Multi-extraction-system intelligent collaborative optimization control system for efficient separation of nickel and cobalt

Abstract

The invention relates to the technical field of industrial process control, in particular to a multi-extraction-system intelligent collaborative optimization control system for efficient separation of nickel and cobalt. According to the system, the water phase ion concentration, the phase interface state and the organic phase health degree are monitored in real time through the enhanced sensing layer module; through an impurity removal section dynamic optimization control module and a separation section intelligent collaborative optimization module, model prediction control and feed-forward-feedback collaborative adjustment are conducted on an impurity removal section and a separation section respectively, and the saponification rate, the O / A ratio and the pH value are dynamically optimized; the global material balance and cost optimization module constructs an economic digital twinborn model to realize comprehensive optimization of the metal yield and cost in the whole process; the intelligent execution and predictive maintenance early warning module is responsible for accurate execution of control instructions, abnormal early warning and maintenance prompting, and an end-to-end intelligent cooperative closed loop from perception, control, optimization to execution is formed. According to the method, the nickel-cobalt separation efficiency, the product purity and the economical efficiency are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of industrial process control technology, specifically to an intelligent collaborative optimization control system for multi-extraction systems aimed at efficient separation of nickel and cobalt. Background Technology

[0002] Hydrometallurgy is a key process for the separation and extraction of nickel and cobalt resources. Among them, multi-stage extraction systems based on extractants such as P204, P507, and C272 are widely used due to their high efficiency and selectivity. In recent years, with the rapid development of intelligent manufacturing and industrial digitalization technologies, the demand for whole-process collaborative optimization and real-time dynamic control in the process control field has been increasing.

[0003] Chinese invention patent CN101813932B discloses a method for predicting and optimizing the composition content of a hydrometallurgical extraction process. This method employs a multi-stage extraction tank hydrometallurgical extraction process and achieves real-time prediction of the composition content of the raffinate by hybrid modeling the hydrometallurgical extraction process. It also provides online optimization guidance for the extraction process. The method includes steps such as data acquisition, selection and standardization of auxiliary variables, establishment of a hybrid model, calibration of the hybrid model, and determination of optimization guidance.

[0004] Currently, multi-source sensing technology, model predictive control, digital twins, and artificial intelligence provide new technical paths for the refined management and overall efficiency improvement of complex industrial processes. Against this backdrop, developing an intelligent collaborative system that integrates sensing, control, optimization, and early warning functions is of great technical significance and development prospects for achieving efficient, stable, and economical operation of the nickel-cobalt extraction process. Summary of the Invention

[0005] The purpose of this invention is to address the problems existing in the background technology by proposing an intelligent collaborative optimization control system for multi-extraction systems aimed at efficient separation of nickel and cobalt.

[0006] The technical solution of this invention: A smart collaborative optimization control system for multi-extraction systems aimed at efficient separation of nickel and cobalt, comprising: The enhanced sensing layer module performs real-time dynamic monitoring of aqueous phase ion concentration, phase interface state, and organic phase health in the multi-stage extraction system, generating multi-source fusion data. The impurity removal section dynamic optimization control module, based on multi-source fusion data, performs model prediction control on the impurity removal section, and dynamically optimizes the saponification rate, the ratio of organic phase to water and the amount of washing acid, so as to achieve efficient removal of impurities and minimize reagent consumption. The intelligent collaborative optimization module for the separation section performs collaborative optimization control that combines feedforward and feedback, dynamically adjusting the pH value, saponifying agent dosage, and the ratio of organic phase to water to improve the selectivity of cobalt-nickel separation and product purity. The global material balance and cost optimization module integrates data from the entire process, builds an economic digital twin model, and conducts comprehensive analysis and global optimization decisions on metal yield and production costs. The intelligent execution and predictive maintenance early warning module transforms optimization strategies into control commands and executes them, while providing anomaly warnings and maintenance prompts based on trend analysis and predictive models.

[0007] Preferably, the process of real-time dynamic monitoring by the enhanced perception layer module specifically includes: High-precision pH sensors and online ion chromatographs or miniature inductively coupled plasma spectrometers are installed at the inlet and outlet of the extractant in the separation section to sample at high frequency and measure the concentration of cobalt ions and nickel ions in the aqueous phase in real time, and calculate the cobalt-nickel separation selectivity index. High-definition industrial cameras are installed in the clarification chambers of the impurity removal and separation sections. Convolutional neural network image recognition models are used to analyze the clarity of the phase interface, the thickness of interface contaminants, and the generation of three phases in real time, and to quantify the interface state indicators. An online near-infrared spectrometer is installed in the separation section to monitor the concentration of the organic phase extractant and the saponification rate in real time, and to calculate the health indicators of the organic phase.

[0008] Preferably, the monitored cobalt ion concentration, nickel ion concentration, separation selectivity index, interface state index, organic phase health index, and real-time aqueous phase pH value are time-series aligned, noise filtered, and feature extracted to generate a unified multi-source data vector for use in the model prediction and control model.

[0009] Preferably, the dynamic optimization process performed by the impurity removal section dynamic optimization control module specifically includes: Based on online ion chromatography and micro inductively coupled plasma spectrometry, the instantaneous load of inlet and outlet impurity ions is quantified in real time, and a comprehensive impurity index and load index are constructed by combining phase boundary image features. By utilizing the changing trends of impurity index and load index, a finite time-domain model predictive control optimizer with soft constraints is used to drive adaptive adjustment of saponification rate. The objective function of the optimizer simultaneously considers the impurity removal objective, the minimization of reagent consumption, and the economic penalty for the load on the separation section. In model predictive control, the recursive least squares method is used to identify the parameters of the predictive model online in order to cope with the fluctuations in feed composition.

