovery chain and the second disturbance recovery chain.

[0068] When the current process is the crushing process, the predicted acceptance status of the grinding and classification process is calculated based on the crushing discharge particle size, allowable change direction, allowable change range, and allowable duration.

[0069] Among them, the predicted acceptance status of the grinding and classification process is the feed particle size of the mill.

[0070] When the current process is grinding and classification, two predicted acceptance states of the flotation process are calculated based on the overflow particle size of the hydrocyclone, the allowable direction of change, the allowable range of change, and the allowable duration.

[0071] Among them, the two predicted acceptance states of the flotation process are the feed particle size and the feed slurry concentration.

[0072] When the current process is flotation, calculate two predicted acceptance states for the thickening and dewatering process based on the tailings grade, allowable change direction, allowable change range, and allowable duration.

[0073] Among them, the two predicted acceptance states of the thickening and dewatering process are the feed concentration and feed flow rate of the thickener.

[0074] It should be noted that the calculation of the predicted acceptance status of the downstream process specifically involves: reading the current value of the output quantity directly related to the downstream acceptance of the current process, and reading the allowable change direction, allowable change magnitude, allowable duration, and state change rate within the current control cycle corresponding to the output quantity; determining the change direction of the output quantity based on the allowable change direction; calculating the executable change quantity within the current control cycle based on the state change rate and allowable duration; and taking the allowable change magnitude as the final change quantity when the executable change quantity is greater than the allowable change magnitude; superimposing the final change quantity along the allowable change direction with the current value to obtain the executable output value at the end of the current control cycle; and inputting the executable output value into the unified mechanism constraint model to obtain the corresponding predicted downstream acceptance status.

[0075] For processes with multiple predicted acceptance states, each predicted acceptance state is compared with its corresponding acceptance boundary, and the comparison data with the most stringent constraint is taken as the final acceptance recursive data for the process.

[0076] Furthermore, the recursive data of each process are connected in sequence according to the process order to obtain a constraint migration chain.

[0077] Each constraint migration element in the constraint migration chain includes at least the current process number, the corresponding downstream process number, the allowed change direction, the allowed change range, the allowed duration, and the corresponding downstream acceptance boundary comparison data.

[0078] Furthermore, a migration status determination is performed on each constraint migration element.

[0079] Specifically, when the predicted acceptance state is between the upper limit of the corresponding target acceptance state and the lower limit of the target acceptance state, and is within the stable operating range, the constraint migration element is marked as a stable migration state.

[0080] When the predicted acceptance state is between the upper limit of the corresponding target acceptance state and the lower limit of the target acceptance state, but has exceeded the stable operating range, the constraint migration element is marked as a migration compression state.

[0081] When the predicted acceptance state exceeds the upper limit of the target acceptance state or falls below the lower limit of the target acceptance state, the constraint migration element is marked as a migration limit violation state.

[0082] It should be noted that the stable operating range is determined by screening stable operating samples from the historical production database that continuously meet production targets for the corresponding processes without equipment alarms, manual intervention, or mismatch between upstream and downstream processes. The process status quantities are sorted from smallest to largest, and samples with a fixed percentage (e.g., 10%) and a fixed percentage (e.g., 10%) are removed. The range between the minimum and maximum values ​​in the remaining samples is then determined as the stable operating range. Removing the fixed percentage is to eliminate a small number of abnormal deviations at both ends of the historical stable samples and retain the main stable samples in the middle as the basis for determining the stable operating range.

[0083] Furthermore, after completing the migration state determination, the recycling state determination is further performed on the constrained migration elements that are in the migration stable state and migration compression state.

[0084] Specifically, the deviation of the predicted acceptance state from the target acceptance state is compared with the current remaining absorbable range of the downstream process to determine whether the deviation can be absorbed in the downstream process; to characterize the absorbability, in this embodiment, the deviation is further analyzed for the first... The process is recursively pushed to the next step. The disturbance recovery criterion value is calculated based on the acceptance results of each process, and the expression is as follows: ; in, For the first The disturbance recovery criterion value corresponding to each process. For the first The current remaining absorbable range width for each process. For the first The process is recursively pushed to the next step. Predicted acceptance status of each process step For the first The target acceptance status quantity corresponding to each process.

