A long-process lead-zinc flotation intelligent control method and system based on multi-source information fusion and adaptive optimization

Abstract

The application relates to a long-process lead-zinc flotation intelligent control method and system based on multi-source information fusion and adaptive optimization. The method comprises the following steps: acquiring multi-source information in a long-process lead-zinc flotation process, processing and fusing the multi-source information to obtain working condition characteristic information; performing state recognition on the current flotation working condition based on the working condition characteristic information, and predicting the control target index of the flotation process, finally obtaining the optimal control set value after adaptive optimization; converting the optimal control set value into an execution instruction and issuing the execution instruction to a flotation execution mechanism to control the long-process lead-zinc flotation process. The application can more accurately represent the current flotation working condition and timely generate a more optimal control instruction, thereby improving the real-time performance, pertinence and stability of the flotation process control, enhancing the adaptive capacity of the system to complex working condition changes and disturbances, and being beneficial to realizing stable and efficient operation of the long-process lead-zinc flotation process and improving the overall control effect.

Description

Technical Field

[0001] This application relates to the field of lead-zinc flotation technology, and in particular to a long-process intelligent control method and system for lead-zinc flotation based on multi-source information fusion and adaptive optimization. Background Technology

[0002] Currently, long-process lead-zinc flotation production generally faces technical challenges such as numerous and highly coupled process variables, difficulty in online detection of key process indicators with significant lag, reliance on manual experience, poor control stability, low level of intelligence, and lack of full-process collaborative optimization capabilities.

[0003] The flotation process is influenced by a multitude of factors, including the properties of the raw ore, its floatability, the particle size distribution of the grinding process, the pulp concentration, pH value, reagent formulation, aeration rate, liquid level, and frothing intensity. These factors exhibit significant nonlinearity, time-varying characteristics, and strong coupling. Particularly when the lead and zinc circuits operate in parallel or in series, significant material-reagent coupling exists between the different circuits. Control actions in one circuit can disturb the other, making it difficult to achieve overall process optimization through localized adjustments.

[0004] The final evaluation indicators of flotation, such as concentrate grade, tailings grade, and metal recovery rate, usually rely on offline sampling and laboratory testing, which have long feedback cycles, typically with a lag of tens of minutes to several hours, and cannot be directly used for real-time closed-loop control. Although some sites are equipped with online analytical instruments, their detection points are limited, and their accuracy is greatly affected by operating conditions, making it difficult to accurately and continuously reflect the overall separation effect.

[0005] Meanwhile, current production mainly relies on operators making manual judgments and adjustments based on foam appearance, tank surface condition, and delayed test results. This method is highly dependent on operational experience, subjective, lacks standardized procedures, and makes it difficult to maintain consistency between shifts. It is prone to adjustment delays, misjudgments, or over-adjustments when ore properties fluctuate, reagent response is delayed, or operating conditions change abruptly, resulting in grade fluctuations, decreased recovery rates, reagent waste, and increased energy consumption. Although some concentrators are equipped with PLC or DCS systems, they can usually only achieve stable control of single variables such as liquid level and flow rate. Most adopt single-loop PID strategies, which cannot perform global optimization for multi-variable coupling, cross-loop disturbances, operating condition migration, and target conflicts in long-process lead-zinc flotation, and it is also difficult to translate process expert experience into repeatable control strategies.

[0006] Therefore, there is an urgent need for an intelligent closed-loop control method that can perceive the entire flotation process in real time, predict key indicators, identify complex working conditions, and combine expert experience and optimization algorithms to achieve coordinated regulation of the entire process, in order to solve the problems of control lag, large fluctuations, reliance on manual labor, and insufficient collaborative optimization in existing technologies. Summary of the Invention

[0007] To address or partially address the problems existing in related technologies, this application provides an intelligent control method and system for long-process lead-zinc flotation based on multi-source information fusion and adaptive optimization. By constructing a closed-loop intelligent control architecture, it achieves multi-source information fusion, operating condition identification, key indicator prediction, and global collaborative optimization control of the long-process lead-zinc flotation process.