[0010] Preferably, based on the dynamic optimization of saponification rate, a fine-tuning strategy for the ratio of organic phase to water is further introduced. The offset of the ratio of organic phase to water is calculated based on the short-cycle prediction results and executed by adjusting the pump speed through feedforward to make the interstage load uniform.

[0011] Preferably, the process of collaborative optimization control performed by the intelligent collaborative optimization module for the separation section specifically includes: Real-time calculation of cobalt-nickel selectivity coefficient, organic phase saponification degree, and separation section load index; A multivariate model predictive control combined with a feedforward correction mechanism is adopted. The predictive model is used and the feedforward correction is performed based on the residual of the impurity removal section output. By optimizing the objective function, the cobalt-nickel selectivity, organic phase saponification degree and reagent consumption are precisely controlled at the same time. The parameters of the separation segment prediction model are updated online using the recursive least squares method, and the control action is automatically adjusted to prioritize interface stability when the interface status index decreases.

[0012] Preferably, when the interface state index is lower than the safety threshold, a control action correction amount is generated at the edge computing node, prioritizing the adjustment of the ratio of organic phase to water or the amount of saponifying agent added, and limiting the rate of change of the control action to ensure interface stability.

[0013] Preferably, the process of comprehensive analysis and global optimization decision-making regarding metal yield and production costs by the global material balance and cost optimization module specifically includes: By integrating the real-time inlet and outlet flow rates, metal concentrations, pH values, saponification degrees, and interface indicators of each unit, a global material balance model is established to dynamically calculate the total metal content and yield of each unit, and then visualize it on a digital twin interface. Collect the flow rates of acids, alkalis, and saponifying agents in each section and the unit metal consumption, construct an economic digital twin model, and integrate the reagent costs, separation section load, and interface risks in a weighted manner to generate a real-time cost curve and a visualization interface. A global optimization objective function is established that simultaneously considers the overall recovery rate, comprehensive cost index, control action variation, and organic phase saponification degree constraint. A reinforcement learning model is then used to generate a cross-segment collaborative optimal control strategy.

[0014] Preferably, the intelligent execution and predictive maintenance early warning module translates optimization strategies into control commands and executes them. Simultaneously, based on trend analysis and predictive models, it provides anomaly warnings and maintenance alerts. The specific steps include: The global optimization control strategy is decomposed into specific pump and valve opening, organic phase to water phase flow ratio setpoint, and saponifier dosing acceleration command, and then sent to the execution layer. Within the defined execution confirmation window, the system status response after execution is monitored in real time, and the execution residual is calculated. When the execution residual exceeds the maximum tolerance, it automatically reverts to a conservative control strategy and generates a manual prompt message.

[0015] Preferably, the intelligent execution and predictive maintenance early warning module translates optimization strategies into control commands and executes them. Simultaneously, the steps of providing anomaly warnings and maintenance alerts based on trend analysis and predictive models specifically include: Real-time monitoring of clarification chamber interface indicators and key equipment operating parameters; Based on real-time collected data on interface clarity, saponification degree, organic phase health, and pump flow rate, the probability of equipment failure and interface anomaly risks in the future time period is calculated using trend analysis and prediction models. When the predicted risk exceeds the set threshold, maintenance or operation adjustment suggestions are automatically generated. Record historical execution actions, early warning events, and economic indicators to train reinforcement learning models and achieve iterative optimization of strategies across production batches.

[0016] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects: This invention designs an intelligent collaborative optimization control system for multi-extraction systems aimed at efficient nickel-cobalt separation. Through an enhanced sensing layer module, it achieves real-time, high-precision monitoring of aqueous phase ion concentration, interface state, and organic phase health throughout the entire process, providing a reliable data foundation for intelligent control. The dynamic optimization control module in the impurity removal section can adaptively adjust the saponification rate, O / A ratio, and washing acid dosage based on real-time monitoring data. This ensures efficient impurity removal while significantly reducing acid and alkali reagent consumption and alleviating the operational load of subsequent separation sections. The intelligent collaborative optimization module in the separation section uses a model predictive control strategy combining feedforward and feedback to dynamically adjust pH value, saponifying agent dosage, and O / A ratio, effectively improving cobalt-nickel separation. The invention improves selectivity and product purity while further optimizing reagent usage efficiency. The global material balance and cost optimization module integrates data from the entire process to construct an economic digital twin model, enabling comprehensive analysis and optimization of metal yield and production costs, supporting operators in making decisions with optimal economic benefits. The intelligent execution and predictive maintenance early warning module ensures the precise execution of optimization strategies and identifies potential equipment and interface anomalies in advance through trend analysis and predictive models, achieving predictive maintenance. Ultimately, this invention forms a complete closed loop from real-time perception, dynamic control, global optimization to intelligent execution, greatly improving the overall efficiency, stability, and economy of the nickel-cobalt separation process, while reducing operational risks and the intensity of manual intervention. Attached Figure Description

[0017] Figure 1 This is a system architecture diagram of an intelligent collaborative optimization control system for multi-extraction systems aimed at efficient separation of nickel and cobalt proposed in this invention. Figure 2 This is a flowchart illustrating the working method of an intelligent collaborative optimization control system for efficient separation of nickel and cobalt in a multi-extraction system proposed in this invention. Detailed Implementation

[0018] Example 1, as Figure 1 As shown, the present invention proposes an intelligent collaborative optimization control system for multi-extraction systems for efficient separation of nickel and cobalt, comprising: an enhanced sensing layer module, a dynamic optimization control module for the impurity removal section, an intelligent collaborative optimization module for the separation section, a global material balance and cost optimization module, and an intelligent execution and predictive maintenance early warning module.

