Carbonate low-temperature rectification load allocation method and system based on multi-objective solution

Abstract

The application discloses a carbonic ester low-temperature rectification load allocation method and system based on multi-target solving, and belongs to the field of carbonic ester rectification process optimization control. The method carries out time stamp alignment on multi-rectification unit process data, generates a dynamic reachable load pipe retaining cross-time sequence relationship by using a dynamic response model, carries out multi-target non-dominated solving in a system dynamic feasible region formed by total feed and shared cold and hot resource constraints, and converts target tracks into load transfer batches for supply and demand load allocation. The required condensing capacity and reboiling capacity for load increase are respectively subjected to resource checking, and the batches are checked, frozen or withdrawn, and the load pipe is updated rolling, so that the shared resource contention and the quality and safety out-of-limit risks are reduced.

Description

Technical Field

[0001] This invention relates to the field of optimized control of carbonate distillation processes, and more particularly to a method and system for load adjustment in low-temperature carbonate distillation based on multi-objective solutions. Background Technology

[0002] Cryogenic distillation of carbonates typically involves multiple distillation units operating in parallel or series. Each unit is affected by variations in feed composition, feed temperature, equipment heat transfer conditions, column hydraulic conditions, and product demand, resulting in different safe processing loads, corresponding energy consumption, product purity, and recovery rates. Furthermore, these multiple distillation units often share cold and heat sources; changes in the load of a single column can alter the condensation and reboiling loads, and through resource sharing, affect other distillation units.

[0003] Existing load allocation methods often employ fixed ratios, independent adjustment of individual towers, or static weight optimization, making it difficult to consistently maintain candidate load trajectories that meet constraints related to quality, hydraulics, reflux, temperature and pressure, condensation, reboiling, and load change rate during dynamic transitions. Static allocation may also combine loads that appear feasible at each time step but are unreachable across time sequences into the target trajectory. Furthermore, when the load increase of the receiving unit and the resource release of the supply unit are not synchronized, it can easily lead to instantaneous competition for shared cooling or heating capacity. When model prediction biases are not used to tighten execution boundaries, it also increases the risk of localized overload, product quality fluctuations, and unnecessary energy consumption.

[0004] Therefore, there is a need for a method and system for adjusting the load of cryogenic carbonate distillation based on multi-objective solution that can overcome the shortcomings of the existing technology. Summary of the Invention

[0005] One objective of this invention is to propose a method and system for load allocation in cryogenic carbonate distillation based on multi-objective solutions. Addressing the difficulty of existing technologies in achieving continuous and executable multi-tower load allocation during transitions with limited shared cold and heat resources, taking into account the dynamic response differences of different distillation units, this invention proposes a combined technical solution that integrates state alignment, construction of dynamically achievable load pipes, multi-objective non-dominated solutions within the system's dynamic feasible domain, batch balancing of supply and load, separate verification of resources, measured feedback verification, and rolling updates. This invention constrains the target trajectory by both the dynamic boundaries of individual towers and the shared resources of the system, and ensures that load increases are consistent with the reserve capacity or measured release of corresponding condensing and reboiling capacities. This effectively reduces resource contention, localized overload, and exceeding quality and safety limits.

[0006] This invention provides a method for load adjustment in cryogenic distillation of carbonates based on multi-objective solution, comprising: S1. Collect process measurement data and shared cold and heat resource availability of multiple distillation units according to the control cycle, and align the process measurement data with timestamps to obtain the current status of each distillation unit; S2. Based on the current state and candidate load sequence, the dynamic response model of each distillation unit is used to predict the quality indicators, equipment status and cold and hot load trajectories, and the dynamic reachable load tube of each distillation unit is obtained by filtering according to the preset operating constraints at each prediction time. S3. Combine the dynamic reachable load tube with the total feed conservation, shared cold capacity, shared heat capacity and product demand constraints to form the system dynamic feasible domain. Within the system dynamic feasible domain, perform multi-objective non-dominated solutions for quality indicators, energy consumption and load action, and select the target load trajectory from the solutions that satisfy the quality constraints and safety constraints in the preset operating constraints according to the bottleneck margin. S4. Based on the difference between the current state and the target load trajectory, generate a load transfer batch and corresponding control instructions that match the load reduction of the supply unit and the load increase of the receiving unit. The load increase is only performed when the required cold and heat resources of the receiving unit are met by the reserve quantity or the release quantity of the supply unit as confirmed by actual measurement. S5. Verify the predicted trajectory based on the measured data during the execution of the load transfer batch, freeze or roll back the current load transfer batch when the constraint margin is insufficient, and update the dynamic reachable load tube for the next control cycle based on the verification result.

[0007] Optionally, S1 includes: collecting feed flow rate, feed composition, feed temperature, column top pressure, column bottom temperature, column pressure drop, reflux flow rate, condensation load, reboiling load, product flow rate, and product composition for each distillation unit; Using the end of the control cycle as the alignment time, window statistics are performed on process measurement data that generate multiple sample values ​​within a control cycle. Product composition test data returned later than the alignment time are backfilled according to the sampling time, and the most recent valid measurement value and its timestamp are used to characterize data that has not been updated in the current control cycle, forming the current state containing data time stamps.

[0008] Optionally, S2 includes: the dynamic response model includes a mechanism model based on material balance, energy balance and gas-liquid balance and a model deviation correction part determined according to historical operating data; The quality indicators include product purity and recovery rate, and the preset operating constraints include lower limit of product purity, lower limit of recovery rate, tray load range or packing load range, reflux ratio range, temperature range, pressure range, upper limit of condensation capacity, upper limit of reboiling capacity, upper limit of load change rate, and lower limit of equipment safety margin. Candidate load sequences that satisfy the preset operating constraints at each prediction time within the prediction time domain are retained, and the set of retained candidate load sequences constitutes the dynamic reachable load tube, so that the dynamic reachable load tube retains the state recursion relationship and load change relationship determined by the dynamic response model between adjacent prediction times; Furthermore, the difference between the measured and predicted values ​​under the same correspondence within the historical control cycle is grouped according to the distillation unit, operating constraint category, load increase or decrease direction, and prediction step size. The measured recovery rate is determined based on the feed flow rate, feed composition, product flow rate, and product composition in the current state. For each group, along the direction that reduces product purity or recovery rate or makes the temperature and pressure state, the hydraulic state in the tower, the condensation load or the reboiling load approach the corresponding operating constraint boundary, determine the one-sided residual boundary that covers the preset proportional residual of the group. The unilateral residual boundary is added to the measurement error boundary of the corresponding measuring device to obtain the constraint-by-constraint safety tightening amount; The corresponding operating constraint boundary is moved toward the feasible side by the constraint tightening amount to correct the dynamically reachable load tube, and the upper limit of the batch is reduced according to the difference in the load capacity of the load unit before and after the correction.

[0009] Optionally, S3 includes: establishing the total feed conservation constraint and the product demand constraint so that the sum of the feed rates of each distillation unit equals the total feed rate of the system, and so that the flow rate and quality indicators of each product meet the product demand setpoints of the current control cycle. The upper limit for shared cooling capacity and the upper limit for shared heating capacity are established based on the sum of the condensation load and the sum of the reboiling load of each distillation unit not exceeding the value after deducting the preset reserve margin from the available amount of the corresponding shared resources. The nondominated solution set is determined by normalizing the quality indicators, unit energy consumption, processing capacity, safety margin, and total load action variation in the same direction. The load trajectory of each distillation unit in each non-dominated solution belongs to the candidate load sequence contained in the corresponding dynamically reachable load tube. For each non-dominated solution, its various operational constraint margins are normalized according to the corresponding allowable range, and the lower bound of the normalized margin is used as the bottleneck margin. The solution with the largest bottleneck margin is selected from the non-dominated solutions that satisfy the quality constraints and safety constraints in the preset operational constraints as the target load trajectory.

[0010] Optionally, S4 includes: determining the maximum batch size for a single load transfer batch based on the load reduction capacity of the supply unit, the load increase capacity of the receiving unit, and the availability of shared cooling and heating capacity; The total load difference is divided into multiple load transfer batches not exceeding the upper limit of the batch, so that the sum of the load reduction of the supplying units in the same load transfer batch is equal to the sum of the load increase of the receiving units. The feed change trajectory for each load transfer batch is determined based on the dynamic response model, and the corresponding set trajectories for reflux, condensation load and reboiling load are determined based on the feed change trajectory, thereby generating the feed command, reflux command, condensation command and reboiling command. Furthermore, the condensing capacity and reboiling capacity required for the load-bearing unit to increase load in each load transfer batch are converted into cold energy token demand and heat energy token demand, respectively. The condensing capacity and reboiling capacity released by the common reserve resources or the load-bearing unit after load reduction and confirmed by actual measurement of refrigerant flow, heat flow, regulating valve position or heat exchange load are converted into available cold energy tokens and available heat energy tokens, respectively. The corresponding load increase command is allowed to be executed only after the loaded unit obtains a quantity of available cooling tokens and available heating tokens that meet the corresponding demand. The state region where the margin of each monitored state to the corresponding operating constraint boundary after being corrected by the per-constraint safety tightening amount reaches the preset target margin lower limit is defined as the target margin region. After the load unit enters the target margin region, the available cold energy token and available heat energy token occupied by the load unit are released. Furthermore, resource release confirmation deadlines and status trend verification conditions are set for each load transfer batch; If the demand for cold or heat tokens to be met by the load reduction release from the load supply unit is not confirmed by actual measurement within the resource release confirmation period, or if any operating state tends to exceed the operating constraint boundary after being corrected by the constraint tightening amount based on the constraint margin of multiple consecutive control cycles, the unexecuted load increase command is frozen or a rollback command is generated according to the executed load increase amount, and the upper limit of the next load transfer batch is reduced to no more than the load increase amount that has been confirmed by actual measurement to be safe to execute.

