Continuous production big data analysis and regulation method for high-temperature stabilizer of cement slurry

Abstract

The present application relates to the intelligent control technology field of continuous production and chemical process of petroleum cementing additives, in particular to a continuous production big data analysis and regulation method for high-temperature stabilizer of cementing slurry, comprising: collecting process state time series data and synchronous control instruction data of a reaction kettle, a jacket heat exchange loop, a stirring driving part and an online viscosity detection part and aligning; constructing a normal working condition benchmark model based on polymerization kinetics constraints and jacket heat exchange heat balance constraints; loading disturbance parameter forward deduction to generate a set of theoretical abnormal states; determining a target abnormal state through similarity calculation of real residual vector and theoretical residual vector set; generating and outputting regulation instructions to a control execution end; the present application realizes stable closed-loop regulation and control of the continuous production process, and reduces the risk of single-point threshold misregulation.

Description

Technical Field

[0001] This invention relates to the field of continuous production of oil cementing additives and intelligent control technology of chemical processes, specifically a big data analysis and control method for continuous production of high-temperature stabilizers for cementing slurry. Background Technology

[0002] High-temperature stabilizers for cement slurry are an important component of the cementing chemical system for oil and gas wells. Their continuous production process involves multiple stages, such as exothermic polymerization reaction, heat transfer in jacket heat exchange, changes in feed rhythm, and gradual evolution of material viscosity. To ensure stable product quality and safe production process, it is usually necessary to monitor and control the continuous reaction vessel, heat exchange circuit, feed system, and stirring system in a coordinated manner. However, when identifying and regulating anomalies in the continuous production process of cement slurry high-temperature stabilizers, existing methods often rely on reaction temperature, valve opening, or individual measurement point thresholds as the main criteria for judgment. While single-point threshold judgment is relatively straightforward, it is difficult to simultaneously consider the coupling relationship between polymerization kinetics changes, heat transfer capacity changes, actuator response deviations, and material rheological states. Furthermore, some methods use historical experience curves or empirical parameters to determine production status, which is easily affected by batch differences in raw materials, ambient temperature fluctuations, sensor delays, and human-set factors, leading to inaccurate anomaly identification. This uncertainty not only affects the timeliness and targeting of continuous production control but may also cause misjudgments of heat transfer anomalies as triggering anomalies and actuator mismatches as process anomalies, adversely impacting the continuous and stable production of cement slurry high-temperature stabilizers. Therefore, there is an urgent need for a big data analysis and control method for the continuous production of cement slurry high-temperature stabilizers that comprehensively considers polymerization reaction kinetics and multivariate coupling relationships, and eliminates reliance on a single historical experience benchmark. This method aims to reduce the risk of misadjustment of single-point thresholds and achieve stable closed-loop control of the continuous production process. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a big data analysis and control method for the continuous production of high-temperature stabilizers for cementing slurry. Specifically, the technical solution of this invention includes: The system collects process status time-series data from the reactor, jacketed heat exchange circuit, stirring drive unit, and online viscosity detection unit, and collects control command data from the feed drive unit, initiator dropping unit, and control valve that are synchronized with the process status time-series data by timestamp. The process status time-series data and control command data are then aligned. Based on the mechanistic constraint rules including polymerization kinetics constraints and jacket heat transfer thermal balance constraints, a benchmark model is constructed to characterize the reference evolution relationship of the continuous production process of cement slurry high temperature stabilizer under normal operating conditions. The perturbation parameters corresponding to the preset perturbation operator set are loaded into the heat transfer parameters, reaction kinetic parameters and / or feed flow parameters of the benchmark model, and forward deduction is performed to generate a theoretical abnormal state set. Extract the multidimensional deviation values ​​between the actual collected values ​​of the process state time series data in the same physical dimension at the same time stamp and the reference state values ​​output by the benchmark model. Determine the actual residual vector based on the multidimensional deviation values, determine the theoretical residual vector set based on the theoretical abnormal state set and the benchmark model, and perform similarity calculation to determine the target abnormal state. Based on the target abnormal state, control instructions are generated and output to the control execution terminal for adjusting the feed flow rate, initiator dripping amount, heat exchange intensity and / or stirring intensity.

[0004] Preferably, the process status timing data is output and collected by the distributed control system. The process status timing data includes reaction temperature data, jacket heat exchange data, stirring load data, and material rheology data. The control command data includes feed drive data, initiator dripping data, and valve opening feedback data. The process status timing data and control command data are time-stamp aligned, missing data compensated, and outlier removed.

[0005] Preferably, a reference temperature evolution relationship is generated based on preset polymerization kinetic constraints; a reference heat transfer evolution relationship is generated based on preset jacket heat transfer thermal balance constraints; and a reference baseline trajectory is generated based on the reference temperature evolution relationship, the reference heat transfer evolution relationship, and the stirring load data. The reference baseline trajectory is used to characterize the reference temperature gradient trajectory and the reference load evolution trajectory of the target continuous production process under normal operating conditions.

[0006] Preferably, a preset set of perturbation operators is obtained, which includes a thermal resistance increase operator, a reaction rate decay operator, and a flow pulsation operator; the perturbation parameters corresponding to the set of perturbation operators are loaded into the baseline model respectively; and the model after loading the perturbation parameters is forward-deduced to generate a theoretical abnormal state set.