[0019] The enhanced sensing layer module deploys a high-precision pH sensor and online ion chromatography or miniature ICP-OES to perform high-frequency sampling of the inlet and outlet aqueous phases of each P507 / C272 stage, enabling real-time monitoring of Co. 2+ Ni 2+ Concentration and ratio; simultaneously, high-definition cameras are installed in the clarification chambers of P204 and P507 to analyze the clarity of the phase interface, the thickness of interface contaminants, and the formation of three phases in real time through image recognition models; an online near-infrared spectrometer is used to monitor the concentration of organic phase extractant and saponification rate in real time, providing high-precision real-time data for the entire process and providing basic sensing input for intelligent control; The dynamic optimization control module for the impurity removal section dynamically optimizes the control of the P204 impurity removal section based on the impurity ion concentration and interface indicators provided by the enhanced sensing layer. By adjusting the saponification rate, O / A ratio and washing acid dosage in real time, it achieves optimal removal of impurity ions while minimizing the consumption of alkali and acid, reducing the load on the separation section, ensuring the efficient and stable impurity removal process, and providing stable raw material conditions for the subsequent separation section. The intelligent collaborative optimization module for the separation segment, specifically targeting the core separation segment of P507 / C272, performs real-time calculation of β. Co / Ni The ratio is used to fine-tune the pH control valve opening through a combination of feedforward and feedback methods, stabilizing the pH within the optimal range; based on Co 2+ The loading condition predictively adjusts the amount of saponifying agent added to ensure that the organic phase maintains the optimal saponification state, and adjusts the O / A ratio and interstage reflux in advance according to changes in feed composition to achieve efficient nickel-cobalt separation; this module operates in coordination with the impurity removal section to form a dynamic optimization closed loop for the entire process; The global material balance and cost optimization module integrates the metal flow rate, reagent consumption, and interface indicators of each unit to build a global material balance and economic digital twin model; it calculates the distribution and yield of cobalt and nickel in each unit in real time, and correlates reagent consumption and energy consumption data to form a cost optimization analysis; operators can evaluate the impact of different control strategies on yield, cost, and interface stability based on the simulation interface, select the most economically efficient solution, and achieve global optimization decision-making. The intelligent execution and predictive maintenance early warning module distributes global optimization strategies to the pump, valve, and saponifying agent dosing system to achieve precise execution; it monitors execution deviations, interface indicators, and saponification degree in real time, and provides early warnings of potential maintenance needs, such as clarification chamber cleaning or pump and valve inspections, through trend analysis and predictive models; this module also supports strategy fine-tuning and simulated operation to achieve synergistic optimization of operational safety, economy, and yield, and records execution data for subsequent reinforcement learning and strategy iteration.

[0020] Example 2, as Figure 2 As shown, the present invention proposes an intelligent collaborative optimization control system for multi-extraction systems aimed at efficient separation of nickel and cobalt, the operation of which includes the following specific implementation steps: S1. Through multi-dimensional high-precision sensors and intelligent data processing, real-time dynamic monitoring of the aqueous phase, interface, and organic phase in the multi-stage extraction system is achieved. Multi-source data is fused, filtered, and feature extracted to generate high-value inputs that can be directly used by predictive control models. This provides a precise, continuous, and implementable data foundation for the subsequent optimization control of the impurity removal and separation stages. The specific implementation process is as follows: S11. High-precision pH sensors and online ICP-OES / IC are installed at the inlets and outlets of P507 and C272 to measure the pH in the aqueous phase in real time. , Concentration and ratio : ; in, This indicates real-time measurement of cobalt ion concentration in the aqueous phase; This indicates real-time measurement of nickel ion concentration in the aqueous phase; Indicates the selectivity index for separation; Dynamic concentration curves are generated through high-frequency sampling, providing basic data for predictive regulation and feedforward control; S12. Install high-definition industrial cameras in clarification chambers P204 and P507, and use convolutional neural networks (CNN) to analyze interface sharpness, three-phase material generation, and contaminant thickness, quantifying the interface state index I. interface,t It enables real-time alarms for interface anomalies and provides optimized control inputs. ; in, Indicates real-time interface status metrics; The interface sharpness index is obtained from image grayscale gradient analysis. The thickness of the interface contaminant is estimated from the CNN segmentation results. The three-phase generation index is calculated using image color distribution and particle recognition algorithms. , and The weighting coefficients are fitted based on actual process experience and historical data. S13. Install an online near-infrared (NIR) spectrometer in the separation section to monitor the concentration of the organic phase extractant and the saponification rate in real time, and calculate the organic phase health index H. org,t By continuously and dynamically evaluating the organic phase to ensure it is in an optimal state, real-time organic phase information is provided for predictive control. ; in, This represents the real-time organic phase health index, ranging from [0,1]. The closer the value is to 1, the closer the organic phase is to its optimal health state. This indicates the current concentration of the organic phase extractant, measured by NIR spectroscopy; This indicates the optimal concentration setpoint, which is determined by the process design. S14. Perform time series alignment, noise filtering, and feature extraction on the multi-source data collected in steps S11-S13 to generate a unified data vector X. t This is used in the MPC predictive control model to achieve continuous, precise, and multi-dimensional control of the impurity removal and separation sections, forming a complete closed-loop sensing system, namely: ; Among them, pH t This indicates the real-time pH value of the aqueous phase.