[0011] Optionally, S5 includes: during the execution of each load transfer batch, comparing the measured pressure, temperature, tower pressure drop and product composition with the predicted values ​​of the predicted trajectory at the corresponding timestamp, and determining the constraint margin based on the distance of each measured value to the corresponding operating constraint boundary; When any constraint margin is less than the lower limit of the corresponding equipment safety margin, an exception is handled, the issuance of subsequent load increase commands is stopped, and the current command is maintained or a rollback command is issued in the opposite direction of the current load transfer batch. The difference between the measured value and the predicted value is written into the deviation correction part of the corresponding dynamic response model to update the predicted trajectory and dynamic reachable load tube for the next control cycle.

[0012] On the other hand, the present invention also provides a carbonate cryogenic distillation load distribution system based on multi-objective solution, comprising: The system comprises the following modules: a state construction module for collecting process measurement data and shared resource availability, aligning them with timestamps to obtain the current state of each distillation unit; a load tube generation module for generating the dynamic reachable load tube for each distillation unit based on the current state, candidate load sequences, and dynamic response model; an objective solution module for constructing the system's dynamic feasible region and performing multi-objective non-dominated solutions within that region to obtain the target load trajectory; a batch execution module for generating load transfer batches and corresponding instructions, and controlling the load increase of the receiving unit based on the measured confirmation of the corresponding shared resource reserve or the corresponding capacity release of the supplying unit, according to the condensing and reboiling capacity required for the load increase of the receiving unit; and a rolling verification module for verifying the predicted trajectory based on measured data, freezing or reverting the current load transfer batch and updating the dynamic reachable load tube for the next control cycle when the constraint margin is insufficient.

[0013] The beneficial effects of this invention are: 1. By retaining the state recursion relationship and load change relationship of the candidate load sequence across the prediction time by the dynamic response model and solving it within the dynamic feasible domain of the system, it is possible to avoid recombining load values ​​that are feasible at each time step but unreachable across time series into the target trajectory.

[0014] 2. By decomposing the target difference across towers into load transfer batches that match the load reduction and load increase, and by performing standby or measured release verification for the required condensing capacity and reboiling capacity respectively, the instantaneous competition for shared cold and heat resources during the multi-tower transition process can be reduced.

[0015] 3. By tightening the dynamic reachable load tube and batch upper limit with unilateral residual limits and measurement error limits, and freezing or withdrawing batches when the measured constraint margin is insufficient, model errors, shared resource status and inter-tower load transfer can be kept consistent in rolling control. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the method for adjusting the load of carbonate cryogenic distillation based on multi-objective solution according to the present invention.

[0017] Figure 2 This is a flowchart of the dynamic reachable load tube generation method in S2 of the present invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figures 1-2 A multi-objective solution-based method for load allocation in cryogenic carbonate distillation includes: S1. Collect process measurement data and shared cold and heat resource availability of multiple distillation units according to the control cycle, and align the process measurement data with timestamps to obtain the current status of each distillation unit; S2. Based on the current state and candidate load sequence, the dynamic response model of each distillation unit is used to predict the quality indicators, equipment status and cold and hot load trajectories, and the dynamic reachable load tube of each distillation unit is obtained by filtering according to the preset operating constraints at each prediction time. S3. Combine the dynamic reachable load tube with the total feed conservation, shared cold capacity, shared heat capacity and product demand constraints to form the system dynamic feasible domain. Within the system dynamic feasible domain, perform multi-objective non-dominated solutions for quality indicators, energy consumption and load action, and select the target load trajectory from the solutions that satisfy the quality constraints and safety constraints in the preset operating constraints according to the bottleneck margin. S4. Based on the difference between the current state and the target load trajectory, generate a load transfer batch and corresponding control instructions that match the load reduction of the supply unit and the load increase of the receiving unit. The load increase is only performed when the required cold and heat resources of the receiving unit are met by the reserve quantity or the release quantity of the supply unit as confirmed by actual measurement. S5. Verify the predicted trajectory based on the measured data during the execution of the load transfer batch, freeze or roll back the current load transfer batch when the constraint margin is insufficient, and update the dynamic reachable load tube for the next control cycle based on the verification result.

[0020] In this specific embodiment, S1 includes: This specific implementation uses three parallel carbonate cryogenic distillation units that share a refrigerant station and a heat transfer station as the allocation objects. The method is jointly executed by a distributed control system, a laboratory information system, and a load allocation server. The unit identifiers of the three distillation units are written into the unit number field of the status record. The load allocation server uses a 5-minute control cycle. At the end of each control cycle, a unique cycle number, alignment time, and data version number are established. The dynamic reachable load pipe version, executed load transfer batch number, and resource token occupancy status output from the previous control cycle are used as prior records for the status construction of the current cycle. The distributed control system reads the feed flow rate, feed temperature, absolute pressure at the top of the column, reboiler temperature, pressure drop inside the column, reflux flow rate, refrigerant flow rate, heat transfer fluid flow rate, condensing load, reboiling load, and product flow rate of each distillation unit at a sampling cycle of 5 seconds. The units for feed flow rate, reflux flow rate, and product flow rate are kg / h, the unit for temperature is degrees Celsius, the unit for pressure and pressure drop inside the column is kPa, and the unit for condensing load and reboiling load is kW. The feed composition and product composition are recorded as mass fractions. The online composition analyzer outputs a measurement value every 60 seconds. The analysis information system receives the feed samples and product samples obtained according to the production batch and saves the sampling time, analysis completion time, component mass fraction, sample number, and review status. The load dispatching server uses the end of the current control cycle as the alignment time. It sorts the measured values ​​of flow rate, temperature, pressure, tower pressure drop, and heat exchange load within a 5-minute window by timestamp. First, it deletes records that exceed the instrument range or quality level due to faults. Then, it takes the time-weighted average of the retained records and forms the fluctuation range with the maximum and minimum values ​​within the window. For the valve position of the control valve, the pump operating status, and the analyzer quality level, it uses the effective value that is closest to the alignment time and no later than that time. For duplicate records with the same timestamp, it retains the record with the largest acquisition sequence number. This forms a process status record with a fixed field order. For product composition test data returned later than the alignment time, the load dispatch server backfills the data according to the sampling time. Instead of using the test completion time as the process occurrence time, it reads the sampling time based on the sample number, backfills the test values ​​to the historical control cycle that covers the sampling time, and adds a backfill version number to the affected historical status record. If the current control cycle has not yet returned test values, it reads the most recently approved valid test value and its sampling timestamp, and marks the composition field as delayed. The historical residual recalculation is triggered only after the test values ​​are returned, so that subsequent residual grouping uses the composition data corresponding to the actual sampling time. Before the timestamp alignment rule is implemented, it is verified by the original records of 30 consecutive control cycles and the event sequence confirmed by manual verification. The allowable cross-cycle mismatch rate is no more than 0.5%, the alignment deviation of flow and temperature and pressure measurement values ​​is no more than 5 seconds, and the alignment deviation of online component values ​​is no more than 60 seconds. If the verification fails to meet any threshold, the original records are kept unchanged and the current status snapshot is stopped. The clock and acquisition sequence number are resynchronized by the data interface and the current cycle window is recalculated. The alignment rule version, verification sample interval, mismatch records and verification conclusions are written into the status construction audit table. For measurement fields that have not been updated within the current control cycle, according to Calculate data timeliness, The symbol represents the first The first distillation unit The data in the measurement field is time-sensitive and the unit is seconds. The symbols represent discrete indices of distillation units with values ​​from 1 to 3, and the units are not applicable. The symbols represent discrete indexes of measurement field categories, and the units are not applicable. 1 to 11 correspond to feed flow rate, feed composition, feed temperature, top pressure, bottom temperature, pressure drop, reflux flow rate, condensate load, reboiler load, product flow rate, and product composition, respectively. The symbol represents the current control cycle alignment time, using a server-wide clock and a timestamp in seconds, accumulated since 00:00 on January 1, 1970. The symbol represents the same timing reference timestamp of the most recent valid measurement of the corresponding field and the unit is s; the status record simultaneously saves the most recent valid value, data timeliness and four types of timeliness markers: valid, delayed, missing or faulty. When the data timeliness of the flow rate, temperature and pressure and heat exchange load fields exceeds 30 s, it is marked as missing. When the data timeliness of the constituent fields exceeds the sum of its most recent planned test interval and 20 min grace period, it is marked as missing. The available shared cooling capacity is obtained by deducting the locked load of non-rectification users and the protection margin of the cooling station from the total cooling capacity of the refrigerant station. The available shared heating capacity is obtained by deducting the locked load of non-rectification users and the protection margin of the boiler side from the total heating capacity of the heat transfer station. The total cooling capacity is calculated based on the refrigerant supply and return flow rate, supply and return enthalpy difference, and the heat exchange efficiency calibrated by the meter. The total heating capacity is calculated based on the heat transfer supply and return flow rate, supply and return enthalpy difference, and the heat exchange efficiency calibrated by the meter. The server saves the available cooling capacity value, available heating capacity value, reserve value, measurement timestamp, flow meter mass level, enthalpy source, and resource version number in the shared resource record. All capacity fields are uniformly converted to kW before being transmitted to subsequent steps. When any distillation unit's feed flow rate, top pressure, bottom temperature, or internal pressure drop is missing or faulty, that unit is not allowed to be used as a load unit in this cycle, and the load of the previous effective cycle will be used as its candidate sequence base point. When the shared cooling capacity or shared heating capacity record is missing, the available quantity in this cycle will be the minimum value of the three most recent effective control cycles, and the release of unconfirmed load supply will be prohibited. If there is still no effective shared resource record for three consecutive control cycles, the generation of load increase candidates will be stopped, and only load reduction or load maintenance candidates will be retained. After the state is constructed, the server writes the unit number, cycle number, alignment time, process measurement value, fluctuation range, composition value, time stamp of each field data, shared available cooling capacity, shared available heating capacity and resource version number into the current state snapshot, and performs mode verification on the field name, unit and dimension. The current state snapshot that passes the verification is passed to S2 as a read-only object, and the backfilled version and time stamp are written to the historical state database at the same time, so that S4 can build residual samples grouped by distillation unit, operating constraint category, load direction and prediction step size.

