An ore dressing equipment process parameter self-adaptive control method and system and a storage medium

Abstract

The application relates to the technical field of beneficiation equipment control, and discloses a beneficiation equipment process parameter self-adaptive control method, a beneficiation equipment process parameter self-adaptive control system and a storage medium. The method comprises the following steps: improving state estimation reliability through redundant measurement data fusion, adopting online parameter identification to make a model track process characteristic changes, utilizing rolling optimization to realize prospective prediction control, suppressing process fluctuations caused by multivariable coupling through hierarchical coordination execution, and continuously optimizing controller parameters based on performance index driven gradient adjustment. The application solves the problems of poor measurement data reliability, fixed model parameters that cannot adapt to changes in ore properties, control response lag and insufficient multivariable coordination control capability in the prior art.

Description

Technical Field

[0001] This application relates to the field of mineral processing equipment control technology, and in particular to an adaptive control method, system and storage medium for process parameters of mineral processing equipment. Background Technology

[0002] Controlling process parameters in mineral processing equipment is crucial for ensuring stable operation and improving technical and economic indicators. Traditionally, this control relies on operators' experience to observe indicators such as froth layer status and concentrate grade, manually adjusting parameters like aeration rate, stirring intensity, and reagent dosage. This manual control method suffers from significant lag and subjectivity. To improve control effectiveness, existing technologies have introduced automatic control systems. These systems use single-point measurement sensors to collect flotation process status parameters and employ PID feedback control algorithms to automatically adjust process parameters based on concentrate grade deviations. Some advanced systems utilize machine learning methods, such as LSTM-DNN neural networks, to establish predictive models between process parameters and technical and economic indicators. Cloud computing platforms are used to process massive amounts of historical data for model training, enabling parameter optimization and control based on prediction results. These existing technologies reduce reliance on manual operation and improve control accuracy to some extent.

[0003] However, existing technologies still have significant shortcomings: First, the state information obtained by single-point measurement is easily affected by local disturbances and measurement noise. When sensors drift or malfunction, there is a lack of effective data correction mechanisms, leading to the control system making incorrect decisions based on unreliable data. Second, fixed-parameter control models cannot adapt to dynamic changes in ore properties. When the hardness, oxidation rate, or mineral composition of the feed ore fluctuates, the parameters of the pre-trained neural network model remain unchanged, causing the model's prediction accuracy to drop sharply and the control effect to deteriorate significantly. Third, the hysteresis feedback control method based on result indicators such as concentrate grade has a slow response speed. It usually takes 15 to 20 minutes from the adjustment of process parameters to the feedback of grade detection results. This control delay causes the flotation process to be in a suboptimal state for a long time when the ore properties change abruptly. Summary of the Invention

[0004] This application provides an adaptive control method, system, and storage medium for process parameters of mineral processing equipment. By constructing a state estimation system based on redundant measurement data fusion, a predictive control model with online parameter identification, a hierarchical coordinated multivariate execution strategy, and a performance index-driven adaptive adjustment mechanism for controller parameters, it solves the problems of poor reliability of measurement data, fixed model parameters that cannot adapt to changes in ore properties, lag in control response, and insufficient multivariate coordinated control capability in the prior art, thereby improving the accuracy, real-time performance, and robustness of process parameter control in mineral processing equipment.

[0005] In a first aspect, this application provides an adaptive control method for process parameters of mineral processing equipment, the adaptive control method for process parameters of mineral processing equipment comprising:

[0006] Step S1: Obtain the first rheological parameter at the inlet of the flotation cell, the second rheological parameter and bubble characteristic parameter in the middle, and the third rheological parameter and electrochemical parameter at the outlet. Perform redundant fusion processing on the first rheological parameter, the second rheological parameter and the third rheological parameter using the weighted least squares method to obtain the fusion state estimate.

[0007] Step S2: Construct a state transition matrix and an input matrix based on the fused state estimate, and use a recursive least squares algorithm to identify the parameter vectors of the state transition matrix and the input matrix online to obtain the updated parameter vector;

[0008] Step S3: Substitute the updated parameter vector into the rolling optimization objective function, perform quadratic programming on the control increment sequence, and obtain the optimal control output;

[0009] Step S4: Decompose the optimal control output into an aeration volume adjustment sequence, a stirring intensity adjustment sequence, and a drug dosage adjustment sequence. Execute the aeration volume adjustment sequence, the stirring intensity adjustment sequence, and the drug dosage adjustment sequence in a hierarchical manner according to their time priority to obtain state response data.

[0010] Step S5: Calculate the tracking deviation index and control change index based on the state response data, and perform gradient adjustment on the output weight and control weight in the rolling optimization objective function to obtain the adjusted weight parameters.

[0011] Secondly, this application provides an adaptive control system for process parameters of mineral processing equipment, the adaptive control system for process parameters of mineral processing equipment comprising:

[0012] The fusion module is used to acquire the first rheological parameter at the inlet of the flotation cell, the second rheological parameter and bubble characteristic parameter in the middle, and the third rheological parameter and electrochemical parameter at the outlet. The first rheological parameter, the second rheological parameter and the third rheological parameter are redundantly fused using the weighted least squares method to obtain the fusion state estimate.

[0013] The identification module is used to construct a state transition matrix and an input matrix based on the fused state estimate, and to identify the parameter vectors of the state transition matrix and the input matrix online using a recursive least squares algorithm to obtain an updated parameter vector.

[0014] The solution module is used to substitute the updated parameter vector into the rolling optimization objective function, perform quadratic programming on the control increment sequence, and obtain the optimal control output.

[0015] The grading module is used to decompose the optimal control output into an aeration volume adjustment sequence, a stirring intensity adjustment sequence, and a drug dosage adjustment sequence, and to execute the aeration volume adjustment sequence, the stirring intensity adjustment sequence, and the drug dosage adjustment sequence in a graded manner according to the time priority to obtain state response data;

[0016] The adjustment module is used to calculate the tracking deviation index and control change index based on the state response data, and to perform gradient adjustment on the output weights and control weights in the rolling optimization objective function to obtain the adjusted weight parameters.

[0017] Thirdly, an adaptive control device for process parameters of a mineral processing equipment is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the adaptive control device for process parameters of the mineral processing equipment to execute the aforementioned adaptive control method for process parameters of the mineral processing equipment.

[0018] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored therein, which, when executed on a computer, cause the computer to perform the aforementioned adaptive control method for process parameters of mineral processing equipment.

[0019] The technical solution provided in this application acquires rheological parameters, bubble characteristic parameters, and electrochemical parameters at three key locations: the inlet, middle, and outlet of the flotation cell. A redundant measurement network is constructed, and a set of constraint equations is established based on the material conservation relationship. Weighted least squares is used to fuse data from multiple measurement points. When an abnormal measurement point is detected, its weight coefficient is automatically reduced, and a corrected fused value is obtained by resolving the problem. This significantly improves the reliability of process state estimation and avoids the defects of single-point measurement being susceptible to local disturbances and sensor failures. A state transition matrix and input matrix are constructed based on the fused state estimate. The parameter vector is identified online and the convergence state is determined using a recursive least squares algorithm, enabling the control model to track the dynamic characteristics of the flotation process in real time. This overcomes the limitation of fixed-parameter models being unable to adapt to fluctuations in ore properties. The updated parameter vector is substituted into the rolling optimization objective function for quadratic programming. An optimization objective is constructed based on the tracking error between the output predicted value sequence and the reference trajectory in the prediction time domain, as well as the cost term of the control increment in the control time domain. This optimization is applied to the aeration rate, ... The optimal control output is solved under the constraints of stirring intensity and drug dosage change rate, realizing forward predictive control. Compared with traditional hysteresis feedback control, the response time is significantly shortened. The optimal control output is decomposed into three adjustment sequences and executed hierarchically according to time priority. The aeration volume adopts a three-stage step method and monitors the rate of state change to achieve safe adjustment. The stirring intensity adopts a ramp tracking method and monitors the state coupling index to avoid over-coupling. The drug dosage is added when the stirring intensity is adjusted to the preset progress. This hierarchical coordinated execution strategy effectively suppresses process fluctuations caused by simultaneous adjustment of multiple variables through time decoupling and amplitude coordination. The tracking deviation index and control change index are calculated based on the state response data to construct a comprehensive performance index to evaluate the control effect. When the performance index is lower than the threshold, the gradient of the output weight and control weight with respect to the comprehensive performance index is calculated. The adjusted weight parameters are obtained by numerically updating according to the gradient direction. This performance index-driven parameter adaptive adjustment mechanism enables the controller to continuously optimize its own parameters according to the actual operating effect and maintain long-term stable control performance.

[0020] The innovation of this application in the field of adaptive control of process parameters in mineral processing equipment lies in the organic integration of multi-level adaptive mechanisms. The weighted least squares fusion algorithm achieves adaptation at the measurement level through redundant constraints and weight adaptation; the recursive least squares identification algorithm achieves adaptation at the model level through covariance updating and convergence determination; the quadratic programming optimization algorithm achieves forward-looking optimization at the control level through rolling time domain and constraint processing; and the gradient descent adjustment algorithm achieves adaptation at the controller level through performance feedback and parameter updating. These four levels of algorithms work together to form a complete adaptive control system. In particular, the gradient descent algorithm in this scheme is not used for model training but for online adjustment of controller parameters. This innovative application enables the control system to have self-improvement capabilities, automatically optimizing controller weight parameters based on comprehensive performance indicators during long-term operation, adapting to the slow drift and seasonal changes in process characteristics. The forgetting factor mechanism of the recursive least squares algorithm makes model identification focus more on recent data and quickly respond to sudden changes in ore properties, while the learning rate of the gradient descent algorithm... The mechanism ensures that controller parameter adjustments remain gradual, avoiding oscillations in control performance due to parameter mutations. The quadratic programming algorithm considers both tracking performance in the prediction time domain and stationarity in the control time domain when solving for the optimal control output, and ensures the physical feasibility of the control output through constraints. This multi-objective optimization characteristic allows the control strategy to pursue accurate tracking while taking into account the stability of control actions and equipment safety. The state change rate monitoring and state coupling index monitoring in the hierarchical execution strategy are essentially a safety protection mechanism based on real-time feedback. When an abnormal process response is detected during the adjustment process, subsequent adjustment actions are immediately suspended or attenuated. This adaptive protection at the execution layer further enhances the robustness of the control system. In summary, this application introduces targeted adaptive algorithms in each stage of data processing, model identification, optimized control, and parameter adjustment to construct a hierarchical, closed-loop adaptive control architecture, fundamentally solving complex control problems faced in the mineral processing process, such as ore property fluctuations, measurement noise interference, multivariable coupling, and long-term performance degradation. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of an embodiment of the adaptive control method for process parameters of mineral processing equipment in this application.

