A method and system for controlling the concentration of dense underflow based on multi-source parameter fusion

Abstract

The application relates to a thickening underflow concentration control method and system based on multi-source parameter fusion, relates to the thickening field, and comprises the following steps: collecting multi-source parameters of a thickening process; establishing an underflow concentration soft measurement model based on the multi-source parameters, estimating the concentration and the uncertainty; taking the concentration estimation value as a controlled variable, adopting an anti-integral saturation PI control law to close-loop adjust the pipe clamp valve opening degree; dynamically calculating the flocculant adding amount based on the overflow turbidity and the rake torque or the mud layer thickness, and rolling optimizing the reagent consumption of the unit solid removal amount; automatically recommending the start and stop number of tables according to the underflow solid flow and the filter processing capacity, and adopting a hysteresis window to cooperatively control; built-in expert rules are used to identify abnormal working conditions and perform disposal, and feedforward compensation is performed according to the feed change. The scheme provided by the application can realize multi-target cooperative optimization of the thickening process, improve the underflow concentration stability, the overflow clarity and the equipment safety, and improve the automation level of the thickening and filtration whole process.

Description

Technical Field

[0001] This application relates to the field of mineral processing thickening technology, and in particular to a method and system for controlling the concentration of thickening underflow based on multi-source parameter fusion. Background Technology

[0002] In related technologies, the thickening process is a key step in solid-liquid separation in mineral processing. Its core objective is to concentrate low-concentration slurry into a high-concentration underflow while obtaining a clear overflow to provide a stable material for downstream filtration or tailings disposal. Parameters such as underflow concentration, overflow turbidity, rake torque, and mud layer thickness are interdependent and difficult to control, directly affecting the working efficiency of subsequent filters, flocculant consumption, and overall production safety.

[0003] Traditional underflow concentration detection relies heavily on manual sampling and offline testing, with sampling intervals typically ranging from 1 to 2 hours. Measurement results lag significantly behind changes in operating conditions. Operators can only make post-hoc adjustments based on the outdated concentration data, unable to address real-time disturbances such as feed fluctuations and changes in flocculation efficiency. This results in significant fluctuations in underflow concentration between 70% and 80%, leading to frequent occurrences of overflow turbidity and excessive flocculant addition.

[0004] Currently, there are also thickening underflow concentration control systems with added automation, but these systems mostly use single-parameter PID control of the underflow valve or fixed-value flocculant addition, resulting in simple control logic. Operators can only subjectively judge the mud layer state based on limited data such as rake torque and underflow pressure, and manually adjust the underflow valve opening and dosage. Different shifts have significantly different operating habits, making it difficult to achieve continuous, stable, and precise control 24 hours a day. Furthermore, they cannot respond quickly and automatically to abnormal conditions such as rake jamming, thickening overflow, or thinning overflow, easily leading to production accidents. Moreover, the thickening process typically acts as a buffer between grinding and filtration, without effective linkage with upstream grinding yield and downstream ceramic filter capacity. Fluctuations in upstream feed solids content and flow rate are directly transmitted to the thickening unit, causing mud layer instability. Simultaneously, irregular changes in underflow concentration and flow rate lead to frequent start-ups and shutdowns of downstream filters, increasing backflushing frequency, raising equipment energy consumption and maintenance costs, and restricting overall production efficiency.

[0005] Meanwhile, existing control schemes often focus only on the underflow concentration, failing to simultaneously consider multiple interdependent objectives such as overflow clarity, rake frame safety torque, flocculant consumption, and compatibility with downstream filtration equipment. When operating conditions fluctuate, the system exhibits poor robustness, significant attenuation of control effectiveness, and difficulty in achieving economical and stable operation of the entire thickening-filtration process.

[0006] In summary, there is an urgent need for an intelligent control method and system for the concentration process that can integrate multiple source parameters, achieve online soft measurement of undercurrent concentration, and possess multi-objective collaborative optimization and automatic handling capabilities for abnormal operating conditions, in order to solve the problems of detection lag, coarse control, and insufficient coordination in the existing technology. Summary of the Invention

[0007] To address or partially address the problems existing in related technologies, this application provides a method and system for controlling the concentration of dense undercurrent based on multi-source parameter fusion, aiming to solve the problems of existing dense control relying on human experience, poor coordination between upstream and downstream processes, and lack of multi-objective collaborative optimization capabilities.

[0008] The first aspect of this application provides a method for controlling the concentration of dense undercurrent based on multi-source parameter fusion, including: Collect multi-source parameters of the thickening process, including the rake frame torque, mud layer thickness, overflow turbidity of the thickener body, underflow pressure, underflow flow rate, online pipeline concentration of the underflow pipeline, feed flow rate and feed concentration on the feed side, and the number of downstream ceramic filters that are in operation. Based on multi-source parameters, a soft measurement model for underflow concentration is established to estimate the underflow concentration and the corresponding uncertainty. The soft measurement model adopts a combination of recursive partial least squares algorithm and Kalman filter, or a combination of autoregressive model with hysteresis term and Kalman filter. The estimated underflow concentration is the controlled variable, and the opening of the pinch valve is the manipulated variable. Closed-loop control is performed according to the set target range of underflow concentration to adjust the opening of the pinch valve. The closed-loop control adopts a proportional-integral control law that resists integral saturation and limits the change step size of the pinch valve opening to suppress liquid-solid disturbances. Based on the dual indicators of overflow turbidity and rake torque or mud layer thickness, the flocculant addition amount is dynamically calculated, and the flocculant consumption value per unit of solids removal is continuously compared to maintain the flocculant consumption value within the historical optimal range. Based on the underflow solids flow rate and the processing capacity of a single filter, the recommended number of ceramic filters to be turned on is calculated, and a hysteresis window strategy is adopted to control the switching of the number of filters. Built-in expert rules identify abnormal operating conditions and take corresponding abnormal handling actions, while also performing feedforward compensation based on changes in feed flow rate and feed concentration.

