A Method and System for Anomaly Correction of Industrial Big Data in the Entire Process of Low Conductivity Resin Preparation

By constructing a multi-dimensional feature baseline library and a dynamic control strategy library, and combining multi-agent collaborative verification, the precise location and correction of abnormal parameters in the preparation process of low conductivity resin were achieved. This solved the problem of lack of stage positioning in parameter correction in the existing technology, and ensured consistency and product stability throughout the entire process.

CN122889107BActive Publication Date: 2026-11-10SHANXI XINGAN NEW MATERIALS CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611368456.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-09-04
Publication Date
2026-11-10
Estimated Expiration
2046-09-04

AI Technical Summary

Technical Problem

In the preparation of low conductivity resins, existing technologies lack stage-specific identification and causal traceability for anomaly identification and parameter correction, which makes it impossible to guarantee the system consistency of each stage of the entire process and affects the stability of product quality.

Method used

A multi-dimensional feature baseline library is constructed, multi-source sensor data is collected in real time, mapped to stage state vectors through a state space model, multi-dimensional membership is determined, a correction template from a dynamic control strategy library is called, and cross-validation is performed based on a multi-agent collaborative verification engine to ensure the validity and consistency of the corrected data.

Benefits of technology

It achieves precise location and root cause correction of abnormal parameters at each stage, ensuring parameter consistency throughout the process, improving product quality stability, and reducing reliance on human experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122889107B_ABST
    Figure CN122889107B_ABST
Patent Text Reader

Abstract

This invention relates to the field of industrial automation control and chemical production process data processing technology, and discloses a method and system for anomaly correction of industrial big data throughout the entire process of low conductivity resin preparation. The method includes: constructing a multi-dimensional feature baseline library; collecting multi-source sensor data and mapping it to corresponding preparation stages as stage state vectors; comparing the stage state vectors with the multi-dimensional feature baseline library to determine multi-dimensional membership and generate anomaly stage vectors; calling a matching correction template from a dynamic control strategy library to adjust parameters and generate corrected data; performing cross-validation on the corrected data through a multi-agent collaborative verification engine, marking the data as valid correction data when all verification agents return a verification pass signal; and adding the valid correction data to the multi-dimensional feature baseline library and updating the correction template. This invention achieves anomaly stage-level localization and full-process system consistency verification, improving correction effectiveness and product quality stability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of industrial automation control and chemical production process data processing technology, specifically to a method and system for anomaly correction of industrial big data in the entire process of low conductivity resin preparation. Background Technology

[0002] Low conductivity resins are increasingly used in electronic packaging, high-frequency communication equipment, and aerospace. Their preparation process typically involves multiple continuous stages, such as batching, polymerization, ion exchange, washing and purification, and drying and molding. There are complex correlations and constraints between key process parameters in each stage, such as reaction temperature, stirring speed, system pressure, and pH value. Currently, existing technical solutions for anomaly monitoring and parameter correction in chemical production processes mostly adopt anomaly identification methods based on fixed rules or single thresholds. That is, an alarm is triggered when a parameter exceeds a preset threshold, and then the operator manually adjusts the relevant parameters based on experience. However, this type of method has obvious drawbacks: anomaly identification and parameter correction are logically independent, and the correction operation lacks stage-specific location and causal tracing of the anomaly's root cause. As a result, the corrected parameters often only meet the index requirements of a single stage, but may disrupt the process balance of other related stages, failing to guarantee the systematic consistency of parameters in all stages of the entire process, and thus affecting the stability of the final product quality. Summary of the Invention

[0003] The purpose of this invention is to provide an industrial big data anomaly correction method and system for the entire process of low conductivity resin preparation, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for correcting industrial big data anomalies in the entire process of low conductivity resin preparation, comprising the following steps: S1: Construct a multi-dimensional feature baseline library for the entire process of low conductivity resin preparation. The multi-dimensional feature baseline library includes standard parameter ranges, parameter change rate thresholds, and correlation constraints between parameters for each preparation stage. S2: Real-time acquisition of multi-source sensor data throughout the entire process of low conductivity resin preparation. Based on the timestamps and equipment identifiers of the preparation process, the acquired multi-source sensor data is divided into the corresponding preparation stages. Based on the preset state space model of each preparation stage, the divided data is mapped into the corresponding stage state vector. S3: Compare the stage state vector of each preparation stage with the standard parameter range and parameter change rate threshold of each preparation stage in the multidimensional feature baseline library, determine the multidimensional membership degree of each parameter, generate an abnormal stage vector based on the multidimensional membership degree, and record the abnormal parameter identifier, the abnormal occurrence stage and the abnormal deviation direction. S4: Based on the abnormal parameter identifier and abnormal occurrence stage in the abnormal stage vector, call the correction template in the dynamic control strategy library that matches the abnormal parameter identifier and abnormal occurrence stage, perform parameter adjustment on the abnormal parameter according to the correction template, and generate corrected data. The dynamic control strategy library stores the parameter adjustment sequence corresponding to different abnormal parameter identifiers in each preparation stage. S5: Synchronize the corrected data to the multi-agent collaborative verification engine for the entire low conductivity resin preparation process. The multi-agent collaborative verification engine contains verification agents corresponding to each preparation stage. Each verification agent performs cross-validation on the corrected data based on the parameter correlation constraint relationship in the multi-dimensional feature baseline library. When all verification agents return a verification pass signal, the corrected data is marked as valid corrected data. S6: The effective calibration data are classified into the multidimensional feature baseline library according to the preparation stage and status label, and the parameter adjustment sequence of the corresponding calibration template in the dynamic regulation strategy library is updated based on the effective calibration data.

[0005] As a preferred embodiment of the present invention, step S2 maps the partitioned data into corresponding stage state vectors based on the preset state space model of each preparation stage, specifically including: Obtain the set of equipment operating status parameters output by all online monitoring devices in the current preparation stage, as well as the set of environmental status parameters of the process area where the current preparation stage is located; The set of equipment operating status parameters and the set of environmental status parameters are input into the preset state space model. The preset state space model stores the expected amplitude range and expected fluctuation rhythm of each parameter under various standard operating conditions at the current preparation stage. The preset state-space model extracts the amplitude deviation of each parameter's actual measured value relative to the expected amplitude range, as well as the rhythm deviation of each parameter's actual measured value's change rhythm relative to the expected fluctuation rhythm, based on the actual measured values ​​of each parameter. The amplitude deviation and rhythm deviation of each parameter are encoded into a binary sequence according to the parameter category, and the binary sequence is output as the stage state vector of the current preparation stage.

[0006] As a preferred embodiment of the present invention, the determination of the multidimensional membership degree of each parameter in step S3 specifically includes: Extract the amplitude deviation and rhythm deviation corresponding to the current parameters from the stage state vector; The amplitude deviation is used as the first dimension, and a first spatial position comparison is performed with the allowable amplitude deviation range of the current parameter in the current preparation stage in the multidimensional feature baseline library to determine the relative position level of the amplitude deviation within the allowable amplitude deviation range. The relative position level is used as the first membership component. The rhythm deviation is used as the second dimension and compared with the allowable rhythm deviation threshold of the current parameter in the current preparation stage in the multidimensional feature baseline library to determine the degree of exceedance of the rhythm deviation relative to the allowable rhythm deviation threshold. The degree of exceedance is used as the second membership component. The first membership component and the second membership component are combined into a multidimensional membership vector, which includes the amplitude dimension membership level and the rhythm dimension membership level. If either the amplitude dimension membership level or the rhythm dimension membership level exceeds the corresponding normal level range, the current parameter is determined to be an abnormal parameter, and the abnormal parameter identifier and abnormal deviation direction are recorded. The abnormal deviation direction is determined by the positive and negative signs of the amplitude deviation and the rhythm deviation.