[0021] S2. A dynamic optimization control mechanism based on online component sensing is constructed around the impurity removal section (P204 system). Through comprehensive input of real-time impurity concentration, phase boundary state, and organic phase health, a customized model is used to calculate the optimal combination of saponification rate, O / A ratio, and washing acid strategy. This establishes a linkage and compensation relationship between the impurity removal load and the stability of the subsequent nickel-cobalt separation section, thereby continuously reducing reagent consumption while ensuring the impurity removal effect. The specific implementation process is as follows: S21. The instantaneous loading of inlet and outlet impurity ions is quantified in real time by online ion chromatography and micro ICP-OES, and a comprehensive impurity index is constructed by combining the phase boundary image characteristics entering the impurity removal section. This establishes a correlation between single-point concentration and interface behavior, i.e., the multi-source vector X... t Perform short-term statistical analysis and rapid reliability assessment to generate the current cleanup segment's operating status set S. t With a set of soft safety constraints C t It is responsible for mapping the sensed signals into constraints and initial state estimates that can be used by the optimizer, including noise correction, short-term trend separation, and confidence interval calculation, specifically: Applying a fast Kalman filter to the sensor readings yields a smoothed estimate: ; and interface metrics With organic phase index Establish soft constraints C t When the threshold is approached, a "penalty term" will be injected into the optimizer cost function to avoid frequent shutdowns. ; ; Generate current load index L t : ; Based on this, the state estimation after output filtering is performed. Soft constraint set C t and load index L t , serving as the initial value and constraint input for mid-term prediction optimization; Among them, I min This indicates the set threshold for the interface status indicator; H min This indicates the set threshold for the health of the organic phase; X represents the multi-source data vector at time t. t The state vector after smoothing / estimation by filters such as Kalman filtering (denoising and short-term prediction correction); and Indicates empirical weights, reflecting the relative impact of each metal on downstream load or separation difficulty (obtained from historical material balance fitting); S load This represents the set of all aqueous phase inlet and outlet nodes equipped with online ion monitoring units within the P204 impurity removal section; S22. The saponification rate is adaptively controlled by utilizing the changing trend of the impurity index, maintaining the effective activity of the organic phase within a narrow range that ensures adsorption efficiency. The optimal saponification target value is solved in real time by coupling the saponification degree obtained from NIR spectral inversion with historical response curves. The amount of alkali added is adjusted using dynamic deviation compensation to achieve the optimal balance between impurity removal effect and reagent consumption. Specifically, a finite-time domain MPC (Model Predictive Control) optimizer with soft constraints is constructed. Its objective function balances impurity removal and reagent / energy consumption minimization, and an economic penalty is imposed on the P507 load. The optimizer uses a simplified dynamic process model as its prediction kernel and allows online identification of model parameters to cope with material fluctuations. Define control variable vector , representing the actions of the saponifying agent dosing rate, O / A ratio adjustment ratio, and washing acid flow rate at time k, respectively; state vector x k Select key measurable and inferable variables (including but not limited to effluent impurity concentration, outlet pH, and interface parameters). Among them, u k u represents the control action vector at time k; soap Indicates the saponifying agent addition flow rate; u O / AThis indicates the adjustment and control amount of the organic / aqueous phase (O / A) ratio; Indicates the flow rate of washing acid (used in the impurity removal washing section); Building a predictive model: Matrix A, B, and E are adaptively identified using recursive least squares (RLS); Where, x k A represents the set of states used to describe the dynamics of the system; B represents the dynamic matrix of the state of the cleanup section itself as a function of time; C represents the control matrix of the cleanup section; D represents the disturbance matrix; E represents the disturbance matrix. k This represents the disturbance vector, which characterizes external disturbances such as fluctuations in feed composition and temperature changes. Construct the MPC cost function J and optimize it within the prediction time domain N: ; Among them, y k+i|k y represents the output value at time k+i predicted from the current time k based on the model; target Indicates the target or objective to be achieved; This indicates the predicted reagent consumption, which is the total reagent consumption estimated based on control actions within the prediction window; This indicates the predicted back-end load, i.e., the load forecast for P507 (e.g., load index L). t (predicted value); Indicates the amount of change in the control action; , , , and The weights of the cost function represent the weights for output bias (quality), reagent consumption (economy), backend load (system coordination), action smoothness, and soft constraint relaxation variables, respectively. k+i represents the slack variable, which is the amount that allows soft constraints to be partially violated; N represents the prediction time domain / window length, which is the number of future time steps that MPC considers during optimization; Its constraints include: Constraint 1: ; ; Constraint 2: Actuator Physical Constraints: and rate limit ; Among them, pH min and pH max These represent the minimum and maximum allowable pH values ​​for the process, respectively; Indicates the upper limit of the rate of change of the control action; u min and u max These represent the physical upper and lower limits of each controlled quantity (pump / valve capacity, process allowable range); Based on this, the optimizer outputs the optimal control trajectory and sends the initial control input to the execution layer; it also outputs the expected reagent consumption and load prediction. S23. Based on the stable saponification rate, a fine-tuning strategy for the O / A ratio is further introduced to address the changes in interphase mass transfer efficiency caused by impurity ion fluctuations. The O / A offset is calculated based on short-cycle prediction results and implemented by adjusting the pump speed via feedforward to homogenize the interstage load. Simultaneously, its output serves as a prerequisite for optimizing subsequent washing acid dosing, ensuring the overall continuity of the control chain. Specifically: Recursive least squares (RLS) is used to recursively identify A and B, i.e., RLS updates are performed on the model parameters: ; in, This represents the model parameter vector, used to describe the set of parameters of dynamic models A, B, and E; K represents the updated model parameter vector; t This represents the gain matrix during parameter updates. This represents the transpose of the regression vector (past state and control); Design a Disturbance Observer (DOB) to identify d k The mutation, when the observer detects When the set threshold is reached, one of the following actions will be triggered: temporarily increase w1 (the target weight of the impurities) and adopt a more conservative reagent dosage to ensure that the effluent meets the standards; or reduce the upper limit of the action rate. To prevent interface crashes caused by rapid actions; S24. Output the aforementioned impurity index, saponification rate status, and O / A adjustment range in a unified manner. By inverting the ratio of actual impurity removal efficiency to acid consumption efficiency in the P204 system, dynamically determine the effective washing amount under minimum acid consumption. Rely on trend tracking to avoid over-washing, and issue adjustment instructions in advance before the outlet impurities suddenly rise, so as to keep the overall operation of the impurity removal section stable and reduce the separation pressure in the later stage. That is, package the first step control action optimized in step S22 and the model correction in step S23 together to form an execution instruction. After execution, verify the execution effect through short-term backtesting. If the execution deviation exceeds the tolerance, trigger a rapid rollback or manual prompt. At the same time, generate operation suggestions and economic assessment (for operator decision-making), and add the execution data to the online learning library for subsequent offline / online weight fine-tuning. The optimal control trajectory and predicted trajectory are sent to the execution layer (pump speed, dosing rate, O / A ratio settings), and an execution confirmation window T is set. confirm Monitor the instantaneous response of the state vector within this window; Calculate the execution residual index e: ; If e > e max(i.e., implement residual tolerance), automatically revert to a conservative control strategy and generate manual prompts (including possible causes and recommended actions); in, This represents the predicted value of key output quantities (such as outlet impurity concentration, outlet pH, interface index, etc.) for the next time t+1 based on the prediction model (the prediction part of MPC) at time t. It is the output prediction calculated by the prediction model and obtained by the observation model. This represents the actual output measured at t+1 (the measured value of the online sensor / analyzer).