[0023] Figure 2 This is a schematic diagram of the parameter identification convergence process in the embodiments of this application;

[0024] Figure 3 This is a schematic diagram of an embodiment of the adaptive control system for process parameters of mineral processing equipment in this application.

[0025] Figure 4 This is a schematic block diagram of the adaptive control device for process parameters of mineral processing equipment in an embodiment of the present invention. Detailed Implementation

[0026] This application provides an adaptive control method, system, and storage medium for process parameters of a mineral processing equipment. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the adaptive control method for process parameters of mineral processing equipment in this application includes:

[0028] Step S1: Obtain the first rheological parameter at the inlet of the flotation cell, the second rheological parameter and bubble characteristic parameter in the middle, and the third rheological parameter and electrochemical parameter at the outlet. Perform redundant fusion processing on the first rheological parameter, the second rheological parameter and the third rheological parameter using the weighted least squares method to obtain the fusion state estimate.

[0029] Step S2: Construct the state transition matrix and input matrix based on the fused state estimates, and use the recursive least squares algorithm to identify the parameter vectors of the state transition matrix and input matrix online to obtain the updated parameter vector;

[0030] Step S3: Substitute the updated parameter vector into the rolling optimization objective function, perform quadratic programming on the control increment sequence, and obtain the optimal control output;

[0031] Step S4: Decompose the optimal control output into an aeration volume adjustment sequence, a stirring intensity adjustment sequence, and a drug dosage adjustment sequence. Execute the aeration volume adjustment sequence, stirring intensity adjustment sequence, and drug dosage adjustment sequence in a hierarchical manner according to their time priority to obtain state response data.

[0032] Step S5: Calculate the tracking deviation index and control change index based on the state response data, and perform gradient adjustment on the output weights and control weights in the rolling optimization objective function to obtain the adjusted weight parameters.

[0033] It is understood that the executing entity of this application can be an adaptive control system for process parameters of mineral processing equipment, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiment uses a server as an example for illustration.

[0034] Specifically, process status data is acquired by deploying measuring devices at three key locations in the flotation cell. At the inlet, the first rheological parameter is collected, reflecting the initial viscosity of the ore slurry. Simultaneously, the second rheological parameter and bubble characteristic parameters are collected in the middle to characterize the dynamic properties of the flotation zone. At the outlet, the third rheological parameter and electrochemical parameters are collected, reflecting the slurry properties during concentrate overflow. Due to noise interference and sensor drift in the measurements, redundant fusion processing is needed to improve data reliability. Specifically, a weighted least squares method is used. First, a constraint relationship between the three rheological parameters is established based on the continuity of materials within the flotation cell. Then, a coefficient matrix is ​​constructed to correlate the rheological parameters at the three measuring points. First, weight coefficients are set for each measuring point to reflect the measurement confidence level. The preliminary fusion value is obtained by minimizing the sum of squared weighted residuals. Then, the residual vector between the actual measured value and the fusion value is calculated. The residuals are standardized to eliminate the influence of dimensions. When the absolute value of the standardized residual of a measuring point exceeds three times the standard deviation, it is judged as an abnormal measuring point. The weight of the measuring point is reduced to one-fifth of the original value, and the weighted least squares solution is re-executed to obtain the corrected fusion value. Finally, the corrected rheological fusion value, bubble characteristic parameters, and electrochemical parameters are combined in columns to form a state vector as the fusion state estimate. This fusion value integrates information from multiple measuring points and eliminates abnormal interference.

[0035] Based on the mathematical description of the fusion state estimation process, the flotation process is abstracted into a state-space form. The state transition matrix describes how the current state evolves to the next state, and the input matrix describes the influence of the control input on the state changes. The elements of these two matrices constitute a parameter vector. Since the characteristics of the flotation process change with the properties of the ore, the parameter vector needs to be identified online. A recursive least squares algorithm is used to continuously update the parameter estimates using the collected input and output data. Specifically, at each sampling time, the control input sequence and the actual output sequence are collected, and the historical state values ​​and input values ​​are combined into a regression vector. Simultaneously, a covariance matrix is ​​maintained. To address the uncertainty in parameter estimation, the regression vector and covariance matrix are substituted into the recursive formula to calculate the identified parameter vector. This vector represents the current estimate of the true parameters. To determine whether the parameters have converged, the Euclidean distance between the current identified parameter vector and the parameter vector at the previous moment is calculated and divided by the parameter vector magnitude to obtain the parameter change rate. When the change rate is less than 2% for 20 consecutive sampling periods, the parameters are determined to have converged to a stable value. At this point, the identified parameter vector is output as the updated parameter vector for subsequent control. Simultaneously, the covariance matrix is ​​updated according to the recursive formula to prepare for identification at the next moment. The entire identification process continuously brings the process model parameters inside the controller closer to the characteristics of the real system.

[0036] The optimal control strategy is calculated using the updated parameter vector. First, the parameter vector is reconstructed by dimension into a state transition matrix and an input matrix. Based on these two matrices, a prediction mechanism is established to predict the output values ​​at each time step in the future, starting from the current time step. During prediction, a recursive calculation is performed. The predicted state value at each step is equal to the state transition matrix multiplied by the previous step's state, plus the input matrix multiplied by the corresponding control input. This maps the state to the output, resulting in a sequence of predicted output values. Then, this sequence is subtracted point by point from the desired reference trajectory to obtain the tracking error at each time step. The squares of the errors at all times are multiplied by the output weights and summed to form the tracking error term. The larger the value of this term, the worse the tracking performance. Simultaneously, the tracking error term is calculated at each time step in the control time domain. The control cost term is obtained by squaring the control increment, multiplying it by the control weight, and summing the results. This term reflects the intensity of the control action. The sum of the two terms constitutes the rolling optimization objective function. To ensure the physical feasibility of the control output, the following constraints are set: the rate of change of aeration volume does not exceed 0.15 cubic meters per minute, the rate of change of stirring intensity does not exceed 15 revolutions per minute, and the rate of change of drug dosage does not exceed 10 grams per ton. These constraints are transformed into inequalities. The optimal control increment sequence is obtained by quadratic programming on the constrained rolling optimization objective function. This sequence contains the control increments at each time point in the control time domain, but only the first element in the sequence is executed. This element is added to the control quantity at the previous time point to obtain the optimal control output at the current time point.

[0037] The optimal control output is separated into three control channels: the aeration volume control command is used to adjust the aeration system at the bottom of the flotation cell, the stirring intensity control command is used to adjust the speed of the stirring motor, and the dosage control command is used to adjust the flow rate of the dosing pump. Subtracting the corresponding control quantity from the previous moment from each command yields the actual adjustment range required, forming the aeration volume adjustment sequence, stirring intensity adjustment sequence, and dosage adjustment sequence, respectively. Due to the different response speeds and coupling effects of the three control channels, a hierarchical execution strategy is adopted. First, the fastest-responding aeration volume adjustment is executed. The aeration volume adjustment sequence is decomposed into three step increments at a ratio of 40%, 35%, and 25%. The first increment is executed at the current moment, the second increment is executed after a 1-second interval, and the third increment is executed after another 1-second interval. Immediately after each increment, the estimated fusion state value is collected. The difference between two adjacent state values ​​is calculated and divided by the time interval to obtain the state change rate. When the rate exceeds 0.08... The remaining step adjustments are paused at the threshold every second to avoid process fluctuations. After the aeration volume adjustment is completed, the stirring intensity adjustment is initiated. Instead of using a step adjustment, the adjustment amount is divided equally into 6 time steps, increasing the adjustment amount by one-sixth every second to achieve ramp tracking. During the adjustment process, the product of the state's sensitivity to the aeration volume and the aeration volume change amplitude is calculated in real time, and the product of the state's sensitivity to the stirring intensity and the stirring intensity change amplitude is added to obtain the state coupling index. When this index exceeds the upper limit of 0.15, the subsequent adjustment amplitude is reduced to 70% of the original to prevent over-coupling. When the stirring intensity adjustment reaches 50% progress, the dosage adjustment is initiated simultaneously. The required reagent volume is calculated based on the adjustment sequence and feed flow rate, and the dosing pump is controlled to complete the dosing within 180 seconds at a constant flow rate. The entire execution process records the control output value and the corresponding fusion state estimate at each moment to form time series data as state response data.

