Electrolyte solvent refining method

Abstract

The application discloses an electrolyte solvent refining method, and belongs to the electrolyte solvent refining field. In order to solve the problem that the purity is unstable due to the fluctuation of trace water, alcohol residues, metal ions and heavy impurities in the deep refining of ethylene carbonate and methyl ethyl carbonate, the application adjusts the path adaptive control action through the linkage of the raw material impurity fingerprint, the hierarchical pretreatment, the differentiated deep refining path, the soft measurement module including the prediction submodel and the migration risk calculation submodule and the model predictive controller, and executes the reprocessing priority when the safety preposition exception, the agent low confidence or the consistency verification failure, so that the technical effect of stably obtaining the battery-grade electrolyte solvent under the batch fluctuation is realized.

Description

Technical Field

[0001] This invention relates to the field of electrolyte solvent purification, and more particularly to a method for electrolyte solvent purification. Background Technology

[0002] Ethylene carbonate and ethyl methyl carbonate are commonly used solvents in lithium-ion battery electrolytes. Their content of water, residual alcohols, metal ions, and heavy impurities can affect subsequent lithium salt dissolution, additive compatibility, and the consistency of the battery system. Existing deep refining processes typically rely on unit operations such as water removal adsorption, metal trapping, protective bed treatment, vacuum distillation, or melt crystallization. Vacuum degree, reflux ratio, tower temperature, recovery ratio, and crystallization sweating conditions are controlled manually or through fixed process windows.

[0003] In actual production, the impurity composition and migration trends of different batches of raw materials fluctuate. If the process is operated according to fixed parameters, trace amounts of moisture may penetrate the dewatering bed, light alcohol components may be carried along with side stream products, metal ions may enter subsequent stages when the collection unit is close to saturation, and heavy impurities may remain in the candidate product stream if the reboiler discharge or crystallization sweating control is insufficient. Disturbances in vacuum, reflux ratio, and column temperature can also alter the separation boundaries between light impurity removal, side stream collection, and heavy impurity removal, making it difficult to maintain consistent product purity and impurity residue across batches.

[0004] The existing control methods lack a mechanism to transform raw material impurity fingerprints, online process detection, historical batch removal efficiency, and product target limits into executable refining actions. They also lack closed-loop rules for generating reprocessing instructions when path selection is abnormal, model is disabled, parameter table is not hit, equipment temperature exceeds limits, or pre-release consistency verification fails. It is difficult to distinguish in a timely manner the specific reasons for needing to extend dehydration, enhance light removal, start metal reprocessing, increase weight removal, or adjust the crystallization sweating endpoint.

[0005] Therefore, there is a need for an electrolyte solvent purification method that can overcome the shortcomings of the existing technology. Summary of the Invention

[0006] One objective of this invention is to propose an electrolyte solvent refining method. Addressing the problem in existing technologies where it is difficult to stably control product purity based on batch impurity fluctuations during the deep refining of battery-grade ethylene carbonate and ethyl methyl carbonate, this invention proposes a method for graded pretreatment based on raw material impurity fingerprints that retain untrunculated standard detection values. The method selects a vacuum distillation or vacuum removal of light and heavy impurities, and a segmented melting crystallization sweating path based on the target solvent type, forming candidate product flows and path status records. A soft measurement module, including a prediction sub-model and a migration risk calculation sub-module, along with a model prediction controller, jointly adjusts vacuum degree, reflux ratio, tower temperature, collection ratio, water removal residence time, resin bed contact time, heavy impurity protection bed contact time, adsorption bed switching, and crystallization sweating endpoint. For branches with invalid fingerprints, pretreatment reprocessing, model disabling, parameter table mismatch, equipment out-of-bounds, safety state triggering, low confidence proxy, or consistency verification failure, the method outputs reprocessing instructions, manual review instructions, or a prohibition on release flag. This invention effectively transforms batch impurity fluctuations into executable refining control actions and reduces the probability of non-conforming materials entering the finished product tank.

[0007] This invention provides a method for refining electrolyte solvents, comprising: S1. Obtain the detection values ​​of moisture, alcohol residue, metal ions and heavy impurities of ethylene carbonate or methyl ethyl carbonate raw materials, and generate raw material impurity fingerprints; S2. Based on the raw material impurity fingerprint, the raw material is sequentially passed through a water removal adsorption unit, a metal ion collection unit, and a heavy impurity protection unit to obtain pretreated material. S3. Select the deep refining path according to the target solvent type, so that methyl ethyl carbonate enters the vacuum distillation path and ethylene carbonate enters the vacuum removal of light and heavy components and the segmented melting crystallization sweating path, and obtain the candidate product flow and path status record. S4. Collect refining process detection data, and input the raw material impurity fingerprint, process detection data, path status record and historical batch removal efficiency into the soft measurement module. The soft measurement module includes a trained prediction sub-model and a migration risk calculation sub-module. The prediction sub-model outputs the product purity prediction value, impurity residue prediction value and confidence interval, and the migration risk calculation sub-module outputs the impurity migration risk index. S5. Input the predicted product purity value, predicted impurity residue value, confidence interval, and impurity migration risk index into the model prediction controller. The model prediction controller generates a path-adaptive refining control action based on the target solvent type and the path status record, and outputs a product release instruction, a reprocessing instruction, or a manual review instruction. When S1 or S2 generates a fingerprint invalidation flag, a preprocessing / reprocessing field, or a safety pre-existing abnormal record, the model prediction controller or release verification module directly outputs a manual review instruction, a reprocessing instruction, or a release prohibition flag, and excludes the product release branch for the corresponding material.

[0008] Optionally, S1 includes: Establish batch identification, target solvent type, feed time, and testing method identification for the same batch of raw materials; Moisture content was determined by moisture detection, peak area of ​​alcohol residue and peak area of ​​heavy impurities were determined by chromatography, and metal ion concentration was determined by metal ion detection. The moisture detection value, alcohol residue peak area, metal ion concentration and heavy impurity peak area are normalized according to the detection dimensions and target limits to generate a raw material impurity fingerprint containing moisture, alcohol, metal and heavy components, and the untruncated standard detection value is retained and written into the batch fingerprint record. When the detection status is pending verification, the target limit table is invalid, or any component is missing, an invalid fingerprint flag and a manual verification instruction are generated, and the corresponding material is prohibited from entering the continuous pretreatment line.

[0009] Optionally, S2 includes: The residence time and bed switching threshold of the dehydration adsorption unit are determined based on the moisture content in the raw material impurity fingerprint. The dehydration adsorption unit uses a molecular sieve or alumina adsorption bed with controlled precipitated metal content. The resin bed contact time and regeneration status of the metal ion collection unit are determined based on the metal composition. The protective bed differential pressure threshold and bypass closure status of the heavy impurity protection unit are determined based on the heavy component. When the outlet detection value of any bed reaches the preset proportion of the corresponding target limit, the current bed is marked as to be switched and the material is introduced into the standby bed, and the pre-treated material and bed status record are obtained. When the detection value at the outlet of any bed layer still reaches the preset proportion of the corresponding target limit after pretreatment, or when the release field of the pretreated material is set to pretreatment reprocessing, a batch-level reprocessing instruction and a pretreatment reprocessing flag are generated, and the product purity prediction value, impurity residue prediction value, confidence interval and impurity migration risk index generated in subsequent S4 are marked as pending review or not participating in the release status.

[0010] Optionally, S3 includes: When the target solvent type is methyl ethyl carbonate, the pretreated material is fed into a vacuum distillation column, and light component discharge stream, side stream candidate product stream and heavy component discharge stream are formed in the order of light component removal at the top of the column, side stream collection and heavy component removal at the bottom of the column, and the side stream collection window is determined according to the side stream location, pressure difference in the column and temperature at the bottom of the column. When the target solvent type is ethylene carbonate, the pretreated material is sent to a segmented melt crystallizer after being depressurized to remove light and heavy components. A crystalline phase is formed in the crystallization section and the impurity-containing mother liquor is discharged in the sweating section to obtain a crystallization candidate product stream and a crystallization state record. The path status record includes candidate product flow records, as well as distillation status records corresponding to vacuum distillation paths or crystallization status records corresponding to segmented melting crystallization sweating paths. The path status records are used for subsequent soft measurement prediction and model predictive control.