[0008] The first aspect of this application provides a smart control method for long-process lead-zinc flotation based on multi-source information fusion and adaptive optimization, comprising the following steps: Acquire multi-source information from the long-process lead-zinc flotation process, process and fuse the multi-source information to obtain condition characteristic information for characterizing the current flotation conditions; Based on the operating condition characteristics, the current flotation condition is identified, and the control target indicators of the flotation process are predicted. The control setpoints for the flotation process are generated based on the operating conditions and control objectives, and the optimal control setpoints are obtained after adaptive optimization. The optimal control setpoint is converted into an execution command and sent to the flotation actuator to control the long-process lead-zinc flotation process.

[0009] Specifically, the system collects feedback information after execution control and updates the state identification and / or adaptive optimization based on the feedback information to form a closed-loop control.

[0010] The processing and fusion of multi-source information includes: The flotation foam image information is denoised, corrected, and the region of interest is extracted. The flotation process parameter information is processed for outliers, compensated for missing values, and standardized. Time alignment and time delay compensation are performed on information from different sampling frequencies; The processed multi-source information is spliced ​​or mapped to construct working condition feature information.

[0011] The process of generating the control setpoint includes: Based on the operating condition, retrieve similar operating condition cases from historical operating cases, and generate recommended control setpoints for the cases based on the operating parameters corresponding to the similar operating condition cases. The recommended control settings are modified according to the preset control rules corresponding to the current operating condition to obtain the control settings.

[0012] Adaptive optimization of the control setpoint includes: Establish a predictive process to characterize future trends in long-process lead-zinc flotation; Input the current operating condition characteristic information, operating condition status identification results and initial control setpoints into the prediction process to obtain the index prediction results for future control cycles; Based on the predicted results of the indicators, the initial control setpoint is iteratively corrected under the preset control constraints to obtain the optimal control setpoint. The optimal control setpoint takes the comprehensive control effect of concentrate grade, recovery rate and reagent consumption as the optimization target.

[0013] Among them, the state recognition and predictive control target indicators are realized based on a multi-source feature fusion model, which includes: LSTM branch used to extract the temporal features of flotation process parameters; DBN branch used to extract features from foam images and / or deep nonlinear features from laboratory data; This is a decision output layer used to fuse the outputs of LSTM and DBN branches and output the results of operating condition identification and / or prediction results of control target-related indicators.

[0014] The control setpoints include at least one of the following: reagent addition amount, aeration amount, slurry level, slurry flow rate, and stirring intensity.

[0015] Among them, the control setpoints for the flotation process are generated based on the operating conditions and control target indicators, including: taking into account the correlation between different operating sections and / or lead circuits and zinc circuits in long-process lead-zinc flotation, and collaboratively determining the control setpoints for each operating section.

[0016] When an abnormality in operating condition identification, prediction result, key input information, or optimal control setpoint is detected, the current adaptive optimization process is stopped, and the system switches to a preset rule control mode and / or a manual confirmation control mode.

[0017] The second aspect of this application provides a long-process lead-zinc flotation intelligent control system based on multi-source information fusion and adaptive optimization, used to execute the aforementioned long-process lead-zinc flotation intelligent control method, including: The data acquisition module is used to acquire foam images, process parameters, and test data in real time. The data fusion and operating condition identification module is used to fuse foam images, process parameters, and test data, and to determine the current flotation status. The intelligent decision-making module is used to generate key operating parameter settings for the lead and zinc circuits based on the current flotation status, and optimize the key operating parameter settings to obtain the optimal settings. The collaborative execution module is used to drive the actions of field equipment according to the optimal set value, so as to achieve collaborative optimization control of the entire flotation process; The feedback update module is used to collect feedback information after execution control and update the judgment process of the current flotation state and / or the generation process of the optimal setpoint based on the feedback information.

[0018] The technical solution provided in this application may include the following beneficial effects: This application provides an intelligent control method and system for long-process lead-zinc flotation based on multi-source information fusion and adaptive optimization. It overcomes excessive reliance on human experience, achieving unmanned, efficient, and stable operation in a "lights-out" factory environment, significantly reducing production fluctuations caused by human factors. Through advanced condition identification and global optimization, the flotation process can continuously operate in its optimal state, effectively improving the stability of lead / zinc concentrate grades and metal recovery rates, resulting in significant economic benefits. Furthermore, it achieves precise and coordinated reagent addition, effectively avoiding reagent waste or insufficiency, and reducing production costs and consumption.

