Tailings thickener underflow concentration automatic detection and control system and method

Abstract

The present application belongs to the technical field of tailings thickening process control, and discloses a tailings thickening machine underflow concentration automatic detection and control system and method; through collecting multi-point turbidity data in the radial direction of the overflow area of the feedwell, early identification of the disturbance of the feed properties is realized; a layered settling model is constructed based on the mud layer interface height and the thickening machine geometric parameters, and the material residence time is accurately calculated; a medicament effect propagation descriptor is created, and the dynamic propagation process of the flocculating agent in the system is tracked; combined with the in-transit cumulative effect evaluation and future concentration prediction, feedforward-feedback composite control is realized; through lagging prediction error analysis, the model parameters are dynamically corrected, and the system performance is continuously optimized. The present method realizes predictive control and accurate adjustment of the underflow concentration, effectively solves the problem of excessive flocculating agent adjustment, and improves the stability and reliability of the tailings treatment process.

Description

Technical Field

[0001] This invention relates to the field of tailings thickening process control technology, and more specifically, to an automatic detection and control system and method for underflow concentration of tailings thickener. Background Technology

[0002] Existing tailings thickener control technology faces the challenge of time-delay control, fundamentally due to the significant lag effect caused by the spatial distance between the mixing point and the reaction point. From the point of addition to the flocculant, to its full mixing with the slurry, to the formation of effective flocs and sedimentation, and finally reflected in the underflow concentration, there is a considerable pure lag time. This inherent long delay in the process traps the control system in a "blind zone operation" predicament, particularly prominent in actual production environments where ore properties fluctuate frequently. When the ore particle size suddenly becomes finer or the solids content fluctuates, the underflow concentration begins to gradually decrease. The control system then increases the flocculant dosage to compensate, but the effect of this adjustment requires a long lag period before it is reflected in the underflow parameters. During this "effect propagation vacuum period," the control system cannot obtain real-time feedback on the adjustment, yet it still makes more aggressive compensations based on the continuously decreasing underflow concentration, leading to excessive flocculant addition. When the early addition effect finally reaches the underflow, the superimposed flocculation effect often causes the underflow concentration to rise sharply to excessively high levels, resulting in blockage of the thickener's bottom discharge system pipes, excessive load on the scraper frame, and even the risk of shutdown. This vicious cycle of "under-adjustment-over-fluctuation-readjustment" frequently occurs in actual production, not only wasting a large amount of flocculant and increasing operating costs, but also seriously affecting the stability of subsequent dewatering processes. Even more problematic is that traditional control systems cannot distinguish the cumulative effects of adjustments made at different times, nor do they possess the ability to anticipate changes in feed properties. This often forces operators to make painful trade-offs between "stability" and "response speed," choosing conservative parameters to avoid system collapse, but at the cost of a significant decrease in control accuracy.

[0003] In view of this, the present invention proposes an automatic detection and control system and method for the underflow concentration of tailings thickener to solve the above problems. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an automatic detection and control method for underflow concentration in tailings thickeners, comprising:

[0005] The real-time operating parameters of the tailings thickener are obtained, including the solid content of the feed slurry, feed flow rate, flocculant dosing rate, mud layer interface height, rake torque, and underflow concentration measurement.

[0006] Multiple turbidity data points were collected radially in the overflow zone of the feed well. The turbidity gradient change rate was calculated on the turbidity data within the sliding time window. Feed property disturbance events were identified and disturbance feature vectors were generated.

[0007] Based on the mud layer interface height and the geometric parameters of the thickener, the thickener settling zone is divided into multiple settling sub-layers along the vertical direction, and the material residence time of each settling sub-layer is calculated in combination with the solid flux balance relationship.

[0008] For each addition event where the change in the flocculant addition rate exceeds a preset minimum increment threshold, a flocculant effect propagation descriptor is created. The flocculant effect propagation descriptor includes the addition increment value, injection timestamp, current sedimentation sublayer index, and remaining effect amount.

[0009] Based on the material residence time of each settling sublayer, the position migration of all active agent effect propagation descriptors between adjacent settling sublayers is promoted in each control cycle. The residual effect of all active descriptors that have not yet reached the underflow outlet is accumulated and recorded as the in-transit cumulative effect value.

[0010] When the disturbance feature vector triggers the dosage adjustment request, the feedforward compensation amount is calculated, the in-transit cumulative effect value is subtracted from the feedforward compensation amount, and a saturation limit constraint is applied to obtain the bounded net adjustment amount and execute the corresponding flocculant dosing operation.

[0011] Based on the remaining propagation time of all active descriptors, a future undercurrent concentration prediction trajectory is constructed. When there is a time node in the future undercurrent concentration prediction trajectory that exceeds the preset concentration limit, an early reduction signal is sent to the current control cycle.

[0012] The measured value of the underflow concentration at the current moment is compared with the predicted value generated at the corresponding historical moment to calculate the lag prediction error. Based on the lag prediction error, the material residence time and reagent effect decay parameters of each settling sublayer are dynamically corrected.

[0013] An automatic detection and control system for underflow concentration in a tailings thickener, comprising a method for automatically detecting and controlling the underflow concentration in a tailings thickener, including:

[0014] The data acquisition module is used to acquire the real-time operating parameters of the tailings thickener, including the solid content of the feed slurry, feed flow rate, flocculant dosing rate, mud layer interface height, rake torque, and underflow concentration measurement.

[0015] The disturbance identification module is used to collect turbidity data at multiple points along the radial direction in the overflow area of ​​the feed well, calculate the turbidity gradient change rate of the turbidity data within the sliding time window, identify feed property disturbance events, and generate disturbance feature vectors.

[0016] The sedimentation stratification module is used to divide the sedimentation zone of the thickener vertically into multiple sedimentation sub-layers based on the mud layer interface height and the geometric parameters of the thickener, and to calculate the material residence time of each sedimentation sub-layer in combination with the solid flux balance relationship.

[0017] The effect tracking module is used to create a descriptor for the propagation of flocculant effects for each dosing event where the change in the flocculant dosing rate exceeds a preset minimum increment threshold.

[0018] The effect propagation update module is used to advance the position migration of all active agent effect propagation descriptors between adjacent sedimentation sub-layers in a control cycle according to the material residence time of each sedimentation sub-layer, and to accumulate the remaining effect of all active descriptors that have not yet reached the underflow outlet, which is recorded as the in-transit cumulative effect value.

[0019] The control decision module is used to calculate the feedforward compensation amount when the disturbance feature vector triggers the dosage adjustment request, subtract the in-transit cumulative effect value from the feedforward compensation amount, apply the saturation limit constraint, obtain the bounded net adjustment amount, and execute the corresponding flocculant dosing operation.

[0020] The prediction and early warning module is used to construct a future undercurrent concentration prediction trajectory based on the remaining propagation time of all active descriptors. When there is a time node in the future undercurrent concentration prediction trajectory that exceeds the preset concentration limit, an early reduction signal is sent to the current control cycle.

[0021] The model self-calibration module is used to compare the measured value of the underflow concentration at the current moment with the predicted value generated at the corresponding historical moment, calculate the lag prediction error, and dynamically correct the material residence time and reagent effect decay parameters of each settling sublayer based on the lag prediction error.

[0022] The modules are connected via wired and / or wireless means to enable data transmission between them.