[0007] As a preferred embodiment of the present invention, step S4, which involves calling a correction template from the dynamic control strategy library that matches the abnormal parameter identifier and the stage of abnormal occurrence, specifically includes: Based on the abnormal occurrence stage in the abnormal stage vector, the corresponding stage strategy sub-library is indexed from the dynamic control strategy library. Based on the abnormal parameter identifier in the abnormal stage vector, a set of candidate correction templates containing abnormal parameter identifiers is selected from the stage strategy sub-library. Each candidate correction template in the set of candidate correction templates corresponds to a different abnormal deviation direction level. Based on the abnormal deviation direction in the abnormal stage vector, a target correction template matching the abnormal deviation direction is selected from the candidate correction template set. The target correction template contains multi-level adjustment instructions arranged in chronological order. Each level of adjustment instruction carries the identification of the adjustment target, the single adjustment amplitude value, and the expected state return check conditions after adjustment.

[0008] As a preferred embodiment of the present invention, step S4, which involves adjusting the abnormal parameters according to the correction template, specifically includes: According to the order of the multi-level adjustment instructions in the target correction template, the adjustment instructions for the single adjustment amplitude value are sent to the execution device corresponding to the adjustment target identifier in sequence; After each level of adjustment instruction is executed, the real-time feedback parameters of the preparation stage where the adjustment target is located are obtained; The real-time feedback parameters are compared with the expected state check conditions carried by the current level adjustment command; if the real-time feedback parameters meet the expected state check conditions, the next level adjustment command is executed; if the real-time feedback parameters do not meet the expected state check conditions, the execution of subsequent adjustment commands is paused, and the currently executed adjustment command sequence and the corresponding real-time feedback parameters are recorded as interruption status data. Once all adjustment instructions at all levels have been executed and the real-time feedback parameters at each level meet the corresponding expected state check conditions, the data after the entire sequence of executed adjustment instructions is output as the corrected data.

[0009] As a preferred embodiment of the present invention, in step S5, each verification agent performs cross-validation on the corrected data based on the parameter correlation constraints in the multidimensional feature baseline library, specifically including: The first verification agent extracts the first adjusted value of the adjusted parameter in the current preparation stage from the corrected data, and queries the identifier of the associated parameter with the adjusted parameter and the target preparation stage where the associated parameter is located based on the parameter correlation constraint relationship in the multidimensional feature baseline library. The first verification agent encapsulates the first adjusted value, the associated parameter identifier, and the target preparation stage into a cross-stage verification request and sends it to the second verification agent corresponding to the target preparation stage. After receiving the cross-stage verification request, the second verification agent extracts the second actual feedback value of the associated parameter from the corrected data, and compares the second actual feedback value with the standard parameter range of the associated parameter in the multidimensional feature baseline library under the target preparation stage to obtain the first comparison result. The second verification agent further compares the actual ratio between the first adjusted value and the second actual feedback value with the linkage ratio range defined in the parameter correlation constraint relationship, and obtains the second comparison result. When both the first and second comparison results indicate that the comparison has passed, the second verification agent returns a verification pass signal to the first verification agent.

[0010] As a preferred embodiment of the present invention, step S5, in which the corrected data is marked as valid corrected data when all verification agents return a verification pass signal, specifically includes: The first verification agent continuously receives verification response signals returned by all verification agents corresponding to each preparation stage, and records the verification pass signal and verification fail signal to the collaborative verification log according to the agent identifier of the signal source; When all verification agents return verification pass signals, the first verification agent extracts the signal timestamps corresponding to each verification agent from the collaborative verification log. After confirming that each verification pass signal meets the preset time consistency window condition based on the signal timestamps, it sends a valid confirmation command to the data marking module. When any verification agent returns a verification failure signal, the first verification agent determines the concentration of verification failure signals based on the number of verification failure signals and the corresponding agent identifiers recorded in the collaborative verification log. When the verification failure signal set originates from a single verification agent and all other verification agents return verification success signals, the first verification agent marks the single verification agent as a temporarily untrusted state, and writes the agent identifier corresponding to the temporarily untrusted state and the verification failure signal into the verification note field of the corrected data, while sending a valid confirmation instruction to the data marking module. After receiving a valid confirmation instruction, the data tagging module obtains the preparation batch identifier and timestamp corresponding to the corrected data, and writes the preparation batch identifier, timestamp, and verification via tag association into the header field of the corrected data to form valid corrected data.

[0011] As a preferred embodiment of the present invention, step S6, which involves updating the parameter adjustment sequence of the corresponding calibration template in the dynamic control strategy library based on valid calibration data, specifically includes: Extract the actual adjustment trajectory and actual recovery time of abnormal parameters from the valid correction data; The actual adjustment trajectory is compared with the parameter adjustment sequence in the calibration template to determine the deviation segment between the actual adjustment trajectory and the parameter adjustment sequence. The deviation segment includes the adjustment magnitude deviation value and the adjustment timing deviation value. The deviation segments are associated and stored in the cumulative deviation record area of ​​the dynamic control strategy library according to the preparation stage identifier and the abnormal parameter identifier. When the cumulative number of deviation segments corresponding to the same preparation stage and the same abnormal parameter identifier in the cumulative deviation record area reaches the preset trigger threshold, the adjustment step size of the corresponding stage in the parameter adjustment sequence is uniformly corrected according to the accumulated multiple deviation segments to generate the correction template after the first update. The updated calibration template is backfilled into the dynamic control strategy library to replace the original calibration template. Valid calibration data generated by applying the updated calibration template in at least one subsequent adjustment cycle is obtained. The feedback adjustment trajectory corresponding to the valid calibration data is compared with the expected feedback trajectory in the updated calibration template to obtain a feedback consistency index. If the feedback consistency index meets the preset steady-state conditions, the updated calibration template is marked as a verified updated template. If the feedback consistency index does not meet the preset steady-state conditions, the process is rolled back to the original calibration template, and a preset trigger threshold is added to the cumulative deviation record area.

[0012] An industrial big data anomaly correction system for the entire process of low conductivity resin preparation, used to execute the industrial big data anomaly correction method for the entire process of low conductivity resin preparation described in this invention, includes: The baseline construction module is used to build a multi-dimensional feature baseline library for the entire process of low conductivity resin preparation. The multi-dimensional feature baseline library includes standard parameter ranges, parameter change rate thresholds, and correlation constraints between parameters for each preparation stage. The data acquisition and state mapping module is used to collect multi-source sensor data in real time throughout the entire process of low conductivity resin preparation. Based on the timestamp of the preparation process and the equipment identifier, the collected multi-source sensor data is divided into the corresponding preparation stages, and the divided data is mapped into the corresponding stage state vector according to the preset state space model of each preparation stage. The anomaly detection module is used to compare the stage state vector of each preparation stage with the standard parameter range and parameter change rate threshold of each preparation stage in the multidimensional feature baseline library, determine the multidimensional membership degree of each parameter, and generate anomaly stage vector based on the multidimensional membership degree. The anomaly stage vector records the anomaly parameter identifier, the stage of anomaly occurrence, and the direction of anomaly deviation. The dynamic correction module is used to call the correction template in the dynamic control strategy library that matches the abnormal parameter identifier and the abnormal occurrence stage according to the abnormal parameter identifier and the abnormal occurrence stage in the abnormal stage vector. The module then performs parameter adjustment on the abnormal parameters according to the correction template and generates corrected data. The dynamic control strategy library stores the parameter adjustment sequence corresponding to different abnormal parameter identifiers in each preparation stage. The multi-agent collaborative verification module is used to synchronize the corrected data to the multi-agent collaborative verification engine. The multi-agent collaborative verification engine contains verification agents corresponding to each preparation stage. Each verification agent performs cross-validation on the corrected data based on the parameter correlation constraints in the multi-dimensional feature baseline library. When all verification agents return a verification pass signal, the corrected data is marked as valid correction data. The baseline update module is used to classify the effective calibration data into the multidimensional feature baseline library according to the preparation stage and state label, and update the parameter adjustment sequence of the corresponding calibration template in the dynamic control strategy library based on the effective calibration data.

