Gold mine flotation full process intelligent monitoring and optimization control method and system

Abstract

The present application relates to the technical field of monitoring and controlling the entire process of gold flotation, and discloses a method and system for intelligent monitoring and optimization control of the entire process of gold flotation. The method comprises: collecting flotation process parameters, and performing classification processing through outlier detection; fusing the classified data to establish a digital twin model; designing a negotiation control architecture based on the model, and adjusting the parameters using a contract network protocol; constructing a soft measurement model and optimization strategy, and optimizing the control parameters through a self-learning mechanism. The present application solves the technical problems in existing gold flotation control systems, such as the difficulty in achieving global optimization due to independent control of units, the inaccurate control decisions due to simple data processing, the difficulty in estimating key quality indicators due to the lack of soft measurement methods, and the poor adaptability due to the limited learning ability of the system.

Description

Technical Field

[0001] The present application relates to the technical field of full-process monitoring and control of gold mine flotation, and in particular to a method and system for intelligent monitoring and optimization control of the full-process of gold mine flotation. Background Art

[0002] Gold flotation is a crucial step in gold beneficiation. Its process includes multiple stages, including crushing, grinding and classification, flotation, and concentrate processing. Traditional gold flotation control systems typically employ unit-level PID control, adjusting parameters for individual control points, such as pH controllers, pulp concentration controllers, and reagent addition controllers. With the advancement of automation technology, intelligent control methods based on fuzzy control, expert systems, and neural networks have been gradually introduced into the gold flotation process, improving control accuracy. Some advanced flotation plants also employ foam image analysis technology based on visual recognition to indirectly evaluate flotation performance by monitoring flotation foam characteristics. They are also beginning to experiment with the application of digital twin technology for virtual simulation and optimization of the gold flotation process.

[0003] However, existing systems mostly adopt a unit-independent control mode and lack a collaborative control mechanism for the entire process. There is insufficient information exchange between the control units, making it difficult to achieve global optimization. Secondly, the existing system handles process parameters in a simple way, the data quality is not high, and the ability to handle outliers and noise is limited, which affects the accuracy of control decisions. Third, the existing system has insufficient adaptability to complex working conditions. When the ore properties change, the equipment ages, and so on, the control effect is significantly reduced, and frequent manual intervention is required. Fourth, the existing system lacks effective soft measurement methods, making it difficult to accurately estimate key quality indicators such as concentrate grade and recovery rate, resulting in unclear control objectives. Finally, the existing system has limited learning capabilities and cannot accumulate optimization experience from historical operating data. The control strategy lacks a self-improvement mechanism.

[0004] In traditional control systems, the internal structure design of the intelligent bodies of each functional unit is not perfect, and there is a lack of effective data processing modules and decision-making reasoning modules, which leads to the inability to accurately perceive the environmental status and make reasonable decisions; at the same time, the loading method of the expert knowledge base, control algorithm library and fuzzy rule library in the existing technology is not flexible enough to adapt to the requirements of different working conditions; in addition, the boundary between the local optimization control layer and the collaborative optimization control layer is not clear, and the control logic of the two layers is cross-confused, resulting in conflicts between local goals and global goals. Summary of the Invention

[0005] The present application provides a method and system for intelligent monitoring and optimization control of the entire gold mine flotation process, which is used to solve the technical problems in existing gold mine flotation control systems, such as the difficulty in achieving global optimization due to independent control of units, inaccurate control decisions due to simple data processing, difficulty in estimating key quality indicators due to the lack of soft measurement methods, and poor adaptability due to limited system learning ability.

[0006] In the first aspect, the present application provides a method for intelligent monitoring and optimization control of the entire gold mine flotation process, which includes: collecting process parameter data in the entire gold mine flotation process through a distributed sensor network, performing outlier detection and hierarchical processing on the collected data to obtain hierarchical processing data; constructing a multi-source heterogeneous data fusion engine based on the hierarchical processing data, standardizing and multi-level fusing the data to obtain a digital twin model of the entire gold mine flotation process; designing a multi-agent negotiation control architecture based on the digital twin model, and using a negotiation mechanism based on a contract network protocol to adjust the control parameters to obtain a global optimization control scheme; constructing an adaptive soft measurement model and a multi-objective optimization control strategy according to the global optimization control scheme, realizing parameter optimization through self-learning and knowledge accumulation mechanism, and obtaining process control parameters.

[0007] In a second aspect, the present application provides a full-process intelligent monitoring and optimization control system for gold flotation, which includes:

[0008] The classification module is used to collect process parameter data of the entire gold flotation process through a distributed sensor network, perform outlier detection and classification processing on the collected data, and obtain classified processing data;

[0009] A fusion module is used to build a multi-source heterogeneous data fusion engine based on the hierarchically processed data, standardize and fuse the data at multiple levels, and obtain a digital twin model of the entire gold mine flotation process;

[0010] An adjustment module is used to design a multi-agent negotiation control architecture based on the digital twin model, and to adjust control parameters using a negotiation mechanism based on the contract network protocol to obtain a globally optimized control solution;

[0011] The construction module is used to construct an adaptive soft measurement model and a multi-objective optimization control strategy according to the global optimization control scheme, realize parameter optimization through self-learning and knowledge accumulation mechanism, and obtain process control parameters.

[0012] In a third aspect, a full-process intelligent monitoring and optimization control device for gold mine flotation is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the full-process intelligent monitoring and optimization control device for gold mine flotation executes the above-mentioned full-process intelligent monitoring and optimization control method for gold mine flotation.

[0013] In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium. When the computer-readable storage medium is run on a computer, the computer executes the above-mentioned method for intelligent monitoring and optimization control of the entire gold flotation process.

[0014] In the technical solution provided by this application, process parameter data is collected through a distributed sensor network, and outlier detection and hierarchical processing are performed on the data, which effectively improves the data quality and solves the problem of simple data processing and low quality in traditional systems. A multi-source heterogeneous data fusion engine is constructed based on the hierarchical processing data, and the data is standardized and multi-level fused to obtain a digital twin model of the entire gold mine flotation process, realizing real-time mapping between the physical flotation system and the virtual system, and providing an accurate information basis for subsequent decision-making. A multi-agent negotiation control architecture is designed based on the digital twin model, and a negotiation mechanism based on the contract network protocol is adopted to adjust the control parameters, effectively solving the problem of insufficient information exchange among the control units in the traditional system and difficulty in achieving global optimization. The negotiation mechanism significantly improves the collaborative performance of the overall system, enabling each functional unit to jointly achieve the global optimal goal while maintaining local stability. An adaptive soft measurement model and a multi-objective optimization control strategy are constructed according to the global optimization control scheme, and parameter optimization is achieved through self-learning and knowledge accumulation mechanisms. This not only solves the problem of the lack of effective soft measurement means in the traditional system, but also overcomes the defect that the control strategy lacks a self-improvement mechanism. In particular, the contract network protocol used in the negotiation control architecture, as a multi-agent negotiation mechanism, fully considers the characteristics of each functional unit in the flotation process and their mutual influence. Through structured message interaction and task decomposition negotiation process, it effectively balances multiple conflicting goals such as grade, recovery rate, and energy consumption. The introduction of the negotiation mechanism enables the system to have distributed decision-making capabilities, significantly reduces the computational burden, and improves the system response speed; and the adaptive soft measurement model realizes accurate estimation of parameters that are difficult to measure directly through multi-model integration and transfer learning methods. The self-learning mechanism enables the system to accumulate optimization experience from historical operation data, improve its adaptability to complex working conditions, and reduce energy consumption and reagent consumption while improving the grade and recovery rate of concentrate. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0016] Figure 1 This is a schematic diagram of an embodiment of the method for intelligent monitoring and optimization control of the entire gold flotation process in the embodiment of the present application;

[0017] Figure 2 This is a schematic diagram of an embodiment of the intelligent monitoring and optimization control system for the entire gold flotation process in the embodiment of the present application;

[0018] Figure 3 It is a schematic block diagram of the structure of the intelligent monitoring and optimization control equipment for the entire process of gold flotation in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The embodiments of the present application provide a method and system for intelligent monitoring and optimization control of the entire process of gold flotation. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or are inherent to these processes, methods, products or devices.

[0020] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the method for intelligent monitoring and optimization control of the entire gold flotation process includes:

[0021] Step S101: collecting process parameter data in the entire gold flotation process through a distributed sensor network, performing outlier detection and classification processing on the collected data to obtain classified processing data;

[0022] Step S102: construct a multi-source heterogeneous data fusion engine based on the hierarchically processed data, standardize and fuse the data at multiple levels, and obtain a digital twin model of the entire gold mine flotation process;

[0023] Step S103: design a multi-agent negotiation control architecture based on the digital twin model, and use a negotiation mechanism based on the contract network protocol to adjust control parameters to obtain a global optimal control solution;

[0024] Step S104: construct an adaptive soft measurement model and a multi-objective optimization control strategy according to the global optimization control scheme, realize parameter optimization through self-learning and knowledge accumulation mechanism, and obtain process control parameters.

[0025] It is understandable that the execution subject of this application can be a gold mine flotation full process intelligent monitoring and optimization control system, or a terminal or server, which is not limited here. The embodiment of this application is described by taking the server as the execution subject as an example.