[0023] The technical effects and advantages of the automatic detection and control system and method for tailings thickener underflow concentration of this invention are as follows:

[0024] This invention achieves a paradigm shift from passive response to proactive prediction in control. It overcomes the limitation of traditional control systems that cannot accurately track the propagation of flocculant effects, establishing a complete digital mapping of the settling process, enabling the control system to "see" the dynamic transmission of the flocculant effect within the thickener. Through early identification and characteristic analysis of feed disturbances, this invention can implement precise intervention before the disturbance affects the underflow, effectively suppressing concentration fluctuations. A prediction mechanism based on the fusion of physical models and real-time data enables the system to accurately predict future concentration trends, allowing for preventative adjustments before potential risks arise, avoiding equipment overload and pipeline blockage accidents caused by excessively high underflow concentrations. A closed-loop self-calibration mechanism endows the system with the ability to continuously learn and optimize, enabling model parameters to adaptively adjust to match constantly changing process conditions and slurry characteristics, ensuring long-term control accuracy. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the automatic detection and control method for underflow concentration of tailings thickener according to the present invention;

[0026] Figure 2 This is a schematic diagram of the automatic detection and control system for underflow concentration of tailings thickener of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] This application provides an automatic detection and control system and method for underflow concentration in a tailings thickener. The system is applicable to intelligent control of underflow concentration in the tailings thickening process, enabling accurate prediction and feedforward-feedback composite control of the underflow concentration. The system's execution entities include, but are not limited to, the following: thickener control platform, slurry treatment automation platform, tailings treatment process system, and intelligent tailings concentration control unit, which can be considered as general control nodes in this application.

[0029] Please see Figure 1 In this embodiment of the invention, the specific implementation process of the automatic detection and control method for underflow concentration of tailings thickener includes:

[0030] Real-time operating parameters of the tailings thickener are acquired, including the solids content of the feed slurry, feed flow rate, flocculant dosing rate, mud interface height, rake torque, and underflow concentration. These real-time operating parameters are the fundamental input data for the control system and are acquired in real time through a multi-channel data acquisition interface. The solids content of the feed slurry reflects the characteristics of the incoming material; the feed flow rate characterizes the material entry rate; the flocculant dosing rate records the usage of the flocculant aid; the mud interface height reflects the solid-liquid separation state in the settling zone; the rake torque indicates the thickness of the bottom mud layer; and the underflow concentration directly reflects the actual state of the control target. These data provide comprehensive process information for subsequent analysis, ensuring the accuracy and timeliness of control decisions.

[0031] Multiple turbidity data points are collected radially in the overflow zone of the feed well. The turbidity gradient change rate is calculated within a sliding time window to identify feed property disturbances and generate disturbance feature vectors. Feed characteristic monitoring is a crucial step in achieving feedforward control, enabling early detection of feed changes through turbidity gradient analysis. The monitoring process involves setting up multiple turbidity sensors at equal radial intervals in the overflow zone of the thickener feed well to collect turbidity data sequences in real time. The system applies sliding window analysis to the collected turbidity data, calculates the first-order difference mean, and constructs a radial gradient distribution. When an abnormal pattern appears in the gradient distribution, the system identifies it as a disturbance in the feed properties and distinguishes the disturbance type based on gradient characteristics. The disturbance feature vector contains a disturbance type identifier and gradient peak information, providing feedforward information for subsequent control adjustments.

[0032] Based on the mud interface height and thickener geometry, the thickener settling zone is vertically divided into multiple settling sub-layers. The material residence time of each sub-layer is calculated using the solids flux balance relationship. Stratified settling zone modeling is a fundamental method for accurately describing material transport dynamics, improving model accuracy through layered calculations. The modeling process first determines the effective settling zone height based on the measured mud interface height. Then, it is divided into multiple settling sub-layers using a geometric progression, with thinner sub-layers near the interface and thicker sub-layers near the bottom to accommodate concentration gradient distribution characteristics. For each sub-layer, the system calculates the flux value based on the thickener cross-sectional area and solids settling velocity corresponding to the layer's center height, and derives the material residence time based on the sub-layer's geometric volume. The accumulated residence times constitute the total propagation lag time from the mud interface to the bottom outlet, providing a time reference for the analysis of reagent effect propagation.

[0033] For each flocculant dosing event where the change in flocculant dosing rate exceeds a preset minimum increment threshold, a flocculant effect propagation descriptor is created. This descriptor includes the dosing increment, injection timestamp, current settling sublayer index, and remaining effect quantity. The flocculant effect propagation descriptor is the core data structure for tracking the dynamic propagation of flocculant effects, enabling accurate modeling of the flocculant effect. The creation process first checks whether the change in flocculant dosing rate exceeds the threshold; for valid changes, a descriptor object is created. During descriptor initialization, the dosing increment is calculated, the initial effective flocculant concentration is assessed based on the current feed conditions, and the flocculation efficiency coefficient is adjusted in conjunction with perturbation characteristics to determine the initial value of the remaining effect quantity. The current sublayer index of the descriptor is set to the sequence number of the topmost settling sublayer, and the creation time is recorded as the injection timestamp. This structured effect description method enables the system to accurately track the dynamic impact of each flocculant adjustment.

[0034] Based on the material residence time of each settling sub-layer, the system advances the migration of all active descriptors of the agent's effect propagation between adjacent settling sub-layers in each control cycle. The remaining effect of all active descriptors that have not yet reached the underflow outlet is accumulated and recorded as the in-transit cumulative effect value. Effect propagation is a dynamic calculation step simulating the transmission of agent effects within the thickener, achieving spatiotemporal tracking of the effect. In each control cycle, the system accumulates the residence time of each descriptor in the current sub-layer. When the residence time of that layer is reached, the descriptor is migrated to the next layer, and the effect retention coefficient of that layer is applied to adjust the remaining effect, simulating the attenuation of flocculant effectiveness during transport. When a descriptor reaches the bottom layer and completes its residence in that layer, it is removed from the active list. The system accumulates the remaining effect of all active descriptors in real time, calculating the in-transit cumulative effect value, which reflects the total effect of the agent that has been added but has not yet reached the underflow, providing an important reference for adjusting the dosage.

[0035] When a disturbance characteristic vector triggers a flocculant dosage adjustment request, a feedforward compensation is calculated. This feedforward compensation is then subtracted from the accumulated effect value before applying a saturation limit constraint to obtain a bounded net adjustment, and the corresponding flocculant dosing operation is executed. Dosage adjustment is a core step in control decision-making, comprehensively considering both feedforward compensation and the accumulated effect to ensure the rationality of control intervention. The adjustment process first calculates the feedforward compensation based on the disturbance characteristic vector, determining the ideal adjustment based on historical empirical relationships with similar disturbances. Then, the accumulated effect value is subtracted to avoid redundant adjustments. Finally, a saturation limit constraint is set based on the current rake torque state to ensure that the adjustment does not lead to excessive underflow concentration and equipment overload. The bounded net adjustment is directly used to execute the flocculant dosing operation, achieving precise response and preventative control to disturbances.

[0036] Based on the remaining propagation time of all active descriptors, a predicted trajectory for future underflow concentration is constructed. When any time point in the predicted trajectory exceeds the preset concentration limit, an early reduction signal is sent to the current control cycle. Concentration prediction trajectory construction is a key technology for achieving predictive control, allowing for early intervention against potential risks through future state prediction. The prediction process first calculates the remaining propagation time of each active descriptor from its current position to the underflow, then projects the effect size of each descriptor onto the corresponding node on the future time axis, accumulating the prediction increments at the same time step to ultimately generate a complete predicted trajectory for future underflow concentration. The system performs threshold detection on the predicted trajectory. When it detects that the concentration may exceed the limit at a future time, it immediately triggers the early reduction mechanism to achieve proactive risk intervention, preventing excessive underflow concentration from causing excessive equipment load or pipeline blockage.