[0011] Optionally, S4 includes: When the target solvent is methyl ethyl carbonate, the distillation process status sequence is generated by aligning the column temperature, column pressure, vacuum degree, reflux ratio, side stream flow rate, internal pressure difference, online moisture detection value, online chromatographic detection value, and candidate product metal detection value according to the sampling time. When the target solvent type is ethylene carbonate, the jacket temperature, crystal layer temperature, mother liquor conductivity, mother liquor alcohol chromatographic detection value, mother liquor heavy impurity chromatographic detection value, online moisture detection value, candidate product flowing metal detection value, crystallization section number and sweating section number are aligned according to the sampling time to generate a crystallization process state sequence; The distillation process state sequence or crystallization process state sequence is combined with the raw material impurity fingerprint, target solvent type, bed state record of pretreated material, path state record, and historical batch removal efficiency into a soft measurement input vector; The prediction sub-model is trained from the process state sequence of historical batches, laboratory retest purity, and laboratory retest impurity residue values, and outputs the product purity prediction value, impurity residue prediction value, and confidence interval corresponding to the current batch. Furthermore, the impurity migration risk index is obtained by weighting the water migration sub-index, alcohol migration sub-index, metal migration sub-index, and heavy impurity migration sub-index; Each migration sub-index is generated based on the corresponding untruncation standard detection value, the corresponding process detection value, the corresponding historical batch removal efficiency, and the corresponding product target limit. The metal migration sub-index preferentially uses the metal detection value of the candidate product stream as the corresponding process detection value. When the metal detection value of the candidate product stream is missing, the metal ion value at the metal trapping bed outlet is used as the proxy process detection value and written into the proxy low confidence flag. The weights of each migration sub-index are determined by the target solvent type and the reasons for non-compliance in historical batches when the number of historical samples reaches the preset minimum number of samples. When the number of historical samples is insufficient, the previous frozen version is used or the default weight reorganization is adopted and a weight rollback flag is generated. When any migration sub-index reaches a preset first threshold, a level of concern is generated; when any migration sub-index reaches a preset second threshold, a level of interception is generated. The preset second threshold is greater than the preset first threshold. Furthermore, the model prediction controller generates control actions based on different migration sub-indices and target solvent types: when the moisture migration sub-indices reach the level of concern, the water removal residence time is extended or a backup water removal bed is switched. When the alcohol migration index reaches the level of concern and the target solvent type is ethyl methyl carbonate, increase the reflux ratio for light removal and narrow the side-line sampling window. When the target solvent type is ethylene carbonate, extend the vacuum removal time or adjust the crystallization sweating endpoint. When the metal migration index reaches the concern level, the resin bed regeneration is effective, and the outlet metal ion detection value does not reach the preset proportion of the corresponding target limit, the resin bed contact time is extended. When the outlet metal ion detection value reaches the preset proportion or the resin bed regeneration effective sign fails, the metal capture and reprocessing is initiated. When the heavy impurity migration index reaches the level of concern and the target solvent type is methyl ethyl carbonate, increase the proportion of heavy removal from the bottom of the column or extend the contact time of the heavy impurity protection bed. When the target solvent type is ethylene carbonate, extend the depressurization time, maintain the state of discharging mother liquor, or extend the contact time of the heavy impurity protection bed. When any migration sub-index reaches the interception level, a reprocessing instruction is generated and the corresponding material is prohibited from entering the finished product tank. Furthermore, for materials that have formed a side-stream candidate product stream or a crystallization candidate product stream and have entered the release verification, before outputting the product release instruction or the reprocessing instruction based on the candidate product stream, the laboratory re-inspection results of the side-stream candidate product stream or the crystallization candidate product stream are checked for consistency with the product purity prediction value, impurity residue prediction value and impurity migration risk index. When the laboratory retest results meet the product target limit, no migration sub-index reaches the preset second threshold, the consistency check passes, and there are no abnormal path selection, disabled model, missing parameter table, out-of-bounds equipment temperature, safety status trigger, prohibited release flag, reprocessing priority flag, or low confidence flag for agents, the product release instruction is output and the corresponding material is imported into the finished product tank. When the laboratory retest results do not meet the target limit of any product, any migration sub-index reaches the preset second threshold, path selection is abnormal, model is disabled, parameter table is not hit, equipment temperature exceeds the limit, safety status is triggered, release is prohibited, reprocessing priority is marked, agent low confidence is marked, or consistency verification fails, a reprocessing instruction is output and the corresponding material is imported into the rework tank or the tank to be manually reviewed. When both the product release condition and the reprocessing condition are met, the reprocessing instruction has a higher priority than the product release instruction. When the fingerprint is invalid, the preprocessing reprocessing is performed, the path selection is abnormal, or the security pre-processing is abnormal, resulting in the failure to form a re-inspectable candidate product flow, the reprocessing instruction, the manual review instruction, or the release prohibition flag is directly output, and the consistency check is not performed.

[0012] Optionally, S5 includes: The model predictive controller uses vacuum degree, reflux ratio, column temperature, product yield, column bottom discharge ratio, water removal residence time, resin bed contact time, heavy impurity protection bed contact time, side-stream product yield window threshold, adsorption bed switching status, and crystallization sweating endpoint as control variables. The column temperature is the temperature setting of the vacuum distillation column or the vacuum removal stage for removing light and heavy components. The product yield is the side-stream product yield or the feed yield into the crystallizer. The column bottom discharge ratio is the discharge ratio of heavy components discharged from the column bottom. Inapplicable control variables are set to disabled or zero according to the target solvent type. The model predictive controller uses the target limit for product purity, the target limit for residual impurities, the equipment operating boundary, and the confidence interval as constraints. When the upper boundary of the confidence interval indicates that the predicted value of any impurity residue reaches the preset proportion of the corresponding target limit and there is a candidate action that is applicable to the path and meets the equipment operation boundary, the model prediction controller increases the control margin corresponding to the impurity and outputs at least one of the following refining control actions: water removal residence time, light reflux ratio, resin bed contact time, metal capture and reprocessing, heavy weight discharge ratio in the tower bottom, heavy impurity protection bed contact time, side stream extraction window, and crystallization sweating endpoint; When the model is disabled, the action constraint has no feasible solution, the safety state is triggered, or the control priority queue is still empty when the upper boundary of any confidence interval reaches the preset proportion of the corresponding target limit, output the reprocessing instruction, the manual review instruction, or the prohibition of release flag. Furthermore, the segmented melting crystallization sweating path includes a crystallization section, a heating sweating section, and a mother liquor discharge section; In the crystallization section, the cooling rate is determined based on the online moisture content, alcohol chromatographic values, heavy impurity chromatographic values, and crystallizer jacket temperature of the ethylene carbonate candidate material entering the segmented melting crystallizer. During the heating and sweating stage, the sweating endpoint score and crystallization state are determined and recorded based on the mother liquor conductivity, chromatographic detection value of heavy impurities in the mother liquor, and crystal layer temperature. The sweating endpoint score is only used as input for the model prediction controller to determine the re-inspection trigger and crystallization sweating endpoint control actions, and is not used as a product qualification judgment. In step S5, when the model prediction controller determines that the conditions for re-inspection are met based on the sweating endpoint score, the predicted value of impurity residue, the confidence interval, and the equipment operation boundary, the crystallization candidate product stream is marked as a state to be re-inspected.

[0013] The beneficial effects of this invention are: 1. This invention normalizes the detection values ​​of moisture, alcohol residue, metal ions and heavy impurities into raw material impurity fingerprints, and retains the untruncation standard detection values ​​to participate in the calculation of historical removal efficiency and migration risk index. This allows water removal adsorption, metal capture and heavy impurity protection to form corresponding residence time, contact time, pressure difference threshold and bed switching state according to the changes in raw material batch, thereby reducing the sensitivity of the fixed pretreatment window to batch fluctuations.

[0014] 2. This invention employs different depth refining paths for methyl ethyl carbonate and ethylene carbonate, and incorporates the side stream extraction window, the bottom weight removal ratio, the depressurization deweighting time, the mother liquor discharge state, and the crystallization sweating endpoint into the path adaptation control actions, so that the separation boundaries between light removal, heavy removal, side stream extraction, and crystallization sweating can be adjusted in combination with the target solvent type and process detection data.

[0015] 3. This invention outputs predicted product purity, predicted impurity residue, and confidence intervals through the prediction sub-model in the soft measurement module, and outputs an impurity migration risk index through the migration risk calculation sub-module. The model prediction controller generates differentiated control actions or reprocessing instructions based on the migration sub-index, path status records, and sweating endpoint scores, so that the migration risks of moisture, alcohols, metal ions, and heavy impurities can be mapped to the control of dehydration, light impurity removal, resin bed contact time, metal capture reprocessing, heavy impurity removal, protective bed contact time, side stream extraction, crystallization sweating, and abnormal reprocessing. It also prevents the corresponding materials from entering the finished product tank when consistency verification fails, model is disabled, parameters are not hit, agent low confidence is triggered, or safety status is triggered. 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 a method for refining electrolyte solvents.

[0017] Figure 2 This is a flowchart of the soft measurement module in step S4 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 An electrolyte solvent purification method, comprising: S1. Obtain the detection values ​​of moisture, alcohol residue, metal ions and heavy impurities of ethylene carbonate or methyl ethyl carbonate raw materials, and generate raw material impurity fingerprints; S2. Based on the raw material impurity fingerprint, the raw material is sequentially passed through a water removal adsorption unit, a metal ion collection unit, and a heavy impurity protection unit to obtain pretreated material. S3. Select the deep refining path according to the target solvent type, so that methyl ethyl carbonate enters the vacuum distillation path and ethylene carbonate enters the vacuum removal of light and heavy components and the segmented melting crystallization sweating path, and obtain the candidate product flow and path status record. S4. Collect refining process detection data, and input the raw material impurity fingerprint, process detection data, path status record and historical batch removal efficiency into the soft measurement module. The soft measurement module includes a trained prediction sub-model and a migration risk calculation sub-module. The prediction sub-model outputs the product purity prediction value, impurity residue prediction value and confidence interval, and the migration risk calculation sub-module outputs the impurity migration risk index. S5. Input the predicted product purity value, predicted impurity residue value, confidence interval, and impurity migration risk index into the model prediction controller. The model prediction controller generates a path-adaptive refining control action based on the target solvent type and the path status record, and outputs a product release instruction, a reprocessing instruction, or a manual review instruction. When S1 or S2 generates a fingerprint invalidation flag, a preprocessing / reprocessing field, or a safety pre-existing abnormal record, the model prediction controller or release verification module directly outputs a manual review instruction, a reprocessing instruction, or a release prohibition flag, and excludes the product release branch for the corresponding material.