[0038] The control performance is evaluated and controller parameters are adjusted based on the state response data. The output variable time series, reflecting the changes in quality and recovery rates over time, is extracted from the state response data. The output series is time-aligned with a set reference trajectory. The difference between the output value and the reference value is calculated at each time point. The sum of the squared differences and divided by the number of time points yields the average tracking error, which is used as the tracking deviation index to quantify the setpoint tracking performance. Simultaneously, the execution trajectories of three control channels are extracted from the state response data. The difference in control quantity between adjacent time points is calculated for each channel. The sum of the squared differences of the three channels at each time point is accumulated and divided by the total number of time points to obtain the average control change rate, which is used as the control change index to reflect the smoothness of the control action. The tracking deviation index is then divided by the target tracking deviation to obtain the tracking performance ratio, and the control change index is divided by the target control change to obtain the control stability ratio. The two ratios are squared separately, multiplied by a weighting coefficient, and added to obtain the comprehensive performance index. This index is then compared with a performance threshold of 0.9. A comparison of zero and above is performed. When the index is less than the threshold, it indicates that the control performance is substandard and the controller parameters need to be adjusted. The relative magnitude of the tracking deviation index and the control change index is analyzed to determine the adjustment direction. If the tracking deviation is greater than 1.2 times the target value and the control change is less than the target value, it indicates insufficient tracking, and the output weight should be increased and the control weight decreased. If the tracking deviation is less than the target value but the control change exceeds 1.3 times the target value, it indicates over-control, and the output weight should be decreased and the control weight increased. The output weight and control weight in the rolling optimization objective function are expanded into a weight parameter vector. The partial derivatives of the comprehensive performance index with respect to each element in the vector are calculated to form a gradient vector. The gradient direction indicates the direction of parameter change that increases the performance index the fastest. The gradient vector is multiplied by the learning rate of 0.05 to obtain the weight adjustment vector. The adjustment vector is added to the current output weight and control weight to complete the parameter update and obtain the adjusted weight parameters. The updated weight parameters will be used in the rolling optimization of the next control cycle, so that the controller gradually adapts to the changes in process characteristics and performance requirements.

[0039] In one specific embodiment, step S1 includes:

[0040] The first rheological parameter at the inlet of the flotation cell, the second rheological parameter in the middle, and the third rheological parameter at the outlet are obtained. A set of constraint equations is constructed based on the material conservation relationship. The first, second, and third rheological parameters are substituted into the set of constraint equations and weighted least squares is used to solve the problem to obtain the preliminary fusion value.

[0041] Calculate the measurement residual vector of the constraint equation system, standardize the measurement residual vector to obtain the standardized residual of each measurement point;

[0042] Based on the standardized residuals, abnormal measurement points are identified. After reducing the weight coefficients corresponding to the abnormal measurement points, the weighted least squares solution is performed again to obtain the corrected fusion value.

[0043] By combining the corrected fusion value with bubble characteristic parameters and electrochemical parameters, an estimated fusion state value is obtained.

[0044] Specifically, the first rheological parameter at the inlet of the flotation cell is obtained by measuring the viscosity of the feed slurry using a rotary viscosity sensor. The second rheological parameter in the middle section is obtained by measuring the viscosity of the slurry in the flotation reaction zone using a similar sensor installed in the middle of the cell. The third rheological parameter at the outlet is obtained by measuring the viscosity of the concentrate slurry using a sensor at the overflow weir. The rheological parameters at the three measuring points are physically correlated because the continuous flow of the slurry in the flotation cell follows the law of conservation of mass. Based on the material conservation relationship, a set of constraint equations is established. The principle of constructing the set of constraint equations is that although the viscosity of the slurry changes due to the addition of air bubbles and mineral separation during the flow from the inlet to the middle and then to the outlet, the change follows a definite physical law. Specifically, the first constraint equation describes the relationship between the inlet and middle measuring points. The first rheological parameter multiplied by the first transfer coefficient should equal the theoretical value of the second rheological parameter. The first transfer coefficient reflects the proportion of viscosity change caused by the addition of air bubbles during the flow of the slurry from the inlet to the middle. The second constraint equation describes the relationship between the measuring points in the middle and the outlet. The second rheological parameter multiplied by the second transfer coefficient should equal the theoretical value of the third rheological parameter. The second transfer coefficient reflects the proportion of viscosity change caused by concentrate separation during the flow of the slurry from the middle to the outlet. These two transfer coefficients were obtained through statistical analysis of a large amount of measurement data during historical normal operation. A first transfer coefficient of 0.92 indicates that the viscosity of the slurry generally decreases to 92% of its original value after flowing through the flotation reaction zone. A second transfer coefficient of 0.88 indicates that the viscosity further decreases to [a higher value] during concentrate overflow. The previous value was 88%. The constraint equations connect three independently measured rheological parameters through the physical constraint of material conservation, forming a mutually restrictive mathematical relationship. The actual measured values ​​of the first, second, and third rheological parameters are substituted into the constraint equations. Due to factors such as random noise, sensor drift, and local disturbances in the slurry during the measurement process, the actual measured values ​​often cannot accurately satisfy the theoretical relationship of the constraint equations. A set of fused values ​​is solved using the weighted least squares method to make the fused values ​​as close as possible to the measured values ​​of each measuring point and to satisfy the physical constraint relationship of the constraint equations as much as possible. The weighting reflects the difference in measurement reliability of different measuring points. The inlet measuring point is located in the feed pipeline where the flow is relatively stable, and the weight is set to 0.25. The middle measuring point is located in the critical flotation reaction zone. The weight that best reflects the process state is set to 0.30. The weight of the outlet measuring point, which is located at the overflow weir and is greatly affected by the foam layer fluctuation, is set to 0.20. The weighted least squares objective function is constructed as follows: the square of the difference between the measured value and the fused value of the first rheological parameter multiplied by the weight 0.25, plus the square of the difference between the measured value and the fused value of the second rheological parameter multiplied by the weight 0.30, plus the square of the difference between the measured value and the fused value of the third rheological parameter multiplied by the weight 0.20. At the same time, the fused value is required to satisfy the constraint equations or minimize the degree of violation of the constraint equations. The partial derivative of the weighted objective function with respect to the fused value is calculated and set to zero to obtain the optimal fused value, i.e., the preliminary fused value, which minimizes the objective function. The preliminary fused value integrates the information from the three measuring points and takes into account the material conservation constraints.

[0045] Calculating the measurement residual vector of the constraint equations is a crucial step in evaluating the measurement deviation at each measuring point. The method for calculating the measurement residual vector involves substituting the preliminary fused value back into the constraint equations to check their satisfaction. For the first constraint equation, the preliminary fused value is multiplied by the first transfer coefficient of 0.92 to obtain the theoretical predicted value for the middle measuring point. This theoretical predicted value is then subtracted from the actual measured value of the second rheological parameter to obtain the first residual component. This residual component reflects the deviation between the measured value at the middle measuring point and the theoretical value calculated based on the inlet measuring point and material conservation. For the second constraint equation, the preliminary fused value is multiplied by the first transfer coefficient of 0.92 and then by the second transfer coefficient of 0.88 to obtain the value for the outlet measuring point. The theoretical predicted value is obtained by subtracting the theoretical predicted value from the actual measured value of the third rheological parameter to obtain the second residual component. This residual component reflects the deviation between the measured value at the outlet measuring point and the theoretical value calculated based on the inlet measuring point and the two-stage material conservation. For the inlet measuring point itself, the measured value of the first rheological parameter is subtracted from the preliminary fusion value to obtain the third residual component. This residual component reflects the direct deviation between the measured value at the inlet measuring point and the fusion value. The three residual components constitute the measurement residual vector. Each element of the residual vector corresponds to the measurement deviation degree of a measuring point. The larger the absolute value of the residual, the worse the consistency between the measured value of that measuring point and the overall data. The residual vector contains the distribution information of the measurement error of the three measuring points.

[0046] Standardizing the measurement residual vector aims to convert residuals with dimensions and magnitudes from different measuring points into dimensionless, uniform, and comparable standardized residuals, facilitating the setting of a unified anomaly detection threshold. The specific standardization method first calculates the statistical characteristics of the residual vector, including the residual mean and residual standard deviation. The residual mean is calculated by adding the three elements of the residual vector and dividing by 3, reflecting the central tendency of the residuals. The residual standard deviation is calculated by taking the square root of the sum of the squares of the differences between each residual and the mean, divided by 3, reflecting the dispersion of the residuals. Then, a standardization transformation is performed on each element of the residual vector. The standardization transformation formula is: subtract the residual mean from the residual value of that element and then divide by the residual standard deviation. The standardized residuals eliminate dimensions, have a mean of zero, and a standard deviation of one, following a standard normal distribution. The standardized residuals of each measuring point correspond to the deviation degree of the entrance, middle, and exit measuring points under a unified standard. The larger the absolute value of the standardized residual, the further the measuring point deviates from the normal distribution.

[0047] The three-sigma criterion is used as the statistical test standard to determine outlier points based on standardized residuals. The three-sigma criterion, based on the normal distribution theory, states that the probability of normal data falling within the range of plus or minus three standard deviations of the mean is 99.73%. When the absolute value of the standardized residual of a measurement point is greater than 3, the data at that point falls into the extreme tail region of the normal distribution and is determined to be an outlier. Causes of outliers include measurement lag due to fouling on the sensor surface, impulse noise introduced by electrical interference, and unrepresentative measurements due to sudden changes in local slurry properties. The weight coefficient corresponding to the outlier point is adjusted from its original value to one-fifth of its original value. This reduction in weight means that when re-performing the weighted least squares solution, the outlier point will negatively impact the fusion result. The impact of the results is greatly weakened, but the abnormal measurement point data is still retained in the calculation instead of being completely removed because although its measurement value has deviation, it still contains some information about the real process. The weighted least squares objective function is reconstructed using the adjusted weight matrix. The weight coefficients of the terms corresponding to abnormal measurement points in the objective function have been updated to the reduced values, while the weights of normal measurement points remain unchanged. The partial derivative of the updated objective function with respect to the fusion value is calculated again and the derivative is set to zero to obtain the corrected fusion value considering the anomaly suppression. Compared with the initial fusion value, the corrected fusion value is less affected by the interference of abnormal measurement points and more accurately reflects the real rheological state of the slurry in the flotation cell. The anomaly detection and weight adjustment mechanism enables the redundant fusion algorithm to have adaptive anti-interference ability.