[0037] The system compares the current undercurrent concentration measurement with the corresponding historical prediction value to calculate the lag prediction error. Based on this lag prediction error, it dynamically corrects the material residence time and reagent effect decay parameters for each settling sublayer. Prediction error correction is the core mechanism of the system's self-learning optimization, continuously improving model accuracy through closed-loop feedback. The correction process first retrieves historical prediction data to find the historical prediction value corresponding to the current moment; then, it calculates the difference between the current and measured values ​​to obtain the lag prediction error; and finally, it performs an exponentially weighted average of the errors over multiple consecutive periods to obtain the smoothed error trend. Based on the sign of the error trend, the system determines the direction of deviation between the actual propagation speed of the reagent effect and the predicted value, and adjusts the material residence time and effect decay parameters for each settling sublayer accordingly. This closed-loop correction mechanism enables the system to adapt to environmental changes and equipment characteristics, continuously improving prediction and control accuracy.

[0038] In this embodiment of the invention, the detailed implementation steps of collecting turbidity data at multiple points along the radial direction in the overflow zone of the feed well, calculating the turbidity gradient change rate of the turbidity data within a sliding time window, identifying feed property disturbance events, and generating disturbance feature vectors include:

[0039] Multiple turbidity sampling points are set at equal intervals radially along the overflow weir of the feed well, and turbidity value sequences of each sampling point are collected synchronously. A reasonable layout of the turbidity sampling network is the foundation for accurately capturing disturbances, achieving spatial resolution through multi-point distribution. The layout process involves setting 5-8 equally spaced turbidity sensors in the radial direction of the overflow weir of the feed well, covering the complete radial range from the inside to the outside. The acquisition system employs a high-precision turbidimeter and synchronous sampling technology to ensure time alignment of data at each point, with a typical sampling frequency of 1-10Hz. The turbidity value sequences are stored with timestamp indexes, forming a spatiotemporal two-dimensional data matrix, providing raw data support for subsequent gradient analysis. This multi-point layout strategy significantly improves the system's spatial resolution capability for disturbances, enabling precise differentiation between local disturbances and global changes.

[0040] Within a preset sliding time window, the first-order difference mean is calculated for the turbidity value sequence at each turbidity sampling point. Time-difference analysis is an effective method for capturing trends, highlighting the rate of change through first-order differencing. The analysis process sets an appropriate sliding time window (typically 30-120 seconds), calculates the difference between consecutive samples in the turbidity sequence at each sampling point within the window, and calculates the mean of these differences to obtain the turbidity change rate at that point. The formula for calculating the first-order difference mean is:

[0041] ;

[0042] in, Sampling points The first difference mean, Sampling points Within the window The turbidity value at that moment. This represents the number of samples within the window. The difference mean directly reflects the rate of change in turbidity; a positive value indicates an increase in turbidity, a negative value indicates a decrease in turbidity, and the absolute value reflects the degree of drastic change.

[0043] The first-order difference mean values ​​of each turbidity sampling point are arranged radially, and the differential gradient between adjacent sampling points is calculated. Spatial gradient analysis is a crucial step in detecting distribution anomalies, revealing changes in the radial distribution pattern. The analysis process involves arranging the first-order difference mean values ​​of each sampling point in radial order to form a radial distribution sequence; then, the differential gradient between adjacent sampling points is calculated to quantify the rate of change in the radial direction. The formula for calculating the differential gradient is:

[0044] ;

[0045] in, For the first and The differential gradient between sampling points, and The first-order difference mean of two adjacent points. The radial distance between the two points is denoted as . This second-order difference essentially analyzes the spatial variation of the turbidity rate of change, and can effectively detect anomalous patterns in the radial distribution.

[0046] When the gradient values ​​at multiple consecutive sampling points in the differential gradient exceed a preset gradient threshold, a feed property disturbance event is determined to have occurred. Disturbance event determination is a decision-making step in turbidity analysis, identifying significant changes through gradient threshold detection. The determination process sets an appropriate gradient threshold (usually determined based on the 3σ principle of historical data statistical distribution). When the gradient values ​​at multiple consecutive sampling points (usually ≥3) simultaneously exceed the threshold, the system determines that a feed property disturbance event has occurred. The requirement of exceeding the threshold at multiple consecutive points ensures the reliability of the determination and eliminates the possibility of single-point noise interference. The determination is implemented using a state machine method. When a valid disturbance is detected, the system enters the disturbance response state, initiating the feature analysis and control adjustment process.

[0047] Based on the spatial distribution morphology of the differential gradient, perturbations of decreasing particle size and abrupt changes in solid concentration are distinguished, and the perturbation type identifier is combined with the gradient peak value to form a perturbation feature vector. Perturbation type identification is the foundation of precise control, distinguishing different property changes through distribution patterns. The identification process analyzes the spatial distribution morphology of the differential gradient and extracts key pattern parameters. When the gradient distribution exhibits a single-peak pattern that gradually decreases from the inside out, it is identified as a decreasing particle size perturbation; this type of perturbation typically shows the most significant change in turbidity on the inner side. When the gradient distribution changes uniformly across the entire radial range or exhibits a multi-peak structure, it is identified as an abrupt change in solid concentration perturbation. The perturbation type identifier (usually 0 for particle size change and 1 for concentration change) and the gradient peak value (taking the absolute value of the maximum gradient) are combined to form the perturbation feature vector, providing qualitative and quantitative composite information for subsequent control. This perturbation classification method based on gradient distribution morphology significantly improves the system's accuracy in identifying different types of perturbations and the specificity of its response.

[0048] In this embodiment of the invention, the detailed implementation steps for dividing the settling zone of the thickener vertically into multiple settling sub-layers and calculating the material residence time of each settling sub-layer in conjunction with the solid flux balance relationship include:

[0049] The effective vertical height of the settling zone is determined by using the mud layer interface height as the upper boundary and the bottom conical outlet of the thickener as the lower boundary. Determining the effective settling zone is the first step in layered modeling, defining the calculation range through boundary definition. The process involves real-time reading of mud layer interface height sensor data, using this as the upper boundary of the settling zone; the bottom conical outlet of the thickener is used as the lower boundary; the vertical distance between these two boundaries is the effective settling zone height. This height represents the area actually involved in the settling process and is the basis for subsequent layered calculations. The system applies low-pass filtering to the interface height data to eliminate the impact of short-term fluctuations and ensure the stability of the division. The accurate definition of the effective settling zone directly affects the accuracy of subsequent residence time calculations and is a key input parameter of the layered model.

[0050] The effective vertical height is divided into N settlement sub-layers in a geometric progression, where the thickness of the settlement sub-layer near the mud interface is less than that near the bottom cone. The non-uniform stratification strategy is an optimization method adapted to the concentration gradient characteristics, improving model accuracy through variable thickness stratification. The division process uses a geometric progression, causing the thickness of each sub-layer to gradually increase from top to bottom, reflecting the physical characteristic of a gradually increasing concentration gradient from the interface to the bottom. The formula for calculating the equal-segment stratification is:

[0051] ;

[0052] in, For the first The thickness of the layer, The thickness of the topmost layer (closest to the interface). Let the common ratio be (usually taken as 1.2-1.5), satisfying... , Total height of the effective settlement zone. Number of sub-layers. Typically, a value of 5-8 is used, determined based on the size of the thickener and the required control precision. This non-uniform stratification strategy enables the model to have higher resolution in interface regions where concentration changes drastically, while reducing computational resource consumption in regions with slow changes, thus optimizing the balance between computational efficiency and model accuracy.