[0021] In this specific embodiment, S2 includes: The load generation module reads the current state snapshot output by S1, and generates candidate load sequences within a prediction time domain of 12 prediction steps and 5 minutes each, using the current feed flow rate of each distillation unit as the starting point of the candidate sequence. In this specific implementation, the feed change at adjacent prediction times is limited to -5%, -2.5%, 0%, +2.5%, or +5% of the current feed flow rate. At the same time, branches with excessively rapid changes are eliminated based on the upper limit of the load change rate in the equipment file, and candidate sequence number, parent node number, feed setting sequence, generation time, and model version are written for each remaining branch. To limit the scale of activities, after the module completes the constraint screening at each prediction layer, it sorts the candidate sequence numbers in descending order of the minimum normalized safety margin up to that layer, ascending order of the cumulative comprehensive heat exchange energy consumption, and lexicographical order of the candidate sequence numbers. Each distillation unit retains a maximum of 200 active branches per layer and retains the parent node chain of each branch completely. Nodes with different states are not merged, and paths of different parent nodes are not spliced. At the start of each control cycle, the maintain load sequence and the safe load reduction sequence generated according to the upper limit of the load change rate are fully generated and verified first, and then the remaining candidate trees are expanded. The candidate tree calculation budget is set to 240 s from the S1 state snapshot. The server stops generating new branches 10 s before the budget deadline and only continues to complete the branches that have entered the model calculation. At the deadline, only the sequence that has covered all 12 prediction steps and passed all constraints is published. If the memory usage reaches 80% of the memory reserved for this task, the calculation budget is exhausted, or the active branch limit is triggered, the load management record is written with the status of limited scale or incomplete calculation, and at least the verified maintain load and safe load reduction sequences are published. Incomplete sequences are not published. If the two types of pre-verified sequences are also not feasible, the load management of this unit is stopped and the load increase of this unit is frozen. The active branch limit, memory threshold, calculation budget and trigger status are written into the versioned real-time calculation configuration. For each feed candidate sequence, the load tube generation module constructs initial trajectories of reflux flow rate, condensing load, and reboiling load under the same candidate number from a validated control baseline table. This table uses unit number, product grade, feed composition range, and load range as lookup keys, and stores the calibrated reflux-feed coefficient, unit feed condensing load coefficient, unit feed reboiling load coefficient, upper and lower limits, unit, calibration sample range, and version number. The reflux-feed coefficient is defined as the reflux mass flow rate of a stable sample divided by the concurrent feed mass flow rate. The dimensional ratio can only be used within the composition, grade, and load range specified in the record, and belongs to a different field from the conventional reflux ratio of reflux flow divided by distillate flow in the distillation field. The module first multiplies the reflux-feed coefficient by the candidate feed flow to obtain the initial reflux value. The conventional reflux ratio in the predicted output is calculated separately by dividing the obtained reflux flow by the model-predicted distillate flow. Then, the initial condensation and reboiling values ​​are obtained by adding the corresponding marginal coefficient and the product of the current measured heat exchange load and the feed change. Subsequently, within the equipment constraints, the bounded coordinate search is performed with a fixed search step size of 1% reflux flow, 2 kW condensation load, and 2 kW reboiling load, in the order of mass constraint error, material balance error, and energy balance error. The search stops when the three types of errors do not exceed 1% of the allowable range, 0.01% of the total system feed, and 1% of the current total heat exchange load, respectively. If the predicted distillate flow is not greater than the lower limit of the flow meter after calibration, there is no matching control reference record, or the process fails to converge after 20 iterations, the candidate is deleted. The four converged control trajectories are saved along with the candidate sequence number. The dynamic response model of each distillation unit consists of a nonlinear mechanism model based on material balance, energy balance, and gas-liquid balance, as well as an output deviation correction vector released by S5. The model follows... Recursive state, The symbol represents the first The distillation unit in the first... The state vector at each predicted time point The symbol represents the corresponding unit in the first position. The state vectors at each prediction time point, in a fixed order, include the target component liquid holdup at the top of the column (kg), the target component liquid holdup at the bottom of the column (kg), the temperature of a representative column section (degrees Celsius), the absolute pressure at the top of the column (kPa), the pressure drop inside the column (kPa), the equivalent thermal state of the condenser (kW), and the equivalent thermal state of the reboiler (kW). The initial values ​​are derived from the S1 measurement and the material energy state estimate at the beginning of the current cycle. The symbol represents a discrete element index with a value of 1 to 3, and the unit is not applicable. The symbol represents a discrete prediction step size index with values ​​from 0 to 11, and the unit is not applicable. The symbol represents the first The state recursive mapping obtained by discretizing the material, energy, and gas-liquid balance equations of each distillation unit, and with no applicable function units, has output components that have the same dimensions as the corresponding components of the state vector. The symbols represent control vectors composed of feed rate (kg / h), reflux rate (kg / h), condensing load (kW), and reboiling load (kW) in a fixed order, obtained from a control reference table and a bounded search. The symbol represents an external input vector derived from the current state of S1, composed in a fixed order of dimensionless mass fraction of feed composition, feed temperature in degrees Celsius, shared available cold energy in kW, and shared available heat energy in kW. The symbol represents the output bias correction vector used in the following output mapping formula but not in the current state recursion formula; The initial values ​​of the model state are determined by the current state snapshot of S1 through steady-state material balance and the most recent model state estimation. For units where the composition field is in a delayed state and the data timeliness does not exceed the sum of the planned test interval and the 20-minute grace period, the most recently approved valid composition value is used and the range of composition variation is extended to the 95th percentile interval of the composition deviation of the most recent 20 valid cycles of the unit. The model is run along the upper and lower boundaries of the interval respectively, and only candidate sequences that satisfy the constraints of the two boundary predictions are retained. For units marked as missing or faulty in feed composition or product composition, they are prohibited from being used as load-bearing units. Only the most recently approved valid value and the approved maximum composition uncertainty interval are used to generate candidates for maintaining load or safe load reduction. If the most recently approved valid value also exceeds the above maximum timeliness or does not exist, the dynamic reachable load tube of the unit is stopped from being published and the composition blockage reason and manual review status are recorded. After each state is recursively processed, according to Calculate the predicted output. The symbol represents the first The first distillation unit The output vector at each prediction time point, in a fixed order, includes dimensionless mass fraction of product purity, dimensionless proportion of recovery rate, dimensionless proportion of tray load or packing load, dimensionless proportion of reflux ratio, representative column section temperature (degrees Celsius), absolute pressure at the top of the column (kPa), pressure drop inside the column (kPa), condensation load (kW), and reboiling load (kW). The symbol represents an output function that maps state and control variables to the nine output components mentioned above, and the function unit is not applicable. Each output component has the same dimensions as its corresponding output vector component. The symbols represent state vectors composed of a fixed order of top liquid holdup (kg), bottom liquid holdup (kg), column section temperature (°C), top pressure (kPa), internal pressure drop (kPa), and equivalent thermal states for condensation and reboiling (kW). The symbols represent control vectors composed of feed rate (kg / h), reflux rate (kg / h), condensing load (kW), and reboiling load (kW) in a fixed order. The symbol represents the output bias correction vector read from the current effective bias version of S5 and composed of the nine output components in the same order and units as described above, using the same published version within 12 prediction steps. The symbol represents a discrete element index with a value of 1 to 3, and the unit is not applicable. The symbol represents a discrete prediction step size index with a value from 0 to 11 and the unit is not applicable; the output record simultaneously saves the predicted value, unit, prediction step size, model version and the output bias correction version used; The material balance of the mechanistic model is closed-loop based on changes in the target carbonate component in the feed, the top product, the bottom product, and the liquid holdup within the column. The energy balance is closed-loop based on feed enthalpy, condensation heat transfer, reboiling heat supply, product enthalpy, and equipment heat dissipation. The gas-liquid balance parameters are calculated by the property package of this unit under the current pressure and composition. The model parameter table uses unit number, packing or tray section, and composition range as search keys to store effective liquid holdup, mass transfer efficiency, heat transfer coefficient, pressure drop coefficient, parameter units, effective range, and version number. The parameters are identified from the stable operating data of the most recent three months and are updated when there is planned maintenance, changes in the property package version, or when the prediction error exceeds the acceptance threshold for five consecutive cycles. During model identification, operational segments within the most recent three months that had effective control commands, valid measured mass levels, and material balance relative errors not exceeding 2% were selected. The first 80% were used as the identification set, and the last 20% as the validation set, arranged chronologically. Bounded nonlinear least squares was used to adjust the effective liquid holdup, mass transfer efficiency, heat transfer coefficient, and pressure drop coefficient. For each sample, the predicted errors for product composition, top pressure, bottom temperature, internal pressure drop, condensate load, and reboiling load were first divided by the corresponding positive allowable span in the current effective operational constraint table. Then, the squares of the six dimensionless errors were summed with an equal weight of one-sixth. Finally, all effective samples in the identification set were analyzed. The arithmetic mean is used as the sole identification target. The normalized span, six non-negative weights, the verification result of the weight sum being 1, the constraint table version, and the model parameter table version are all written into the identification configuration. Identification is rejected if any span is missing, zero, or negative, or if any weight is missing or the weight sum deviates from 1 by more than one millionth. The identification parameters must fall within the valid range given by the device file. If the average absolute error of any key output on the validation set exceeds 5% of the corresponding allowable range, the new version is rejected. During runtime, the current state and candidate control vectors are read first, and then state recursion and output mapping are executed in the order of prediction step size, thereby forming a definite model inference path. The load tube generation module checks the lower limit of product purity, lower limit of recovery rate, tray load range or packing load range, reflux ratio range, temperature range, pressure range, upper limit of pressure drop in the tower, upper limit of condensing capacity, upper limit of reboiling capacity, upper limit of load change rate, and lower limit of equipment safety margin at each predicted time. Each boundary is read from the approved operating constraint table, which uses unit number, product grade, and operating range as keys to store upper and lower limits, units, effective time, instrument error limits, and safety tightening version. If no matching record is found for the search key, the corresponding candidate sequence is deleted without using the adjacent grade boundary as a substitute. For lower limit constraints, the predicted value minus the lower limit after safety tightening is used as the constraint margin. For upper limit constraints, the upper limit after safety tightening is used minus the predicted value as the constraint margin. For constraints with both upper and lower limits, the minimum value of the two margins is taken. If the product purity, recovery rate, or equipment safety margin at any prediction time is lower than its lower limit, or if the hydraulic, temperature and pressure, reflux, condensation, reboiling, and rate of change constraints exceed the limits, the candidate sequence and its derived subsequent branches are immediately deleted, and the deletion step size, constraint type, and exceedance amount are recorded. After completing the time-by-time screening, all candidate sequences covering the entire prediction time domain are retained and arranged from largest to smallest according to the minimum safety margin of the entire time domain. If the minimum safety margins of the entire time domain are the same, they are arranged from smallest to largest according to the comprehensive condensation and reboiling energy consumption. If they are still the same, they are arranged in alphabetical order according to the candidate sequence number. This sorting only determines the order of subsequent combination search and record writing, and does not delete sequences that have passed the constraint screening. Therefore, the same candidate set, model version and constraint version obtain a unique and verifiable dynamic reachable load tube order. Candidate load sequences that satisfy all preset operating constraints at each prediction time within the prediction time domain are written into the dynamic reachable load tube of the corresponding distillation unit. The load tube records and retains the candidate sequence number, feed rate at each step, state vector, quality index, equipment status, condensation load, reboiling load, constraint margin, parent node relationship, and load change relationship between adjacent steps. Since subsequent steps can only select complete candidate sequence numbers and cannot arbitrarily splice load values ​​from different sequences at a single time, the cross-time-series reachability relationship of state recursion and load change is retained along with the dynamic reachable load tube. When a distillation unit does not have an increase sequence covering all 12 predicted steps, the load tube still retains the hold load or reduce load sequence that satisfies the constraints, and marks the unit as prohibited from being loaded; when none of the distillation units have the combination sequence required for the total system feed, the longest continuous feasible step and the first blocking constraint are written into the infeasibility cause record, and S3 only solves for the combination of hold current load and safe reduce load without extrapolating the model boundary; After the screening is completed, the load tube generation module summarizes the calculation content of each dynamic reachable load tube and writes it into the load tube library. The output objects include the cycle number, unit number, model version, constraint table version, deviation correction version, candidate sequence set and infeasibility reason record. This output serves as the source of discrete trajectory for constructing the dynamic feasible domain of the system in S3, and as a common baseline for batch upper limit correction, prediction verification and next cycle rolling update in S4 and S5.