[0048] The corrected fusion value, combined with bubble characteristic parameters and electrochemical parameters, forms a complete process state description. The bubble characteristic parameter, the Sutter mean diameter, is measured using an ultrasonic detection device, reflecting the bubble size distribution during flotation. The Sutter mean diameter is defined as six times the ratio of bubble swarm volume to surface area, characterizing the statistical average characteristics of bubble size. The electrochemical parameter, pulp conductivity, is measured using an electromagnetic probe, reflecting the ion concentration and chemical environment state in the pulp. High conductivity indicates a high concentration of soluble ions, affecting reagent adsorption and mineral surface electrical properties. These three types of parameters characterize the flotation process from different physical dimensions. The corrected fusion value describes the pulp rheological properties, reflecting the pulp viscous resistance and flow characteristics. The properties of the bubble feature parameters describe the gas-liquid interface characteristics, reflecting the geometric conditions of the collision and adhesion between the bubble and the mineral particles. The electrochemical parameters describe the chemical environment state, reflecting the chemical background of the reagent action. The corrected fusion value is used as the first element of the state vector, the bubble feature parameters as the second element, and the electrochemical parameters as the third element. The three elements are arranged and combined in the form of a column vector to obtain a fusion state estimation vector containing three numerical elements. This vector comprehensively represents the integrated operating state of the flotation process at the current moment in three dimensions: rheological properties, bubble morphology, and chemical environment. The fusion state estimation vector serves as the basic data source for the state feedback signal of the control system.

[0049] In one specific embodiment, step S2 includes:

[0050] Construct a state-space expression based on the fused state estimate, and expand the state transition matrix and input matrix in the state-space expression into a parameter vector;

[0051] Collect the control input sequence and the actual output sequence, construct the regression vector and covariance matrix, substitute the regression vector and covariance matrix into the recursive least squares formula to identify the parameters, and obtain the identified parameter vector.

[0052] Calculate the rate of change of the identification parameter vector with respect to the parameter vector at the previous moment, determine the convergence state of the parameters based on the rate of change of the parameters, and obtain the convergence determination result;

[0053] When the convergence determination result meets the convergence condition, the identified parameter vector is output as the updated parameter vector, and the covariance matrix is ​​updated.

[0054] Specifically, the state-space expression comprises two parts: state equations and output equations. The state equations describe how the state vector at the next moment is jointly determined by the current state vector and the control input vector. The output equations describe how the output vector is obtained by mapping the state vector and the control input vector. The general form of the state equations is that the state vector at the next moment equals the state transition matrix multiplied by the current state vector plus the input matrix multiplied by the current control input vector. The state transition matrix describes the dynamic evolution of the process itself, reflecting the mutual influence between state variables. The input matrix describes the strength of the control input's effect on state changes, reflecting how the aeration rate, stirring intensity, and reagent dosage change the flotation process state. The output equation is in the form that the output vector equals the output matrix multiplied by the state vector plus the direct transfer matrix multiplied by the control input vector. The input and output vectors contain concentrate quality and recovery indicators. The fused state estimates are used as real-time values ​​of the state vectors and substituted into the state-space expression. Expanding the state transition matrix and input matrix in the state-space expression into parameter vectors means arranging all elements of the matrix sequentially by rows or columns into a long vector. The state transition matrix is ​​5 rows and 5 columns with 25 elements, the input matrix is ​​5 rows and 3 columns with 15 elements, the output matrix is ​​2 rows and 5 columns with 10 elements, and the direct transfer matrix is ​​2 rows and 3 columns with 6 elements. All elements of the four matrices are concatenated to form a parameter vector containing 56 elements. Each element of the parameter vector corresponds to a parameter to be identified in the state-space model. These parameters reflect the dynamic characteristics of the flotation process and change with the ore properties, requiring online identification.

[0055] Acquiring the control input sequence and the actual output sequence forms the data foundation for parameter identification. The control input sequence includes historical values ​​of aeration rate, stirring intensity, and reagent dosage at multiple consecutive sampling times. The actual output sequence includes measured values ​​of concentrate quality indicators and recovery rate indicators at corresponding times. Data from the most recent 10 sampling times are selected to form the identification data window. Constructing the regression vector involves combining historical state values ​​and historical input values ​​in a specific way. The regression vector is constructed by concatenating the state vector and the control input vector from the previous time step column-wise according to the state equation. The dimension of the regression vector equals the dimension of the state vector plus the dimension of the control input vector, i.e., 5 plus 3 equals 8. The covariance matrix describes the uncertainty of parameter estimation and the correlation between parameters. The covariance matrix is ​​a 56x56 symmetric positive definite matrix initialized as an identity matrix multiplied by a large constant of 100 to indicate that the initial parameters are completely unknown. The regression vector and covariance matrix are substituted into the recursive least squares formula for parameter identification. The identification algorithm, using a recursive least squares formula, comprises three sub-steps: gain calculation, parameter update, and covariance update. Gain calculation involves multiplying the covariance matrix by the regression vector, dividing by the forgetting factor, adding the transpose of the regression vector, multiplying by the covariance matrix, and then multiplying by the regression vector again. The forgetting factor, set to 0.98, indicates that a decay weight is applied to older data, making the identification focus more on recent data. Parameter update involves adding the previous time step parameter vector to the gain multiplied by the prediction error. The prediction error equals the actual measured output minus the predicted output calculated based on the previous time step parameters and the current input. Covariance update involves subtracting the gain from the previous time step covariance matrix, multiplying by the transpose of the regression vector, multiplying by the covariance matrix, and then dividing the whole result by the forgetting factor. The recursive least squares algorithm continuously refines the parameter estimates through rolling calculations at each time step, bringing the parameter vector closer to the true value. The resulting identification parameter vector serves as the latest estimate of the process model parameters at the current time step. The identification parameter vector is continuously updated to track changes in process characteristics as new data is added.

[0056] The rate of change of the identified parameter vector compared to the parameter vector at the previous moment is used to determine whether the parameter identification has converged to a stable value. The method for calculating the rate of change is to subtract the parameter vector at the previous moment from the current identified parameter vector to obtain the parameter increment vector. The Euclidean norm of the parameter increment vector is then calculated to obtain the total magnitude of the parameter change. This is then divided by the Euclidean norm of the parameter vector at the previous moment for normalization to obtain the relative rate of change. The Euclidean norm is the square root of the sum of the squares of all elements of the vector, reflecting the overall size of the vector. The rate of change is a dimensionless quantity reflecting the drasticness of parameter updates. The convergence state of the parameters is determined based on the rate of change. When the rate of change is small... When the relative change of the parameter is less than 2%, the parameter identification is considered to have entered the convergence state. However, a single fulfillment of the condition is not enough to determine convergence. It needs to be met continuously for multiple sampling periods to confirm. The number of periods for continuously meeting the threshold is set to 20. The parameter change rate at each time moment is recorded to form a judgment sequence. It is checked whether the judgment sequence of the most recent 20 time moments is all true. If all are true, the convergence judgment result is that convergence has been obtained. If any one is false, the convergence judgment result is that convergence has not been obtained. The convergence judgment mechanism prevents temporary parameter fluctuations from being mistakenly judged as convergence and ensures the stability and reliability of the identified parameters.

[0057] When the convergence determination result meets the convergence condition, i.e., convergence is determined, the parameter output operation is executed. The identified parameter vector is output as the updated parameter vector for subsequent controller use. The updated parameter vector is the parameter estimate that has been fully identified and converged, accurately reflecting the true dynamic characteristics of the current flotation process. The updated parameter vector is reconstructed into the form of a state transition matrix and an input matrix by performing the inverse operation of parameter vector expansion. The first 25 elements are arranged in 5 rows and 5 columns to restore the state transition matrix, and the 26th to 40th elements are arranged in 5 rows and 3 columns to restore the input matrix. Subsequent predictive control calculations use the reconstructed matrix and update the covariance matrix to prepare for identification at the next time step. The covariance matrix is ​​updated according to the recursive least squares covariance update formula. The updated covariance matrix reflects the uncertainty of the current parameter estimation. The smaller the diagonal elements of the covariance matrix, the more accurate the corresponding parameter estimation. The parameter identification process continues, and a recursive calculation is performed at each sampling time. When the ore properties change, causing the process characteristics to change, the parameter vector will deviate from the convergence state and the rate of change exceeds the threshold. The identification algorithm automatically detects the parameter change and restarts the convergence judgment process. Online parameter identification enables the internal model of the controller to always track the actual process characteristics, solving the problem that the fixed parameter model in the existing technology cannot adapt to the fluctuation of ore properties.

[0058] See Figure 2 , Figure 2 This is a schematic diagram of the parameter identification convergence process in the embodiments of this application. Figure 2The graph illustrates the variation of the parameter change rate with the sampling period during online parameter identification using the recursive least squares algorithm. The horizontal axis represents the sampling period, and the vertical axis represents the parameter change rate (%). The dashed line represents the 2% convergence threshold. As shown in the graph, the parameter change rate rapidly decreases from approximately 15% in the initial stage. Within 30-70 periods, the parameter change rate falls below the threshold and remains stable. When it remains below 2% for 20 consecutive periods, the system determines that the parameters have converged. Around the 70th period, due to changes in ore properties, the parameter change rate rises to approximately 6%. After detecting this change in process characteristics, the identification algorithm restarts the convergence determination. Within 85-100 periods, the parameter change rate again rapidly decreases and tends to stabilize. This verifies that the recursive least squares algorithm possesses the adaptive tracking capability to process characteristic changes. This mechanism enables the internal model of the controller to track the dynamic characteristics of the flotation process in real time, solving the problem that fixed-parameter models cannot adapt to fluctuations in ore properties.