[0009] Optionally, a soft-sensor model for underflow concentration is established to estimate the underflow concentration and the corresponding uncertainty, including: It integrates underflow pressure, online pipeline concentration, rake torque, mud layer thickness, feed flow rate, and feed concentration; A regression model between input features and bottom flow concentration was established using a recursive partial least squares algorithm. The Kalman filter was then used to filter and reduce noise in the model output to determine the estimated bottom flow concentration and the confidence interval. Alternatively, a combination of an autoregressive model with hysteresis terms and a Kalman filter can be used: In the formula, This is the estimated underflow concentration at the current time t. These are the autoregressive coefficients, representing the weights of the linear contribution of past estimates to the current estimate. The input feature coefficients represent the linear contribution weights of each input feature to the current estimate. The j-th input feature is The value at time, This is the noise term.

[0010] Optionally, closed-loop control is performed based on a set target range for underflow concentration to adjust the opening of the pinch valve, including: Set the mass fraction of the target range for underflow concentration; By employing a proportional-integral control law that resists integral saturation, the opening degree of the pinch valve is obtained, which is the valve opening degree that should be output in the current sampling period. : In the formula, , This represents the difference between the target concentration and the estimated concentration. To determine the target undercurrent concentration, Here, k represents the estimated bottom flow concentration; k is the current discrete time step. This is the proportional gain coefficient. This is the integral gain coefficient. , These are the upper and lower limit constraint values ​​for the opening of the pinch valve. As a saturation function, the calculation results are restricted to... Within the range; Limiting the single-step change in the opening of the pinch valve ,in The preset step size threshold; Configure an underflow pressure auxiliary control loop to monitor underflow pressure. ,like Exceeding the preset feasible range If ], then the valve opening action of the main circuit is limited.

[0011] Optionally, the flocculant dosage can be dynamically calculated and the flocculant consumption per unit of solids removed can be compared on a rolling basis, including: The sensitivity coefficients of overflow turbidity, rake torque, and mud layer thickness to flocculant addition amount were estimated online, respectively. Dynamically calculate the adjustment value for flocculant dosage: In the formula, , , This represents the sensitivity coefficient of overflow turbidity, rake torque, and mud layer thickness to the amount of flocculant added. , , These represent the overflow turbidity setting, the rake frame torque setting, and the mud layer thickness setting, respectively. , , This indicates overflow turbidity, rake torque, and mud layer thickness. Set upper and lower limits and a rate of change constraint for the amount of flocculant added; The flocculant consumption per unit of solids removal is calculated on a rolling basis and compared with the historical best range. The current consumption value is maintained within the historical best range by adjusting the sensitivity coefficient or set value.

[0012] Optionally, the recommended number of ceramic filter units to be operated is calculated, and a hysteresis window strategy is used to control the switching of the number of filter units, including: Based on the predicted underflow solids flow rate and the processing capacity of a single filter, the recommended number of filters to operate is calculated: In the formula, The bottom flow rate, The processing capacity of a single ceramic filter. For safety factor; The hysteresis window console number switching is used, and the additional criteria are: The criteria for reducing the number of platforms are: In the formula, , They are respectively Time period and Average underflow solids flow over a period of time This represents the number of ceramic filter units currently in operation. To increase the hysteresis, To reduce the hysteresis during platform shutdown; When it is detected that the rake frame torque is too high, the mud layer thickness is increasing, the underflow pressure is too high, and the estimated underflow concentration is still near the upper limit of the target range, the filter should be turned on first. During the 2-3 minute buffer period after switching the number of filters, the change step size of the tightening clamp valve opening and the change in the amount of flocculant added are observed.

[0013] Optionally, built-in expert rules identify abnormal operating conditions and perform corresponding abnormal handling actions, including: The early signs of excessive mud layer and rake jamming are identified by the following conditions: the rake frame torque increases to near the alarm threshold, the mud layer thickness increases, the underflow pressure increases, and the overflow turbidity is normal. The corrective actions are to gradually increase the underflow valve opening to increase the underflow rate, suspend the slight increase in flocculant addition, and trigger pulse sludge discharge if necessary. In the case of overflow turbidity, the identification condition is that the overflow turbidity continuously exceeds the maximum value and exceeds the set time; the handling action is to gradually increase the amount of flocculant added and slightly close the valve to reduce the bottom flow rate. If it still exceeds the limit, the feed rate is reduced or the upstream linkage is requested. In cases of insufficient dilution or filtration, the identification conditions are: the estimated underflow concentration is lower than the lower limit of the target range, the underflow pressure drops, and the downstream filter is under high load. The corrective actions are to close the underflow valve, check the dosage and feed solids content, and if necessary, reduce the number of filters or reduce the frequency of filtration suction. In case of sensor malfunction, the identification conditions are: the deviation between the estimated online pipeline concentration and the underflow concentration exceeds the threshold and continues for a period of timeout, or the signal is lost; the action is to switch to soft measurement single source, activate slow valve mode, and prompt for verification.