[0085] It should be noted that the current remaining absorbable interval width is determined by reading the target acceptance state upper limit, target acceptance state lower limit, and current actual value corresponding to the current actual state quantity of the downstream process, and based on the deviation direction of the predicted acceptance state relative to the target acceptance state quantity. When the predicted acceptance state is higher than the target acceptance state quantity, the distance between the current actual value and the target acceptance state upper limit is determined as the current remaining absorbable interval width. When the predicted acceptance state is lower than the target acceptance state quantity, the distance between the current actual value and the target acceptance state lower limit is determined as the current remaining absorbable interval width.

[0086] when When the value is greater than or equal to 1, the corresponding constraint migration element is marked as fully recycled; when When the value is greater than 0 and less than 1, the corresponding constraint migration element is marked as a limited recovery state; when the predicted acceptance state is already in the migration over-limit state, or the current remaining absorbable range of the downstream process is zero, the corresponding constraint migration element is marked as an unrecoverable state.

[0087] It should be noted that, The significance is to characterize the coverage capacity of the current remaining absorbable range width of the downstream process relative to the current deviation, where, This indicates that the width of the remaining absorbable range downstream is equal to the current deviation, which is a critical state where it can be completely absorbed. Therefore, 1 is used as the boundary between the complete recovery state and the limited recovery state. This indicates that the current remaining absorbable range width downstream is zero, meaning that the downstream has no absorption capacity. Therefore, 0 is taken as the natural boundary between recyclable and non-recyclable.

[0088] When there are multiple predicted acceptance states in the downstream process, the disturbance recovery criterion value corresponding to each predicted acceptance state is calculated, and the minimum value is taken as the final disturbance recovery criterion value of the process.

[0089] Furthermore, the constrained migration elements that are in a fully recycled state are connected in series according to the process sequence to form the first disturbance recycling chain.

[0090] The constrained migration elements in the limited recovery state are connected in series according to the process sequence to form a second disturbance recovery chain.

[0091] For constraint migration elements that are in an unrecoverable state, they are not written into the first and second perturbation recycling chains, but their state flags are retained for the elimination determination of subsequent candidate control action sets.

[0092] S4. Construct a set of candidate control actions based on the process executable boundary, constraint migration chain, and disturbance recovery chain. Combine this with the process intent sequence, perform cross-process linkage reasoning and screening through the large mineral processing model, generate a unique action chain, and execute it.

[0093] Read the open instruction table of the control layer and the list of adjustable parameters of the actuator.

[0094] The control layer open instruction table is used to give the adjustment objects and corresponding adjustment directions that are allowed to be issued within the current control cycle. The actuator adjustable parameter list is used to give the adjustable parameter name, adjustable range and adjustment step size corresponding to each adjustment object. The adjustable parameters corresponding to the crushing process, grinding and classification process, flotation process and thickening and dewatering process are expanded in the process sequence to form the action candidate parameter table.

[0095] Furthermore, based on the action candidate parameter table, single-process candidate actions are generated for each process.

[0096] Specifically, for the crushing process, the crushing discharge particle size adjustment action is generated based on the allowable change direction and allowable change range corresponding to the crushed discharge particle size.

[0097] For the grinding and classification process, based on the overflow particle size of the hydrocyclone, the corresponding slurry concentration, the allowable direction of change, and the allowable range of change, the mill feed rate adjustment action, the mill power adjustment action, the hydrocyclone feed pressure adjustment action, and the water feed rate adjustment action are generated.

[0098] For the flotation process, based on the allowable direction and range of change of tailings grade, pulp pH, aeration rate, and reagent addition rate, pulp pH adjustment action, aeration rate adjustment action, and reagent addition rate adjustment action are generated.