[0019] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0020] The above and other objects, features and advantages of this application will become more apparent from the more detailed description of exemplary embodiments thereof in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments thereof.

[0021] Figure 1 This is a schematic flowchart of the intelligent control method for long-process lead-zinc flotation shown in the embodiments of this application; Figure 2 This is a schematic diagram of the structure of the intelligent control system for long-process lead-zinc flotation shown in the embodiments of this application. Detailed Implementation

[0022] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to make this application more thorough and complete, and to fully convey the scope of this application to those skilled in the art.

[0023] Please see Figure 1 This embodiment provides an intelligent control method for long-process lead-zinc flotation based on multi-source information fusion and adaptive optimization, including the following steps: S1. Obtain multi-source information from the long-process lead-zinc flotation process, process and fuse the multi-source information to obtain condition characteristic information for characterizing the current flotation conditions.

[0024] In this embodiment, the multi-source information includes one or more of the following: foam image information, flotation process parameter information, and laboratory data information.

[0025] Among them, foam image information can be collected in real time by an industrial camera set above the flotation cell; process parameter information can be obtained by field instruments, PLC or DCS system, including reagent addition amount, aeration amount, slurry level, slurry flow rate, slurry concentration, pH value, stirring current, valve opening degree, pump frequency, etc.; test data information can come from online detection device and / or offline test results, including raw ore grade, concentrate grade, tailings grade, recovery rate and other indicators.

[0026] Since different information sources differ in sampling frequency, time delay, dimensional range, and noise level, preprocessing of various types of information is necessary before fusion.

[0027] For foam image information, methods such as noise reduction, brightness correction, distortion correction, and region of interest extraction can be used for processing. Based on this, features such as foam size, texture, color, movement speed, stability, and distribution uniformity can be extracted.

[0028] For process parameter information, outlier removal, missing value compensation, and standardization can be performed. Preferably, sampled values ​​that exceed a preset threshold range or deviate from historical statistical patterns are identified as outliers and corrected using nearest-time interpolation, moving average, or historical valid values. For missing values, linear interpolation, nearest-neighbor imputation, or model estimation can be used for compensation. For process parameters with different dimensions, standardization can be used to unify them.

[0029] For example, the j-th process parameter can be standardized in the following way: in, These are the original parameter values ​​at time t. and These are the statistical mean and standard deviation of the corresponding parameters, respectively. To prevent extremely small constants with a denominator of zero.

[0030] For laboratory data, since it is usually updated in a low frequency and has a large detection time lag, time mapping and time lag compensation can be performed according to its corresponding production time to make it correspond to the current control time.

[0031] Furthermore, time alignment and time delay compensation are performed on information from different sampling frequencies to unify the features of foam images, process parameters, and laboratory data to the same control time. After time alignment, feature stitching, mapping fusion, or dimensionality reduction mapping methods can be used to construct the operating condition feature information.

[0032] In one embodiment, the operating condition characteristic information can be represented as: in, Indicates features of a bubble image. Indicates process parameter characteristics, Indicates the characteristics of laboratory data. Represents the fusion mapping function, This represents the conditional information used to characterize the current flotation operation.

[0033] Through the above processing, multiple types of information reflecting the apparent state of flotation, the operating state of the process, and the state of the result indicators can be uniformly represented as operating condition characteristic information, providing an input basis for subsequent operating condition identification and control decisions.

[0034] S2. Based on the operating condition characteristic information, identify the current flotation operating condition and predict the control target indicators of the flotation process.

[0035] In this embodiment, the current flotation condition is identified based on operating condition characteristic information, and the control target indicators of the flotation process are predicted. The control target indicators may include one or more of the following: concentrate grade, tailings grade, metal recovery rate, concentrate yield, and overall beneficiation efficiency.