[0022] In this specific embodiment, S3 includes: The objective solver module reads the dynamic reachable load tubes of the three distillation units under the same cycle number, treats a complete candidate sequence of each unit as a discrete decision object, and constructs a system combination record with the candidate sequence number, feed rate of each prediction step, product flow rate, product purity, recovery rate, condensation load, reboiling load, and constraint margin. When generating the combination, only sequences with the same prediction start point, prediction step number, model version family, and shared resource version are connected. Combinations with inconsistent versions are marked as incomparable and exit the current cycle solver. At each prediction time, a total feed conservation constraint is established based on the sum of the feed rates of each distillation unit equaling the total system feed rate. Specifically, this is done according to... implement, The symbol represents the first The first distillation unit The feed rate at each predicted time point, in kg / h. The symbol represents the total system feed demand at the same predicted time, and the unit is kg / h. The symbol represents the number of distillation units involved in the preparation, and is a dimensionless positive integer, which is taken as 3 in this specific implementation. The symbols represent discrete distillation unit indices with values ​​from 1 to 3, and the units are not applicable. The symbol represents the discrete prediction step size index with a value of 0 to 11 and the unit is not applicable; the numerical closure error allowed by the solver is set to one ten-thousandth of the total system feed requirement, and combinations exceeding this error are directly judged as infeasible; Shared cooling capacity constraints according to implement, The symbol represents the first The first distillation unit The condensing load at each predicted time point, in kW. The symbol represents the shared cooling capacity available from S1, and the unit is kW. The symbols represent cooling station fluctuations and the reserve cooling capacity of unmodeled users, with units of kW. The symbol represents the number of dimensionless distillation units, and in this specific embodiment, it is taken as 3. The symbols represent discrete distillation unit indices with values ​​from 1 to 3, and the units are not applicable. The symbol represents the discrete prediction step size index with a value of 0 to 11 and the unit is not applicable; the shared heat constraint adopts the same checking process and compares the sum of the reboiling load of each unit with the value of the shared heat availability after deducting the heat reserve margin. The cold energy reserve margin and the heat energy reserve margin are respectively taken as 10% of the corresponding resource availability and written into the solution configuration of this cycle. Product demand constraints are aggregated by product brand for each unit product flow, requiring that the aggregated flow not be lower than the demand setpoint of the production plan at the corresponding forecast time. Simultaneously, the purity and recovery rate of each unit product included in this aggregated flow must meet the S2 quality constraints. The production plan record uses the cycle number and product brand as keys to store demand flow, minimum purity, minimum recovery rate, effective period, and plan version. The quality target table uses the cycle number, product brand, and effective period as lookup keys to store product purity targets, recovery rate targets, approver, version number, and default value status. Quality targets must not be lower than the minimum purity and minimum recovery rate in the same version of the production plan. If no matching valid record is found in any table, the multi-objective solution for this cycle is terminated without using expired targets. If any version changes during the solution process, the current population evaluation is terminated and restarted from the candidate combination generation stage. For each system combination, five initial targets are calculated according to the following rules: At each prediction step, first calculate the product purity target minus the predicted purity for each unit; if the difference is positive, take the difference; otherwise, take zero. Then divide by the positive purity allowable range in the quality target table. Calculate the non-negative deficit of the recovery rate target relative to the predicted recovery rate using the same rules and divide by the positive recovery rate allowable range. Multiply each of the two dimensionless deficits by 0.50 and add them together. Then, weight the sum by the proportion of the predicted product flow rate of this unit to the total predicted product flow rate of the three units to obtain the quality loss for this step. Divide the sum of the condensation load and reboiling load of the three units by the predicted product flow rate of the three units in the same step. The total product flow rate is used to obtain the unit product comprehensive energy consumption. The heat exchange load is in kW and the product flow rate is in kg / h, resulting in a unit of kWh / kg. Only the predicted product flow rate of units that meet the quality constraints of this cycle in terms of predicted product purity and recovery rate is summarized, and the total flow rate of qualified products is multiplied by -1 to obtain the negative qualified product flow rate in kg / h, so that different unit allocations and quality results correspond to different treatment benefits. For all normalized constraint margins stored in S2, 0.20 is subtracted from each margin. If the difference is positive, the difference is taken; otherwise, zero is taken. The maximum value among all non-negative differences is taken as the safety margin gap for this step. Quality loss, unit product comprehensive energy consumption, negative throughput, and safety margin gap are all calculated using a fixed weighted average of 1 / 12 across 12 prediction steps. The fifth objective is calculated by summing the absolute values ​​of the differences in feed rates between adjacent prediction steps for each of the three units, yielding the total load action variation in kg / h. The allowable span for purity and recovery rate is read from the same version of the quality objective table as the production plan and must be greater than zero. If the total predicted product flow rate at any step is not greater than the calibrated lower limit of the product flow meter, or if any allowable span is missing or not positive, the combination is deemed unevaluable. The above objective calculation rules, fixed weights, quality objective table version, and anomaly flags are written into the solver configuration. Quality loss, unit energy consumption, safety margin gap, and total load action variation are all preferred to be as small as possible. The total throughput, multiplied by a negative sign, is also preferred to be as small as possible. Regarding the first Project target and number One feasible combination, according to Perform normalization in the same direction. The symbol represents the normalization objective value and is a dimensionless number. The symbol represents the first The first combination The original target values ​​are defined as follows: Target Index 1 corresponds to the dimensionless mass loss caused by the difference in product purity and recovery rate; Target Index 2 corresponds to the unit energy consumption obtained by dividing the condensation and reboiling loads by the product flow rate, in kWh / kg; Target Index 3 corresponds to the negative qualified product flow rate in kg / h; Target Index 4 corresponds to the dimensionless safety margin gap; and Target Index 5 corresponds to the total variation in load action in kg / h. The symbol represents the first feasible combination sample in this period. The minimum value of the project target, and the unit corresponds one-to-one with the target index mentioned above. The symbols represent the corresponding maximum values, and the units correspond one-to-one with the target indices mentioned above. The symbol represents the positive zero-difference protection quantity, which is expressed in dimensionless, kWh / kg, kg / h, dimensionless, and kg / h units according to the target index, and is taken as one millionth of the approved allowable positive span of the target. The evaluation of the corresponding target is terminated when the allowable span is not positive. The symbol represents a discrete target index with a value from 1 to 5, and the unit is not applicable. The symbol represents a discrete combination index from 1 to the total number of feasible combinations in the current period, and the unit is not applicable; the smaller the normalized value, the better the corresponding objective, and the calculation result is written into the objective vector field of the combination record. The objective is solved using a non-dominated sorting genetic algorithm with elite retention. The population size is set to 160, and the maximum number of iterations is set to 120. The individual code consists of the complete candidate sequence numbers of three units. In the initial population, the candidate sequences of each unit are first arranged in the output order of S2, and then generated in the cyclically increasing combination order of the three sequence positions, skipping duplicate individuals and hard-constrained infeasible individuals, until 160 individuals are obtained. If there are fewer than 160 individuals, feasible individuals obtained can be used to fill the gap in lexicographical order, but each copy retains the same objective value. Parent selection adopts a binary tournament, comparing non-dominated level, crowding distance and individual code lexicographical order in turn. The unit-level crossover trigger rate is set to 0.80, and the individual mutation trigger rate is set to 0.10. The crossover operation only swaps the unit-level sequence numbers, and the mutation operation uses pseudo-random numbers to re-extract a complete sequence from the same unit's dynamic reachable load tube. This period's pseudo-random seed is the unsigned integer represented by the first 16 hexadecimal characters of the SHA-256 digest of the concatenated text containing the period number, the three load tube content summaries, the model version, the constraint version, and the solution configuration version. The seed, pseudo-random number generator version, and call count are saved with the solution configuration digest. Repair operations prioritize replacing unit sequences with corresponding load increase / decrease capabilities based on the total feed gap, and re-check the total feed, product demand, shared cooling capacity, and shared heating capacity hard constraints after each operation. In each generation, the parent and child generations are merged and re-sorted for non-dominated status. The elite archive retains all feasible non-dominated individuals; when there are more than 160 individuals, they are sorted by crowding distance from largest to smallest and then broken up in lexicographical order by individual codes, and the first 160 are truncated. After 120 iterations, 20 consecutive generations, or 45 iterations since S3, the non-dominated frontier no longer changes. The calculation stops when the deadline is reached. If the first two conditions are not met at the deadline, the result is selected from the current feasible elite file according to the bottleneck margin rule described later. If the elite file is empty, the branch where the system's dynamic feasible domain is empty is switched. The same input, version, seed and calculation deadline generation will result in the same non-dominated solution set. If one combination is not inferior to another combination in all five normalized objectives and is superior in at least one objective, then the former combination dominates the latter, and the objective solving module forms a non-dominated solution set accordingly. When the total feed and shared resource constraints cannot be satisfied simultaneously, the module does not disguise the solution as a feasible solution with a penalty value, but instead outputs an independent system dynamic feasible domain as empty, hard constraint conflict category, conflict prediction step size, and a flag prohibiting the generation of load increase batches. No target load trajectory that can be executed normally by S4 is generated. The module searches for a safe hold or load reduction combination that only satisfies equipment safety, product quality, and the upper limit of shared cooling and heating capacity. The deviation allowed for this downgraded combination is only the total feed or product demand of the production plan. Its planning gap is recorded and marked as hold or load reduction permission only. If the downgraded combination does not exist, all new instructions are frozen and the most recently confirmed safety setting is maintained. When S4 receives the empty domain state, it must not generate load increase instructions, but can only execute the above-marked hold or load reduction permissions until the operator adjusts the production plan, the shared resource availability is updated, or the state change in the next control cycle triggers a re-solution starting from S2. For each cell, each prediction step, and each running constraint of each non-dominated solution, read the constraint margin stored in S2 and divide it by the allowable range of that constraint, according to... Determine the bottleneck margin. The symbol represents the first The bottleneck margin of each non-dominated solution is dimensionless. The symbol represents the first The first distillation unit Class execution constraints in the first The dimensionless margin from each prediction time to the corresponding boundary The symbol indicates the number of times the operation constraint table is accessed. Unit 1, the 1st The class constraints, product brand, and positive allowable span read from the running range to which the prediction step belongs are used to generate the corresponding... The constraint version and boundary used must correspond one-to-one, and the corresponding margin unit must be the same and greater than zero. The symbol represents a discrete element index with a value of 1 to 3, and the unit is not applicable. The symbols represent 11 categories of discrete constraint indices, including product purity, recovery rate, tray or packing load, reflux ratio, temperature, pressure, pressure drop inside the column, condensation load, reboiling load, load change rate, and equipment safety margin. The units are not applicable. The symbol represents a discrete prediction step size index with values ​​from 0 to 11, and the unit is not applicable. The symbol represents the discrete solution index from 1 to the total number of non-dominated solutions in this round, and the unit is not applicable; when the span is missing, zero or negative, the corresponding solution is marked as unevaluable and removed from the candidate; the normalization margin less than zero indicates that it has exceeded the limit and removes the solution from the candidate; the larger the bottleneck margin, the larger the relative margin that the most unfavorable constraint still retains. The target solution module first removes any solutions that do not meet quality constraints or equipment safety constraints from the non-dominated solution set, and then selects the solution with the largest bottleneck margin as the target load trajectory. When the bottleneck margin difference of multiple solutions does not exceed one ten-thousandth, the solution with the smallest unit product comprehensive energy consumption value is selected first, and then the solution with the smallest total load action variation value is selected. If they are still the same, the unique result is determined according to the lexicographical order of the candidate sequence number, so that the same input and version can obtain a definite output. The selected results are written to the target load trajectory object according to the prediction step size. This object contains the target feed rate of each unit, the expected product flow rate and composition, the expected recovery rate, the return flow rate, the condensation load, the reboiling load, the constraint margin, the selected candidate sequence number, the bottleneck constraint location, the bottleneck margin, the model version, the constraint version, and the solution configuration summary. The target load trajectory and the summary of the non-dominated solution set are saved to the solution result library. Only when the solution status is normal feasible, the target load trajectory serves as the sole input for S4 to calculate the difference between the supply and load and generate control commands. When the solution status is empty, the aforementioned execution of the disable flag and the safety maintenance or load reduction object constraint S4 permissions are used instead.