[0014] Optionally, feedforward compensation can be triggered simultaneously based on changes in feed flow rate and feed concentration, including: The product of feed flow rate and feed concentration The step change generates a feedforward pulse for the bottom flow valve opening. Feedforward pulse for flocculant addition ,in , The empirical coefficients for online identification.

[0015] Optional, sensor health monitoring, comparing the undercurrent concentration estimate with the soft measurement estimate. With online pipeline concentration When the deviation between the two exceeds the set threshold and the duration exceeds the set time, the fusion weight of the corresponding sensor data is reduced and a calibration alarm is issued.

[0016] A second aspect of this application provides a dense undercurrent concentration control system based on multi-source parameter fusion, comprising: The data acquisition unit is used to acquire multi-source parameters in real time; The soft measurement and state estimation unit is used to build a soft measurement model for underflow concentration, predict the estimated value and uncertainty of underflow concentration, and perform sensor health monitoring. The underflow valve closed-loop control unit is used to perform closed-loop control with step size limitation and underflow pressure assistance, using the underflow concentration estimate as the controlled variable and the pinch valve opening as the control variable. The flocculant adaptive control unit is used to dynamically calculate the flocculant addition amount based on two indicators and perform rolling optimization of unit consumption. The filter co-control unit is used to calculate the recommended number of filters to be turned on, and to execute the hysteresis window switching and priority addition logic; Expert rules and feedforward compensation units are used to identify abnormal operating conditions and trigger corresponding handling actions, as well as to trigger feedforward compensation based on changes in feed. The system deployment and maintenance module is used for phase management of shadow mode, semi-closed-loop debugging, full-closed-loop debugging, and continuous optimization. In shadow mode, the soft measurement model and control logic are monitored but no control commands are output. The model calibration and initial parameter tuning are completed in 1-2 weeks. Semi-closed-loop debugging involves opening the underflow valve closed-loop control, limiting small step adjustments, and verifying the stability of concentration control. Full-closed-loop deployment involves enabling the normal adjustment step size, and implementing flocculant adaptive control, filter collaborative control, and abnormal operating condition handling logic. The continuous optimization phase involves periodically replaying the running data and iteratively updating the model parameters and control rules.

[0017] The technical solution provided in this application may include the following beneficial effects: By integrating multi-source data to construct a soft measurement model for underflow concentration, real-time accurate estimation and closed-loop feedback control of concentration are achieved, stabilizing the underflow concentration within the target range. A dual-index adaptive strategy based on overflow turbidity and rake torque or mud layer thickness dynamically optimizes the flocculant dosage, continuously maintaining the optimal consumption range for unit solids removal, effectively reducing reagent consumption and overflow turbidity. The system recommends the number of units to start and stop based on underflow solids flow rate and filter processing capacity, employing hysteresis window collaborative control. It also incorporates abnormal operating condition identification and expert handling rules, as well as a feed feedforward compensation mechanism, enhancing robustness to feed fluctuations and abnormal operating conditions. This comprehensively improves the automation and intelligence level of the entire mineral processing thickening and filtration process, enhancing the overall operating efficiency of the thickening process.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0019] The above and other objects, features and advantages of this application will become more apparent from the more detailed description of exemplary embodiments thereof in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments thereof.

[0020] Figure 1 This is a schematic flowchart illustrating the dense undercurrent concentration control method based on multi-source parameter fusion in an embodiment of this application. Detailed Implementation

[0021] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to make this application more thorough and complete, and to fully convey the scope of this application to those skilled in the art.

[0022] The technical solutions of the embodiments of this application are described in detail below with reference to the accompanying drawings.

[0023] In some implementations, see Figure 1 A method for controlling the concentration of dense undercurrent based on multi-source parameter fusion, comprising: S101. Collect multi-source parameters of the thickening process; the multi-source parameters include the rake frame torque, mud layer thickness, overflow turbidity of the thickener body, underflow pressure, underflow flow rate, online pipeline concentration of the underflow pipeline, feed flow rate and feed concentration on the feed side, and the number of downstream ceramic filters that have been turned on. Specifically, multi-source parameters are acquired in real time at a frequency of not less than 1 Hz, including: Thickener body: rake frame torque M, mud layer thickness Overflow turbidity ; Underflow pipe: Underflow pressure Bottom flow rate Online pipeline concentration ; Feed side: Feed flow rate Feed concentration ; Downstream: Number of ceramic filter units already in operation .

[0024] Perform data validity verification, check the rationality of each measurement value, mark the data as invalid if a value exceeds the physical limit, and trigger the sensor alarm after 3 consecutive invalid values.

[0025] Digital filtering employs a moving average filter on valid data, with a window length of 5 sampling points to suppress high-frequency noise. The filtered values ​​are stored in a circular buffer for subsequent use.

[0026] Write all raw values, filtered values, and timestamps into a real-time database, and retain the data from the last 30 days for continuous optimization.

