Method and system for precise control of concentration in an automated lye preparation process

Abstract

The present application relates to the field of automation control technology, and more particularly to a method and system for precise concentration control in an automatic lye preparation process, which collects lye preparation process parameters, constructs an evolution trajectory prediction model of lye concentration in a rolling time domain based on historical data, inputs current process parameters to predict future trajectories, constructs a multi-objective coordinated optimization problem with the dual objectives of minimizing concentration deviation integral and minimizing adjustment energy consumption, solves the Pareto frontier, selects a compensation sequence that meets the concentration convergence rate constraint and has the minimum energy consumption, and applies the solvent flow compensation amount and solute addition rate compensation amount of the first control cycle to the regulating valve and the driving device, respectively, and repeats the rolling at the next sampling time until the concentration converges. This method realizes precise control of lye concentration and reduces adjustment energy consumption.

Description

Technical Field

[0001] This invention relates to the field of automation control technology, and in particular to a method and system for precise concentration control in the automated preparation of alkali solutions. Background Technology

[0002] In the preparation of industrial alkali solutions, concentration control is a crucial step in ensuring the stability of the production process and product quality. Current conventional practices generally employ feedback control strategies based on fixed thresholds. For example, after setting a concentration deviation range, a proportional-integral-derivative (PID) controller is used to compensate for solvent flow rate or solute dosage in real time. Some systems introduce feedforward control, using material balance equations to estimate the dosage, but this relies on accurate static models. Furthermore, manual adjustment based on experience is still widely used for auxiliary parameter tuning, especially in multi-batch, multi-condition scenarios.

[0003] The aforementioned conventional methods have the following drawbacks. First, PID control and feedforward control are essentially linearization methods, making them ill-suited to the strong nonlinearity and time-delay characteristics of alkaline reaction processes—for example, solubility fluctuations caused by temperature changes and mixing inhomogeneities due to solvent flow fluctuations can cause the actual concentration response to deviate significantly from expectations. Second, existing control schemes typically treat concentration deviation as the sole optimization objective, neglecting the energy costs of the adjustment process. Frequent and excessive actions of the solute dosing drive and solvent flow regulating valve not only increase mechanical losses but also waste power and materials, accumulating a significant operational burden over long-term production. Therefore, an automated method that can simultaneously improve concentration control accuracy and reduce adjustment energy consumption is urgently needed. Summary of the Invention

[0004] The present invention provides a method and system for precise concentration control in the automated preparation of alkali solutions, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a method for precise concentration control during automated alkali solution preparation, comprising: The process parameters of the alkali solution preparation process are collected, including alkali solution temperature, alkali solution concentration, solvent flow rate and solute addition acceleration rate; based on historical process parameters and alkali solution concentration sequence, an evolution trajectory prediction model of alkali solution concentration in a preset rolling time domain is constructed; the current process parameters are input into the evolution trajectory prediction model to predict the evolution trajectory sequence of alkali solution concentration in the future rolling time domain. Within the rolling time domain, a multi-objective coordinated optimization problem is constructed with the dual optimization objectives of minimizing the concentration deviation integral and minimizing the regulation energy consumption. The concentration deviation integral is determined by the time-domain integral of the deviation between the predicted concentration and the target concentration at each moment in the evolution trajectory sequence. The regulation energy consumption is determined by the weighted sum of squares of the flow compensation increment and the acceleration rate compensation increment. The Pareto front of this optimization problem is solved, and the compensation sequence that satisfies the preset concentration convergence rate constraint and minimizes the regulation energy consumption is selected. The solvent flow rate compensation and solute addition acceleration compensation in the first control cycle of the compensation sequence are applied to the solvent flow rate regulating valve and the solute addition drive device, respectively. At the next sampling moment, the process parameters are updated and the rolling time domain is slid forward by one control cycle. The trajectory prediction and compensation sequence solution process is repeated until the measured concentration at the alkali outlet converges to the preset target concentration.

[0006] Based on historical process parameters and alkali concentration sequences, a model for predicting the evolution trajectory of alkali concentration within a preset rolling time domain is constructed, including: Temporal coupling features are extracted from the temperature, concentration, flow rate and acceleration rate at each sampling time in the historical process parameter sequence. The temporal subsequence of each parameter is truncated along the temporal direction by a sliding time window of a preset length. The normalized cross-correlation data between each pair of time subsequences of all parameter pairs are calculated. The cross-correlation data of all parameter pairs are combined into a multi-parameter coupling feature vector at that sampling time. The multi-parameter coupling feature vectors at each sampling time are aligned with the corresponding alkaline concentration change rate obtained by the difference of the alkaline concentration sequence in time sequence. A state space embedding model is constructed with the multi-parameter coupling feature vectors as input and the alkaline concentration change rate as output. The nonlinear coupling effect between the multi-parameters is encoded as the transition driving term of the concentration evolution trajectory in the current state through the state transition matrix, so that the state transition direction is related to the intensity distribution of the multi-parameter coupling feature vectors. Using the state-space embedding model as the core structure of the evolution trajectory prediction model, the evolution trajectory prediction model starts from the initial state corresponding to the current process parameters, and recursively calculates the state transition driving terms at each moment along the rolling time domain for each control cycle and updates the concentration state to obtain the evolution trajectory sequence. Each state transition implicitly represents the coupled driving relationship between temperature, flow rate and dosing acceleration rate on alkali concentration.

[0007] The evolution trajectory prediction model starts from the initial state corresponding to the current process parameters, recursively calculates the state transition driving terms at each time point along the rolling time domain for each control cycle, and updates the concentration state to obtain the evolution trajectory sequence, including: The process parameters at the current sampling time are converted into the current multi-parameter coupling feature vector. In the multi-parameter coupling feature vector space of the historical process parameter sequence, all historical operating points whose normalized distance to the current multi-parameter coupling feature vector is less than the preset neighborhood radius are retrieved. The group of historical operating points with the closest distance is selected in ascending order of distance, and the subsequent concentration evolution trajectory fragments corresponding to each historical operating point are extracted. The reciprocal of the normalized distance between the current multi-parameter coupled feature vector and the feature vectors of each retrieved historical working point is used as the fusion weight of each trajectory segment. After aligning the concentration evolution trajectory segments corresponding to each historical working point according to the time axis, the fusion is performed by weighted fusion using the fusion weight to generate an initial evolution trajectory sequence. After obtaining the online measured value of the alkali concentration in each control cycle, the residual between the predicted concentration corresponding to the cycle in the initial evolution trajectory sequence and the online measured value is calculated. The residual is used as a feedback correction term, which is decayed and propagated along the remaining cycles that have not yet been executed in the rolling time domain to correct the subsequent predicted concentration, thereby obtaining the online corrected evolution trajectory sequence. The initial evolution trajectory sequence is replaced by the online corrected evolution trajectory sequence.

[0008] Within the rolling time domain, a multi-objective coordinated optimization problem is constructed with minimizing the integral of concentration deviation and minimizing regulation energy consumption as dual optimization objectives, including: For the deviation between the predicted concentration and the preset target concentration at each moment in the evolution trajectory sequence, a weighted window function that monotonically decreases from near to far along the rolling time domain is applied. The concentration deviation integral objective function is constructed by discretely summing the weighted deviation along the rolling time domain. The weighted window function assigns the highest weight to the deviation closest to the current sampling moment and a lower weight to the deviation far from the current sampling moment. The solvent flow rate compensation increment and solute addition acceleration rate compensation increment to be solved in each control cycle within the rolling time domain are divided by the corresponding preset normalization coefficients for dimension normalization. The weighted sum of squares of the normalized flow rate compensation increment and the addition acceleration rate compensation increment is accumulated cycle by cycle along the rolling time domain to construct the energy consumption adjustment objective function. The concentration deviation integral objective function and the adjustment energy consumption objective function together constitute the multi-objective coordinated optimization problem. The decision variables of the multi-objective coordinated optimization problem are the solvent flow compensation amount and the solute addition acceleration rate compensation amount in each control cycle within the rolling time domain. The constraints of the multi-objective coordinated optimization problem include the physical stroke range constraint of the solvent flow regulating valve and the rotational speed range constraint of the solute addition drive device.

[0009] Solve the Pareto front of this optimization problem, and select the compensation sequence that satisfies the preset concentration convergence rate constraint and minimizes adjustment energy consumption, including: Using the concentration deviation integral objective function value and the adjustment energy consumption objective function value as the first coordinate axis and the second coordinate axis respectively, the Pareto front of the multi-objective coordinated optimization problem is solved in the objective space. The Pareto front consists of a set of mutually non-dominated solutions, and each non-dominated solution corresponds to a complete set of solvent flow rate compensation and solute addition acceleration rate compensation. On the Pareto front, the concentration convergence rate corresponding to each nondominated solution is calculated. The concentration convergence rate is characterized by the rate of decrease of the deviation between the predicted concentration at the end of the evolution trajectory sequence corresponding to the nondominated solution and the preset target concentration. Nondominated solutions with a concentration convergence rate lower than the preset concentration convergence rate constraint are screened out, and a subset of candidate nondominated solutions that satisfy the constraint are retained. In the candidate non-dominated solution subset, the non-dominated solution with the smallest adjustment energy consumption objective function value is selected. The solvent flow rate compensation and solute addition acceleration rate compensation for each control cycle corresponding to the non-dominated solution are arranged sequentially according to the control cycle order to form an initial compensation sequence. The variation range of solvent flow rate compensation and solute addition acceleration rate compensation during adjacent cycles in the initial compensation sequence are respectively subjected to smoothing constraint processing to ensure that the rate of change of compensation along the control cycle does not exceed the preset upper limit of the dynamic response rate of the adjustment mechanism. The resulting sequence after smoothing is used as the compensation sequence.

