All-time-space multimode optimization method for beneficiation process

Abstract

The application discloses a whole-time-space multi-modal optimization method for a beneficiation process, and relates to the technical field of multi-modal data fusion, and comprises the following steps: collecting original perception data, extracting ore grindability characteristic values, target particle size cumulative proportion, foam surface moving speed, stable foam area proportion and tailing gold grade characteristic values; according to the dynamic delay length of materials between positions, the characteristic values are aligned in cross-process time and space, a process state vector is generated, and the current belonging process state interval is determined; combining the process state vector, the process state interval and the current executed working condition, a plurality of candidate control action sequences are generated, and dominant risk relief screening is performed, the reduction amplitude of each candidate control action sequence on the current dominant risk is compared, and the final control action sequence is determined and executed. The method realizes accurate correlation of cross-process multi-modal data, and can improve process state recognition accuracy and control decision pertinence.

Description

Technical Field

[0001] This invention relates to the field of multimodal data fusion technology, and in particular to a spatiotemporal multimodal optimization method for mineral processing. Background Technology

[0002] In the field of mineral processing, gold ore beneficiation typically consists of multiple continuous processes, including feeding, grinding and classification, flotation, and product discharge, with each process closely coupled through slurry flow. Traditional process optimization and condition monitoring methods mainly rely on independently collected physical sensor data (such as particle size, concentration, and flow rate) from each process, as well as human on-site experience for control. With the development of industrial internet and machine vision technology, online sensing methods based on multimodal data such as foam images and ore images are gradually being applied, providing a richer information foundation for analyzing the beneficiation status from a holistic process perspective.

[0003] However, when integrating cross-process and multimodal data, the aforementioned conventional methods typically assume that the data from each monitoring point are naturally aligned in time, neglecting the significant dynamic delays caused by changes in transport distance and flow rate between different processes. This spatiotemporal mismatch can lead to biases in the characterization of the same batch of ore, such as incorrectly attributing upstream feed fluctuations to downstream flotation results. Furthermore, conventional methods often use single indicators or fixed rules to generate control strategies, making it difficult to effectively distinguish the primary and secondary contradictions in the current process (such as tailings loss versus concentrate contamination), resulting in weak risk targeting in control actions and affecting the overall coordination of optimization decisions. Summary of the Invention

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

[0005] Therefore, this invention provides a spatiotemporal multimodal optimization method for mineral processing to solve the problems of spatiotemporal mismatch of multimodal sensing data across processes leading to distorted process state representation and unclear response of control actions to process-dominant risks in existing technologies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a multi-modal optimization method for a mineral processing process across time and space, comprising: Step S1, collecting raw sensing data on the feed location, grinding and classification location, flotation location, and concentrate and tailings discharge locations during gold ore beneficiation, and extracting ore grindability characterization values, cumulative proportion of target particle size, froth surface movement speed, stable froth area proportion, and tailings gold grade characterization values; Step S2, performing cross-process spatiotemporal alignment of the characterization values ​​based on the dynamic delay time of the material at each location, generating a process state vector, and determining the current process state interval; Step S3, generating multiple candidate control action sequences based on the process state vector, process state interval, and currently executed conditions, and screening for dominant risk mitigation; Step S4, comparing the reduction magnitude of each candidate control action sequence on the current dominant risk, determining the final control action sequence, and executing it.

[0007] As a preferred embodiment of the all-time-space multimodal optimization method for the mineral processing process described in this invention, the original sensing data includes ore image data, feed mass flow rate data, feed particle size data, mill active power data, slurry concentration data, classification product particle size data, foam image sequence, tailings detection data, and concentrate detection data.

[0008] As a preferred embodiment of the all-time-space multimodal optimization method for the mineral processing process described in this invention, the specific steps for extracting the ore grindability characterization value, the cumulative proportion of the target particle size, the foam surface movement speed, the stable foam area proportion, and the tailings gold grade characterization value are as follows: Extracting the ore grindability characterization value based on feed mass flow rate data, feed particle size data, and mill active power data; extracting the cumulative proportion of the target particle size based on the particle size data of the classification products; extracting the foam surface movement speed based on the foam texture displacement between adjacent foam image frames in the foam image sequence; extracting the stable foam area proportion based on the boundary continuity and area change rate of the foam region at adjacent times; and extracting the tailings gold grade characterization value based on the tailings gold grade detection values ​​within multiple consecutive acquisition cycles.

[0009] As a preferred embodiment of the multimodal optimization method for the mineral processing process described in this invention, the cross-process spatiotemporal alignment includes: dividing the feed position to the grinding and classification position, the grinding and classification position to the flotation position, and the flotation position to the concentrate and tailings discharge position into a first process segment, a second process segment, and a third process segment; obtaining the effective volume and real-time volumetric flow rate of each process segment; calculating the basic delay time of each process segment based on the ratio of the effective volume and real-time volumetric flow rate of each process segment; correcting the basic delay time based on the load deviation of each process segment to obtain the dynamic delay time of each process segment; and, based on the dynamic delay time and using the time corresponding to the tailings gold grade characterization value as a benchmark, performing backtracking matching on the ore grindability characterization value, the cumulative proportion of the target particle size, the foam surface movement speed, and the stable foam area proportion to obtain a cross-process spatiotemporal alignment feature sequence.

[0010] As a preferred embodiment of the all-temporal-space multimodal optimization method for the mineral processing process described in this invention, the generation of the process state vector includes: obtaining the ore grindability characterization value, the cumulative proportion of the target particle size, the foam surface movement speed, the stable foam area proportion, and the tailings gold grade characterization value from the cross-process spatiotemporal aligned feature sequence; generating a particle size dissociation state quantity based on the ratio of the cumulative proportion of the target particle size to the target particle size cumulative proportion benchmark value; generating a foam separation state quantity based on the combination relationship of the foam surface movement speed, the stable foam area proportion, and the foam speed benchmark value; generating a tailings risk state quantity based on the combination relationship of the tailings gold grade characterization value and the particle size dissociation state quantity; generating an inclusion risk state quantity based on the ratio of the stable foam area proportion and the foam surface movement speed; and combining the particle size dissociation state quantity, the foam separation state quantity, the tailings risk state quantity, and the inclusion risk state quantity to form the process state vector.

[0011] As a preferred embodiment of the all-time-space multimodal optimization method for the mineral processing process described in this invention, the determination of the current process state interval includes: comparing the tailings risk state quantity and the tailings risk threshold in the process state vector, and comparing the particle size dissociation state quantity and the particle size lower limit threshold; when the tailings risk state quantity is greater than the tailings risk threshold and the particle size dissociation state quantity is less than the particle size lower limit threshold, the current process state interval is determined to be a tailings run-dominated interval; comparing the inclusion risk state quantity and the inclusion risk threshold, and comparing the foam separation state quantity and the foam separation lower limit threshold; when the inclusion risk... When the risk state quantity is greater than the inclusion risk threshold and the foam separation state quantity is less than the foam separation lower limit threshold, the current process state interval is determined to be the inclusion-dominant interval. The particle size dissociation state quantity and the particle size upper limit threshold are compared, and the inclusion risk state quantity and the inclusion risk threshold are also compared. When the particle size dissociation state quantity is greater than the particle size upper limit threshold and the inclusion risk state quantity is greater than the inclusion risk threshold, the current process state interval is determined to be the over-grinding-dominant interval. When the determination conditions for the tail-running-dominant interval, the inclusion-dominant interval, and the over-grinding-dominant interval are not met, the current process state interval is determined to be the stable operation interval.

[0012] As a preferred embodiment of the all-time-space multimodal optimization method for the mineral processing process described in this invention, the generation of multiple candidate control action sequences includes: reading the feed setpoint, water replenishment setpoint, mill load setpoint, classification pressure setpoint, collector addition setpoint, frother addition setpoint, slurry pH setpoint, and aeration setpoint within the current control cycle to form a vector of currently executed operating conditions; determining the current dominant risk type and the correction direction of each control parameter based on the process state interval; calculating the action correction amount based on the deviation of the state quantity in the process state vector from the threshold; superimposing the action correction amounts at different amplitude ratios to generate a candidate control action sequence; and synchronously correcting the relevant control parameters in the candidate control action sequence according to the parameter linkage relationship to obtain a set of candidate control action sequences.