[0027] S102. Based on multi-source parameters, establish a soft measurement model for underflow concentration to estimate the underflow concentration and the corresponding uncertainty. The soft measurement model adopts a combination of recursive partial least squares algorithm and Kalman filter, or a combination of autoregressive model with hysteresis term and Kalman filter. Specifically, multiple source parameters are fused to construct the input feature vector: (1) Perform soft measurement model calculations, providing two optional models: A recursive partial least squares algorithm was used to establish a regression model between the input features and the bottom flow concentration. A Kalman filter was then used to filter and reduce noise in the model output to determine the estimated bottom flow concentration and its confidence interval. Forward computation: ,in, This represents the estimated undercurrent concentration output by the RPLS model at the current time t. This is a vector of regression coefficients, which is recursively updated each time an offline test value is received. Using the RPLS output as the observation value, a Kalman filter is further applied to reduce noise and estimate uncertainty. The filtering recursion process is as follows: Kalman filter time update: , ,in For the prior state estimate of time k, The prior estimate of the error covariance for the current state. The process noise covariance reflects the uncertainty during the system's state transition process; Measurement Update: , , ;in, Kalman gain is used to balance the weights of predicted and observed values. To measure the noise covariance, which reflects the reliability of RPLS observations, For the posterior estimate of the state at the current moment, Let be the posterior estimation error covariance at the current time. Output underflow concentration estimate Uncertainty .

[0028] A combination of an autoregressive model with hysteresis and a Kalman filter was used to determine the estimated bottom current concentration and its confidence interval. A combination of an autoregressive model with hysteresis and a Kalman filter was used to determine the estimated bottom current concentration and its confidence interval. Autoregressive model with lag terms: (2) In the formula, These are the autoregressive coefficients, representing the weights of the linear contribution of past estimates to the current estimate. The input feature coefficients represent the linear contribution weights of each input feature to the current estimate. The j-th input feature is The value at time, Noise term; The Kalman filtering steps are as described above, and the output is... and uncertainty.

[0029] Sensor health monitoring compares the concentration estimates output by soft measurement. With online pipeline concentration detection value If the deviation between the two exceeds the preset threshold and the duration exceeds the set time, the fusion weight of the pipeline concentration meter will be reduced, for example, its weight will be set to zero or reduced, and a calibration alarm will be issued; if the pipeline concentration meter signal is lost, it will immediately switch to a soft measurement mode that does not use the signal.

[0030] Mud layer thickness trend estimation, based on the mass balance formula to estimate the rate of change in solid inventory: (3) In the formula, Let A be the density of the solid and A be the area of ​​the settling zone. If the inventory change rate is positive, that is If the mud layer rises and continues for a certain period of time, it is predicted that the mud layer will rise. When it is predicted that the mud layer will exceed the set upper limit in the near future, pre-dumping of mud will be triggered in advance to carry out feedforward control.

[0031] S103. Using the estimated underflow concentration as the controlled variable and the pinch valve opening as the manipulated variable, closed-loop control is executed according to the set target range of underflow concentration to adjust the pinch valve opening. The closed-loop control adopts a proportional-integral control law that resists integral saturation and limits the change step size of the pinch valve opening to suppress liquid-solid disturbance. Specifically, based on the current estimated bottom flow concentration... and target concentration Calculation error: (4) Closed-loop control is executed based on the set target range for underflow concentration to adjust the pinch valve opening. A proportional-integral control law with anti-integral saturation is used to obtain the pinch valve opening, i.e., the valve opening that should be output in the current sampling period. : (5) In the formula, This represents the difference between the target concentration and the estimated concentration. To determine the target undercurrent concentration, Here, k represents the estimated bottom flow concentration; k is the current discrete time step. This is the proportional gain coefficient. This is the integral gain coefficient. 、 These are the upper and lower limit constraint values ​​for the opening of the pinch valve. As a saturation function, the calculation results are restricted to... Within the range; When the previous output If the limit value has been reached and the current error direction is consistent with the saturation direction, then the accumulation of the integral term is stopped to prevent the integration depth from saturating.

[0032] Variation step size limit: To prevent excessively rapid valve action from causing liquid-solid disturbances, the single-step variation amount is limited. , ,in, The preset step size threshold; Bottom flow pressure-assisted limiting, monitoring bottom flow pressure Is it within the preset feasible range? ]Inside; like If the valve is closed further, the opening must not be reduced to ensure that the slurry in the pipeline has sufficient conveying pressure. like If the valve is not opened further, the opening degree must not be increased to prevent the risk of pipe bursting due to excessive pipeline pressure. when No additional restrictions are imposed when the interval is within the range.

[0033] The calculated Send to the pinch valve positioner and update historical values ​​for use in the next cycle.