[0007] Preferably, the thermal resistance increase operator is used to characterize the state of decreased heat transfer capacity, the reaction rate decay operator is used to characterize the state of decreased activity, and the flow pulsation operator is used to characterize the flow pulsation state caused by wear of the feed pump and / or feed control valve.

[0008] Preferably, based on process state time series data and benchmark model, the actual residual vector is determined; based on theoretical abnormal state set and benchmark model, the theoretical residual vector set is determined; the actual residual vector and theoretical residual vector set are compared to generate similarity results; based on the similarity results, the target abnormal state is determined.

[0009] Preferably, when the similarity result is less than a preset confidence threshold, the current deviation represented by the actual residual vector is determined as a background disturbance state; when the similarity result is greater than or equal to the preset confidence threshold and less than the preset execution threshold, the current deviation represented by the actual residual vector is determined as a state to be verified, and the current control parameters remain unchanged; when the similarity result is greater than or equal to the preset execution threshold, the anomaly type represented by the theoretical residual vector corresponding to the maximum similarity value in the theoretical residual vector set is determined as the target anomaly state, wherein the preset execution threshold is greater than the preset confidence threshold.

[0010] Preferably, the similarity calculation uses the dynamic time warping algorithm or the Mahalanobis distance metric algorithm.

[0011] Preferably, a pre-established abnormal state-control parameter mapping strategy library is invoked based on the target abnormal state; control instructions are generated based on the abnormal state-control parameter mapping strategy library; and the control instructions are output to the control execution terminal to adjust the initiator dropping parameters, heat exchange control parameters, feed drive parameters and / or stirring drive parameters.

[0012] Preferably, the control execution end is a feed pump, an initiator dripping pump, a jacket control valve, and a stirring drive controller. The abnormal state-control parameter mapping strategy library stores the correspondence between abnormal types and the adjustment range of initiator dripping amount, jacket valve opening, feed pump speed, and stirring drive speed.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention enables the reaction state at a certain moment and the corresponding control action to be mapped to the same production segment through timestamp alignment, missing value compensation and outlier removal. It can effectively reduce the interference of sampling asynchrony, sensor instantaneous noise and abnormal feedback from the execution end on anomaly identification, and avoid ignoring the coupling relationship between heat exchange, mixing and rheological states by relying solely on a single temperature measurement point for judgment. 2. This invention constructs a benchmark model based on polymerization kinetic constraints and jacket heat exchange thermal balance constraints. The reference benchmark trajectory is generated by the reference temperature evolution relationship, the reference heat transfer evolution relationship, and the stirring load. It can use the temperature gradient, heat transfer response, and load evolution that should be presented under normal initiation, normal heat transfer, and normal mixing conditions as a reference, instead of simply relying on historical average curves or experience templates. This reduces the impact of raw material batch differences, ambient temperature fluctuations, and human-set factors on the determination of production status and improves the consistency of the mechanism of normal operating condition reference. 3. This invention sets up thermal impedance increase operator, reaction rate decay operator and flow pulsation operator, and loads the disturbance parameters into the heat transfer parameters, reaction kinetic parameters and feed flow parameters of the benchmark model for forward extrapolation. This can generate a theoretical abnormal state set in advance, such as heat transfer capacity reduction, activity reduction and feed flow pulsation. This makes the abnormal state no longer limited to the exceedance of a single measuring point, but forms an abnormal reference spectrum of the combined performance of temperature, heat transfer, stirring load and material rheology, thereby improving the pertinence and interpretability of abnormal source classification. 4. This invention extracts the difference between the actual collected process state and the same benchmark model to obtain the actual residual vector, and extracts the difference between the theoretical abnormal state and the same benchmark model to obtain the theoretical residual vector set. Then, it uses a dynamic time warping algorithm or a Mahalanobis distance metric algorithm to calculate similarity. It can complete the matching between the actual deviation and the theoretical anomaly map under a unified reference. It can adapt to time misalignment situations such as reaction thermal inertia, jacket response lag and online viscosity detection delay, and also consider the correlation of multiple variables such as temperature, heat exchange and load. This improves the anti-interference ability of anomaly identification and reduces the risk of misjudging heat exchange anomalies as triggering anomalies and misjudging execution mismatch as process anomalies. Attached Figure Description