[0099] For the thickening and dewatering process, underflow discharge adjustment actions and flocculant addition adjustment actions are generated based on the allowable change direction and allowable change range of underflow concentration and underflow flow rate.

[0100] For each single-process candidate action, multiple action levels are generated by discretizing the corresponding adjustment step size in the list of adjustable parameters of the actuator, and each action level is recorded as a candidate action.

[0101] Furthermore, a first round of screening is conducted on each candidate action.

[0102] Specifically, each candidate action is compared with the process executable boundary of the corresponding process; when the adjustment direction corresponding to the candidate action is inconsistent with the allowed change direction, the candidate action is eliminated; when the adjustment amount corresponding to the candidate action exceeds the allowed change range, the candidate action is eliminated; when the expected duration corresponding to the candidate action exceeds the allowed duration, the candidate action is eliminated; after the first round of screening, a set of candidate actions within the boundary is obtained.

[0103] Furthermore, a second round of screening is performed on the set of candidate actions within the boundary.

[0104] Specifically, each candidate action is substituted into the corresponding constraint migration chain to calculate the downstream receiving data after the candidate action is passed along the process sequence; when the migration state corresponding to the downstream receiving data is marked as migration limit exceeded, the candidate action is eliminated; when the recovery state corresponding to the downstream receiving data is marked as non-recoverable, the candidate action is eliminated; when the recovery state corresponding to the downstream receiving data is marked as limited recovery, the candidate action is retained and marked as limited candidate action; when the recovery state corresponding to the downstream receiving data is marked as fully recovered, the candidate action is retained and marked as priority candidate action, thus obtaining a set of candidate control actions.

[0105] The candidate control action set includes at least a priority candidate action set and a limit candidate action set.

[0106] In order to characterize the executable priority of candidate actions within the current control cycle, in this embodiment, the ratio of the remaining absorbable interval width of the downstream process corresponding to the candidate action and the deviation of the predicted acceptance state of the downstream process corresponding to the candidate action from the target acceptance state is used as the migration recovery closure criterion value.

[0107] Among them, a migration recovery closure criterion value greater than or equal to 1 indicates that after the current candidate action is taken, the width of the remaining absorbable interval of the downstream process is greater than or equal to the deviation amount; a migration recovery closure criterion value greater than 0 and less than 1 indicates that after the current candidate action is taken, the width of the remaining absorbable interval of the downstream process is less than the deviation amount, but still has some absorption capacity.

[0108] Calculate the migration recovery closure criterion value for each candidate action in the priority candidate action set and the limit candidate action set, and sort them from largest to smallest.

[0109] Furthermore, the process intent sequence and the candidate control action set are input together into the large-scale mineral processing model.

[0110] Among them, the mineral processing large model is based on the order and intensity of the process intent elements of each process in the process intent sequence, and performs combined reasoning on the candidate actions in the candidate control action set according to the process sequence to obtain cross-process linkage action combination data.

[0111] Specifically, from the sorted set of priority candidate actions, the candidate action corresponding to the process intent element of the crushing process is selected as the first action unit; the first action unit is substituted into the constraint migration chain, and the candidate action that is consistent with the process intent element of the grinding and classification process and does not cause mismatch in downstream acceptance is selected as the second action unit; then the candidate actions corresponding to the flotation process and the thickening and dewatering process are determined in sequence; when there is no candidate action that satisfies the complete recovery state in a certain process, the candidate action with the largest migration recovery closure criterion value is selected from the set of limited candidate actions as the alternative action unit; the action units determined according to the process sequence are connected in series to form a full-process control action chain.

[0112] Furthermore, the consistency of constraint migration and the closure of disturbance recovery are checked for the entire process control action chain.