[0023] In this specific embodiment, S4 includes: The batch execution module reads the current S1 status, the S3 solution status, the target load trajectory, and the S2 dynamic reachable load tube. Only when the S3 solution status is normal and feasible, the load difference of each unit is determined by subtracting the current feed amount from the feed amount at the first predicted time of the target. Units with negative differences are registered as supply units and their load reduction amount is recorded. Units with positive differences are registered as load-receiving units and their load increase amount is recorded. Units with an absolute difference value not exceeding one-thousandth of the current feed amount retain their original settings. When the S3 solution status is that the system's dynamic feasible domain is empty, the batch execution module locks the load increase permission and does not read non-existent target load trajectories. It only generates hold or reduce records according to the safety hold or reduce objects marked by S3. If S3 does not generate the object, all new instructions are frozen. The supply, load, and degradation records jointly store the unit number, current load, target load, difference direction, execution permission, dynamic reachable load tube version, and target trajectory or degradation object version. Before determining the batch upper limit, the batch execution module reads the measured values ​​and corresponding predicted values ​​from the historical state database, and then executes them according to... Forming residuals The symbol represents the first The first distillation unit Class operation constraints in load direction and prediction step size The predicted residuals are given with the same units as the corresponding measured values. The symbol represents the measured field paired with the actual timestamp. The symbols represent the predicted output fields of S2 with the same timestamp. Constraint categories 1 and 2 correspond to product purity and recovery rate, respectively, with units of dimensionless proportions. Category 3 corresponds to temperature, with units of degrees Celsius. Categories 4 and 5 correspond to pressure and pressure drop in the tower, respectively, with units of kPa. Categories 6 and 7 correspond to condensing load and reboiling load, respectively, with units of kW. The measured fields are derived from the current state of S1 or the recovery rate calculation records described below, while the predicted fields are derived from the corresponding components of the S2 output vector. The symbol represents a discrete element index with a value of 1 to 3, and the unit is not applicable. The symbol represents a discrete constraint category index with values ​​from 1 to 7, and the unit is not applicable. The symbol represents a discrete directional code that takes values ​​for increasing or decreasing load, and the unit is not applicable. The symbol represents a discrete prediction step size index with a value of 0 to 11 and the unit is not applicable; only historical period samples with valid model version, constraint version and measurement quality bits are used. The actual recovery rate is based on calculate, The symbol represents the first The target carbonate component recovery rate for each distillation unit is a dimensionless number. The symbol represents the product flow rate and the unit is kg / h. The symbol represents the mass fraction of the target carbonate component in the product and is a dimensionless number. The symbol indicates the feed flow rate and the unit is kg / h. The symbol represents the mass fraction of the target carbonate component in the feed and is a dimensionless number. The symbol represents the discrete distillation unit index with a value of 1 to 3 and the unit is not applicable; when the feed flow rate or feed composition is missing, zero, or the mass position is invalid, the recovery rate residual for that cycle is not generated and the sample is marked as unusable for tightening updates; The residual samples are grouped by distillation unit, operating constraint category, load increase or decrease direction, and prediction step size. The residual direction is uniformly converted to a positive value increase, i.e., a direction that is closer to the unfavorable boundary. For product purity and recovery rate, the predicted value is subtracted from the measured value. For temperature, pressure, column pressure drop, condensing load and reboiling load with upper limits, the measured value is subtracted from the predicted value. For conditions with lower limits, the predicted value is subtracted from the measured value. Each group retains the most recent 60 effective control cycles and requires at least 20 samples. If the sample is insufficient, the approved tightening amount of the previous version is used instead of extrapolating from a small number of samples. For each group of adverse direction residuals, sort them in ascending order of value and take the one-sided quantile value covering 95% of the samples, then follow... The amount of safety tightening by each constraint is obtained. The symbol represents the safety tightening amount of the corresponding group, and the unit is the same as the first group. Class constraints are the same. The symbol represents the 95th percentile of the unfavorable direction residuals for that group. The symbol represents the first The first distillation unit The measurement error limits determined by calibration for this type of measuring device. The symbol represents a discrete element index with a value of 1 to 3, and the unit is not applicable. The symbols represent discrete constraint categories indexed by values ​​from 1 to 7, coded in the aforementioned order of purity, recovery rate, temperature, pressure, pressure drop in the column, condensation load, and reboiling load, and the units are not applicable. The symbol represents a discrete directional code that takes values ​​for increasing or decreasing load, and the unit is not applicable. The symbol represents the discrete prediction step size index with a value of 0 to 11 and the unit is not applicable; the quantile value and the measurement error boundary use the same unit, and after addition, they are trimmed according to the allowable range of the corresponding constraint and must not be negative; For lower limit constraints, the original lower limit is moved to the feasible side with increasing value by a corresponding safety tightening amount. For upper limit constraints, the original upper limit is moved to the feasible side with decreasing value by a corresponding safety tightening amount. The dynamic reachable load tube of the load-bearing unit is re-checked using the correction boundary. The maximum load increase corresponding to the deleted sequence after correction is deducted from the load-bearing capacity. The batch upper limit is reduced accordingly by the difference between the load increase before and after correction. The tightening table saves the unit number, constraint type, direction, prediction step size, number of samples, quantile value, measurement error boundary, tightening amount, applicable model version and update time. The batch execution module determines the upper limit of a single batch based on the revised available load capacity, available load capacity, shared cooling capacity, and available shared heating capacity. calculate, The symbol indicates the upper limit of the feed load for a single load transfer batch, and the unit is kg / h. The symbol represents the sum of the reduced load capacity after corrections for all load-bearing units, and the unit is kg / h. The symbol represents the sum of the increased load capacity after correction of all loaded units, and the unit is kg / h. The symbol represents the available cooling capacity after deducting reserve capacity and occupied tokens, and the unit is kW. The symbol represents available heat after deducting reserve and occupied tokens, and the unit is kW. The symbol represents the maximum positive marginal coefficient of condensing capacity required per unit feed increment among the load candidates, and the unit is kW per kg / h. The symbol represents the maximum positive marginal coefficient of the corresponding reboiling capacity and the unit is kW per kg / h; the two marginal coefficients are calculated only from the effective adjacent load points in the S2 load increase direction where the feed difference is greater than zero and the heat exchange load difference is positive and must be greater than zero. If either coefficient is missing, zero or negative, the corresponding resource conversion is not performed and the load candidate is frozen until a verified positive marginal coefficient is obtained. Servers are arranged in descending order of bottleneck margin and in descending order of load supply capacity. The total load difference is divided into multiple batches, each not exceeding the upper limit of a single batch, and then... Verify batch balancing. The symbol represents the first Each batch of supply unit