[0034] S104. Based on the dual indicators of overflow turbidity and rake frame torque or mud layer thickness, dynamically calculate the amount of flocculant added, and continuously compare the flocculant consumption value per unit of solid removal, so as to maintain the flocculant consumption value in the historical optimal range. Specifically, it is executed on a cycle independent of valve control, achieving dynamic adjustment of flocculant addition based on two indicators: overflow turbidity and rake torque or mud layer thickness. Get the current overflow turbidity Rake frame torque M, mud layer thickness and their corresponding settings , , The set value is manually set according to the process requirements; Online sensitivity coefficient identification was performed, and recursive least squares (RLS) with a forgetting factor was used to estimate the overflow turbidity. Rake frame torque M, mud layer thickness flocculant addition amount Sensitivity coefficient , , Taking the overflow turbidity channel as an example, the identification model is as follows: (6) in, and Let represent the changes in turbidity and the changes in dosage at adjacent time points, respectively. The recursive least squares recursive formula with forgetting factor is as follows: , (7) (8) (9) In the formula, Forgetting factor; Rake frame torque M, mud layer thickness The identification methods are similar, and respectively obtain , ; Dynamically calculate the adjustment value for flocculant dosage: (10) Set upper and lower limits and a rate of change constraint for the amount of flocculant added: (11) In the formula, This represents the original calculated value of the unconstrained flocculant addition amount. This represents the output value of the flocculant addition amount in the previous control cycle; Apply amplitude limit: ,in, , These are the minimum and maximum allowable values ​​for the amount of flocculant to be added, respectively. Apply a rate of change constraint, if Then it is limited to Inside, This represents the maximum allowable change at each step.

[0035] Will Send to the frequency converter or flow regulating valve of the flocculant dosing pump.

[0036] Unit consumption rolling optimization calculates flocculant consumption per unit of solids removal using a fixed time window: (12) In the formula, This represents the flocculant consumption per unit of solids removed. From The time integral of the flocculant addition amount within the time window up to time t, i.e., the total amount of flocculant consumed during that period. The duration of the sliding window; Save the U values ​​from the most recent multiple windows, and take the 30th to 70th percentiles as the historical best interval. If the current U is higher than the upper limit of the interval, increase it proportionally. and This strengthens the regulatory effect of torque and mud layer thickness on chemical dosing, thereby reducing the dosage; if the current U is below the lower limit of the interval, the dosage is reduced proportionally. , To avoid insufficient medication leading to turbidity.

[0037] S105. Based on the underflow solids flow rate and the processing capacity of a single filter, calculate the recommended number of ceramic filters to be turned on, and use a hysteresis window strategy to control the switching of the number of filters. Specifically, it is executed on an independent cycle, coordinating the number of downstream ceramic filter units that are turned on; Calculate the average solids flow rate and read the bottom flow rate at the current moment from the database. and bottom flow concentration estimates Calculate the bottom solid flow rate ; Based on the underflow solids flow rate and the processing capacity of a single filter, calculate the theoretical number of filters required: (13) In the formula, The processing capacity of a single ceramic filter. For safety reasons, This represents the floor function, ensuring that the processing capacity is not less than the actual solid flow rate; To prevent frequent start-stop cycles, a hysteresis window strategy is used to control the switching of the number of filter units, reading the number of currently active ceramic filter units from the filter controller. ; Calculate the most recent first time period Average underflow solids flow If satisfied If the current number of units has not reached the maximum available number, then add one ceramic filter and start the buffer period timer. After adding the filter, skip the unit reduction judgment. for Average underflow solids flow over a period of time To increase the hysteresis; If the additional conditions are not met, then calculate the most recent second time period. Average underflow solids flow If satisfied If the current number of devices is greater than 1, then one device will be shut down, and a buffer period timer will be started; among which, for Average underflow solids flow over a period of time To reduce the hysteresis during platform shutdown; Prioritize starting up filters in case of anomalies. Even if the hysteresis threshold for starting up has not been reached, one filter should be started up first if the following abnormal signs are detected simultaneously: excessive rake torque (M); excessive mud layer thickness. The pressure rose above the preset level within the recently set time period; undercurrent pressure. Too high; estimated underflow concentration It is still near the upper limit of the target range; if all the above conditions are met and there are idle filters, immediately add one and record the abnormal priority addition event.

[0038] During the buffer period after the switchover, control is tightened temporarily to mitigate disturbances, including limiting the step size of the underflow valve opening. The dosage is temporarily reduced to a smaller proportion of the normal value; the change rate limit of the flocculant addition is temporarily reduced, and the original parameters are restored after the buffer period.

[0039] S106: Built-in expert rules identify abnormal operating conditions and perform corresponding abnormal handling actions, while also providing feedforward compensation based on changes in feed flow rate and feed concentration.

[0040] Specifically, expert rules are executed within each sampling period and have a higher priority than conventional closed-loop control. When abnormal operating conditions are detected, expert rules can directly modify control commands or switch control modes. Expert rules include: Excessive mud thickness is a precursor to rake jamming. The identification condition is that the rake frame torque M rises to near the rake frame torque alarm threshold. Mud layer thickness Continued rise, bottom pressure The turbidity rises, while the overflow turbidity increases. Normally, there is no runoff. The response is to initiate a segmented sludge discharge strategy, increasing the valve opening by a fixed small step in each control cycle, pausing the slight increase in flocculant addition, until the rake torque M stops rising. If the rake torque M continues to rise for several consecutive cycles, pulse sludge discharge is executed, briefly opening the valve to a larger opening before resuming operation, while simultaneously pausing flocculant addition.