[0013] As a preferred embodiment of the all-time-space multimodal optimization method for the mineral processing process described in this invention, the dominant risk mitigation screening includes: establishing the dominant risk response relationship of each candidate control action sequence based on the current dominant risk type; calculating the dominant risk prediction value after the execution of each candidate control action sequence based on the parameter change of each candidate control action sequence relative to the currently executed working condition vector; calculating the dominant risk mitigation amount corresponding to each candidate control action sequence based on the current dominant risk deviation and the dominant risk prediction value; comparing the dominant risk mitigation amount of each candidate control action sequence with the dominant risk mitigation lower limit threshold, and retaining candidate control action sequences whose dominant risk mitigation amount is not lower than the dominant risk mitigation lower limit threshold; calculating the total action change of each retained candidate control action sequence relative to the currently executed working condition vector, and removing candidate control action sequences whose total action change is higher than the action disturbance upper limit, thereby obtaining a set of effective candidate control action sequences.

[0014] As a preferred embodiment of the all-time-space multimodal optimization method for mineral processing described in this invention, the step of comparing the reduction magnitude of each candidate control action sequence on the current dominant risk includes: determining the current dominant risk benchmark based on the current process state interval, and calculating the dominant risk reduction magnitude corresponding to each candidate control action sequence; sorting each candidate control action sequence according to the dominant risk reduction magnitude; when the difference in the dominant risk reduction magnitude of multiple candidate control action sequences does not exceed the equivalent threshold, the candidate control action sequence with the smallest total disturbance is taken as the highest priority.

[0015] As a preferred embodiment of the all-time-space multimodal optimization method for the mineral processing process described in this invention, the step of determining and executing the final control action sequence includes: determining the candidate control action sequence with the highest priority as the final control action sequence; decomposing the final control action sequence into control commands; setting a maximum allowable change amount within a single control cycle for each control command; when the parameter change amount of the control command exceeds the maximum allowable change amount, executing to the intermediate value of the maximum allowable change amount, and continuing to approach the final control action sequence in the next control cycle.

[0016] The beneficial effects of this invention are as follows: by calculating the dynamic delay time of each process segment and backtracking to match the upstream characterization value, the spatiotemporal alignment of multimodal data across processes is achieved, improving the accuracy of process state characterization; by constructing a process state vector and determining the dominant risk interval, the automatic identification of the dominant risk type in the mineral processing process is realized, providing a basis for targeted regulation; by generating candidate regulation action sequences based on the dominant risk type and deviation degree and screening them according to the risk mitigation amount, the directional response of regulation actions to the current dominant risk is realized, improving the targeting of decision-making. 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 multi-modal optimization method for the entire time and space of the mineral processing process.

[0019] Figure 2 This is a schematic diagram of cross-process spatial and temporal alignment.

[0020] Figure 3 This is a schematic diagram of process state vector generation and process state interval determination.

[0021] Figure 4 This is a schematic diagram illustrating the selection and execution of action sequences.

[0022] Figure 5 This is a comparison chart of continuous sampling of gold grade in tailings. 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 As one embodiment of the present invention, this embodiment provides a multi-modal optimization method for a mineral processing process across all time and space, comprising the following steps: S1. Collect raw sensing data on the feeding location, grinding and classification location, flotation location, and concentrate and tailings discharge location during the gold ore beneficiation process, and extract the ore grindability characterization value, the cumulative proportion of target particle size, the foam surface movement speed, the stable foam area proportion, and the tailings gold grade characterization value.

[0027] Furthermore, data collection points were set up along the gold ore beneficiation process.

[0028] The data collection locations are, in order, the ore feeding location, the grinding and classification location, the flotation location, the concentrate discharge location, and the tailings discharge location, and a uniform time stamp is added to the data collected at each location.

[0029] At the feed location, ore image data, feed mass flow rate data, and feed particle size data are collected before the ore enters the grinding process.

[0030] Preferably, an industrial camera is fixedly installed above the feed conveyor belt to continuously acquire images of the surface of the feed ore.

[0031] A belt scale is installed below the feed conveyor belt to continuously collect the feed mass flow rate per unit time.

[0032] An online particle size detection device is installed at the corresponding position of the feed conveyor belt to continuously collect particle size distribution data of the feed ore.

[0033] Ultimately, raw sensing data is collected at the ore location to characterize the characteristics of the raw gold ore currently entering the beneficiation process.

[0034] In the grinding and classification stage, data on mill power, mill current, slurry concentration, and particle size of the classification products are collected.

[0035] Preferably, the active power and operating current of the mill are read directly from the mill drive circuit.

[0036] A concentration meter is installed at the ore feed pipeline of the hydrocyclone to continuously collect the slurry concentration.

[0037] An online particle size detection device is installed at the overflow outlet of the hydrocyclone to continuously collect the particle size distribution curve of the slurry after grinding and classification.

[0038] Finally, raw sensing data is collected at the grinding and classification locations to characterize the current grinding and classification status.

[0039] In the flotation phase, a sequence of images of foam in the flotation cell is acquired.

[0040] Preferably, an industrial camera is fixedly installed above the liquid surface in the coarse selection tank to continuously acquire foam image frames at a constant sampling frequency.

[0041] The images have a defined time interval between adjacent frames, providing the original image basis for extracting the foam surface movement speed and stabilizing the foam area ratio.

[0042] Specifically, concentrate and tailings test data were collected at the concentrate discharge location and the tailings discharge location, respectively.

[0043] Preferably, automatic samplers are installed on the concentrate discharge pipeline and the tailings discharge pipeline respectively, and concentrate samples and tailings samples are obtained at fixed time intervals.

[0044] Among them, concentrate testing data is used to establish a link between concentrate and tailings results.

[0045] Tailings testing data is used to extract gold grade characterization values ​​from tailings.

[0046] In this embodiment, in order to ensure that data from different locations correspond on the same timeline, time markers are uniformly assigned to the data output from each acquisition location.

[0047] Thus, a set of original sensory data is formed at a unified moment.

[0048] The raw sensing data set includes ore image data, feed mass flow rate data, feed particle size data, mill active power data, slurry concentration data, classification product particle size data, foam image sequence, tailings detection data, and concentrate detection data.

[0049] Furthermore, after completing the collection of raw sensing data, the grindability characterization values ​​of the ore are extracted from the data of the ore feeding location and the grinding and classification location.

[0050] The grindability characterization value is used to characterize the degree of difficulty or ease of grinding of the gold ore currently entering the beneficiation process under continuous on-site production conditions to achieve the desired grinding effect.

[0051] Specifically, image preprocessing is performed on the ore image data collected at the ore feeding location.

[0052] The image preprocessing includes background removal, ore region segmentation, and ore boundary recognition, in order to remove the belt background and retain only the ore region.

[0053] Subsequently, based on the multiple ore contours obtained from the segmentation, the equivalent particle size of each ore block is calculated, and the particle size distribution of the feed within the current sampling time window is statistically analyzed.

[0054] The median particle size of the feed particle size distribution is defined as the characteristic particle size of the feed at the current moment, denoted as . .

[0055] Simultaneously, the belt scale detection results within the same sampling time window are read to obtain the feed mass flow rate, which is recorded as... .

[0056] The active power of the mill within the same sampling time window at the grinding and classification location is read and denoted as... .

[0057] In this embodiment, in order to eliminate the direct amplification effect of different feed flow rates on the power value, the mill specific power consumption corresponding to a unit feed mass is calculated.

[0058] The specific power consumption of the mill is expressed as: ; in, Indicates time The mill's specific power consumption.

[0059] In this embodiment, the grindability characterization value of ore is defined as the ratio of mill specific power consumption to feed characteristic particle size. That is, under the current production conditions, the higher the specific power consumption level corresponding to ore with a unit characteristic particle size, the more difficult the ore is to grind.

[0060] The grindability characterization value of ore is expressed as: ; in, Indicates time The grindability characterization value of the ore.

[0061] Combining the two equations, the ore grindability characterization value is expressed as: ; It should be noted that the higher the ore grindability value, the higher the energy required to achieve the desired crushing and refining effect under the current on-site grinding conditions, and the more difficult the ore is to grind; the lower the ore grindability value, the easier the ore is to grind.