[0014] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart illustrating the big data analysis and control method for continuous production of high-temperature stabilizer for cement slurry provided in an embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0016] A big data analysis and control method for continuous production of high-temperature stabilizers for cementing slurry includes the following steps: S1. Collect process status time-series data from the reactor, jacketed heat exchange circuit, stirring drive unit and online viscosity detection unit, and collect control command data from the feed drive unit, initiator dropping unit and control valve synchronized with the process status time-series data according to the timestamp; and perform alignment processing on the process status time-series data and control command data. S2. Based on the mechanistic constraint rules including polymerization kinetics constraints and jacket heat transfer thermal balance constraints, a benchmark model is constructed to characterize the reference evolution relationship of the continuous production process of cement slurry high temperature stabilizer under normal operating conditions. S3. Load the perturbation parameters corresponding to the preset perturbation operator set into the heat transfer parameters, reaction kinetic parameters and / or feed flow parameters of the benchmark model, and perform forward deduction to generate a theoretical abnormal state set; S4. Extract the multidimensional deviation values ​​between the actual collected values ​​of the process state time series data in the same physical dimension under the same time stamp and the reference state values ​​output by the benchmark model. Determine the actual residual vector based on the multidimensional deviation values, determine the theoretical residual vector set based on the theoretical abnormal state set and the benchmark model, and perform similarity calculation to determine the target abnormal state. S5. Generate control instructions based on the target abnormal state and output them to the control execution terminal for adjusting the feed flow rate, initiator dripping amount, heat exchange intensity and / or stirring intensity; This embodiment provides a full-process mechanism for big data analysis and control of continuous production of high-temperature stabilizers for cementing slurry, such as... Figure 1 As shown; specifically, this embodiment takes a continuous production line of a chemical production base in an oilfield as the main scenario. The production line is used to prepare a high-temperature stabilizer for cementing slurry that participates in polymerization with monomers containing sulfonic acid groups. The site includes a continuous reactor, a jacketed heat exchange circuit, a monomer feed pump, an initiator drip pump, a stirring motor, an online viscosity detection component, and a distributed control system. The technical difficulty of this production process lies in the fact that the polymerization reaction itself generates heat of reaction, the jacket circuit will continuously remove heat, and the viscosity of the material will gradually change with the polymerization process. Therefore, it is difficult to determine whether it is normal thermal inertia, batch difference of raw materials, or early abnormalities such as equipment scaling and decreased initiator activity based solely on a single temperature change trend. The system continuously collects process status time-series data from reactor temperature measuring points, jacket inlet and outlet temperature and flow measuring points, stirring motor current or torque measuring points, and online viscosity detection components. Simultaneously, it collects control command data such as monomer feeding frequency, initiator dropping rate, jacket valve opening, and stirring speed setting on the same time axis. Since the sampling cycles of the field control loop and detection components may be different, for example, the temperature is updated every few seconds while the viscosity detection is lagging, the system first aligns the data of each channel according to the timestamp, so that the reaction state at a certain moment and the control action that caused the state can correspond to the same production segment. The system constructs a benchmark model based on polymerization kinetics constraints and jacket heat exchange thermal balance constraints. The benchmark model does not directly use historical production data as an empirical template, but instead uses the temperature evolution, heat removal, and viscosity growth trends that materials should exhibit under normal initiation, normal heat transfer, and normal mixing conditions as a reference. For high-temperature stabilizers of cement slurry, the polymerization rate affects the intensity of heat release, the jacket heat exchange capacity affects the rate of temperature drop, and the stirring load and viscosity change together reflect the degree of polymerization of the material. Therefore, this benchmark model is used to describe the reference evolution relationship between reactor temperature gradient, jacket heat exchange capacity, stirring load and material rheological state under normal operating conditions. After the baseline model is formed, the system loads the perturbation parameters corresponding to the preset perturbation operators into the model. For example, the heat transfer parameters are changed to simulate the heat transfer decrease after scaling on the inner wall of the reactor, the reaction kinetic parameters are changed to simulate insufficient heat release after the initiator activity decreases, and the feed flow rate parameters are changed to simulate pulsating feed after the feed pump wears out. The system performs forward deduction on the loaded model to obtain multiple theoretical abnormal states. The perturbation parameters here are