indices is a non-empty set and the unit is not applicable. The symbol represents the first Each batch must be a non-empty set of loaded cell indices with no applicable units. The elements of both sets must be cell indices 1 to 3 and must be mutually exclusive within the same batch. A batch is not generated if either set is empty. Symbols represent supply units In the The load reduction for each batch, in kg / h. Symbols represent loaded units In the The load increase for each batch, in kg / h. The symbol represents a discrete batch index numbered sequentially from 1 to the total number of batches in the current period, and the unit is not applicable. The symbol represents the discrete distillation unit index with a value of 1 to 3 and the unit is not applicable; the last paired quantity is the minimum value of the remaining quantities of both parties, and the remaining difference that cannot be balanced is carried over to the next control cycle and is not written into the control command. For each balanced batch, the batch endpoint can only be selected from adjacent predicted nodes of the same retained candidate sequence in S2, or from values ​​within a continuously reachable interval between adjacent nodes verified by the same model version. The dynamic response model divides the endpoint feed change into three equal-slope set trajectories for three control cycles, and synchronously generates corresponding set trajectories according to the model-predicted reflux ratio, condensation load, and reboiling load. The same state recursion, nine-item output calculation, and all 11 types of operational constraint screening are re-executed for the three intermediate set points. After all three pass, an independent batch prediction trajectory version is generated. If any intermediate point fails, the batch quantity is halved and regenerated and screened. This halving is repeated three times consecutively. If the batch still fails to pass, the batch is cancelled and no token is used. The control instruction record includes batch number, unit number, execution direction, start and end time, feed setpoint sequence, reflux setpoint sequence, condensation setpoint sequence, reboiler setpoint sequence, rate of change limit, pre-token condition, source candidate sequence number, batch prediction trajectory version and cancellation status. When the next control cycle is re-solved, the new target trajectory takes priority over the remaining amount of the old batch that has not yet been issued. The old batch instructions that have not been issued are cancelled. The measured and effective supply and load status is used as the initial state of the new cycle S1. The tokens that have been used remain occupied until the corresponding old instructions are cancelled or the resource is confirmed by actual measurement. Then, the balance and rearrangement are re-adjusted according to the new target trajectory. In this specific implementation, 5 kW condensing capacity is defined as a cooling capacity token, and 5 kW reboiling capacity is defined as a heating capacity token. The load demand of the load-bearing unit is calculated by dividing the difference in condensing load and reboiling load before and after the predicted load increase by the corresponding token capacity and then rounding up. The public standby resource token is generated from the unoccupied standby capacity in the S1 shared resource record. The load release token is only generated after the load reduction command is executed, when the heat exchange load reduction calculated by the refrigerant flow rate or heat flow rate and the supply-return enthalpy difference is confirmed by actual measurement, and when the direction of the regulating valve position change is consistent with the capacity release direction. Each token records the resource category, source unit, capacity value, measurement timestamp, occupied batch, and validity period. Only after the loading unit obtains a quantity of available cold energy tokens and available heat energy tokens that meet the corresponding demand can the corresponding load increase command be executed. Only then will the batch execution module release the pre-lock of its feed, reflux, condensation, and reboiling load increase commands. If any type of token is insufficient or token pairing is not feasible, the current setting of the loading unit will be maintained. The supply unit will only execute a load reduction microstep not exceeding one-quarter of the batch upper limit when the existing inventory of the product tank can cover the product demand of two control cycles and the total feed gap of the system after the microstep load reduction does not exceed the transient allowable amount of the production plan. The new tokens generated by the supply release will enter the available queue sorted by measurement timestamp. Tokens shall not be used repeatedly between two batches, nor shall they be replaced by heat energy tokens or vice versa. Each load supply / reduction microstep and a load loading microstep use a timestamp handshake. After the load supply release is confirmed by actual measurement, a matching load loading command must be issued within one control cycle. The batch execution module rearranges the subsequent feed, reflux, condensation, and reboiling trajectories of the batch according to the actual effective time and rechecks the total feed, product demand, and shared resource constraints. If matching cannot be performed within the specified time, the load supply unit is restored to the setting before the microstep, or the S3 system combination is recalled based on the current actual measurement state. The actual load supply waiting period, the rearranged prediction timestamp, and the command effective time are all transmitted to S5. S5 marks the residual caused purely by token waiting as a time-series residual and excludes it from the model bias update. For each load transfer batch, a resource release confirmation period and status trend verification conditions are set. In this specific implementation, the resource release confirmation period is set to two control cycles. Within the period, the batch execution module checks the measured heat exchange load decrease of the supply unit cycle by cycle. At the same time, it reads the normalized constraint margin of three consecutive control cycles. When the linear trend points to the unfavorable boundary and extrapolating the trend by two control cycles will result in a value lower than the boundary after safety tightening, it is determined that the status trend does not meet the verification conditions. The unexecuted load increase instructions are frozen, and feed, reflux, condensation, and reboiling reversal instructions are generated in the opposite direction of the executed load increase. If the token is not confirmed by actual measurement within the confirmation period, the same freezing process is also performed, and the batch upper limit of the next load transfer batch is reduced to no more than the load increase that has been confirmed by actual measurement and can be safely executed. The target margin zone is jointly determined by the normalized margins of product purity, recovery rate, temperature, pressure, pressure drop in the tower, condensing load, and reboiling load up to the safe tightening boundary. In this specific implementation, it is required that each normalized margin is not less than 0.20 for three consecutive control cycles. After the loaded unit enters this zone, the batch execution module first recalculates the total shared cooling capacity and shared heating capacity based on the latest refrigerant flow rate, heating flow rate, supply and return enthalpy difference, and heat exchange load. The measured condensing load increment and reboiling load increment after the loaded unit is added are transferred to the current cycle's continuous resource occupancy baseline. Then, the corresponding batch token is cancelled and subsequent available tokens are generated based on the recalculated net remaining capacity. If the resource recalculation fails, the measurement quality bit is invalid, or the resource version is inconsistent, the original token is retained and subsequent additions are prohibited. Tokens that have not entered the target margin zone are retained until the batch withdrawal is completed or the operator performs a safe release after authorization verification. The load transfer batch object is written with the load supply and load receiving relationship, balance verification results, tightening version, token requirement, token source, actual measured confirmation value, control command, freeze or pullback status, and the upper limit of the next batch for each batch. Control commands that have been released from the pre-lock are sent to the distributed control system in batch order. The batch object and the corresponding prediction trajectory are simultaneously transmitted to S5 for verification of prediction results and updating of the dynamic reachable load tube for the next control cycle according to the actual execution timestamp.