[0062] Furthermore, after extracting the ore grindability characterization values, the cumulative proportion of the target particle size is extracted using the particle size data of the graded products collected at the grinding and classification locations.

[0063] The cumulative proportion of the target particle size is used to characterize the proportion of particles in the slurry after grinding and classification that are within the target particle size range, so as to reflect whether the current grinding and classification results meet the particle size conditions required for subsequent flotation.

[0064] Specifically, the particle size distribution curve output by the online particle size detection device at the overflow outlet of the hydrocyclone is read.

[0065] Based on the lower and upper limits of the target particle size, the cumulative mass of particles whose particle size falls within the lower and upper limits of the target particle size range is calculated. Then, the total mass of all detected particles at the current sampling time is calculated, and the ratio of the cumulative mass of particles to the total mass of all detected particles is taken as the cumulative proportion of the target particle size.

[0066] It should be noted that the lower limit of the target particle size is determined by grinding, particle size screening, and monomer liberation detection of the target gold ore sample, which corresponds to the minimum particle size at which the gold-bearing minerals begin to achieve an effective liberation rate. The obtained minimum particle size is set as the lower limit of the target particle size, and the value range is usually 0.019~0.038mm. The upper limit of the target particle size is determined by grinding, particle size screening, and flotation recovery of the target gold ore sample, which corresponds to the maximum particle size under the condition of ensuring effective liberation of the gold-bearing minerals and no decrease in flotation recovery rate. The maximum particle size is set as the upper limit of the target particle size, and the value range is 0.074~0.15mm.

[0067] Furthermore, after completing the foam image acquisition, the foam image sequence acquired at the flotation location is analyzed frame by frame to extract the foam surface movement speed.

[0068] Among them, the foam surface movement velocity is used to characterize the average displacement velocity of the foam layer on the surface of the flotation cell per unit time, thereby reflecting the dynamic flow characteristics of the foam layer.

[0069] Specifically, bubble image frames are acquired at two adjacent sampling times.

[0070] Gray-level normalization and noise suppression are performed on each frame of the bubble image to reduce the impact of illumination changes and random noise on recognition; then, bubble texture region recognition is performed on the bubble image to obtain multiple traceable bubble texture blocks in the current frame.

[0071] Furthermore, the same foam texture blocks are matched between two adjacent frames to obtain the displacement of each foam texture block in the image plane.

[0072] In this embodiment, the average displacement length of the foam texture blocks involved in the matching in two adjacent frames is calculated and divided by the time interval between the two frames to obtain the foam surface movement speed.

[0073] It should be noted that the speed of foam surface movement essentially represents the average displacement of the foam layer surface per unit time; the greater the speed of foam surface movement, the faster the foam layer moves; the smaller the speed of foam surface movement, the slower the foam layer moves.

[0074] Furthermore, after extracting the foam surface movement speed, the percentage of stable foam area is extracted from the same foam image sequence.

[0075] Among them, the stable foam area ratio is used to characterize the proportion of the total foam area in the current time window where the boundary is continuous and the area change rate of adjacent frames is not greater than the foam stability threshold, so as to reflect the stability of the foam layer.

[0076] Specifically, the foam region in a single frame of foam image is segmented to obtain the total foam region area of ​​the current frame.

[0077] Subsequently, in two adjacent frames, the rate of change of the foam area is calculated for each foam region.

[0078] For a bubble region, if the current bubble region maintains continuous boundaries between two adjacent frames and the rate of change of area is not greater than the bubble stability threshold, then the current bubble region is determined to be a stable bubble region.

[0079] Sum the areas of all regions identified as stable bubble regions to obtain the area of ​​the stable bubble region.

[0080] It should be noted that the foam stability threshold is determined by annotating and analyzing the foam images continuously collected during the target gold ore flotation process, statistically analyzing the distribution of the area change rate of stable foam regions between adjacent image frames, and setting the upper limit of the area change rate of the samples with stable coverage as the foam stability threshold. The foam stability threshold usually does not have a uniform fixed value. In a preferred embodiment, the foam stability threshold usually ranges from 0.05 to 0.15.

[0081] It should be noted that the larger the proportion of stable foam area, the higher the proportion of foam area in a stable state on the surface of the flotation cell at the current moment; the smaller the proportion of stable foam area, the faster the foam area changes and the lower the stability of the foam layer.

[0082] Furthermore, at the tailings discharge location, an automatic sampler acquires tailings samples at a fixed sampling cycle and detects the gold grade of the tailings at each sampling time.

[0083] In order to reduce the impact of single sampling fluctuations on process analysis, this embodiment performs a moving average processing on the tailings gold grade detection values ​​over multiple consecutive sampling periods to obtain tailings gold grade characterization values.

[0084] Finally, after extracting the ore grindability characterization value, the cumulative proportion of target particle size, the foam surface movement speed, the proportion of stable foam area, and the tailings gold grade characterization value, five types of characteristic quantities were obtained.

[0085] S2. Based on the dynamic delay time of the material at each location, the characterization value is aligned across processes in time and space to generate a process state vector and determine the current process state interval.

[0086] Furthermore, the gold ore beneficiation process is divided into three adjacent process segments.

[0087] Specifically, the first process stage from the ore feeding position to the grinding and classification position.

[0088] The second stage, from the grinding and classification position to the flotation position.

[0089] The third stage, from the flotation position to the concentration and tailings discharge position.

[0090] To ensure that the extracted characterization values ​​correspond to the same batch of materials, the dynamic delay time of each of the three process segments needs to be determined separately.

[0091] In this embodiment, the dynamic delay time of the first process segment is defined as the time it takes for the ore to enter the grinding process from the feeding position and finally form a detection result at the grinding and classification position.

[0092] The dynamic delay time of the second process stage is defined as the time it takes for the slurry to enter the flotation cell after grinding and classification and form foam image features at the flotation position.

[0093] The dynamic delay time of the third process stage is defined as the time it takes for the flotation pulp to move from the flotation position to the concentrate and tailings discharge position and form the test results.

[0094] In this embodiment, the basic delay time of each of the three process segments is calculated by the ratio of effective volume to real-time volumetric flow rate.

[0095] Specifically, the first process segment Second process segment The third process stage The base latency can be expressed as: ; in, Indicates time The Dynamic delay time for each process segment Indicates the first Each process segment corresponds to the effective volume of the equipment and pipelines. Indicates time No. Real-time volumetric flow rate of each process segment Indicates the process segment number, =1,2,3.

[0096] It should be noted that when the effective volume of a process segment is constant, the larger the real-time volumetric flow rate, the shorter the time required for the material to pass through the current process segment; the smaller the real-time volumetric flow rate, the longer the time required for the material to pass through the current process segment. Therefore, the ratio of effective volume to real-time volumetric flow rate can be used to characterize the dynamic delay time of the current process segment online.

[0097] Furthermore, in order to improve the accuracy of dynamic delay calculation.

[0098] In this embodiment, the base delay is also corrected by incorporating a load correction factor.

[0099] The load correction factor is used to reflect the impact of slurry retention, foam accumulation, or load fluctuations inside the equipment on the actual delay.

[0100] Specifically, for the first The corrected dynamic delay time for each process segment is expressed as follows: ; in, Indicates time The The corrected dynamic delay time for each process segment Indicates the first Correction coefficients for each process segment Indicates time The Load deviation of each process segment.

[0101] It should be noted that the load correction factor is... .

[0102] It should be noted that the correction coefficients for the first stage were obtained by simultaneously collecting data on the actual passage time, feed mass flow rate, slurry concentration, and mill current between the feed position and the grinding and classification position under stable production conditions, and then using least squares fitting to minimize the error between the corrected dynamic delay time and the actual passage time. The correction coefficients for the second stage were obtained by simultaneously collecting data on the actual passage time, slurry flow rate, hydrocyclone pressure, and flotation cell level between the grinding and classification position and the flotation position under stable production conditions, and then using least squares fitting to minimize the error between the corrected dynamic delay time and the actual passage time. The correction coefficients for the third stage were obtained by simultaneously collecting data on the actual passage time, slurry flow rate, froth layer thickness, and tailings discharge flow rate between the flotation position and the concentrate and tailings discharge positions under stable production conditions, and then using least squares fitting to minimize the error between the corrected dynamic delay time and the actual passage time.