[0013] As a preferred embodiment of the present invention, the dynamic correction module calls a correction template from the dynamic control strategy library that matches the abnormal parameter identifier and the stage of abnormal occurrence, and specifically performs the following actions: Based on the abnormal occurrence stage in the abnormal stage vector, the corresponding stage strategy sub-library is indexed from the dynamic control strategy library. Based on the abnormal parameter identifier in the abnormal stage vector, a set of candidate correction templates containing abnormal parameter identifiers is selected from the stage strategy sub-library. Each candidate correction template in the set of candidate correction templates corresponds to a different abnormal deviation direction level. Based on the abnormal deviation direction in the abnormal stage vector, a target correction template matching the abnormal deviation direction is selected from the candidate correction template set. The target correction template contains multi-level adjustment instructions arranged in chronological order. Each level of adjustment instruction carries the identification of the adjustment target, the single adjustment amplitude value, and the expected state return check conditions after adjustment. According to the order of the multi-level adjustment instructions in the target correction template, the adjustment instructions for the single adjustment amplitude value are sent to the execution device corresponding to the adjustment target identifier in sequence; After each level of adjustment instruction is executed, the real-time feedback parameters of the preparation stage where the adjustment target is located are obtained; The real-time feedback parameters are compared with the expected state check conditions carried by the current level adjustment command; if the real-time feedback parameters meet the expected state check conditions, the next level adjustment command is executed; if the real-time feedback parameters do not meet the expected state check conditions, the execution of subsequent adjustment commands is paused, and the currently executed adjustment command sequence and the corresponding real-time feedback parameters are recorded as interruption status data. Once all adjustment instructions at all levels have been executed and the real-time feedback parameters at each level meet the corresponding expected state check conditions, the data after the entire sequence of executed adjustment instructions is output as the corrected data.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention establishes a stage state mapping and multi-dimensional membership determination mechanism covering the entire process of low conductivity resin preparation, thereby achieving precise stage-level positioning of abnormal parameters. This avoids the logical independence between anomaly identification and parameter correction in traditional methods. Based on the anomaly stage vector, the correction template in the dynamic control strategy library is called to perform parameter adjustment, making the correction operation stage-specific and root cause-oriented, thus improving the effectiveness of the correction.

[0015] 2. This invention deploys a multi-agent collaborative verification engine corresponding to each preparation stage. Each verification agent performs cross-stage cross-verification of the corrected data based on the correlation constraints between parameters in the multi-dimensional feature baseline library, ensuring that the corrected parameters maintain system consistency across all stages of the process. This overcomes the technical challenge that correction in a single stage may disrupt the process balance of other related stages, thus ensuring the stability of product quality.

[0016] 3. This invention establishes a closed-loop optimization mechanism based on feedback from actual correction effects by classifying effective correction data into a multidimensional feature baseline library according to preparation stage and status labels, and updating the parameter adjustment sequence of the corresponding correction template in the dynamic control strategy library based on the effective correction data. This enables the continuous evolution and dynamic adaptation of correction strategies and reduces the dependence on human experience intervention. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall process of the industrial big data anomaly correction method for the entire process of low conductivity resin preparation according to the present invention. Figure 2 This is a structural framework diagram of the industrial big data anomaly correction system for the entire process of low conductivity resin preparation according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1 like Figure 1 As shown, this invention provides a method for correcting industrial big data anomalies throughout the entire process of low conductivity resin preparation, including the following steps: S1: Construct a multi-dimensional feature baseline library for the entire process of low conductivity resin preparation. The multi-dimensional feature baseline library includes standard parameter ranges, parameter change rate thresholds, and correlation constraints between parameters for each preparation stage. S2: Real-time acquisition of multi-source sensor data throughout the entire process of low conductivity resin preparation. Based on the timestamps and equipment identifiers of the preparation process, the acquired multi-source sensor data is divided into the corresponding preparation stages. Based on the preset state space model of each preparation stage, the divided data is mapped into the corresponding stage state vector. S3: Compare the stage state vector of each preparation stage with the standard parameter range and parameter change rate threshold of each preparation stage in the multidimensional feature baseline library, determine the multidimensional membership degree of each parameter, generate an abnormal stage vector based on the multidimensional membership degree, and record the abnormal parameter identifier, the abnormal occurrence stage and the abnormal deviation direction. S4: Based on the abnormal parameter identifier and abnormal occurrence stage in the abnormal stage vector, call the correction template in the dynamic control strategy library that matches the abnormal parameter identifier and abnormal occurrence stage, perform parameter adjustment on the abnormal parameter according to the correction template, and generate corrected data. The dynamic control strategy library stores the parameter adjustment sequence corresponding to different abnormal parameter identifiers in each preparation stage. S5: Synchronize the corrected data to the multi-agent collaborative verification engine for the entire low conductivity resin preparation process. The multi-agent collaborative verification engine contains verification agents corresponding to each preparation stage. Each verification agent performs cross-validation on the corrected data based on the parameter correlation constraint relationship in the multi-dimensional feature baseline library. When all verification agents return a verification pass signal, the corrected data is marked as valid corrected data. S6: The effective calibration data are classified into the multidimensional feature baseline library according to the preparation stage and status label, and the parameter adjustment sequence of the corresponding calibration template in the dynamic regulation strategy library is updated based on the effective calibration data.

[0020] Furthermore, in step S2, the partitioned data is mapped to the corresponding stage state vectors based on the preset state space model for each preparation stage, specifically including: Obtain the set of equipment operating status parameters output by all online monitoring devices in the current preparation stage, as well as the set of environmental status parameters of the process area where the current preparation stage is located; The set of equipment operating status parameters and the set of environmental status parameters are input into the preset state space model. The preset state space model stores the expected amplitude range and expected fluctuation rhythm of each parameter under various standard operating conditions at the current preparation stage. The preset state-space model extracts the amplitude deviation of each parameter's actual measured value relative to the expected amplitude range, as well as the rhythm deviation of each parameter's actual measured value's change rhythm relative to the expected fluctuation rhythm, based on the actual measured values ​​of each parameter. The amplitude deviation and rhythm deviation of each parameter are encoded into a binary sequence according to the parameter category, and the binary sequence is output as the stage state vector of the current preparation stage.

[0021] Furthermore, step S3 determines the multidimensional membership degree of each parameter, specifically including: Extract the amplitude deviation and rhythm deviation corresponding to the current parameters from the stage state vector; The amplitude deviation is used as the first dimension, and a first spatial position comparison is performed with the allowable amplitude deviation range of the current parameter in the current preparation stage in the multidimensional feature baseline library to determine the relative position level of the amplitude deviation within the allowable amplitude deviation range. The relative position level is used as the first membership component. The rhythm deviation is used as the second dimension and compared with the allowable rhythm deviation threshold of the current parameter in the current preparation stage in the multidimensional feature baseline library to determine the degree of exceedance of the rhythm deviation relative to the allowable rhythm deviation threshold. The degree of exceedance is used as the second membership component. The first membership component and the second membership component are combined into a multidimensional membership vector, which includes the amplitude dimension membership level and the rhythm dimension membership level. If either the amplitude dimension membership level or the rhythm dimension membership level exceeds the corresponding normal level range, the current parameter is determined to be an abnormal parameter, and the abnormal parameter identifier and abnormal deviation direction are recorded. The abnormal deviation direction is determined by the positive and negative signs of the amplitude deviation and the rhythm deviation.

[0022] Furthermore, step S4 involves calling a correction template from the dynamic control strategy library that matches the abnormal parameter identifier and the stage of the abnormality, specifically including: Based on the abnormal occurrence stage in the abnormal stage vector, the corresponding stage strategy sub-library is indexed from the dynamic control strategy library. Based on the abnormal parameter identifier in the abnormal stage vector, a set of candidate correction templates containing abnormal parameter identifiers is selected from the stage strategy sub-library. Each candidate correction template in the set of candidate correction templates corresponds to a different abnormal deviation direction level. Based on the abnormal deviation direction in the abnormal stage vector, a target correction template matching the abnormal deviation direction is selected from the candidate correction template set. The target correction template contains multi-level adjustment instructions arranged in chronological order. Each level of adjustment instruction carries the identification of the adjustment target, the single adjustment amplitude value, and the expected state return check conditions after adjustment.