[0022] S3, the intelligent collaborative optimization control of the separation section (P507 / C272) integrates enhanced sensing data and the output of the impurity removal section to achieve dynamic adjustment of pH, saponifying agent dosage, O / A ratio, and interface state. It employs a combination of predictive control, feedforward correction, adaptive parameter updates, and closed-loop execution to improve cobalt-nickel selectivity and product purity, while reducing reagent consumption and operational risks. This end-to-end collaborative optimization control is implemented as follows: S31. Real-time separation status assessment and key index calculation: Calculate the cobalt-nickel selectivity coefficient, saponification degree, and load index to provide accurate real-time status input for subsequent MPC control, i.e., combining the multi-source sensing vector from step S1. Based on the optimal cleanup action output in step S2, calculate the current key state indicators of the separation segment, including: Cobalt-nickel selectivity coefficient : ; Degree of saponification of organic phase S org,t : ; And obtain the safety indicators of the clarification chamber interface. and organic phase health This is used to determine whether rapid intervention is triggered. Combined with the current load index L output in step S2 t Generate the load index of the separation section : ; in, This indicates the selectivity ratio of cobalt to nickel in the organic phase; This indicates the concentration of cobalt ions measured in the organic phase; Indicates the concentration of nickel ions in the organic phase; It represents the degree of saponification of the organic phase, which is the ratio of the pre-saponification concentration (concentration of saponifying component / saponifying agent in the carrier) to the optimal saponification concentration in the organic phase. This indicates the equivalent concentration of the saponifying agent (or saponified product) measured online. This indicates the optimal saponification concentration (nominal value) determined by the process, used for normalization; Indicates the load index of the separation section; S32. Execute the feedforward-feedback coordinated MPC control program, using a predictive model combined with the residuals from the impurity removal section for feedforward correction. By optimizing the objective function, selectivity, saponification degree, and reagent consumption are simultaneously controlled to achieve high-precision dynamic adjustment of the separation section. That is, by using multivariate predictive control (MPC) combined with a feedforward correction mechanism, fine adjustment of the pH, saponifying agent dosage, and O / A ratio in the separation section is achieved. Building a predictive model: ; in, The dynamic response matrix of the separation section (P507 / C272) is represented, which describes the intrinsic dynamics of the changes in states such as cobalt-nickel selectivity, saponification degree, and interface parameters as the product changes. This represents the control input matrix for the separation section, describing the effects of adjustments such as pH, saponifying agent dosage, and O / A ratio on the state of the separation section. This represents the disturbance matrix, which describes the impact of external disturbances from feed fluctuations and the output of the impurity removal section on the state of the separation section. Represents a state vector, a set of key process state variables used for prediction; This represents the disturbance vector, which is an external disturbance / unmodeled input, such as fluctuations in feed composition, residuals from the upstream impurity removal section, etc. This represents the control vector, which is the set of control actions defined in the MPC. This represents the predicted state vector; Adjust the initial action using the residual e from the output of the noise removal section in step S2: ; Constructing the MPC optimization objective function : ; in, This indicates the control action after feedforward correction, within the original optimized trajectory. Add feedforward correction to the existing action; K ff This represents the feedforward gain matrix, which is the gain matrix that converts upstream residuals or load anomalies into control action corrections. This indicates the desired selective target value (process / product quality requirement), such as the minimum ratio that needs to be achieved to ensure cobalt purity; The weighting coefficients of the selectivity bias term determine how much importance the optimizer attaches to the β bias. This represents the normalized index of predicted organic phase saponification degree; This represents the weighting coefficient for the saponification deviation term; This indicates the predicted reagent consumption. The weighting coefficient (economic weight) of reagent consumption items is used to balance quality and cost; This represents the control increment vector; This indicates the weight of the incremental penalty control; S33. The predictive model parameters are updated online using the recursive least squares method, and dynamically corrected by combining the interface and organic phase health indicators. When anomalies occur, the control actions are automatically adjusted to ensure interface stability and excellent selectivity, achieving cross-segment synergistic optimization. Specifically: Recursive least squares (RLS) is used to predict the model parameters. , and Online identification is used to address fluctuations in feed volume and changes in the output of the impurity removal section; If the health of the interface or organic phase declines: Prioritize adjusting the O / A ratio or the amount of saponifying agent added to protect interfacial stability; Generate correction actions at edge computing nodes: ; in, This represents the control action adjustment vector calculated based on the amount of interface degradation (correction for increasing / decreasing saponifier, O / A ratio, or pump speed). This represents the interface-driven correction gain matrix or vector, which maps interface deviations to control correction values. This represents the interface safety threshold (minimum acceptable interface score). If the score is below this value, the interface is considered to be in a warning / dangerous state. This indicates a function that takes the positive part only. A positive correction is generated when the interface value is below the threshold; otherwise, the value is 0 (no trigger). Once an abnormal event is detected, conservative mode is triggered, limiting the rate of action. To ensure interface stability; S34, Optimize the action The data is distributed to the pump and valve execution layer to monitor the deviation between the execution effect and the predicted state in real time, triggering a rollback or prompting the operator. Combined with economic indicators, the data is used to calculate and inform the operator's decision-making. Simultaneously, real-time data is fed back to the state assessment and model update, achieving closed-loop control, dynamic correction, and predictive early warning. Specifically: The optimized and corrected control actions are sent down to the pump and valve execution layer (acid and alkali pump, saponifying agent pump, O / A ratio regulating pump). Real-time monitoring of the deviation between execution results and predicted status. If the threshold is exceeded, the following actions will be triggered: a short-term conservative action will be performed to revert to the previous state; an operator prompt will be given (e.g., "P507 Level 2 interface thickness has reached the warning value, cleanup is recommended"); and the feedforward gain K will be automatically fine-tuned. ff ; Simultaneously calculate economic indicators: ; in, It represents the residual measure after the separation segment is executed in a short-term closed loop, indicating the magnitude of the difference between the predicted state and the actual measured state; This represents the predicted value based on the controller's action (i.e., the expected system state vector at the next moment). This indicates that after the action is executed, a confirmation window T will appear. confirm The state vector obtained from actual internal measurements (filtered / verified); This represents the cumulative projected cost over the forecast period T, used to assess the economic consequences of the current control strategy (reagent cost + back-end load cost). This represents the load penalty factor, which is the load index of the separation section. The weighting factors, mapped to equivalent costs, reflect the economic costs of back-end loads (such as future processing, heat consumption, and downtime risks).