[0010] The solvent flow rate compensation and solute dosing acceleration compensation in the first control cycle of the compensation sequence are applied to the solvent flow rate regulating valve and the solute dosing drive device, respectively, including: The measured valve opening of the current solvent flow regulating valve is obtained, and the solvent flow compensation amount of the first control cycle is superimposed on the measured valve opening to obtain the target valve opening. According to the preset valve position-flow characteristic curve of the solvent flow regulating valve, the target valve opening is converted into the corresponding regulating valve drive electrical signal, and the regulating valve drive electrical signal is output to the drive end of the solvent flow regulating valve. The measured rotation speed of the current solute dosing drive device is obtained, the solute dosing acceleration compensation amount of the first control cycle is converted into the corresponding rotation speed increment, the rotation speed increment is superimposed on the measured rotation speed to obtain the target rotation speed, and the target rotation speed is converted into the corresponding drive device rotation speed control signal according to the preset drive device rotation speed-dosing acceleration characteristic curve and output to the control terminal of the solute dosing drive device. When outputting the control valve drive electrical signal and the drive device speed control signal, the start time of the same control cycle is used as the synchronization reference time. An advance trigger offset matching the response lag time of each actuator is applied to the two signals respectively, so that the solvent flow rate adjustment action and the solute addition acceleration rate adjustment action are executed synchronously in time, avoiding the introduction of additional parameter coupling disturbances due to the difference in adjustment timing.

[0011] After synchronizing the solvent flow rate adjustment and solute dosing rate adjustment in time, the process also includes: Within the preset delay monitoring window after compensation execution, the measured concentration sequence of alkaline solution outlet is continuously collected. The measured concentration sequence is subjected to time-series difference calculation along the time axis to obtain the measured concentration change trend vector. The expected concentration sequence within the time period corresponding to the preset delay monitoring window is extracted from the evolution trajectory sequence. The expected concentration sequence is subjected to time-series difference calculation along the time axis to obtain the expected concentration change trend vector. The angle between the two change trend vectors is calculated as the trend deviation angle. If the absolute value of the trend deviation angle is less than the preset deviation tolerance limit, the current compensation sequence is determined to be valid, the compensation amount of the subsequent control cycles in the compensation sequence remains unchanged, and the compensation amount of each subsequent cycle is executed in the order of the control cycles. If the absolute value of the trend deviation angle is not less than the preset deviation tolerance upper limit, it is determined that the current compensation action is interfered with by unmodeled coupling factors. The execution of the compensation amount for the remaining control cycles in the compensation sequence is paused. The current input parameters are replaced with the latest collected process parameters and measured concentrations. The trajectory deduction of the evolution trajectory prediction model and the solution process of the multi-objective coordinated optimization problem are retried. The current compensation sequence is completely replaced with the newly generated compensation sequence, and execution restarts from the first control cycle of the new sequence.

[0012] A second aspect of the present invention provides a concentration precision control system for an automated alkali solution preparation process, comprising: The parameter acquisition unit is used to acquire process parameters of the alkali solution preparation process, including alkali solution temperature, alkali solution concentration, solvent flow rate and solute addition rate. The model building unit is used to construct a model for predicting the evolution trajectory of alkali concentration within a preset rolling time domain based on historical process parameters and alkali concentration sequences. The trajectory prediction unit is used to input the current process parameters into the evolution trajectory prediction model to predict the evolution trajectory sequence of the alkali concentration in the future rolling time domain. An optimization construction unit is used to construct a multi-objective coordinated optimization problem in the rolling time domain with the dual optimization objectives of minimizing the concentration deviation integral and minimizing the regulation energy consumption. The concentration deviation integral is determined by the time domain integral of the deviation between the predicted concentration and the target concentration at each moment in the evolution trajectory sequence. The regulation energy consumption is determined by the weighted sum of squares of the flow compensation increment and the acceleration rate compensation increment. The frontier solving unit is used to solve the Pareto front of the optimization problem and select the compensation sequence that satisfies the preset concentration convergence rate constraint and minimizes the adjustment energy consumption. The compensation execution unit is used to apply the solvent flow rate compensation amount and the solute dosing acceleration rate compensation amount of the first control cycle in the compensation sequence to the solvent flow rate regulating valve and the solute dosing drive device, respectively. The rolling update unit is used to update the process parameters at the next sampling time and slide the rolling time domain forward by one control cycle, repeating the trajectory prediction and compensation sequence solution process until the measured concentration of the alkali solution outlet converges to the preset target concentration.

[0013] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0015] The present invention significantly improves the concentration control accuracy of alkali solution preparation. Through rolling time-domain evolution trajectory prediction and multi-objective coordinated optimization, the integral deviation between the predicted concentration and the target concentration is minimized, while also considering adjustment energy consumption, ultimately achieving stable convergence of the outlet concentration to the preset value. This method effectively suppresses concentration overshoot and oscillation, reducing drastic fluctuations in solvent flow rate and solute addition rate while ensuring convergence rate. It avoids the shortcomings of traditional PID control parameters, such as difficulty in tuning and poor adaptability, and is particularly suitable for the nonlinear and large hysteresis process characteristics in alkali solution preparation.

[0016] The dual-objective optimization mechanism introduces a Pareto front solution strategy, coordinating the concentration deviation integral and regulation energy consumption as independent conflicting objectives, and automatically selecting the optimal compensation sequence that satisfies the concentration convergence rate constraint. This method significantly saves energy while improving control accuracy, reducing solute waste and solvent consumption, and extending the service life of the regulating valve and dosing device. The concentration deviation integral is calculated cumulatively along the time domain, making the control action more sensitive to long-term trends and effectively responding to disturbances and time-varying parameter fluctuations.

[0017] The tight integration of rolling time domain and model prediction enables control commands to be dynamically refreshed in each sampling period, avoiding mismatch issues caused by fixed-parameter models. This method requires no additional hardware modifications; it can be embedded into existing PLC or DCS systems simply by upgrading the algorithm. It can be widely applied in industrial scenarios requiring continuous alkali solution preparation, such as chemical, pharmaceutical, and water treatment industries. Attached Figure Description

[0018] Figure 1This is a flowchart illustrating the method for precise concentration control during the automated alkali solution preparation process according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the method for constructing an evolutionary trajectory prediction model according to an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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.

[0020] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0021] Figure 1 This is a flowchart illustrating the method for precise concentration control during the automated alkali solution preparation process according to an embodiment of the present invention. The method for precise concentration control during the automated alkali solution preparation process provided by this invention includes: The process parameters of the alkali solution preparation process are collected, including alkali solution temperature, alkali solution concentration, solvent flow rate and solute addition acceleration rate; based on historical process parameters and alkali solution concentration sequence, an evolution trajectory prediction model of alkali solution concentration in a preset rolling time domain is constructed; the current process parameters are input into the evolution trajectory prediction model to predict the evolution trajectory sequence of alkali solution concentration in the future rolling time domain. Within the rolling time domain, a multi-objective coordinated optimization problem is constructed with the dual optimization objectives of minimizing the concentration deviation integral and minimizing the regulation energy consumption. The concentration deviation integral is determined by the time domain integral of the deviation between the predicted concentration and the target concentration at each moment in the evolution trajectory sequence. The regulation energy consumption is determined by the weighted sum of squares of the flow compensation increment and the acceleration rate compensation increment. The Pareto front of the optimization problem is obtained, and the compensation sequence that satisfies the preset concentration convergence rate constraint and minimizes adjustment energy consumption is selected. The solvent flow rate compensation and solute addition acceleration rate compensation in the first control cycle of the compensation sequence are applied to the solvent flow rate regulating valve and the solute addition drive device, respectively. The process parameters are updated at the next sampling time and the rolling time domain is slid forward by one control cycle. The trajectory prediction and compensation sequence solution process is repeated until the measured concentration at the alkali outlet converges to the preset target concentration.

[0022] Figure 2This is a flowchart illustrating the method for constructing an evolution trajectory prediction model according to an embodiment of the present invention. Based on historical process parameters and alkali concentration sequences, an evolution trajectory prediction model for alkali concentration within a preset rolling time domain is constructed, including: Temporal coupling features are extracted from the temperature, concentration, flow rate and acceleration rate at each sampling time in the historical process parameter sequence. The temporal subsequence of each parameter is truncated along the temporal direction by a sliding time window of a preset length. The normalized cross-correlation data between each pair of time subsequences of all parameter pairs are calculated. The cross-correlation data of all parameter pairs are combined into a multi-parameter coupling feature vector at that sampling time. The multi-parameter coupling feature vectors at each sampling time are aligned with the corresponding alkaline concentration change rate obtained by the difference of the alkaline concentration sequence in time sequence. A state space embedding model is constructed with the multi-parameter coupling feature vectors as input and the alkaline concentration change rate as output. The nonlinear coupling effect between the multi-parameters is encoded as the transition driving term of the concentration evolution trajectory in the current state through the state transition matrix, so that the state transition direction is related to the intensity distribution of the multi-parameter coupling feature vectors. Using the state-space embedding model as the core structure of the evolution trajectory prediction model, the evolution trajectory prediction model starts from the initial state corresponding to the current process parameters, and recursively calculates the state transition driving terms at each moment along the rolling time domain for each control cycle and updates the concentration state to obtain the evolution trajectory sequence. Each state transition implicitly represents the coupled driving relationship between temperature, flow rate and dosing acceleration rate on alkali concentration.

[0023] In the actual production process of automated alkali solution preparation, there are complex dynamic coupling relationships among process parameters such as temperature, flow rate, and dosing rate. The coordinated changes of these parameters have a significant impact on the evolution trend of alkali solution concentration. In order to accurately predict the trajectory of alkali solution concentration changes in the future rolling time domain, it is necessary to mine the coupling characteristics between multiple parameters from the time series data of historical process parameters and alkali solution concentration, and establish a predictive model that can describe the dynamic evolution law of concentration.

[0024] In the temporal coupling feature extraction stage, historical process parameter data collected from the distributed control system are typically recorded with a fixed sampling period, usually 5 to 30 seconds. For each sampling moment, it is necessary to obtain the parameter values ​​at that moment and several moments prior, forming an observation window with temporal depth. Specifically, a sliding time window of a preset length is set, the length of which is determined based on the dynamic response characteristics of the preparation process, typically 20 to 50 sampling periods. Moving this sliding window forward along the time axis, the temporal subsequences of the four parameters—temperature, concentration, flow rate, and dosing acceleration—are extracted at each sampling moment. For example, when the sliding window length is 30 sampling periods, the temperature temporal subsequence at a certain sampling moment includes the temperature measurements at that moment and the 29 moments prior; the temporal subsequences of other parameters are constructed in the same way.