[0023] Furthermore, in step S4, parameter adjustments are performed on the abnormal parameters based on the correction template, specifically including: According to the order of the multi-level adjustment instructions in the target correction template, the adjustment instructions for the single adjustment amplitude value are sent to the execution device corresponding to the adjustment target identifier in sequence; After each level of adjustment instruction is executed, the real-time feedback parameters of the preparation stage where the adjustment target is located are obtained; The real-time feedback parameters are compared with the expected state check conditions carried by the current level adjustment command; if the real-time feedback parameters meet the expected state check conditions, the next level adjustment command is executed; if the real-time feedback parameters do not meet the expected state check conditions, the execution of subsequent adjustment commands is paused, and the currently executed adjustment command sequence and the corresponding real-time feedback parameters are recorded as interruption status data. Once all adjustment instructions at all levels have been executed and the real-time feedback parameters at each level meet the corresponding expected state check conditions, the data after the entire sequence of executed adjustment instructions is output as the corrected data.

[0024] Furthermore, in step S5, each verification agent performs cross-validation on the corrected data based on the parameter correlation constraints in the multidimensional feature baseline library, specifically including: The first verification agent extracts the first adjusted value of the adjusted parameter in the current preparation stage from the corrected data, and queries the identifier of the associated parameter with the adjusted parameter and the target preparation stage where the associated parameter is located based on the parameter correlation constraint relationship in the multidimensional feature baseline library. The first verification agent encapsulates the first adjusted value, the associated parameter identifier, and the target preparation stage into a cross-stage verification request and sends it to the second verification agent corresponding to the target preparation stage. After receiving the cross-stage verification request, the second verification agent extracts the second actual feedback value of the associated parameter from the corrected data, and compares the second actual feedback value with the standard parameter range of the associated parameter in the multidimensional feature baseline library under the target preparation stage to obtain the first comparison result. The second verification agent further compares the actual ratio between the first adjusted value and the second actual feedback value with the linkage ratio range defined in the parameter correlation constraint relationship, and obtains the second comparison result. When both the first and second comparison results indicate that the comparison has passed, the second verification agent returns a verification pass signal to the first verification agent.

[0025] Furthermore, in step S5, when all verification agents return a verification pass signal, the corrected data is marked as valid corrected data, specifically including: The first verification agent continuously receives verification response signals returned by all verification agents corresponding to each preparation stage, and records the verification pass signal and verification fail signal to the collaborative verification log according to the agent identifier of the signal source; When all verification agents return verification pass signals, the first verification agent extracts the signal timestamps corresponding to each verification agent from the collaborative verification log. After confirming that each verification pass signal meets the preset time consistency window condition based on the signal timestamps, it sends a valid confirmation command to the data marking module. When any verification agent returns a verification failure signal, the first verification agent determines the concentration of verification failure signals based on the number of verification failure signals and the corresponding agent identifiers recorded in the collaborative verification log. When the verification failure signal set originates from a single verification agent and all other verification agents return verification success signals, the first verification agent marks the single verification agent as a temporarily untrusted state, and writes the agent identifier corresponding to the temporarily untrusted state and the verification failure signal into the verification note field of the corrected data, while sending a valid confirmation instruction to the data marking module. After receiving a valid confirmation instruction, the data tagging module obtains the preparation batch identifier and timestamp corresponding to the corrected data, and writes the preparation batch identifier, timestamp, and verification via tag association into the header field of the corrected data to form valid corrected data.

[0026] Furthermore, in step S6, the parameter adjustment sequence of the corresponding calibration template in the dynamic control strategy library is updated based on the effective calibration data, specifically including: Extract the actual adjustment trajectory and actual recovery time of abnormal parameters from the valid correction data; The actual adjustment trajectory is compared with the parameter adjustment sequence in the calibration template to determine the deviation segment between the actual adjustment trajectory and the parameter adjustment sequence. The deviation segment includes the adjustment magnitude deviation value and the adjustment timing deviation value. The deviation segments are associated and stored in the cumulative deviation record area of ​​the dynamic control strategy library according to the preparation stage identifier and the abnormal parameter identifier. When the cumulative number of deviation segments corresponding to the same preparation stage and the same abnormal parameter identifier in the cumulative deviation record area reaches the preset trigger threshold, the adjustment step size of the corresponding stage in the parameter adjustment sequence is uniformly corrected according to the accumulated multiple deviation segments to generate the correction template after the first update. The updated calibration template is backfilled into the dynamic control strategy library to replace the original calibration template. Valid calibration data generated by applying the updated calibration template in at least one subsequent adjustment cycle is obtained. The feedback adjustment trajectory corresponding to the valid calibration data is compared with the expected feedback trajectory in the updated calibration template to obtain a feedback consistency index. If the feedback consistency index meets the preset steady-state conditions, the updated calibration template is marked as a verified updated template. If the feedback consistency index does not meet the preset steady-state conditions, the process is rolled back to the original calibration template, and a preset trigger threshold is added to the cumulative deviation record area.

[0027] like Figure 2 As shown, the present invention also provides an industrial big data anomaly correction system for the entire process of low conductivity resin preparation, used to perform any of the methods described above, including: The baseline construction module is used to build a multi-dimensional feature baseline library for the entire process of low conductivity resin preparation. The multi-dimensional feature baseline library includes standard parameter ranges, parameter change rate thresholds, and correlation constraints between parameters for each preparation stage. The data acquisition and state mapping module is used to collect multi-source sensor data in real time throughout the entire process of low conductivity resin preparation. Based on the timestamp of the preparation process and the equipment identifier, the collected multi-source sensor data is divided into the corresponding preparation stages, and the divided data is mapped into the corresponding stage state vector according to the preset state space model of each preparation stage. The anomaly detection module is used to compare the stage state vector of each preparation stage with the standard parameter range and parameter change rate threshold of each preparation stage in the multidimensional feature baseline library, determine the multidimensional membership degree of each parameter, and generate anomaly stage vector based on the multidimensional membership degree. The anomaly stage vector records the anomaly parameter identifier, the stage of anomaly occurrence, and the direction of anomaly deviation. The dynamic correction module is used to call the correction template in the dynamic control strategy library that matches the abnormal parameter identifier and the abnormal occurrence stage according to the abnormal parameter identifier and the abnormal occurrence stage in the abnormal stage vector. The module then performs parameter adjustment on the abnormal parameters according to the correction template and generates corrected data. The dynamic control strategy library stores the parameter adjustment sequence corresponding to different abnormal parameter identifiers in each preparation stage. The multi-agent collaborative verification module is used to synchronize the corrected data to the multi-agent collaborative verification engine. The multi-agent collaborative verification engine contains verification agents corresponding to each preparation stage. Each verification agent performs cross-validation on the corrected data based on the parameter correlation constraints in the multi-dimensional feature baseline library. When all verification agents return a verification pass signal, the corrected data is marked as valid correction data. The baseline update module is used to classify the effective calibration data into the multidimensional feature baseline library according to the preparation stage and state label, and update the parameter adjustment sequence of the corresponding calibration template in the dynamic control strategy library based on the effective calibration data.