[0020] In this specific embodiment, S1 includes: The refining control system reads the batch identifier, target solvent type, feed tank number, feed time, supplier batch number, and testing method identifier of the raw material to be refined from the batch management module in the raw material tank area, and establishes a batch fingerprint record for the same batch in the laboratory information management system. The target solvent type field is set to methyl ethyl carbonate or ethylene carbonate. The batch fingerprint record includes... , , Moisture standard test value Standard test values ​​for alcohols Metal standard test value Standard test values ​​for heavy impurities The batch fingerprint record, along with the detection status field, is stored in the batch database of the refining control system and serves as input to the S2 preprocessing unit. The detection module performs three types of tests on the same batch of raw materials. Moisture detection uses a coulometric moisture meter and the output unit is [unit missing]. Moisture content was determined by chromatographic analysis using gas chromatography or liquid chromatography, with outputs of alcohol residue peak area and heavy impurity peak area, respectively. Metal ion detection was performed using inductively coupled plasma mass spectrometry or ion chromatography, with outputs in units of [unit missing]. The concentration of metal ions; the retesting rule table is searched by impurity category, detection method identifier, and target solvent type, with fields including data source priority and outlier threshold. The system specifies the number of parallel samples, version number, effective date, and expiration flag; when there are three parallel tests for the same test item, the system categorizes them by impurity type. Calculate the relative deviation ,in This refers to a single test value for the same testing item. For a target limit or normalized benchmark that is greater than zero within the same unit, if Then delete the test result. The verification rule table is provided, and the version of the verification rule table is written into the batch fingerprint record; When the same impurity category has online detection values, laboratory retest values, and third retest values ​​simultaneously, the batch fingerprint generation module establishes a set of candidate detection values. The system first filters candidate values ​​whose units have been converted to be consistent and whose status field is valid; if the candidate set is empty, the system writes the impurity category into the missing field and outputs the re-examination task; if the candidate set is not empty, the system selects the appropriate data source according to the data source priority in the re-examination rule table. Laboratory retest values ​​have higher priority than online test values, and third retest values ​​have higher priority than laboratory retest values. When two candidate values ​​have the same priority and the same value, either record is retained and marked as parallel. When the priorities are the same but the values ​​are different, the result is determined by... The conservative selection result was determined and the detection status was set to "pending review". Write batch fingerprint records; Refined control system reads target limit table The detected values ​​are normalized, and the target limit table is defined. The fields include target solvent type, impurity category, target limit, normalized baseline limit, unit, detection method identifier, valid version and effective date, and moisture limit. Limits for alcohols Metal limits and heavy impurity limits All from and Matching valid records; all impurity limits entering the denominator or normalized benchmark All values ​​must be greater than 0. When the regulatory or internal control field is written as "not detected" or 0, the system reads the positive value of the method quantitation limit from the same record as the quantitation limit. Used for normalization, the original limit value is only used for pass / fail determination; if and If all values ​​are missing, zero, or non-positive, the system will write the detection status field as pending review and stop generating raw material impurity fingerprints that can enter S2; The system is based on the formula Generate the untruncated ratio and according to The impurity components that enter the pretreatment classification are generated, among which The components are represented, in order: water, alcohols, metal ions, and heavy impurities. Take a positive target limit When the target limit is expressed as undetectable or 0, a positive normalized benchmark is used. , and Using the same testing method and the same unit, Used for bed parameter classification, untruncation ratio and standard test value The fingerprints are retained in the batch fingerprint records for calculating the historical removal efficiency and migration risk index of S4; To avoid batch fingerprint incomparability caused by different detection methods, the system reads the method calibration table based on impurity type, source detection method, and target standard method. The method calibration table Including impurity categories Source detection methods Target Standard Method Calibration sample source, calibration slope Calibration intercept The following parameters must be specified: unit, valid range, output non-negative boundary, calibration validity period, and version number. If the testing method labeling for the same batch is inconsistent with the target limit table, then it shall be handled according to... Original detection values Converted to standard test value ,in The effective upper bound of the output given by the method calibration table; when When the original detection value before conversion exceeds the valid range or the output after conversion fails to meet the non-negative boundary, the system writes a pending review status and does not generate a passable fingerprint. The batch fingerprint generation module will , , , , , , and Encapsulated as a raw material impurity fingerprint in a fixed field order. The target limit table version, method calibration table version, retest rule table version, retest status, and test timestamp are all written into the batch fingerprint record, along with the raw material impurity fingerprint. After being written, it is locked as the pretreatment basis for the current batch, allowing S2 to determine the operating parameters of the water removal adsorption unit, the metal ion capture unit, and the heavy impurity protection unit. When the detection status field indicates "pending verification," "target limit table record invalid," or "any component missing," the system sets the batch status to "fingerprint invalid" and prohibits materials from entering the continuous pretreatment line. Simultaneously, it generates a re-inspection task and a manual verification instruction. The manual verification instruction includes... The system identifies missing impurity categories, invalid table versions, and prohibited release flags, and generates an invalid fingerprint flag. This flag marks the subsequently generated product purity prediction, impurity residue prediction, confidence interval, and impurity migration risk index as not eligible for release. When all four components are complete and the detection status field is valid, the system output includes... , and The raw material impurity fingerprint package is sent to the preprocessing sequence control module of S2; The invalid fingerprint branch also generates batch anomaly records. The fields include the source of the anomaly, missing fields, target limit table version, method calibration table version, re-verification rule table version, re-verification task number, and re-verification completion status; before the re-verification is completed, If a batch is identified as a pre-safety anomaly by the S5 release verification module, the model prediction controller will not read it as a candidate for release, but will only allow the generation of instructions for manual review or reprocessing. When the re-verified and supplemented detection values ​​regenerate a valid raw material impurity fingerprint, the system will establish the next batch fingerprint record with a new detection timestamp, without overwriting the already written anomaly record. .

[0021] In this specific embodiment, S2 includes: After receiving the raw material impurity fingerprint package output by S1, the preprocessing sequence control module reads the bed configuration table. and preprocessing parameter table Bed configuration table This includes the following information: water removal adsorption bed number, adsorption material batch number, metal content precipitated per unit mass, metal ion trapping resin bed number, resin regeneration status, heavy impurity protection bed number, upper limit of protection bed differential pressure, standby bed number, most recent switchover time, version number, effective time and failure flag, and a pretreatment parameter table. The data includes component range, residence time, contact time, differential pressure threshold, bypass status, switching threshold, version number, effective time, and failure flag. Both tables are searched according to the target solvent type and bed material version. If the target solvent is not found, the data is invalid, or the version is inconsistent, the pretreatment control quantity is not calculated and the pretreatment material record is written to the pending review status. The object to be measured is the same batch of pretreatment material flowing through the outlet of each bed. The measurement points are set sequentially at the outlet of the dewatering bed, the outlet of the metal collection bed, and the outlet of the heavy impurity protection bed. The water removal control module is based on the moisture content. Determine the residence time of the water removal adsorption unit The calculation rule is as follows ,in Preprocessing parameter table The base residence time that matches the target solvent type. This represents the increase in residence time corresponding to a unit of excess component. Moisture content generated for S1; Pretreatment parameter table limits and If the boundary condition is not met, then no calculation is performed. And write the parameter table miss status; when If it does not reach 1, press Run, when When the moisture content at the bed outlet reaches 1, the residence time is extended and the frequency of bed outlet moisture detection is switched from the regular cycle to the frequent cycle. The dehydration adsorption unit employs a molecular sieve or alumina adsorption bed with controlled precipitated metal content. The precipitated metal content is determined by immersion testing in a blank solvent before the adsorption material enters the production process. ,only Material release limits not met Only the bed floor record of the specified bed floor can be selected as the current bed floor; if the current bed floor record is not specified, the bed floor record will be selected as the current bed floor. achieve Or the cumulative treatment volume of the bed reaches the bed life threshold. The pretreatment sequence control module marks the current bed as to be switched and calls the backup dewatering bed; The metal capture control module is based on the metal composition. Determine the resin bed contact time and regeneration state, the contact time is according to Calculation, where and From the preprocessing parameter table And satisfy , ;like Missing or The system does not calculate the resin bed contact time and writes it to the pending review status; the resin bed status field includes the treated volume, number of regenerations, outlet metal ion detection value, and regeneration validity mark. When the outlet metal ion detection value reaches the metal target limit... preset ratio When this happens, the system switches the resin bed to a reprocessing priority state and directs the current batch to the standby resin bed; The heavy impurity protection control module is based on the heavy impurity content. Determine the differential pressure threshold of the protective bed And the bypass is closed, the calculation rule is as follows ,in Based on the upper limit of the basic pressure difference, The minimum permissible differential pressure threshold that is greater than zero. This is the differential pressure tightening factor when the heavy component exceeds the limit, and its unit is consistent with the differential pressure; the pretreatment parameter table specifies... and If any boundary condition is not met, then no calculation is performed. The status of the protective bed will be written to be reviewed; when The peak area of ​​heavy impurities at the outlet of the protective bed reaches the target limit. preset ratio At this time, the bypass state is forcibly written as closed to allow the material to pass through the heavy impurity protection bed; The pretreatment sequence control module generates a valve interlock list according to the order of water removal adsorption unit, metal ion capture unit and heavy impurity protection unit. The list fields include inlet valve number, outlet valve number, standby bed valve number, allowable flow rate, allowable differential pressure, bed status and abnormal handling action. When the detection value of any bed outlet reaches the preset proportion of the corresponding target limit, the system executes the switching rule of opening the standby bed inlet valve first and then closing the current bed outlet valve, and writes the current bed status as waiting to be regenerated or waiting to be replaced. The outlet detection of each bed is triggered by an online sampling valve according to the flow direction of the same batch of material. The measurement object at the outlet of the dewatering bed is the liquid solvent after water removal and adsorption, and the output unit is [unit missing]. Export moisture value The metal trapping bed outlet measurement object is the liquid solvent after contact with the resin bed, and the output unit is... Export metal ion value The heavy impurity guard bed outlet measurement object is the liquid solvent after passing through the guard bed, and the outlet chromatogram heavy impurity response is output. The system compares them separately. , and When any ratio reaches the corresponding preset ratio The result of the bed layer switching determination is obtained and written into the bed layer status record; The material after passing through three pretreatment units is written into the pretreatment material record. The pre-treated material record Including batch identification, target solvent type, and water removal residence time. Metal trapping contact time Protective bed differential pressure threshold The system monitors the following parameters: bed switching status, outlet moisture content, outlet metal ion content, outlet chromatographic heavy impurity peak area, bed configuration table version, and pretreatment parameter table version. If any outlet detection value still reaches the preset proportion of the corresponding target limit, or if the bed configuration table version or pretreatment parameter table version is missing or invalid, then... The release field is set to preprocessing or pending review; when When the release field indicates preprocessing passed, the preprocessing sequence control module sends the preprocessed material to the deep refining path selection module of S3 and simultaneously sends the bed status record; when When the release field is pre-processing reprocessing, the system will import the material into the rework tank and generate a batch-level reprocessing instruction and a pre-processing reprocessing flag. The reprocessing instruction includes the category of excessive impurities, the corresponding bed number, the suggested switching action, and the prohibition on entering the finished product tank flag. It will also mark the product purity prediction value, impurity residue prediction value, confidence interval, and impurity migration risk index generated in S4 as pending review or not participating in the release status, so that early pre-processing anomalies no longer wait for the candidate product stream release branch. The preprocessing and reprocessing branch also generates preprocessing exception records. Fields include Number, out-of-limit bed layer, out-of-limit detection ratio, standby bed switching result, rework tank number, and no-release sign; S4 reads This record is only used as a sample for updating historical batch removal efficiency and bed status; it is not output for product release. , or S5 read The reprocessing instruction is prioritized over the product release instruction, and the corresponding valve status, bed status, and rework tank number are written into the batch control log. After the reprocessing tank completes the reprocessing, a new batch flow record is re-established from S1 or S2. The control log field is used for the calibration of the bed switching threshold for subsequent batches.