[0041] Turbidity overflow, meaning the overflow turbidity exceeds the standard, is identified by the overflow turbidity. And the duration exceeds the set threshold; the treatment action is to increase the amount of flocculant added in small increments, while slightly closing the underflow valve to prolong the residence time of the slurry in the thickener; if the overflow turbidity after a period of time... If the feed rate is still higher than the target value, a request will be made to the upstream process to reduce the feed rate; if this is still ineffective, an alarm will be triggered to prompt manual intervention.

[0042] The identification conditions are: dilution, insufficient filter supply. Below the lower limit of the target range If the concentration drops below the threshold within a short period of time, and the downstream filter is under high load, the appropriate action is to close the underflow valve to increase the underflow concentration; at the same time, check the amount of flocculant added, and increase it appropriately if it is too low; if the concentration still does not recover after the set time, it is recommended that the operator reduce the number of filters in operation or reduce the filter speed.

[0043] Sensor anomaly, identification condition is online pipeline concentration Compared with soft measurement estimates If the deviation exceeds the threshold and the duration exceeds the set time, or Signal disconnection. The response is to automatically switch the control mode to use only soft measurement estimates, no longer relying on the pipeline concentration meter, temporarily tighten the underflow valve step limit, and issue a sensor calibration alarm.

[0044] In some implementations, corresponding to the aforementioned application function implementation method embodiments, this application also provides a dense undercurrent concentration control system based on multi-source parameter fusion and corresponding embodiments.

[0045] A multi-source parameter fusion-based thickening underflow concentration control system is deployed in the thickening section of a mineral processing plant, interacting with the field DCS distributed control system. It includes: a data acquisition unit, a soft measurement and state estimation unit, an underflow valve closed-loop control unit, a flocculant adaptive control unit, a filter co-operation control unit, an expert rule and feedforward compensation unit, and a system deployment and maintenance module. All units can be integrated into an industrial control computer or edge intelligent gateway, communicating with field instruments, PLCs, and the DCS via OPC UA or Modbus TCP protocols.

[0046] The data acquisition module is used to acquire observable measurements of the thickening process, with a sampling frequency of not less than 1Hz. Specifically, it acquires the rake frame torque, mud layer thickness / material level, and overflow turbidity of the thickener; the underflow pressure, underflow flow rate, and online pipeline concentration of the underflow pipeline; the feed flow rate and feed concentration on the feed side; and the number of downstream ceramic filters that have been turned on.

[0047] The hierarchical control module adopts a hierarchical control structure that combines soft-sensor concentration-underflow valve loop, flocculant dual-index adaptive control, and filter unit number coordination, including: The basic protection layer is used to implement rake frame torque protection, overflow turbidity protection, and equipment / signal health anomaly handling logic: When the rake frame torque is greater than or equal to the alarm threshold, the upper limit of chemical dosing is limited and the underflow valve is slowly opened to increase the underflow rate, and a segmented sludge discharge strategy is triggered if necessary; When the rake frame torque is greater than or equal to the trip threshold, the valve is forcibly opened to a safe opening, chemical dosing is suspended, and an alarm is triggered for manual inspection; When the overflow turbidity is greater than the maximum overflow turbidity, the chemical dosage is slightly increased in a short period of time and the underflow valve is slightly closed to prolong the material residence time. If the limit is continuously exceeded, the feed rate is reduced and the process is investigated; The backoff logic is triggered by equipment anomalies such as sensor disconnection, valve jamming, and empty dosing tank.

[0048] In the closed-loop main control layer, the undercurrent concentration obtained through soft-sensing / online fusion is used as the controlled variable, and the pinch valve opening is used as the manipulated variable. The target concentration is set to 75%. A proportional-integral control law with anti-integral saturation is adopted to obtain the pinch valve opening, while limiting the step size of the opening change. Suppress liquid-solid disturbances. The auxiliary circuit aims to stabilize the underflow pressure and sets [a certain value] for the underflow pressure. [Feasible range] When the range is exceeded, the valve opening action of the main circuit is limited.

[0049] The optimized coordination layer includes a flocculant adaptive adjustment unit and a filter unit coordination unit: flocculant adjustment is based on dual-index adaptive control of turbidity, torque, and sludge layer thickness, and the sensitivity coefficient of each parameter to the dosage is identified online. , , The dosage is dynamically adjusted, and the dosage for solid removal per unit volume is compared online to maintain it within the historical optimal range. The number of filters is coordinated based on the current / predicted underflow solids flow rate and the processing capacity of a single filter. The number of filters is switched using a hysteresis window to avoid frequent start-stop. When the rake torque is too high, the mud layer thickness increases, the underflow pressure is too high, and the estimated underflow concentration is still near the upper limit, filters are added first to avoid material accumulation in the thickener.

[0050] The soft measurement and state estimation unit is used to achieve soft measurement of underflow concentration and estimation of mud layer thickness: the soft measurement of underflow concentration integrates input features such as underflow pressure, online pipeline concentration, and rake torque, and uses a soft measurement model with a combination of recursive partial least squares algorithm and Kalman filter, or a combination of autoregressive model with hysteresis term and Kalman filter to output the estimated flow concentration and uncertainty. When the deviation between the soft measurement concentration and the online pipeline concentration exceeds the threshold and continues to time out, the corresponding sensor weight is reduced and an alarm calibration is triggered; the mud layer thickness estimation predicts the mud layer rising trend through a simple mass balance formula and triggers valve position and chemical dosing pre-action in advance.