[0026] Specifically, a distributed sensor network collects process parameter data from the entire gold flotation process. Outlier detection and classification are performed on the data collected by the sensor network. In practice, process parameter data, such as pH, slurry concentration, grinding fineness, and foam characteristics, are classified according to their rate of change. Fast-changing parameters, such as pH and aeration volume, are sampled at a high frequency of 500ms, while slowly changing parameters, such as slurry concentration, are sampled at a low frequency of 5s. This categorized sampling method avoids data redundancy and information loss associated with traditional fixed-frequency sampling. The collected data is then analyzed for outliers using an improved Z-score method. This method calculates the normalized distance between the parameter value and the sliding window mean. When this distance exceeds a threshold of 4.5, the data point is marked as an outlier. For example, if the pH value suddenly jumps from a normal value of 8.2 to 11.5 during a flotation run, the calculated Z-score is 5.3, exceeding the threshold of 4.5 and thus marking the data point as an outlier. The detected outliers are then replaced, and the replacement values are calculated using the weighted average method for the 20 valid data points before and after. The weight coefficient is determined by the exponential decay function, so that the closer the data is to the outlier, the smaller the weight. The data is subjected to noise elimination by the fourth-order Kalman filter algorithm, and the filter parameter matrix is adjusted according to the characteristics of different parameters. The filtered data is graded using the data importance scoring function. The scoring function comprehensively considers three factors: the influence of the parameters on the process, the amplitude of the change, and the frequency of fluctuation. Parameter data with a score greater than 0.8 are ultimately classified as level one, parameter data with a score between 0.5 and 0.8 are classified as level two, and parameter data with a score below 0.5 are classified as level three, forming hierarchical processing data with priority.

[0027] A multi-source, heterogeneous data fusion engine was built based on the hierarchical processing data to standardize and fuse the data at multiple levels. First, the hierarchical processing data was converted to units: grinding fineness data was converted from μm to mm, slurry concentration data was converted from percentage to decimal, and potential data was converted to standard potential units to ensure consistent units. The data with consistent units was then normalized to adjust the numerical ranges of different parameters to the 0-1 range, eliminating the impact of dimensional differences on subsequent analysis. The normalized data was time-aligned to align data with different sampling frequencies according to the system clock reference point and interpolate sampling intervals to form standardized data that was consistent in both time and space. Within the standardized data, a weighted average of the data from multiple sensors measuring the same parameter was calculated, with weights assigned based on the historical accuracy of each sensor. This yielded a data-level fusion result. Correlation analysis of the data-level fusion results identified strong correlations between parameter combinations, such as the correlation coefficient between slurry pH and froth stability, which reached 0.87. Key features characterizing the flotation process were then extracted to generate feature-level fusion results. The feature-level fusion results are combined with the expert knowledge of flotation technology to establish the correspondence between the physical flotation system and the virtual system, and obtain a digital twin model of the entire gold mine flotation process.

[0028] A multi-agent negotiation control architecture was designed based on a digital twin model, employing a negotiation mechanism based on a contract network protocol for control parameter adjustment. The gold flotation process was first divided into four main functional units: the crushing control unit, the grinding and classification control unit, the flotation control unit, and the concentrate processing unit. Each unit was equipped with an independent decision-making controller and actuator. An intelligent agent was constructed by loading an expert knowledge base, a control algorithm library, and a fuzzy rule library into each functional unit. The internal structure of the intelligent agent comprises a data processing module, a decision-making and reasoning module, and an execution control module, with a two-layer structure consisting of a local optimization control layer and a collaborative optimization control layer. The expert knowledge base stores control knowledge for each functional unit, including process parameter control ranges, exception handling rules, and optimization strategies. For example, control knowledge such as the crusher feed size range of 10-30 mm, the slurry concentration range of 35%-45%, and the flotation pH range of 7-9, along with corresponding exception handling rules. The control algorithm library includes PID control algorithms, fuzzy control algorithms, and model predictive control algorithms. The fuzzy rule library uses process parameter deviations and deviation change rates as inputs and performs fuzzy reasoning to derive control adjustment values. Based on the digital twin model, each agent is assigned initial control objectives. The concentrate processing unit agent generates task requests based on production metrics, including target concentrate grade and recovery rates. These task requests are then passed down through the hierarchy. Upon receiving these requests, the flotation unit agent evaluates the current operating conditions and generates a flotation plan and resource requirements, including parameters such as flotation reagent dosage, pH value, and aeration volume. Each agent then decomposes and negotiates tasks according to the interaction rules defined by the contract network protocol. The Pareto multi-objective optimization method is used to balance conflicting objectives and reach a consensus on a globally optimal control solution.

[0029] Based on a global optimization control scheme, an adaptive soft-sensing model and a multi-objective optimization control strategy were constructed, achieving parameter optimization through self-learning and knowledge accumulation mechanisms. First, a multi-model integrated soft-sensing system was constructed to estimate key quality indicators such as concentrate grade and recovery, which are difficult to measure directly. The soft-sensing model consists of a theoretical model based on flotation dynamics and a regression model based on historical data. The results of these two models are weighted and fused to produce the final estimated value. The quality indicators estimated by the soft-sensing system are combined with actual process parameters to construct a hierarchical control structure consisting of a real-time control layer, a tactical optimization layer, and a strategic decision-making layer. The control cycles of each layer are 100 ms, 2 hours, and 8 hours, respectively. The real-time control layer uses an improved model predictive control algorithm with a prediction horizon of 30 steps and a control horizon of 5 steps. The prediction model is dynamically updated based on the corresponding relationship between process parameters and quality indicators. The tactical optimization layer applies a multi-objective optimization algorithm with four optimization objectives: maximizing concentrate grade, maximizing recovery, minimizing energy consumption, and minimizing reagent consumption. The objective function weights are dynamically adjusted based on current production needs. Real-time evaluation of control effectiveness is performed. By comparing the deviation between actual production indicators and optimization targets, a database of operational cases is established, encompassing operating condition characteristics, control parameter configurations, and control effects. Based on this database and soft-sensor feedback, model parameters are incrementally updated. Control strategies are adjusted based on the control effect evaluation results, forming a self-learning closed loop and achieving continuously optimized process control parameters.

[0030] In an embodiment of the present application, process parameter data is collected through a distributed sensor network, and outlier detection and hierarchical processing are performed on the data, which effectively improves the data quality and solves the problem of simple data processing and low quality in traditional systems; a multi-source heterogeneous data fusion engine is constructed based on the hierarchical processing data, and the data is standardized and multi-level fused to obtain a digital twin model of the entire gold mine flotation process, realizing real-time mapping between the physical flotation system and the virtual system, and providing an accurate information basis for subsequent decision-making; a multi-agent negotiation control architecture is designed based on the digital twin model, and a negotiation mechanism based on the contract network protocol is adopted to adjust the control parameters, which effectively solves the problem of insufficient information exchange among the control units in the traditional system and difficulty in achieving global optimization. The negotiation mechanism significantly improves the collaborative performance of the overall system, enabling each functional unit to achieve the global optimal goal while maintaining local stability; an adaptive soft measurement model and a multi-objective optimization control strategy are constructed according to the global optimization control scheme, and parameter optimization is achieved through self-learning and knowledge accumulation mechanisms, which not only solves the problem of lack of effective soft measurement means in the traditional system, but also overcomes the defect of lack of self-improvement mechanism of the control strategy. In particular, the contract network protocol used in the negotiation control architecture, as a multi-agent negotiation mechanism, fully considers the characteristics of each functional unit in the flotation process and their mutual influence. Through structured message interaction and task decomposition negotiation process, it effectively balances multiple conflicting goals such as grade, recovery rate, and energy consumption. The introduction of the negotiation mechanism enables the system to have distributed decision-making capabilities, significantly reduces the computational burden, and improves the system response speed; and the adaptive soft measurement model realizes accurate estimation of parameters that are difficult to measure directly through multi-model integration and transfer learning methods. The self-learning mechanism enables the system to accumulate optimization experience from historical operation data, improve its adaptability to complex working conditions, and reduce energy consumption and reagent consumption while improving the grade and recovery rate of concentrate.

[0031] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0032] The collected process parameter data of pH value, slurry concentration, grinding fineness, and foam characteristics are classified according to the rate of change. The sampling frequency is set to 500ms / time for the fast-changing parameters and 5s / time for the slow-changing parameters.

[0033] The improved Z-score method is applied to the classified process parameter data for outlier detection. By calculating the standardized distance between the parameter value and the sliding window mean, data points exceeding the threshold of 4.5 are marked as outliers.

[0034] The detected outlier data point is replaced by the weighted average of the 20 valid data points before and after, and the weight coefficient is calculated by the exponential decay function;

[0035] The fourth-order Kalman filter algorithm is applied to the data after replacing the outliers to eliminate noise, and the filter parameter matrix is adjusted according to the data type;

[0036] The filtered data is graded based on the data importance scoring function, which comprehensively considers the three factors of parameter influence on process, variation range and fluctuation frequency;

[0037] Parameter data with a score greater than 0.8 is classified as level one, parameter data with a score between 0.5 and 0.8 is classified as level two, and parameter data with a score lower than 0.5 is classified as level three, forming hierarchical processing data with priority.