[0022] In this specific embodiment, S3 includes: The deep refining path selection module reads the pre-processed material records output by S2. and target solvent type field And read the separate path configuration table. The separate path configuration table This includes the target solvent type, vacuum distillation column number, column top pressure setting range, side stream sampling plate position, column bottom discharge valve number, melt crystallizer number, crystallization section temperature program, evaporation section temperature program, mother liquor discharge valve number, side stream scoring weight, side stream sampling threshold, version number, effective time, and failure flag; candidate path set. by and The record consists of a matching and version-valid record, if If the product is methyl ethyl carbonate, then the reduced pressure distillation route is selected; if the product is ethylene carbonate, then the reduced pressure removal of light and heavy components and the segmented melting crystallization and sweating route is selected. When the candidate path set When multiple valid records exist, the deep refinement path selection module sorts them by path status field, equipment availability flag, and the cleaning completion time of the previous batch. Paths with a valid equipment availability flag are listed first, and paths with the same cleaning completion time are selected in ascending order of equipment number. Missing, candidate path set When the status field is empty or the two path status fields conflict, the module does not feed material to the distillation column or crystallizer, but instead records the pre-processed material. Write to the pending confirmation queue and output the path selection exception flag; In the vacuum distillation path of ethyl methyl carbonate, the column control module feeds the pretreated material into the vacuum distillation column and controls the pressure at the top of the column. Tower top temperature Side line temperature Tower temperature Reflux ratio Lateral flow and pressure difference inside the tower Write the distillation state vector Then The position of the side plate, the status of the bottom reboiler discharge valve, and the sampling time are written into the distillation status record. The light component discharge stream is formed in the light component removal section, the side stream candidate product stream is formed in the side stream section, and the heavy component discharge stream is formed in the bottom section. The flow rate, sampling time, and online chromatographic detection value of the three streams are recorded respectively. The side-stream extraction window is determined by the side-stream location, the pressure difference inside the tower, and the temperature at the bottom of the tower. The tower control module calculates the window score. ,in Indicates temperature separation margin. Use temperature units and , Indicates the differential pressure margin. and Using the same differential pressure unit and , This indicates the degree to which the lateral flow rate approaches the set flow rate. , and allowable deviation Using the same flow unit and When any denominator field is not matched or is not greater than zero, the tower control module does not calculate the side-line extraction window score and imports the corresponding time period material into the transition tank; all three components are normalized to dimensionless quantities before entering the weighted summation, and the weights are... , and From the separate path configuration table and satisfy , , and Lateral sampling threshold It is also given and satisfied by the same frozen version separate path configuration table. If the weights or thresholds do not meet the above boundaries, then no calculation is performed. And write the path selection exception flag, A larger value indicates that the sideline window is more suitable as a candidate product stream; when Reaching the lateral line extraction threshold Furthermore, if the online moisture detection value, alcohol residue chromatographic value, and heavy impurity chromatographic value all fail to meet the corresponding target limits, the tower control module opens the side-stream collection valve and writes the material into the side-stream candidate product stream record. When two consecutive sampling times Same and all reach At that time, the pressure difference margin inside the tower is used. If the sampling time with a larger value is used as the sampling starting point, then... If they are still the same, then the record with the later sampling time is used as the starting point for sampling to reduce entrainment during the transition period; when Not achieved When the pressure difference inside the tower reaches the upper limit or the temperature of the tower bottom reaches the equipment operating boundary, the system closes the side line outlet valve and introduces the material during this period into the transition tank or tower bottom heavy component discharge. In the ethylene carbonate pathway, the light and heavy removal module first checks the pre-treated material records. The process of removing light and heavy components under reduced pressure yields ethylene carbonate candidate material that enters the segmented melting crystallizer. The crystallization control module then records this candidate material into the crystallization state record. The fields include candidate material number, feed temperature, and jacket temperature. Crystal layer temperature Mother liquor conductivity Chromatographic detection values ​​of alcohols in mother liquor Chromatographic detection values ​​of heavy impurities in mother liquor Online moisture detection value Metal detection values ​​of candidate products , Crystallization segment numbering and sweating segment numbering; The cooling rate in the crystallization section is determined based on online moisture detection values, alcohol chromatographic detection values, heavy impurity chromatographic detection values, and crystallizer jacket temperature. The controller first generates the normalized moisture content based on the ratio of the detected value to the target limit in the same unit. Normalized amount of alcohols Normalized amount of heavy impurities The target limits are the moisture limits in the target limit table. Limits for alcohols and heavy impurity limits And all are greater than zero; then calculate the cooling coefficient. And order ,in Based on the cooling rate, , and These are the adjustment coefficients of water, alcohol, and heavy impurities on crystal growth, respectively, and all are not less than zero. Located between 0 and 1; when the online detection value approaches the target limit, the cooling rate is reduced to prolong the crystal impurity removal time; The sweating stage, which involves heating up the temperature, determines the sweating endpoint based on the mother liquor's conductivity, chromatographic detection values ​​of heavy impurities in the mother liquor, and the crystal layer temperature. The crystallization control module then calculates the sweating endpoint score. ,in and All adopted , and All concentrations were calculated using peak area or external standard conversion under the same chromatographic method. , and All values ​​are in Celsius, and all three ratios are dimensionless components normalized to the same unit. , and From the crystallization parameter table and the sum is 1, The larger the value, the closer the sweating endpoint is to the conditions for retesting; the chromatographic detection value of alcohol in the mother liquor. Instead of reaching the sweating endpoint score, it enters the S4 alcohol migration index; The crystallization parameter table was obtained by fitting crystallization segment samples, mother liquor samples, and laboratory retesting results from historical batches of ethylene carbonate. Fields include target solvent type, crystallizer number, etc. , , , , , The threshold for the end of sweating Version number and update trigger conditions; among which , , and , The dimensionless range is between 0 and 1; the crystallization parameter table uses laboratory re-inspection pass / fail labels as the fitting target and adopts the most recent version before freezing. The crystallization segment and mother liquor samples from each batch of ethylene carbonate were determined. and When the sample size does not reach the minimum sample size specified in the crystallization parameter table, If the temperature boundary does not meet the above conditions, or if the crystallization parameter table does not match the current crystallizer number, the system will keep the mother liquor drain valve open and mark the crystallization candidate product stream as unconfirmed. when Reaching the threshold of sweating endpoint At that time, the crystallization control module does not directly perform product qualification judgment, but instead... The following information should be recorded in the crystallization status log: mother liquor conductivity status, mother liquor alcohol chromatographic status, mother liquor heavy impurity chromatographic status, crystal layer temperature status, mother liquor drain valve status, and crystallization candidate product stream number. And output candidates for re-inspection; when the crystal layer temperature exceeds the temperature program boundary of the sweating section and the mother liquor conductivity reaches The chromatographic detection value of heavy impurities in the mother liquor reached If the crystallization parameter table is not matched, the system keeps the mother liquor drain valve open and sets the path status field to crystallization anomaly. Subsequently, S4 generates the prediction result, and S5 determines the result based on the same... ,same The safety priority table determines instructions for re-inspection, extended sweating, or reprocessing; For cases where segmented melting crystallization sweating path is not feasible to control, the crystallization control module compares the crystallizer temperature program boundary, mother liquor drain valve feedback, sweating section duration upper limit, and crystallization parameter table validity status; when Below , Higher than Inconsistent feedback and issuance status between the mother liquid discharge valve and the issued status; sweating duration exceeding the upper limit. or At that time, the system does not calculate the new sweating endpoint action, but instead... The sweating endpoint score status is set to infeasible, and the following flags are written: maintain mother liquor discharge, stop pending retesting, and reprocessing candidate flag; this infeasible status is recorded along with the candidate product flow. Send to S4 and S5 so that subsequent controllers can directly enter the safety priority branch without repeatedly trying to close the mother liquor drain valve; The deep refinement path selection module ultimately outputs candidate product flow records. and path status record ,in Includes target solvent type, path identifier, and side-stream candidate product stream records. Or crystallization candidate product stream record Light component discharge flow, heavy component discharge flow, sampling time, candidate triggers for retesting, path selection anomaly flag, reprocessing instruction field and separation path configuration table version; Write the corresponding status record according to the path type. When the target solvent type is ethyl methyl carbonate... Including distillation state records And side-line sampling window scoring, when the target solvent type is ethylene carbonate Including crystallization state records And the sweating endpoint score status; when the path selection is abnormal, the separation path configuration table version is missing, or the crystallization parameter table is not matched, the reprocessing instruction field is set and follows. and Send to S4 and S5.