[0025] After obtaining the time-series subsequences of each parameter, the normalized cross-correlation quantity between the time-series subsequences of any two parameters is calculated. This quantity quantifies the correlation between the two parameters during the time-series evolution. For parameters... and parameters Given a time-series subsequence, calculate different time delays. Cross-correlation function under This function describes the parameters. The value and parameters at the current moment In delay The correlation between values ​​at later time points. The maximum value of the cross-correlation function within the delay range reflects the strongest correlation between the two parameters. Dividing this maximum value by the product of the standard deviations of the two time series subsequences yields the normalized cross-correlation quantity. The four process parameters can be paired to form six parameter pairs: temperature-concentration, temperature-flow rate, temperature-feeding acceleration rate, concentration-flow rate, concentration-feeding acceleration rate, and flow rate-feeding acceleration rate. Calculate the normalized cross-correlation data for each of these six parameter pairs. , , , , , These six values ​​are arranged in a fixed order to form the multi-parameter coupled feature vector at that sampling time. The magnitude of each component of this feature vector reflects the coupling strength of the corresponding parameter pair within the current time window. When the mutual correlation coefficient of a parameter pair is close to 1, it indicates that the two parameters exhibit strong synchronous changes during this period; when the value is close to 0, it indicates that the changes of the two parameters are relatively independent.

[0026] After extracting the multi-parameter coupled feature vectors for all sampling times, it is necessary to establish the mapping relationship between these coupled features and the rate of change of alkali concentration. The rate of change of alkali concentration is obtained by performing a first-order difference on the historical alkali concentration sequence, that is, subtracting the concentration value of the previous time from the concentration value of the current time, and then dividing by the sampling period to obtain the concentration change rate at that time. The multi-parameter coupled feature vectors at each sampling time are then used. Rate of change of alkali concentration at corresponding time point Aligned chronologically, these samples form a set of input-output pairs. These pairs reveal how the coupling states between process parameters drive the rate of change in alkali concentration.

[0027] The state-space embedding model is constructed based on dynamic systems theory, treating the alkali concentration and its rate of change as state variables of the system, and the multi-parameter coupled feature vector as the external input driving the evolution of this state. The system state vector is defined as follows. ,in The current concentration, The concentration change rate. State transition matrix. Describes the free evolution characteristics of the system without external input, input matrix Describe the driving effect of multi-parameter coupled eigenvectors on state evolution. The discrete form of the state-space model is: ,in Index of sampling time. State transition matrix. The off-diagonal elements reflect the interaction between concentration and the rate of change of concentration. (Input matrix) Each row of elements maps the six components of the multi-parameter coupled feature vector to the increments of the concentration state and the concentration change rate state, respectively.

[0028] The state transition matrix can be identified from the input-output data in the set using the least squares method or recursive least squares algorithm and historical samples. and input matrix The element values ​​are determined. The identification process aims to minimize the mean square error between the concentration change rate predicted by the model and the actual measured concentration change rate. When the historical data sample size is sufficient and covers different operating conditions, the identified matrix parameters can accurately encode the nonlinear coupling effects between multiple parameters. State transition driving term. Indicates the first At each sampling time, the multi-parameter coupled eigenvector exerts a driving effect on the system state through a linear transformation of the input matrix. The direction and strength of this driving term are determined by the cross-correlation data of each parameter pair at the current time. For example, when the cross-correlation data of temperature and flow rate increases significantly at a certain time, the corresponding driving term component will strengthen, causing the concentration change rate to shift in a certain direction.

[0029] After using the state-space embedding model as the core structure of the evolution trajectory prediction model, the prediction model starts running from the initial state corresponding to the current process parameters. The temperature, concentration, flow rate, and acceleration rate at the current sampling moment constitute the initial observation window, and the multi-parameter coupled feature vector at the current moment is extracted using the aforementioned sliding time window method. and the currently measured concentration of alkali solution The rate of change of concentration obtained from the concentration difference between the two time points. Combined into an initial state vector Starting from this initial state, the state transition drivers and state updates for each future time period are recursively calculated along the preset rolling time domain, cycle by cycle.

[0030] The length of the rolling time domain is set according to the response time and control accuracy requirements of the preparation process. A typical rolling time domain contains 5 to 15 control cycles, with the duration of each control cycle consistent with the sampling period. In the first control cycle, the initial state is known. and the initial multi-parameter coupled feature vector Through the state transition equation Calculate the state vector at the first future time step. The first component of this state vector is the predicted alkali concentration at the first future moment. To continue the recursion to the second future time, it is necessary to estimate the multi-parameter coupled feature vector of the first future time. Since the actual future process parameters have not yet been measured, a strategy of assuming the control sequence remains constant can be adopted. This means assuming the solvent flow rate and solute dosing rate maintain their current values ​​in the rolling time domain, and the temperature is extrapolated using a first-order inertial element based on the thermal inertia characteristics of the preparation equipment. Based on these assumptions about the future process parameters, the sliding time window is updated and calculated. and through Obtain the state of the second future moment. Repeat this recursive process until the entire rolling time domain is covered, to obtain the evolution trajectory sequence of the predicted concentrations at future times. ,in This represents the total number of control cycles within the rolling time domain.

[0031] In this recursive process, each state transition implicitly represents the coupled driving relationship between temperature, flow rate, and dosing acceleration on the alkali concentration. State transition driving term. At every moment The data is recalculated to reflect the real-time state of the multi-parameter coupling characteristics at that moment. When temperature fluctuations cause changes in the solute dissolution rate, the cross-correlation between temperature and the acceleration rate changes, thereby altering the values ​​of the corresponding components in the driving term and adjusting the predicted concentration change rate accordingly. Fluctuations in flow rate affect the residence time and mixing degree of the solvent in the reactor; an increase in the cross-correlation between flow rate and concentration will enhance the concentration change trend through the driving term. Adjustments to the acceleration rate directly change the amount of solute entering the system per unit time; the cross-correlation between the acceleration rate and flow rate reflects the coordination between the two, and changes in this cross-correlation adjust the magnitude of the concentration change rate through the driving term. In this way, the evolutionary trajectory prediction model can dynamically capture the comprehensive impact of multi-parameter coupling on concentration evolution in the rolling time domain, providing an accurate concentration prediction basis for subsequent multi-objective coordinated optimization.

[0032] The evolution trajectory prediction model starts from the initial state corresponding to the current process parameters, recursively calculates the state transition driving terms at each time point along the rolling time domain for each control cycle, and updates the concentration state to obtain the evolution trajectory sequence, including: The process parameters at the current sampling time are converted into the current multi-parameter coupling feature vector. In the multi-parameter coupling feature vector space of the historical process parameter sequence, all historical operating points whose normalized distance to the current multi-parameter coupling feature vector is less than the preset neighborhood radius are retrieved. The group of historical operating points with the closest distance is selected in ascending order of distance, and the subsequent concentration evolution trajectory fragments corresponding to each historical operating point are extracted. The reciprocal of the normalized distance between the current multi-parameter coupled feature vector and the feature vectors of each retrieved historical working point is used as the fusion weight of each trajectory segment. After aligning the concentration evolution trajectory segments corresponding to each historical working point according to the time axis, the fusion is performed by weighted fusion using the fusion weight to generate an initial evolution trajectory sequence. After obtaining the online measured value of the alkali concentration in each control cycle, the residual between the predicted concentration corresponding to the cycle in the initial evolution trajectory sequence and the online measured value is calculated. The residual is used as a feedback correction term, which is decayed and propagated along the remaining cycles that have not yet been executed in the rolling time domain to correct the subsequent predicted concentration, thereby obtaining the online corrected evolution trajectory sequence. The initial evolution trajectory sequence is replaced by the online corrected evolution trajectory sequence.

[0033] In the dynamic control of alkali solution preparation, accurately predicting the concentration evolution trajectory in the future rolling time domain is a prerequisite for achieving precise control. Based on the established evolution trajectory prediction model, when making multi-step predictions starting from the current process state, it is necessary to complete the feature space mapping of the current operating condition and the intelligent retrieval of similar historical operating conditions.

[0034] The process parameters measured at the current sampling time, such as alkali temperature, solvent flow rate, and solute addition acceleration rate, are used to extract coupling interaction information between parameters based on previously established cross-correlation analysis results. This information is then transformed into a multi-parameter coupled feature vector for the current moment. This feature vector contains information on the coupling strength of different parameters under different time delays, enabling the characterization of the dynamic characteristics of the current operating condition in a high-dimensional feature space. In the historical process parameter database, each historical sampling time has undergone the same feature transformation process, forming a complete set of historical operating condition feature vectors. By calculating the normalized Euclidean distance between the current feature vector and each element in the historical feature vector set, the similarity between the current operating condition and the operating conditions at various historical moments can be quantified. A neighborhood radius threshold is set. All historical operating conditions with a normalized distance less than the threshold are selected. These operating conditions represent historical moments that are highly similar to the current dynamic state. The selected historical operating conditions are then sorted in ascending order of their normalized distance from the current operating condition, and the closest one is selected. A set of historical operating conditions is created by taking 10 historical operating conditions as a similar operating condition set. For each historical operating condition point in the similar operating condition set, its corresponding time interval is extracted and then continuously... The sequence of measured alkali concentration values ​​for each control cycle serves as a fragment of the concentration evolution trajectory corresponding to that operating condition. These historical trajectory fragments accurately record the concentration evolution patterns under similar operating conditions, implying the dynamic response law of the system near the current operating condition.

[0035] To fully utilize the evolutionary information from multiple similar historical conditions, a weighted fusion strategy based on the reciprocal of distance is adopted to generate the initial predicted trajectory. For the ... There are 3 similar operating point locations, and their normalized distances to the current operating point are denoted as _ . The fusion weight of the trajectory segment corresponding to this working point is defined as follows: This ensures that the sum of all weights is 1. This weighting principle reflects the rationale that historical conditions more recent in time contribute more to current predictions. The historical trajectory segments are aligned on the timeline, for the first segment within the scrolling time domain. One control cycle ( The concentration values ​​for that period in the trajectory segments corresponding to each similar working point are weighted and summed according to the fusion weights to obtain the initial predicted concentration for that period. ,in Indicates the first The first historical trajectory segment The concentration values ​​for each period are calculated sequentially. The initial predicted concentrations for each period within the rolling time domain are then calculated to form an initial evolution trajectory sequence. This sequence integrates evolutionary patterns from multiple similar historical conditions, exhibiting stronger predictive robustness compared to a single historical trajectory.