[0103] It should be noted that the load deviation is defined as the normalized difference between the current operating condition value and the current process segment's baseline operating condition value, used to characterize the degree of deviation of the current operating condition from the stable operating condition.

[0104] Furthermore, after obtaining the dynamic delay duration of each process segment, the characterization values ​​collected at different locations and times are mapped to the same batch of materials, forming a cross-process spatiotemporal alignment result.

[0105] In this embodiment, the gold grade characterization value of the tailings at each time point, based on the tailings discharge location, is used as the result time benchmark for the current batch of materials. Then, the flotation location, grinding and classification location, and feed location characteristics corresponding to the tailings discharge location are traced back upstream in sequence.

[0106] Specifically, the flotation position characterization value corresponding to the tailings gold grade characterization value at the same time is taken from time 1. .

[0107] The grinding and classification position value corresponding to the tailings gold grade characterization value at the same time is taken from time 1. .

[0108] The corresponding feed location value for the tailings gold grade at the same time is taken from time 10. .

[0109] Finally, the corresponding cross-process spatiotemporal alignment feature sequence at the same moment is represented as: ; in, Representation of time Gold grade characterization value of tailings The corresponding cross-process spatiotemporal alignment feature sequence, This represents the upstream ore grindability characterization value corresponding to the current tailings results. This indicates the cumulative percentage of the grinding and classification target particle size corresponding to the current tailings results. This indicates the foam surface movement velocity corresponding to the current tailings result. This indicates the percentage of stable foam area corresponding to the current tailings results. This represents the current gold grade value of the tailings.

[0110] It should be noted that, by aligning the feature sequences across processes in time and space, this embodiment no longer directly combines data collected at different locations at the same time. Instead, it utilizes the dynamic delay time of each process segment to backtrack and pair the upstream characterization values ​​with the current result time, so that the same set of aligned characterization values ​​corresponds as closely as possible to the same batch of gold mineral materials, thereby avoiding data mismatch caused by process delays.

[0111] Furthermore, in this embodiment, all characterization values ​​are stored discretely according to a uniform sampling period. When the backtracking time falls between two sampling times, the characterization value at the corresponding time is obtained by linear interpolation between adjacent times.

[0112] Furthermore, after obtaining the spatiotemporal alignment feature sequence across processes, in order to facilitate a unified state determination of the current mineral processing process, in this embodiment, a process state vector is constructed based on the spatiotemporal alignment feature sequence across processes.

[0113] Among them, the process state vector is used to centrally characterize the dissociation state, foam separation state, tailings risk state, and inclusion risk state of the gold ore beneficiation process at the result time.

[0114] Among them, the particle size dissociation state quantity is obtained by directly normalizing the cumulative proportion of the target particle size through the benchmark value of the cumulative proportion of the target particle size.

[0115] It should be noted that the cumulative percentage benchmark value of the target particle size is obtained by extracting the stable operating average or median value of the cumulative percentage of particles within the target particle size range from the historical classification product particle size data of the gold ore beneficiation production line when it is operating stably and the gold grade of the tailings and the flotation recovery index meet the process requirements.

[0116] It should be noted that when the particle size dissociation state value is greater than 1, it indicates that the cumulative proportion of the target particle size before the current batch of material enters the flotation is relatively high; when the particle size dissociation state value is less than 1, it indicates that the cumulative proportion of the target particle size before the current batch of material enters the flotation is relatively low.

[0117] Among them, the foam separation state quantity is used to simultaneously reflect the motion and stability of the flotation foam layer.

[0118] In this embodiment, the ratio of the foam surface moving speed to the foam speed reference value is coupled with the stable foam area ratio to form a foam sorting state quantity.

[0119] It should be noted that the foam velocity benchmark value is obtained by statistically analyzing historical foam image data of gold mine beneficiation production lines when the recovery rate is stable, the concentrate grade is stable, and the foam layer maintains continuous transport and effective sorting capacity. The foam surface movement speed at each moment is extracted, and the mean or median value of the stable operating range of the foam surface movement speed at each moment is calibrated.

[0120] It should be noted that when both the foam surface movement speed and the proportion of stable foam area are at a reasonable level, the foam sorting state quantity is high; if the foam surface movement speed is too low or the proportion of stable foam area is too low, the foam sorting state quantity will decrease.

[0121] Among them, the tailings risk status quantity is used to characterize the risk of tailings gold loss in the current batch of materials.

[0122] In this embodiment, the cumulative proportion of the target particle size and the gold loss in the tailings are considered to be mutually restrictive.

[0123] The tailings risk state quantity is defined as a combination function of the tailings gold grade characterization value and the grain size dissociation state quantity, that is, the ratio of the tailings gold grade characterization value and the grain size dissociation state quantity is used as the tailings risk state quantity.

[0124] It should be noted that if the gold grade of the tailings increases while the cumulative proportion of the target particle size is low, the tailings risk status will increase significantly; conversely, if the cumulative proportion of the target particle size is high while the gold grade of the tailings is low, the tailings risk status will be small.

[0125] Among them, the inclusion risk state quantity is used to characterize the possibility that the current bubble state will lead to gangue inclusion or concentrate inclusion.

[0126] In this embodiment, a high proportion of stable foam area and a low foam surface movement speed generally indicate increased foam layer retention and a greater risk of inclusion. Therefore, the inclusion risk state quantity is defined as: ; in, Indicates time The mixed risk state quantity.

[0127] It should be noted that when the proportion of stable foam area is high and the foam surface movement speed is low, the amount of mixed risk increases; when the proportion of stable foam area and the foam surface movement speed are in a relatively coordinated state, the amount of mixed risk decreases.

[0128] Furthermore, after obtaining the process state vector, the current process state interval is determined based on the relationship between each state variable and the interval boundary.

[0129] Among them, the process state range is used to characterize the dominant process contradiction type in the current mineral processing process, so as to generate targeted control actions.

[0130] In this embodiment, the current process state range includes at least the tail-running range, the inclusion range, the over-wearing range, and the stable operation range.

[0131] Specifically, when the tailings risk state quantity is greater than the tailings risk threshold and the particle size dissociation state quantity is less than the lower limit threshold of the particle size, the current process state interval is determined to be the tailings run-dominant interval.

[0132] When the inclusion risk state quantity is greater than the inclusion risk threshold and the foam sorting state quantity is less than the lower limit threshold of foam sorting, the current process state interval is determined to be the inclusion-dominant interval.

[0133] It should be noted that the lower limit threshold for particle size distribution is determined by statistically analyzing the cumulative percentage of the target particle size distribution that meets the tailings gold grade control requirements, based on the target particle size range corresponding to the effective liberation of gold-bearing minerals in gold mines and historical online particle size data during stable production. This lower limit threshold is typically 60% to 90% of the baseline value of the cumulative percentage of the target particle size distribution. The lower limit threshold for foam separation is determined by statistically analyzing historical foam images and operating data of the gold mine beneficiation production line under stable recovery and concentrate grade conditions. The minimum state quantity corresponding to the foam layer maintaining continuous transport and effective separation capacity is determined as the lower limit threshold for foam separation. This minimum state quantity is typically 70% to 95% of the average value of the foam separation state quantity under stable operating conditions.

[0134] When the particle size dissociation state quantity is greater than the upper limit threshold of the particle size and the inclusion risk state quantity is higher than the inclusion risk threshold, the current process state interval is determined to be the over-grinding dominant interval.

[0135] When none of the judgment conditions are met, and the tailings risk status quantity, the inclusion risk status quantity, and the foam separation status quantity are all within their respective allowable ranges, the current process status interval is determined to be a stable operating interval.

[0136] Furthermore, to avoid frequent switching of state intervals near the boundaries.

[0137] In this embodiment, continuous operation is preferred. The method of confirming the switching interval is only determined after all sampling periods are consistent.

[0138] That is, only when the same process state range is continuous The current process state interval will only be updated to the process state interval if the corresponding judgment condition is met within each sampling period.