specifically manifested as numerical biases, proportional coefficients, or time-series fluctuation functions corresponding to each perturbation operator. They are used to perform algebraic superposition or product transformation on the heat transfer parameters, reaction kinetic parameters, and / or feed flow parameters in the baseline model to change the baseline settings of the model under normal operating conditions. The theoretical abnormal state here can be understood as a reference set of abnormal states pre-generated by mechanistic knowledge. It reflects the combined performance of a certain type of abnormality in terms of temperature, heat transfer, stirring load, and viscosity, rather than alarming based on a single temperature point. Furthermore, the system extracts the difference between the actual collected process state and the reference state of the benchmark model to obtain the actual residual vector. Specifically, the difference extraction involves algebraically subtracting the actual collected values ​​of the same physical dimension at the same time stamp from the reference state values, and then normalizing them according to the engineering range corresponding to each physical quantity, thus forming dimensionless residual values. The extraction of multidimensional deviation values ​​is performed online in real time in step S4, and the specific extraction calculation formula is as follows: in For multidimensional deviation values, These are the actual collected values ​​of the process state time series data. This represents the reference state value output by the baseline model at the corresponding timestamp. The engineering range quantization coefficients are used to correspond to the physical dimensions; and the differences between each theoretical abnormal state and the same benchmark model are extracted to obtain the theoretical residual vector set; for ease of understanding, a certain production segment can be simplified into three observation dimensions: temperature deviation, heat transfer deviation and stirring load deviation; If the actual residual vector exhibits a combination of slow temperature rise, lag in jacket heat exchange response relative to the reference heat exchange evolution, and stirring load fluctuation rate below the preset fluctuation threshold, and a theoretical residual vector also shows the same sequential relationship and directionality, then the system considers the actual deviation to be closer to an anomaly of decreased heat exchange capacity. If the actual residual mainly manifests as insufficient heat release, viscosity growth with a lag time greater than the preset delay threshold, and weakened jacket demand, then it is closer to an anomaly of decreased activity. The system determines the most likely target anomaly state through similarity calculation, then generates control commands and outputs them to the corresponding control execution end, such as adjusting the feed flow rate, initiator dropping amount, heat exchange intensity, or stirring intensity. The handling method under abnormal conditions is as follows: if any key measuring point, including the reactor temperature measuring point, jacket inlet and outlet temperature and flow measuring point, and stirring motor current or torque measuring point, is continuously lost, the online viscosity detection component is in the cleaning and maintenance state, or the deviation between the control command and the field valve position feedback exceeds the preset fault tolerance range, the system will not directly perform forced closed-loop adjustment, but will mark the current segment as a low confidence data segment, maintain the operation of the original basic control loop, and output a verification prompt to the operator station; The original basic control loop here refers to the single-loop PID closed-loop control logic preset at the bottom layer of the distributed control system. It relies only on the real-time feedback of a single sensor to maintain the basic operation of the equipment, and is not affected by the target abnormal state and control commands. If the actual residual does not match all theoretical residuals, the system will temporarily classify the deviation as an unclassified background disturbance or unknown operating condition to avoid misjudging the ambient temperature drift and short-term pressure fluctuation of cooling water as production abnormalities. When a batch of high-temperature resistant cement slurry stabilizer is continuously prepared on the above production line, the deviation of the reactor temperature from the target trajectory is less than the first preset deviation threshold. However, the heat exchange response under the same valve opening on the jacket side is slower, and the stirring load does not show synchronous abnormality. Instead of immediately increasing the initiator or changing the feed, the system compares the actual deviation with the theoretical residual of the pre-generated heat exchange capacity reduction. After confirming that its shape is closer to the heat transfer reduction caused by internal wall scaling, the system prioritizes outputting the jacket heat exchange adjustment command and limits the initiator dripping to exceed the preset change rate limit. When the deviation is greater than or equal to the first preset deviation threshold, the reactor temperature abnormality alarm is triggered and the emergency cooling program is started. The purpose of this step is to establish a normal production reference using physical mechanisms and then use disturbance deduction to form an abnormal reference, so that the control system can identify early abnormality types in the residual space, reduce the risk of misadjustment caused by relying solely on single-point thresholds, and thus achieve stable closed-loop control of the continuous production process of cement slurry high-temperature stabilizer.