[0036] Preferably, state recognition and control target index prediction are based on a multi-source feature fusion model, which includes: LSTM branch used to extract the temporal features of flotation process parameters; DBN branch used to extract features from foam images and / or deep nonlinear features from laboratory data; This is a decision output layer used to fuse the outputs of LSTM and DBN branches and output the results of operating condition identification and / or prediction results of control target-related indicators.

[0037] Among them, the LSTM branch is used to characterize the dynamic changes of process parameters in the time dimension, which is suitable for capturing the temporal correlation and hysteresis effects in the flotation process; the DBN branch is used to extract deep features from foam image information and test data information, enhancing the ability to characterize complex nonlinear working conditions; the decision output layer is used to fuse the output results of different branches to form the identification results of the current working condition and the prediction results of subsequent control target indicators.

[0038] In one embodiment, the fused feature vector is obtained by fusing the temporal characteristics of process parameters and the deep nonlinear characteristics: Based on the fused feature vector, the current operating condition identification result and the prediction result of the control target index in the future control cycle can be output.

[0039] The operating conditions can be categorized into one or more types based on actual production needs, such as stable operating conditions, disturbed operating conditions, low reagent levels, high reagent levels, abnormal aeration, and fluctuating liquid levels. Those skilled in the art can adjust or expand these categories according to the specific process structure, ore properties, and on-site control requirements.

[0040] This step allows the raw sensing information to be further transformed into predictive results of operating conditions and target indicators that have control significance, providing a basis for the generation and optimization of control setpoints.

[0041] S3. Generate control setpoints for the flotation process based on the operating conditions and control target indicators, and obtain the optimal control setpoints after adaptive optimization.

[0042] In this embodiment, control setpoints for the flotation process are generated based on the operating conditions and control target indicators, and the optimal control setpoints are obtained after adaptive optimization.

[0043] The control setpoints include at least one of the following: reagent addition amount, aeration amount, and slurry level. In long-process lead-zinc flotation applications, the control setpoints can also be set separately for lead circuits and zinc circuits, as well as for different operating sections such as roughing, scavenging, and cleaning, thereby achieving coordinated control of the entire process.

[0044] Preferably, the process of generating control setpoints includes: generating case-recommended control setpoints based on historical operation cases, and making corrections based on preset control rules.

[0045] First, based on the current operating condition characteristics and status, similar operating condition cases are retrieved from the historical operating case database. This database stores historical operating condition characteristics, corresponding operating conditions, historical operating parameters, and actual control effects. After retrieving similar cases, recommended control setpoints are generated based on the corresponding operating parameters.

[0046] In one embodiment, similarity can be calculated using weighted distance: in, Let j be the j-th component of the current operating condition characteristic. Let j be the j-th component of the feature corresponding to the i-th historical case. These are the feature weights.

[0047] Based on the retrieved similar operating case examples, the corresponding operating parameters can be weighted and calculated to obtain the recommended control settings for the case.

[0048] Based on this, the recommended control setpoints are modified according to the preset control rules corresponding to the current operating conditions to obtain the control setpoints. Preferably, the preset control rules can be empirical rules, expert rules, or fuzzy control rules. If fuzzy control rules are used, the control quantities such as the dosage of reagent, the aeration rate, and the liquid level can be increased or decreased based on the operating condition deviation, the index deviation, and the trend of change.

[0049] After rule correction, the initial control setpoint is obtained, and then the initial control setpoint is adaptively optimized to obtain the optimal control setpoint.

[0050] Furthermore, a prediction process is established to characterize the future trends of the long-process lead-zinc flotation process. Current operating condition characteristics, operating condition identification results, and initial control setpoints are input into the prediction process to obtain the predicted index results for the future control cycle. Based on this, the initial control setpoints are iteratively corrected under preset control constraints according to the predicted index results to obtain the optimal control setpoints.

[0051] Preferably, the adaptive optimization employs a rolling optimization method, jointly solving for the control setpoints over multiple future control cycles, and selecting the control setpoint corresponding to the current control cycle as the optimal control setpoint for actual execution. The preset control constraints include at least one of the following: control variable value range constraints, control variable change amplitude constraints, control variable change rate constraints, and actuator operation constraints.