[0024] In this specific embodiment, S5 includes: During the execution of each load transfer batch, the rolling verification module locks the predicted trajectory, mechanism model version, output deviation correction version, tightening amount version, and control command version used in that batch. It receives measured values ​​of feed flow rate, product flow rate, top pressure, bottom temperature, internal pressure drop, condensing load, and reboiling load at 5-second intervals and online product composition values ​​at 60-second intervals. It reads the most recently approved and valid feed composition from the current state of S1, and the data timeliness does not exceed the specified upper limit. The feed composition and product composition obtained from the test are backfilled according to the sampling timestamp of S1. The module locates the corresponding timestamp of the predicted trajectory according to the actual command effective time plus the equipment communication delay correction amount. The load waiting period recorded in S4 uses the rearranged predicted timestamp. Measured records that cannot be matched after more than one sampling cycle are marked as mismatched and do not participate in model updates. For each paired timestamp, the module uses the recovery rate calculation formula described in S4 to generate a recovery rate measurement record using the feed flow rate, product flow rate, feed composition, and product composition backfilled with the same timestamp or the sampling timestamp. It also saves the four inputs, their respective quality levels, composition sample timestamp, calculation timestamp, and the composition version used. If the denominator of the feed flow rate or feed composition is not greater than zero, any input is missing, there is a fault, the time limit is exceeded, or the timestamp cannot be paired, the recovery rate at that moment is marked as invalid, the normalized margin calculation and deviation update for that item are stopped, the old recovery rate is not used to replace it, the loaded unit retains the current setting and new loads are prohibited until a new valid recovery rate record is obtained, and other constraints with valid measured values ​​continue to be checked. For each monitored state, according to Calculate the normalized margin of the bilateral constraints. Symbols represent time. No. The first distillation unit The normalization constraint margin for the class state is a dimensionless number. The symbols represent the measured fields with valid mass values ​​in the current state of S1. Constraint categories 1 and 2 represent dimensionless product purity and recovery rate, respectively; category 3 represents temperature in degrees Celsius; categories 4 and 5 represent pressure and pressure drop in the column in kPa; and categories 6 and 7 represent condensation load and reboiling load in kW. The symbol represents the lower boundary of the same unit, read from the currently valid running constraint table and modified by the S4 safety tightening version. The symbol represents the upper boundary of the same unit obtained from the same table and the same tightening version. The symbol represents the difference between the upper and lower boundaries, and must be in the same unit as the measured value and greater than zero. The symbols represent discrete distillation unit indices with values ​​from 1 to 3, and the units are not applicable. The symbol represents a discrete constraint category index with values ​​from 1 to 7, and the unit is not applicable. The symbol represents the actual measured timestamp in seconds, accumulated from 00:00 on January 1, 1970, using the server's unified clock. When the upper boundary is not greater than the lower boundary, the boundary is missing, or the version is inconsistent, the calculation of the constraint margin is stopped, the corresponding load is frozen, and the constraint table is recorded as abnormal. In the case of a single boundary, only the distance to the boundary is calculated and then divided by the corresponding approved allowable range that is greater than zero. The lower limit of the equipment safety margin is read from the operating constraint table according to the unit number and constraint category. In this specific embodiment, the normalized safety margin lower limits of product purity, recovery rate, temperature, pressure, pressure drop in the tower, condensing load and reboiling load are all set to 0.10. When any measured normalized margin is lower than its lower limit, or when three consecutive sampling points show a monotonically decreasing trend and extrapolating one control cycle based on the current slope will result in a lower limit, the rolling verification module marks the current batch as having insufficient constraint margin and stops issuing load increase instructions for this batch and subsequent batches that have not yet been executed. If the normalized margin is still non-negative but below the lower limit of the safety margin, the module freezes the current feed, reflux, condensation, and reboiling settings and maintains the achieved safe operating point, waiting for a control cycle to recalibrate; if the normalized margin is negative, the product purity or recovery rate exceeds the quality boundary, the pressure drop in the tower exceeds the hydraulic upper limit, or the condensation and reboiling loads exceed the equipment capacity upper limit, the module generates a pullback trajectory in the opposite direction of the load change amount executed in the current batch, the pullback change rate does not exceed the upper limit of the load change rate of S2, and the condensation and reboiling settings are restored to the safe values ​​before the batch execution. Measured and predicted values ​​are used to form verification records by unit, constraint category, and prediction timestamp. Tokens marked with S4 awaiting time-series residuals do not enter the deviation update. When the measurement quality position is valid, the control command is actually effective, and there is no manual intervention, according to... Update the output deviation correction. The symbol represents the first value generated and written into the output deviation version to be released in this calculation. The first distillation unit Class prediction output bias correction amount The symbols represent the corresponding correction amounts read from the current valid output deviation version. Constraint categories 1 and 2 correspond to product purity and recovery rate, with units of dimensionless proportions; category 3 corresponds to temperature, with units of degrees Celsius; categories 4 and 5 correspond to pressure and pressure drop within the column, with units of kPa; and categories 6 and 7 correspond to condensation load and reboiling load, with units of kW. The symbol represents the number of times the current valid update coefficient version is read. The class outputs dimensionless update coefficients. When the system is first deployed or no verified coefficient version is available, the default value is 0.20. The symbol indicates that the measured value of the same timestamp minus S2 has been superimposed. Since the remaining residuals obtained from the predicted output have the same units according to the above categories, this formula will not repeatedly deduct the old bias from these residuals. The symbol represents a discrete element index with a value of 1 to 3, and the unit is not applicable. The symbol represents a discrete output category index with a value of 1 to 7 and the unit is not applicable; the update amount is limited to ±20% of the range allowed by the corresponding running constraints, and residuals exceeding the range are only entered into the abnormal sample library without directly changing the model; For the residuals generated by the product composition test backfill, the module replays the deviation correction calculation from the corresponding sampling time to the current time after the test review is completed, but does not retroactively rewrite the control instructions that have already been executed; for the high-frequency residuals of pressure, temperature, tower pressure drop and heat exchange load, the median value is first taken within one control cycle, and then used as the deviation correction input for that cycle, so as to reduce the impact of a single pulse measurement on the model. The deviation correction objects after backfilling or replaying are all generated with a new version and the previous version is retained for rollback; The deviation correction section performs calibration every 60 effective control cycles. The first 48 cycles are used as the calibration set, and the last 12 cycles are used as the validation set. Within the range of 0.05 to 0.30, update coefficients for each output category are searched at 0.05 intervals to minimize the sum of the normalized average absolute errors of product composition, recovery rate, pressure, temperature, column pressure drop, condensate load, and reboiling load on the calibration set. Each type of error is divided by the positive allowable span of the corresponding category in the current effective operating constraint table. If multiple update coefficients have the same error, the update coefficient with the smallest value is used. If any category's allowable span is missing or not positive, or if an effective sample is not found... If there are fewer than 20 valid coefficient versions, or if any output error in the validation set exceeds 5% of the corresponding allowable operating range, the version to be released will be rejected and the previous valid version will continue to be used. The current valid coefficient version released after calibration will replace the initial default value of 0.20. The coefficient version records seven types of coefficients: version number, effective period, product composition, recovery rate, pressure, temperature, pressure drop in the tower, condensing load and reboiling load, various allowable span versions, validation status and rollback version, and is bound to the output deviation correction version generated using the coefficient. When calibration is rejected, version is revoked or rolled back, the deviation update formula will continue to read the previous valid coefficient version. The model verification record also saves the current state vector, the actual effective feed, reflux, condensation and reboiling control vectors, feed composition and temperature disturbance vectors, and the predicted output vector. When updating the model, only records with complete input fields and successfully matched timestamps are used. The running inference still generates a state vector without deviation terms first according to the mechanism state recursion function determined in S2, and then generates the basic output by the output mapping function and superimposes the current effective output deviation vector. The deviation correction part only corrects the nine predicted output components without entering the state recursion, and does not change the material balance, energy balance and gas-liquid balance structure. The seven types of monitored output deviations released in S5 are written into the S2 output deviation vector according to the same category. The two unmeasured components, tray or packing load and reflux ratio, remain at the previous verified value. Before the release of the new deviation correction version, the module uses the validation set to generate 12-step predicted trajectories before and after the update, compares the errors in product purity, recovery rate, temperature, pressure, pressure drop in the column, condensing load, and reboiling load, and re-executes all operating constraint screening. The new version is only released if the overall normalized error is not higher than that of the old version and the actual out-of-bounds trajectory in the validation set is not misjudged as a feasible trajectory. Otherwise, the new version is withdrawn, the rejection reason is recorded, and the previous valid version continues to be used. At the start of the next control cycle, the load tube generation module replaces the predicted initial state of the previous cycle with the latest measured current state, replaces the old version with the new output deviation correction version that has been verified and released by S5, and reads the constraint-by-constraint safety tightening amount recalculated by S4 based on the most recent 60 effective control cycles. Candidate sequences that were previously frozen or withdrawn due to insufficient constraint margin are marked as restricted samples. When the same unit, direction and constraint category are frozen twice in a row, the candidate load change level in the corresponding direction is reduced from 5% to 2.5% until the normalized margin is not less than 0.20 for three consecutive control cycles before the original candidate level is restored. The updated dynamic response model re-performs state recursion, output calculation, and time-by-time constraint filtering for the 12 prediction steps to generate a new dynamic reachable load tube version. If the mean absolute error of product purity or recovery rate on the validation set after the model update exceeds 10% of the previous version, the deviation correction version is withdrawn and the previous effective model is used. At the same time, the latest safety tightening amount is retained so that the model accuracy regression does not relax the safety boundary already determined by the residuals and measurement errors. The rolling verification result object stores the batch number, measured timestamp, predicted timestamp, each measured value, predicted value, residual, normalized constraint margin, reason for freeze or pullback trigger, actual pullback amount, version before and after deviation correction, new version of dynamic reachable load management, and token status. This object is written to the historical state database and provides S1 with the data timeliness and backfill basis for the next cycle, provides S2 with the updated model and initial state, and provides S4 with residual grouping and the measured and confirmed safe load increase, thereby completing the closed loop of prediction, execution, verification, and re-solution according to the control cycle.