[0017] Further, S1 includes the following sub-steps: S11. Acquire process state timing data output by the distributed control system; wherein, the process state timing data includes reaction temperature data, jacket heat exchange data, stirring load data, and material rheology data; S12. Acquire control command data synchronized with the process state timing data; wherein, the control command data includes feed drive data, initiator dripping data, and valve opening feedback data; S13. Perform timestamp alignment, missing value compensation, and outlier removal on the process state timing data and control command data.

[0018] This embodiment provides a production data acquisition and preprocessing mechanism. Specifically, if only the reactor temperature is used as input, in continuous polymerization production, temperature changes are easily mistaken for changes in reaction state, while ignoring the actual process reflected by heat removal on the jacket side, stirring load, and material viscosity. Therefore, this embodiment further introduces process state timing data and synchronous control command data, and completes alignment, compensation, and rejection before entering subsequent analysis. The process status time-series data output by the distributed control system includes reaction temperature data, jacket heat exchange data, stirring load data, and material rheology data; the reaction temperature data reflects the combined result of polymerization exothermic and heat dissipation; the jacket heat exchange data can be characterized by jacket medium flow rate, inlet and outlet temperature difference, valve feedback, etc., to indicate whether the heat has been effectively removed. The stirring load data can be reflected by changes in motor current, torque or power, which is used to indirectly characterize the material viscosity and mixing resistance; the material rheological data is output by online viscosity detection components or rheological detection units, which is used to explain the degree of polymerization and product state; the control command data includes feed drive data, initiator dripping data and valve opening feedback data, which are used to explain what control actions the production system has applied at that moment. During data processing, the system first organizes data from different sources using a unified timeline. For example, a production segment is simplified into three sampling times: the first time records temperature and pump frequency, the second time records jacket valve position and stirring load, and the third time records viscosity. If viscosity detection lags behind temperature detection, the system assigns it to the corresponding production segment according to the equipment's known response delay, rather than mechanically splicing all sampled values ​​directly according to arrival time. For short-term missing data, the system uses the trend of adjacent stable segments for compensation. For outliers that clearly do not conform to physical boundaries, such as a negative jacket flow rate at a certain moment or a sudden change in motor current outside the equipment's rated range, the system marks and removes them, excluding them from subsequent residual analysis. Furthermore, to avoid semantic confusion between control command data and valve opening feedback data, in this embodiment, valve opening feedback data is processed as execution result data collected synchronously with control commands. It is included in the control action-related data along with feed drive data and initiator dripping data. That is, feed drive data and initiator dripping data represent the action settings issued by the control system, while valve opening feedback data represents the actual response result of the execution end to the action settings. After aligning the timestamps, the system uses the feedback data to verify whether the control action has actually been implemented, thus ensuring that the subsequent residual analysis uses the executed control effects rather than control commands that were only issued but did not take effect. For example, when the set opening of the jacketed valve is increased but the feedback opening does not change synchronously, the system will interpret this segment as an execution-end abnormality rather than a heat exchange mechanism abnormality. Furthermore, missing data compensation is preferably performed only for short-term missing data whose duration does not exceed a preset compensable window. The preset compensable window can be set to the length of several sampling periods according to the sampling period of key variables and process inertia. Missing data exceeding this window is not included as a valid estimate in subsequent residual calculations. For outlier removal, in addition to screening using the equipment's rated range and physical boundaries, it is also preferable to combine the rate of change of adjacent time windows and the linkage relationship of multiple variables for verification. That is, when a single measurement point suddenly changes but other related measurement points do not show corresponding changes, the measurement point is preferentially regarded as an acquisition anomaly. Through the above processing, the data entering S2 to S4 can retain the temporal causal relationship of continuous production segments and avoid miswriting the execution end mismatch as process state deviation. Furthermore, to ensure consistency in terminology throughout the text, in this embodiment, the online viscosity detection component and the rheology detection unit refer to the same type of online detection hardware used to output material rheology data, and will not be distinguished as different devices in subsequent embodiments; material rheology data is used as a unified data name to encompass viscosity values ​​and their corresponding rheological characterization results; at the same time, control command data is a unified expression in the embodiments, and the control action related data referred to in the explanation of this embodiment is only used to illustrate that it simultaneously includes the joint verification meaning of issuing settings and execution feedback, and does not introduce new independent data categories; therefore, in subsequent embodiments, the jacket valve position feedback, valve feedback, and valve opening feedback data appearing in a fault-tolerant processing embodiment are all uniformly understood as valve opening feedback data; The handling procedures for abnormal situations are as follows: If the duration of the missing value exceeds the preset compensable window, the system will no longer replace the actual detection value with the estimated value, but will instead suspend the determination of the abnormality type for that segment; if the duration of the discrepancy between a critical control feedback value and the issued command exceeds the preset execution end response timeout threshold, for example, if the valve opening command has changed but the feedback has not changed, the system will mark the segment as an execution end feedback abnormality and will prioritize prompting on-site inspection of the valve or communication link; if multiple sensors drift simultaneously, the system will activate basic safety control logic to maintain the temperature and pressure within the permissible range of the process. In the aforementioned production line, the cooling water temperature drops at night, causing a slow change in the jacket heat exchange data. Simultaneously, the online viscosity detection component updates the data at fixed intervals longer than the reaction temperature sampling period. The system groups the reactor temperature, jacket inlet and outlet temperature difference, stirring motor current, viscosity value, monomer pump frequency, initiator dripping rate, and jacket valve feedback according to the same batch time axis. When a viscosity value changes more than the preset jump threshold due to sampling bubbles, the system discards it based on the fact that it does not change synchronously with the stirring load. If the jacket valve setting changes but the feedback does not change at the same time, the system marks this segment as an abnormal response at the execution end and does not continue to use it for heat exchange capacity determination. The purpose of this step is to provide consistent and reliable time-series input for subsequent benchmark modeling and residual comparison, thereby reducing the interference of sampling asynchrony, sensor transient noise and execution feedback anomalies on anomaly identification.

[0019] Further, S2 includes the following sub-steps: S21. Generate a reference temperature evolution relationship based on preset polymerization kinetic constraints; S22. Generate a reference heat transfer evolution relationship based on preset jacket heat transfer thermal balance constraints; S23. Generate a reference baseline trajectory based on the reference temperature evolution relationship, the reference heat transfer evolution relationship, and the stirring load data; wherein, the reference baseline trajectory is used to characterize the reference temperature gradient trajectory and the reference load evolution trajectory of the target continuous production process under normal operating conditions.