[0023] In this specific embodiment, S4 includes: The soft measurement module uses the raw material impurity fingerprint generated by S1 S2 generates pre-processed material records Candidate product stream records generated by S3 For upstream input, at the start of the current batch, freeze the target limit table, method calibration table, retest rule table, bed configuration table, pretreatment parameter table, separation path configuration table, metal proxy correction table, risk weight table, risk level threshold table, control margin mapping table, action constraint table, equipment operation boundary table, control response table, crystallization parameter table, consistency deviation threshold table, safety priority table, and soft sensor model version, and freeze the version number to be written into the batch-level calculation context. The re-inspection results, deviation records, prohibited actions, control response table updates, metal proxy correction table updates, control margin mapping table updates, consistency deviation threshold table updates, safety priority table updates, and background model updates generated in the current batch are only written to the version used in the next batch and do not replace the current batch. ; The soft measurement module includes a prediction sub-model, a migration risk calculation sub-module, and a version freeze manager. The prediction sub-model receives soft measurement input vectors. Only product purity prediction values ​​are generated. Predicted residual values ​​of each impurity and confidence interval The migration risk calculation submodule reads the upper boundary of the confidence interval, the untruncated ratio retained by S1, the process detection value, and the historical batch removal efficiency from the output of the prediction sub-model, and generates the migration sub-index. and impurity migration risk index The version freeze manager ensures that the prediction sub-model version, risk weight table version, and risk level threshold table version are computed in the same batch context. Maintain internal consistency; Data alignment module based on sampling time To generate a path-adapted process state sequence for the index, a distillation process state sequence is generated when the target solvent type is methyl ethyl carbonate. ,in The values ​​are measured online moisture content at the side or top of the tower, and the units are... , For online chromatographic response or external standard conversion of alcohol residue concentration, For heavy impurities, the online chromatographic response or external standard-converted concentration is obtained from the candidate product stream or effluent sampling point at the same sampling time in S3 and verified by the method calibration table. Convert to the corresponding scope of the target limit table, The metal detection value of the candidate product stream or the metal ion retention value at the outlet of the metal trapping bed after agent correction; when the target solvent type is ethylene carbonate, a crystallization process state sequence is generated. ,in The concentration of alcohols in the mother liquor was calculated based on the chromatographic response or external standard. The concentration of heavy impurities in the mother liquor is calculated using either the chromatographic response or the external standard conversion, and is then compared with the positive denominator. and match; When any sampling field is missing, the system uses the previous valid sample value for no more than two sampling periods. If the field is still missing after two sampling periods, the period is marked as a disabled model. For metal categories, if a candidate product stream metal detection value exists in the current control period, that detection value is used as... If there are no candidate product stream metal detection values ​​in the current control cycle, then the metal ion value at the outlet of the same batch of metal trapping bed will be used. Through agent correction relationship The proxy process detection value is obtained, where , and Metal proxy correction table from the frozen version, and the proxy branch writes a low confidence flag; if the metal proxy correction table is not hit, , If the value exceeds the applicable range of the agent or the export metal ion value is also missing, the metal migration sub-index will enter the manual review branch and trigger the reprocessing instruction candidate. The soft measurement input vector is obtained by merging the process state sequence of path adaptation, raw material impurity fingerprint, target solvent type, bed state record, candidate product flow record, historical batch removal efficiency, and batch-level computational context, denoted as . ,in In this specific embodiment, a sliding window represents the state sequence of a distillation process or a crystallization process. Set to 30; Historical Batch Removal Efficiency Table This includes fields such as impurity type, standard test value for uncut feed, retest value for discharge, removal efficiency, operating path, bed version, control action version, and whether the test is handled by a proxy. The availability of metal detection values ​​is determined based on the priority of candidate product streams. If a metal detection value exists in a candidate product stream and the detection status is valid, then... Take this detection value; if the metal detection value of the candidate product stream is missing, but the metal ion value at the outlet of the metal trapping bed is missing... If it exists, the system reads the frozen version of the metal proxy correction table. ,according to Generate proxy process detection values, where , , The applicable range for the agent is given by the metal agent correction table; when The proxy correction output is clipped to 0. Exceeding the applicable range or agent correction table version In case of inconsistency, the system writes a low confidence flag to the agent and excludes the product release branch in the S5 release verification. The model training data comes from the process state sequences of completed batches, laboratory retest purity, laboratory retest impurity residue values, and control action records. The training module trains the model separately according to the target solvent type, and divides the training set and validation set by batch. When the upper quantile coverage of any impurity category on the validation set does not reach the target coverage... At this time, the model version is not released, the system continues to use the previous valid model version, and puts the soft measurement module of this batch into a conservative safety state. The conservative safety state only allows the current vacuum degree, reflux ratio and extraction ratio to be maintained, and does not allow the release of new products. The confidence intervals of the prediction sub-model are obtained by calibration using the residual quantiles of the validation set. The training module statistically analyzes the residuals according to the target solvent type and the running path. And read from the upper quantile of the absolute value of the residuals in the validation set. ,generate and upper boundary When the mean absolute error of any impurity category exceeds the release threshold given in the consistency deviation threshold table, or when the validation set coverage fails to reach the target coverage. If the verification sample does not cover the current target solvent type's running path, the model version will not be released, and the system will use the previous valid version or enter a conservative and safe state. Historical batch removal efficiency is calculated by impurity category, and the calculation rule is as follows: ,in The standard test value for uncut feed retained for S1. This indicates the residual value of impurities in laboratory retesting, with both the numerator and denominator using the same unit for the same impurity category. Dimensionless efficiency from 0 to 1; when When the value is zero, exceeds the valid range of the method calibration table, or is missing from the laboratory retest value, the removal efficiency of this impurity category will not be included in the current model input and will be written into the missing flag. The impurity migration risk index calculation module calculates migration sub-indices based on the corresponding untruncation ratio, corresponding process detection value, corresponding historical batch removal efficiency, and corresponding product target limit. The calculation rules are as follows: ,in The untruncated ratio, The upper limit given by the risk weight table; The For the corresponding process detection value, in the methyl ethyl carbonate pathway , , In the ethylene carbonate pathway , , All metal categories are preferred. , For the corresponding product target limit or normalized benchmark that is greater than zero, Weighting for historical batch removal efficiency , , and The risk weights are from the frozen version and sum to 1. This is a dimensionless risk value; an increase in the value indicates an increased risk of the corresponding impurity migrating to the candidate product stream. In the risk formula , and All are penalty terms for impurities that are close to or exceed the target limit, and all use the same type of positive denominator that is greater than zero. , To remove penalty terms for inefficiency, all four penalty terms are positive; among them It is only used for S2 bed parameter classification and is not used as a migration risk penalty item. It is not used for ranking in the profit score, but rather as a risk input that triggers an increase in the attention level, interception level, and control margin; when multiple migration sub-indices reach the attention level at the same time, the system prioritizes the interception level and then the control margin. The data is written into the control priority queue in the order of numerical value priority and later sampling time priority; The risk weighting table includes fields such as target solvent type, impurity category, risk item weighting, impurity category weighting, historical non-compliance reasons percentage, model version, and number of statistical windows. Minimum number of samples Trimming boundaries , , and The validity period and update trigger conditions must be met. and For each impurity category Risk item index The set of values ​​is These correspond to penalties for uninterrupted feed, process detection, historical removal efficiency, and confidence upper bound, respectively. Indicates freezing the statistics window Intrinsic Impurity Category When a batch fails to meet the standards, it is classified as a risk item. For cause counts, compound causes are split into corresponding risk items according to the manual re-inspection cause field, and the sum of the split counts for the same batch under the same impurity category is 1. The system first counts them by... Calculate the clipping weights, then... Normalization, making , , and The sum is 1; when the number of historical samples has not reached , If the denominator is zero after trimming, the system will use the risk item weight reorganization of the previous frozen version. If there is no previous frozen version, the default weight reorganization with the four risk items evenly distributed will be adopted, and a weight rollback flag will be written into the risk index record. For the risk level output, the system also generates a control margin. , and according to Generate control margin coefficients, where , , and Control margin mapping table from the frozen version; the control margin mapping table is searched by impurity category, target solvent type, and model version, and is limited accordingly. and If the boundary is not met, no action can be released and it is written to the queue for review; when any Reaching the preset margin threshold At that time, the soft measurement module writes the impurity category into the control priority queue for the S5 model prediction controller to select an action; The control priority queue also generates path action hint fields for S5 to read. The fields include target solvent type, migration sub-index category, concern level, interception level, control margin, and candidate action; when the alcohol migration sub-index reaches the concern level and the target solvent type is methyl ethyl carbonate, The writing process increases the light reflux ratio and narrows the side-stream extraction window; when the target solvent type is ethylene carbonate... The process can be modified by extending the decompression time for light component removal or adjusting the crystallization sweating endpoint; when the metal migration index reaches the level of concern, the resin bed regeneration is effective, and the outlet metal ion detection value does not reach the target limit preset ratio. The resin bed contact time is extended, and when the outlet metal ion detection value reaches the preset ratio or the resin bed regeneration validity indicator fails, [the process is initiated]. Write into the metal trapping and reprocessing; when the heavy impurity migration index reaches the level of interest and the target solvent type is ethyl methyl carbonate... Increase the proportion of waste material removed from the reboiler or extend the contact time of the protective bed; when the target solvent type is ethylene carbonate. The program can be modified to extend the decompression and degrafting time, maintain the mother liquor discharge state, or extend the contact time with the protective bed; when any migration index reaches the interception level... Simultaneously, a priority flag for reprocessing and a flag prohibiting entry into the finished product tank are written; Before outputting the product release instruction or the reprocessing instruction based on the candidate product stream, the soft measurement module also generates a pre-release consistency check field. The fields include candidate product stream number, tank number to be re-inspected, predicted purity, predicted impurity residue, upper boundary of confidence interval, migration sub-index, interception level flag, model version, target limit table version, method calibration table version, re-inspection rule table version, bed configuration table version, preprocessing parameter table version, separation path configuration table version, metal surrogate correction table version, risk weight table version, risk level threshold table version, control margin mapping table version, control response table version, action constraint table version, equipment operation boundary table version, crystallization parameter table version, consistency deviation threshold table version, and safety priority table version; all of the above version fields are from the current batch. Transmission, Instead of directly releasing materials, the data is used as a comparison benchmark when the S5 release verification module reads the laboratory re-inspection results. This ensures that both the product release instructions and reprocessing instructions for products that have already formed a candidate product flow have a predictive basis for the same batch. When the soft measurement module reads an invalid fingerprint flag, a pre-processing reprocessing flag, a path selection anomaly flag, or a security pre-examination anomaly record and no re-inspectable candidate product flow has been formed, the soft measurement module does not generate a data release. And write a consistency check skip flag so that the subsequent release check module does not perform the consistency check; When the model is disabled, the risk level threshold table is not hit, or the control priority queue is empty but the upper boundary of any confidence interval reaches the corresponding target limit, the soft measurement module will write the batch into the conservative safety state. The output fields of the conservative safety state include the prohibition flag, the maintain current extraction flag, the manual review task number, and the rework tank candidate number. This state is interpreted by S5 as not allowing entry into the finished product tank. The model inference module will at each sampling time The following table versions are included: Model Version, Target Limit Table Version, Method Calibration Table Version, Revalidation Rule Table Version, Bed Configuration Table Version, Pretreatment Parameter Table Version, Separation Path Configuration Table Version, Metal Proxy Correction Table Version, Risk Weight Table Version, Risk Level Threshold Table Version, Control Margin Mapping Table Version, Control Response Table Version, Action Constraint Table Version, Equipment Operation Boundary Table Version, Crystallization Parameter Table Version, Consistency Deviation Threshold Table Version, and Safety Priority Table Version, along with the input vector. Write the inference task number together. and only in the same The following generation , and If any frozen version is updated in the background during inference, the current inference and subsequent sampling of the current batch will still use the updated version. Once the version number is output, the background update will only enter the candidate version pool for the next batch. If the inference task times out or the output fields are incomplete, the soft measurement module will not generate a pre-release consistency check field, but will instead write the model disabled state as disabled and generate a reprocessing candidate state. The soft measurement module outputs soft measurement result records. The fields include batch identifier, target solvent type, and predicted product purity value. Predicted residual values ​​of each impurity Confidence interval Migration sub-index Impurity migration risk index Risk level and control margin Consistency check field before release The table includes the following formats: model version, target limit table version, method calibration table version, verification rule table version, bed configuration table version, pretreatment parameter table version, separation path configuration table version, metal proxy correction table version, risk weight table version, risk level threshold table version, control margin mapping table version, control response table version, action constraint table version, equipment operation boundary table version, crystallization parameter table version, consistency deviation threshold table version, safety priority table version, and model disabled status. When the model disabled status is active, When the model is disabled and passed to S5, the system will only allow the current sampling status to be maintained and trigger manual review if the model is disabled.