[0028] Furthermore, the dynamic correction module calls the correction template in the dynamic control strategy library that matches the abnormal parameter identifier and the stage of the abnormality, and specifically performs the following actions: Based on the abnormal occurrence stage in the abnormal stage vector, the corresponding stage strategy sub-library is indexed from the dynamic control strategy library. Based on the abnormal parameter identifier in the abnormal stage vector, a set of candidate correction templates containing abnormal parameter identifiers is selected from the stage strategy sub-library. Each candidate correction template in the set of candidate correction templates corresponds to a different abnormal deviation direction level. Based on the abnormal deviation direction in the abnormal stage vector, a target correction template matching the abnormal deviation direction is selected from the candidate correction template set. The target correction template contains multi-level adjustment instructions arranged in chronological order. Each level of adjustment instruction carries the identification of the adjustment target, the single adjustment amplitude value, and the expected state return check conditions after adjustment. According to the order of the multi-level adjustment instructions in the target correction template, the adjustment instructions for the single adjustment amplitude value are sent to the execution device corresponding to the adjustment target identifier in sequence; After each level of adjustment instruction is executed, the real-time feedback parameters of the preparation stage where the adjustment target is located are obtained; The real-time feedback parameters are compared with the expected state check conditions carried by the current level adjustment command; if the real-time feedback parameters meet the expected state check conditions, the next level adjustment command is executed; if the real-time feedback parameters do not meet the expected state check conditions, the execution of subsequent adjustment commands is paused, and the currently executed adjustment command sequence and the corresponding real-time feedback parameters are recorded as interruption status data. Once all adjustment instructions at all levels have been executed and the real-time feedback parameters at each level meet the corresponding expected state check conditions, the data after the entire sequence of executed adjustment instructions is output as the corrected data.

[0029] Example 2 Based on the above-mentioned industrial big data anomaly correction method for the entire process of low conductivity resin preparation, this embodiment further refines and supplements the explanation of the stage state vector mapping, multidimensional membership determination and dynamic correction template calling process in steps S2 to S4, combined with the actual working conditions in the low conductivity resin preparation process.

[0030] Taking the entire preparation process of a low conductivity resin production base in East China as an example, the preparation process of the low conductivity resin at this base includes the following stages: batching, polymerization reaction, ion exchange, washing and purification, and drying and molding. Each preparation stage is equipped with online monitoring devices such as temperature sensors, pressure sensors, pH sensors, conductivity sensors, and flow meters. During the execution of step S2, the data acquisition and status mapping module acquires the set of equipment operating status parameters output by all online monitoring devices in real time at a acquisition frequency of 1 time / s. At the same time, it acquires the ambient temperature and humidity parameters of the process area where the current preparation stage is located. The key control parameters of the polymerization reaction stage include reactor temperature, stirring speed, reactor pressure, and pH value of the reaction system. The corresponding expected amplitude ranges are 78℃ to 82℃, 120rpm to 150rpm, 0.25MPa to 0.35MPa, and 6.8 to 7.2, respectively. The corresponding expected fluctuation rhythm is that the change amplitude of each parameter in two adjacent acquisition cycles is 5% of the width of the expected amplitude range.

[0031] The preset state-space model stores the expected amplitude range and expected fluctuation rhythm of each parameter under standard operating conditions for the polymerization reaction stage. When the actual measured value of each parameter is input at a certain acquisition time, the preset state-space model extracts the amplitude deviation and rhythm deviation of each parameter one by one. At a certain acquisition time, the actual measured value of the reactor temperature is 84.5℃, which exceeds the upper limit of the expected amplitude range of 78℃ to 82℃ by 2.5℃. This excess is taken as the amplitude deviation of the temperature parameter. At the same time, the reactor temperature at this time has increased by 1.8℃ compared with the previous acquisition time. The maximum allowable change between adjacent cycles of the expected temperature fluctuation rhythm at this stage is 0.2℃. This 1.8℃ is taken as the rhythm deviation of the temperature parameter. After encoding the amplitude deviation and rhythm deviation of each parameter into a binary sequence according to the parameter category, the stage state vector representing the current operating state of the polymerization reaction stage is output.

[0032] During step S3, the anomaly detection module extracts the amplitude deviation and rhythm deviation corresponding to the current parameters from the stage state vector of the polymerization reaction stage. The amplitude deviation of the reactor temperature parameter is +2.5℃, and the rhythm deviation is +1.8℃. The anomaly detection module uses the amplitude deviation of +2.5℃ as the first dimension and performs a first spatial position comparison with the allowable amplitude deviation range of the temperature parameter under the polymerization reaction stage in the multidimensional feature baseline library. This allowable amplitude deviation range is -2℃ to +2℃. The amplitude deviation of +2.5℃ exceeds the upper limit of the allowable amplitude deviation range by 0.5℃. According to the pre-divided deviation levels, exceeding the upper limit by 0.5℃ corresponds to a relative position level of 2, which is used as the first membership component. At the same time, the anomaly detection module uses the rhythm deviation of +1.8℃ as the second dimension and performs a second spatial position comparison with the allowable rhythm deviation threshold of the temperature parameter under the polymerization reaction stage in the multidimensional feature baseline library. A second spatial position comparison is performed. The allowable rhythm deviation threshold is 1.0℃. The rhythm deviation of +1.8℃ exceeds the allowable rhythm deviation threshold by 0.8℃. According to the pre-defined exceedance levels, exceeding 0.8℃ corresponds to exceedance level 3. This level is used as the second membership component. The first membership component level 2 and the second membership component level 3 are combined into a multidimensional membership vector. The amplitude dimension membership level of this multidimensional membership vector is level 2, and the rhythm dimension membership level is level 3. The normal level range for both is level 0 to level 1. Currently, both the amplitude dimension membership level 2 and the rhythm dimension membership level 3 exceed the normal level range. The anomaly detection module determines that the reactor temperature parameter is an abnormal parameter and records the abnormal parameter as a temperature anomaly. The anomaly occurs in the polymerization reaction stage. The direction of the abnormal deviation is determined as a positive deviation based on the positive and negative signs of the amplitude deviation and the rhythm deviation.

[0033] During step S4, the dynamic correction module identifies the aggregation reaction stage as the abnormal occurrence stage in the abnormal stage vector. It then indexes the corresponding stage strategy sub-library from the dynamic control strategy library. This sub-library stores a set of candidate correction templates corresponding to each abnormal parameter identifier under the aggregation reaction stage. Further, based on the abnormal parameter identifier being a temperature anomaly, the dynamic correction module filters out a set of candidate correction templates containing a temperature anomaly identifier from the stage strategy sub-library. The three candidate correction templates in this set correspond to three abnormal deviation direction levels: slight positive deviation, moderate positive deviation, and severe positive deviation. The dynamic correction module selects a target correction template corresponding to moderate positive deviation based on the level corresponding to a positive deviation direction exceeding the specified amplitude. This target correction template includes… The system includes three levels of adjustment instructions arranged in chronological order. The first-level adjustment instruction targets the cooling water regulating valve of the polymerization reactor, with a single adjustment increment of 15%. The expected post-adjustment condition is that the reactor temperature decreases at a rate of 0.5℃ / min to 1.0℃ / min. The second-level adjustment instruction targets the frequency converter of the polymerization reactor's agitator motor, with a single adjustment increment of 10 rpm. The expected post-adjustment condition is that the uniformity of the reaction system's temperature distribution returns to the standard range. The third-level adjustment instruction targets the cooling water regulating valve of the polymerization reactor, with a single adjustment increment of 10%. The expected post-adjustment condition is that the reactor temperature drops back to the 80℃±1℃ range.

[0034] The dynamic correction module executes the three-level adjustment instructions sequentially. After executing the first-level adjustment instruction, the opening of the cooling water regulating valve increases by 15%. The dynamic correction module acquires the real-time feedback parameters of the reactor temperature in the subsequent three acquisition cycles. These real-time feedback parameters show that the reactor temperature decreases at a rate of 0.7℃ / min, meeting the expected return condition of 0.5℃ / min to 1.0℃ / min. The second-level adjustment instruction is then executed. After executing the second-level adjustment instruction, the stirring speed decreases by 10 rpm. The temperature distribution uniformity index acquired by the dynamic correction module returns to the standard range. The third-level adjustment instruction is then executed. After the third-level adjustment instruction is completed, the reactor temperature drops back to 79.5℃, meeting the expected return condition of 80℃±1℃. Furthermore, after all three-level adjustment instructions are executed, the real-time feedback parameters of each level meet the corresponding expected return condition. The dynamic correction module outputs the data after the entire sequence of executed adjustment instructions as the corrected data.