[0038] Specifically, the collected process parameter data is classified based on their rate of change. Parameters such as pH, pulp concentration, grinding fineness, and foam characteristics change at different rates during the flotation process, necessitating targeted sampling frequencies. The rate of change is determined based on the time derivative of the parameter, calculating the rate of change over a period of time. Parameters greater than a set threshold are defined as rapidly changing parameters, such as pH and aeration volume, and are sampled at a high frequency of 500ms / time. Parameters less than a set threshold are defined as slowly changing parameters, such as pulp concentration and grinding fineness, and are sampled at a low frequency of 5s / time. This classified sampling strategy addresses the data redundancy or information loss issues associated with traditional fixed-frequency sampling and is consistent with the actual variation characteristics of process parameters. After being classified by rate of change, the collected data enters the outlier detection phase. An improved Z-score method is used to detect outliers in process parameter data. While the traditional Z-score method calculates the global mean and standard deviation, the improved Z-score method uses a sliding window mechanism. For each data point, the mean and standard deviation of the preceding N points are calculated, and then the normalized distance between that point and the window mean is calculated. Specifically, the 30 data points preceding the current data point are used as a sliding window, the mean and standard deviation of the data within the window are calculated, and then the Z value of the current data point is calculated: the current value minus the window mean and divided by the window standard deviation. When the absolute value of the Z value exceeds the threshold of 4.5, the data point is marked as an outlier. Compared to traditional methods, the improved Z-score method is more adaptable to the dynamic changes in parameters during the flotation process and can effectively identify local anomalies.

[0039] For the detected outlier data points, the weighted average replacement method is used for processing. Take 10 valid data points before and after the outlier point, a total of 20 points, and calculate the weight coefficient of each point through the exponential decay function. The form of the exponential decay function is weight Wi = , where i represents the distance between the data point and the outlier, and λ is the decay rate parameter, typically set to 0.2, so that data points farther from the outlier have a greater weight, and vice versa. This approach takes into account the potential impact of outliers on surrounding data points, mitigating this impact by reducing the weights of nearby points, ensuring that the replacement value is more consistent with normal trends.

[0040] After outlier replacement, the data undergoes noise removal using a fourth-order Kalman filter. Kalman filtering is a recursive estimation algorithm that builds a system state model and a measurement model, combining prior estimates with current measurements to arrive at an optimal estimate. A fourth-order Kalman filter uses a four-dimensional state vector, consisting of the parameter value, first-order derivative, second-order derivative, and third-order derivative. This allows for a more accurate description of the dynamic trends of the parameters. The filter parameter matrices, including the process noise covariance matrix Q and the measurement noise covariance matrix R, require adjustment for different data types. For rapidly changing pH data, R is set to a smaller value to increase confidence in the measured values; for more stable slurry concentration data, Q is set to a smaller value to increase confidence in the model predictions. By appropriately adjusting these parameters, the Kalman filter algorithm can effectively remove noise from various process parameter data. The filtered data then enters a classification process, where it is classified based on a data importance scoring function. This scoring function comprehensively considers three factors: the parameter's impact on the process, the magnitude of its variation, and the frequency of fluctuation. The parameter's impact on the process is a weighted coefficient determined based on expert experience and statistical analysis of historical data, indicating the parameter's influence on flotation performance. The amplitude of variation is the degree to which the parameter value deviates from the normal operating range. The frequency of fluctuation is the number of times the parameter value fluctuates within a unit of time. The scoring function weights and sums these three factors according to a certain ratio to obtain a parameter data importance score.

[0041] Parameter data is graded according to their importance scores. Parameter data with a score greater than 0.8 is classified as level one, indicating parameters that have a significant impact on the process and require high-priority processing, such as pH value and reagent addition amount. Parameter data with a score between 0.5 and 0.8 is classified as level two, indicating parameters that have a certain impact on the process but are not the most critical, such as foam height and aeration volume. Parameter data with a score below 0.5 is classified as level three, indicating auxiliary parameters with less impact or slow changes, such as temperature and pressure. The graded data has different transmission priorities and processing strategies, ensuring that key parameter data can be transmitted and processed in a timely manner, solving the problems of indiscriminate data processing and delayed key information in traditional systems.

[0042] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0043] Perform unit conversion on the classification processing data, convert the grinding fineness data into millimeters, convert the slurry concentration data into decimal form, and convert the potential data into standard potential units to obtain data with unified units;

[0044] Normalize the data with unified units, adjust the numerical ranges of different parameters to the same interval, eliminate the influence of dimensional differences on subsequent analysis, and obtain normalized data;

[0045] Perform time alignment on the normalized data, align data with different sampling frequencies according to the system clock, and interpolate and fill the sampling intervals to obtain standardized data that is consistent in time and space;

[0046] The weighted average calculation is performed on the data of the same parameter measured by multiple sensors in the standardized data, and the weight is assigned according to the historical accuracy of each sensor to obtain the data-level fusion result;

[0047] Analyze the correlation between different parameters in the data-level fusion results, identify parameter combinations with strong correlation, extract key features that characterize the flotation process status, and obtain feature-level fusion results;

[0048] The feature-level fusion results are combined with the expert knowledge of flotation technology to judge the current flotation process status, establish the correspondence between the physical flotation system and the virtual system, and obtain a digital twin model of the entire gold mine flotation process.

[0049] Specifically, data is converted to different units. Data collected by different sensors often have different units. For example, grinding fineness data may be expressed in microns (μm) and needs to be converted to millimeters (mm) by dividing the original value by 1000. Slurry concentration data is often expressed as a percentage and needs to be converted to decimal form by dividing the percentage by 100. Potential data may vary depending on the reference electrode and needs to be converted to standard potential units by adding or subtracting the corresponding reference electrode potential difference. The purpose of unit conversion is to establish a unified measurement standard to avoid data interpretation errors and calculation errors caused by different units. Even after unit standardization, data still have dimensional differences, and the numerical ranges of different parameters vary greatly. For example, pH values are typically in the range of 0-14, while grinding fineness can range from tens to hundreds of microns. Therefore, data with unified units needs to be normalized to adjust the numerical ranges of different parameters to the same range. Normalization uses the maximum and minimum value normalization method. For each parameter, the maximum and minimum values of its historical data are found, and the current value is then mapped to the interval [0,1]. The calculation process is to subtract the historical minimum from the current value, then divide it by the difference between the historical maximum and minimum values. For parameters without clear upper and lower limits, a sliding window method is used to normalize the parameters by taking the maximum and minimum values over a recent period. This normalization eliminates the impact of dimensional differences on subsequent analysis and allows direct comparison and calculation between different parameters.

[0050] Normalized data also faces the problem of time inconsistency because different parameters have different sampling frequencies. For example, pH is sampled every 500ms, while slurry concentration is sampled every 5s, resulting in data timestamp misalignment. To address this issue, time alignment is employed. First, the system reference clock is determined, typically using the highest sampling frequency as the reference. Then, all parameter data is aligned to the reference clock. Parameters with sampling frequencies lower than the reference clock require interpolation. Interpolation methods include linear interpolation, cubic spline interpolation, and nearest neighbor interpolation. The appropriate interpolation method should be selected based on the parameter's variation characteristics. Linear interpolation is suitable for parameters with relatively gradual variations. It calculates the linear relationship between two adjacent sampling points and derives the parameter value at the intermediate time. Cubic spline interpolation is suitable for parameters that require curvature continuity. Nearest neighbor interpolation is suitable for discrete parameters. After time alignment, all parameter data remains consistent on the time axis, forming standardized data that is consistent in time and space.

[0051] Standardized data may contain multiple sensors measuring the same parameter, such as pH sensors distributed at different locations, or a primary and backup sensor installed at a single location. In these cases, a weighted average calculation is required to obtain a unique, accurate value for the parameter. Weighting is based on the historical accuracy of each sensor. The calculation method involves analyzing the deviation between each sensor's historical data and the standard sample's measurement value. Sensors with smaller deviations are more accurate and are assigned a higher weight. Factors such as sensor operating time and failure rate are also considered, with sensors with stable operation receiving a higher weight. The weighted average calculation formula is: the final parameter value is the sum of each sensor's measurement value multiplied by the corresponding weight coefficient. This approach effectively fuses multi-source data, improves the accuracy of parameter estimation, and produces a data-level fusion result. The data-level fusion result contains a large number of parameters, and the next challenge is to extract key features from it. This approach involves analyzing the correlations between different parameters and identifying parameter combinations with strong correlations. Correlation analysis is used to calculate the Pearson correlation coefficient between parameters. A correlation coefficient close to 1 or -1 indicates a strong correlation, while a coefficient close to 0 indicates a weak correlation. For highly correlated parameter groups, only one parameter can be retained as a representative, or a combined feature can be constructed; for weakly correlated but important parameters, each can be retained individually. In this way, the data dimension is reduced, key features that characterize the flotation process state are extracted, and feature-level fusion results are formed.