[0059] In one specific embodiment, step S3 includes:

[0060] The updated parameter vector is decomposed into a state transition matrix and an input matrix according to the dimension, and the output prediction value sequence at each time point in the prediction time domain is calculated based on the state transition matrix and the input matrix.

[0061] The difference between the output predicted value sequence and the reference trajectory at the corresponding time is calculated point by point, and the weighted square sum of the differences at each time is obtained to obtain the tracking error term;

[0062] The control cost term is obtained by weighted summation of the control increments at each time point in the control time domain. The tracking error term is added to the control cost term to form the rolling optimization objective function.

[0063] The upper limits of the rate of change of aeration volume, the rate of change of stirring intensity, and the rate of change of drug dosage are set as constraints. The rolling optimization objective function is solved by quadratic programming to obtain the control increment sequence. The first element of the control increment sequence is added to the control quantity at the previous time step to obtain the optimal control output.

[0064] Specifically, decomposing the update parameter vector into a state transition matrix and an input matrix by dimension refers to restoring it from a long vector form to a matrix form. The update parameter vector contains 56 elements. The first 25 elements are filled in a 5x5 grid to reconstruct the state transition matrix, and the 26th to 40th elements are filled in a 5x3 grid to reconstruct the input matrix. The state transition matrix describes the evolution of the state vector from the current time step to the next time step, and the input matrix describes the driving effect of the control input vector on the state changes. Based on the state transition matrix and the input matrix, the output predicted value sequence at each time step in the prediction time domain is calculated. The prediction time domain is set to 10 sampling cycles. The period represents a 10-step prediction of the future. The prediction calculation adopts a recursive method. The first step of the predicted state is equal to the state transition matrix multiplied by the current state vector plus the input matrix multiplied by the assumed first step control input. The second step of the predicted state is equal to the state transition matrix multiplied by the first step predicted state plus the input matrix multiplied by the assumed second step control input. This process is repeated until the tenth step to obtain 10 predicted state vectors. The predicted state vectors are multiplied by the output matrix to obtain the corresponding output predicted values. The output predicted values ​​include the predicted values ​​of concentrate quality indicators and recovery rate indicators. The output predicted values ​​at the 10 time points are arranged in chronological order to form the output predicted value sequence.

[0065] Calculating the point-by-point difference between the output predicted value sequence and the reference trajectory at the corresponding time point is the basis for evaluating tracking performance. The reference trajectory is the expected output change path, containing the target values ​​for concentrate quality and recovery rate at each time point. The reference trajectory is determined by the production plan and reflects the process indicator requirements. For each time point within the prediction time domain, the difference between the output predicted value and the reference trajectory is calculated. The difference at the first time point is the output predicted value at the first time point minus the reference value at the first time point; the difference at the second time point is the output predicted value at the second time point minus the reference value at the second time point. This process is repeated to obtain 10 difference vectors. Each difference vector contains two components: quality deviation and recovery rate deviation. The weighted sum of the squared differences at each time point is calculated. The weighting reflects the differences in importance of different time points and different output variables. The weight of the quality indicator is set to 100 because it is directly related to product quality, and the weight of the recovery rate indicator is set to 80 because it affects economic benefits. The weight of the long-term prediction time is slightly lower than that of the near-term time point to reflect the increase in prediction uncertainty over time. Specifically, the weighted error at the first time point is obtained by multiplying the square of the quality deviation at the first time point by the weight of 100 and adding the square of the recovery rate deviation at the first time point by the weight of 80. The weighted errors at 10 time points are accumulated to obtain the tracking error term. The larger the value of the tracking error term, the more serious the deviation between the predicted output and the expected trajectory.

[0066] The control cost term is obtained by summing the weighted squares of the control increments at each time point within the control time domain. The control time domain is set to three sampling periods, indicating that only the first three control actions are optimized. The control time domain is shorter than the prediction time domain because the uncertainty of long-term control actions is high and they are not executed immediately. The control increment is the difference between the control values ​​at adjacent time points, reflecting the magnitude of the change in control action. The control increment at the first time point is the control output at the first time point minus the control value at the current time point; the control increment at the second time point is the control output at the second time point minus the control output at the first time point; and the control increment at the third time point is the control output at the third time point minus the control output at the second time point. Each control increment includes three components: the aeration volume increment, the stirring intensity increment, and the drug dosage increment. The control increments at the three control time points are summed by weighted squares, with the aeration volume increment weighted... The weight of the aeration system increment is set to 1.5 because the aeration system responds quickly and changes need to be limited. The weight of the stirring intensity increment is set to 2.0 because the large power of the stirring motor changes too quickly and affects the equipment life. The weight of the drug dosage increment is set to 1.0 because the drug adjustment is relatively mild. The control cost at the first moment is calculated by multiplying the square of the aeration volume increment by 1.5, adding the square of the stirring intensity increment by 2.0, and adding the square of the drug dosage increment by 1.0. The control costs at the three moments are accumulated to obtain the control cost term. The larger the value of the control cost term, the more drastic the control action. The tracking error term is added to the control cost term to form the rolling optimization objective function. The objective function comprehensively considers the requirements of both tracking performance and control stability. Minimizing the objective function means finding the optimal control strategy that can make the output track the reference trajectory while maintaining the stability of the control action.

[0067] Upper limits for the rates of change in aeration volume, stirring intensity, and dosage are set as constraints to ensure the physical feasibility of the control output. The upper limit for the rate of change in aeration volume is set at 0.15 cubic meters per minute, indicating that the adjustment range of the aeration volume per unit time does not exceed this value to prevent bubble layer instability. The upper limit for the rate of change in stirring intensity is set at 15 revolutions per minute, indicating that the speed adjustment rate is limited by the acceleration and deceleration capabilities of the motor. The upper limit for the rate of change in dosage is set at 10 grams per ton, representing a constraint on the flow rate adjustment range of the dosing pump. These three upper limits are transformed into inequality constraints on the control increment: the absolute value of the increment in aeration volume is less than or equal to 0.15 cubic meters per minute, the absolute value of the increment in stirring intensity is less than or equal to 15 revolutions per minute, and the absolute value of the increment in dosage is less than or equal to 10 grams per ton. Within the control time domain, there are nine control increment components at three time points, each corresponding to a pair of upper and lower bound constraints. A quadratic programming approach is used to solve the rolling optimization objective function. Quadratic programming solves optimization problems with quadratic objective functions and linear constraints. The tracking error term and control cost term in the objective function are both control increments. The quadratic function is constrained by linear inequalities of control increments. The quadratic programming solver iteratively calculates the optimal control increment sequence that minimizes the objective function and satisfies all constraints using the effective set method or interior point method. The control increment sequence contains 9 control increment components at 3 time points. Taking the first element of the control increment sequence, which contains the aeration volume increment, stirring intensity increment, and drug dosage increment at the first time point, these three increments are added to the corresponding control quantities at the previous time point. The aeration volume increment at the previous time point is added to the aeration volume increment at the current time point to obtain the optimal aeration volume output at the current time point. The stirring intensity increment at the previous time point is added to the stirring intensity increment at the current time point to obtain the optimal stirring intensity output at the current time point. The drug dosage increment at the previous time point is added to the drug dosage increment at the current time point to obtain the optimal drug dosage output at the current time point. The three output values ​​constitute the optimal control output vector. Rolling optimization is performed once per sampling period, implementing only the first step of control increment each time. The optimization problem is resolved based on the new state measurement in the next period. The rolling optimization mechanism enables the controller to continuously adjust the control strategy according to the process state feedback to achieve closed-loop optimal control.

[0068] In one specific embodiment, step S4 includes:

[0069] The optimal control output is decomposed into aeration volume control command, stirring intensity control command and drug dosage control command according to the control channel. The difference between each control command and the control quantity at the previous moment is calculated to obtain the aeration volume adjustment sequence, stirring intensity adjustment sequence and drug dosage adjustment sequence.

[0070] The inflation volume adjustment sequence is decomposed into three step increments. Each step increment is executed sequentially at a set time interval. After each step is executed, the fusion state estimate is collected, the state change rate is calculated, and the subsequent step increment execution is paused when the state change rate exceeds the threshold.

[0071] After the aeration volume adjustment sequence is completed, the stirring intensity adjustment sequence is converted into a ramp tracking command. The stirring intensity is gradually adjusted according to the set slope, and the state coupling index is monitored simultaneously. When the state coupling index exceeds the upper limit, the adjustment amplitude is reduced.

[0072] When the stirring intensity adjustment sequence reaches the preset progress, the drug dosage adjustment sequence is started to add drugs, and the execution trajectory of each control channel and the estimated fusion state value at the corresponding time are recorded to obtain the state response data.

[0073] Specifically, decomposing the optimal control output by control channel means splitting the control output vector containing three control variables into three independent control commands. The optimal control output vector contains three elements: aeration amount, stirring intensity, and drug dosage. The first element is extracted as the aeration amount control command, the second element as the stirring intensity control command, and the third element as the drug dosage control command. The difference between each control command and the control quantity at the previous moment is calculated. Subtracting the aeration amount from the aeration amount control command at the previous moment yields the aeration amount adjustment range, subtracting the stirring intensity from the stirring intensity control command at the previous moment yields the stirring intensity adjustment range, and subtracting the drug dosage from the drug dosage control command at the previous moment yields the drug dosage adjustment range. The three adjustment ranges constitute the single-element aeration amount adjustment sequence, stirring intensity adjustment sequence, and drug dosage adjustment sequence, respectively. A positive value in the adjustment sequence indicates an increase in the control quantity, and a negative value indicates a decrease in the control quantity.