[0052] In one embodiment, optimizing the objective function can comprehensively consider the tracking effect of the control target index, the degree of deviation of the control quantity, and the smoothness of the control action, for example: in, The reference value for the control target indicator is represented by Q, R, and S, which represent the weight matrices for indicator tracking, control deviation, and control stability, respectively.

[0053] After solving the constraint optimization problem, the optimal control setpoint at the current moment is obtained: By using the above methods, optimal control settings that are more suitable for the current long-process lead-zinc flotation state can be generated, taking into account current operating conditions, historical experience, rule knowledge, and future trends.

[0054] S4. Convert the optimal control setpoint into an execution command and send it to the flotation actuator to control the long-process lead-zinc flotation process.

[0055] In this embodiment, the optimal control setpoint is converted into an execution command and sent to the flotation actuator to control the long-process lead-zinc flotation process.

[0056] Flotation actuators may include reagent dosing devices, aeration devices, level control devices, slurry conveying devices, and other field execution units related to flotation control. Depending on the execution method and interface requirements of different controlled objects, the optimal control setpoint can be converted into pump frequency commands, valve opening commands, level control commands, or other equipment control commands.

[0057] In one embodiment, the optimal control setpoint and the execution command can be mapped through a preset conversion relationship: in, This represents the optimal control setpoint. Indicates the execution of the instruction. This represents a function that converts a set value into an execution instruction.

[0058] Preferably, before issuing the execution command, amplitude limit verification, safety range judgment, and equipment executability verification can be performed to ensure that the execution command is within the allowable operating range of the field equipment. The verified execution command can be issued to the corresponding actuator via PLC, DCS, industrial Ethernet, or fieldbus, thereby realizing online control of the long-process lead-zinc flotation process.

[0059] Furthermore, it also includes S5, which collects feedback information after execution control and updates the state recognition and / or adaptive optimization based on the feedback information to form closed-loop control.

[0060] In this embodiment, after the control is executed, feedback information is further collected, and the state identification and / or adaptive optimization are updated based on the feedback information to form closed-loop control.

[0061] Feedback information includes one or more of the following: process parameter feedback after control execution, foam image change feedback, online detection index feedback, laboratory result feedback, and actuator status feedback. By comparing the feedback information with the pre-control prediction results and control setpoints, the current control effect can be evaluated, and the status identification model, prediction process, and optimization parameters can be updated accordingly.

[0062] For example, when there is a deviation between the actual control target index after execution and the prediction result, the prediction process parameters can be corrected according to the deviation; when the feedback information indicates that the working condition category has changed, the working condition status identification result can be updated again; when the control effect is inconsistent with the expectation, the optimization weight, rule parameters or case weight can also be adjusted.

[0063] In one embodiment, the prediction error can be expressed as: in, This represents the control target index obtained from actual testing. This represents the prediction result at time t for time t+1.

[0064] Based on this prediction error, the parameters of the state identification model, prediction model, and optimization parameters can be updated online or periodically to enhance the system's adaptability to changes in ore properties, equipment status, and external disturbances.

[0065] This step enables the method to form a closed-loop control structure of "sensing-identification-prediction-optimization-execution-feedback update", thereby improving the real-time performance, stability and continuous optimization capability of the long-process lead-zinc flotation process control.

[0066] In this embodiment, the method can be used for the control of a single flotation unit, as well as for the coordinated control of different operating sections and loops in long-process lead-zinc flotation. For scenarios where there is a significant coupling relationship between the lead and zinc loops, or between roughing and cleaning, multiple control units can be incorporated into a unified optimization framework to collaboratively solve the control setpoints for each operating section, thus taking into account both local indicators and the overall control effect of the entire process.

[0067] In addition, in practical applications, when abnormalities are detected in operating condition identification, prediction results, key input information, or the optimal control setpoint exceeds the safe execution range, the current adaptive optimization process can be stopped and switched to a preset rule control mode and / or a manual confirmation control mode to improve the safety of system operation and engineering feasibility.