[0113] Specifically, the full-process control action chain is sequentially substituted into the constraint migration chain first, and it is checked whether the predicted承接状态 after the adjustment action of each process acts is always within the corresponding承接 boundary; when the predicted承接状态 corresponding to any process exceeds the承接 boundary, it is determined that the full-process control action chain fails the constraint migration consistency check; the full-process control action chain that passes the constraint migration consistency check is sequentially substituted into the first disturbance recovery chain and the second disturbance recovery chain, and it is checked whether all disturbances transmitted along the process sequence fall into the complete recovery state or the amplitude-limited recovery state, and no unrecoverable state exists; when the disturbance corresponding to any process falls into the unrecoverable state, it is determined that the full-process control action chain fails the disturbance recovery closure check.

[0114] The full-process control action chain that passes both the constraint migration consistency check and the disturbance recovery closure check is retained and determined as the unique execution action chain.

[0115] It should be noted that, in this embodiment, when there are multiple full-process control action chains that pass the double check at the same time, the full-process control action chain with the largest product of the migration recovery closure criterion values of each process is preferentially selected as the unique execution action chain; the reason is that the larger the product of the migration recovery closure criterion values, the more sufficient the overall recoverable space of the full-process control action chain in the downstream processes corresponding to each process, and the lower the risk of承接 mismatch formed after transmission along the process sequence.

[0116] In this embodiment, in order to verify the influence of different control schemes on the stability of downstream承接, a process-based comparative analysis including stable working condition scenarios, tight承接 scenarios and disturbance injection scenarios is constructed based on the same batch of historical operation samples, and three schemes of local adjustment, mechanism constraint and linkage reasoning are respectively executed in each control cycle; as Figure 5 it reflects the change of the predicted承接 state with the control cycle under different schemes; in the figure, the solid lines correspond to the instantaneous predicted承接 states of the three schemes of local adjustment, mechanism constraint and linkage reasoning respectively, and the dashed lines correspond to the rolling mean of each scheme; it can be seen that in the stable working condition stage, each scheme fluctuates around the target interval, but after entering the tight承接 stage and the disturbance injection stage, the fluctuation amplitude of the local adjustment scheme increases significantly, and the instantaneous value is more prone to peak spikes; the mechanism constraint scheme has a certain inhibitory effect on fluctuation; the overall fluctuation of the linkage reasoning scheme is relatively gentler, and the rolling mean is more concentrated, which shows that under the synergistic effect of the unified mechanism constraint model, the constraint migration chain and the candidate control action set, it can more effectively constrain the upstream adjustment behavior and maintain the continuous stability of the downstream承接 state, thereby exhibiting a strong synchronous constraint capability of upstream adjustment and downstream承接.

[0117] It should be noted that, in this embodiment, the training steps of the large-scale mineral processing model are as follows: Continuous full-process operation records are extracted from the historical production database in the order of control cycles. The current operating condition base, process intent sequence, process executable boundary, constraint migration chain, first disturbance recovery chain, second disturbance recovery chain, and candidate control action set for multiple consecutive control cycles are extracted using a sliding window method to construct continuous input samples. The actual execution action chain corresponding to the endpoint of the continuous input samples, along with the concentrate grade, recovery rate, throughput, unit energy consumption, acceptance status, and recovery status after the action execution, are constructed as supervision labels. The samples are divided into training sample intervals, validation sample intervals, and test sample intervals in chronological order to ensure the continuity of the process evolution sequence and the upstream and downstream acceptance sequence during training. After normalizing or encoding each state quantity, boundary quantity, and action quantity in the continuous input samples, they are input into the large-scale mineral processing model. Preferably, the large-scale mineral processing model is constructed using a two-layer gated recurrent network and a fully connected recursive output structure. The optimizer preferably uses an adaptive moment estimation optimization method, with a fixed initial learning rate (e.g., 0.0). 01), and when the validation loss does not decrease for several consecutive training rounds, the learning rate is decreased by a decay coefficient (e.g., 0.5); the working condition feature mapping, process intention mapping, boundary constraint mapping, transfer propagation mapping, recovery closure mapping and candidate action mapping are executed in sequence to form the fusion state representation corresponding to each process; then the action recursive decoding is executed in the process order of crushing, grinding and classification, flotation and thickening and dewatering to output the benchmark candidate action chain; the benchmark candidate action chain is compared with the supervision label, and the action selection loss, acceptance consistency loss, recovery closure loss and process result loss are calculated respectively, and the various losses are combined into the total training loss; backpropagation is performed on the mineral processing large model according to the total training loss to calculate the gradient of each trainable parameter and iteratively update it; after completing each round of training, the validation loss is calculated by calling the corresponding sample in the validation sample interval. When the validation loss continues to decrease, training continues. When the validation loss no longer decreases within a fixed number of rounds or the acceptance consistency loss and recovery closure loss increase continuously, training stops, and the model parameters corresponding to the minimum validation loss are retained to obtain the trained mineral processing large model.