[0036] In actual control execution, the online measured value of the alkali concentration at the end of each control cycle can be obtained. This measured value reflects the actual system response under the current control action, and inevitably deviates from the predicted value of the corresponding cycle in the initial predicted trajectory. Fully utilizing this online feedback information to correct subsequent predictions can significantly improve the accuracy of trajectory prediction. Let the... One control cycle ( Obtain online measured values ​​of alkali concentration. Calculate the prediction residual for this period. The residual contains system dynamics information that the model failed to capture, as well as the effects of random disturbances. To make reasonable use of this residual to correct subsequent predictions, a time-domain decay propagation mechanism is introduced: for the ... The first cycle after One cycle ( Define the residual decay factor. ,in This is the decay time constant, reflecting the time-domain propagation characteristics of residual information. This factor decays exponentially with increasing prediction step size, reflecting the objective law that recent residuals have a greater impact on subsequent predictions, while their impact gradually weakens in the long term. For the remaining periods in the rolling time domain that have not yet been executed, residual corrections are applied periodically. The online corrected version was obtained. Periodic concentration prediction. Corrections are made sequentially for all remaining periods, resulting in an online-corrected evolutionary trajectory sequence. .

[0037] As control execution progresses, a trajectory correction is triggered each time a new online measured value is obtained. In the... After the periodic correction is completed, the rolling time domain slides forward one cycle, discarding the first executed cycle and adding a new cycle at the end of the time domain. Cycle prediction. The predicted values ​​for the new supplementary cycle are generated through the same historical operating condition retrieval and trajectory fusion process. In the first... After obtaining new online measured values ​​for the period, calculate the new residual for that period. The aforementioned decay propagation process is repeated to correct subsequent remaining cycle predictions. This rolling correction mechanism forms a closed loop of "prediction-execution-feedback-correction," enabling the evolution trajectory prediction to continuously integrate the latest online information and adaptively track dynamic changes in the system.

[0038] In practical applications, neighborhood radius The selection of search criteria needs to balance the search scope and similarity requirements. Too small a value will result in an insufficient number of similar work condition points being retrieved, while too large a value will introduce historical trajectories with low similarity, affecting prediction accuracy. Typically, an appropriate value is determined through cross-validation based on the distribution density of the feature vector space of historical data. Value. Number of similar working conditions. The choice of quantity also requires a trade-off: too few quantities make it difficult to fully utilize historical information, while too many introduce noise. General settings The value is between 5 and 20, with the specific value determined based on the scale of historical data and the frequency of changes in operating conditions. (Decay time constant) This reflects the dynamic memory characteristics of the system. For alkali preparation systems with slower response times, the [capacity] can be appropriately increased. This allows residual correction to take effect over a longer time domain; for fast-response systems, a smaller correction is used. value.

[0039] The entire trajectory prediction process is executed once at each sampling time, providing accurate information on future concentration evolution for subsequent multi-objective coordinated optimization. The online-corrected evolution trajectory sequence serves as the basis for calculating the concentration deviation integral in the optimization problem, directly affecting the quality of optimization control decisions. By integrating historical experience from similar operating conditions with online feedback correction, this prediction method combines the flexibility of data-driven approaches with the adaptability of online correction. It can maintain high prediction accuracy under complex conditions such as changing operating conditions and model mismatch, laying a reliable predictive foundation for the precise control of alkali concentration.

[0040] Within the rolling time domain, a multi-objective coordinated optimization problem is constructed with minimizing the integral of concentration deviation and minimizing regulation energy consumption as dual optimization objectives, including: For the deviation between the predicted concentration and the preset target concentration at each moment in the evolution trajectory sequence, a weighted window function that monotonically decreases from near to far along the rolling time domain is applied. The concentration deviation integral objective function is constructed by discretely summing the weighted deviation along the rolling time domain. The weighted window function assigns the highest weight to the deviation closest to the current sampling moment and a lower weight to the deviation far from the current sampling moment. The solvent flow rate compensation increment and solute addition acceleration rate compensation increment to be solved in each control cycle within the rolling time domain are divided by the corresponding preset normalization coefficients for dimension normalization. The weighted sum of squares of the normalized flow rate compensation increment and the addition acceleration rate compensation increment is accumulated cycle by cycle along the rolling time domain to construct the energy consumption adjustment objective function. The concentration deviation integral objective function and the adjustment energy consumption objective function together constitute the multi-objective coordinated optimization problem. The decision variables of the multi-objective coordinated optimization problem are the solvent flow compensation amount and the solute addition acceleration rate compensation amount in each control cycle within the rolling time domain. The constraints of the multi-objective coordinated optimization problem include the physical stroke range constraint of the solvent flow regulating valve and the rotational speed range constraint of the solute addition drive device.

[0041] After predicting the evolution trajectory of alkali concentration, the control objective needs to be transformed into a solvable multi-objective optimization problem. The core of this optimization problem lies in simultaneously considering the accuracy of concentration control and operational economy. By rationally designing the objective function and constraints, precise and stable control of the preparation process can be achieved.

[0042] To construct the concentration deviation integral objective function, the concentration prediction values ​​for each control cycle within the rolling time domain are obtained from the evolution trajectory prediction model. It is assumed that the rolling time domain includes... Each control cycle provides a corresponding predicted concentration value. The deviation between the predicted concentration and the target concentration for each control cycle is calculated; this deviation directly reflects the accuracy of concentration control in the future. To emphasize the importance of near-term control effectiveness, a time-varying weighted window function is introduced to differentiate the weights of the deviations at each time point. This weighted window function uses an exponential decay form, and for each control cycle index... The corresponding weighting coefficient Represented as: ,in The decay rate parameter determines how quickly the weight decreases over time. When hour, This indicates that the next cycle immediately following the current moment receives the highest weight; as... Increase The gradual decrease reflects an emphasis on short-term control effectiveness. The weighted deviations of each period are accumulated along the rolling time domain to obtain the concentration deviation integral objective function. : ,in To preset the target concentration, For the first The predicted concentration value for each control period. Minimizing this objective function means minimizing the weighted cumulative amount of concentration deviation over the entire rolling time domain, ensuring that the prepared concentration approaches and remains near the target value as quickly as possible.

[0043] To construct the objective function for regulating energy consumption, it is necessary to quantify the severity of the control actions. In the preparation of alkali solution, the control actions are manifested in adjusting the opening of the solvent flow regulating valve and regulating the rotation speed of the solute dosing drive device. Define the... The solvent flow compensation increment for each control cycle is The incremental compensation for the solute addition rate is Since traffic volume and acceleration rate have different dimensions and numerical magnitudes, direct summation can lead to one variable dominating the optimization process. Therefore, a normalization process is introduced to... Divide by the preset flow normalization coefficient ,Will Divide by the preset acceleration rate normalization coefficient This brings the two to a similar numerical scale. The normalized compensation increments are respectively and Furthermore, to balance the impact of the two control actions on total energy consumption, a weighting coefficient is introduced. and These correspond to the energy consumption weights for flow rate regulation and acceleration rate regulation, respectively. The energy consumption objective function is as follows: It is obtained by accumulating the weighted sum of squares of the normalized compensation increments for each period within the rolling time domain: Minimizing the objective function prompts the optimization algorithm to minimize the magnitude of compensation actions while meeting concentration control requirements, thereby avoiding equipment wear and increased energy consumption caused by frequent and large adjustments.

[0044] Combining the two objective functions above constitutes a multi-objective coordination optimization problem. The mathematical form of this problem can be expressed as: The decision variable is the solvent flow compensation amount for each control cycle within the rolling time domain. and solute addition acceleration compensation amount .because and There is often a trade-off between the two – pursuing a rapid reduction in concentration deviation requires a large control action, while reducing the magnitude of the control action can delay concentration convergence. Therefore, this problem is a typical multi-objective optimization problem.

[0045] In optimization problems, physical constraints of the actual equipment need to be introduced as inequality constraints. For a solvent flow control valve, its opening degree is limited by the valve's mechanical stroke. Assume the minimum valve opening degree is... The maximum opening is During a certain control cycle The current flow rate setting is recorded as The actual flow rate after compensation is It must meet the following requirements: For solute dosing drive devices (such as variable frequency screw feeders), there is also a lower limit to their rotational speed. and upper limit Record the first The benchmark acceleration rate of the cycle is The actual acceleration rate after applying compensation is It must meet the following requirements: The above constraints ensure that the compensation scheme obtained by optimization is feasible within the physical capabilities of the equipment, thus avoiding control failure or equipment damage caused by exceeding the equipment limits.

[0046] In practice, a multi-objective evolutionary algorithm or a weighted method is used to solve this optimization problem. If a weighted method is used, a tradeoff coefficient can be introduced. Transform a multi-objective problem into a single-objective problem: By adjusting The value of can be flexibly balanced between concentration accuracy and energy consumption. If a multi-objective evolutionary algorithm (such as NSGA-II) is used, a set of Pareto optimal solutions can be directly obtained, with each solution corresponding to a set of compensation sequences under different trade-offs. The compensation sequence that satisfies the preset concentration convergence rate constraint and minimizes energy consumption is selected from the Pareto front as the final implementation scheme. The concentration convergence rate constraint is typically expressed as requiring the concentration deviation to decrease below a certain threshold within a specified time domain, ensuring that the preparation process reaches a steady state within an acceptable time.

[0047] Normalization coefficient and The selection is usually based on the statistical characteristics of historical operational data. For example, The flow rate adjustment range can be taken as the standard. A certain proportion, The acceleration rate adjustment range can be selected. A certain proportion causes the normalized compensation increment values ​​to be concentrated in a certain range. Near the range. Energy consumption weighting coefficient. and The setting needs to be combined with the energy consumption characteristics of the actual equipment. If the energy consumption of the flow regulating valve is large, it should be increased appropriately. If the frequent start-stop of the dosing device results in high losses, then increase... .