[0052] The feature-level fusion results are combined with expert knowledge of the flotation process to establish a digital twin model of the entire gold mine flotation process. First, the expert knowledge is structured, including the causal relationships between parameters, the mapping between parameters and process states, and the rules for determining abnormal states. Then, based on the feature-level fusion results, expert knowledge rules are applied to determine the current flotation process status. Process states fall into three main categories: normal, sub-healthy, and abnormal. Each category is further subdivided into multiple specific states, such as over-foaming, under-foaming, and insufficient reagents. The state judgment results are used to establish a correspondence between the physical flotation system and the virtual system. The virtual system is updated in real time based on the status of the physical system, forming a digital twin model. The digital twin model not only captures the current state but also historical trends and predicts the future.

[0053] For example, in a gold flotation production line, if the pH sensor records a pH value of 8.2, the slurry concentration sensor records a concentration of 42%, the grinding fineness sensor records a fineness of 65 μm, and foam image analysis indicates a foam stability of 0.75, units are first converted: the pH remains unchanged, the slurry concentration is converted to 0.42, and the grinding fineness is converted to 0.065 mm. The data is then normalized: for a historical pH range of [6.5, 9.5], the normalized value is (8.2-6.5) / (9.5-6.5)=0.57; for a historical slurry concentration range of [0.35, 0.50], the normalized value is (0.42-0.35) / (0.50-0.35)=0.47. Similar normalization is performed for grinding fineness and foam stability. Because these parameters have different sampling frequencies, time alignment is performed, using a 500ms interval as the benchmark, and linear interpolation is performed to fill in the gaps between low-frequency sampling data, such as slurry concentration. For multiple pH sensors, weights are assigned based on their historical accuracy, for example, sensor A is weighted 0.6 and sensor B is weighted 0.4. The resulting fused pH value is 8.2 × 0.6 + 8.1 × 0.4 = 8.16. Correlation analysis revealed a strong correlation between pH and foam stability, with a correlation coefficient of 0.87. Therefore, the two are combined into a single feature. A significant correlation also exists between grinding fineness and recovery rate. These features are combined with expert knowledge rules to determine the current process state as "normal-ideal foam," corresponding to the normal operating range in the digital twin model. The flotation state within the next hour is then predicted based on parameter trends. This data fusion approach effectively addresses the problems of data heterogeneity, information silos, and inaccurate state judgment in traditional flotation control systems.

[0054] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0055] The entire gold flotation process is divided into four main functional units: crushing control unit, grinding and classification control unit, flotation control unit and concentrate processing control unit. Each unit is equipped with an independent decision controller and actuator.

[0056] Each functional unit is loaded with an expert knowledge base, a control algorithm library, and a fuzzy rule library, and an intelligent agent corresponding to each functional unit is constructed. Each intelligent agent has a local optimization control layer and a collaborative optimization control layer.

[0057] Based on the digital twin model, each agent is assigned an initial control target. The concentrate processing unit agent generates task requests based on production indicator requirements, including target values for concentrate grade and recovery rate.

[0058] The task request is passed hierarchically. After receiving the request, the flotation unit agent evaluates the current working conditions and generates a flotation plan and resource requirements including the amount of flotation reagent, pH value and aeration volume.

[0059] According to the interaction rules set by the contract network protocol, each agent decomposes and negotiates tasks through a message passing mechanism, uses the Pareto multi-objective optimization method to weigh conflicting objectives, and each agent submits a plan in turn and evaluates the plan;

[0060] The scheme in the negotiation process is iteratively optimized. When the scheme converges or reaches the maximum number of iterations, the negotiation is terminated. Each intelligent agent signs the execution contract, determines the control parameter configuration, and obtains the global optimized control scheme.

[0061] Specifically, the entire gold flotation process is divided into control units with distinct functions. Based on the characteristics of the gold flotation process, the entire process is divided into four main functional units: the crushing control unit, the grinding and grading control unit, the flotation control unit, and the concentrate processing control unit. The crushing control unit is responsible for the initial processing of the raw ore, including the control of equipment such as jaw crushers and cone crushers. The grinding and grading control unit is responsible for further grinding and grading the crushed ore, including the control of equipment such as ball mills and cyclones. The flotation control unit separates valuable minerals from gangue, including the control of equipment such as flotation machines and reagent addition systems. The concentrate processing control unit is responsible for post-processing such as dehydration and drying of the flotation concentrate, including the control of equipment such as thickeners and filters. Each functional unit is equipped with an independent decision controller and actuator. The decision controller is responsible for receiving sensor data, performing decision calculations, and outputting control commands. The actuator is responsible for converting control commands into device actions, such as adjusting valve opening and changing motor speed.

[0062] Each functional unit is loaded with an expert knowledge base, a control algorithm base, and a fuzzy rule base to construct a corresponding intelligent agent. The expert knowledge base is a structured representation of empirical knowledge, containing normal parameter ranges, exception handling strategies, and optimization experience. For example, rules such as "When the flotation cell foam is dark, increase the collector dosage." The control algorithm base includes PID control algorithms, fuzzy control algorithms, and model predictive control algorithms. PID control algorithms are suitable for highly linear applications, fuzzy control algorithms are suitable for applications with strong nonlinearity but clear expert experience, and model predictive control algorithms are suitable for applications with multiple inputs and outputs and a clear mathematical model. The fuzzy rule base contains decision-making rules based on fuzzy logic, such as "If the pH value is low and the degree of lowness is large, significantly increase the amount of alkali agent added." After loading these three libraries, an intelligent agent with perception, decision-making, and execution capabilities is formed. Each intelligent agent is internally divided into two layers: a local optimization control layer and a collaborative optimization control layer. The local optimization control layer handles the internal parameter stability control and local target optimization of the unit, and adopts a closed-loop feedback control method; the collaborative optimization control layer is responsible for interacting with other unit intelligent agents, coordinating the consistency of the unit with the global goal, and adopting a predictive control and negotiation mechanism.

[0063] Initial control objectives are assigned to each agent based on the digital twin model. The digital twin model incorporates a mathematical model of the entire gold flotation process and historical operating data, enabling a quantitative description of the relationships between various parameters. During the initial control objective assignment process, the concentrate processing unit agent generates a task request based on production indicator requirements. This request, typically derived from the production plan or operator settings, includes target values for concentrate grade and recovery rate. Concentrate grade represents the percentage of valuable metals in the concentrate, while recovery represents the percentage of valuable metals recovered from the original ore. These two metrics typically exhibit a trade-off relationship. The concentrate processing unit agent packages these target values into a task request message, which is then transmitted to the upstream unit. The task request is passed through the hierarchy, first to the flotation unit agent. Upon receiving the request, the flotation unit agent evaluates whether the current operating conditions meet the target requirements. This evaluation involves calculating the expected concentrate grade and recovery rate under the current operating conditions, comparing them with the target values, and analyzing the extent and causes of the discrepancies. If there are any gaps, the flotation unit agent calculates the parameter adjustments necessary to achieve the target based on the digital twin model. This generates a flotation plan that includes parameters such as flotation reagent dosage, pH value, and aeration volume. It also calculates the resources required to implement this plan, such as reagent consumption and energy consumption, to create a resource requirement list. The flotation plan and resource requirements are then communicated as a whole to the grinding and classification unit agent.

[0064] According to the interaction rules set by the Contract Network protocol, each agent performs task decomposition and negotiation. The Contract Network protocol is a protocol for task allocation in multi-agent systems and consists of four main phases: task notification, bidding, awarding, and results reporting. In gold mine flotation control, the protocol specifies message types and interaction processes. For example, the concentrate processing unit sends a task request message, and the flotation unit replies with a solution response message. Agents negotiate through this message-passing mechanism. When conflicting objectives are encountered, they use the Pareto multi-objective optimization method to balance them. Pareto multi-objective optimization is a method for handling multiple conflicting objectives. A solution is considered Pareto optimal if no other solution can increase the value of another objective function without decreasing the value of at least one objective function. In the flotation process, multiple objectives, such as concentrate grade and recovery rate, production efficiency and energy consumption, often require trade-offs. Each agent submits its own optimization solution in turn and evaluates the solutions of other agents based on criteria such as feasibility, reasonable resource requirements, and goal satisfaction. The solutions generated during the negotiation process are iteratively optimized, with adjustments and evaluations made at each iteration until a termination condition is reached. Termination conditions include solution convergence and reaching the maximum number of iterations. Solution convergence refers to the finding of a stable solution when the change in the solution over several consecutive iterations is less than a set threshold. Reaching the maximum number of iterations is necessary to avoid infinite loops and is typically set at 20-30. After negotiation terminates, each agent "signs" an execution contract, confirming acceptance of the final solution and clarifying their respective execution responsibilities and objectives. The execution contract includes specific control parameter configurations, such as the crusher feed rate setpoint, ball mill speed setpoint, and flotation reagent dosage setpoint, to form a globally optimized control solution.

[0065] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0066] Independent data processing modules, decision-making and reasoning modules, and execution control modules are constructed for the crushing control unit, grinding and classification control unit, flotation control unit, and concentrate processing control unit, forming their own intelligent agent infrastructure.