[0025] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0026] This invention inputs the state alignment results and candidate load sequences of each distillation unit into a dynamic response model. Dynamically attainable loads are screened based on mass, energy consumption, hydraulics, temperature and pressure, and condensation and reboiling trajectories. Then, the unit boundaries are combined with total feed and shared resource constraints to perform a multi-objective non-dominated solution. This combination confines the multi-objective selection within the system's dynamic feasible region, preserving cross-time-series reachability, and the resulting target trajectories can be directly used for subsequent load differential decomposition.

[0027] This invention corrects the predicted boundary and batch upper limit by adjusting the safety tightening amount of each constraint, and uses cold energy tokens and heat energy tokens to characterize the measured release capacity of public backup resources or load supply units. After the load receiving unit obtains the corresponding resources, it performs load increase, and releases the token after entering the target margin area; when a timeout is confirmed or the status tends to exceed the limit, the system freezes, reverts, and reduces the next batch, thereby improving the consistency between dynamic transition commands and the actual equipment and shared resource status.

Claims

1. A method for adjusting the load of carbonate cryogenic distillation based on multi-objective solution, characterized in that, include: S1. Collect process measurement data and shared cold and heat resource availability of multiple distillation units according to the control cycle, and align the process measurement data with timestamps to obtain the current status of each distillation unit; S2. Based on the current status and candidate load sequence, use the dynamic response model of each distillation unit to predict the quality indicators, equipment status and cold and hot load trajectories, and filter them step by step according to the preset operating constraints to obtain the dynamic reachable load pipe of each distillation unit. S3. Combine the dynamic reachable load pipe with the total feed conservation, shared cooling capacity, shared heating capacity and product demand constraints to form the system dynamic feasible domain. Within the system dynamic feasible domain, perform multi-objective non-dominated solutions for quality indicators, energy consumption and load action, and select the target load trajectory from the solutions that satisfy the quality constraints and safety constraints in the preset operating constraints according to the bottleneck margin. S4. Based on the difference between the current state and the target load trajectory, generate load transfer batches and corresponding control commands that match the load reduction of the supply unit and the load increase of the receiving unit. Load increase is only executed when the required cooling and heating resources of the receiving unit are met by the reserve quantity or the release quantity of the supply unit as confirmed by actual measurement. S5. Verify the predicted trajectory based on the measured data during the execution of the load transfer batch. If the constraint margin is insufficient, freeze or roll back the current load transfer batch, and update the dynamic reachable load tube for the next control cycle based on the verification results.

2. The method for adjusting the load of carbonate cryogenic distillation based on multi-objective solution according to claim 1, characterized in that, S1 includes: Collect the feed flow rate, feed composition, feed temperature, column top pressure, column bottom temperature, column pressure drop, reflux flow rate, condensation load, reboiling load, product flow rate, and product composition for each distillation unit; Using the end of the control cycle as the alignment time, window statistics are performed on process measurement data that generate multiple sample values ​​within a control cycle. Product composition test data returned later than the alignment time are backfilled according to the sampling time, and the most recent valid measurement value and its timestamp are used to characterize data that has not been updated in the current control cycle, forming the current state containing data time stamps.

3. The method for adjusting the load of carbonate cryogenic distillation based on multi-objective solution according to claim 1, characterized in that, S2 include: The dynamic response model includes a mechanistic model based on material balance, energy balance, and gas-liquid balance, as well as a model deviation correction part determined based on historical operating data. The quality indicators include product purity and recovery rate, and the preset operating constraints include lower limit of product purity, lower limit of recovery rate, tray load range or packing load range, reflux ratio range, temperature range, pressure range, upper limit of condensation capacity, upper limit of reboiling capacity, upper limit of load change rate, and lower limit of equipment safety margin. Candidate load sequences that satisfy the preset operating constraints at each prediction time within the prediction time domain are retained, and the set of retained candidate load sequences constitutes the dynamic reachable load tube, so that the dynamic reachable load tube retains the state recursion relationship and load change relationship determined by the dynamic response model between adjacent prediction times.

4. The method for adjusting the load of carbonate cryogenic distillation based on multi-objective solution according to claim 1, characterized in that, S3 includes: The total feed conservation constraint and the product demand constraint are established so that the sum of the feed rates of each distillation unit equals the total feed rate of the system, and the flow rate and quality indicators of each product meet the product demand setpoints of the current control cycle. The upper limit for shared cooling capacity and the upper limit for shared heating capacity are established based on the sum of the condensation load and the sum of the reboiling load of each distillation unit not exceeding the value after deducting the preset reserve margin from the available amount of the corresponding shared resources. The nondominated solution set is determined by normalizing the quality indicators, unit energy consumption, processing capacity, safety margin, and total load action variation in the same direction. The load trajectory of each distillation unit in each non-dominated solution belongs to the candidate load sequence contained in the corresponding dynamically reachable load tube. For each non-dominated solution, its various operational constraint margins are normalized according to the corresponding allowable range, and the lower bound of the normalized margin is used as the bottleneck margin. The solution with the largest bottleneck margin is selected from the non-dominated solutions that satisfy the quality constraints and safety constraints in the preset operational constraints as the target load trajectory.

5. The method for adjusting the load of carbonate cryogenic distillation based on multi-objective solution according to claim 1, characterized in that, S4 includes: The maximum load limit for a single load transfer batch is determined based on the load reduction capacity of the supply unit, the load increase capacity of the receiving unit, and the available capacity of shared cooling and heating. The total load difference is divided into multiple load transfer batches not exceeding the upper limit of the batch, so that the sum of the load reduction of the supplying units in the same load transfer batch is equal to the sum of the load increase of the receiving units. The feed change trajectory for each load transfer batch is determined based on the dynamic response model, and the corresponding set trajectories for reflux, condensation load and reboiling load are determined based on the feed change trajectory, thereby generating the feed command, reflux command, condensation command and reboiling command.

6. The method for adjusting the load of carbonate cryogenic distillation based on multi-objective solution according to claim 1, characterized in that, S5 include: During each load transfer batch, the measured pressure, temperature, tower pressure drop and product composition are compared with the predicted values ​​of the predicted trajectory at the corresponding timestamp, and the constraint margin is determined based on the distance of each measured value to the corresponding operating constraint boundary. When any constraint margin is less than the lower limit of the corresponding equipment safety margin, an exception is handled, the issuance of subsequent load increase commands is stopped, and the current command is maintained or a rollback command is issued in the opposite direction of the current load transfer batch. The difference between the measured value and the predicted value is written into the deviation correction part of the corresponding dynamic response model to update the predicted trajectory and dynamic reachable load tube for the next control cycle.

7. The method for adjusting the load of carbonate cryogenic distillation based on multi-objective solution according to claim 5, characterized in that, The residuals formed by the difference between the measured value and the predicted value under the same correspondence within the historical control cycle are grouped according to the distillation unit, operating constraint category, load increase or load decrease direction, and prediction step size. The measured value of recovery rate is determined according to the feed flow rate, feed composition, product flow rate and product composition in the current state. For each group, along the direction that reduces product purity or recovery rate or makes the temperature and pressure state, the hydraulic state in the tower, the condensation load or the reboiling load approach the corresponding operating constraint boundary, determine the one-sided residual boundary that covers the preset proportional residual of the group. The unilateral residual boundary is added to the measurement error boundary of the corresponding measuring device to obtain the constraint-by-constraint safety tightening amount; The corresponding operating constraint boundary is moved toward the feasible side by the constraint tightening amount to correct the dynamically reachable load tube, and the upper limit of the batch is reduced according to the difference in the load capacity of the load unit before and after the correction.

8. The method for adjusting the load of carbonate cryogenic distillation based on multi-objective solution according to claim 7, characterized in that, In each load transfer batch, the condensing capacity and reboiling capacity required for the load-bearing unit to increase the load are converted into the demand for cooling tokens and the demand for heating tokens, respectively. The condensing capacity and reboiling capacity released by the common reserve resources or the load-bearing unit after load reduction and confirmed by actual measurements of refrigerant flow, heat flow, regulating valve position or heat exchange load are converted into available cooling tokens and available heating tokens, respectively. The corresponding load increase command is allowed to be executed only after the loaded unit obtains a quantity of available cooling tokens and available heating tokens that meet the corresponding demand. The state region where the margin of each monitored state to the corresponding operating constraint boundary after the constraint tightening amount is corrected reaches the preset target margin lower limit is defined as the target margin region. After the load unit enters the target margin region, the available cold energy token and available heat energy token occupied by the load unit are released.

9. The method for adjusting the load of carbonate cryogenic distillation based on multi-objective solution according to claim 8, characterized in that, Set resource release confirmation deadlines and status trend verification conditions for each load transfer batch; If the demand for cold or heat tokens to be met by the load reduction release from the load supply unit is not confirmed by actual measurement within the resource release confirmation period, or if any operating state tends to exceed the operating constraint boundary after being corrected by the constraint tightening amount based on the constraint margin of multiple consecutive control cycles, the unexecuted load increase command is frozen or a rollback command is generated according to the executed load increase amount, and the upper limit of the next load transfer batch is reduced to no more than the load increase amount that has been confirmed by actual measurement to be safe to execute.

10. A multi-objective solution-based carbonate cryogenic distillation load allocation system, used to execute the multi-objective solution-based carbonate cryogenic distillation load allocation method according to any one of claims 1 to 9, characterized in that, include: The state construction module is used to collect process measurement data and the availability of shared resources and align them with timestamps to obtain the current state of each distillation unit; The load tube generation module is used to generate the dynamic reachable load tubes for each distillation unit based on the current state, candidate load sequence, and dynamic response model. The objective solution module is used to construct the dynamic feasible region of the system and perform multi-objective non-dominated solution within the dynamic feasible region of the system to obtain the target load trajectory; The batch execution module is used to generate load transfer batches and corresponding instructions, and to control the load increase of the load-receiving unit based on the measured confirmation of the corresponding shared resource reserve or the corresponding capacity release of the supply unit, according to the condensation capacity and reboiling capacity required for the load increase of the load-receiving unit. The rolling verification module is used to verify the predicted trajectory based on the measured data, and to freeze or roll back the current load transfer batch and update the dynamic reachable load tube for the next control cycle when the constraint margin is insufficient.

Images