[0139] It should be noted that the example value of m is generally 2 to 5, preferably 3. This value can filter out misjudgments of process state quantities near the interval boundary caused by detection noise and short-term fluctuations, avoid frequent switching of state intervals, and will not significantly reduce the response speed of operating condition switching due to excessively long continuous confirmation cycles.

[0140] S3. Based on the process state vector, process state interval, and currently executed operating conditions, generate multiple candidate control action sequences and perform primary risk mitigation screening.

[0141] Furthermore, read the operating condition data that has been issued and is in the execution state within the current control cycle to form the currently executed operating condition.

[0142] The currently executed operating condition is used to characterize the actual execution level of each process parameter in the gold mine beneficiation production line at the current moment, serving as a benchmark for generating candidate control action sequences.

[0143] In this embodiment, the currently executed operating conditions include the feed setting value, water replenishment setting value, mill load setting value, classification pressure setting value, collector addition setting value, frother addition setting value, slurry pH setting value, and aeration volume setting value.

[0144] Furthermore, after obtaining the currently executed operating condition vector, the dominant risk type within the current control cycle is determined based on the current process state interval obtained from the judgment, and the action correction direction for the dominant risk type is established.

[0145] In this embodiment, when the current process state interval is the tailings-dominated interval, it indicates that the main contradiction in the current gold ore beneficiation process is that the tailings risk state is too high and the particle size liberation state is too low.

[0146] The dominant risk type in the process state range where tailings run is the dominant range is defined as tailings loss risk.

[0147] Actions to mitigate tailings loss risk include reducing feed shock, increasing effective liberation, and enhancing the uplift capacity of valuable minerals.

[0148] At the process parameter level, the control actions specifically include lowering the feed setting value, increasing the water replenishment setting value or the mill load setting value, adjusting the classification pressure setting value, in order to increase the cumulative proportion of the target particle size, and increasing the collector addition setting value and the aeration amount setting value.

[0149] When the current process state range is dominated by inclusions, it indicates that the main contradiction in the current gold ore beneficiation process is that the amount of inclusion risk is too high and the amount of foam separation is too low.

[0150] The dominant risk type where the process state range is dominated by inclusions is defined as inclusion risk.

[0151] The actions to regulate the risk of entrainment specifically include reducing the degree of foam layer retention, suppressing non-selective entrainment, and improving the dynamic coordination of the foam layer.

[0152] At the process parameter level, the control actions specifically include reducing the foaming agent addition setting value, adjusting the aeration volume setting value, adjusting the slurry pH setting value, and simultaneously adjusting the collector addition setting value as needed.

[0153] When the current process state range is dominated by over-grinding, it indicates that the main contradiction in the current gold ore beneficiation process is that the amount of particle size liberation is too high and the amount of inclusion risk is also too high.

[0154] The dominant risk type when the process state range is dominated by over-grinding is defined as the risk of fine mud inclusion caused by over-grinding.

[0155] Actions to control the risk of fine mud inclusion caused by over-grinding include reducing the tendency for fine particles to continue to form and reducing the entrainment effect of the foam layer on fine mud.

[0156] At the process parameter level, the control actions specifically include reducing the mill load setting, adjusting the water supply setting and the stage pressure setting, and reducing the foaming agent addition setting.

[0157] When the current process state range is a stable operating range, it means that all current state quantities are within the allowable range.

[0158] The dominant risk type where the process state range is a stable operating range is defined as having no obvious dominant risk.

[0159] The measures taken to regulate the situation do not pose a significant risk; specifically, the current operational status quo will be maintained.

[0160] Furthermore, after determining the dominant risk type and regulatory actions, in order to ensure that the candidate regulatory action sequence can reflect not only the risk type but also the risk level.

[0161] In this embodiment, the correction amount for the control action is constructed by combining the deviation degree of each state quantity in the process state vector.

[0162] In this embodiment, for any process parameter, an action correction amount is constructed based on the degree of deviation of the current state quantity from the corresponding target boundary.

[0163] Specifically, let the first The action correction amount for each process parameter is: The basic correction amount for the process parameters is expressed as: ; in, Indicates time Next The basic correction amount for each process parameter Indicates the first Correction coefficients for each process parameter, This indicates the current dominant risk deviation.

[0164] It should be noted that the correction coefficients for process parameters include correction coefficients for feed setting value, water replenishment setting value, mill load setting value, classification pressure setting value, collector addition setting value, frother addition setting value, slurry pH setting value, and aeration rate setting value.

[0165] It should be noted that the correction coefficient for the feed setpoint is obtained by statistically analyzing the correspondence between changes in the feed setpoint and changes in tailings risk status and particle size dissociation status in historical production data, and by taking the dominant risk change rate caused by a unit feed adjustment; the correction coefficient for the water replenishment setpoint is obtained by statistically analyzing the correspondence between changes in the water replenishment setpoint and changes in the cumulative proportion of the target particle size, slurry concentration, and tailings risk status in historical production data, and by taking the dominant risk change rate caused by a unit water replenishment adjustment; the correction coefficient for the mill load setpoint is obtained by statistically analyzing the correspondence between changes in the mill load setpoint and changes in ore grindability characterization value, cumulative proportion of the target particle size, and over-grinding risk in historical production data, and by taking the dominant risk change rate caused by a unit mill load adjustment; the correction coefficient for the classification pressure setpoint is obtained by statistically analyzing the correspondence between changes in the classification pressure setpoint and changes in classification particle size distribution, cumulative proportion of the target particle size, and tailings risk status in historical production data, and by taking the dominant risk change rate caused by a unit classification pressure adjustment; the correction coefficient for the collector addition setpoint is also determined by statistically analyzing the correspondence between changes in the classification pressure setpoint and changes in classification particle size distribution, cumulative proportion of the target particle size, and tailings risk status in historical production data, and by taking the dominant risk change rate caused by a unit classification pressure adjustment; the correction coefficient for the collector addition setpoint is determined by statistically analyzing the correspondence between changes in the collector addition setpoint and changes in the classification particle size distribution, cumulative proportion of the target particle size, and tailings risk status in historical production data, and by taking the dominant risk change rate caused by a unit collector addition setpoint. The positive coefficient is obtained by statistically analyzing the correlation between changes in the collector addition setpoint and the tailings gold grade characterization value, foam separation state value, and tailings risk state value in historical production data, and taking the dominant risk change rate caused by a unit collector adjustment amount; the correction coefficient for the frother addition setpoint is obtained by statistically analyzing the correlation between changes in the frother addition setpoint and the foam surface movement speed, stable foam area ratio, and inclusion risk state value in historical production data, and taking the dominant risk change rate caused by a unit frother adjustment amount; the correction coefficient for the slurry pH setpoint is obtained by statistically analyzing the correlation between changes in the slurry pH setpoint and the foam separation state value, inclusion risk state value, and tailings risk state value in historical production data, and taking the dominant risk change rate caused by a unit pH adjustment amount; the correction coefficient for the aeration volume setpoint is obtained by statistically analyzing the correlation between changes in the aeration volume setpoint and the foam surface movement speed, stable foam area ratio, and foam separation state value in historical production data, and taking the dominant risk change rate caused by a unit aeration volume adjustment amount.

[0166] Furthermore, the dominant risk deviation is determined based on the current process state range, specifically through steps A1-A4 as follows: A1. When the process state range is dominated by tailings, the deviation of the dominant risk is defined as the tailings risk. The excess of the state value relative to the tailings risk threshold is expressed as: ; in, This indicates the risk threshold for tailings.

[0167] It should be noted that the tailings risk threshold is determined by statistically analyzing the historical tailings risk status of the gold ore beneficiation production line under continuous and stable operation conditions, corresponding to the qualified tailings gold grade range, and taking the upper quantile value of the historical tailings risk status as the tailings risk threshold. The value range is usually the 75% to 95% quantile value.

[0168] A2. When the process state range is the inclusion-dominant range, the dominant risk deviation is defined as the amount by which the inclusion risk state value exceeds the inclusion risk threshold, expressed as: ; in, This indicates the threshold for mixed risk.

[0169] It should be noted that the inclusion risk threshold is calculated by statistically analyzing the historical inclusion risk status of the gold mine beneficiation production line under continuous and stable operation, when the corresponding concentrate grade is qualified and there is no obvious inclusion anomaly in the foam layer. The upper quantile value of the inclusion risk status is taken as the inclusion risk threshold, and the value range is usually 75% to 95%.