[0020] This embodiment provides a benchmark model construction mechanism. Specifically, after completing the multi-source data acquisition, if the historical average curve is directly used as the normal template, once the historical data is mixed with scaling, raw material batch deviation, or temporary operator adjustment where the thermal resistance increase ratio is lower than the alarm threshold, the obtained template itself will deviate from the real normal working conditions. Therefore, this embodiment uses polymerization kinetics and jacket heat exchange thermal balance as constraints to construct a reference benchmark trajectory that is closer to the normal production mechanism. The system first generates a reference temperature evolution relationship based on preset polymerization kinetic constraints. For the polymerization process of this type of high-temperature stabilizer, the initiator generates heat of reaction when initiating monomer polymerization. The rate of heat release is related to the initiation efficiency, monomer concentration, reaction temperature, and residence time. Under normal circumstances, the temperature will not jump randomly, but should show a gradual change in coordination with the feed and initiator addition. Based on this, the system generates a reference temperature gradient, indicating how the reactor temperature should rise, maintain, or fall under the current feed settings. The system then generates a reference heat transfer evolution relationship based on the jacket heat exchange heat balance constraint. The jacket loop removes or replenishes heat through cooling or heating media. The heat transfer effect is affected by the jacket flow rate, inlet and outlet temperature difference, valve opening, and vessel wall heat transfer status. Under normal conditions without significant scaling or valve failure, changes in the jacket valve opening should cause a predictable heat transfer response. Based on this, the system forms a reference heat transfer evolution relationship to distinguish between changes in reaction exothermic response and changes in heat transfer capacity. The system generates a reference baseline trajectory by combining stirring load data. The stirring load is not a separate alarm quantity, but an auxiliary process characteristic related to viscosity growth and mixing resistance. If the polymerization reaction proceeds normally, the material viscosity and stirring load will usually show a consistent trend with the degree of polymerization. If the temperature changes significantly but the stirring load does not change accordingly, it may indicate that the cause is more related to thermal factors than material polymerization. The reference baseline trajectory is essentially a set of reference evolution sequences on the same time axis, which includes temperature gradient evolution curves, heat transfer response evolution curves, and load evolution curves, rather than a single temperature threshold. Specifically, polymerization kinetic constraints and jacket heat transfer balance constraints together constitute the underlying dynamic control equations of the baseline model. The reference baseline trajectory is the time-series characteristic curve output by the baseline model in forward solving under normal process parameters and current feed conditions. The logical relationship between the two is that the baseline model is the internal physical mechanism rule, and the reference baseline trajectory is the external data performance of the baseline model in a specific production segment. Furthermore, to avoid the reference model, reference temperature evolution relationship, reference heat transfer evolution relationship, and reference reference trajectory being understood as multiple independent models throughout the text, this embodiment uniformly defines their hierarchical relationship: the reference model is the overall normal operating condition characterization model established by S2; the reference temperature evolution relationship and the reference heat transfer evolution relationship are the temperature constraint sub-relationship and heat transfer constraint sub-relationship in the reference model, respectively; and the reference reference trajectory is the time-series reference result output by the reference model under a given production segment and control settings. Accordingly, the reference state reference output reference trajectory appearing in subsequent embodiments will all be interpreted as the output of this reference model and will no longer be interpreted as a new independent model or a new independent trajectory library. Furthermore, considering that the feature description emphasizes the reference temperature gradient trajectory and the reference load evolution trajectory, and this embodiment also describes the reference heat transfer evolution relationship, this embodiment further defines the relationship between the two as follows: the reference heat transfer evolution relationship serves as one of the internal constraints for generating the reference baseline trajectory, and is used to jointly determine the formation of the reference temperature gradient trajectory and the reference load evolution trajectory. Its role can be reflected in the heat transfer deviation dimension during subsequent residual extraction, but it does not change the definition of the core output object of the reference baseline trajectory. In one fault-tolerant implementation: if the current production formula is switched, for example, the monomer ratio, target solid content, or initiator type is changed, the system does not follow the baseline trajectory of the previous formula, but instead loads the mechanism parameters corresponding to the new formula; if the stirring load measuring point is under maintenance or the motor inverter feedback is unavailable, the system can temporarily reduce the participation of this dimension in the generation of the baseline trajectory and indicate a decrease in the confidence of subsequent judgments; if the jacket heat exchange loop is in a manual bypass state, the system stops using this stage as a normal heat exchange reference update source; When producing a batch of conventional high-temperature stabilizers on the aforementioned production line, the system generates a reference relationship based on the current monomer feed rate and initiator dropping status to determine the reactor temperature should rise steadily and be maintained within a control range; simultaneously, it generates a reference relationship for heat removal based on the jacket water flow rate and valve opening; and combined with the trend of the stirring motor current slowly increasing with viscosity, it forms a reference baseline trajectory for this batch; when the ambient temperature decreases at night, causing a change in the jacket inlet water temperature, this change is explained separately through heat transfer constraints, rather than being simply attributed to an abnormal reaction. The purpose of this step is to create a normal operating condition reference with chemical mechanism constraints, so that subsequent residuals no longer only indicate differences from historical averages, but also differences from normal polymerization and normal heat exchange mechanisms.

[0021] Further, S3 includes the following sub-steps: S31. Obtain a preset set of perturbation operators; wherein, the set of perturbation operators includes a thermal resistance increase operator, a reaction rate decay operator, and a flow pulsation operator; S32. Load the perturbation parameters corresponding to the set of perturbation operators into the baseline model respectively; S33. Perform forward deduction on the model after loading the perturbation parameters to generate a theoretical abnormal state set.