[0053] For each settling sublayer, the solids flux is calculated based on the cross-sectional area of ​​the thickener corresponding to its center height and the solids settling velocity at that height. Solids flux calculation is the basis for residence time derivation, estimating material transfer rate through geometric and kinetic parameters. The calculation process first determines the height of the center point of each sublayer, then calculates the cross-sectional area at that height based on the thickener's geometry (usually a cylindrical upper section and a conical lower section); simultaneously, the solids settling velocity at that height is estimated based on empirical formulas or measured data, and the solids flux is calculated by combining both factors. The solids flux calculation formula is:

[0054] ;

[0055] in, For the first Solid flux of the layer (t / h) Let m be the cross-sectional area (m²) of the concentrator at the center height of this layer. The solid settling velocity at that height (m / h) The solid concentration (t / m³) of this layer is given. Settling velocity is usually related to solid concentration. The system constructs a velocity-concentration empirical model based on historical data to accurately estimate the settling velocity of each layer.

[0056] Based on the solids flux value and geometric volume of each sublayer, the material residence time of that sublayer is calculated. Residence time calculation is a crucial step in describing transport dynamics, deriving time parameters through the relationship between flux and volume. The calculation process first determines the geometric volume of each sublayer, considering variations in the shape of the thickener; then, combining this with the solids flux value of that layer, the time required for material to pass through that layer, i.e., the residence time, is derived. The formula is:

[0057] ;

[0058] in, For the first Material residence time in the layer (h), The geometric volume (m³) of this layer. The solid concentration (t / m³) of this layer. This represents the solids flux (t / h) of the layer. This formula essentially calculates the ratio of the mass of solids stored in the layer to the mass flow rate of solids passing through the layer, directly reflecting the average residence time of the material in that layer.

[0059] The total propagation lag time from the mud interface to the bottom outlet is obtained by summing the residence times of all settling sublayers. The total propagation lag time is a comprehensive description of the system's delay characteristics; a global time model is established through summation. The calculation process simply involves summing the residence times of each sublayer to obtain the total time for material or effects to propagate from the interface to the bottom outlet. The formula is:

[0060] ;

[0061] in, For the total propagation lag time, For the first Material residence time in the layer, The total number of sub-layers represents the total propagation lag time. This is a key characteristic parameter of the system, directly influencing the time window setting of the control strategy and the time span of the prediction model, providing a time frame for describing the propagation of the agent effect. This multi-layered cumulative time model more accurately describes the dynamic characteristics of complex settlement systems than a simplified single-delay model, significantly improving the accuracy of prediction and control.

[0062] In this embodiment of the invention, the detailed implementation steps for creating a drug effect propagation descriptor include:

[0063] The difference between the flocculant dosing rate in the current control cycle and the dosing rate in the previous control cycle is calculated and recorded as the dosing increment. Increment calculation is the trigger step for descriptor creation, identifying effective adjustments through difference analysis. The calculation process compares the flocculant dosing rates of adjacent control cycles; when the absolute value of the difference exceeds a preset threshold, it is recorded as an effective dosing increment. The increment value can be positive (increased dosing) or negative (decreased dosing), directly reflecting the direction and intensity of the control intervention. The system uses filtering technology to process the raw dosing rate data, reducing the impact of measurement noise and ensuring the accuracy of the increment calculation. This increment-based triggering mechanism avoids over-responding to minor fluctuations, focusing on capturing significant control behaviors and improving the efficiency and significance of descriptor tracking.

[0064] The initial effective concentration of the reagent is calculated based on the ratio of the incremental dosage to the current feed flow rate. This initial concentration calculation is fundamental to quantifying reagent efficacy, determining the initial effect through the relationship between the dosage increment and flow rate. The calculation process involves dividing the incremental dosage by the current feed flow rate to obtain the increase in reagent concentration per unit feed volume, i.e., the initial effective concentration. The formula is:

[0065] ;

[0066] in, This represents the initial effective drug concentration. The increment value (g / min) is used. The feed flow rate is t / min. This calculation method takes into account the dynamic changes in feed conditions, ensuring consistent assessment of the agent effect under different flow rates, and providing a standardized base value for subsequent effect quantity calculations.

[0067] By combining the disturbance type identifier in the disturbance feature vector with the corresponding flocculation efficiency coefficient, the product of the initial effective reagent concentration and the flocculation efficiency coefficient is assigned as the initial value of the residual effect. Efficiency adjustment is a key step in adapting to different disturbances, and effect evaluation is optimized through type matching. The adjustment process first looks up the corresponding coefficient from a preset efficiency coefficient table based on the type identifier (particle size change or concentration change) in the disturbance feature vector; then, this coefficient is multiplied by the initial effective concentration to obtain the initial value of the residual effect. The flocculation efficiency coefficient reflects the actual difference in effectiveness of the same dosage under different slurry conditions. Finer particle size usually corresponds to a smaller coefficient (requiring more reagent), while increased concentration corresponds to a larger coefficient (increased reagent utilization). This dynamic adjustment mechanism based on disturbance type enables the system to make more accurate reagent evaluations for different property changes, significantly improving the adaptability and effectiveness of control.

[0068] The current settling sublayer index of the descriptor for the propagation of the agent effect is set to the sequence number of the top-level settling sublayer, and the current time is recorded as the injection timestamp. Initialization configuration is the final step in descriptor creation, setting the initial state and time base. The configuration process initializes the descriptor's sublayer index to 1 (top-level), indicating that the agent effect has just begun entering the settling zone; simultaneously, the current system time is recorded as the injection timestamp, serving as a reference point for subsequent time calculations. The system employs a unified time standard and indexing mechanism to ensure that all descriptors operate within the same framework, facilitating coordinated management and comparative analysis. The complete initialization of the descriptor ensures that it contains all the information needed to track the propagation of the agent effect: dosage (represented by the residual effect quantity), location (represented by the sublayer index), and time (represented by the timestamp), laying the foundation for subsequent dynamic propagation simulations.

[0069] In this embodiment of the invention, the detailed implementation steps of advancing the position migration of all active agent effect propagation descriptors between adjacent sedimentary sublayers in a controlled cycle, and accumulating the residual effect of all active descriptors that have not yet reached the underflow outlet, and recording it as the in-transit cumulative effect value, include:

[0070] For each active agent effect propagation descriptor, its residence time in the current settling sublayer is accumulated. This residence time accumulation forms the basis for descriptor advancement, controlling the migration pace through time tracking. The accumulation process updates the residence time of each descriptor in each control cycle, with an increment equal to the duration of one control cycle (typically in minutes). The system maintains a list of active descriptors, each containing complete status information, including the current sublayer index, residence time in that layer, and remaining effect quantity. The accumulation of residence time enables precise time management of the agent effect transmission process in each sublayer, providing a temporal basis for location migration judgment. This time-accumulation-based advancement mechanism conforms to the physical laws of material transport, ensuring consistency between the simulation process and the actual system dynamics.

[0071] When the residence time reaches the material residence time of the current settling sublayer, the settling sublayer index of the descriptor is incremented to the next layer, and the residence time is reset to zero. Position migration is the core step in descriptor advancement, simulating the propagation process through inter-layer transfer. During each control cycle, the migration process checks whether the residence time of each descriptor has reached or exceeded the material residence time of the current sublayer; if so, the sublayer index is incremented by 1 (migrating down one layer), and the residence time is reset to zero, starting the time accumulation in the new sublayer. This condition-triggered migration mechanism accurately simulates the settling process of materials under gravity, reflecting the different transport rate characteristics of each sublayer. The system achieves spatial position tracking of descriptors within the settling zone through index increment, intuitively presenting the dynamic distribution of the reagent effect within the thickener.