[0068] Please see Figure 2 Corresponding to the aforementioned embodiments of the intelligent control method for long-process lead-zinc flotation, this embodiment also provides an intelligent control system for long-process lead-zinc flotation based on multi-source information fusion and adaptive optimization, which executes the aforementioned intelligent control method for long-process lead-zinc flotation. This system includes: The data acquisition module is used to acquire foam images, process parameters, and test data in real time. The data fusion and operating condition identification module is used to fuse foam images, process parameters, and test data, and to determine the current flotation status. The intelligent decision-making module is used to generate key operating parameter settings for the lead and zinc circuits based on the current flotation status, and optimize the key operating parameter settings to obtain the optimal settings. The collaborative execution module is used to drive the actions of field equipment according to the optimal set value, so as to achieve collaborative optimization control of the entire flotation process; The feedback update module is used to collect feedback information after execution control and update the judgment process of the current flotation state and / or the generation process of the optimal setpoint based on the feedback information.

[0069] In one specific implementation, such as Figure 2 The intelligent control system for long-process lead-zinc flotation shown here is based on the construction of a closed-loop intelligent control architecture of "perception-decision-execution". The overall system architecture from top to bottom includes: Intelligent decision-making layer: performs data fusion, status identification, and optimization calculation.

[0070] Collaborative execution layer: issues decision instructions to the executing agencies.

[0071] Basic perception layer: responsible for collecting real-time data throughout the entire process.

[0072] The specific components and connections of this system are as follows: 1. Data Acquisition Module Machine vision unit: A high-definition industrial camera is installed above each key flotation cell (especially the coarsening and cleaning cells) to capture images of the foam. This unit uses Fourier transform image processing algorithms to extract visual feature parameters such as foam flow rate, size distribution, texture features, color, and stability online.

[0073] Process instrumentation unit: includes an online XRF analyzer (for rapid estimation of pulp grade at key nodes), pH meter, pulp concentration meter, flow meter, level sensor, etc., used to acquire online physicochemical parameters related to flotation process control such as grade, pH value, pulp concentration, air intake, and level.

[0074] Production data interface: Integrates with upper-level management systems (such as MES) to obtain information such as raw ore ratio and offline test data (concentrate grade, recovery rate).

[0075] 2. Data Fusion and Operating Condition Recognition Module This module receives all data from the multi-source information sensing module and has a built-in feature fusion model based on Deep Belief Network (DBN) or Long Short-Term Memory Network (LSTM) to fuse image features, process parameters and test data in a higher dimension to generate a "super feature vector" that can comprehensively characterize the current flotation conditions.

[0076] The model architecture consists of three parts: Feature extraction branch: LSTM is used to extract the temporal dynamic features of flotation process parameters; DBN is used for unsupervised extraction of foam image features and deep nonlinear features of test data.

[0077] High-dimensional feature fusion layer: The LSTM temporal features and DBN deep features are concatenated and fused in a high-dimensional space to form a comprehensive feature vector.

[0078] Decision output layer: Through mapping of the fully connected layer, output the prediction of flotation indicators (grade, recovery rate) or the result of operating condition identification.

[0079] The training process is mainly as follows: DBN Unsupervised Pre-training: Restricted Boltzmann Machine (RBM) is trained layer by layer to learn the data distribution and initialize weights.

[0080] Supervised training of LSTM: Using time series parameters as input, it learns the temporal dependencies.

[0081] Global joint fine-tuning: Fusing two types of features, the entire network is optimized end-to-end using test values ​​as labels to minimize prediction error.

[0082] 3. Intelligent Decision-Making Module The intelligent decision-making module adopts a hybrid intelligent decision-making mechanism: Case-Based Reasoning (CBR) Unit: Stores a large number of historical best-practice operation cases (each case includes operating condition characteristics, operating parameters, and performance indicators). When the system identifies the current operating condition, the CBR unit retrieves the most similar historical case from the case library and uses its operating parameters as the initial recommended values.

[0083] Fuzzy rule base: It encapsulates the operational experience of domain experts and forms fuzzy rules of "IF (condition) THEN (action)". For example, "IF foam flow rate is slow AND foam color is dark THEN increase the amount of collector appropriately".

[0084] The adaptive optimizer uses a dynamic, mechanism- and data-driven hybrid model as the predictive model. Employing either Model Predictive Control (MPC) or Reinforcement Learning (RL) algorithms, it aims to predict the optimal combination of concentrate grade and recovery rate over a future period. It performs rolling optimization on the initial setpoints given by the CBR and fuzzy rules to calculate the final optimal control setpoints. This optimizer can adjust its parameters online according to changes in process characteristics, possessing self-learning and adaptive capabilities.