[0035] This embodiment achieves precise location and graded correction of temperature anomalies during the polymerization reaction stage in the preparation of low conductivity resin through the above-described processing method. By executing multi-level adjustment commands step by step and providing real-time feedback and verification, it avoids parameter overshoot or production fluctuations that may be caused by a single large adjustment, thereby improving the stability and reliability of the correction process.

[0036] Example 3 This embodiment further refines and supplements the explanation of the multi-agent collaborative verification, effective correction data marking, and correction template update process involved in steps S5 to S6, taking into account the multi-process linkage characteristics of the entire low conductivity resin preparation process.

[0037] Taking the low conductivity resin production base in East China as the application scenario, the entire process of low conductivity resin preparation at the base includes five preparation stages. The multi-agent collaborative verification engine is deployed with five verification agents, namely agent A corresponding to the batching stage, agent B corresponding to the polymerization reaction stage, agent C corresponding to the ion exchange stage, agent D corresponding to the washing and purification stage, and agent E corresponding to the drying and molding stage. The verification agents establish a bidirectional data communication link through industrial Ethernet. Any verification agent can initiate cross-stage verification requests to other verification agents and receive corresponding verification response signals.

[0038] During step S5, agent B in the polymerization reaction stage is used as the first verification agent. After the dynamic correction module completes the correction of the abnormal temperature in the polymerization reaction stage and outputs the corrected data, agent B extracts the first adjusted value of the adjusted parameter in the polymerization reaction stage, namely the reactor temperature, from the corrected data. The value is 79.5℃. Agent B queries the related parameters that have a correlation constraint with the adjusted parameter according to the parameter correlation constraint relationship in the multidimensional feature baseline library. The correlation constraint relationship between the polymerization reaction stage temperature and the ion exchange stage feed temperature stored in the multidimensional feature baseline library is as follows: for every 1℃ increase in the discharge temperature of the polymerization reaction stage, the feed temperature of the ion exchange stage increases by 0.6℃ to 0.8℃, and the standard range of the feed temperature of the ion exchange stage is 45℃ to 55℃. Based on this, agent B finds that the related parameter is identified as the feed temperature of the ion exchange stage, the target preparation stage where the related parameter is located is the ion exchange stage, and the corresponding target verification agent is agent C.

[0039] Agent B encapsulates the first adjusted value of 79.5℃, the associated parameter identified as the feed temperature of the ion exchange stage, and the target preparation stage as the ion exchange stage into a cross-stage verification request, and sends it to Agent C via industrial Ethernet. After receiving the cross-stage verification request, Agent C extracts the second actual feedback value of the feed temperature of the ion exchange stage from the corrected data. The second actual feedback value is 49.2℃. Agent C performs a first comparison between the second actual feedback value of 49.2℃ and the standard parameter range of 45℃ to 55℃ for the feed temperature of the ion exchange stage in the multidimensional feature baseline library. 49.2℃ is within the standard parameter range of 45℃ to 55℃, and the first comparison result indicates that it passes.

[0040] Agent C further performs a second comparison based on the linkage ratio range of 0.6 to 0.8 between the polymerization reaction stage temperature and the ion exchange stage feed temperature in the multidimensional feature baseline library. The first adjusted value of the polymerization reaction stage temperature is 79.5℃, which is 3.0℃ lower than the original value of 82.5℃. According to the linkage ratio range of 0.6 to 0.8, the theoretical linkage change range of the ion exchange stage feed temperature is 1.8℃ to 2.4℃. The original ion exchange stage feed temperature was 51.5℃, and the second actual feedback value after correction is 49.2℃, with an actual change range of 2.3℃, which is within the linkage change range of 1.8℃ to 2.4℃ defined by the linkage ratio range of 0.6 to 0.8. Based on this, Agent C determines that the second comparison result indicates a pass. With both the first and second comparison results indicating a pass, Agent C returns a verification pass signal to Agent B.

[0041] Agent B continuously receives verification response signals returned by the verification agents corresponding to all five preparation stages. Agents A, D, and E all return verification pass signals after going through their respective verification logic. Agent B records the verification pass signals and verification fail signals to the collaborative verification log according to the agent identifier of the signal source. When all five verification agents have returned verification pass signals, Agent B extracts the signal timestamps corresponding to each verification agent from the collaborative verification log. After confirming that the timestamps of each verification pass signal are within the preset time consistency window, i.e., all signals arrive within 1.5 seconds, Agent B sends a valid confirmation command to the data marking module.

[0042] In another production batch, agents A, B, D, and E all returned verification pass signals, while agent C returned a verification fail signal. The reason for this failure was that the actual variation between the second actual feedback value of the feed temperature in the ion exchange stage (49.2℃) and the first adjusted value of the polymerization reaction stage temperature (79.5℃) was 2.3℃. However, the direction of change between the reference value before and after correction of the feed temperature in the ion exchange stage was opposite to the direction of change defined by the linkage ratio range. Agent B, based on the fact that the number of verification fail signals recorded in the collaborative verification log was one and the corresponding agent identifier was agent C, determined that the verification failed. The failure signals originated from a single verification agent, Agent C, while the other four verification agents all returned verification pass signals. Agent B marked Agent C as temporarily untrusted and associated Agent C's identifier and the corresponding failure signal in the verification annotation field of the corrected data. At the same time, it sent a valid confirmation command to the data marking module. After receiving the valid confirmation command, the data marking module obtained the preparation batch identifier (20260628B3) and the timestamp (14:32:18) corresponding to the corrected data. It associated the preparation batch identifier, timestamp, and verification pass tag in the header field of the corrected data to form valid corrected data.

[0043] Agent C further performs a second comparison based on the linkage ratio range of 0.6 to 0.8 between the polymerization reaction stage temperature and the ion exchange stage feed temperature in the multidimensional feature baseline library. The first adjusted value of the polymerization reaction stage temperature is 79.5℃, which is 3.0℃ lower than the original value of 82.5℃. According to the linkage ratio range of 0.6 to 0.8, the theoretical linkage change range of the ion exchange stage feed temperature is 1.8℃ to 2.4℃. The original ion exchange stage feed temperature was 51.5℃, and the second actual feedback value after correction is 49.2℃, with an actual change range of 2.3℃, which is within the linkage change range of 1.8℃ to 2.4℃ defined by the linkage ratio range of 0.6 to 0.8. Based on this, Agent C determines that the second comparison result indicates a pass. With both the first and second comparison results indicating a pass, Agent C returns a verification pass signal to Agent B.

[0044] Agent B continuously receives verification response signals returned by the verification agents corresponding to all five preparation stages. Agents A, D, and E all return verification pass signals after going through their respective verification logic. Agent B records the verification pass signals and verification fail signals to the collaborative verification log according to the agent identifier of the signal source. When all five verification agents have returned verification pass signals, Agent B extracts the signal timestamps corresponding to each verification agent from the collaborative verification log. After confirming that the timestamps of each verification pass signal are within the preset time consistency window, i.e., all signals arrive within 1.5 seconds, Agent B sends a valid confirmation command to the data marking module.

[0045] In another production batch, agents A, B, D, and E all returned verification pass signals, while agent C returned a verification fail signal. The reason for this failure was that the actual variation between the second actual feedback value of the feed temperature in the ion exchange stage (49.2℃) and the first adjusted value of the polymerization reaction stage temperature (79.5℃) was 2.3℃. However, the direction of change between the reference value before and after correction of the feed temperature in the ion exchange stage was opposite to the direction of change defined by the linkage ratio range. Agent B, based on the fact that the number of verification fail signals recorded in the collaborative verification log was one and the corresponding agent identifier was agent C, determined that the verification failed. The failure signals originated from a single verification agent, Agent C, while the other four verification agents all returned verification pass signals. Agent B marked Agent C as temporarily untrusted and associated Agent C's identifier and the corresponding failure signal in the verification annotation field of the corrected data. At the same time, it sent a valid confirmation command to the data marking module. After receiving the valid confirmation command, the data marking module obtained the preparation batch identifier (20260628B3) and the timestamp (14:32:18) corresponding to the corrected data. It associated the preparation batch identifier, timestamp, and verification pass tag in the header field of the corrected data to form valid corrected data.