[0067] Compile the process parameter control range, exception handling rules and optimization strategies for each functional unit into structured knowledge items, store them in the expert knowledge base of the corresponding intelligent agent, and establish a knowledge retrieval and matching mechanism;

[0068] Build a control algorithm library within each intelligent agent, including PID control algorithm, fuzzy control algorithm and model predictive control algorithm, and set algorithm selection logic to automatically switch the optimal control algorithm according to changes in working conditions;

[0069] Each intelligent agent is loaded with a fuzzy rule base, and the process parameter deviation value and deviation change rate are taken as input, matched with the rule base after fuzzy processing, and the fuzzy conclusion of the control quantity adjustment value is obtained through fuzzy reasoning;

[0070] A local optimization control layer is constructed within each intelligent agent. The local optimization control layer receives the sensor data of the unit, executes the local control loop, maintains the stability of the parameters within the unit, and optimizes the local objective function.

[0071] A collaborative optimization control layer is constructed for each intelligent agent. The collaborative optimization control layer receives messages from other intelligent agents, processes global target constraints, coordinates the working relationship between the unit and adjacent units, and participates in the negotiation process among multiple intelligent agents.

[0072] Specifically, independent data processing modules, decision-making and reasoning modules, and execution control modules are constructed for the crushing control unit, grinding and classification control unit, flotation control unit, and concentrate processing control unit, respectively. The data processing module receives raw sensor data and performs preprocessing, including outlier detection, data filtering, and data normalization. The decision-making and reasoning module performs state assessment and control calculations based on the processed data, outputting control decisions. The execution control module converts control decisions into specific execution commands and sends them to the corresponding execution devices. For example, in the flotation control unit, the data processing module receives sensor data such as pH value, redox potential, flotation reagent flow rate, and slurry concentration, converting the data format, unit standardization, and time alignment. The decision-making and reasoning module uses this data to calculate the current flotation operating conditions and determine whether parameters such as reagent addition amount and stirring speed need to be adjusted. The execution control module converts the calculated reagent addition amount into a flow control signal for the reagent pump and a frequency control signal for the stirring speed converter. This modular design enables each functional unit to operate independently and work collaboratively as a whole, forming an intelligent agent infrastructure. Compiling the process parameter control ranges, exception handling rules, and optimization strategies for each functional unit into structured knowledge items requires knowledge representation and organization. Structured knowledge items adopt a "condition-action" format, with the condition describing the operating condition characteristics and the action describing the corresponding control measures. For example, for a flotation unit, a knowledge item might be "Condition: pH value below 7.5 and reduced foam volume; Action: Increase lime addition until pH returns to within the range of 8.0 ± 0.2." Each knowledge item is assigned a unique identifier and labeled based on attributes such as scope of application and priority. The compiled knowledge items are stored in the expert knowledge base of the corresponding agent, and a knowledge retrieval and matching mechanism is established. Knowledge retrieval uses inverted indexing technology to quickly locate relevant knowledge items based on keywords. Knowledge matching uses similarity calculation to match the current operating condition with the conditions described in the knowledge item to identify the most similar item. This similarity calculation considers the proximity of parameter values, parameter importance, and the matching degree of operating condition characteristics. When certain operating parameters change or anomalies occur, the knowledge retrieval and matching mechanism is automatically triggered to identify the appropriate handling strategy for the current situation, providing a basis for decision-making reasoning.

[0073] Building a control algorithm library within each intelligent agent requires implementing multiple control algorithms and establishing a selection mechanism. The control algorithm library includes three main algorithms: PID control, fuzzy control, and model predictive control. The PID control algorithm calculates the control output based on the error and its rate of change and is suitable for control objects with high linearity and simple models. The fuzzy control algorithm uses fuzzy rule reasoning to make control decisions and is suitable for control objects with strong nonlinearity, where precise mathematical models are difficult to establish but where extensive empirical knowledge is available. The model predictive control algorithm predicts future outputs based on a system model and solves an optimization problem to obtain a control sequence. It is suitable for complex control objects with multiple variables, strong coupling, and constraints. Algorithm selection logic is set up to automatically switch to the optimal control algorithm based on changing operating conditions. This algorithm selection logic is based on the characteristics of the current control object, the control objective requirements, and the historical performance of each algorithm. This is achieved through an evaluation function that calculates the suitability of each algorithm under the current operating conditions and selects the most suitable algorithm. This evaluation function considers factors such as control accuracy, response speed, and robustness, and dynamically adjusts the weights of each factor based on historical data.

[0074] To load a fuzzy rule base onto each agent, a fuzzy control process must be designed. First, the process parameter deviation (E) and the deviation change rate (EC) are used as inputs to the fuzzy controller. These two values are calculated by calculating the difference between the current parameter value and the setpoint and its rate of change over time. E and EC are then fuzzified, mapping the precise numerical values to fuzzy sets, such as "negative large," "negative small," "zero," "positive small," and "positive large" fuzzy linguistic variables. Fuzzification utilizes membership functions, with triangular, trapezoidal, and Gaussian functions being common. The appropriate function form is selected based on the parameter characteristics. The fuzzified inputs are then matched against rules in the fuzzy rule base. These rules employ an "if-then" formula, such as "If E is negative large and EC is positive small, then U is negative medium," where U is the controlled variable adjustment value. Fuzzy reasoning is then used to calculate the fuzzy set of the controlled variable. Fuzzy reasoning methods include Mamdani and Sugeno, depending on specific requirements. Finally, defuzzification is used to convert fuzzy sets back into precise values. Common defuzzification methods include the centroid method and the maximum membership method. This allows the fuzzy controller to determine appropriate control adjustments based on parameter deviations and trends, achieving intelligent control. A local optimization control layer is constructed within each intelligent agent, enabling it to independently handle the control tasks of its own unit. The local optimization control layer receives sensor data and first undergoes preprocessing, including filtering, outlier detection, and compensation. The data then enters the state estimation phase, where it estimates state variables that are difficult to measure directly using a Kalman filter or observer. The state estimation result is compared with the setpoint to calculate the control deviation. Based on this deviation, an appropriate control algorithm is selected to calculate the control output. The control output is also subject to constraints to ensure it remains within the physical limitations of the actuator. The local optimization control layer also optimizes the local objective function, which typically includes metrics such as control accuracy, stability, and resource consumption. Optimization algorithms such as gradient descent and simulated annealing are used to determine the optimal control parameters. Through a closed-loop feedback control mechanism, the local optimization control layer continuously monitors control effectiveness and adjusts the control strategy to maintain stable operation of the parameters within the unit.

[0075] A collaborative optimization control layer is constructed for each agent to achieve coordination and cooperation between units. The collaborative optimization control layer is responsible for interacting with other agents and handling global issues involving multiple units. The collaborative optimization control layer receives messages from other agents, including status information, demand requests, and negotiation proposals, and parses and processes them. For global objective constraints, the collaborative optimization control layer decomposes them into local constraints executable by the unit and checks their compatibility with the unit's objectives. If there are conflicts, the local objective function needs to be adjusted to align with the global objective. The collaborative optimization control layer is also responsible for coordinating the working relationship between the unit and adjacent units, such as ensuring material flow matching and process parameter integration. When participating in multi-agent negotiations, the collaborative optimization control layer generates response plans based on its own capabilities and constraints, evaluates the plans of other units, and participates in the iterative negotiation process until consensus is reached. The decision results of the collaborative optimization control layer are passed to the local optimization control layer for execution, forming a hierarchical control structure.

[0076] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0077] The flotation expert experience knowledge is encoded in the form of condition-action. The condition part includes the parameter value range and state description, and the action part includes the control operation and adjustment strategy, forming a structured knowledge entry.

[0078] Assign unique identifiers to the structured knowledge items of each functional unit, perform multi-dimensional annotation according to the functional unit, knowledge type and applicable conditions, and construct a knowledge index table;

[0079] A knowledge retrieval system is built based on inverted index technology, using process parameter names, value ranges, and exception types as retrieval keywords, corresponding to a set of identifiers for related knowledge items.

[0080] A knowledge matching algorithm based on weighted cosine similarity is designed to convert the current working condition parameters into feature vectors, calculate the similarity with the conditional feature vectors in the knowledge items, and obtain the matching ranking results;

[0081] Build a knowledge entry activation mechanism. When a certain operating condition parameter deviates from the normal range or an abnormal state occurs, the retrieval process of related knowledge entries is automatically triggered, and knowledge entries with a similarity exceeding the threshold of 0.85 are marked as activated.

[0082] Conflict resolution rules are applied to multiple activated knowledge items. When different knowledge items give conflicting suggestions, the optimal knowledge item is selected as the decision basis based on the priority, applicability accuracy and historical success rate of the knowledge items.

[0083] Specifically, flotation expert knowledge represents years of accumulated practical experience by flotation engineers and operators, including techniques for adjusting process parameters, methods for handling abnormal conditions, and process optimization strategies. Converting this knowledge into structured knowledge items that can be processed by computers requires encoding using a condition-action format. The condition describes the operating conditions that trigger the application of the knowledge, including parameter value ranges and state descriptions, such as "pH between 8.2 and 8.5 and pulp concentration below 35%," "foam color is gray and foam layer thickness is less than 5 cm," etc. The action describes the control operations and adjustment strategies to address these conditions, such as "increase collector addition by 15% while maintaining pH," "reduce aeration to 0.6 m³ / min and increase agitation intensity to 280 rpm," etc. This encoding method intuitively reflects the expert's problem-solving approach, transforming implicit knowledge into explicit expressions and forming structured knowledge items.