[0085] Specifically, the optimizer model is a hybrid model architecture, consisting of a mechanistic model and a data-driven model in parallel. The mechanistic part describes the basic physical laws of the process, including flotation kinetics, material balance, and reagent action mechanisms. The data-driven part is a DBN-LSTM feature fusion model that fits nonlinear, time-varying, and uncertain characteristics. The two are fused through weighted or residual compensation to form a global prediction model.

[0086] Operation process: Input: foam image features, process parameters, test values, and operating condition category; Mechanism model output: basic predicted values; Data-driven model compensates for nonlinear errors and time delays; Fusion output: prediction of future multi-step concentrate grade and recovery rate.

[0087] Optimal setpoint calculation: The objective function is to optimize the comprehensive performance index of grade and recovery rate; MPC: The optimal control sequence is solved by rolling within the control constraints; RL: The optimal action strategy is learned through interactive trial and error with the environment; The initial values ​​given by CBR and fuzzy rules are iteratively corrected to obtain the optimal setpoint sequence.

[0088] Actual control: The optimized setpoints are converted into field-executable parameters through actuator mapping relationships, such as reagent dosage (collector, inhibitor, frother), aeration rate, slurry level, and frothing speed. These parameters are then sent to the collaborative execution layer to achieve closed-loop control.

[0089] 4. Collaborative Execution Module This module receives control settings from the intelligent decision-making module and distributes them to various actuators, including: frequency converters or valves of the reagent addition device, aeration valves, liquid level regulating gates, pump pool water supply valves, etc.

[0090] 5. Feedback Update Module The system also establishes a closed-loop feedback mechanism to feed back the results of the execution (data collected again through the perception layer and subsequent test results) to the intelligent decision-making module, which is used to evaluate the control effect, update the case library and optimize the model, so as to realize the continuous evolution of the system.

[0091] A large lead-zinc ore beneficiation plant with a processing capacity of 2000 tons / day has a long flotation process, severe lead-zinc intermingling, and stagnant recovery rates. After applying the system described in this invention, the following results were achieved: System Deployment: A total of 12 machine vision units were installed on the roughing and cleaning tanks for lead and zinc, and connected to the existing online pH meters, concentration meters, and XRF analyzers. An intelligent decision server was deployed in the central control room to implement closed-loop control of 6 reagent addition points (collectors, inhibitors, etc.) and 8 aeration control points.

[0092] Implementation process: 1. Data Acquisition and Model Initialization: First, a month-long pure data collection was conducted to accumulate data under different working conditions, which was used to initialize the case library and train the hybrid prediction model.

[0093] 2. Trial operation and parameter tuning: When the system enters the trial operation phase, operators can confirm or fine-tune the set values ​​given by the system on the control room interface. The system learns from these human interventions and further optimizes the decision-making logic.

[0094] 3. Fully automatic operation: After the trial run is stable, switch to fully automatic mode.

[0095] Effect comparison: Indicator stability: The standard deviation of zinc concentrate grade decreased from 0.8% to 0.3%, and the operating curve became very smooth.

[0096] Improved recovery rate: Despite a slight decrease in the zinc grade of the raw ore, the zinc recovery rate steadily increased from 96.4% to 96.8%.

[0097] Chemical savings: The consumption of zinc loop collector (butyl xanthate) per ton of ore decreased by about 3%.

[0098] Labor productivity: The flotation workshop has basically achieved "unmanned" inspection and operation, requiring only one person per shift for monitoring, which greatly reduces labor intensity.

[0099] This example fully demonstrates the significant application value of this system in improving the economic indicators of long-process lead-zinc flotation technology, reducing costs and increasing efficiency.