[0024] In this specific embodiment, S5 includes: The model predictive controller receives and records the soft measurement results output by S4. And read the device operation boundary table Action constraint table Control Response Table Product Target Limit Table Method Calibration Table Risk Level Threshold Table Crystallization parameter table Consistency Deviation Threshold Table and security priority table Equipment operation boundary table This includes lower and upper limits for vacuum, reflux ratio, tower temperature, side stream extraction ratio, bottom weight removal ratio, water retention time, resin bed contact time, guard bed contact time, side stream window threshold range, minimum interval for adsorption bed switching, and crystallization sweating temperature program boundaries, along with an action constraint table. This includes control step size, action hold time, interlocking conditions, path applicability flags, rules for handling inapplicable control quantities, prohibited actions, and purity gap normalization benchmarks. Complex processing path, control response table Including response matrix , , Control scaling matrix Predicted step size and version number; Control Response Table This is generated by the model maintenance module after the batch ends. The data source includes the process state sequence and control action records of historical batches. The model maintenance module identifies raw material impurity fingerprints, bed status records, candidate product flow laboratory retest results, and action execution feedback; it also identifies these by target solvent type and path identifier. , and ,make To obtain the minimum, among which The non-negative regularization coefficient is used to identify samples that are divided into training and validation sets in batches. When the one-step prediction error of the validation set exceeds the control response table release threshold, the response direction is opposite to the known separation mechanism, or the number of path samples is lower than the minimum number of samples, a new... Do not release; the current batch will continue to use the previously frozen valid version or enter the reprocessing branch. The controller is reading First, a validity cleaning process is performed. The cleaning rules are as follows: delete sampling rows where the model is disabled; delete prediction records whose timestamps are more than two sampling periods earlier than the current control period; and add records lacking versions of the target limit table, method calibration table, verification rule table, bed configuration table, preprocessing parameter table, separation path configuration table, risk level threshold table, or crystallization parameter table to the pending review queue. Only records with the following versions are retained: model version number, target limit table version, method calibration table version, verification rule table version, bed configuration table version, preprocessing parameter table version, separation path configuration table version, metal proxy correction table version, risk weight table version, risk level threshold table version, control margin mapping table version, control response table version, action constraint table version, equipment operation boundary table version, crystallization parameter table version, consistency deviation threshold table version, and safety priority table version, all of which must be identical to the current batch version. Consistent records; for duplicate records from the same batch at the same sampling time, if all versions are consistent... If they match, the record with the later timestamp is retained; if any version is inconsistent, it is written to the queue for review. (After cleaning) If empty, the controller generates a reprocessing instruction and keeps the current valve and extraction ratio unchanged, prohibits the corresponding material from entering the finished product tank, and excludes the batch of material from the product release branch; Model predictive controller constructs control vector: ; in Indicates the vacuum setting. This indicates the reflux ratio setting. This indicates the tower temperature or the temperature setting for the depressurization and separation of light and heavy components. This indicates the proportion of material extracted from the side stream or the proportion of material entering the crystallizer. This indicates the proportion of weight removed from the column bottom or the proportion of weight removed by vacuum decompression. Indicates the residence time excluding water. Indicates the resin bed contact time. Indicates the contact time of the heavy impurity protective bed. Indicates the lateral window threshold. Indicates the switching status of the adsorption bed. This represents the endpoint state of crystallization sweating; each component of the control vector is given by the action constraint table, which specifies the allowable step size, upper and lower limits, and holding time, and exceeds the equipment operation boundary table. The candidate actions were directly deleted; The controller executes path adaptation rules based on the target solvent type. When the target solvent type is ethyl methyl carbonate, Disabled and not involved in the objective function, when the target solvent type is ethylene carbonate and it is in the crystallization or sweating stage. , and Set to zero or retain value and not as an adjustable action; Action constraint table Record the path applicability flag for each control variable; do not generate execution commands for non-applicable control variables. The controller records soft measurement results. The predicted product purity, predicted impurity residue, confidence interval, and migration sub-index are used as state variables to construct a state vector: ; in and The measured values ​​of the process status are indicated by the subscript. , , and These represent water, alcohols, metal ions, and heavy impurities, respectively. In each control cycle, the model predictive controller uses a rolling response model to predict future states, with the following recursion rule: ,in , and The control response table for the frozen version is identified by the target solvent type and the current path. Read, The disturbance vector is composed of the raw material impurity fingerprint and the bed state; if the control response table version is the same as... If there is inconsistency, the control response table is not hit, or the recursive result exceeds the effective range of the model, the controller stops solving and generates a complex processing instruction. Model predictive controller solves the dimensionless objective function ,in To predict the step size, The target limit for product purity, The MPC purity notch normalization benchmark is given by the action constraint table and , The target limit for impurity residue, For the same impurity category, the target limit or normalized benchmark is greater than zero. The control quantity is scaled to a matrix such that each diagonal element is the reciprocal of the allowable step size or upper / lower limit span of the corresponding control quantity. Only control quantities applicable to the path and allowed to have a step size or upper / lower limit span greater than zero are allowed to enter. Disable control variables and zero-span control variables by removing them from the objective function. , and The weights are derived from the action constraint table and are dimensionless; the action constraint table is limited. Each impurity category involved in the constraint and , If the weight is missing, negative, or any The controller stops solving and enters the branch for review or reprocessing. The smaller the objective function, the smaller the predicted purity gap, the smaller the impurity residue exceeding the limit, and the smaller the action disturbance. For control optimization only; consistency verification is to be performed separately. Both are from the current batch The versions of the action constraint table and consistency deviation threshold table for freezing are given; Controller constraints include , , , , , , , as well as ,in , When there is no feasible solution to the constraint, the controller freezes the side line sampling or crystallization for re-inspection output and generates a reprocessing instruction; When the moisture migration sub-index reaches a level of concern, the controller prioritizes extending the water retention time or switching to a backup dewatering bed. Specifically, it reads the S2 bed status record; if the current dewatering bed is not in a state awaiting switching and the minimum switching interval for the adsorption bed has been met, then... Set as a backup bed for switching; if the switching interval is not met, then... Increase the step size according to the action constraint table and increase the lateral movement ratio. Reduce the step size by one. When the alcohol migration index reaches a level of concern, the controller selects the appropriate action based on the target solvent type. When the target solvent type is ethyl methyl carbonate, the controller increases the light solvent removal reflux ratio and narrows the side-stream sampling window. Specifically, it... Increase the allowed step size, and read the window tightening coefficient from the frozen version of the action constraint table. and according to Update the side window threshold, where ;when When there is no feasible sampling window for two consecutive control cycles after the input is missing, negative, or projected, the controller will not continue to increase the threshold but will generate a verification or reprocessing instruction; when the pressure difference inside the tower... When the differential pressure reaches the upper limit in the equipment operating boundary table, the controller no longer increases the reflux ratio, but instead reduces the side-stream extraction ratio and introduces the material for the corresponding time period into the transition tank; when the target solvent type is ethylene carbonate and the material is in the depressurization and de-heavy phase, the controller extends the de-heavy holding time or reduces the feed ratio into the crystallizer; when the material is in the crystallization or sweating section, the controller maintains... and The risk of alcohol entrainment is reduced by maintaining a prohibited state and extending the sweating process or continuing to drain the mother liquor; When the metal migration sub-index reaches a level of concern, the controller reads the outlet metal ion detection value and regeneration effectiveness indicator of the metal trapping resin bed. If the resin bed is in an effective regeneration state and the outlet metal ion detection value has not reached the preset proportion of the metal target limit, then... The step size is increased according to the action constraint table; if the outlet metal ion detection value reaches the preset ratio or the regeneration effective mark fails, the controller outputs a metal capture and reprocessing instruction and introduces the candidate product stream into the rework tank. When the heavy impurity migration index reaches the concern level, the controller selects the action based on the target solvent type. When the target solvent type is ethyl methyl carbonate, the controller increases the reboiler weight removal ratio or extends the heavy impurity protection bed contact time, with the selection rule being to first read from the equipment operation boundary table. and And confirm Then compare the temperature margin of the tower bottom. Pressure differential margin with protective bed ,in The protective bed differential pressure threshold that is greater than zero, determined by S2. To protect the current differential pressure measurement value of the bed, both units are the same; if Temperature boundary not hit or The controller does not calculate or And generate instructions to be reviewed or processed; when Not less than Prioritize increasing the proportion of weight removed from the bottom of the tower. Greater than Time priority extension When the target solvent is ethylene carbonate, the controller extends the depressurization and deweighting holding time, keeps the mother liquor discharge valve open, or extends the holding time. and the weight ratio of the tower bottom It is only used in the decompression and weight removal stage, and not in the crystallization or sweating stages; For the candidate product stream of ethylene carbonate crystallization, the controller reads the crystallization status record according to the crystallization segment, the heating and sweating segment, and the mother liquor discharge segment in the segmented melt crystallization sweating path. The results of S4 soft measurement and the sweating endpoint score of S3 are used to determine the crystallization sweating endpoint. The upper boundary of the confidence interval for all predicted impurity residue values ​​is satisfied. And the sweating endpoint score Reaching the threshold of sweating endpoint At that time, Set the sweating process to complete and output a re-inspection command, where and The instruction to be re-inspected only indicates that the material has entered the laboratory re-inspection process and does not indicate that the product is qualified; if the metal migration sub-index uses a proxy low confidence flag, the crystal layer temperature exceeds the limit, any migration sub-index reaches the preset second threshold, any If the crystallization parameter table is not matched, then the safety priority table... Set the reprocessing command to the highest priority and import it into the rework tank. If only the conductivity of the mother liquor or the chromatographic value of heavy impurities reaches the limit and the interception is not triggered, extend the sweating process and continue to discharge the mother liquor. For materials that have formed a side-stream or crystallization candidate product stream and entered the release verification process, the release verification module reads the laboratory re-inspection results before outputting the product release instruction or the reprocessing instruction based on the candidate product stream, and sets the re-inspection purity... and retesting residual impurities Recording soft measurement results Perform consistency verification; consistency deviation threshold table Provided by impurity category, target solvent type, detection method identifier, model version, and frozen batch version. And provide the detection method according to the target solvent type and purity. and purity deviation normalization benchmark ,in It is only used for normalization of consistency deviation in the release process; when a match is missed, the consistency check field is set to failure. The release check module calculates the residual impurity deviation. and purity deviation ,when Meets product purity target limits, all Meet the corresponding product target limits, all and When there is no interception level migration sub-index, and there are no path selection anomalies, model disablement, parameter table miss, equipment temperature out of bounds, safety status triggering, prohibition of release flag, reprocessing priority flag, and agent low confidence flag, the system outputs a product release command and imports the material into the finished product tank. When laboratory retest results fail to meet any product target limit, any migration sub-index reaches the preset second threshold, path selection anomaly occurs, model is disabled, parameter table is not hit, equipment temperature exceeds limits, safety status is triggered, release is prohibited flag is activated, reprocessing priority flag is activated, proxy low confidence flag is activated, or consistency deviation occurs. To reach the corresponding Purity deviation achieve If the consistency deviation threshold table is not matched, the release verification module outputs a reprocessing instruction and imports the corresponding material into the rework tank or the tank awaiting manual review, while excluding the corresponding material from the product release branch; if the fingerprint is invalid, preprocessing reprocessing, path selection is abnormal, or safety pre-conditioning is abnormal, resulting in the failure to form a re-inspectable candidate product flow, the release verification module does not read the laboratory re-inspection consistency verification field and directly outputs a reprocessing instruction, a manual review instruction, or a release prohibition flag; if the product release condition and any reprocessing condition are met simultaneously, the safety priority table... The reprocessing instructions are to cover the product release instructions, and deviation records are to be written into the historical batch removal efficiency table. And the model retraining sample pool, for S4 to update the model version and risk weight table for subsequent batches; The model predictive controller will ultimately control the recording of actions. Write to the batch control log, with fields including batch identifier, control period, and control vector. Objective function value The system records the constraint hit items, risk level, release instructions, reprocessing instructions, imported tank number, and action execution results. The batch control log serves as a source for statistical analysis of historical removal efficiency and non-conformity reasons for subsequent batches. When the execution result is inconsistent with the issued action, the system records valve feedback, pump frequency feedback, and manual intervention flags, and adds the action to the prohibited action set in the next 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 raw material impurity fingerprints, process state sequences of path adaptation, path state records, historical batch removal efficiency, and product target limits into a soft measurement module. It uses the predicted values ​​and confidence intervals output by the prediction sub-model and the impurity migration risk index output by the migration risk calculation sub-module to determine the control margin, so that the migration chain of various impurities from the raw material end to pretreatment, deep refining, candidate product flow, and finished product release has a traceable control basis.