[0118] In summary, this invention achieves the effect of synchronous constraint between upstream regulation and downstream acceptance by unifying the mechanism constraint model and constraint migration chain; it achieves the effect of hierarchical identification of the degree of disturbance absorption by migration state, recovery state and disturbance recovery chain; and it achieves the effect of coordinated optimization of the whole process control actions by cross-process linkage reasoning of candidate control action set and mineral processing big model.

[0119] 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 for optimizing a large-scale mineral processing model with mechanistic models as constraints, characterized in that, include: Collect and preprocess operational data from the entire process of crushing, grinding and classification, flotation and thickening and dewatering to construct a current operating condition baseline. Based on the current operating conditions and production targets, a sequence of process intentions arranged in the order of process succession is generated, and a unified mechanism constraint model is input to form the process executable boundary. Based on the process executable boundary, the upstream and downstream acceptance relationship is recursively deduced to generate a constraint migration chain. Combined with the current operating condition, the migration status and recovery status are determined to form the first disturbance recovery chain and the second disturbance recovery chain. Candidate control action sets are constructed based on the process executable boundary, constraint migration chain, and disturbance recovery chain. Combined with the process intent sequence, cross-process linkage reasoning and screening are performed through the mineral processing big model to generate a unique execution action chain and execute it. The generated constraint migration chain includes: The direction of output change is determined based on the allowed direction of change. The executable change amount within the current control cycle is calculated based on the rate of change of state and the allowed duration. When the executable change amount is greater than the allowed change range, the allowed change range is taken as the final change amount. The final change is added to the current value of the output along the allowed change direction to obtain the executable output value at the end of the control cycle; Input the executable output value into the unified mechanism constraint model to obtain the corresponding downstream predicted acceptance status; The downstream predicted acceptance status corresponding to each process is connected in series according to the process sequence to form a constraint migration chain; The determination of the migration status includes: When the predicted acceptance state in the constraint migration chain is between the upper limit of the corresponding target acceptance state and the lower limit of the target acceptance state, and is within the stable operating range, the corresponding constraint migration element is marked as a migration stable state. When the predicted acceptance state is between the upper limit of the corresponding target acceptance state and the lower limit of the target acceptance state, but exceeds the stable operating range, the corresponding constraint migration element will be marked as a migration compression state. When the predicted acceptance state exceeds the upper limit of the target acceptance state or falls below the lower limit of the target acceptance state, the corresponding constraint migration element is marked as a migration limit violation state.

2. The large-scale mineral processing model optimization method with mechanistic model as a constraint as described in claim 1, characterized in that, The construction of the current operating condition base includes: The collected data from the entire process of crushing, grinding and classification, flotation and thickening dewatering are uniformly time-stamped, outlier-screened and bound to detection data. Combined with the operating status data of key equipment, the data are integrated according to the process sequence to construct the current operating condition base for the current control cycle.