[0048] Using the aforementioned multi-objective coordinated optimization framework, a set of optimal compensation sequences can be obtained within each rolling time-domain window, and this sequence will be available in the future. A global plan was developed for adjusting solvent flow rate and solute addition acceleration rate within each control cycle. Based on the rolling optimization principle, only the compensation amount for the first control cycle in this sequence was calculated. and In practice, this mechanism operates on the actuator, re-collecting process parameters at the next sampling moment, updating the evolution trajectory prediction, sliding the rolling time domain forward by one cycle, and resolving the optimization problem to achieve dynamic feedback adjustment. This iterative closed-loop optimization mechanism effectively addresses model uncertainties and external disturbances, ensuring that the alkali concentration always evolves precisely along the preset trajectory and ultimately converges to the target concentration.

[0049] Solve the Pareto front of this optimization problem, and select the compensation sequence that satisfies the preset concentration convergence rate constraint and minimizes adjustment energy consumption, including: Using the concentration deviation integral objective function value and the adjustment energy consumption objective function value as the first coordinate axis and the second coordinate axis respectively, the Pareto front of the multi-objective coordinated optimization problem is solved in the objective space. The Pareto front consists of a set of mutually non-dominated solutions, and each non-dominated solution corresponds to a complete set of solvent flow rate compensation and solute addition acceleration rate compensation. On the Pareto front, the concentration convergence rate corresponding to each nondominated solution is calculated. The concentration convergence rate is characterized by the rate of decrease of the deviation between the predicted concentration at the end of the evolution trajectory sequence corresponding to the nondominated solution and the preset target concentration. Nondominated solutions with a concentration convergence rate lower than the preset concentration convergence rate constraint are screened out, and a subset of candidate nondominated solutions that satisfy the constraint are retained. In the candidate non-dominated solution subset, the non-dominated solution with the smallest adjustment energy consumption objective function value is selected, and the solvent flow rate compensation amount and solute addition acceleration rate compensation amount of each control cycle corresponding to the non-dominated solution are arranged in the order of the control cycles to form an initial compensation sequence. The changes in solvent flow rate compensation and solute addition acceleration rate compensation during adjacent cycles in the initial compensation sequence are respectively subjected to smoothing constraint processing to ensure that the rate of change of compensation along the control cycle does not exceed the preset upper limit of the dynamic response rate of the adjustment mechanism. The resulting sequence after smoothing is used as the compensation sequence.

[0050] After constructing the multi-objective coordinated optimization problem and setting the objective function, it is necessary to identify a compensation sequence from the Pareto optimal solution set that simultaneously satisfies the constraints of rapid concentration convergence and reasonable energy consumption. In the two-dimensional objective space, the horizontal axis represents the concentration deviation integral objective function value, and the vertical axis represents the adjustment energy consumption objective function value. A multi-objective evolutionary algorithm is used to solve this optimization problem. Specifically, a non-dominated sorting genetic algorithm is adopted, with a population size of 200, 150 generations, a crossover probability of 0.9, and a mutation probability of 0.1. After iterative evolution, a set of non-dominated solutions is obtained. This set of solutions forms a curve extending from the upper left to the lower right in the objective space, which is the Pareto front. Each point on this front represents a compensation strategy. Points on the left side of the front correspond to aggressive strategies with smaller concentration deviation integrals but higher adjustment energy consumption, while points on the right side of the front correspond to conservative strategies with smaller adjustment energy consumption but larger concentration deviation integrals.

[0051] After obtaining the set of non-dominated solutions on the Pareto front, the concentration convergence characteristics corresponding to each non-dominated solution need to be evaluated one by one. The solvent flow compensation for each control cycle corresponding to that non-dominated solution should be extracted. Compensation amount for solute addition rate Substituting this into the evolutionary trajectory prediction model, the predicted concentration sequence at each time point within the rolling time domain is recalculated. Particular attention is paid to the last time point... Predicted concentration for each control cycle Calculate the predicted concentration and the target concentration at that moment. deviation To quantify the concentration convergence rate, a time-domain average concentration bias is introduced. The calculation method is to sum the concentration deviations of each period in the rolling time domain and then divide by the total number of periods. Concentration convergence rate Defined as the improvement ratio of the concentration deviation at the end of the time domain to the average concentration deviation, i.e. ,in The duration of a single control cycle. If the concentration convergence rate... Below the preset convergence rate threshold If the compensation strategy corresponding to the non-dominated solution has advantages in energy consumption or deviation integration, it cannot achieve rapid concentration approximation within a specified time, and therefore it is removed from the candidate set.

[0052] After convergence rate screening, a subset of candidate non-dominated solutions that satisfy the constraints is retained. Within this subset, each candidate solution ensures that the concentration rapidly approaches the target value at the end of the time domain. At this point, further optimization from an energy consumption economy perspective is required. Each non-dominated solution in the candidate subset is traversed, and its corresponding energy consumption adjustment objective function value is read. Select The nondominated solution with the smallest value is taken as the optimal solution, and the solvent flow compensation amount for each control cycle corresponding to this solution is... Compensation amount for solute addition rate Arranged chronologically, an initial compensation sequence is formed. Theoretically, this sequence can achieve rapid concentration convergence with minimal energy consumption in the rolling time domain. However, since no explicit constraint is imposed on the rate of change of compensation between adjacent cycles during the optimization process, the compensation amount between some cycles may jump too much, exceeding the dynamic response capability of the regulating mechanism.

[0053] To ensure the feasibility of the compensation sequence in actual implementation, the initial compensation sequence needs to be smoothed and constrained. For the solvent flow rate compensation sequence, the rate of change of flow rate between adjacent periods is calculated. ,in To control the periodic index. If the absolute value of the flow rate change rate in a certain period... Exceeding the preset upper limit of the dynamic response rate of the flow control valve If the flow compensation for that period and its adjacent periods is adjusted locally, the adjustment method uses moving average filtering to adjust the flow compensation for that period and its adjacent periods. The flow compensation amount for each cycle is replaced with the weighted average of the flow compensation amounts for the current cycle and the cycles preceding and following it, for a total of three cycles, with weighting coefficients of 0.25, 0.5, and 0.25, respectively. For the first cycle and the second cycle at the time domain boundary... For each cycle, a two-point weighted average is used, with weighting coefficients of 0.4 and 0.6. After one smoothing process, the flow rate change rate between each cycle is recalculated. If there are still cases exceeding the limit, the filtering operation is repeated until the flow rate change rate between all cycles meets the requirements. .

[0054] A similar smoothing constraint is applied to the solute addition rate compensation sequence. The rate of change of addition rate between adjacent periods is calculated. This is compared with the preset upper limit of the dynamic response rate of the solute dosing drive device. Compare them. When the absolute value of the rate of change of the acceleration rate in a certain period. Exceed At that time, a three-point moving average filter is used to smooth the acceleration rate compensation for the current period and adjacent periods. The filter weight coefficients are consistent with the settings in the flow compensation smoothing process, and the boundary periods also use a two-point weighted average. The filtering operation is performed iteratively until the acceleration rate change rate between all periods meets the constraint conditions. During the smoothing process, to avoid over-filtering that weakens the control effect of the compensation sequence, the upper limit of the number of filtering iterations is set to 5. If the dynamic response rate constraint cannot be fully met after 5 iterations, the target space is moved towards the conservative direction of the Pareto front, and a suboptimal non-dominated solution is selected to reconstruct the initial compensation sequence and perform smoothing.

[0055] After smoothing constraint processing, the resulting flow rate compensation sequence and solute dosing acceleration compensation sequence are recombined to form the final compensation sequence. The compensation amounts for each control cycle in this sequence not only meet the concentration convergence rate requirements and energy minimization objectives, but also ensure smooth and continuous changes in the compensation amount over time, conforming to the physical constraints of the regulating mechanism. At each sampling moment of the rolling control, the solvent flow rate compensation amount and solute dosing acceleration compensation amount of the first control cycle in the compensation sequence are extracted and superimposed on the baseline flow rate setpoint and baseline dosing acceleration rate, respectively. The flow control valve opening command and solute dosing drive device speed command for the current cycle are calculated and sent to the actuator via fieldbus or analog signals, achieving dynamic compensation and regulation of the alkali preparation process.

[0056] At the next sampling time, the collected process parameters such as alkali temperature, alkali concentration, solvent flow rate, and solute addition acceleration rate are updated. The rolling time domain is then shifted forward by one control cycle, and the complete process of evolution trajectory prediction, multi-objective coordinated optimization, Pareto front solution, convergence rate screening, energy-optimal solution selection, and compensation sequence smoothing constraint processing is re-executed. Through the iterative rolling time domain optimization and feedback correction mechanism, the compensation sequence can be continuously optimized according to the real-time changes in the process state, maintaining the minimization of concentration deviation and the economy of energy consumption adjustment throughout the entire alkali preparation process. When the deviation between the measured concentration at the alkali outlet and the preset target concentration is less than the allowable error threshold for multiple consecutive sampling cycles, it is determined that the concentration has converged to the target value. The system exits the rolling optimization cycle and switches to a constant value maintenance control mode, restarting the rolling optimization process only when the concentration deviation is detected to exceed the threshold again.

[0057] In solving multi-objective coordinated optimization problems, the distribution of the Pareto front directly reflects the inherent trade-off between concentration control accuracy and regulation energy consumption. Through convergence rate constraint screening and energy minimization optimization, the optimal compensation strategy meeting process requirements can be automatically identified from numerous non-dominated solutions. Smoothing constraint processing further ensures the feasibility and stability of the compensation sequence in field execution, avoiding frequent operation and mechanical wear of the regulating mechanism due to drastic changes in compensation. The entire optimization process considers control performance, energy economy, and execution constraints, providing complete multi-objective decision support for precise concentration control in automated alkali preparation processes.

[0058] The solvent flow rate compensation and solute dosing acceleration compensation in the first control cycle of the compensation sequence are applied to the solvent flow rate regulating valve and the solute dosing drive device, respectively, including: The measured valve opening of the current solvent flow regulating valve is obtained, and the solvent flow compensation amount of the first control cycle is superimposed on the measured valve opening to obtain the target valve opening. According to the preset valve position-flow characteristic curve of the solvent flow regulating valve, the target valve opening is converted into the corresponding regulating valve drive electrical signal, and the regulating valve drive electrical signal is output to the drive end of the solvent flow regulating valve. The measured rotation speed of the current solute dosing drive device is obtained, the solute dosing acceleration compensation amount of the first control cycle is converted into the corresponding rotation speed increment, the rotation speed increment is superimposed on the measured rotation speed to obtain the target rotation speed, and the target rotation speed is converted into the corresponding drive device rotation speed control signal according to the preset drive device rotation speed-dosing acceleration characteristic curve and output to the control terminal of the solute dosing drive device. When outputting the control valve drive electrical signal and the drive device speed control signal, the start time of the same control cycle is used as the synchronization reference time. An advance trigger offset matching the response lag time of each actuator is applied to the two signals respectively, so that the solvent flow rate adjustment action and the solute addition acceleration rate adjustment action are executed synchronously in time, avoiding the introduction of additional parameter coupling disturbances due to the difference in adjustment timing.