[0023] S4. Based on the sensing data and optimization control results from steps S1 to S3, an end-to-end global material balance and cost optimization closed loop is constructed to achieve synergistic optimization of the yield, economy, and operational safety of the nickel-cobalt multi-stage extraction system. Furthermore, digital twins and predictive early warning systems support operational decision-making and simulation optimization. The specific implementation process is as follows: S41. Integrate real-time flow rate, concentration, pH, saponification degree, and interface indicators from each unit to establish a global material balance model, dynamically calculate the total metal content and unit yield, and generate a visual digital twin interface to track the distribution and flow of nickel and cobalt throughout the entire process in real time. Specifically: Real-time collection of inlet and outlet flow rates, metal concentrations, interface parameters, saponification degree, and pH values ​​of each unit (P204 for impurity removal and P507 / C272 for separation) is achieved, while integrating the impurity removal optimization actions output from step S2 with the separation section optimization actions from step S3. Establish continuous-time global material balance for each metallic element (Co, Ni): ; And calculate the unit yield: ; The digital twin interface visualizes the material flow, yield of each unit, and amount of remaining metal, providing a real-time dynamic map that shows the distribution and flow of metal in each unit. in, Represents the total amount of metal at time t; and Let i and n represent the inlet and outlet flow rates of the i-th unit, respectively. and N represents the metal concentration at the inlet and outlet of the i-th unit; units This refers to all units in the nickel-cobalt multi-stage extraction system (e.g., each unit in the impurity removal section P204 and the separation section P507 / C272). This represents the total amount of metal at time t+1; This indicates the unit yield, which is the efficiency of metal recovery per unit. S42. Collect the flow rates of acid / alkali and saponifying agent in each section and the unit metal consumption, construct an economic digital twin model, and integrate the reagent cost, separation section load, and interface risk in a weighted manner to generate a real-time cost curve and a visualization interface. This allows operators to simulate economic changes under different control strategies. Specifically: Collect the flow rates of acids, alkalis, and saponifying agents at each stage, and obtain the unit consumption of metal reagents: ; Building an economic digital twin model: ; Where, N reagents This refers to all reagent units included in the cost, such as NaOH, NH3, acid, etc.; N interfaces This represents all key phase interfaces monitored in the system, such as the interfaces of each clarification chamber and extraction section; This represents the flow rate or consumption of the i-th reagent at time t; P represents the amount of metallic reagent consumed per unit. i Indicates the unit price of the i-th reagent; Indicates the load index of the separation section; Indicates the change in interface metrics; and This represents the weighting coefficient, balancing reagent cost, separation section load, and interface risk in the overall C... t The weights in C can be adjusted according to the operation priority; t A comprehensive cost indicator representing time t; Based on this, a unit metal cost curve, a total cost trend chart, and the consumption percentage of each unit are generated; S43. Based on material balance and cost digital twins, establish a global optimization objective function, simultaneously considering yield, cost, changes in control actions, and saponification degree constraints. Utilize hybrid optimization or reinforcement learning to generate cross-segment collaborative optimal control strategies, and provide visualization simulation and strategy distribution interfaces. Specifically: Establish a global optimization objective function J global : ; Where, N org Indicates the number of sections in the organic phase flow path; This represents the global optimization objective function; Indicates the overall recovery rate throughout the entire process; Indicates the change in control action for each segment; This indicates the change in the degree of saponification of the organic phase; , , and Indicates the optimization weights; Using a reinforcement learning (RL) model, control parameters (pH, O / A ratio, saponifying agent dosage) for each unit are generated, while simultaneously satisfying interface stability, degree of saponification, and safety constraints. This achieves global yield-economy-operational co-optimization and outputs the optimal control action. To the execution layer; S44. The global optimization strategy is distributed to pumps, valves, and O / A regulating pumps for execution. Execution deviations are collected in real time, predictive warnings and operational suggestions are generated, and hypothetical strategy simulation and report generation are supported. All data is recorded in the database to form a continuous optimization closed loop, achieving dynamic and coordinated control of yield, economy, and safety. Specifically: global optimization strategy The data is sent to pumps, valves, and O / A ratio regulating pumps to collect execution deviations in real time. If the expected overall yield decline exceeds the threshold, the operator will be notified in advance to adjust the strategy. If reagent consumption is abnormal or the risk of interface indicators increases, the feedforward gain will be automatically fine-tuned or a conservative backoff mode will be triggered.