3. The large-scale mineral processing model optimization method with mechanistic model as constraint as described in claim 1, characterized in that, The process intent sequence includes: Read the production target corresponding to the current control cycle, and expand the process state quantities corresponding to crushing, grinding and classification, flotation and thickening and dewatering in the current working condition base according to the process sequence. The target acceptance status is determined based on the production target corresponding to each process, and the target acceptance status quantity is extracted from the target acceptance status. Combined with the deviation of the actual status quantity of each process from the target acceptance status, the allowable change direction and process intention intensity value of each process are determined. The permissible direction of change and the intensity value of the process intention for each process are connected in series according to the process sequence to form a process intention sequence.

4. The large-scale mineral processing model optimization method with mechanistic model as constraint as described in claim 3, characterized in that, The unified mechanism constraint model includes: A unified mechanism constraint model is constructed using a piecewise linear succession mapping model; The piecewise linear transfer mapping model includes establishing a transfer mapping relationship from the output state quantity of the upstream process to the input state quantity of the downstream process based on historical transfer samples, and correcting the transfer mapping relationship through boundary correction parameters. The boundary correction parameters include the reversible margin, the degree of compression downstream, the degree of amplification downstream, and the adjustable margin of the equipment.

5. The large-scale mineral processing model optimization method with mechanistic model as a constraint as described in claim 1, characterized in that, The executable boundaries of the formation process include: Based on the actual state quantity, target acceptance state quantity, stable operating range, available margin for correction, and equipment adjustability of each process, determine the allowable direction and allowable magnitude of change for each process within the current control cycle. Based on the rate of change of state in adjacent time periods, the degree of compression of downstream acceptance, and the degree of amplification of downstream sensitivity, the allowable duration of each process in the current control cycle is determined. The permissible direction of change, permissible range of change, permissible duration of change, and downstream acceptance restrictions for each process are combined to form the process executable boundary.

6. The large-scale mineral processing model optimization method with mechanistic model as a constraint as described in claim 1 or 5, characterized in that, The formation of the first disturbance recovery chain and the second disturbance recovery chain includes: For constrained migration elements in the stable migration state and the compressed migration state, calculate the deviation of the predicted acceptance state from the target acceptance state, and calculate the current remaining absorbable interval width of the downstream process. The ratio of the current remaining absorbable interval width to the deviation is used as the disturbance recovery criterion value; When the disturbance recovery criterion value is not less than 1, the corresponding constraint migration element is marked as fully recovered. When the disturbance recovery criterion value is greater than 0 and less than 1, the corresponding constraint migration element is marked as a limited recovery state; When the predicted acceptance status is in the migration limit exceedance state, or the current remaining absorbable interval width of the downstream process is zero, the corresponding constraint migration element will be marked as unrecoverable. The constrained migration elements that are in a fully recycled state are connected in series according to the process sequence to form the first disturbance recycling chain. The constrained migration elements in the limited recovery state are connected in series according to the process sequence to form a second disturbance recovery chain.

7. The large-scale mineral processing model optimization method with mechanistic model as constraint as described in claim 1, characterized in that, The construction of the candidate control action set includes: Candidate actions are generated based on the adjustable parameters of each process, and the candidate actions are screened by combining the process executable boundary, constraint migration chain and disturbance recovery chain of the corresponding process. Candidate actions that satisfy boundary constraints and do not lead to migration exceeding limits or becoming unrecoverable are retained and divided into priority candidate actions and limit candidate actions according to their recovery status, forming a set of candidate control actions.

8. The large-scale mineral processing model optimization method with mechanistic model as constraint as described in claim 7, characterized in that, The generation and execution of a unique chain of actions includes: The process intent sequence and the candidate control action set are input together into the large mineral processing model; Based on the order and intensity of the process intent elements in each process in the process intent sequence, the mineral processing big model performs cross-process linkage combination reasoning on the candidate actions in the candidate control action set to generate a full-process control action chain. Constraint migration consistency verification and disturbance recovery closure verification are performed on the entire process control action chain. The entire process control action chain, which is verified by constraint migration consistency check and disturbance recovery closure check, is determined as the unique execution action chain and executed.

Images