[0059] After optimizing the compensation sequence, the control command corresponding to the first control cycle in the sequence needs to be accurately applied to the field actuator. The control process of the solvent flow control valve involves reading the current valve opening status and accurately setting the target opening. A smart valve position transmitter installed in the field is used to acquire the valve opening signal of the solvent flow control valve in real time. This signal is usually expressed as a percentage, denoted as [missing information]. Its value ranges from 0% to 100%. The solvent flow compensation amount for the first control cycle obtained from the optimization solution will be... The target valve position opening is obtained by directly adding the current valve position opening. The calculation formula is: , Considering the mechanical limitation on the physical opening of the control valve, boundary constraints need to be applied to the calculated target valve opening. If If the opening exceeds the valve's allowable range, it will be corrected according to the saturation limit principle to ensure that the final output target opening is within the valve's safe operating range.

[0060] The valve position-flow rate relationship of solvent flow control valves typically exhibits nonlinear characteristics, especially in the small and large opening regions, where the sensitivity of flow rate to valve position differs significantly. In engineering implementation, the valve position-flow rate characteristic curve of this control valve was established through offline calibration experiments. The calibration process involved measuring the actual flow rate output at different valve opening degrees, collecting at least 15 data points covering the full opening range to form a discrete characteristic dataset. A piecewise cubic Hermitian interpolation method was then used to fit the discrete data points, generating a continuous valve position-flow rate mapping function. ,in This represents the nonlinear mapping relationship after fitting.

[0061] The calculated target valve position opening As input, this mapping relationship determines the corresponding expected flow rate value. The actuator of the control valve typically uses standard industrial control signals, commonly in the form of 4-20mA current signals or 0-10V voltage signals. Based on the interface specification of the control valve actuator, a linear conversion relationship is established between the valve opening degree and the drive electrical signal. If the actuator uses a 4-20mA current signal, the conversion formula is: , in This is the current signal output to the control valve's drive end, measured in milliamperes (mA). This current signal is transmitted to the control valve's intelligent positioner via an industrial fieldbus or hardwired connection. The positioner drives the valve stem actuator based on the current signal value, achieving precise valve position adjustment.

[0062] The control process of the solute dosing drive device follows a similar logical architecture. The dosing drive device is typically a variable frequency drive screw pump or metering pump, whose rotational speed directly determines the solute dosing rate. The current rotational speed of the drive device is acquired in real time via a photoelectric encoder mounted on the motor shaft. The unit is revolutions per minute (rpm). The solute addition acceleration compensation amount obtained from the optimization solution for the first control cycle. It needs to be converted into the corresponding speed increment. This conversion process depends on the relationship between the rotational speed and acceleration rate of the drive unit.

[0063] For positive displacement metering pumps, the acceleration rate and rotational speed typically exhibit a linear relationship, with the conversion factor determined by the pump's displacement per revolution. Let the displacement per revolution of the metering pump be denoted as... (Unit: ml / rpm), then the conversion relationship between the speed increment and the acceleration compensation is: , The coefficient 60 is used to convert the acceleration unit from milliliters per second to milliliters per minute to match the unit system of the rotational speed. The calculated speed increment is then added to the current measured speed to obtain the target speed. Similarly, boundary constraint checks are required for the target speed to ensure it does not exceed the rated speed range of the drive unit. The speed control of the drive unit is achieved through a frequency converter, which receives standard frequency setting signals or communication protocol commands. If the frequency converter uses an analog frequency setting method, a mapping relationship between the speed and frequency setting signals needs to be established. For a feed pump driven by an asynchronous motor, there is an approximately proportional relationship between speed and frequency; the conversion formula is: , in This is the frequency setpoint (in Hertz) output to the frequency converter. This represents the number of pole pairs of the motor. If the frequency converter uses a communication protocol control method, the target speed data is encapsulated according to industrial communication protocol formats such as Modbus RTU or Profibus, and sent to the control terminal of the frequency converter through the communication interface.

[0064] In actual industrial formulation processes, there are significant differences in the dynamic response characteristics of solvent flow control valves and solute dosing actuators. Due to the mechanical movement of the valve stem and fluid dynamics involved, the response time of the control valve is typically between 2 and 5 seconds. In contrast, the speed regulation of the dosing actuator is mainly affected by the response speed of the frequency converter and the acceleration characteristics of the motor, with a typical response time between 1 and 3 seconds. If this difference in response time is not compensated for, it will lead to a timing misalignment between the adjustment actions of the solvent flow rate and the solute dosing acceleration rate, causing instantaneous concentration fluctuations during the formulation process.

[0065] To achieve timing synchronization of the adjustment actions of the two actuators, an early trigger offset mechanism is introduced. The start time of the control cycle is used as the synchronization reference time. This moment corresponds to the planned execution start point of the first control cycle in the compensation sequence. For the solvent flow control valve, the average response lag time is calculated based on its historical action data. When outputting the control valve drive electrical signal, the actual output timing is advanced to [time missing]. This ensures that the moment the regulating valve completes its action corresponds exactly to the start of the control cycle.

[0066] The solute dosing drive unit employs the same advance triggering strategy. The average time from receiving the control command to the rotational speed stabilizing to the target value is statistically analyzed. The actual output time of the drive unit speed control signal is advanced to... This differentiated early triggering setting ensures that, despite the different inherent response characteristics of the two actuators, the actual effective time of their adjustment actions can be aligned.

[0067] Accurate determination of the advance trigger offset relies on the accurate identification of the dynamic characteristics of the actuator. During the commissioning phase before system operation, step response experiments are conducted to identify the response characteristics of the control valve and the drive device. Given a step control signal of a given amplitude, the entire process curve of the actuator from receiving the command to achieving stable output is recorded, and the characteristic response time is extracted using a time constant fitting method. Considering that the actuator's response time will slowly change with load conditions and equipment aging, step response tests are periodically performed during long-term system operation to update the advance trigger offset parameters and maintain the effectiveness of timing synchronization control.

[0068] At the hardware implementation level of control signal output, a high-precision real-time clock module is used as the time reference source. This clock module has a microsecond-level time resolution, enabling precise timing of advance trigger offsets. The control system's task scheduler calculates the predetermined output time of each control signal based on the start time of the current control cycle and the advance trigger offsets of each actuator, and, with the support of the real-time operating system, triggers the signal output operation according to the calculated time.

[0069] The output of control signals also needs to consider the transmission delay of the fieldbus or communication link. If the control system and the actuator are connected via Ethernet or fieldbus, there is a millisecond-level delay in the transmission of communication messages in the network. In setting the advance trigger offset, the communication delay needs to be included as an additional component in the calculation to ensure that the timing synchronization requirements are still met when the control command arrives at the actuator control end after transmission through the communication link.

[0070] The effectiveness of timing synchronization control can be verified by online monitoring of the actual change curves of solvent flow rate and solute dosing acceleration rate. When the adjustment actions of the two actuators are correctly synchronized, the flow rate curve and the dosing acceleration rate curve should show a synchronous upward or downward trend, and the inflection points on the time axis should basically coincide. If a significant time misalignment is observed between the two curves, the advance trigger offset parameter needs to be recalibrated until a satisfactory timing alignment effect is achieved.

[0071] Through the above-mentioned complete control chain design from compensation instructions to actuator actions, it is ensured that the control strategy obtained by optimization calculation can be faithfully executed at the physical level, avoiding secondary disturbances introduced by timing mismatch or nonlinear characteristics in the execution links, and providing reliable execution guarantee for the precise control of concentration in the alkali preparation process.

[0072] After synchronizing the solvent flow rate adjustment and solute dosing rate adjustment in time, the process also includes: Within the preset delay monitoring window after compensation execution, the measured concentration sequence of alkaline solution outlet is continuously collected. The measured concentration sequence is subjected to time-series difference calculation along the time axis to obtain the measured concentration change trend vector. The expected concentration sequence within the time period corresponding to the preset delay monitoring window is extracted from the evolution trajectory sequence. The expected concentration sequence is subjected to time-series difference calculation along the time axis to obtain the expected concentration change trend vector. The angle between the two change trend vectors is calculated as the trend deviation angle. If the absolute value of the trend deviation angle is less than the preset deviation tolerance limit, the current compensation sequence is determined to be valid, the compensation amount of the subsequent control cycles in the compensation sequence remains unchanged, and the compensation amount of each subsequent cycle is executed in the order of the control cycles. If the absolute value of the trend deviation angle is not less than the preset deviation tolerance upper limit, it is determined that the current compensation action is interfered with by unmodeled coupling factors. The execution of the compensation amount for the remaining control cycles in the compensation sequence is paused. The current input parameters are replaced with the latest collected process parameters and measured concentrations. The trajectory deduction of the evolution trajectory prediction model and the solution process of the multi-objective coordinated optimization problem are retried. The current compensation sequence is completely replaced with the newly generated compensation sequence, and execution restarts from the first control cycle of the new sequence.

[0073] Once the compensation sequence begins execution, to ensure that the adjustment actions are coordinated with actual operating condition changes and to prevent the preset compensation sequence from failing due to unmodeled disturbances or sudden changes in operating conditions, the system enters a short real-time verification phase after the compensation action of the first control cycle. During this phase, an online concentration detection instrument is installed at the outlet of the alkali preparation system. This instrument uses a conductivity sensor or densitometer to continuously measure the actual concentration of the alkali. The system has a preset delay monitoring window, the duration of which is typically selected as the cumulative time of three to five control cycles to ensure that sufficient data points can be collected within the window to analyze the trend of concentration changes. Within this monitoring window, the data acquisition module continuously reads the measured concentration at the alkali outlet at a fixed sampling frequency, forming a sequence of measured concentrations arranged in chronological order. The time interval between each sampling point in this sequence is typically between several seconds and tens of seconds, depending on the response timescale and process characteristics of the preparation system.