[0170] A3. When the process state range is dominated by over-grinding, the dominant risk deviation is defined as the amount by which the particle size dissociation state quantity exceeds the upper limit threshold of the particle size, expressed as: ; in, This indicates the upper limit threshold for the granularity level.

[0171] It should be noted that the upper limit threshold of the particle size is calculated by converting the cumulative proportion of the target particle size corresponding to the gold-bearing minerals in the gold ore when they have achieved effective liberation and have not shown obvious over-grinding and mudification into a particle size liberation state quantity, and taking the upper limit value of the particle size liberation state quantity as the upper limit threshold of the particle size, which is usually 1.05~1.15.

[0172] A4. During the stable operating range, the deviation of the dominant risk is taken as zero.

[0173] Furthermore, in order to ensure that the direction of regulatory actions corresponds to the current dominant risks.

[0174] In this embodiment, for the first The process parameters are combined with the direction factor to achieve the first process parameter. The action correction amount for each process parameter is expressed as: ; in, This represents the direction factor, which takes the value of 1 or -1. It is used to characterize whether the current process parameters should be adjusted to increase or decrease the direction of the current dominant risk.

[0175] For example, in the tail-end dominant zone, the direction factor of the feed setting value can be -1, which means reducing the feed setting value; the direction factor of the collector addition setting value can be 1, which means increasing the collector addition setting value.

[0176] Furthermore, after constructing the action correction amounts for each process parameter, multiple candidate control action sequences are generated based on the currently executed operating condition vector.

[0177] Each candidate control action sequence corresponds to a different parameter correction combination, which represents a set of executable control actions that may be taken within the current control cycle.

[0178] In this embodiment, the standard correction vector includes corrections for feed setting value, water replenishment setting value, mill load setting value, classification pressure setting value, collector addition setting value, frother addition setting value, slurry pH setting value, and aeration setting value.

[0179] By using the standard correction vector as the center and the action amplitude ratio coefficient, multiple candidate control action sequences are generated.

[0180] The candidate regulatory action sequence is represented as follows: ; in, Indicates time The A candidate sequence of regulatory actions, This indicates the currently executed condition vector. Indicates the first The amplitude ratio coefficients corresponding to each candidate regulatory action sequence This represents the standard correction vector.

[0181] It should be noted that the amplitude ratio coefficient is obtained by statistically analyzing the historical adjustment records of the gold ore beneficiation production line during the stable operation phase, and extracting the proportion of the actual effective adjustment amplitude of each process parameter within a single control cycle to the corresponding standard correction amount.

[0182] Ultimately, a set of multiple candidate regulatory action sequences is formed.

[0183] Furthermore, after forming a set of candidate regulatory action sequences, action sequences that cannot effectively alleviate the current dominant risk are removed from the set.

[0184] In this embodiment, candidate regulatory action sequences are screened based on the dominant risk mitigation amount.

[0185] In this embodiment, a local response relationship is established between candidate control action sequences and process state variables.

[0186] Specifically, the difference between the current dominant risk deviation and the dominant risk prediction value of each candidate regulatory action sequence is used as the dominant risk mitigation amount of each candidate regulatory action sequence.

[0187] The dominant risk mitigation amount is expressed as: ; in, Indicates the first The dominant risk mitigation amount corresponding to each candidate regulatory action sequence. Indicates execution of the first Dominant risk prediction value after a candidate regulatory action sequence.

[0188] Furthermore, in this embodiment, the dominant risk prediction value is estimated using local linear response estimation.

[0189] Specifically, the response coefficient vector corresponding to the current dominant risk includes the local response coefficients of the feed setting value, water replenishment setting value, mill load setting value, classification pressure setting value, collector addition setting value, frother addition setting value, slurry pH setting value, and aeration volume setting value to the current dominant risk.

[0190] It should be noted that the local response coefficient of the current dominant risk is obtained by recording the change value of the dominant risk caused by each process parameter under a unit adjustment step under stable production conditions, and identifying the local regression slope of the dominant risk relative to each process parameter using historical operating data within a sliding time window. The identification result is used as the local response coefficient of the current dominant risk.

[0191] The predicted dominant risk value after the execution of each candidate regulatory action sequence is expressed as follows: ; in, Indicates time The dominant risk response coefficient vector.

[0192] Substituting the formula for the dominant risk prediction value into the formula for the dominant risk mitigation amount, we get: ; Therefore, in this embodiment, the dominant risk mitigation amount essentially represents the degree of expected risk improvement brought about by the candidate control action sequence relative to the current executed operating condition.

[0193] After obtaining the dominant risk mitigation amount corresponding to all candidate regulatory action sequences, the candidate regulatory action sequences are screened.

[0194] Specifically, only candidate control action sequences that satisfy the condition that the amount of mitigation of the dominant risk is not less than the lower limit threshold of the dominant risk mitigation are retained as effective candidate control action sequences for mitigating the current dominant risk.

[0195] Candidate control action sequences that do not meet the screening criteria are removed from the candidate control action sequence set.

[0196] It should be noted that the lower limit threshold for mitigating the dominant risk is determined by statistically analyzing the distribution of the actual decrease in dominant risk after the implementation of each candidate control action sequence based on historical sample data of the gold mine beneficiation production line under continuous and stable operating conditions. The minimum effective decrease that meets the requirements for improving recovery rate or tailings gold grade is determined as the lower limit threshold for mitigating the dominant risk. The value range is usually 3% to 15% of the current dominant risk benchmark.

[0197] Furthermore, to improve the stability of the screening results, a continuity constraint is added, which requires that the total change of the candidate control action sequence relative to the currently executed working condition vector does not exceed the upper limit of the action disturbance.

[0198] When the total change exceeds the upper limit of the action disturbance, even if the dominant risk mitigation amount of the candidate control action sequence meets the requirements, it is still removed to avoid causing excessive instantaneous disturbance to the on-site working conditions.

[0199] Ultimately, a set of candidate regulatory action sequences that satisfy the lower threshold for dominant risk mitigation and the upper threshold for action perturbation is obtained.

[0200] It should be noted that the upper limit of motion disturbance is usually based on the historical adjustment data of each process parameter in the continuous control cycle, the equipment response capability and the test results of the operating condition stability test. It is calculated to be the upper limit of the total motion change that will not cause fluctuations in grinding and classification, instability of the foam layer or sudden changes in the reagent system. The upper limit is determined as the upper limit of motion disturbance. The value range is usually 5% to 20% of the total set value of the current executed operating condition corresponding to the sum of the absolute values ​​of the single-cycle changes of each control parameter.

[0201] It should also be noted that parameter linkage means that when one control parameter changes, other related control parameters are adjusted synchronously according to a fixed coupling rule to ensure process continuity and overall stability.

[0202] It should also be noted that the dominant risk response relationship refers to the mapping relationship between the parameter changes of the candidate control action sequence and the current dominant risk prediction changes, which is used to calculate the dominant risk prediction value and mitigation amount after the action sequence is executed.

[0203] S4. Compare the reduction effects of each candidate regulatory action sequence on the current dominant risk, determine the final regulatory action sequence, and execute it.

[0204] Furthermore, based on the current process state range, determine the current dominant risk benchmark.

[0205] Among them, the dominant risk benchmark is used as a unified reference value for comparing the reduction effects of various candidate regulatory action sequences.

[0206] In this embodiment, when the current process state interval is the tailings-dominant interval, the tailings risk state quantity is used as the current dominant risk benchmark quantity.

[0207] When the current process state range is the inclusion-dominant range, the inclusion risk state quantity is used as the current dominant risk benchmark quantity.

[0208] When the current process state range is dominated by over-grinding, the amount by which the particle size dissociation state exceeds the upper limit threshold of the particle size is used as the current dominant risk benchmark.

[0209] When the current process state range is a stable operating range, the current dominant risk benchmark is set to zero.

[0210] Furthermore, after determining the current dominant risk benchmark, for each candidate control action sequence in the set of effective candidate control action sequences, the predicted reduction magnitude of the current dominant risk after execution is calculated.