[0072] During each interlayer migration, the remaining effect of the descriptor is multiplied by the effect retention coefficient corresponding to that settling sublayer. The effect retention coefficient characterizes the proportion of flocculant effectiveness attenuation caused by shear within that sublayer. Effect attenuation is a key aspect of propagation simulation, and the transmission loss is quantified using the retention coefficient. The attenuation process is simulated by applying the effect retention coefficient during interlayer migration of the descriptor, mimicking the effectiveness loss of the flocculant during transport. The formula is:

[0073] ;

[0074] in, This represents the residual effect after entering the next layer. This represents the residual effect size of the current layer. The effect retention coefficient for the current layer ( The retention coefficient reflects the differences in shear environment among different sublayers; typically, the coefficient is smaller in the lower sublayers (due to more significant attenuation) because the shear strength increases with increasing concentration. This layer-based simulation of effect attenuation accurately reflects the physical process of flocculant degradation under shear, significantly improving the realism of the agent effect propagation model.

[0075] When the settling sub-layer index of a descriptor exceeds the lowest-level sequence number, the descriptor is removed from the active list. List management is the final stage of the descriptor lifecycle, and automatic cleanup is achieved through index checks. The management process checks the sub-layer indices of all descriptors in each control cycle. When a descriptor's index value exceeds the lowest-level number (indicating that the entire propagation process has been completed), it is deleted from the active list. This automatic removal mechanism based on completion conditions ensures efficient management of the active list and avoids invalid descriptors consuming computational resources. The system retains key information about removed descriptors for historical analysis, while simultaneously freeing up memory space, optimizing operational efficiency and resource utilization. Complete descriptor lifecycle management ensures that the system can accurately track the entire impact of each reagent adjustment, from injection to reaching the underflow outlet.

[0076] The process iterates through all descriptors in the active list, sums their residual effect sizes, and obtains the cumulative effect value in transit. Effect accumulation is the summation step in the overall impact assessment, and it is performed globally through list traversal. The accumulation process simply sums the residual effect sizes of all descriptors in the active list to obtain the cumulative effect value in transit, using the following formula:

[0077] ;

[0078] in, This is the cumulative effect value during transit. For the first Residual effect size of each active descriptor This represents the total number of currently active descriptors. The cumulative effect value in transit directly reflects the total effect of the reagents "already added but not yet reached the bottom flow" in the system, and is a key parameter for evaluating the current state of the system and deciding on subsequent control. This global effect evaluation method based on descriptor sets quantifies the cumulative effect of multiple independent additions in a unified manner, providing the control system with the ability to comprehensively consider the influence of historical behavior, and significantly improving the foresight and accuracy of control decisions.

[0079] In this embodiment of the invention, when a disturbance feature vector triggers a dosage adjustment request, the detailed implementation steps for calculating the feedforward compensation amount, subtracting the in-transit cumulative effect value from the feedforward compensation amount, and applying a saturation limiting constraint to obtain the bounded net adjustment amount include:

[0080] Based on the disturbance type identifier and gradient peak value in the disturbance feature vector, and combined with the correspondence between the undercurrent concentration deviation and the dosage adjustment under similar historical disturbances, the feedforward compensation amount is calculated. Feedforward compensation calculation is the initial step in control adjustment, determining the basic response through empirical mapping. The calculation process first extracts the type identifier and gradient peak value from the disturbance feature vector; then, it queries the historical experience database maintained by the system to find the historical case closest to the current disturbance characteristics; finally, based on the response effect of the historical case, it determines the feedforward compensation amount suitable for the current situation. The system uses a combination of case reasoning and interpolation calculation to construct a continuous mapping relationship between discrete historical data points, improving the accuracy and smoothness of the compensation amount calculation. This feedforward calculation method based on historical experience fully utilizes the system's learning ability, applying past successful experiences to new similar situations, achieving intelligent decision-making through "learning by analogy," and greatly improving the system's response speed and accuracy to known types of disturbances.

[0081] Subtracting the accumulated effect value in transit from the feedforward compensation amount yields the original net adjustment. The summation process is a crucial step to avoid redundant adjustments, taking into account the in-transit effect through subtraction. The process subtracts the feedforward compensation amount (ideal adjustment amount) from the accumulated effect value in transit (adjustments implemented but not yet reached) to obtain the net adjustment requirement after considering historical impacts. The formula is:

[0082] ;

[0083] in, This is the original net adjustment. This is the feedforward compensation amount. This is the cumulative effect value during transit. This superposition processing mechanism, which takes into account historical behavior, effectively avoids control redundancy and over-response, solves the "repeated adjustment" problem commonly encountered by traditional control systems when dealing with long-delay systems, and significantly improves the stability and efficiency of control.

[0084] The saturation limit is set based on a comparison between the current rake torque and a preset torque threshold. When the rake torque exceeds the preset threshold, the saturation limit is lowered. Safety limiting is a protective element in control decisions, preventing over-adjustment through load monitoring. The limiting process first compares the current rake torque with a preset safety threshold (typically 75-85% of the rated torque); then, it dynamically adjusts the saturation limit based on the comparison result. When the torque approaches the threshold, the system automatically reduces the maximum allowable adjustment to prevent potential overload risks. The limit is calculated using a piecewise function, setting different limiting strategies for different torque ranges to ensure the safety and smoothness of control intervention. This adaptive limiting mechanism based on equipment status adds a safety layer to the control system, avoiding the risk of equipment damage caused by aggressive control while maintaining necessary control flexibility.

[0085] The initial net adjustment is limited to between zero and the upper limit of the saturation threshold to obtain a bounded net adjustment. The range constraint is the final decision-making constraint, and executability is ensured through bilateral limiting. The constraint process employs a simple range truncation operation to ensure the adjustment is within a reasonable range; the formula is as follows:

[0086] ;

[0087] in, This is a bounded net adjustment. This is the original net adjustment. The upper limit is set to the saturation limit. The lower limit is set to zero to prevent negative adjustments in special circumstances (when the cumulative effect in transit is much greater than the feedforward compensation), ensuring the physical feasibility of the control action. The dynamic setting of the upper limit ensures that control will not cause equipment overload. This bilateral limiting mechanism keeps control intervention within a safe and effective range, a crucial guarantee for reliable system operation. The bounded net adjustment, as the final decision result, is directly used to execute the flocculant dosing operation, achieving a complete closed loop from analysis to control.

[0088] In this embodiment of the invention, the detailed implementation steps for constructing the predicted trajectory of future bottom current concentration based on the remaining propagation time of all active descriptors include:

[0089] For each active agent effect propagation descriptor, the cumulative residence time of material from its current settling sublayer index to the bottommost sublayer is calculated and recorded as the remaining propagation time for that descriptor. The remaining time calculation is a fundamental step in prediction, assessing the arrival time through accumulation. The calculation process for each active descriptor starts from its current sublayer and accumulates the residence time of material in that layer and all layers below it to obtain the estimated time for that descriptor to reach the underflow outlet from its current location. The formula is:

[0090] ;

[0091] in, For descriptor The remaining propagation time, For descriptor The current sub-index, For the first Material residence time in the layer, This is the lowest-level sequence number. The remaining propagation time reflects the "reaching the expected" state of each descriptor and is a key parameter for constructing the timeline projection, providing time positioning for subsequent predictions.

[0092] Starting from the current moment, the residual effect sizes of each descriptor are projected onto corresponding time nodes on a future timeline according to their respective residual propagation times. Timeline projection is the core step in forecast construction, establishing a future impact distribution through mapping relationships. The projection process first establishes a future timeline starting from the current moment, covering the longest residual propagation time; then, it maps the residual effect size of each descriptor to the future time point corresponding to its residual propagation time, forming a discrete effect distribution. The projection uses precise time mapping, ensuring that the impact of each descriptor is accurately located to its expected arrival time, providing a temporal organizational framework for subsequent cumulative calculations. This time projection method based on a physical propagation model gives the forecast clear physical meaning and interpretability, significantly improving the reliability of the forecast results.