[0024] S5. Based on the global optimization strategy and material, reagent, and interface data output in step S4, intelligent execution, closed-loop feedback, predictive maintenance warning, and economic-operational coordinated adjustment of the nickel-cobalt multi-stage extraction system are achieved, forming an end-to-end intelligent control and continuous optimization closed loop. This improves product yield, reduces reagent consumption, and ensures operational safety. The specific implementation process is as follows: S51. The global optimization strategy is decomposed into specific execution instructions such as pump valves, O / A ratio, and saponifying agent addition. Deviations are corrected through real-time measurement data in a closed loop, prioritizing interface clarity and saponification stability to achieve end-to-end precise execution and safety constraints. Specifically: The global control strategy generated in step S4 It is decomposed into execution instructions for each unit. Each instruction corresponds to specific pump valve opening, O / A ratio adjustment amount, saponifying agent dosing acceleration rate, etc., which can directly control hardware execution. Real-time acquisition of execution data (flow rate, concentration, pH, interface clarity) and calculation of execution deviation. If the deviation exceeds the allowable range, automatic fine-tuning of pump valve opening and control gain is performed to achieve closed-loop correction. Prioritize ensuring the optimal stability and saponification of the P507 / C272 interface in the separation section, and automatically limit the flow rate increase or adjust the O / A ratio based on the interface clarity; S52. Real-time monitoring of key equipment and clarification chamber interface indicators; establishment of a risk prediction model based on flow rate, saponification degree, and reagent data; early warning of potential faults or interface problems; provision of operational suggestions; and realization of a cross-unit, global safety maintenance closed loop, specifically: Real-time monitoring of key interface nodes such as pumps, valves, clarification chambers and extraction sections is performed, and image recognition models are used to analyze interface fouling thickness, three-phase formation and organic phase health. A trend analysis and prediction model is constructed. Based on real-time collected data such as interface clarity, saponification degree, organic phase health, pump flow rate and reagent dosage, the probability of risk in future time periods is calculated using historical data and reinforcement learning algorithms. The model outputs potential abnormal indicators of devices or interfaces. When the predicted risk exceeds the threshold, maintenance or adjustment suggestions are automatically generated to achieve predictive early warning and dynamic operation optimization. S53. Based on actual execution data and early warning information, calculate economic indicators, automatically fine-tune pump speed, O / A ratio, and saponifying agent dosage to ensure synergistic optimization of yield, cost, and interface safety. Simultaneously, it can simulate the effects of different strategies, providing decision support for operators. Specifically: Using the digital twin from step S4, combined with the execution deviation e t Based on the early warning information, calculate the unit metal cost and overall economic indicators under the current operation: ; If the cost or yield deviates from expectations, the system will automatically adjust the O / A ratio, pump speed, and saponifying agent dosage; prioritize maintaining interface health and saponification degree, while minimizing reagent consumption and energy consumption. in, This represents the unit metal cost and overall economic indicators at the current time t; This includes, but is not limited to, the indirect costs of operating the separation section (including energy consumption, pump and valve operating load, and potential maintenance costs); This represents the total economic cost of reagent consumption; This represents the ratio of the actual consumption of the i-th reagent at time t to the metal production. This indicates the actual load index of the separation section, which can comprehensively represent the changes in pump speed, flow rate, O / A ratio, and interface pressure or saponification degree; P i This represents the unit price of the i-th reagent; S54 records execution actions, early warning events, and economic indicators, uses historical data to train reinforcement learning models, achieves cross-batch strategy iterative optimization, improves system adaptability and intelligence, and provides self-learning closed-loop optimization capabilities for future production.

[0025] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A smart collaborative optimization control system for multi-extraction systems aimed at efficient separation of nickel and cobalt, characterized in that, include: The enhanced sensing layer module performs real-time dynamic monitoring of aqueous phase ion concentration, phase interface state, and organic phase health in the multi-stage extraction system, generating multi-source fusion data. The impurity removal section dynamic optimization control module, based on multi-source fusion data, performs model prediction control on the impurity removal section, and dynamically optimizes the saponification rate, the ratio of organic phase to water and the amount of washing acid, so as to achieve efficient removal of impurities and minimize reagent consumption. The intelligent collaborative optimization module for the separation section performs collaborative optimization control that combines feedforward and feedback, dynamically adjusting the pH value, saponifying agent dosage, and the ratio of organic phase to water to improve the selectivity of cobalt-nickel separation and product purity. The global material balance and cost optimization module integrates data from the entire process, builds an economic digital twin model, and conducts comprehensive analysis and global optimization decisions on metal yield and production costs. The intelligent execution and predictive maintenance early warning module transforms optimization strategies into control commands and executes them, while providing anomaly warnings and maintenance prompts based on trend analysis and predictive models.