[0074] The inflation volume adjustment sequence is decomposed into three step increments to avoid bubble layer instability caused by sudden changes in inflation volume. The decomposition ratios are set to 40%, 35%, and 25%. The first step increment is the inflation volume adjustment sequence value multiplied by 0.4, the second step increment is the inflation volume adjustment sequence value multiplied by 0.35, and the third step increment is the inflation volume adjustment sequence value multiplied by 0.25. The sum of the three step increments equals the original adjustment sequence value. Each step increment is executed sequentially at a set time interval of 1 second. At the current moment, the first step increment is executed, increasing the inflation volume from the previous moment's value by the first step increment. After a 1-second interval, the second step increment is executed, further increasing the inflation volume based on the first step increment. After another 1-second interval, the third step increment is executed to complete the entire adjustment. After the first stage is executed, the estimated fusion state value is collected. The estimated fusion state value is collected 1 second after the first stage is executed and recorded as the first collected value. The estimated fusion state value is collected 1 second after the second stage is executed and recorded as the second collected value. The dynamic response of the process is evaluated by calculating the rate of change of state. The rate of change of state is the current collected estimated fusion state value minus the previously collected estimated fusion state value and divided by the time interval of 1 second. The rate of change of state after the first stage is the first collected value minus the state value before execution and divided by 1 second. When the rate of change of state exceeds the threshold of 0.08 seconds, it is determined that the process response is too fast and there is a risk of instability. The subsequent step execution is paused. Pausing means that the second and third stage step amounts are no longer executed and the current inflation amount remains unchanged. The pause mechanism prevents the foam layer from collapsing or the tank from flooding due to the adjustment of inflation amount.

[0075] After the inflation volume adjustment sequence is completed, it is determined whether all three step increments have been executed or the process is prematurely paused due to an excessive rate of change in state. The end time of inflation volume adjustment is used as the start time of stirring intensity adjustment. The stirring intensity adjustment sequence is converted into a ramp tracking command. The ramp tracking command means that the control quantity changes linearly at a constant rate rather than a step change. The ramp rate is set to the stirring intensity adjustment sequence value divided by the ramp time of 6 seconds. During the ramp time, the stirring intensity increases by one-sixth of the adjustment range per second. The stirring intensity is gradually adjusted according to the set ramp rate. At the end of the inflation volume adjustment, the stirring intensity remains at the value of the previous moment. After 1 second, the stirring intensity increases by one ramp unit. After 2 seconds, the stirring intensity increases by two ramp units. After 6 seconds, the stirring intensity increases by a total of six ramp units, reaching the target value and completing the adjustment. The state coupling is monitored synchronously. The state coupling index describes the combined influence of changes in aeration volume and stirring intensity on the process state. The state coupling index is calculated as the sensitivity of the state to aeration volume multiplied by the amplitude of the aeration volume change, plus the sensitivity of the state to stirring intensity multiplied by the amplitude of the stirring intensity change. The sensitivity is extracted from the identified input matrix to reflect the state change caused by the unit control quantity change. The state coupling index is calculated once per second during the stirring intensity ramp adjustment process. When the state coupling index exceeds the upper limit of 0.15, it is determined that the combined effect of aeration volume and stirring intensity is too strong and there is a risk of coupled oscillation. The adjustment amplitude is then attenuated. The attenuation operation is to reduce the stirring intensity increment at subsequent time moments to 70% of the original value, and the slope is adjusted accordingly to the original slope multiplied by 0.7. After attenuation, the stirring intensity continues to be adjusted at a gentler rate until the attenuation is completed or triggered again.

[0076] When the stirring intensity adjustment sequence reaches the preset progress, the reagent dosage adjustment sequence is initiated. The preset progress is set to 50%, meaning that reagent addition begins when the stirring intensity adjustment is halfway complete. The total stirring intensity adjustment time is 6 seconds, and the 50% progress is reached after 3 seconds. At this point, the reagent dosage adjustment sequence's injection operation is initiated. The reagent dosage adjustment sequence is a single value representing the total amount of reagent to be added or reduced. The injection operation is achieved by controlling the dosing pump. The required reagent volume is calculated based on the reagent dosage adjustment sequence and the feed flow rate. The reagent volume equals the reagent dosage adjustment sequence multiplied by the flotation cell slurry volume divided by the reagent density. The injection time window is set to 180 seconds, and the dosing pump flow rate is controlled to be the reagent volume divided by 180 seconds to achieve constant-speed injection. The dosing pump starts when the stirring intensity adjustment reaches 50% and continuously injects reagent for 180 seconds to complete the reagent addition. The execution trajectory of each control channel is recorded. The inflation volume execution trajectory records the inflation volume value at each moment during the inflation volume adjustment process, including the values ​​at the three step moments. The stirring intensity execution trajectory records the stirring intensity value per second during the ramp adjustment process. The drug dosage execution trajectory records the cumulative drug dosage value added during the injection process. The estimated value of the fusion state at the corresponding moment is recorded. The estimated value of the fusion state is collected at regular intervals after each step of inflation volume, per second during the stirring intensity adjustment process, and during the drug injection process. The execution trajectory data and the estimated value data are linked by timestamp to form a time series dataset. The dataset includes time column, inflation volume column, stirring intensity column, drug dosage column, rheological state column, bubble characteristic column, and electrochemical state column to obtain the state response data. The state response data completely records the execution process of the control action and the trajectory of process state changes, reflecting the control effect.

[0077] In one specific embodiment, step S5 includes:

[0078] The output sequence is extracted from the state response data. The output sequence is time-aligned with the reference trajectory, and the difference is calculated point by point. The difference is squared, summed, and averaged to obtain the tracking deviation index.

[0079] The execution trajectory of each control channel is extracted from the state response data. The control change index is obtained by summing the squares of the control quantity differences between adjacent time points of each control channel and taking the average value.

[0080] The tracking deviation index is calculated as a ratio to the target tracking deviation, and the control change index is calculated as a ratio to the target control change. The two ratios are then weighted and summed to obtain the comprehensive performance index.

[0081] The performance status is determined based on the comprehensive performance index. When the comprehensive performance index is lower than the threshold, the gradient of the output weight and control weight in the rolling optimization objective function with respect to the comprehensive performance index is calculated. The output weight and control weight are then updated numerically according to the gradient direction to obtain the adjusted weight parameters.

[0082] Specifically, extracting the output sequence from the state response data refers to filtering the time history of the output variables from the recorded time series data. The state response data contains multiple columns of data, among which the output variable columns correspond to the concentrate quality index and the recovery rate index. These two columns of data are extracted and arranged in chronological order to form the output sequence. The output sequence is a two-dimensional array, with each row corresponding to a time point and containing two elements: the quality value and the recovery rate value. The output sequence is then time-aligned with a reference trajectory, which is a preset expected output value time series that also contains the target quality and target recovery rate at each time point. Time alignment ensures that each time point in the output sequence and the reference trajectory corresponds one-to-one. After alignment, point-by-point difference is performed. The calculation process involves calculating the quality difference at the first moment, which is the output sequence's quality value at the first moment minus the target quality of the reference trajectory at the first moment. The recovery rate difference at the first moment is the output sequence's recovery rate value at the first moment minus the target recovery rate of the reference trajectory at the first moment. This process is repeated for all moments to obtain a difference sequence. The differences are squared and summed. The sum of the squared quality differences at each moment is the quality error summation. The sum of the squared recovery rate differences at each moment is the recovery rate error summation. The sum of the two sums is the total error summation. The average error summation is obtained by dividing the total error summation by the total number of moments. The average error summation is used as a tracking deviation index to quantify the accuracy of the output tracking reference trajectory.

[0083] Extracting the execution trajectory of each control channel from the state response data involves extracting the time series data of the aeration volume, stirring intensity, and drug dosage columns respectively. The aeration volume execution trajectory records the aeration volume value at each moment, the stirring intensity execution trajectory records the stirring intensity value at each moment, and the drug dosage execution trajectory records the cumulative drug dosage value at each moment. The difference between the control values ​​at adjacent moments of each control channel is squared and summed. For the aeration volume channel, the first aeration volume change value is obtained by subtracting the aeration volume at the first moment from the aeration volume at the second moment, and the second aeration volume change value is obtained by subtracting the aeration volume at the second moment from the aeration volume at the third moment. The difference between all adjacent moments is calculated in sequence, and the squares of all aeration volume change values ​​are summed to obtain the sum of squares of aeration volume change. The stirring intensity channel and the drug dosage channel are calculated using the same method to obtain their respective sums of squares of change. The sums of squares of change of the three control channels are added to obtain the total sum of squares of control change. The average control change is obtained by dividing the total sum of squares of control change by the total number of differences. The average control change is used as a control change index to quantify the severity of the control action.

[0084] The tracking deviation index is calculated as a ratio to the target tracking deviation, where the target tracking deviation is a preset acceptable tracking error level. Dividing the tracking deviation index by the target tracking deviation yields the tracking performance ratio. A ratio of 1 indicates that the tracking performance meets the target level, a ratio greater than 1 indicates insufficient tracking performance, and a ratio less than 1 indicates that the tracking performance exceeds the target. The control change index is calculated as a ratio to the target control change, where the target control change is a preset acceptable control fluctuation level. Dividing the control change index by the target control change yields the control stability ratio. A ratio of 1 indicates that control stability meets the target, a ratio greater than 1 indicates excessive control, and a ratio less than 1 indicates sufficient control stability. The two ratios are then weighted and squared, and the sum of the squares of the tracking performance ratio multiplied by a weighting factor of 0.55 and the squares of the control stability ratio multiplied by a weighting factor of 0.45 yields the comprehensive performance index. The weighting factor allocation reflects the control objective of prioritizing tracking performance over control stability. The comprehensive performance index is a dimensionless scalar comprehensive evaluation of the overall performance of the control system.