[0084] Each structured knowledge item in each functional unit is assigned a unique identifier and annotated multi-dimensionally. The unique identifier uses a coding scheme such as "FU-KT-SN," where FU represents the functional unit code (e.g., CR for crushing unit, GR for grinding and classification unit, FL for flotation unit, CC for concentrate processing unit), KT represents the knowledge type code (e.g., PR for parameter range knowledge, EH for exception handling knowledge, and OS for optimization strategy knowledge), and SN represents the sequence number. For example, "FL-EH-023" represents the 23rd exception handling knowledge item for the flotation unit. Multi-dimensional annotation describes the attributes of the knowledge item, including functional unit attributes, knowledge type attributes, and applicability condition attributes. Functional unit attributes indicate the process unit to which the knowledge item applies; knowledge type attributes are categorized into parameter range knowledge, exception handling knowledge, and optimization strategy knowledge; and applicability condition attributes describe the specific operating conditions to which the knowledge item applies, such as the ore grade range and equipment type. This multi-dimensional annotation constructs a structured knowledge index table, facilitating subsequent knowledge retrieval. A knowledge retrieval system is constructed based on inverted indexing technology. The process parameter names, value ranges, and anomaly types are used as search keywords, which are mapped to a set of identifiers for related knowledge items. Inverted indexing is a key technology in information retrieval, mapping words in a document to a list of documents containing the word. In the expert knowledge base, the inverted index maps keywords to a set of knowledge item identifiers containing the keyword. In its implementation, keywords are first extracted from the knowledge items. These keywords include process parameter names (such as "pH value" and "collector"), value ranges (such as "less than 7.0" and "greater than 20 g / t"), and anomaly types (such as "foam instability" and "concentrate grade decline"). A mapping table is then created from keywords to knowledge item identifiers. A single keyword may correspond to multiple knowledge items. Finally, a term index is constructed, recording the knowledge items in which each keyword appears, its position within the conditional part, and its weight. When searching for relevant knowledge, a keyword is entered, and the inverted index quickly finds the set of knowledge item identifiers containing the keyword. Results are then sorted and returned based on relevance. A knowledge matching algorithm based on weighted cosine similarity is designed to calculate the similarity between the current operating conditions and the knowledge items. Weighted cosine similarity is an extension of cosine similarity, taking into account the differences in weights of different features. First, the current operating condition parameters are converted into feature vectors. The dimension of the feature vector is consistent with the number of conditional features defined in the knowledge entry, and each dimension represents the value of a parameter or state feature. Similarly, the conditional part in the knowledge entry is converted into a conditional feature vector. Then, the weight vector is set according to the importance of the parameters, and more critical parameters are given higher weights. For example, key parameters such as pH value and collector dosage have higher weights, while auxiliary parameters such as temperature and pressure have lower weights. The weighted cosine similarity of the two vectors is calculated to obtain the matching value. The closer the matching value is to 1, the more similar the current operating condition is to the conditions described in the knowledge entry.To improve computational efficiency, vector space compression technology is used to reduce the original high-dimensional feature vectors to lower dimensions through principal component analysis, preserving key feature information while reducing computational complexity. For continuous-valued parameters, fuzzy interval mapping is used to handle numerical differences; for discrete state features, equality judgment or similarity measurement is used. Ultimately, a ranking of the matching degree between the current working condition and each knowledge item is obtained.

[0085] A knowledge item activation mechanism is constructed to proactively trigger the retrieval of relevant knowledge items when operating conditions change. This knowledge item activation mechanism is based on two triggering methods: parameter excursion triggering and state change triggering. Parameter excursion triggering occurs when a process parameter deviates from its normal range, triggering the retrieval of relevant knowledge items. For example, a drop in pH from the normal range of 7.5-8.5 to 7.2 triggers the retrieval of knowledge items related to low pH values. State change triggering occurs when a significant change in process conditions occurs, such as a change from stable to unstable foam quality. The retrieval process employs a two-stage strategy. In the first stage, an inverted index is used to quickly filter a set of potentially relevant knowledge items. In the second stage, each knowledge item in the candidate set is calculated for similarity with the current operating conditions. Entries with a similarity exceeding a threshold of 0.85 are marked as active. The threshold of 0.85 is an empirical value that ensures sufficient relevant knowledge items are found while filtering out less relevant ones. Active knowledge items enter the next stage of conflict resolution, providing a basis for the final decision. Conflict resolution rules are applied to multiple active knowledge items to resolve conflicts between them. Under complex operating conditions, it's common for multiple knowledge items to be activated simultaneously, offering different or even conflicting recommendations. For example, one knowledge item recommends increasing the collector dosage, while another recommends decreasing it. Conflict resolution rules are based on three key factors: the knowledge item's priority, scope accuracy, and historical success rate. Priority is an intrinsic property of a knowledge item, assigned by the knowledge engineer during coding, reflecting its importance and reliability. Scope accuracy indicates how closely the conditions described by the knowledge item match the current operating conditions; a more precise match is assigned a higher weight. Historical success rate is the percentage of successful applications of the knowledge item in past applications; a higher success rate carries a higher weight. Combining these three factors, a comprehensive score is calculated for each active knowledge item, and the highest-scoring knowledge item is selected as the basis for decision-making. If multiple knowledge items have similar scores and compatible recommendations, a compromise strategy is adopted, such as taking a weighted average of their recommendations. If the recommendations are incompatible, the highest-scoring knowledge item is retained.

[0086] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0087] A multi-model integrated soft-sensing system was constructed based on a global optimization control scheme to estimate five key quality indicators: concentrate grade, recovery rate, mineral surface hydrophobicity, and particle-bubble adhesion probability. The soft-sensing model includes a flotation kinetics theory model and a historical data regression model.

[0088] The quality indicators estimated by the soft measurement system are combined with the actual process parameters to build a hierarchical control structure, including a real-time control layer, a tactical optimization layer, and a strategic decision-making layer. The control cycles of each layer are set to 100ms, 2 hours, and 8 hours respectively.

[0089] An improved model predictive control algorithm is used for the real-time control layer, with the prediction time domain set to 30 steps and the control time domain set to 5 steps. The prediction model is updated based on the correspondence between process parameters and quality indicators.

[0090] A multi-objective optimization algorithm is applied to the tactical optimization layer, setting four optimization objectives: maximizing concentrate grade, maximizing recovery rate, minimizing energy consumption, and minimizing reagent consumption. The objective function weights are dynamically adjusted according to current production needs.

[0091] Conduct real-time evaluation of control effects, and establish an operation case library including working condition characteristics, control parameter configuration, and control effects by comparing the deviation between actual production indicators and optimization targets;

[0092] Based on the operation case library and soft measurement feedback, the model parameters are incrementally updated through the transfer learning method, and the control strategy is adjusted according to the control effect evaluation results to form a self-learning closed loop and obtain the process control parameters.

[0093] Specifically, soft sensing refers to a method that indirectly estimates key quality indicators that are difficult to measure directly through measurable process parameters. In gold flotation, five key quality indicators—concentrate grade, recovery rate, mineral surface hydrophobicity, particle-bubble adhesion probability, and concentrate quality—are difficult to measure online in real time, yet are crucial for control decisions. Soft sensing systems employ a multi-model integration strategy, encompassing two main categories: flotation kinetics theoretical models and historical data regression models. Flotation kinetics theoretical models, based on the physical and chemical principles of flotation, establish mathematical equations describing the interaction between particles and bubbles, deriving quality indicator estimates by solving differential equations. Historical data regression models, based on historical operating data, establish a mapping between process parameters and quality indicators. These models include multivariate linear regression models, support vector regression models, and neural network models. Multi-model integration improves the accuracy and robustness of estimates by weighting the outputs of each model. Weights are assigned based on the performance of each model on a validation dataset, with higher weights assigned to better performance.

[0094] By combining quality indicators estimated by a soft-sensor system with actual process parameters, a hierarchical control structure consisting of three control levels is constructed. The real-time control layer is responsible for stable control of process parameters within each unit, such as closed-loop control of pH, reagent dosage, and agitation speed. The control cycle is set to 100ms to ensure rapid response to parameter perturbations. The tactical optimization layer is responsible for medium-timescale process optimization, such as dynamic adjustment of grinding fineness, flotation time, and reagent formulation. The control cycle is set to 2 hours, with multiple optimizations performed within a single shift. The strategic decision-making layer is responsible for long-timescale production planning, such as throughput allocation and energy consumption planning. The control cycle is set to 8 hours, and adjustments are typically made once per shift. This hierarchical control structure reduces system complexity by separating time scales, allowing each level to focus on control tasks appropriate to its time scale. A modified model predictive control algorithm is used for the real-time control layer, which incorporates several improvements over standard model predictive control. First, the prediction horizon is set to 30 steps, predicting the system behavior for 30 sampling periods into the future. The control horizon is set to 5 steps, calculating the control input sequence for the next five sampling periods. The prediction model is the core of the algorithm. It describes the correspondence between process parameters and quality indicators and is represented by a state-space model. The prediction model is not fixed but is dynamically updated based on real-time data. Update methods include recursive least squares and moving window batch processing. The recursive least squares method uses newly acquired data to make small adjustments to the model parameters to maintain model accuracy; the moving window batch processing method regularly uses data from the most recent period to re-estimate the model parameters to adapt to changes in operating conditions. The improved model predictive control algorithm also introduces a constraint processing mechanism to ensure that the control output is within the physical limitations of the actuator, as well as a soft constraint mechanism that allows for conditional violation of some constraints when necessary.