[0027] Compared to refining methods that rely solely on fixed vacuum levels, fixed reflux ratios, or fixed crystallization sweating times, the improvement of this invention lies in using soft measurement prediction results and migration indexes as constraints and action selection criteria for the model predictive controller. Furthermore, it generates reprocessing instructions with higher priority than release for path anomalies, model disabling, parameter misses, equipment out-of-bounds issues, low agent confidence, safety state triggering, and consistency verification failure states. This allows for the dynamic generation of dewatering residence time, light reflux ratio, resin bed contact time, metal capture reprocessing, tower bottom discharge ratio, heavy impurity protection bed contact time, side-stream sampling window, and crystallization sweating endpoint for different batches based on the detection data of the same batch.

Claims

1. A method for refining electrolyte solvent, characterized in that, include: S1. Obtain the detection values ​​of moisture, alcohol residue, metal ions and heavy impurities of ethylene carbonate or methyl ethyl carbonate raw materials, and generate raw material impurity fingerprints; S2. Based on the raw material impurity fingerprint, the raw material is sequentially passed through the water removal adsorption unit, the metal ion collection unit, and the heavy impurity protection unit to obtain the pretreated material. S3. Select the deep refining path according to the target solvent type, so that methyl ethyl carbonate enters the vacuum distillation path and ethylene carbonate enters the vacuum removal of light and heavy components and the segmented melting crystallization sweating path, and obtain the candidate product flow and path status record. S4. Collect refining process detection data, input raw material impurity fingerprint, process detection data, path status record and historical batch removal efficiency into the soft measurement module. The soft measurement module includes a trained prediction sub-model and a migration risk calculation sub-module. The prediction sub-model outputs the product purity prediction value, impurity residue prediction value and confidence interval, and the migration risk calculation sub-module outputs the impurity migration risk index. S5. Input the predicted product purity value, predicted impurity residue value, confidence interval, and impurity migration risk index into the model prediction controller. The model prediction controller generates path-adaptive refining control actions based on the target solvent type and path status record, and outputs product release instructions, reprocessing instructions, or instructions for manual review. When S1 or S2 generates a fingerprint invalidation flag, a preprocessing / reprocessing field, or a safety pre-existing abnormal record, the model prediction controller or release verification module directly outputs instructions for manual review, reprocessing instructions, or a prohibited release flag, and excludes the product release branch for the corresponding material.

2. The electrolyte solvent purification method according to claim 1, characterized in that, S1 includes: Establish batch identification, target solvent type, feed time, and testing method identification for the same batch of raw materials; Moisture content was determined by moisture detection, peak area of ​​alcohol residue and peak area of ​​heavy impurities were determined by chromatography, and metal ion concentration was determined by metal ion detection. The moisture detection value, alcohol residue peak area, metal ion concentration and heavy impurity peak area are normalized according to the detection dimensions and target limits to generate a raw material impurity fingerprint containing moisture, alcohol, metal and heavy components, and the untruncated standard detection value is retained and written into the batch fingerprint record. When the detection status is pending verification, the target limit table is invalid, or any component is missing, an invalid fingerprint flag and a manual verification instruction are generated, and the corresponding material is prohibited from entering the continuous pretreatment line.