[0211] In this embodiment, the difference between the dominant risk benchmark and the dominant risk prediction value after the corresponding candidate control action sequence is used as the dominant risk reduction magnitude of the corresponding candidate control action sequence.

[0212] In this embodiment, the magnitude of the dominant risk reduction is essentially equal to the amount of dominant risk mitigation brought about by the candidate control action sequence relative to the currently executed operating condition.

[0213] To avoid the significant deterioration of other state variables caused by selecting candidate control action sequences based solely on the magnitude of a single dominant risk reduction.

[0214] In this embodiment, let the first After the execution of the candidate regulatory action sequence, the th... The predicted change value of each non-dominant state variable is The corresponding total amount of non-dominant disturbance is expressed as: ; in, Indicates the first The total disturbance of all non-dominant state variables by each candidate regulatory action sequence. Indicates the first After the execution of the candidate regulatory action sequence, the th... Predicted changes in each non-dominant state variable.

[0215] It should be noted that the predicted changes of different non-dominant state quantities are normalized before being summed by direct summation of absolute values.

[0216] It should be noted that the predicted change values ​​of non-dominant state variables are obtained by establishing the local response relationship between each non-dominant state variable and each process parameter based on historical operating data within the sliding time window, and by substituting the parameter change of the candidate control action sequence relative to the currently executed operating condition vector into the calculation formula of the local response coefficient of the corresponding non-dominant state variable.

[0217] Among them, the total amount of non-dominant disturbance is used to prioritize the candidate control action sequence with smaller disturbance to the non-dominant state quantity when multiple candidate control action sequences have similar dominant risk reduction magnitudes.

[0218] Furthermore, after obtaining the dominant risk reduction magnitude of each candidate regulatory action sequence, the effective candidate regulatory action sequence set is prioritized.

[0219] The basic principle of sorting is to prioritize candidate control action sequences with larger reductions in dominant risks; when the reductions in dominant risks are similar, the total amount of non-dominant disturbances and the continuity of operating conditions are then compared.

[0220] In this embodiment, when the difference in the dominant risk reduction magnitude of multiple candidate control action sequences does not exceed the equivalent threshold, the priority of the two is no longer distinguished solely by the dominant risk reduction magnitude. Instead, when comparing the total amount of non-dominant disturbances, the candidate control action sequence with the larger one will be retained first.

[0221] It should be noted that the equivalent threshold is determined by the difference distribution of the dominant risk reduction magnitude corresponding to different candidate control action sequences within the historical control cycle of the gold mine beneficiation production line, combined with process measurement error and natural fluctuation of operating conditions. The reduction magnitude difference that is less than the comprehensive fluctuation upper limit is determined as the equivalent threshold, and the value range is usually 1% to 8%.

[0222] Furthermore, in order to ensure the smoothness of the control action issuance process, it is preferable to construct the total action change amount.

[0223] Specifically, let the first The change in the total action of each candidate control action sequence relative to the currently executed operating condition is expressed as: ; in, Indicates the first The total change in action of each candidate control action sequence relative to the currently executed working condition vector. Indicates the first In the candidate regulatory action sequence, the first The set values ​​of each process parameter, This indicates the th element in the currently executed condition vector. The set values ​​of each process parameter.

[0224] When two candidate control action sequences satisfy both the requirement that the magnitude of the reduction in dominant risk is approximately equal and the total amount of non-dominant disturbance is approximately equal, the candidate control action sequence with the smaller total change in action should be selected first to ensure the continuity of operating conditions.

[0225] Furthermore, after comparing the priorities of each candidate regulatory action sequence, the candidate regulatory action sequence with the highest priority is selected as the final regulatory action sequence.

[0226] Specifically, in the first selection layer, the magnitude of the reduction in dominant risk is compared, and the candidate control action sequence with the largest magnitude of reduction in dominant risk is retained first.

[0227] In the second selection layer, among the candidate control action sequences retained in the first selection layer, the total amount of non-dominant disturbances is compared, and the candidate control action sequence with the smallest total amount of non-dominant disturbances is preferentially retained.

[0228] In the third optimization layer, among the candidate regulatory action sequences retained from the second optimization layer, the total change in action is compared, and the candidate regulatory action sequence with the smallest total change in action is retained first.

[0229] Therefore, the final sequence of control actions is determined by a hierarchical optimization rule that prioritizes dominant risks, followed by non-dominant disturbances, and then action continuity.

[0230] Furthermore, after determining the final control action sequence, the final control action sequence is not issued all at once. Instead, it is broken down into parameter adjustment instructions corresponding to different execution objects and then sent to each execution mechanism for execution in a fixed order.

[0231] Specifically, adjustment instructions related to grinding and classification are first issued, including feed setting values, water replenishment setting values, mill load setting values, and classification pressure setting values, in order to prioritize changing the particle size liberation state before entering flotation.

[0232] Further adjustment instructions related to the reagent system were issued, including set values ​​for collector addition, frother addition, and pulp pH, to further correct the mineral floatability and foam properties based on the adjustment of the grinding and classification status.

[0233] Finally, an instruction to adjust the aeration volume setting is issued to make a final correction to the dynamic state of the flotation foam layer.

[0234] During execution, each set value is implemented through the corresponding actuator.

[0235] Specifically, the feed setting value is achieved by adjusting the speed or opening of the feed equipment; the water replenishment setting value is achieved by adjusting the opening of the water replenishment valve; the mill load setting value is achieved by adjusting the mill control loop; the classification pressure setting value is achieved by controlling the relevant pumps or valves; the collector addition setting value and the frother addition setting value are achieved by controlling the corresponding reagent metering pumps; the pulp pH setting value is achieved by controlling the acid and alkali addition; and the aeration rate setting value is achieved by the flotation aeration control device.

[0236] Furthermore, to reduce the risk of mutations during the execution process.

[0237] In this embodiment, the speed is limited for each set value by a speed limit execution rule.

[0238] Specifically, when the parameter change in the final control action sequence exceeds the maximum allowable change, the control will only proceed up to the intermediate value corresponding to the current maximum allowable change, and will continue to approach the target value in the next control cycle.

[0239] It should be noted that the maximum permissible variation of the feed setpoint, water replenishment setpoint, mill load setpoint, and classification pressure setpoint is determined by calibration based on the single-cycle response capability of the corresponding actuator under rated operating conditions, the equipment's safe operating boundary, and stable adjustment over multiple consecutive control cycles. The maximum permissible variation is the maximum parameter change that will not cause sudden changes in slurry concentration, mill overload, or classification pressure instability. The value range is typically 2% to 6% of the current setpoint of the corresponding parameter. The maximum permissible variation of the collector addition setpoint, frother addition setpoint, slurry pH setpoint, and aeration setpoint is determined by calibration based on the single-cycle output capability of the reagent metering device and aeration device, the stable range of flotation froth state, and on-site step adjustment. The maximum permissible variation is the maximum parameter change that will not cause violent fluctuations in the froth layer, excessive reagent impact, or abnormal shifts in slurry pH. The value range is typically 1% to 8% of the current setpoint of the corresponding parameter.

[0240] In this embodiment, Figure 5 This reflects the changes in gold grade of tailings under continuous sampling conditions at the industrial site, compared with control conditions and after adopting the method of this invention.

[0241] like Figure 5 As shown, under conventional experience-based control conditions, the gold grade of tailings fluctuates significantly, with noticeable upward spikes in certain periods. This indicates that when on-site operating conditions are affected by fluctuations in feed, changes in grinding and classification status, and disturbances in flotation froth conditions, the risk of tailings loss is difficult to suppress in a timely manner. However, after adopting the method of this invention, the overall gold grade of tailings significantly decreases, the frequency of abnormal peaks decreases, and the moving average curve becomes more stable.

[0242] In summary, by aligning the results of ore feeding, grinding and classification, flotation and tailings across processes in time and space, this invention can more accurately correlate upstream conditions with downstream results. Combined with the determination of the dominant risk range, it can generate targeted control actions, thereby improving the accuracy of tailings loss risk identification and response efficiency, and ultimately reducing tailings gold loss and improving on-site operational stability.