[0095] A multi-objective optimization algorithm is applied to the tactical optimization layer, with four optimization objectives set: maximizing concentrate grade, maximizing recovery rate, minimizing energy consumption, and minimizing reagent consumption. These objectives are often conflicting. For example, increasing concentrate grade often reduces recovery rate, while reducing energy consumption can affect processing capacity. The multi-objective optimization algorithm addresses these conflicts by dynamically adjusting the objective function weights. The objective function weights change based on current production needs. For example, the weight of the grade target is increased when market prices are high, the weight of the recovery target is increased when the ore grade is low, and the weight of the energy consumption target is increased during peak electricity prices. The weight adjustment process takes into account market factors, resource constraints, and equipment status, forming an adaptive weight allocation mechanism. The optimization algorithm uses an improved genetic algorithm with a real number coding scheme, a population size of 100, a crossover probability of 0.85, and a mutation probability of 0.15. The optimal control parameter combination is solved through multi-generation evolution.

[0096] Control effectiveness is evaluated in real time, and a case study library is established. The evaluation process compares the deviation between actual production indicators and optimization targets to calculate control performance indicators, including concentrate grade deviation, recovery rate deviation, energy consumption exceedance rate, and reagent usage efficiency. When the deviation exceeds a set threshold, a control strategy adjustment is triggered. Simultaneously, the current operating conditions, control parameter configuration, and control effectiveness evaluation results are recorded in the case study library. The case study library's data structure includes operating condition characteristics (such as ore properties and equipment status), control parameters (set values for each control parameter), and effectiveness evaluation (key indicators such as concentrate grade, recovery rate, and energy consumption). The case study library records both successful and failed cases, providing a comprehensive reference for subsequent optimization. A case similarity calculation method is established so that when encountering new operating conditions, similar cases can be found in the case study library and their control experience can be leveraged. Based on the case study library and soft sensor feedback, model parameters are incrementally updated through transfer learning, and the control strategy is adjusted based on the control effectiveness evaluation results. Transfer learning refers to fine-tuning an existing model on new data rather than retraining it. This method retains the knowledge in the model that is relevant to the new data and only updates the parts with larger deviations. The incremental update process first calculates the prediction error of the current model, then adjusts the update step size based on the error size and pattern, and finally updates the model parameters. Parameter updates use gradient descent or recursive least squares to ensure that the accuracy of the model continues to improve over time. Control strategy adjustments are based on the results of the control effect evaluation, such as increasing the robustness parameters of the control algorithm when the grade fluctuates greatly, and adjusting the optimization target weights when the recovery rate is insufficient. The entire update and adjustment process forms a self-learning closed loop, and the system can continuously accumulate experience from practice to improve the control effect.

[0097] The above describes the method for intelligent monitoring and optimization control of the entire gold flotation process in the embodiment of the present application. The following describes the intelligent monitoring and optimization control system for the entire gold flotation process in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the intelligent monitoring and optimization control system for the entire gold flotation process includes:

[0098] The classification module 201 is used to collect process parameter data in the entire gold flotation process through a distributed sensor network, perform outlier detection and classification processing on the collected data, and obtain classified processing data;

[0099] The fusion module 202 is used to build a multi-source heterogeneous data fusion engine based on the hierarchical processing data, standardize and fuse the data at multiple levels, and obtain a digital twin model of the entire gold mine flotation process;

[0100] An adjustment module 203 is used to design a multi-agent negotiation control architecture based on the digital twin model, and to adjust control parameters using a negotiation mechanism based on the contract network protocol to obtain a global optimal control solution;

[0101] The construction module 204 is used to construct an adaptive soft measurement model and a multi-objective optimization control strategy according to the global optimization control scheme, realize parameter optimization through self-learning and knowledge accumulation mechanism, and obtain process control parameters.

[0102] Through the collaborative efforts of these components, process parameter data is collected through a distributed sensor network, and outlier detection and hierarchical processing are performed on the data, effectively improving data quality and addressing the simple and low-quality data processing methods found in traditional systems. A multi-source, heterogeneous data fusion engine is constructed based on the hierarchically processed data, standardizing and fusing the data at multiple levels to generate a digital twin model of the entire gold flotation process. This achieves real-time mapping between the physical and virtual flotation systems, providing an accurate information basis for subsequent decision-making. A multi-agent negotiation control architecture is designed based on the digital twin model, and a negotiation mechanism based on the contract network protocol is used to adjust control parameters. This effectively addresses the issues of insufficient information exchange between control units and the difficulty in achieving global optimization in traditional systems. The negotiation mechanism significantly enhances the overall system's collaborative performance, enabling each functional unit to achieve the global optimal goal while maintaining local stability. Based on the global optimization control scheme, an adaptive soft-sensing model and a multi-objective optimization control strategy are constructed, achieving parameter optimization through self-learning and knowledge accumulation mechanisms. This addresses both the lack of effective soft-sensing methods in traditional systems and the lack of self-improvement mechanisms in control strategies. In particular, the contract network protocol used in the negotiation control architecture, as a multi-agent negotiation mechanism, fully considers the characteristics of each functional unit in the flotation process and their mutual influence. Through structured message interaction and task decomposition negotiation process, it effectively balances multiple conflicting goals such as grade, recovery rate, and energy consumption. The introduction of the negotiation mechanism enables the system to have distributed decision-making capabilities, significantly reduces the computational burden, and improves the system response speed; and the adaptive soft measurement model realizes accurate estimation of parameters that are difficult to measure directly through multi-model integration and transfer learning methods. The self-learning mechanism enables the system to accumulate optimization experience from historical operation data, improve its adaptability to complex working conditions, and reduce energy consumption and reagent consumption while improving the grade and recovery rate of concentrate.

[0103] above Figure 2 The intelligent monitoring and optimization control system for the entire flotation process of gold mines in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The intelligent monitoring and optimization control device for the entire flotation process of gold mines in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0104] Figure 3This is a schematic diagram of the structure of a full-process intelligent monitoring and optimization control device for gold flotation, provided by an embodiment of the present invention. The full-process intelligent monitoring and optimization control device 300 may vary significantly depending on configuration or performance. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors), a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage medium 330 may be either transient or persistent storage. The program stored in the storage medium 330 may include one or more modules (not shown), each of which may include a series of instructions for the full-process intelligent monitoring and optimization control device 300. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, executing the series of instructions stored in the storage medium 330 on the full-process intelligent monitoring and optimization control device 300 to implement the steps of the above-described method for full-process intelligent monitoring and optimization control of gold flotation.

[0105] The gold mine flotation full process intelligent monitoring and optimization control device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. It will be understood by those skilled in the art that Figure 3 The structure of the gold mine flotation full-process intelligent monitoring and optimization control equipment shown does not constitute a limitation on the gold mine flotation full-process intelligent monitoring and optimization control equipment provided by the present invention, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0106] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions. When the instructions are executed on a computer, the computer executes the steps of the method for intelligent monitoring and optimization control of the entire gold mine flotation process.

[0107] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0108] 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 the present invention, or the portion 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 for enabling a gold mine flotation full-process intelligent monitoring and optimization control device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0109] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for intelligent monitoring and optimization control of the entire gold flotation process, characterized in that: This includes collecting process parameter data from the entire gold flotation process through a distributed sensor network, performing outlier detection and classification processing on the collected data, and obtaining classified processing data; A multi-source heterogeneous data fusion engine is built based on hierarchical data processing, which is then standardized and integrated at multiple levels to create a digital twin model of the entire gold mine flotation process. A multi-agent negotiation control architecture is designed based on the digital twin model. A negotiation mechanism based on the contract network protocol is used to adjust control parameters to obtain a global optimization control solution. This involves dividing the entire gold flotation process into four main functional units: a crushing control unit, a grinding and classification control unit, a flotation control unit, and a concentrate processing control unit. Each unit is equipped with an independent decision controller and actuator. An expert knowledge base, a control algorithm library, and a fuzzy rule library are loaded into each functional unit, and an agent corresponding to each functional unit is constructed. Each agent has a local optimization control layer and a collaborative optimization control layer. Based on the digital twin model, each agent is assigned an initial control objective. The concentrate processing unit agent generates a task request based on production indicator requirements, including a target value for concentrate grade and recovery rate. The digital twin model includes a mathematical model of the entire gold flotation process and historical operating data. Task requests are transmitted hierarchically. After receiving the request, the flotation unit agent evaluates the current operating conditions and generates a flotation plan and resource requirements, including flotation agent dosage, pH value, and aeration volume. According to the interaction rules set by the contract network protocol, each agent decomposes and negotiates tasks through a message passing mechanism. The Pareto multi-objective optimization method is used to weigh conflicting objectives. Each agent submits a plan in turn and evaluates the plan. Iteratively optimize the solutions during the negotiation process and terminate the negotiation when the solution converges or the maximum number of iterations is reached. The termination conditions include convergence and the maximum number of iterations. Each intelligent agent signs the execution contract, determines the control parameter configuration, and obtains the globally optimized control solution. According to the global optimization control scheme, an adaptive soft measurement model and a multi-objective optimization control strategy are constructed. Parameter optimization is achieved through self-learning and knowledge accumulation mechanisms to obtain process control parameters.