[0051] The expert rule module is used to identify abnormal operating conditions and trigger corresponding actions. It runs in parallel with the closed-loop / optimization layer. It only tightens the feasible region, adjusts the weights, or issues a one-time action when an abnormality is triggered, and then returns the closed-loop maintenance.

[0052] The system deployment and maintenance module is used for phase management of shadow mode, semi-closed-loop debugging, full-closed-loop debugging, and continuous optimization. In shadow mode, the soft measurement model and control logic are monitored but no control commands are output. The model calibration and initial parameter tuning are completed in 1-2 weeks. Semi-closed-loop debugging involves opening the underflow valve closed-loop control, limiting small step adjustments, and verifying the stability of concentration control. Full-closed-loop deployment involves enabling the normal adjustment step size, and implementing flocculant adaptive control, filter collaborative control, and abnormal operating condition handling logic. The continuous optimization phase involves periodically replaying the running data and iteratively updating the model parameters and control rules.

[0053] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for controlling the concentration of dense undercurrent based on multi-source parameter fusion, characterized in that, Collect multi-source parameters of the thickening process, including the rake frame torque, mud layer thickness, overflow turbidity of the thickener body, underflow pressure, underflow flow rate, online pipeline concentration of the underflow pipe, feed flow rate and feed concentration on the feed side, and the number of downstream ceramic filters that are in operation. Based on the aforementioned multi-source parameters, a soft measurement model for undercurrent concentration is established to estimate the undercurrent concentration and the corresponding uncertainty. The soft measurement model employs a combination of recursive partial least squares algorithm and Kalman filter, or a combination of autoregressive model with hysteresis term and Kalman filter. The underflow concentration estimate is used as the controlled variable, and the pinch valve opening is used as the manipulated variable. Closed-loop control is performed according to the set target range of underflow concentration to adjust the pinch valve opening. The closed-loop control adopts a proportional-integral control law with anti-integral saturation and limits the change step size of the pinch valve opening to suppress liquid-solid disturbance. Based on the dual indicators of overflow turbidity and rake frame torque or mud layer thickness, the amount of flocculant added is dynamically calculated, and the flocculant consumption value per unit of solids removal is continuously compared to maintain the flocculant consumption value within the historical optimal range. Based on the underflow solids flow rate and the processing capacity of a single filter, the recommended number of ceramic filters to be turned on is calculated, and a hysteresis window strategy is adopted to control the switching of the number of filters. Built-in expert rules identify abnormal operating conditions and take corresponding abnormal handling actions, while also performing feedforward compensation based on changes in feed flow rate and feed concentration.

2. The method for controlling the concentration of dense undercurrent based on multi-source parameter fusion according to claim 1, characterized in that, The establishment of a soft-sensor model for underflow concentration, and the estimation of underflow concentration and corresponding uncertainty, includes: It integrates underflow pressure, online pipeline concentration, rake torque, mud layer thickness, feed flow rate, and feed concentration; A regression model between input features and bottom flow concentration was established using a recursive partial least squares algorithm. The Kalman filter was then used to filter and reduce noise in the model output to determine the estimated bottom flow concentration and the confidence interval. Alternatively, a combination of an autoregressive model with hysteresis terms and a Kalman filter can be used: In the formula, This is the estimated underflow concentration at the current time t. These are the autoregressive coefficients, representing the weights of the linear contribution of past estimates to the current estimate. The input feature coefficients represent the linear contribution weights of each input feature to the current estimate. The j-th input feature is The value at time, This is the noise term.

3. The method for controlling the concentration of dense undercurrent based on multi-source parameter fusion according to claim 1, characterized in that, The step of performing closed-loop control to adjust the clamp valve opening based on a set target range for underflow concentration includes: Set the mass fraction of the target range for underflow concentration; By employing a proportional-integral control law that resists integral saturation, the opening degree of the pinch valve is obtained, which is the valve opening degree that should be output in the current sampling period. : In the formula, , This represents the difference between the target concentration and the estimated concentration. To determine the target undercurrent concentration, Here, k represents the estimated bottom flow concentration; k is the current discrete time step. This is the proportional gain coefficient. This is the integral gain coefficient. , These are the upper and lower limit constraint values ​​for the opening of the pinch valve. As a saturation function, the calculation results are restricted to... Within the range; Limiting the single-step change in the opening of the pinch valve ,in The preset step size threshold; Configure an underflow pressure auxiliary control loop to monitor underflow pressure. ,like Exceeding the preset feasible range If ], then the valve opening action of the main circuit is limited.

4. The method for controlling the concentration of dense undercurrent based on multi-source parameter fusion according to claim 1, characterized in that, The flocculant dosage is dynamically calculated, and the flocculant consumption per unit of solids removal is compared on a rolling basis, including: The sensitivity coefficients of the overflow turbidity, the rake frame torque, and the mud layer thickness to the amount of flocculant added are estimated online, respectively. Dynamically calculate the adjustment value for flocculant dosage: In the formula, , , This represents the sensitivity coefficient of overflow turbidity, rake torque, and mud layer thickness to the amount of flocculant added. , , These represent the overflow turbidity setting, the rake frame torque setting, and the mud layer thickness setting, respectively. , , This indicates overflow turbidity, rake torque, and mud layer thickness. Set upper and lower limits and a rate of change constraint for the amount of flocculant added; The flocculant consumption value per unit of solids removal is calculated on a rolling basis and compared with the historical best range. The current consumption value is maintained within the historical best range by adjusting the sensitivity coefficient or the set value.