[0243] In summary, this invention achieves spatiotemporal alignment of multimodal data across processes by calculating the dynamic delay duration of each process segment and backtracking to match upstream characterization values, thereby improving the accuracy of process state characterization. By constructing process state vectors and determining dominant risk intervals, it enables automatic identification of dominant risk types in the mineral processing process, providing a basis for targeted regulation. By generating candidate regulation action sequences based on dominant risk types and deviation degrees and screening them according to risk mitigation amounts, it achieves directional responses of regulation actions to current dominant risks, enhancing the targeting of decision-making.

[0244] 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 multi-modal optimization method for a mineral processing process across all time and space, characterized in that, include: Step S1: Collect raw sensing data of the feeding position, grinding and classification position, flotation position, and concentrate and tailings discharge position during the gold ore beneficiation process, and extract the ore grindability characterization value, the cumulative proportion of target particle size, the froth surface movement speed, the stable froth area proportion, and the tailings gold grade characterization value. Step S2: Align the characterization values ​​across processes in time and space based on the dynamic delay time of the material at each location, generate a process state vector, and determine the current process state interval. Step S3: Based on the process state vector, process state interval, and currently executed operating conditions, generate multiple candidate control action sequences and perform dominant risk mitigation screening; Step S4: Compare the reduction magnitude of each candidate regulatory action sequence on the current dominant risk, determine the final regulatory action sequence, and execute it.

2. The multi-modal optimization method for the mineral processing process across all time and space as described in claim 1, characterized in that, The raw sensing data includes ore image data, feed mass flow rate data, feed particle size data, mill active power data, slurry concentration data, classification product particle size data, foam image sequence, tailings detection data, and concentrate detection data.

3. The multi-modal optimization method for the mineral processing process across all time and space as described in claim 2, characterized in that, The specific steps for extracting the ore grindability characterization value, the cumulative proportion of the target particle size, the foam surface movement speed, the stable foam area proportion, and the tailings gold grade characterization value are as follows: Based on feed mass flow rate data, feed particle size data, and mill active power data, the grindability characterization value of the ore is extracted. Based on the particle size data of the graded products, the cumulative proportion of the target particle size is extracted; Based on the displacement of foam texture between adjacent foam image frames in a foam image sequence, the movement speed of the foam surface is extracted. Based on the boundary continuity and area change rate of the foam region at adjacent time points, the proportion of stable foam area is extracted. Based on the tailings gold grade detection values ​​over multiple consecutive collection cycles, the tailings gold grade characterization values ​​are extracted.

4. The multi-modal optimization method for the mineral processing process across all time and space as described in claim 1 or 3, characterized in that, The cross-process spatiotemporal alignment includes: The process is divided into three stages: the first stage, the second stage, and the third stage, from the feed position to the grinding and classification position, from the grinding and classification position to the flotation position, and from the flotation position to the concentrate and tailings discharge position. The effective volume and real-time volumetric flow rate of each stage are obtained respectively. Calculate the basic delay time for each process segment based on the ratio of the effective volume to the real-time volumetric flow rate of each process segment; The basic delay time is corrected based on the load deviation of each process segment to obtain the dynamic delay time of each process segment. Based on the dynamic delay duration, and taking the time corresponding to the gold grade characterization value of tailings as the benchmark, the ore grindability characterization value, the cumulative proportion of target particle size, the foam surface movement speed, and the proportion of stable foam area are back-matched to obtain the spatiotemporal alignment feature sequence across processes.

5. The multi-modal optimization method for the mineral processing process across all time and space as described in claim 3, characterized in that, The generated process state vector includes: Obtain the ore grindability characterization value, target particle size cumulative proportion, foam surface movement speed, stable foam area proportion, and tailings gold grade characterization value from the cross-process spatiotemporal alignment feature sequence. The particle size dissociation state quantity is generated based on the ratio of the cumulative proportion of the target particle size to the baseline value of the cumulative proportion of the target particle size. Based on the combination of foam surface movement speed, stable foam area ratio, and foam speed benchmark value, foam sorting state quantities are generated. Based on the combination relationship between the gold grade characterization value of tailings and the particle size dissociation state quantity, the tailings risk state quantity is generated. Based on the ratio of the stable foam area ratio to the foam surface movement speed, the inclusion risk state quantity is generated; The particle size dissociation state, foam separation state, tailings risk state, and inclusion risk state are combined to form a process state vector.

6. The multi-modal optimization method for the mineral processing process across all time and space as described in claim 5, characterized in that, The determination of the current process state range includes: Compare the tailings risk state quantity and tailings risk threshold in the process state vector, and compare the particle size dissociation state quantity and the particle size lower limit threshold. When the tailings risk status quantity is greater than the tailings risk threshold and the particle size dissociation status quantity is less than the lower limit threshold of the particle size, the current process status interval is determined to be the tailings run-dominant interval. The mixture risk state quantity and mixture risk threshold are compared, and the foam sorting state quantity and foam sorting lower limit threshold are compared; When the inclusion risk state quantity is greater than the inclusion risk threshold and the foam sorting state quantity is less than the foam sorting lower limit threshold, the current process state interval is determined to be the inclusion-dominant interval. Compare the particle size dissociation state quantity with the particle size upper limit threshold, and compare the inclusion risk state quantity with the inclusion risk threshold; When the particle size dissociation state quantity is greater than the upper limit threshold of the particle size and the inclusion risk state quantity is greater than the inclusion risk threshold, the current process state interval is determined to be the over-grinding dominant interval. When the conditions for determining the tail-running dominant zone, the inclusion dominant zone, and the over-wearing dominant zone are not met, the current process state zone is determined to be a stable operating zone.

7. The multi-modal optimization method for the mineral processing process across all time and space as described in claim 1, characterized in that, The generation of multiple candidate regulatory action sequences includes: Read the feed setting value, water replenishment setting value, mill load setting value, classification pressure setting value, collector addition setting value, frother addition setting value, slurry pH setting value, and aeration volume setting value within the current control cycle to form the current executed working condition vector; Based on the process status range, determine the current dominant risk type and the correction direction for each control parameter; The action correction amount is calculated based on the degree of deviation of the state variables in the process state vector from the threshold. The action correction amounts are superimposed at different amplitude ratios to generate candidate control action sequences; Based on the parameter linkage relationship, the relevant control parameters in the candidate control action sequence are synchronously corrected to obtain a set of candidate control action sequences.

8. The multi-modal optimization method for the mineral processing process across all time and space as described in claim 7, characterized in that, The primary risk mitigation screening includes: Based on the current dominant risk type, establish the dominant risk response relationship for each candidate regulatory action sequence; Based on the parameter changes of each candidate control action sequence relative to the currently executed operating condition vector, calculate the dominant risk prediction value after the execution of each candidate control action sequence; Based on the current deviation of the dominant risk and the predicted value of the dominant risk, calculate the amount of mitigation of the dominant risk corresponding to each candidate control action sequence; The dominant risk mitigation amount and the dominant risk mitigation lower limit threshold of each candidate control action sequence are compared, and candidate control action sequences with a dominant risk mitigation amount not lower than the dominant risk mitigation lower limit threshold are retained. Calculate the total change in action of each candidate control action sequence after retention relative to the currently executed working condition vector, remove candidate control action sequences whose total change in action exceeds the upper limit of action disturbance, and obtain a set of effective candidate control action sequences.

9. The multi-modal optimization method for the mineral processing process as described in claim 8, characterized in that, The comparison of the reduction effect of each candidate regulatory action sequence on the current dominant risk includes: Determine the current dominant risk benchmark based on the current process state range, and calculate the dominant risk reduction magnitude corresponding to each candidate control action sequence; The candidate regulatory action sequences are ranked according to the magnitude of the dominant risk reduction. When the difference in the dominant risk reduction magnitude of multiple candidate control action sequences does not exceed the equivalent threshold, the candidate control action sequence with the smallest total disturbance is given the highest priority.

10. The multi-modal optimization method for the mineral processing process across all time and space as described in claim 9, characterized in that, The process of determining and executing the final control action sequence includes: The candidate regulatory action sequence with the highest priority is determined as the final regulatory action sequence; The final control action sequence is decomposed into control instructions; The maximum allowable change in a single control cycle is set for each adjustment command. When the parameter change of the adjustment command exceeds the maximum allowable change, the command is executed to the intermediate value of the maximum allowable change and continues to approach the final control action sequence in the next control cycle.

Images