[0074] For the acquired measured concentration sequence, time-series differencing calculations are required to extract the dynamic characteristics of concentration changes over time. Assume that a total of [number missing] samples were collected within the monitoring window. There are 1 concentration measurement points, which are recorded sequentially as follows: The corresponding sampling times are respectively Temporal difference calculations obtain the concentration change rate by dividing the concentration increment between adjacent sampling points by the time interval, forming a set of discrete rate-of-change values. Specifically, for the ... From the sampling point to the... The concentration change rate over the time interval between sampling points is Arrange the rates of change between all adjacent sampling points sequentially to form a vector representing the trend of measured concentration changes. The dimension of this vector is... Each component represents the instantaneous change trend of concentration within the corresponding time period. To reduce the impact of measurement noise on trend judgment, a low-pass filter can be applied to the measured concentration sequence before the difference calculation, or the rate of change vector can be smoothed by a moving average after the difference calculation to ensure that the extracted trend features have high stability and representativeness.

[0075] Simultaneously, it is necessary to extract the expected concentration sequence corresponding to the preset delay monitoring window from the previously estimated evolution trajectory sequence. The evolution trajectory sequence predicts the concentration for each future control cycle within the rolling time domain. After the compensation action of the first control cycle is executed, the actual time period corresponds to several predicted points in the trajectory sequence. The predicted concentration values ​​within this time range are extracted from the trajectory sequence to form the expected concentration sequence, denoted as... The time axis of this sequence is aligned with the measured concentration sequence, maintaining a one-to-one correspondence in time. The expected concentration sequence is also subjected to time-series differencing, using the same differencing method as the measured sequence, to obtain the expected concentration change rate for each time period, thus constructing the expected concentration change trend vector. This vector also has a dimension of [missing information]. Each component reflects the rate of concentration change within the corresponding time period as predicted by the ideal model. This expected trend vector essentially represents the direction and speed of concentration change that the compensation sequence was originally intended to guide.

[0076] Having obtained the measured concentration change trend vector and the expected concentration change trend vector, the angle between them is calculated to quantify the degree of consistency between the actual and expected responses. The angle between the two vectors can be obtained through the geometric relationship of the vector dot product. Let the measured trend vector be... The expected trend vector is Then the included angle between the two satisfy The inner product term is the sum of the products of corresponding components of the two vectors, and the denominator is the Euclidean norm of each vector. To facilitate judgment and threshold setting, the included angle can be calculated directly. The inverse cosine value of the angle reflects the degree of deviation between the actual and expected directions of concentration change: if the angle is close to zero, it indicates that the actual concentration change trend is highly consistent with the model prediction; if the angle is large, it means that the actual operating conditions are significantly disturbed by unmodeled factors, and the originally set compensation sequence is no longer applicable. This angle is the trend deviation angle, and its absolute value is the key basis for judging whether the compensation sequence continues to be effective.

[0077] A preset upper limit for deviation tolerance is set, which is usually determined based on process stability and control accuracy requirements. For example, when the absolute value of the trend deviation angle is less than 15 degrees, the actual response can be considered to be within a reasonable disturbance range of the expected trajectory, and the compensation sequence is still effective. At this time, the system determines that the current compensation sequence can continuously and effectively guide the alkali concentration toward the target value, maintains the compensation amount of subsequent control cycles in the compensation sequence unchanged, and gradually applies the solvent flow rate compensation increment and solute dosing rate compensation increment of the remaining cycles to the corresponding actuators according to the original planned control cycle sequence. This maintenance strategy can ensure the continuity and stability of the control strategy, avoid unnecessary repeated optimization calculations, reduce the system's computational burden and adjustment frequency, and ensure that the entire preparation process converges in a stable dynamic process.

[0078] If the absolute value of the trend deviation angle is not less than the preset deviation tolerance upper limit, it indicates that the actual concentration response deviates significantly from the expected trajectory. This deviation can be caused by various factors, such as sudden changes in raw material concentration or purity, fluctuations in solvent feed temperature, deviations in actuator response characteristics from calibration parameters, changes in mixing effect due to residues in the pipeline, or the sudden action of other coupling factors not included in the model. In this case, continuing to execute subsequent compensation amounts in the original compensation sequence will not guarantee that the concentration converges towards the target value, and may even lead to a further deviation in concentration. To address this situation, it is immediately determined that the current compensation action is interfered with by unmodeled coupling factors, and the execution of compensation amounts for the remaining control cycles in the compensation sequence is suspended. After the suspension, subsequent adjustment commands are no longer sent to the solvent flow control valve and solute dosing drive device as originally planned to avoid the cumulative effects of erroneous compensation.

[0079] After pausing the original compensation sequence, the latest acquired process parameters and measured concentrations are used as the new starting point, replacing the original input parameters. At this time, the data acquisition module re-reads the current values ​​of alkali temperature, alkali concentration, solvent flow rate, and solute dosing rate. These parameters reflect the latest operating status under the effects of compensation execution and external disturbances. Using these updated process parameters, the trajectory prediction model is re-triggered. Based on the actual process state at the current moment, the model predicts the concentration evolution trajectory for each control cycle in the future rolling time domain, generating a completely new expected concentration sequence. This sequence can more accurately reflect the concentration change trend under current operating conditions and correct the prediction bias caused by disturbances in the previous model.

[0080] After the new trajectory is generated, the solution process for the multi-objective coordinated optimization problem is restarted synchronously. Based on the newly generated concentration evolution trajectory sequence, the concentration deviation integral and regulation energy consumption are recalculated to construct a new multi-objective optimization problem. A new Pareto front is solved, and the compensation sequence that satisfies the concentration convergence rate constraint and minimizes regulation energy consumption is selected. This re-solution process is exactly the same as the initial optimization process, except that the input conditions are updated, and the optimization results can be adapted to the current actual operating conditions. The newly generated compensation sequence completely replaces the original compensation sequence. The solvent flow rate compensation and solute addition acceleration rate compensation in the first control cycle of the new sequence are immediately applied to the actuator, and execution restarts from the first control cycle of the new sequence. This replacement and restart mechanism enables adaptive adjustment of the control strategy, which can promptly correct the regulation direction when the operating conditions change abruptly or the model mismatch occurs, ensuring that the alkali concentration always evolves along the path converging to the target concentration.

[0081] The entire process forms a closed-loop adaptive control logic: after each compensation sequence begins execution, the actual concentration change trend is continuously monitored. Once the trend deviation is found to exceed the tolerance range, the compensation scheme is immediately replanned and re-executed, ensuring that the entire preparation process maintains high-precision concentration control under dynamic operating conditions. This real-time verification and dynamic adjustment mechanism significantly improves the system's robustness and anti-interference ability, effectively copes with unavoidable non-ideal factors in actual production environments, and guarantees the achievement of the goal of precise alkali concentration control.

[0082] A second aspect of the present invention provides a concentration precision control system for an automated alkali solution preparation process, comprising: The parameter acquisition unit is used to acquire process parameters of the alkali solution preparation process, including alkali solution temperature, alkali solution concentration, solvent flow rate and solute addition rate. The model building unit is used to construct a model for predicting the evolution trajectory of alkali concentration within a preset rolling time domain based on historical process parameters and alkali concentration sequences. The trajectory prediction unit is used to input the current process parameters into the evolution trajectory prediction model to predict the evolution trajectory sequence of the alkali concentration in the future rolling time domain. An optimization construction unit is used to construct a multi-objective coordinated optimization problem in the rolling time domain with the dual optimization objectives of minimizing the concentration deviation integral and minimizing the regulation energy consumption. The concentration deviation integral is determined by the time domain integral of the deviation between the predicted concentration and the target concentration at each moment in the evolution trajectory sequence. The regulation energy consumption is determined by the weighted sum of squares of the flow compensation increment and the acceleration rate compensation increment. The frontier solving unit is used to solve the Pareto front of the optimization problem and select the compensation sequence that satisfies the preset concentration convergence rate constraint and minimizes the adjustment energy consumption. The compensation execution unit is used to apply the solvent flow rate compensation amount and the solute dosing acceleration rate compensation amount of the first control cycle in the compensation sequence to the solvent flow rate regulating valve and the solute dosing drive device, respectively. The rolling update unit is used to update the process parameters at the next sampling time and slide the rolling time domain forward by one control cycle, repeating the trajectory prediction and compensation sequence solution process until the measured concentration of the alkali solution outlet converges to the preset target concentration.

[0083] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0084] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0085] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for precise concentration control during automated alkali solution preparation, characterized in that, include: Collect process parameters for the preparation of alkali solution, including alkali solution temperature, alkali solution concentration, solvent flow rate, and solute addition rate; Based on historical process parameters and alkali concentration sequences, an evolution trajectory prediction model for alkali concentration within a preset rolling time domain is constructed; the current process parameters are input into the evolution trajectory prediction model to predict the evolution trajectory sequence of alkali concentration within the future rolling time domain. Within the rolling time domain, a multi-objective coordinated optimization problem is constructed with the dual optimization objectives of minimizing the concentration deviation integral and minimizing the regulation energy consumption. The concentration deviation integral is determined by the time-domain integral of the deviation between the predicted concentration and the target concentration at each moment in the evolution trajectory sequence. The regulation energy consumption is determined by the weighted sum of squares of the flow compensation increment and the acceleration rate compensation increment. The Pareto front of this optimization problem is solved, and the compensation sequence that satisfies the preset concentration convergence rate constraint and minimizes the regulation energy consumption is selected. The solvent flow rate compensation and solute addition acceleration compensation in the first control cycle of the compensation sequence are applied to the solvent flow rate regulating valve and the solute addition drive device, respectively. At the next sampling moment, the process parameters are updated and the rolling time domain is slid forward by one control cycle. The trajectory prediction and compensation sequence solution process is repeated until the measured concentration at the alkali outlet converges to the preset target concentration.