[0085] Performance status is determined based on comprehensive performance indicators, with a performance threshold of 0.90 set as the criterion. When the comprehensive performance indicator is greater than or equal to 0.90, the performance is considered satisfactory and controller parameters do not need adjustment. When the comprehensive performance indicator is lower than 0.90, the performance is considered unsatisfactory and controller parameters need adjustment. The gradients of the output weights and control weights in the rolling optimization objective function with respect to the comprehensive performance indicator are calculated. The gradient describes the rate of change of the comprehensive performance indicator caused by a small change in the weight parameters. The gradient of the output weight is the partial derivative of the comprehensive performance indicator with respect to the output weight, and the gradient of the control weight is the partial derivative of the comprehensive performance indicator with respect to the control weight. The gradient calculation uses a numerical differentiation method, where the output weight is increased by a small amount and then recalculated. The comprehensive performance index is calculated. The difference between the new index and the original index is divided by the weight increment to obtain an approximate value of the output weight gradient. The control weight gradient is calculated in the same way. The output weight and control weight are updated numerically according to the gradient direction. The gradient direction indicates the direction of parameter change that increases the comprehensive performance index. The updated output weight value is equal to the current output weight plus the learning rate multiplied by the output weight gradient. The updated control weight value is equal to the current control weight plus the learning rate multiplied by the control weight gradient. The learning rate is set to 0.05, which is the control parameter adjustment step size. The updated output weight and control weight are used as the adjusted weight parameters in the rolling optimization of the next control cycle. The adaptive adjustment of the weight parameters enables the controller performance to be continuously optimized.

[0086] In one specific embodiment, the performance status is determined based on the comprehensive performance index. When the comprehensive performance index is lower than a threshold, the gradients of the output weights and control weights in the rolling optimization objective function with respect to the comprehensive performance index are calculated. The output weights and control weights are then numerically updated according to the gradient direction to obtain the adjusted weight parameters, including:

[0087] The comprehensive performance index is compared with the performance threshold. When the comprehensive performance index is less than the performance threshold, it is determined to be a performance failure state. The relative deviation of the tracking deviation index and the control change index is extracted.

[0088] Based on the deviation relationship between the tracking deviation index and the target tracking deviation, and the deviation relationship between the control change index and the target control change, determine the direction of output weight adjustment and the direction of control weight adjustment.

[0089] The output weights and control weights in the rolling optimization objective function are combined into a weight parameter vector. The gradient vector is obtained by taking the partial derivative of the comprehensive performance index with respect to the weight parameter vector.

[0090] Multiply the gradient vector by the learning rate to obtain the weight adjustment amount. Add the weight adjustment amount to the current output weight and the current control weight to obtain the adjusted weight parameters.

[0091] Specifically, a performance threshold of 0.90 is set as the minimum standard for the control system to be considered qualified. When the comprehensive performance index value is greater than or equal to 0.90, it indicates that the overall performance of the control system in terms of tracking performance and control stability meets expectations and no adjustment of controller parameters is required. When the comprehensive performance index is less than 0.90, it is determined that the performance is substandard and the controller parameters need to be adjusted to improve performance. The relative deviation of the tracking deviation index and the control change index is extracted. The relative deviation describes the degree of deviation between the actual index and the target index. The relative deviation of the tracking deviation is the tracking deviation index minus the target tracking deviation and divided by the target tracking deviation. A positive ratio indicates that the tracking error exceeds the target and the tracking performance needs to be improved. A negative ratio indicates that the tracking error is less than the target and the tracking performance is good. The relative deviation of the control change is the control change index minus the target control change and divided by the target control change. A positive ratio indicates that the control fluctuation exceeds the target and the control stability needs to be enhanced. A negative ratio indicates that the control fluctuation is less than the target and the control is stable enough. The two relative deviation values ​​provide diagnostic information for adjusting the controller parameters.

[0092] The direction of output weight adjustment and control weight adjustment are determined based on the deviation relationship between the tracking deviation index and the target tracking deviation, and the deviation relationship between the control change index and the target control change. The direction of output weight adjustment depends on the tracking performance requirements. When the tracking deviation index is greater than 1.2 times the target tracking deviation, it indicates that the tracking error is too large, and the control change index is less than the target control change, indicating that the control action is not aggressive enough. It is determined that the output weight should be increased to make the controller pay more attention to tracking performance, while the control weight should be decreased to allow the control action to be more aggressive. When the tracking deviation index is less than the target tracking deviation, but the control change index is greater than 1.3 times the target control change, it indicates that the tracking performance is good, but the control is too drastic. It is determined that the output weight should be decreased to reduce the tracking requirements, while the control weight should be increased to suppress control fluctuations. The direction of output weight adjustment is either to increase or decrease. The direction of control weight adjustment is opposite to that of output weight, reflecting the trade-off between tracking performance and control stability. After the adjustment direction is determined, the specific adjustment range needs to be calculated.

[0093] The output weights and control weights in the rolling optimization objective function are combined into a weight parameter vector. The output weights contain two elements: quality index weight and recovery rate index weight. The control weights contain three elements: aeration volume increment weight, stirring intensity increment weight, and drug dosage increment weight. The five weight parameters are arranged in sequence to form the weight parameter vector. The partial derivatives of the comprehensive performance index with respect to the weight parameter vector are calculated. The partial derivatives describe the change in the comprehensive performance index caused by a unit change in the weight parameters. Since there is a complex nonlinear relationship between the comprehensive performance index and the weight parameters, it is difficult to obtain an analytical expression for the partial derivatives. Therefore, a numerical differentiation method is used to calculate the partial derivatives. A small perturbation is added to the first element of the weight parameter vector, the quality index weight, for example, by increasing the current value by 1%. Keeping the other weight parameters unchanged, the control loop is re-executed to calculate the new comprehensive performance index. The new index minus the original index gives the index increment. The index increment is divided by the weight perturbation to obtain the partial derivative of the quality index weight. Perturbations are applied to the five weight parameters in sequence to calculate the corresponding partial derivatives. The five partial derivatives are arranged in sequence to form a gradient vector. The gradient vector indicates the direction in which the weight parameters change will cause the comprehensive performance index to increase the fastest.

[0094] The weight adjustment is obtained by multiplying the gradient vector by the learning rate. The learning rate is a scalar parameter that controls the step size of the weight adjustment. A learning rate of 0.05 means that each adjustment is 5% of the gradient direction. An excessively large learning rate will cause over-adjustment and oscillation, while an excessively small learning rate will lead to slow convergence. Each element of the gradient vector is multiplied by the learning rate to obtain the adjustment amount of the corresponding weight parameter. The quality index weight adjustment amount is the quality index weight gradient multiplied by 0.05, and the recovery rate index weight adjustment amount is the recovery rate index weight gradient multiplied by 0.05. The three control increment weight adjustments are calculated using the same method. The five adjustments constitute the weight adjustment vector. The weight adjustments are superimposed on the current output weight and the current control weight. The current quality index weight is added to the corresponding adjustment amount to obtain the updated quality index weight, and the current recovery rate index weight is added to the corresponding adjustment amount to obtain the updated recovery rate index weight. The three control increment weights are also updated by adding their respective adjustments. The updated five weight parameters are used as adjusted weight parameters in the rolling optimization of the next control cycle. The gradient descent method enables the controller parameters to be gradually adjusted along the performance improvement direction.

[0095] The above describes the adaptive control method for process parameters of mineral processing equipment in the embodiments of this application. The following describes the adaptive control system for process parameters of mineral processing equipment in the embodiments of this application. Please refer to [link / reference]. Figure 3 One embodiment of the adaptive control system for process parameters of mineral processing equipment in this application includes:

[0096] The fusion module is used to acquire the first rheological parameter at the inlet of the flotation cell, the second rheological parameter and bubble characteristic parameter in the middle, and the third rheological parameter and electrochemical parameter at the outlet. The first rheological parameter, the second rheological parameter and the third rheological parameter are redundantly fused using the weighted least squares method to obtain the fusion state estimate.

[0097] The identification module is used to construct a state transition matrix and an input matrix based on the fused state estimate, and to identify the parameter vectors of the state transition matrix and the input matrix online using a recursive least squares algorithm to obtain an updated parameter vector.

[0098] The solution module is used to substitute the updated parameter vector into the rolling optimization objective function, perform quadratic programming on the control increment sequence, and obtain the optimal control output.

[0099] The grading module is used to decompose the optimal control output into an aeration volume adjustment sequence, a stirring intensity adjustment sequence, and a drug dosage adjustment sequence, and to execute the aeration volume adjustment sequence, the stirring intensity adjustment sequence, and the drug dosage adjustment sequence in a graded manner according to the time priority to obtain state response data;

[0100] The adjustment module is used to calculate the tracking deviation index and control change index based on the state response data, and to perform gradient adjustment on the output weights and control weights in the rolling optimization objective function to obtain the adjusted weight parameters.

[0101] above Figure 3 The adaptive control system for process parameters of the mineral processing equipment in this embodiment of the invention is described in detail from the perspective of modular functional entities. The adaptive control device for process parameters of the mineral processing equipment in this embodiment of the invention is described in detail from the perspective of hardware processing.