[0046] The baseline update module refills the initially updated calibration template into the dynamic control strategy library to replace the original calibration template, and obtains the effective calibration data generated by applying the initially updated calibration template in the subsequent five adjustment cycles. The baseline update module compares the feedback adjustment trajectory corresponding to the effective calibration data with the expected feedback trajectory in the initially updated calibration template to obtain a feedback consistency index. This feedback consistency index is defined as 90% of the proportion of the maximum deviation between the actual adjustment trajectory and the expected feedback trajectory falling within the range of -0.15℃ to +0.15℃ in the subsequent five adjustment cycles. The preset steady-state condition is 85% of the proportion of the maximum deviation falling within the range of -0.2℃ to +0.2℃. Currently, the feedback consistency index meets the preset steady-state condition at 90%. The baseline update module marks the initially updated calibration template as a verified updated template. If the proportion of the maximum deviation between the actual adjustment trajectory and the expected feedback trajectory falling within the range of -0.2℃ to +0.2℃ is less than 85% in the subsequent five adjustment cycles, the feedback consistency index does not meet the preset steady-state condition. The baseline update module rolls back to the original calibration template and increases the preset trigger threshold of the cumulative deviation recording area from 5 times to 8 times.

[0047] This embodiment achieves automatic verification of cross-stage parameter correlation consistency and provisional isolation of untrusted agents through the above processing method. At the same time, through the dual guarantee mechanism of cumulative deviation driving and feedback consistency verification, the adaptive optimization and safe rollback of the correction template are realized, which improves the reliability and robustness of the correction template update.

[0048] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A method for correcting industrial big data anomalies in the entire process of low conductivity resin preparation, characterized in that, Includes the following steps: S1: Construct a multi-dimensional feature baseline library for the entire process of low conductivity resin preparation. The multi-dimensional feature baseline library includes standard parameter ranges, parameter change rate thresholds, and correlation constraints between parameters for each preparation stage. S2: Real-time acquisition of multi-source sensor data throughout the entire process of low conductivity resin preparation. Based on the timestamps and equipment identifiers of the preparation process, the acquired multi-source sensor data is divided into the corresponding preparation stages. Based on the preset state space model of each preparation stage, the divided data is mapped into the corresponding stage state vector. S3: Compare the stage state vector of each preparation stage with the standard parameter range and parameter change rate threshold of each preparation stage in the multidimensional feature baseline library, determine the multidimensional membership degree of each parameter, generate an abnormal stage vector based on the multidimensional membership degree, and record the abnormal parameter identifier, the abnormal occurrence stage and the abnormal deviation direction. Among them, multidimensional membership degree includes: Extract the amplitude deviation and rhythm deviation corresponding to the current parameters from the stage state vector; The amplitude deviation is used as the first dimension, and a first spatial position comparison is performed with the allowable amplitude deviation range of the current parameter in the current preparation stage in the multidimensional feature baseline library to determine the relative position level of the amplitude deviation within the allowable amplitude deviation range. The relative position level is used as the first membership component. The rhythm deviation is used as the second dimension and compared with the allowable rhythm deviation threshold of the current parameter in the current preparation stage in the multidimensional feature baseline library to determine the degree of exceedance of the rhythm deviation relative to the allowable rhythm deviation threshold. The degree of exceedance is used as the second membership component. The first membership component and the second membership component are combined into a multidimensional membership vector, which includes the amplitude dimension membership level and the rhythm dimension membership level. S4: Based on the abnormal parameter identifier and abnormal occurrence stage in the abnormal stage vector, call the correction template in the dynamic control strategy library that matches the abnormal parameter identifier and abnormal occurrence stage, perform parameter adjustment on the abnormal parameter according to the correction template, and generate corrected data. The dynamic control strategy library stores the parameter adjustment sequence corresponding to different abnormal parameter identifiers in each preparation stage. S5: Synchronize the corrected data to the multi-agent collaborative verification engine for the entire low conductivity resin preparation process. The multi-agent collaborative verification engine contains verification agents corresponding to each preparation stage. Each verification agent performs cross-validation on the corrected data based on the parameter correlation constraint relationship in the multi-dimensional feature baseline library. When all verification agents return a verification pass signal, the corrected data is marked as valid corrected data. S6: The effective calibration data are classified into the multidimensional feature baseline library according to the preparation stage and status label, and the parameter adjustment sequence of the corresponding calibration template in the dynamic regulation strategy library is updated based on the effective calibration data.

2. The method according to claim 1, characterized in that, In step S2, the partitioned data is mapped to the corresponding stage state vector according to the preset state space model of each preparation stage. Specifically, this includes: Obtain the set of equipment operating status parameters output by all online monitoring devices in the current preparation stage, as well as the set of environmental status parameters of the process area where the current preparation stage is located; The set of equipment operating status parameters and the set of environmental status parameters are input into the preset state space model. The preset state space model stores the expected amplitude range and expected fluctuation rhythm of each parameter under various standard operating conditions at the current preparation stage. The preset state-space model extracts the amplitude deviation of each parameter's actual measured value relative to the expected amplitude range, as well as the rhythm deviation of each parameter's actual measured value's change rhythm relative to the expected fluctuation rhythm, based on the actual measured values ​​of each parameter. The amplitude deviation and rhythm deviation of each parameter are encoded into a binary sequence according to the parameter category, and the binary sequence is output as the stage state vector of the current preparation stage.

3. The method according to claim 1, characterized in that, The determination of the multidimensional membership degree of each parameter in step S3 includes: If either the amplitude dimension membership level or the rhythm dimension membership level exceeds the corresponding normal level range, the current parameter is determined to be an abnormal parameter, and the abnormal parameter identifier and abnormal deviation direction are recorded. The abnormal deviation direction is determined by the positive and negative signs of the amplitude deviation and the rhythm deviation.

4. The method according to claim 1, characterized in that, Step S4 involves calling a correction template from the dynamic control strategy library that matches the abnormal parameter identifier and the stage of the abnormality. Specifically, this includes: Based on the abnormal occurrence stage in the abnormal stage vector, the corresponding stage strategy sub-library is indexed from the dynamic control strategy library. Based on the abnormal parameter identifier in the abnormal stage vector, a set of candidate correction templates containing abnormal parameter identifiers is selected from the stage strategy sub-library. Each candidate correction template in the set of candidate correction templates corresponds to a different abnormal deviation direction level. Based on the abnormal deviation direction in the abnormal stage vector, a target correction template matching the abnormal deviation direction is selected from the candidate correction template set. The target correction template contains multi-level adjustment instructions arranged in chronological order. Each level of adjustment instruction carries the identification of the adjustment target, the single adjustment amplitude value, and the expected state return check conditions after adjustment.

5. The method according to claim 1, characterized in that, Step S4, which involves adjusting the abnormal parameters based on the correction template, specifically includes: According to the order of the multi-level adjustment instructions in the target correction template, the adjustment instructions for the single adjustment amplitude value are sent to the execution device corresponding to the adjustment target identifier in sequence; After each level of adjustment instruction is executed, the real-time feedback parameters of the preparation stage where the adjustment target is located are obtained; The real-time feedback parameters are compared with the expected state check conditions carried by the current level adjustment command; if the real-time feedback parameters meet the expected state check conditions, the next level adjustment command is executed; if the real-time feedback parameters do not meet the expected state check conditions, the execution of subsequent adjustment commands is paused, and the currently executed adjustment command sequence and the corresponding real-time feedback parameters are recorded as interruption status data. Once all adjustment instructions at all levels have been executed and the real-time feedback parameters at each level meet the corresponding expected state check conditions, the data after the entire sequence of executed adjustment instructions is output as the corrected data.