[0022] This embodiment provides a perturbation operator injection and theoretical abnormal state generation mechanism. Specifically, when there is only a normal reference trajectory, although the system can know that the current production has deviated from the normal state, it is still difficult to explain the source of the deviation. Different anomalies may exhibit similar characteristics in certain temperature segments. For example, a decrease in heat transfer and an increase in reaction exothermics may both lead to an increase in reactor temperature. Therefore, this embodiment introduces a preset set of perturbation operators to transform the characteristics of historical typical anomaly mechanisms into a predictable anomaly reference state. The set of perturbation operators acquired by the system includes at least the thermal resistance increase operator, the reaction rate decay operator, and the flow pulsation operator. The thermal resistance increase operator acts on the heat transfer parameters of the baseline model to simulate the slowed heat removal caused by fouling of the vessel wall, deterioration of jacket heat transfer, or contamination of the heat transfer surface. The reaction rate decay operator acts on the reaction kinetic parameters to simulate insufficient polymerization exothermics caused by improper initiator storage, blockage of the dripping pipeline, or reduced effective activity of the initiator. The flow pulsation operator acts on the feed flow parameters to simulate periodic or irregular feed fluctuations caused by metering pump wear, valve core jamming, or air bubbles in the pipeline. To illustrate the data flow, the reference state output by the benchmark model within a short time window can be simplified into three items: temperature trend, heat transfer response, and load trend. After the thermal resistance is increased and disturbances are added, the theoretical abnormal state may manifest as the temperature lagging behind the reference trajectory, the heat transfer response weakening after the jacket valve changes, and the stirring load not necessarily changing synchronously. After loading the reaction rate decay perturbation, the theoretical abnormal state may manifest as insufficient heat release, slowed viscosity growth, and reduced jacket heat exchange demand; after loading the flow rate pulsation perturbation, the theoretical abnormal state may manifest as periodic fluctuations in temperature, stirring load, or viscosity that are related to feed fluctuations and whose amplitude is within a preset range; the system performs forward deduction on the loaded model to form a set of theoretical abnormal states. In one fault-tolerant implementation: if the equipment corresponding to a certain disturbance operator is not activated in the current process section, for example, if a certain standby feed pump is not in operation, the system will not generate a flow pulsation abnormal state related to that equipment; if multiple disturbances may occur simultaneously, the system can generate a combined disturbance state, but requires that their similarity reach a higher confidence level before outputting control commands; if the current process is in the start-up, shutdown or cleaning stage, the system will suspend the use of the disturbance map of continuous steady-state production to avoid misjudging normal transition processes as abnormalities; After the aforementioned production line had been running for several weeks, maintenance records showed that there was a slight risk of scaling in the reactor. The system injected increased thermal resistance disturbances on the normal baseline trajectory, forming a theoretical abnormal state of slowed heat exchange. At the same time, based on the risk of batch replacement of initiator, reaction rate decay disturbances were injected, forming a theoretical abnormal state of insufficient polymerization exothermics. Furthermore, based on the cumulative running time of the feed pump, flow pulsation disturbances were injected, forming a theoretical abnormal state of unstable feed. The purpose of this step is to pre-build an interpretable database of abnormal states so that subsequent deviations can not only be detected but also attributed to specific process or equipment causes, thereby supporting targeted closed-loop control.

[0023] Furthermore, the thermal resistance increase operator is used to characterize the state of decreased heat transfer capacity, the reaction rate decay operator is used to characterize the state of decreased activity, and the flow pulsation operator is used to characterize the flow pulsation state caused by wear of the feed pump and / or feed control valve.

[0024] This embodiment provides a physical meaning limitation mechanism for perturbation operators. Specifically, in the previous embodiment, although a theoretical abnormal state has been generated through perturbation operators, if there is no clear correspondence between the perturbation operators and the actual equipment mechanism, the subsequent control terminal may not be able to determine whether to adjust the jacket, initiator, or feed pump. Therefore, this embodiment further limits the specific industrial state represented by each perturbation operator. The thermal resistance increase operator is used to characterize the state of decreased heat transfer capacity. This state usually corresponds to polymer deposition on the inner wall of the reactor, contamination of the jacket heat exchange surface, and local blockage of the cooling medium channel. In terms of physical performance, even if the jacket valve opening is increased, the heat removed per unit time is less than the rated heat flux preset by the baseline model under normal conditions. The reactor temperature is more likely to show a slow drop or local heat accumulation. The system classifies such changes as heat exchange side anomalies, rather than immediately identifying them as excessively fast reactions. The reaction rate decay operator is used to characterize the state of decreased initiation activity. This state usually corresponds to situations such as improper initiator storage temperature, decay of active ingredients, micro-blockage of the dripping pipeline, or insufficient delivery by the dripping pump. Physically, the polymerization process slows down after the monomer enters the reactor, the exothermic reaction weakens, and the viscosity increase of the material may also be slower than the reference state. The system classifies such changes as reaction kinetic anomalies, thereby avoiding masking the initiation problem by simply heating. The flow pulsation operator is used to characterize the flow pulsation state caused by wear of the feed pump and / or feed control valve. This state usually corresponds to the aging of the metering pump diaphragm, poor sealing of the check valve, and stuck valve core of the feed control valve. In terms of physical performance, the feed rate fluctuates around the set value, resulting in slight fluctuations in reactant concentration and residence time, which in turn causes fluctuations in temperature, stirring load or viscosity related to the feed rhythm. In one fault-tolerant implementation: if the increase in thermal resistance and the decrease in reaction rate are similar, the system does not directly execute two control actions in opposite directions, but instead performs secondary confirmation by combining feedback from the jacket valve, cooling water temperature difference, and viscosity growth trend; if flow pulsation only appears in sensor readings and is not reflected in pump frequency, valve position feedback, or material status, the system prioritizes marking it as a flow meter measurement anomaly; if on-site personnel confirm that the abnormal characteristics have disappeared after equipment maintenance, the system updates the label for that period, but does not use the abnormal segment before maintenance as normal baseline data; In the aforementioned production line, the system detected a delay between temperature deviation and changes in jacket valve opening, but the initiator dripping and viscosity increase maintained a normal trend. Therefore, it tended to classify this as a state of decreased heat transfer capacity characterized by increased thermal resistance. If the temperature release was insufficient and the viscosity increase was slower in another batch, the system tended to classify it as a state of decreased initiation activity characterized by decreased reaction rate. If the temperature and load showed slight fluctuations related to the feed pump stroke, it was classified as a state of flow pulsation. The purpose of this mechanism is to establish a traceable correspondence between abstract disturbances and real equipment failures or process anomalies, so that the anomaly identification results can directly serve subsequent execution-end control and on-site maintenance judgment.