[0093] At each discrete time step on the future time axis, the effects of all descriptors reaching the underflow outlet at that time step are accumulated to obtain the predicted concentration increment for that time step. Time-point accumulation is a crucial step in trajectory generation, constructing a single-point prediction through the superposition of concurrent effects. The accumulation process traverses each discrete time step on the future time axis (usually in units of control periods), identifies all descriptors expected to reach the underflow at that time step, and accumulates their residual effects to obtain the predicted concentration increment for that time point. The formula is:

[0094] ;

[0095] in, For time points The predicted concentration increment, For descriptor The residual effect size, summed under the condition that the descriptor is expected to be at time point Reaching the bottom outlet. This method of accumulating effects at specific points in time intuitively reflects the concentrated impact at each moment, forming the basic framework of the predicted trajectory.

[0096] The predicted concentration increments at each time step are superimposed onto the current undercurrent concentration measurement to generate a predicted trajectory for future undercurrent concentrations. Baseline superposition is the final step in completing the trajectory, constructing a complete sequence through incremental accumulation. The superposition process uses the current undercurrent concentration measurement as a baseline, sequentially accumulating the predicted increments at each time point to form a continuous predicted trajectory. The formula is:

[0097] ;

[0098] in, For time points Predicted underflow concentration, This is the current measured value of the underflow concentration. For time points The system predicts the concentration increment. The predicted trajectory visually displays the future trend of underflow concentration, providing a time window for risk prediction and early intervention. The system implements threshold monitoring on the predicted trajectory. When it detects that the concentration may exceed the preset upper limit at a certain time in the future, it triggers a reduction signal in advance to achieve proactive intervention and avoid the risk of excessive equipment load or pipeline blockage caused by excessive underflow concentration. This predictive control strategy based on a physical model significantly improves the system's risk prevention capabilities and operational stability.

[0099] In this embodiment of the invention, the detailed steps for comparing the current measured value of the undercurrent concentration with the predicted value generated at the corresponding historical time to calculate the lag prediction error include:

[0100] In each control cycle, the system stores the predicted trajectory of future undercurrent concentrations and the timestamp of its generation. Predictive storage serves as the data preparation step for error analysis, establishing a basis for comparison through historical records. The storage process saves newly generated predicted trajectories and their timestamps to the historical database in each control cycle, forming a complete historical prediction history. The system employs an efficient time-series storage structure to ensure rapid data retrieval and long-term availability. Stored content includes key information such as the predicted trajectory's time point values, the absolute timestamp of its generation, and system state parameters, providing comprehensive data support for subsequent prediction evaluation and model optimization. This systematic prediction recording mechanism enables the system to "review history" and "verify itself," serving as the infrastructure for self-learning optimization.

[0101] At the current moment, the system retrieves the predicted concentration value corresponding to the current moment from the future undercurrent concentration prediction trajectory generated during the control cycle prior to the total propagation lag time. Historical retrieval is the data acquisition stage for error calculation, finding the corresponding prediction through time matching. The retrieval process first calculates the backtracking time point, i.e., the current moment minus the total propagation lag time, to find the historical moment when a prediction should have been made; then, it searches for the corresponding predicted value for the current moment in the prediction trajectory generated at that historical moment. The retrieval employs precise time indexing and interpolation techniques to ensure accurate retrieval of the corresponding historical prediction, even if the actual time point does not perfectly match the discrete storage point. This time backtracking mechanism based on a physical model ensures the physical rationality of the comparison, reflecting the true relationship between "past predictions for the present" and "the current actual situation."

[0102] The difference between the current measured underflow concentration and the predicted concentration is calculated and denoted as the lag prediction error. Error calculation is a core step in model evaluation, quantifying prediction accuracy through difference analysis. The calculation process simply involves subtracting the current measured underflow concentration from the historical predicted value to obtain the lag prediction error. The formula is:

[0103] ;

[0104] in, This is the lag prediction error. This is the current measured concentration value. These are historical predicted values. The lag prediction error directly reflects the accuracy of the prediction model; a positive value indicates that the actual concentration is higher than the prediction (prediction is too low), and a negative value indicates that the actual concentration is lower than the prediction (prediction is too high). This error calculation method based on delay matching overcomes the limitation of traditional real-time error calculation, which cannot be applied to long-latency systems, and provides an effective performance indicator for model optimization.

[0105] An exponentially weighted moving average is applied to the lag prediction errors over multiple consecutive control cycles to obtain a smoothed error trend value. Trend smoothing is an optimization step in noise suppression, extracting a stable pattern through weighted averaging. The smoothing process applies an exponentially weighted moving average algorithm to the lag prediction errors over multiple consecutive control cycles to reduce the impact of random fluctuations and extract a stable error trend. The formula is:

[0106] ;

[0107] in, For a moment The smoothing error trend value, For a moment The lag prediction error, For a moment The smoothing error trend value, The smoothing coefficient is typically set between 0.1 and 0.3. The exponential weighting method assigns higher weight to recent errors while preserving historical trend information, balancing response speed and stability. The smoothed error trend value better reflects the model's systematic bias, providing a reliable basis for parameter correction, avoiding over-response to instantaneous fluctuations, and improving the robustness and effectiveness of the correction.

[0108] In this embodiment of the invention, the detailed implementation steps for dynamically correcting the material residence time and reagent effect attenuation parameters of each settling sublayer based on the hysteresis prediction error include:

[0109] When the smoothed error trend value is positive, it is determined that the actual arrival time of the reagent effect at the underflow is earlier than the predicted value. The material residence time of each settling sublayer is then shortened according to a preset adjustment step size. Time acceleration correction is an adjustment step in model optimization, addressing the earlier effect by shortening parameters. The correction process first determines the sign of the smoothed error trend. When it is positive (actual concentration higher than predicted), it indicates that the effect propagates faster than the model predicts. Then, the material residence time of each settling sublayer is shortened according to a preset adjustment step size (usually 1-5% of the current value), making the model's prediction of propagation speed closer to reality. The correction adopts a proportional allocation strategy, determining the specific adjustment amount based on the relative contribution of each layer to ensure the balance and physical rationality of the correction. This directional adjustment mechanism based on the error sign enables the model to adaptively track changes in system characteristics, continuously optimize time parameters, and improve the time accuracy of predictions.

[0110] When the smoothed error trend value is negative, it is determined that the actual time for the reagent effect to reach the underflow is later than the predicted value. The material residence time of each settling sublayer is then extended by a preset adjustment step size. Time deceleration correction is a complementary step in model optimization, addressing the delay effect by extending the parameters. During the correction process, when the smoothed error trend is negative (actual concentration lower than predicted), it indicates that the effect propagation speed is slower than the model anticipates. Then, the material residence time of each settling sublayer is extended by the same preset adjustment step size, allowing the model to more accurately reflect the actual propagation delay. The extension adjustment also employs a proportional allocation strategy to ensure the systematic and coherent nature of the correction. This two-way adjustment mechanism enables the system to perform appropriate self-corrections under different conditions, avoiding oscillations or divergences that may result from unidirectional adjustment, and improving the stability and convergence of model optimization.