2. The method according to claim 1, characterized in that, Based on historical process parameters and alkali concentration sequences, a model for predicting the evolution trajectory of alkali concentration within a preset rolling time domain is constructed, including: Temporal coupling features are extracted from the temperature, concentration, flow rate and acceleration rate at each sampling time in the historical process parameter sequence. The temporal subsequence of each parameter is truncated along the temporal direction by a sliding time window of a preset length. The normalized cross-correlation data between each pair of time subsequences of all parameter pairs are calculated. The cross-correlation data of all parameter pairs are combined into a multi-parameter coupling feature vector at that sampling time. The multi-parameter coupling feature vectors at each sampling time are aligned with the corresponding alkaline concentration change rate obtained by the difference of the alkaline concentration sequence in time sequence. A state space embedding model is constructed with the multi-parameter coupling feature vectors as input and the alkaline concentration change rate as output. The nonlinear coupling effect between the multi-parameters is encoded as the transition driving term of the concentration evolution trajectory in the current state through the state transition matrix, so that the state transition direction is related to the intensity distribution of the multi-parameter coupling feature vectors. Using the state-space embedding model as the core structure of the evolution trajectory prediction model, the evolution trajectory prediction model starts from the initial state corresponding to the current process parameters, and recursively calculates the state transition driving terms at each moment along the rolling time domain for each control cycle and updates the concentration state to obtain the evolution trajectory sequence. Each state transition implicitly represents the coupled driving relationship between temperature, flow rate and dosing acceleration rate on alkali concentration.

3. The method according to claim 2, characterized in that, The evolution trajectory prediction model starts from the initial state corresponding to the current process parameters, recursively calculates the state transition driving terms at each time point along the rolling time domain for each control cycle, and updates the concentration state to obtain the evolution trajectory sequence, including: The process parameters at the current sampling time are converted into the current multi-parameter coupling feature vector. In the multi-parameter coupling feature vector space of the historical process parameter sequence, all historical operating points whose normalized distance to the current multi-parameter coupling feature vector is less than the preset neighborhood radius are retrieved. The group of historical operating points with the closest distance is selected in ascending order of distance, and the subsequent concentration evolution trajectory fragments corresponding to each historical operating point are extracted. The reciprocal of the normalized distance between the current multi-parameter coupled feature vector and the feature vectors of each retrieved historical working point is used as the fusion weight of each trajectory segment. After aligning the concentration evolution trajectory segments corresponding to each historical working point according to the time axis, the fusion is performed by weighted fusion using the fusion weight to generate an initial evolution trajectory sequence. After obtaining the online measured value of the alkali concentration in each control cycle, the residual between the predicted concentration corresponding to the cycle in the initial evolution trajectory sequence and the online measured value is calculated. The residual is used as a feedback correction term, which is decayed and propagated along the remaining cycles that have not yet been executed in the rolling time domain to correct the subsequent predicted concentration, thereby obtaining the online corrected evolution trajectory sequence. The initial evolution trajectory sequence is replaced by the online corrected evolution trajectory sequence.

4. The method according to claim 1, characterized in that, Within the rolling time domain, a multi-objective coordinated optimization problem is constructed with minimizing the integral of concentration deviation and minimizing adjustment energy consumption as dual optimization objectives, including: For the deviation between the predicted concentration and the preset target concentration at each moment in the evolution trajectory sequence, a weighted window function that monotonically decreases from near to far along the rolling time domain is applied. The concentration deviation integral objective function is constructed by discretely summing the weighted deviation along the rolling time domain. The weighted window function assigns the highest weight to the deviation closest to the current sampling moment and a lower weight to the deviation far from the current sampling moment. The solvent flow rate compensation increment and solute addition acceleration rate compensation increment to be solved in each control cycle within the rolling time domain are divided by the corresponding preset normalization coefficients for dimension normalization. The weighted sum of squares of the normalized flow rate compensation increment and the addition acceleration rate compensation increment is accumulated cycle by cycle along the rolling time domain to construct the energy consumption adjustment objective function. The concentration deviation integral objective function and the adjustment energy consumption objective function together constitute the multi-objective coordinated optimization problem. The decision variables of the multi-objective coordinated optimization problem are the solvent flow compensation amount and the solute addition acceleration rate compensation amount in each control cycle within the rolling time domain. The constraints of the multi-objective coordinated optimization problem include the physical stroke range constraint of the solvent flow regulating valve and the rotational speed range constraint of the solute addition drive device.

5. The method according to claim 4, characterized in that, Solve the Pareto front of this optimization problem, and select the compensation sequence that satisfies the preset concentration convergence rate constraint and minimizes adjustment energy consumption, including: Using the concentration deviation integral objective function value and the adjustment energy consumption objective function value as the first coordinate axis and the second coordinate axis respectively, the Pareto front of the multi-objective coordinated optimization problem is solved in the objective space. The Pareto front consists of a set of mutually non-dominated solutions, and each non-dominated solution corresponds to a complete set of solvent flow rate compensation and solute addition acceleration rate compensation. On the Pareto front, the concentration convergence rate corresponding to each nondominated solution is calculated. The concentration convergence rate is characterized by the rate of decrease of the deviation between the predicted concentration at the end of the evolution trajectory sequence corresponding to the nondominated solution and the preset target concentration. Nondominated solutions with a concentration convergence rate lower than the preset concentration convergence rate constraint are screened out, and a subset of candidate nondominated solutions that satisfy the constraint are retained. In the candidate non-dominated solution subset, the non-dominated solution with the smallest adjustment energy consumption objective function value is selected. The solvent flow rate compensation and solute addition acceleration rate compensation for each control cycle corresponding to the non-dominated solution are arranged sequentially according to the control cycle order to form an initial compensation sequence. The variation range of solvent flow rate compensation and solute addition acceleration rate compensation during adjacent cycles in the initial compensation sequence are respectively subjected to smoothing constraint processing to ensure that the rate of change of compensation along the control cycle does not exceed the preset upper limit of the dynamic response rate of the adjustment mechanism. The resulting sequence after smoothing is used as the compensation sequence.

6. The method according to claim 1, characterized in that, The solvent flow rate compensation and solute dosing acceleration compensation in the first control cycle of the compensation sequence are applied to the solvent flow rate regulating valve and the solute dosing drive device, respectively, including: The measured valve opening of the current solvent flow regulating valve is obtained, and the solvent flow compensation amount of the first control cycle is superimposed on the measured valve opening to obtain the target valve opening. According to the preset valve position-flow characteristic curve of the solvent flow regulating valve, the target valve opening is converted into the corresponding regulating valve drive electrical signal, and the regulating valve drive electrical signal is output to the drive end of the solvent flow regulating valve. The measured rotation speed of the current solute dosing drive device is obtained, the solute dosing acceleration compensation amount of the first control cycle is converted into the corresponding rotation speed increment, the rotation speed increment is superimposed on the measured rotation speed to obtain the target rotation speed, and the target rotation speed is converted into the corresponding drive device rotation speed control signal according to the preset drive device rotation speed-dosing acceleration characteristic curve and output to the control terminal of the solute dosing drive device. When outputting the control valve drive electrical signal and the drive device speed control signal, the start time of the same control cycle is used as the synchronization reference time. An advance trigger offset matching the response lag time of each actuator is applied to the two signals respectively, so that the solvent flow rate adjustment action and the solute addition acceleration rate adjustment action are executed synchronously in time, avoiding the introduction of additional parameter coupling disturbances due to the difference in adjustment timing.

7. The method according to claim 6, characterized in that, After synchronizing the solvent flow rate adjustment and solute dosing rate adjustment in time, the process also includes: Within the preset delay monitoring window after compensation execution, the measured concentration sequence of alkaline solution outlet is continuously collected. The measured concentration sequence is subjected to time-series difference calculation along the time axis to obtain the measured concentration change trend vector. The expected concentration sequence within the time period corresponding to the preset delay monitoring window is extracted from the evolution trajectory sequence. The expected concentration sequence is subjected to time-series difference calculation along the time axis to obtain the expected concentration change trend vector. The angle between the two change trend vectors is calculated as the trend deviation angle. If the absolute value of the trend deviation angle is less than the preset deviation tolerance limit, the current compensation sequence is determined to be valid, the compensation amount of the subsequent control cycles in the compensation sequence remains unchanged, and the compensation amount of each subsequent cycle is executed in the order of the control cycles. If the absolute value of the trend deviation angle is not less than the preset deviation tolerance upper limit, it is determined that the current compensation action is interfered with by unmodeled coupling factors. The execution of the compensation amount for the remaining control cycles in the compensation sequence is paused. The current input parameters are replaced with the latest collected process parameters and measured concentrations. The trajectory deduction of the evolution trajectory prediction model and the solution process of the multi-objective coordinated optimization problem are retried. The current compensation sequence is completely replaced with the newly generated compensation sequence, and execution restarts from the first control cycle of the new sequence.

8. A concentration precision control system for automated alkali solution preparation, used to implement the method as described in any one of claims 1-7, characterized in that, include: The parameter acquisition unit is used to acquire process parameters of the alkali solution preparation process, including alkali solution temperature, alkali solution concentration, solvent flow rate and solute addition rate. The model building unit is used to construct a model for predicting the evolution trajectory of alkali concentration within a preset rolling time domain based on historical process parameters and alkali concentration sequences. The trajectory prediction unit is used to input the current process parameters into the evolution trajectory prediction model to predict the evolution trajectory sequence of the alkali concentration in the future rolling time domain. An optimization construction unit is used to construct a multi-objective coordinated optimization problem in the rolling time domain with the dual optimization objectives of minimizing the concentration deviation integral and minimizing the regulation energy consumption. The concentration deviation integral is determined by the time domain integral of the deviation between the predicted concentration and the target concentration at each moment in the evolution trajectory sequence. The regulation energy consumption is determined by the weighted sum of squares of the flow compensation increment and the acceleration rate compensation increment. The frontier solving unit is used to solve the Pareto front of the optimization problem and select the compensation sequence that satisfies the preset concentration convergence rate constraint and minimizes the adjustment energy consumption. The compensation execution unit is used to apply the solvent flow rate compensation amount and the solute dosing acceleration rate compensation amount of the first control cycle in the compensation sequence to the solvent flow rate regulating valve and the solute dosing drive device, respectively. The rolling update unit is used to update the process parameters at the next sampling time and slide the rolling time domain forward by one control cycle, repeating the trajectory prediction and compensation sequence solution process until the measured concentration of the alkali solution outlet converges to the preset target concentration.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images