2. The gold flotation whole process intelligent monitoring and optimization control method according to claim 1 is characterized in that: The process parameter data of the entire gold flotation process are collected through a distributed sensor network, and the collected data are subjected to outlier detection and classification processing to obtain classified processing data, including: classifying the collected process parameter data of pH value, pulp concentration, grinding fineness, and foam characteristics according to the change rate, setting the sampling frequency of fast-changing parameters to 500ms / time, and setting the sampling frequency of slowly changing parameters to 5s / time; applying the improved Z-score method to the classified process parameter data for outlier detection, and marking the data points exceeding the threshold of 4.5 as outliers by calculating the standardized distance between the parameter value and the sliding window mean; The outlier data points are replaced with the weighted average of the 20 valid data points before and after, and the weight coefficient is calculated by the exponential decay function; the fourth-order Kalman filter algorithm is applied to the data after replacing the outliers to eliminate noise, and the filter parameter matrix is adjusted according to the data type; the filtered data is graded based on the data importance scoring function, and the scoring function comprehensively considers three factors: the parameter's influence on the process, the amplitude of change, and the frequency of fluctuation; the parameter data with a score greater than 0.8 is classified as level one, the parameter data with a score between 0.5 and 0.8 is classified as level two, and the parameter data with a score below 0.5 is classified as level three, forming hierarchical processing data with priority.

3. The gold flotation whole process intelligent monitoring and optimization control method according to claim 1 is characterized in that: A multi-source heterogeneous data fusion engine is built based on the hierarchical processing data. The data is standardized and integrated at multiple levels to obtain a digital twin model of the entire gold flotation process. This includes: unit conversion of the hierarchical processing data, converting the grinding fineness data into millimeters, converting the slurry concentration data into decimal form, and converting the potential data into standard potential units to obtain data with unified units; normalization of the data with unified units, adjusting the numerical ranges of different parameters to the same range, eliminating the impact of dimensional differences on subsequent analysis, and obtaining normalized data; and time alignment of the normalized data, aligning data with different sampling frequencies according to the system clock. , interpolate and fill the sampling intervals to obtain standardized data that is consistent in time and space; perform weighted average calculation on the data of the same parameter measured by multiple sensors in the standardized data, assign weights according to the historical accuracy of each sensor, and obtain data-level fusion results; analyze the correlation between different parameters in the data-level fusion results, identify parameter combinations with strong correlation, extract key features that characterize the flotation process status, and obtain feature-level fusion results; combine the feature-level fusion results with the knowledge of flotation process experts to judge the current flotation process status, establish the correspondence between the physical flotation system and the virtual system, and obtain a digital twin model of the entire gold mine flotation process.

4. The gold flotation whole process intelligent monitoring and optimization control method according to claim 1 is characterized in that: Each functional unit is loaded with an expert knowledge base, a control algorithm library, and a fuzzy rule library, and an intelligent agent corresponding to each functional unit is constructed. Each intelligent agent has a local optimization control layer and a collaborative optimization control layer, including: Independent data processing modules, decision-making and reasoning modules, and execution control modules are constructed for the crushing control unit, grinding and classification control unit, flotation control unit, and concentrate processing control unit, forming their own intelligent agent infrastructure. The process parameter control range, exception handling rules, and optimization strategies for each functional unit are compiled into structured knowledge items and stored in the expert knowledge base of the corresponding intelligent agent, establishing a knowledge retrieval and matching mechanism. A control algorithm library is constructed within each intelligent agent, including the PID control algorithm, fuzzy control algorithm and model predictive control algorithm, and the algorithm selection logic is set to automatically switch the optimal control algorithm according to changes in working conditions; a fuzzy rule library is loaded for each intelligent agent, and the process parameter deviation value and the deviation change rate are taken as input, matched with the rule library after fuzzy processing, and a fuzzy conclusion of the control quantity adjustment value is obtained through fuzzy reasoning; a local optimization control layer is constructed within each intelligent agent, and the local optimization control layer receives the sensor data of this unit, executes the local control loop, maintains the stability of the parameters within this unit, and optimizes the local objective function; a collaborative optimization control layer is constructed for each intelligent agent, and the collaborative optimization control layer receives messages from other intelligent agents, processes global target constraints, coordinates the working relationship between this unit and adjacent units, and participates in the negotiation process between multiple intelligent agents.

5. The gold flotation whole process intelligent monitoring and optimization control method according to claim 4 is characterized in that: The process parameter control range, exception handling rules and optimization strategies of each functional unit are compiled into structured knowledge items and stored in the expert knowledge base of the corresponding intelligent agent. A knowledge retrieval and matching mechanism is established, including: encoding the flotation expert experience knowledge in a condition-action form, where the condition part includes the parameter value range and state description, and the action part includes the control operation and adjustment strategy, forming a structured knowledge item; assigning a unique identifier to the structured knowledge item of each functional unit, and performing multi-dimensional annotation according to the functional unit, knowledge type and applicable conditions to construct a knowledge index table; building a knowledge retrieval system based on inverted index technology, and using the process parameter name, value range, and exception type as retrieval keywords. , corresponding to the set of identifiers of related knowledge items; a knowledge matching algorithm based on weighted cosine similarity is designed to convert the current working condition parameters into feature vectors, calculate the similarity with the conditional feature vectors in the knowledge items, and obtain the matching ranking results; a knowledge item activation mechanism is constructed. When a certain working condition parameter deviates from the normal range or an abnormal state occurs, the retrieval process of related knowledge items is automatically triggered, and the knowledge items with a similarity exceeding the threshold of 0.85 are marked as activated; conflict resolution rules are applied to knowledge items in multiple activated states. When different knowledge items give conflicting suggestions, the optimal knowledge item is selected as the decision basis based on the priority, accuracy of applicable scope, and historical success rate of the knowledge item.

6. The gold flotation whole process intelligent monitoring and optimization control method according to claim 1 is characterized in that: According to the global optimization control scheme, an adaptive soft measurement model and a multi-objective optimization control strategy are constructed. Parameter optimization is achieved through self-learning and knowledge accumulation mechanisms to obtain process control parameters, including: constructing a multi-model integrated soft measurement system based on the global optimization control scheme to estimate five key quality indicators: concentrate grade, recovery rate, mineral surface hydrophobicity, and particle-bubble adhesion probability. The soft measurement model includes a flotation dynamics theory model and a historical data regression model; combining the quality indicators estimated by the soft measurement system with the actual process parameters to construct a hierarchical control structure, including a real-time control layer, a tactical optimization layer, and a strategic decision-making layer. The control cycles of each layer are set to 100ms, 2 hours, and 8 hours, respectively; and adopting an improved model predictive control for the real-time control layer. The algorithm sets the prediction time domain to 30 steps and the control time domain to 5 steps. The prediction model is updated according to the correspondence between process parameters and quality indicators. The multi-objective optimization algorithm is applied to the tactical optimization layer, and four optimization objectives are set: maximizing concentrate grade, maximizing recovery rate, minimizing energy consumption, and minimizing reagent consumption. The objective function weights are dynamically adjusted according to current production needs. The control effect is evaluated in real time. By comparing the deviation between actual production indicators and optimization objectives, an operation case library containing operating condition characteristics, control parameter configuration, and control effect is established. Based on the operation case library and soft measurement feedback, the model parameters are incrementally updated through the transfer learning method, and the control strategy is adjusted according to the control effect evaluation results to form a self-learning closed loop and obtain process control parameters.

7. A gold mine flotation full process intelligent monitoring and optimization control system, characterized in that: The method for realizing the full-process intelligent monitoring and optimization control of gold ore flotation according to any one of claims 1 to 6, the full-process intelligent monitoring and optimization control system of gold ore flotation comprises: The classification module is used to collect process parameter data of the entire gold flotation process through a distributed sensor network, perform outlier detection and classification processing on the collected data, and obtain classified processing data; The fusion module is used to build a multi-source heterogeneous data fusion engine based on hierarchical processing data, standardize and fuse the data at multiple levels, and obtain a digital twin model of the entire gold mine flotation process; The adjustment module is used to design a multi-agent negotiation control architecture based on the digital twin model, and uses a negotiation mechanism based on the contract network protocol to adjust control parameters and obtain a globally optimized control solution; The building module is used to construct an adaptive soft measurement model and a multi-objective optimization control strategy according to the global optimization control scheme, realize parameter optimization through self-learning and knowledge accumulation mechanism, and obtain process control parameters.

8. An intelligent monitoring and optimization control device for the entire flotation process of gold mines, characterized in that: The invention comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the method for intelligent monitoring and optimization control of the entire gold flotation process according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the processor executes the gold mine flotation full-process intelligent monitoring and optimization control method as described in any one of claims 1 to 6.

Images