3. The electrolyte solvent purification method according to claim 1, characterized in that, S2 include: The residence time and bed switching threshold of the dehydration adsorption unit are determined based on the moisture content in the raw material impurity fingerprint. The dehydration adsorption unit uses a molecular sieve or alumina adsorption bed with controlled precipitated metal content. The resin bed contact time and regeneration status of the metal ion collection unit are determined based on the metal composition. The protective bed differential pressure threshold and bypass closure status of the heavy impurity protection unit are determined based on the heavy component. When the outlet detection value of any bed reaches the preset proportion of the corresponding target limit, the current bed is marked as to be switched and the material is introduced into the standby bed, and the pre-treated material and bed status record are obtained. When the detection value at the outlet of any bed layer still reaches the preset proportion of the corresponding target limit after pretreatment, or when the release field of the pretreated material is set to pretreatment reprocessing, a batch-level reprocessing instruction and a pretreatment reprocessing flag are generated, and the product purity prediction value, impurity residue prediction value, confidence interval and impurity migration risk index generated in subsequent S4 are marked as pending review or not participating in the release status.

4. The electrolyte solvent purification method according to claim 1, characterized in that, S3 include: When the target solvent type is methyl ethyl carbonate, the pretreated material is fed into a vacuum distillation column, and light component discharge stream, side stream candidate product stream and heavy component discharge stream are formed in the order of light component removal at the top of the column, side stream collection and heavy component removal at the bottom of the column. The side stream collection window is determined according to the side stream position, pressure difference in the column and temperature at the bottom of the column, and a distillation status record is formed. When the target solvent type is ethylene carbonate, the pretreated material is sent to a segmented melt crystallizer after being depressurized to remove light and heavy components. A crystalline phase is formed in the crystallization section and the impurity-containing mother liquor is discharged in the sweating section to obtain a crystallization candidate product stream and a crystallization state record. The path status record includes candidate product flow records, as well as distillation status records corresponding to vacuum distillation paths or crystallization status records corresponding to segmented melting crystallization sweating paths. The path status records are used for subsequent soft measurement prediction and model predictive control.

5. The electrolyte solvent purification method according to claim 1, characterized in that, S4 include: When the target solvent is methyl ethyl carbonate, the distillation process status sequence is generated by aligning the column temperature, column pressure, vacuum degree, reflux ratio, side stream flow rate, internal pressure difference, online moisture detection value, online chromatographic detection value, and candidate product metal detection value according to the sampling time. When the target solvent type is ethylene carbonate, the jacket temperature, crystal layer temperature, mother liquor conductivity, mother liquor alcohol chromatographic detection value, mother liquor heavy impurity chromatographic detection value, online moisture detection value, candidate product flowing metal detection value, crystallization section number and sweating section number are aligned according to the sampling time to generate a crystallization process state sequence; The distillation process state sequence or crystallization process state sequence is combined with the raw material impurity fingerprint, target solvent type, bed state record of pretreated material, path state record, and historical batch removal efficiency into a soft measurement input vector; The prediction sub-model is trained from the process state sequence of historical batches, laboratory retest purity, and laboratory retest impurity residue values, and outputs the product purity prediction value, impurity residue prediction value, and confidence interval corresponding to the current batch.

6. The electrolyte solvent purification method according to claim 1, characterized in that, S5 include: The model predictive controller uses vacuum degree, reflux ratio, column temperature, product yield, column bottom discharge ratio, water removal residence time, resin bed contact time, heavy impurity protection bed contact time, side-stream product yield window threshold, adsorption bed switching status, and crystallization sweating endpoint as control variables. The column temperature is the temperature setting of the vacuum distillation column or the vacuum removal stage for removing light and heavy components. The product yield is the side-stream product yield or the feed yield into the crystallizer. The column bottom discharge ratio is the discharge ratio of heavy components discharged from the column bottom. Inapplicable control variables are set to disabled or zero according to the target solvent type. The model predictive controller uses the target limit for product purity, the target limit for residual impurities, the equipment operating boundary, and the confidence interval as constraints. When the upper boundary of the confidence interval indicates that the predicted value of any impurity residue reaches the preset proportion of the corresponding target limit and there is a candidate action that is applicable to the path and meets the equipment operation boundary, the model prediction controller increases the control margin corresponding to the impurity and outputs at least one of the following refining control actions: water removal residence time, light reflux ratio, resin bed contact time, metal capture and reprocessing, heavy weight discharge ratio in the tower bottom, heavy impurity protection bed contact time, side stream extraction window, and crystallization sweating endpoint; When the model is disabled, the action constraints have no feasible solution, the safety state is triggered, or the control priority queue is still empty when the upper boundary of any confidence interval reaches the preset proportion of the corresponding target limit, output a reprocessing instruction, a manual review instruction, or a prohibition release flag.

7. The electrolyte solvent purification method according to claim 5, characterized in that, The impurity migration risk index is obtained by weighting the moisture migration sub-index, alcohol migration sub-index, metal migration index, and heavy impurity migration index. Each migration sub-index is generated based on the corresponding untruncation standard detection value, the corresponding process detection value, the corresponding historical batch removal efficiency, and the corresponding product target limit. The metal migration sub-index preferentially uses the metal detection value of the candidate product stream as the corresponding process detection value. When the metal detection value of the candidate product stream is missing, the metal ion value at the metal trapping bed outlet is used as the proxy process detection value and written into the proxy low confidence flag. The weights of each migration sub-index are determined by the target solvent type and the reasons for non-compliance in historical batches when the number of historical samples reaches the preset minimum number of samples. When the number of historical samples is insufficient, the previous frozen version is used or the default weight reorganization is adopted and a weight rollback flag is generated. When any migration sub-index reaches a preset first threshold, a level of concern is generated; when any migration sub-index reaches a preset second threshold, a level of interception is generated. The preset second threshold is greater than the preset first threshold.

8. The method for refining electrolyte solvent according to claim 7, characterized in that, The model prediction controller generates control actions based on different migration indexes and target solvent types: When the moisture migration sub-index reaches the level of concern, extend the water removal retention time or switch to a standby water removal bed. When the alcohol migration index reaches the level of concern and the target solvent type is ethyl methyl carbonate, increase the reflux ratio for light removal and narrow the side-line sampling window. When the target solvent type is ethylene carbonate, extend the vacuum removal time or adjust the crystallization sweating endpoint. When the metal migration index reaches the concern level, the resin bed regeneration is effective, and the outlet metal ion detection value does not reach the preset proportion of the corresponding target limit, the resin bed contact time is extended. When the outlet metal ion detection value reaches the preset proportion or the resin bed regeneration effective sign fails, the metal capture and reprocessing is initiated. When the heavy impurity migration index reaches the level of concern and the target solvent type is methyl ethyl carbonate, increase the proportion of heavy removal from the bottom of the column or extend the contact time of the heavy impurity protection bed. When the target solvent type is ethylene carbonate, extend the depressurization time, maintain the state of discharging mother liquor, or extend the contact time of the heavy impurity protection bed. When any migration sub-index reaches the interception level, a reprocessing instruction is generated and the corresponding material is prohibited from entering the finished product tank.

9. The method for refining electrolyte solvent according to claim 4, characterized in that, The segmented melting crystallization sweating path includes a crystallization section, a heating sweating section, and a mother liquor discharge section; In the crystallization section, the cooling rate is determined based on the online moisture content, alcohol chromatographic values, heavy impurity chromatographic values, and crystallizer jacket temperature of the ethylene carbonate candidate material entering the segmented melting crystallizer. During the heating and sweating stage, the sweating endpoint score and crystallization state are determined and recorded based on the mother liquor conductivity, chromatographic detection value of heavy impurities in the mother liquor, and crystal layer temperature. The sweating endpoint score is only used as input for the model prediction controller to determine the re-inspection trigger and crystallization sweating endpoint control actions, and is not used as a product qualification judgment. In step S5, when the model prediction controller determines that the conditions for re-inspection are met based on the sweating endpoint score, the predicted value of impurity residue, the confidence interval, and the equipment operation boundary, the crystallization candidate product stream is marked as a state to be re-inspected.

10. The method for refining electrolyte solvent according to claim 8, characterized in that, For materials that have formed a side-line candidate product stream or a crystallization candidate product stream and entered the release verification, before issuing the product release instruction or the reprocessing instruction based on the candidate product stream, the laboratory re-inspection results of the side-line candidate product stream or the crystallization candidate product stream are checked for consistency with the product purity prediction value, impurity residue prediction value and impurity migration risk index. When the laboratory retest results meet the product target limit, no migration sub-index reaches the preset second threshold, the consistency check passes, and there are no abnormal path selection, disabled model, missing parameter table, out-of-bounds equipment temperature, safety status trigger, prohibited release flag, reprocessing priority flag, or low confidence flag for agents, the product release instruction is output and the corresponding material is imported into the finished product tank. When the laboratory retest results do not meet the target limit of any product, any migration sub-index reaches the preset second threshold, path selection is abnormal, model is disabled, parameter table is not hit, equipment temperature exceeds the limit, safety status is triggered, release is prohibited, reprocessing priority is marked, agent low confidence is marked, or consistency verification fails, a reprocessing instruction is output and the corresponding material is imported into the rework tank or the tank to be manually reviewed. When both the product release condition and the reprocessing condition are met, the reprocessing instruction has a higher priority than the product release instruction. When a fingerprint is invalid, preprocessing or reprocessing is performed, path selection is abnormal, or security pre-processing is abnormal, resulting in the failure to form a re-inspectable candidate product flow, the reprocessing instruction, the manual review instruction, or the prohibition release flag is directly output, and the consistency verification is not performed.

Images