[0111] The attenuation parameters of the reagent effect for each settling sublayer are recalculated based on the adjusted material residence time. Updating these parameters is a crucial step in maintaining model consistency, ensuring internal coordination through correlated adjustments. The update process recalculates the reagent effect attenuation parameters for each settling sublayer based on the adjusted material residence time, ensuring consistency between time and effect models. The calculations utilize the physical characteristic equations of the thickener, considering the impact of residence time variations on the shear environment, and derive the corresponding effect retention coefficients. This parameter linkage update mechanism guarantees the physical consistency among parameters within the model, avoiding internal model inconsistencies caused by updating some parameters while lagging others, thus improving the overall physical rationality and predictive ability of the model.

[0112] The updated material residence time and reagent effect decay parameters are applied to the subsequent propagation calculations of all active reagent effect propagation descriptors. Parameter application is the execution phase of the model update, ensuring consistent implementation through global updates. The application process uses the updated parameter values ​​for the propagation calculations of all active descriptors, including core operations such as location migration judgment and effect decay assessment. The system ensures the atomicity and consistency of parameter updates, avoiding computational chaos caused by partial applications. Global parameter application allows the model optimization effect to be immediately reflected in the system's prediction and control behavior, accelerating the learning-optimization-application closed loop and improving the system's adaptive speed and effectiveness. This continuous self-optimization mechanism enables the system to continuously adapt to changes in process conditions and drift in equipment characteristics, maintaining long-term high-precision control capabilities.

[0113] The above describes the automatic detection and control method for underflow concentration of tailings thickener in the embodiments of this application. The following describes the automatic detection and control system for underflow concentration of tailings thickener in the embodiments of this application. Please refer to [link to relevant documentation]. Figure 2 One embodiment of the automatic detection and control system for tailings thickener underflow concentration in this application includes:

[0114] The data acquisition module is used to acquire the real-time operating parameters of the tailings thickener, including the solid content of the feed slurry, feed flow rate, flocculant dosing rate, mud layer interface height, rake torque, and underflow concentration measurement.

[0115] The disturbance identification module is used to collect turbidity data at multiple points along the radial direction in the overflow area of ​​the feed well, calculate the turbidity gradient change rate of the turbidity data within the sliding time window, identify feed property disturbance events, and generate disturbance feature vectors.

[0116] The sedimentation stratification module is used to divide the sedimentation zone of the thickener vertically into multiple sedimentation sub-layers based on the mud layer interface height and the geometric parameters of the thickener, and to calculate the material residence time of each sedimentation sub-layer in combination with the solid flux balance relationship.

[0117] The effect tracking module is used to create a descriptor for the propagation of flocculant effects for each dosing event where the change in the flocculant dosing rate exceeds a preset minimum increment threshold.

[0118] The effect propagation update module is used to advance the position migration of all active agent effect propagation descriptors between adjacent sedimentation sub-layers in a control cycle according to the material residence time of each sedimentation sub-layer, and to accumulate the remaining effect of all active descriptors that have not yet reached the underflow outlet, which is recorded as the in-transit cumulative effect value.

[0119] The control decision module is used to calculate the feedforward compensation amount when the disturbance feature vector triggers the dosage adjustment request, subtract the in-transit cumulative effect value from the feedforward compensation amount, apply the saturation limit constraint, obtain the bounded net adjustment amount, and execute the corresponding flocculant dosing operation.

[0120] The prediction and early warning module is used to construct a future undercurrent concentration prediction trajectory based on the remaining propagation time of all active descriptors. When there is a time node in the future undercurrent concentration prediction trajectory that exceeds the preset concentration limit, an early reduction signal is sent to the current control cycle.

[0121] The model self-calibration module is used to compare the measured value of the underflow concentration at the current moment with the predicted value generated at the corresponding historical moment, calculate the lag prediction error, and dynamically correct the material residence time and reagent effect decay parameters of each settling sublayer based on the lag prediction error.

[0122] The modules are connected via wired and / or wireless means to enable data transmission between them.

[0123] This invention achieves intelligent monitoring and precise control of the underflow concentration in tailings thickeners throughout the entire process by acquiring real-time parameters, analyzing turbidity gradients, constructing multi-layer sedimentation models, tracking the propagation of reagent effect descriptors, and implementing prediction-feedback closed-loop correction. The adaptive propagation model of this invention can accurately predict changes in underflow concentration, effectively identify feed disturbances, and implement feedforward control.

[0124] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0125] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0126] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. An automatic detection and control method for underflow concentration in a tailings thickener, characterized in that, include: The real-time operating parameters of the tailings thickener are obtained, including the solid content of the feed slurry, feed flow rate, flocculant dosing rate, mud layer interface height, rake torque, and underflow concentration measurement. Multiple turbidity data points are collected radially in the overflow zone of the feed well. The turbidity gradient change rate is calculated on the turbidity data within a sliding time window to identify feed property disturbance events and generate disturbance feature vectors. Based on the mud layer interface height and the thickener's geometric parameters, the thickener's settling zone is vertically divided into multiple settling sub-layers, and the material residence time of each settling sub-layer is calculated in conjunction with the solid flux balance relationship. For each addition event where the change in the flocculant addition rate exceeds a preset minimum increment threshold, a flocculant effect propagation descriptor is created. The flocculant effect propagation descriptor includes the addition increment value, injection timestamp, current sedimentation sublayer index, and remaining effect amount. Based on the material residence time of each settling sub-layer, the position migration of all active agent effect propagation descriptors between adjacent settling sub-layers is advanced in a controlled cycle, and the residual effect of all active descriptors that have not yet reached the underflow outlet is accumulated and recorded as the in-transit cumulative effect value. When the disturbance feature vector triggers a dosage adjustment request, the feedforward compensation amount is calculated, the in-transit cumulative effect value is subtracted from the feedforward compensation amount, a saturation limit constraint is applied, a bounded net adjustment amount is obtained, and the corresponding flocculant dosing operation is performed. Based on the remaining propagation time of all active descriptors, a future undercurrent concentration prediction trajectory is constructed. When there is a time node in the future undercurrent concentration prediction trajectory that exceeds the preset concentration upper limit, an early reduction signal is sent to the current control cycle. The measured value of the underflow concentration at the current moment is compared with the predicted value generated at the corresponding historical moment to calculate the lag prediction error. Based on the lag prediction error, the material residence time and reagent effect decay parameters of each sedimentation sublayer are dynamically corrected.

2. The method for automatic detection and control of underflow concentration in a tailings thickener according to claim 1, characterized in that, The process of collecting turbidity data at multiple points radially along the overflow zone of the feed well, calculating the turbidity gradient change rate of the turbidity data within a sliding time window, identifying feed property disturbance events, and generating disturbance feature vectors includes: Multiple turbidity sampling points are set at equal intervals radially along the overflow weir of the feed well, and the turbidity value sequence of each turbidity sampling point is collected synchronously; Within a sliding time window of a preset length, the first-order difference mean of the turbidity value sequence of each turbidity sampling point is calculated; Arrange the first-order difference mean values ​​of each turbidity sampling point according to their radial positions, and calculate the difference gradient between adjacent sampling points; When the gradient values ​​of multiple consecutive sampling points in the differential gradient exceed a preset gradient threshold, it is determined that the feed property disturbance event has occurred. Based on the spatial distribution pattern of the differential gradient, the perturbation is distinguished as either a particle size reduction type or a solid concentration abrupt change type, and the perturbation type identifier and gradient peak value are combined to form the perturbation feature vector.