5. The method for controlling the concentration of dense undercurrent based on multi-source parameter fusion according to claim 1, characterized in that, The calculation of the recommended number of ceramic filter units to be operated, and the use of a hysteresis window strategy to control the switching of the number of filter units, includes: Based on the predicted underflow solids flow rate and the processing capacity of a single filter, the recommended number of filters to operate is calculated: In the formula, The bottom flow rate, The processing capacity of a single ceramic filter. For safety factor; The hysteresis window console number switching is used, and the additional criteria are: The criteria for reducing the number of platforms are: In the formula, , They are respectively Time period and Average underflow solids flow over a period of time This represents the number of ceramic filter units currently in operation. To increase the hysteresis, To reduce the hysteresis during platform shutdown; When it is detected that the rake frame torque is too high, the mud layer thickness is increasing, the underflow pressure is too high, and the estimated underflow concentration is still near the upper limit of the target range, the filter should be turned on first. During the minute buffer period after switching the number of filters, the change step size of the tightening clamp valve opening and the change in the amount of flocculant added are recorded.

6. The method for controlling the concentration of dense undercurrent based on multi-source parameter fusion according to claim 1, characterized in that, The built-in expert rules identify abnormal operating conditions and perform corresponding abnormal handling actions, including: The conditions for identifying excessive mud thickness and the precursor to rake jamming are: the rake frame torque rises to the alarm threshold, the mud thickness increases, the underflow pressure increases, and the overflow turbidity is normal. The corrective actions are to gradually increase the underflow valve opening to increase the underflow rate, suspend the slight increase in flocculant addition, or trigger pulse sludge discharge. In the case of overflow turbidity, the identification condition is that the overflow turbidity continuously exceeds the maximum value and exceeds the set time; the handling action is to gradually increase the amount of flocculant added and slightly close the valve to reduce the bottom flow rate. If it still exceeds the limit, the feed rate is reduced or the upstream linkage is requested. In cases of insufficient dilution or filtration, the identification conditions are: the estimated underflow concentration is lower than the lower limit of the target range, the underflow pressure drops, and the downstream filter is under high load. The corrective actions are to close the underflow valve, check the dosage and feed solids content, and if necessary, reduce the number of filters or reduce the frequency of filtration suction. In case of sensor malfunction, the identification conditions are: the deviation between the estimated online pipeline concentration and the underflow concentration exceeds the threshold and continues for a period of timeout, or the signal is lost; the action is to switch to soft measurement single source, activate slow valve mode, and prompt for verification.

7. The method for controlling the concentration of dense undercurrent based on multi-source parameter fusion according to claim 1, characterized in that, The simultaneous triggering of feedforward compensation based on changes in feed flow rate and feed concentration includes: The product of feed flow rate and feed concentration The step change generates a feedforward pulse for the bottom flow valve opening. Feedforward pulse for flocculant addition ,in , The empirical coefficients for online identification.

8. The method for controlling the concentration of dense undercurrent based on multi-source parameter fusion according to claim 1, characterized in that, Also includes: Sensor health monitoring compares the estimated undercurrent concentration values ​​obtained from soft measurement. With the online pipeline concentration When the deviation between the two exceeds the set threshold and the duration exceeds the set time, the fusion weight of the corresponding sensor data is reduced and a calibration alarm is issued.

9. A dense undercurrent concentration control system based on multi-source parameter fusion, used to execute the dense undercurrent concentration control method based on multi-source parameter fusion as described in any one of claims 1-8, characterized in that, include: The data acquisition unit is used to acquire multi-source parameters in real time; The soft measurement and state estimation unit is used to build a soft measurement model for underflow concentration, predict the estimated value and uncertainty of underflow concentration, and perform sensor health monitoring. The underflow valve closed-loop control unit is used to perform closed-loop control with step size limitation and underflow pressure assistance, using the underflow concentration estimate as the controlled variable and the pinch valve opening as the control variable. The flocculant adaptive control unit is used to dynamically calculate the flocculant addition amount based on two indicators and perform rolling optimization of unit consumption. The filter co-control unit is used to calculate the recommended number of filters to be turned on, and to execute the hysteresis window switching and priority addition logic; Expert rules and feedforward compensation units are used to identify abnormal operating conditions and trigger corresponding handling actions, as well as to trigger feedforward compensation based on changes in feed. The system deployment and maintenance module is used for phase management of shadow mode, semi-closed-loop debugging, full-closed-loop debugging, and continuous optimization. The shadow mode is where the soft measurement model and control logic only monitor without outputting control commands, and the model calibration and initial parameter tuning are completed in 1-2 weeks. The semi-closed-loop debugging involves opening the underflow valve closed-loop control, limiting small step adjustments, and verifying the stability of concentration control. The full-closed-loop deployment involves enabling the normal adjustment step size, and implementing flocculant adaptive control, filter collaborative control, and abnormal operating condition handling logic. The continuous optimization phase involves periodically replaying the running data and iteratively updating the model parameters and control rules.

Images