[0100] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0101] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0102] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A long-process intelligent control method for lead-zinc flotation based on multi-source information fusion and adaptive optimization, characterized in that, Includes the following steps: Multi-source information is acquired during the long-process lead-zinc flotation process, and the multi-source information is processed and fused to obtain condition characteristic information for characterizing the current flotation conditions. Based on the aforementioned operating condition characteristic information, the current flotation operating condition is identified, and the control target indicators of the flotation process are predicted. Based on the operating conditions and the control target indicators, control setpoints for the flotation process are generated, and the optimal control setpoints are obtained after adaptive optimization. The optimal control setpoint is converted into an execution command and sent to the flotation actuator to control the long-process lead-zinc flotation process.

2. The intelligent control method for long-process lead-zinc flotation according to claim 1, characterized in that, The method further includes: collecting feedback information after execution control, and updating the state identification and / or the adaptive optimization based on the feedback information to form closed-loop control.

3. The intelligent control method for long-process lead-zinc flotation according to claim 1, characterized in that, The processing and fusion of the multi-source information includes: The flotation foam image information is denoised, corrected, and the region of interest is extracted. The flotation process parameter information is processed for outliers, compensated for missing values, and standardized. Time alignment and time delay compensation are performed on information from different sampling frequencies; The processed multi-source information is spliced ​​or mapped to construct working condition feature information.

4. The intelligent control method for long-process lead-zinc flotation according to claim 1, characterized in that, The process of generating the control setpoint includes: Based on the operating condition status, retrieve operating condition cases similar to the current operating condition from historical operating cases, and generate recommended control settings based on the operating parameters corresponding to the similar operating condition cases. The recommended control setting value for the case is modified according to the preset control rules corresponding to the current operating condition to obtain the control setting value.

5. The intelligent control method for long-process lead-zinc flotation according to claim 1, characterized in that, Adaptive optimization of the control setpoint includes: Establish a predictive process to characterize future trends in long-process lead-zinc flotation; The current operating condition characteristic information, operating condition status identification results and initial control setpoints are input into the prediction process to obtain the index prediction results for future control cycles. Based on the predicted results of the indicators, the initial control setpoint is iteratively corrected under preset control constraints to obtain the optimal control setpoint. The optimal control setpoint uses the comprehensive control effect of concentrate grade, recovery rate and reagent consumption as the optimization objective.

6. The intelligent control method for long-process lead-zinc flotation according to claim 1, characterized in that, The state identification and prediction of the control target index are achieved based on a multi-source feature fusion model, which includes: LSTM branch used to extract the temporal features of flotation process parameters; DBN branch used to extract features from foam images and / or deep nonlinear features from laboratory data; A decision output layer used to fuse the outputs of the LSTM branch and the DBN branch and output the operating condition identification result and / or the prediction result of the control target related index.

7. The intelligent control method for long-process lead-zinc flotation according to claim 1, characterized in that, The control settings include at least one of the following: reagent addition amount, aeration amount, slurry level, slurry flow rate, and stirring intensity.

8. The intelligent control method for long-process lead-zinc flotation according to claim 1, characterized in that, The step of generating control setpoints for the flotation process based on the operating conditions and the control target indicators includes: collaboratively determining the control setpoints for each operating segment by combining the correlation between different operating segments and / or lead circuits and zinc circuits in long-process lead-zinc flotation.

9. The intelligent control method for long-process lead-zinc flotation according to claim 1, characterized in that, When an abnormality in operating condition identification, prediction result, key input information, or optimal control setpoint is detected, the current adaptive optimization process is stopped, and the system switches to preset rule control mode and / or manual confirmation control mode.

10. A long-process lead-zinc flotation intelligent control system based on multi-source information fusion and adaptive optimization, used to execute the long-process lead-zinc flotation intelligent control method as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire foam images, process parameters, and test data in real time. The data fusion and operating condition identification module is used to fuse the foam image, process parameters and test data, and determine the current flotation state; The intelligent decision-making module is used to generate key operating parameter settings for the lead circuit and zinc circuit based on the current flotation state, and to optimize the key operating parameter settings to obtain the optimal settings. The collaborative execution module is used to drive the field equipment to operate according to the optimal set value, so as to achieve collaborative optimization control of the entire flotation process; The feedback update module is used to collect feedback information after execution control, and update the judgment process of the current flotation state and / or the generation process of the optimal set value based on the feedback information.

Images