3. The method for automatic detection and control of underflow concentration in a tailings thickener according to claim 1, characterized in that, The process of dividing the thickener's settling zone vertically into multiple settling sub-layers and calculating the material residence time of each settling sub-layer based on the solids flux balance relationship includes: The effective vertical height of the settling zone is determined by taking the height of the mud layer interface as the upper boundary and the bottom cone outlet of the thickener as the lower boundary. The effective vertical height is divided into N settlement sub-layers in a geometric progression, wherein the thickness of the settlement sub-layer near the mud layer interface is less than the thickness of the settlement sub-layer near the bottom cone. For each of the settling sublayers, the solid flux value of the sublayer is calculated based on the cross-sectional area of ​​the thickener corresponding to the center height of the sublayer and the solid settling velocity at that height. Based on the solid flux value of each sublayer and the geometric volume of that sublayer, the material residence time of that sublayer is calculated. The total propagation lag time from the mud interface to the bottom outlet is obtained by summing up the material residence times of all settling sublayers.

4. The method for automatic detection and control of underflow concentration in a tailings thickener according to claim 1, characterized in that, The creation of the drug effect propagation descriptor includes: Calculate the difference between the flocculant dosing rate in the current control cycle and the dosing rate in the previous control cycle, and record it as the dosing increment value; The initial effective reagent concentration is calculated based on the ratio of the incremental dosage to the current feed flow rate; By combining the disturbance type identifier in the disturbance feature vector, the corresponding flocculation efficiency coefficient is queried, and the product of the initial effective agent concentration and the flocculation efficiency coefficient is assigned as the initial value of the residual effect quantity; Set the current sedimentation sub-layer index of the drug effect propagation descriptor to the sequence number of the top sedimentation sub-layer, and record the current time as the injection timestamp.

5. The method for automatic detection and control of underflow concentration in a tailings thickener according to claim 1, characterized in that, The process of advancing the positional migration of all active agent effect propagation descriptors between adjacent sedimentary sublayers in a controlled cycle, and accumulating the residual effect of all active descriptors that have not yet reached the underflow outlet, is recorded as the in-transit cumulative effect value, including: For each active agent effect propagation descriptor, accumulate its residence time in the current sedimentation sublayer; When the residence time reaches the material residence time of the settling sublayer, the settling sublayer index of the descriptor is incremented to the next layer, and the residence time is reset to zero. During each interlayer migration, the residual effect of the descriptor is multiplied by the effect retention coefficient corresponding to the settlement sublayer; When the sinking sub-level index of a descriptor exceeds the lowest level sequence number, the descriptor is removed from the active list; Iterate through all descriptors in the active list, sum their remaining effect sizes, and obtain the in-transit cumulative effect value.

6. The method for automatic detection and control of underflow concentration in a tailings thickener according to claim 1, characterized in that, When the disturbance feature vector triggers a dosage adjustment request, a feedforward compensation amount is calculated. After subtracting the in-transit cumulative effect value from the feedforward compensation amount, a saturation limiting constraint is applied to obtain a bounded net adjustment amount, including: Based on the disturbance type identifier and gradient peak value in the disturbance feature vector, and combined with the correspondence between the undercurrent concentration deviation and the dosage adjustment under similar historical disturbances, the feedforward compensation amount is calculated. Subtract the in-transit cumulative effect value from the feedforward compensation amount to obtain the original net adjustment amount; The upper limit of saturation is set based on the comparison result between the current rake frame torque and the preset torque threshold. When the rake frame torque exceeds the preset torque threshold, the upper limit of saturation is reduced. The bounded net adjustment is obtained by limiting the original net adjustment amount between zero and the upper limit of the saturation limit.

7. The method for automatic detection and control of underflow concentration in a tailings thickener according to claim 1, characterized in that, The construction of the future undercurrent concentration prediction trajectory based on the remaining propagation time of all active descriptors includes: For each active agent effect propagation descriptor, calculate the cumulative value of the material residence time from its current settling sub-layer index to the bottom sub-layer, and record it as the remaining propagation time of the descriptor; Starting from the current moment, the residual effect size of each descriptor is projected onto the corresponding time node on the future time axis according to its corresponding residual propagation time; At each discrete time step of the future time axis, all descriptor effects that reach the underflow outlet at that time step are accumulated to obtain the predicted concentration increment for that time step; The predicted concentration increments at each time step are superimposed onto the current undercurrent concentration measurement to generate the predicted trajectory of the future undercurrent concentration.

8. The method for automatic detection and control of underflow concentration in a tailings thickener according to claim 3, characterized in that, The step of comparing the current measured value of the undercurrent concentration with the predicted value generated at the corresponding historical time and calculating the lag prediction error includes: In each control cycle, the predicted trajectory of the future undercurrent concentration and the timestamp of its generation time are stored. At the current moment, retrieve the predicted concentration value corresponding to the current moment from the future undercurrent concentration prediction trajectory generated by the control cycle prior to the total propagation lag time; The difference between the current measured value of the underflow concentration and the predicted concentration value is calculated and denoted as the hysteresis prediction error; An exponentially weighted moving average is applied to the lag prediction error over multiple consecutive control cycles to obtain a smoothed error trend value.

9. The method for automatic detection and control of underflow concentration in a tailings thickener according to claim 8, characterized in that, The dynamic correction of the material residence time and reagent effect decay parameters of each settling sublayer based on the hysteresis prediction error includes: When the smoothed error trend value is positive, it is determined that the actual time when the agent effect reaches the underflow is earlier than the estimated value, and the material residence time of each settling sub-layer is shortened according to the preset adjustment step size. When the smoothed error trend value is negative, it is determined that the actual time when the agent effect reaches the bottom flow is later than the estimated value, and the material residence time of each of the settling sub-layers is extended according to the preset adjustment step size. Recalculate the agent effect attenuation parameters for each settling sublayer based on the adjusted material residence time; The updated material residence time and reagent effect decay parameters are applied to the subsequent propagation calculations of all active reagent effect propagation descriptors.

10. An automatic detection and control system for underflow concentration of a tailings thickener, used to implement the automatic detection and control method for underflow concentration of a tailings thickener as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to acquire the real-time operating parameters of the tailings thickener, including the solid content of the feed slurry, feed flow rate, flocculant dosing rate, mud layer interface height, rake torque, and underflow concentration measurement. The disturbance identification module is used to collect turbidity data at multiple points along the radial direction in the overflow area of ​​the feed well, calculate the turbidity gradient change rate of the turbidity data within a sliding time window, identify feed property disturbance events, and generate disturbance feature vectors. The sedimentation stratification module is used to divide the sedimentation zone of the thickener vertically into multiple sedimentation sub-layers based on the mud layer interface height and the geometric parameters of the thickener, and to calculate the material residence time of each sedimentation sub-layer in combination with the solid flux balance relationship. The effect tracking module is used to create a descriptor for the propagation of flocculant effects for each dosing event where the change in the flocculant dosing rate exceeds a preset minimum increment threshold. The effect propagation update module is used to advance the position migration of all active agent effect propagation descriptors between adjacent sedimentation sub-layers in a controlled cycle according to the material residence time of each sedimentation sub-layer, and to accumulate the remaining effect of all active descriptors that have not yet reached the underflow outlet, which is recorded as the in-transit cumulative effect value. The control decision module is used to calculate the feedforward compensation amount when the disturbance feature vector triggers the dosage adjustment request, subtract the in-transit cumulative effect value from the feedforward compensation amount, apply the saturation limit constraint, obtain the bounded net adjustment amount, and execute the corresponding flocculant dosing operation. The prediction and early warning module is used to construct a future undercurrent concentration prediction trajectory based on the remaining propagation time of all active descriptors. When there is a time node in the future undercurrent concentration prediction trajectory that exceeds the preset concentration upper limit, an early reduction signal is sent to the current control cycle. The model self-calibration module is used to compare the measured value of the underflow concentration at the current moment with the predicted value generated at the corresponding historical moment, calculate the lag prediction error, and dynamically correct the material residence time and reagent effect decay parameters of each sedimentation sublayer based on the lag prediction error.

Images