[0102] Reference Figure 4 This invention also provides an adaptive control device for process parameters of a mineral processing equipment. This adaptive control device can be a server, and its internal structure can be as follows: Figure 4 As shown. The adaptive control device for process parameters of the mineral processing equipment includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor in this computer design provides computing and control capabilities. The memory of the adaptive control device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the adaptive control device stores the data corresponding to this embodiment. The network interface of the adaptive control device is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0103] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the adaptive control device for process parameters of mineral processing equipment to which the present invention is applied.

[0104] 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, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the adaptive control method for process parameters of the mineral processing equipment.

[0105] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0106] 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, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a mineral processing equipment process parameter adaptive control device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0107] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An adaptive control method for process parameters of mineral processing equipment, characterized in that, The method includes: Step S1: Obtain the first rheological parameter at the inlet of the flotation cell, the second rheological parameter and bubble characteristic parameter in the middle, and the third rheological parameter and electrochemical parameter at the outlet. Perform redundant fusion processing on the first rheological parameter, the second rheological parameter and the third rheological parameter using the weighted least squares method to obtain the fusion state estimate. Step S2: Construct a state transition matrix and an input matrix based on the fused state estimate, and use a recursive least squares algorithm to identify the parameter vectors of the state transition matrix and the input matrix online to obtain the updated parameter vector; Step S3: Substitute the updated parameter vector into the rolling optimization objective function, perform quadratic programming on the control increment sequence, and obtain the optimal control output; Step S4: Decompose the optimal control output into an aeration volume adjustment sequence, a stirring intensity adjustment sequence, and a drug dosage adjustment sequence. Execute the aeration volume adjustment sequence, the stirring intensity adjustment sequence, and the drug dosage adjustment sequence in a hierarchical manner according to their time priority to obtain state response data. Step S5: Calculate the tracking deviation index and control change index based on the state response data, and perform gradient adjustment on the output weights and control weights in the rolling optimization objective function to obtain adjusted weight parameters. This includes: extracting the output sequence from the state response data, aligning the output sequence with the reference trajectory in time, calculating the difference point by point, summing the squares of the differences, and taking the average to obtain the tracking deviation index; extracting the execution trajectory of each control channel from the state response data, summing the squares of the control quantity differences between adjacent time points of each control channel, and taking the average to obtain the control change index; calculating the ratio of the tracking deviation index to the target tracking deviation, calculating the ratio of the control change index to the target control change, and summing the weighted squares of the two ratios to obtain the comprehensive performance index; determining the performance status based on the comprehensive performance index; when the comprehensive performance index is lower than a threshold, calculating the gradient of the output weights and control weights in the rolling optimization objective function with respect to the comprehensive performance index, and updating the output weights and control weights numerically according to the gradient direction to obtain adjusted weight parameters.

2. The adaptive control method for process parameters of mineral processing equipment according to claim 1, characterized in that, Step S1 includes: The first rheological parameter at the inlet of the flotation cell, the second rheological parameter in the middle, and the third rheological parameter at the outlet are obtained. A set of constraint equations is constructed based on the material conservation relationship. The first rheological parameter, the second rheological parameter, and the third rheological parameter are substituted into the set of constraint equations and weighted least squares solution is performed to obtain the preliminary fusion value. Calculate the measurement residual vector of the constraint equation system, and standardize the measurement residual vector to obtain the standardized residual of each measurement point; Based on the standardized residual, abnormal measurement points are identified. After reducing the weight coefficients corresponding to the abnormal measurement points, a weighted least squares solution is performed again to obtain the corrected fusion value. The corrected fusion value is combined with the bubble characteristic parameters and the electrochemical parameters to obtain the fusion state estimate.

3. The adaptive control method for process parameters of mineral processing equipment according to claim 1, characterized in that, Step S2 includes: A state-space expression is constructed based on the fused state estimate, and the state transition matrix and input matrix in the state-space expression are expanded into a parameter vector. Collect the control input sequence and the actual output sequence, construct the regression vector and the covariance matrix, and substitute the regression vector and the covariance matrix into the recursive least squares formula to identify the parameters and obtain the identified parameter vector. Calculate the rate of change of the identified parameter vector with respect to the parameter vector at the previous moment, determine the convergence state of the parameters based on the rate of change of the parameters, and obtain the convergence determination result; When the convergence determination result satisfies the convergence condition, the identification parameter vector is output as the update parameter vector, and the covariance matrix is ​​updated.

4. The adaptive control method for process parameters of mineral processing equipment according to claim 1, characterized in that, Step S3 includes: The updated parameter vector is decomposed into a state transition matrix and an input matrix according to the dimensions, and the output predicted value sequence at each time point in the prediction time domain is calculated based on the state transition matrix and the input matrix. The output predicted value sequence is compared with the reference trajectory at the corresponding time point by point. The weighted square sum of the differences at each time point is then obtained to obtain the tracking error term. The control increments at each moment in the control time domain are summed by weighted squares to obtain the control cost term. The tracking error term is added to the control cost term to form the rolling optimization objective function. The upper limits of the rate of change of aeration volume, the rate of change of stirring intensity, and the rate of change of drug dosage are set as constraints. The rolling optimization objective function is solved by quadratic programming to obtain the control increment sequence. The first element of the control increment sequence is added to the control quantity at the previous moment to obtain the optimal control output.

5. The adaptive control method for process parameters of mineral processing equipment according to claim 1, characterized in that, Step S4 includes: The optimal control output is decomposed into aeration volume control command, stirring intensity control command, and drug dosage control command according to the control channel. The difference between each control command and the control quantity at the previous moment is calculated to obtain the aeration volume adjustment sequence, stirring intensity adjustment sequence, and drug dosage adjustment sequence. The aeration volume adjustment sequence is decomposed into three step quantities. Each step quantity is executed sequentially at a set time interval. After each execution, the fusion state estimate is collected, and the state change rate is calculated. When the state change rate exceeds a threshold, the subsequent step execution is paused. After the aeration volume adjustment sequence is completed, the stirring intensity adjustment sequence is converted into a ramp tracking command. The stirring intensity is gradually adjusted according to the set slope, and the state coupling index is monitored synchronously. When the state coupling index exceeds the upper limit, the adjustment amplitude is attenuated. When the stirring intensity adjustment sequence reaches the preset progress, the drug dosage adjustment sequence is initiated, the execution trajectory of each control channel and the estimated fusion state value at the corresponding time are recorded to obtain state response data.

6. The adaptive control method for process parameters of mineral processing equipment according to claim 1, characterized in that, The process involves determining the performance status based on the comprehensive performance index. When the comprehensive performance index is below a threshold, the gradients of the output weights and control weights in the rolling optimization objective function with respect to the comprehensive performance index are calculated. The output weights and control weights are then numerically updated according to the gradient direction to obtain adjusted weight parameters, including: The comprehensive performance index is compared with the performance threshold. When the comprehensive performance index is less than the performance threshold, it is determined that the performance is substandard. The relative deviation between the tracking deviation index and the control change index is extracted. Based on the deviation relationship between the tracking deviation index and the target tracking deviation, and the deviation relationship between the control change index and the target control change, the direction of output weight adjustment and the direction of control weight adjustment are determined. The output weights and control weights in the rolling optimization objective function are combined into a weight parameter vector. The gradient vector is obtained by taking the partial derivative of the comprehensive performance index with respect to the weight parameter vector. The gradient vector is multiplied by the learning rate to obtain the weight adjustment amount. The weight adjustment amount is then added to the current output weight and the current control weight to obtain the adjusted weight parameters.

7. An adaptive control system for process parameters of mineral processing equipment, characterized in that, For implementing the adaptive control method for process parameters of mineral processing equipment as described in any one of claims 1 to 6, the adaptive control system for process parameters of mineral processing equipment comprises: The fusion module is used to acquire the first rheological parameter at the inlet of the flotation cell, the second rheological parameter and bubble characteristic parameter in the middle, and the third rheological parameter and electrochemical parameter at the outlet. The first rheological parameter, the second rheological parameter and the third rheological parameter are redundantly fused using the weighted least squares method to obtain the fusion state estimate. The identification module is used to construct a state transition matrix and an input matrix based on the fused state estimate, and to identify the parameter vectors of the state transition matrix and the input matrix online using a recursive least squares algorithm to obtain an updated parameter vector. The solution module is used to substitute the updated parameter vector into the rolling optimization objective function, perform quadratic programming on the control increment sequence, and obtain the optimal control output. The grading module is used to decompose the optimal control output into an aeration volume adjustment sequence, a stirring intensity adjustment sequence, and a drug dosage adjustment sequence, and to execute the aeration volume adjustment sequence, the stirring intensity adjustment sequence, and the drug dosage adjustment sequence in a graded manner according to the time priority to obtain state response data; The adjustment module is used to calculate the tracking deviation index and control change index based on the state response data, and to perform gradient adjustment on the output weights and control weights in the rolling optimization objective function to obtain adjusted weight parameters. This includes: extracting the output sequence from the state response data; aligning the output sequence with the reference trajectory in time and calculating the difference point by point; summing the squares of the differences and taking the average to obtain the tracking deviation index; extracting the execution trajectory of each control channel from the state response data; summing the squares of the control quantity differences between adjacent time points of each control channel and taking the average to obtain the control change index; calculating the ratio of the tracking deviation index to the target tracking deviation, and calculating the ratio of the control change index to the target control change; summing the weighted squares of the two ratios to obtain a comprehensive performance index; determining the performance status based on the comprehensive performance index; and when the comprehensive performance index is below a threshold, calculating the gradient of the output weights and control weights in the rolling optimization objective function with respect to the comprehensive performance index, and updating the output weights and control weights numerically according to the gradient direction to obtain adjusted weight parameters.

8. An adaptive control device for process parameters of a mineral processing plant, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the adaptive control method for process parameters of the mineral processing equipment as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to execute the adaptive control method for process parameters of the mineral processing equipment as described in any one of claims 1 to 6.

Images