2. The intelligent collaborative optimization control system for multi-extraction systems for efficient separation of nickel and cobalt according to claim 1, characterized in that, The process of real-time dynamic monitoring by the enhanced perception layer module specifically includes: High-precision pH sensors and online ion chromatographs or miniature inductively coupled plasma spectrometers are installed at the inlet and outlet of the extractant in the separation section to sample at high frequency and measure the concentration of cobalt ions and nickel ions in the aqueous phase in real time, and calculate the cobalt-nickel separation selectivity index. High-definition industrial cameras are installed in the clarification chambers of the impurity removal and separation sections. Convolutional neural network image recognition models are used to analyze the clarity of the phase interface, the thickness of interface contaminants, and the generation of three phases in real time, and to quantify the interface state indicators. An online near-infrared spectrometer is installed in the separation section to monitor the concentration of the organic phase extractant and the saponification rate in real time, and to calculate the health indicators of the organic phase.

3. The intelligent collaborative optimization control system for multi-extraction systems for efficient separation of nickel and cobalt according to claim 2, characterized in that, The monitored cobalt ion concentration, nickel ion concentration, separation selectivity index, interface state index, organic phase health index, and real-time aqueous phase pH value are time-series aligned, noise filtered, and feature extracted to generate a unified multi-source data vector for use in the model prediction and control model.

4. The intelligent collaborative optimization control system for multi-extraction systems for efficient separation of nickel and cobalt according to claim 3, characterized in that, The dynamic optimization process performed by the impurity removal section dynamic optimization control module specifically includes: Based on online ion chromatography and micro inductively coupled plasma spectrometry, the instantaneous load of inlet and outlet impurity ions is quantified in real time, and a comprehensive impurity index and load index are constructed by combining phase boundary image features. By utilizing the changing trends of impurity index and load index, a finite time-domain model predictive control optimizer with soft constraints is used to drive adaptive adjustment of saponification rate. The objective function of the optimizer simultaneously considers the impurity removal objective, the minimization of reagent consumption, and the economic penalty for the load on the separation section. In model predictive control, the recursive least squares method is used to identify the parameters of the predictive model online in order to cope with the fluctuations in feed composition.

5. The intelligent collaborative optimization control system for multi-extraction systems for efficient separation of nickel and cobalt according to claim 4, characterized in that, Based on the dynamic optimization of saponification rate, a fine-tuning strategy for the ratio of organic phase to water is further introduced. The offset of the ratio of organic phase to water is calculated based on the short-cycle prediction results and implemented by adjusting the pump speed through feedforward to make the interstage load uniform.

6. The intelligent collaborative optimization control system for multi-extraction systems for efficient separation of nickel and cobalt according to claim 5, characterized in that, The process of collaborative optimization control performed by the intelligent collaborative optimization module for the separation section specifically includes: Real-time calculation of cobalt-nickel selectivity coefficient, organic phase saponification degree, and separation section load index; A multivariate model predictive control combined with a feedforward correction mechanism is adopted. The predictive model is used and the feedforward correction is performed based on the residual of the impurity removal section output. By optimizing the objective function, the cobalt-nickel selectivity, organic phase saponification degree and reagent consumption are precisely controlled at the same time. The parameters of the separation segment prediction model are updated online using the recursive least squares method, and the control action is automatically adjusted to prioritize interface stability when the interface status index decreases.

7. The intelligent collaborative optimization control system for multi-extraction systems for efficient separation of nickel and cobalt according to claim 6, characterized in that, When the interface status index is lower than the safety threshold, control action corrections are generated at the edge computing node. Priority is given to adjusting the ratio of the organic phase to water or the amount of saponifying agent added, and the rate of change of control actions is limited to ensure interface stability.

8. The intelligent collaborative optimization control system for multi-extraction systems for efficient separation of nickel and cobalt according to claim 7, characterized in that, The process of comprehensive analysis and global optimization decision-making regarding metal yield and production costs by the global material balance and cost optimization module specifically includes: By integrating the real-time inlet and outlet flow rates, metal concentrations, pH values, saponification degrees, and interface indicators of each unit, a global material balance model is established to dynamically calculate the total metal content and yield of each unit, and then visualize it on a digital twin interface. Collect the flow rates of acids, alkalis, and saponifying agents in each section and the unit metal consumption, construct an economic digital twin model, and integrate the reagent costs, separation section load, and interface risks in a weighted manner to generate a real-time cost curve and a visualization interface. A global optimization objective function is established that simultaneously considers the overall recovery rate, comprehensive cost index, control action variation, and organic phase saponification degree constraint. A reinforcement learning model is then used to generate a cross-segment collaborative optimal control strategy.

9. The intelligent collaborative optimization control system for multi-extraction systems for efficient separation of nickel and cobalt according to claim 8, characterized in that, The intelligent execution and predictive maintenance early warning module translates optimization strategies into control commands and executes them. Simultaneously, based on trend analysis and predictive models, it provides anomaly warnings and maintenance alerts. The specific steps include: The global optimization control strategy is decomposed into specific pump and valve opening, organic phase to water phase flow ratio setpoint, and saponifier dosing acceleration command, and then sent to the execution layer. Within the defined execution confirmation window, the system status response after execution is monitored in real time, and the execution residual is calculated. When the execution residual exceeds the maximum tolerance, it automatically reverts to a conservative control strategy and generates a manual prompt message.

10. The intelligent collaborative optimization control system for multi-extraction systems for efficient separation of nickel and cobalt according to claim 9, characterized in that, The intelligent execution and predictive maintenance early warning module translates optimization strategies into control commands and executes them. Furthermore, the steps for providing anomaly warnings and maintenance alerts based on trend analysis and predictive models include: Real-time monitoring of clarification chamber interface indicators and key equipment operating parameters; Based on real-time collected data on interface clarity, saponification degree, organic phase health, and pump flow rate, the probability of equipment failure and interface anomaly risks in the future time period is calculated using trend analysis and prediction models. When the predicted risk exceeds the set threshold, maintenance or operation adjustment suggestions are automatically generated. Record historical execution actions, early warning events, and economic indicators to train reinforcement learning models and achieve iterative optimization of strategies across production batches.

Images