6. The method according to claim 1, characterized in that, In step S5, each verification agent performs cross-validation on the corrected data based on the parameter correlation constraints in the multidimensional feature baseline library, specifically including: The first verification agent extracts the first adjusted value of the adjusted parameter in the current preparation stage from the corrected data, and queries the identifier of the associated parameter with the adjusted parameter and the target preparation stage where the associated parameter is located based on the parameter correlation constraint relationship in the multidimensional feature baseline library. The first verification agent encapsulates the first adjusted value, the associated parameter identifier, and the target preparation stage into a cross-stage verification request and sends it to the second verification agent corresponding to the target preparation stage. After receiving the cross-stage verification request, the second verification agent extracts the second actual feedback value of the associated parameter from the corrected data, and compares the second actual feedback value with the standard parameter range of the associated parameter in the multidimensional feature baseline library under the target preparation stage to obtain the first comparison result. The second verification agent further compares the actual ratio between the first adjusted value and the second actual feedback value with the linkage ratio range defined in the parameter correlation constraint relationship, and obtains the second comparison result. When both the first and second comparison results indicate that the comparison has passed, the second verification agent returns a verification pass signal to the first verification agent.

7. The method according to claim 1, characterized in that, In step S5, when all verification agents return a verification pass signal, the corrected data is marked as valid corrected data. This specifically includes: The first verification agent continuously receives verification response signals returned by all verification agents corresponding to each preparation stage, and records the verification pass signal and verification fail signal to the collaborative verification log according to the agent identifier of the signal source; When all verification agents return verification pass signals, the first verification agent extracts the signal timestamps corresponding to each verification agent from the collaborative verification log. After confirming that each verification pass signal meets the preset time consistency window condition based on the signal timestamps, it sends a valid confirmation command to the data marking module. When any verification agent returns a verification failure signal, the first verification agent determines the concentration of verification failure signals based on the number of verification failure signals and the corresponding agent identifiers recorded in the collaborative verification log. When the verification failure signal set originates from a single verification agent and all other verification agents return verification success signals, the first verification agent marks the single verification agent as a temporarily untrusted state, and writes the agent identifier corresponding to the temporarily untrusted state and the verification failure signal into the verification note field of the corrected data, while sending a valid confirmation instruction to the data marking module. After receiving a valid confirmation instruction, the data tagging module obtains the preparation batch identifier and timestamp corresponding to the corrected data, and writes the preparation batch identifier, timestamp, and verification via tag association into the header field of the corrected data to form valid corrected data.

8. The method according to claim 1, characterized in that, Step S6 involves updating the parameter adjustment sequence of the corresponding calibration template in the dynamic control strategy library based on the effective calibration data. Specifically, this includes: Extract the actual adjustment trajectory and actual recovery time of abnormal parameters from the valid correction data; The actual adjustment trajectory is compared with the parameter adjustment sequence in the calibration template to determine the deviation segment between the actual adjustment trajectory and the parameter adjustment sequence. The deviation segment includes the adjustment magnitude deviation value and the adjustment timing deviation value. The deviation segments are associated and stored in the cumulative deviation record area of ​​the dynamic control strategy library according to the preparation stage identifier and the abnormal parameter identifier. When the cumulative number of deviation segments corresponding to the same preparation stage and the same abnormal parameter identifier in the cumulative deviation record area reaches the preset trigger threshold, the adjustment step size of the corresponding stage in the parameter adjustment sequence is uniformly corrected according to the accumulated multiple deviation segments to generate the correction template after the first update. The updated calibration template is backfilled into the dynamic control strategy library to replace the original calibration template. Valid calibration data generated by applying the updated calibration template in at least one subsequent adjustment cycle is obtained. The feedback adjustment trajectory corresponding to the valid calibration data is compared with the expected feedback trajectory in the updated calibration template to obtain a feedback consistency index. If the feedback consistency index meets the preset steady-state conditions, the updated calibration template is marked as a verified updated template. If the feedback consistency index does not meet the preset steady-state conditions, the process is rolled back to the original calibration template, and a preset trigger threshold is added to the cumulative deviation record area.

9. An industrial big data anomaly correction system for the entire process of low conductivity resin preparation, used to execute the method according to any one of claims 1-8, characterized in that, include: The baseline construction module is used to build a multi-dimensional feature baseline library for the entire process of low conductivity resin preparation. The multi-dimensional feature baseline library includes standard parameter ranges, parameter change rate thresholds, and correlation constraints between parameters for each preparation stage. The data acquisition and state mapping module is used to collect multi-source sensor data in real time throughout the entire process of low conductivity resin preparation. Based on the timestamp of the preparation process and the equipment identifier, the collected multi-source sensor data is divided into the corresponding preparation stages, and the divided data is mapped into the corresponding stage state vector according to the preset state space model of each preparation stage. The anomaly detection module is used to compare the stage state vector of each preparation stage with the standard parameter range and parameter change rate threshold of each preparation stage in the multidimensional feature baseline library, determine the multidimensional membership degree of each parameter, and generate anomaly stage vector based on the multidimensional membership degree. The anomaly stage vector records the anomaly parameter identifier, the stage of anomaly occurrence, and the direction of anomaly deviation. The dynamic correction module is used to call the correction template in the dynamic control strategy library that matches the abnormal parameter identifier and the abnormal occurrence stage according to the abnormal parameter identifier and the abnormal occurrence stage in the abnormal stage vector. The module then performs parameter adjustment on the abnormal parameters according to the correction template and generates corrected data. The dynamic control strategy library stores the parameter adjustment sequence corresponding to different abnormal parameter identifiers in each preparation stage. The multi-agent collaborative verification module is used to synchronize the corrected data to the multi-agent collaborative verification engine. The multi-agent collaborative verification engine contains verification agents corresponding to each preparation stage. Each verification agent performs cross-validation on the corrected data based on the parameter correlation constraints in the multi-dimensional feature baseline library. When all verification agents return a verification pass signal, the corrected data is marked as valid correction data. The baseline update module is used to classify the effective calibration data into the multidimensional feature baseline library according to the preparation stage and state label, and update the parameter adjustment sequence of the corresponding calibration template in the dynamic control strategy library based on the effective calibration data.

10. The system according to claim 9, characterized in that, The dynamic correction module calls the correction template in the dynamic control strategy library that matches the abnormal parameter identifier and the stage of the abnormality, and specifically performs the following actions: Based on the abnormal occurrence stage in the abnormal stage vector, the corresponding stage strategy sub-library is indexed from the dynamic control strategy library. Based on the abnormal parameter identifier in the abnormal stage vector, a set of candidate correction templates containing abnormal parameter identifiers is selected from the stage strategy sub-library. Each candidate correction template in the set of candidate correction templates corresponds to a different abnormal deviation direction level. Based on the abnormal deviation direction in the abnormal stage vector, a target correction template matching the abnormal deviation direction is selected from the candidate correction template set. The target correction template contains multi-level adjustment instructions arranged in chronological order. Each level of adjustment instruction carries the identification of the adjustment target, the single adjustment amplitude value, and the expected state return check conditions after adjustment. According to the order of the multi-level adjustment instructions in the target correction template, the adjustment instructions for the single adjustment amplitude value are sent to the execution device corresponding to the adjustment target identifier in sequence; After each level of adjustment instruction is executed, the real-time feedback parameters of the preparation stage where the adjustment target is located are obtained; The real-time feedback parameters are compared with the expected state check conditions carried by the current level adjustment command; if the real-time feedback parameters meet the expected state check conditions, the next level adjustment command is executed; if the real-time feedback parameters do not meet the expected state check conditions, the execution of subsequent adjustment commands is paused, and the currently executed adjustment command sequence and the corresponding real-time feedback parameters are recorded as interruption status data. Once all adjustment instructions at all levels have been executed and the real-time feedback parameters at each level meet the corresponding expected state check conditions, the data after the entire sequence of executed adjustment instructions is output as the corrected data.

Citation Information

Patent Citations

  • Industrial equipment energy consumption optimization method based on sensor data and fuzzy neural network

    CN121187131A

  • Distributed Additive Manufacturing Platform for Value Chain Networks

    US20230080545A1