[0025] Further, S4 includes the following sub-steps: S41. Determine the actual residual vector based on the process state time series data and the benchmark model; S42. Determine the theoretical residual vector set based on the theoretical abnormal state set and the benchmark model; S43. Perform similarity calculation between the actual residual vector and the theoretical residual vector set to generate similarity results; S44. Determine the target abnormal state based on the similarity results.

[0026] This embodiment provides a dual-track residual comparison mechanism. Specifically, after generating the normal baseline and theoretical abnormal state, if the real data is directly compared with each abnormal simulation curve, it may still be affected by the difference in raw material batches, changes in ambient temperature, or zero-point drift of the sensor. Therefore, this embodiment first extracts the differences between the real data and the theoretical abnormal state relative to the same baseline model, and then performs similarity calculation at the residual level. The actual residual vector is formed by the difference between the process state time series data and the reference output of the benchmark model, which contains the deviation of the actual production site from the normal mechanism; the theoretical residual vector set is formed by the difference between each theoretical abnormal state and the benchmark model, which represents the deviation pattern that a certain type of abnormality should present when no background noise from the site is mixed in; both use the same benchmark as a reference, which can reduce the overall offset caused by changes in production settings. Specifically, the residual characteristics within a certain time window can be represented by a three-dimensional feature vector, with the three dimensions corresponding to the direction of temperature deviation, the direction of heat transfer response deviation, and the direction of stirring load deviation, respectively. For example, the actual residual may be characterized by a higher temperature, a weaker heat transfer response, and a basically normal load. In the theoretical residual set, the type of increased thermal resistance also exhibits a similar pattern, while the type of decreased reaction rate is more likely to be characterized by a lower temperature and a slower load growth. The system calculates the similarity between the actual residual and each theoretical residual to obtain matching results for different anomaly types, and selects the anomaly type that meets the judgment criteria as the target anomaly state. In one fault-tolerant implementation: if the actual residual amplitude is lower than the sensor noise level, the system does not classify it as an anomaly, but marks it as negligible fluctuation; if multiple theoretical residuals and actual residuals have close similarity, the system does not immediately output strong control commands, but enters the verification process, continues to observe the next time window or requests additional auxiliary data; if the actual residuals lack a stable trend in various dimensions, such as irregular jumps in temperature, viscosity and load, the system prioritizes checking the data acquisition link and power supply interference, rather than identifying it as a process anomaly. In the aforementioned production line, the system found that the actual temperature curve shifted slightly upward relative to the baseline trajectory, but the stirring load did not increase rapidly accordingly; at the same time, the heat exchange response was not as good as the reference state after the jacket valve opening was increased; after the system extracted the actual residual, it compared it with the three types of theoretical residuals: increased thermal resistance, decreased reaction rate, and flow pulsation. It found that it was closer to the type of residuals with increased thermal resistance in terms of trend and sequence, so the decrease in heat exchange capacity was selected as a candidate target abnormal state. The purpose of this step is to transform the complex identification of continuous production anomalies into a comparison problem between actual deviations and theoretical fault maps, thereby improving the interpretability and anti-interference capability of early anomaly identification.

[0027] Further, S44 includes: when the similarity result is less than a preset confidence threshold, determining the current deviation represented by the actual residual vector as a background disturbance state; when the similarity result is greater than or equal to the preset confidence threshold and less than a preset execution threshold, determining the current deviation represented by the actual residual vector as a state to be verified, and keeping the current control parameters unchanged; when the similarity result is greater than or equal to the preset execution threshold, determining the anomaly type represented by the theoretical residual vector corresponding to the maximum similarity value in the theoretical residual vector set as the target anomaly state; wherein, the preset execution threshold is greater than the preset confidence threshold.

[0028] This embodiment provides a hierarchical judgment mechanism. Specifically, in the previous embodiment, the system was able to obtain similarity results. However, if only a single judgment threshold is set, two types of problems may easily occur: if the judgment threshold is too low, environmental fluctuations will be treated as an anomaly and adjustments will be made frequently; if the judgment threshold is too high, early faults may be missed. Therefore, this embodiment sets a preset confidence threshold and a preset execution threshold, and makes the execution threshold higher than the confidence threshold to form three levels: background disturbance, state to be verified, and target abnormal state. When the similarity result is less than the preset confidence threshold, it indicates that the actual residual lacks a stable correspondence with any known theoretical anomaly spectrum. At this time, the current deviation is more likely to be a short-term change in cooling water inlet temperature, ambient temperature drift, small sensor noise, or normal fluctuations in raw material batches. The system determines it as a background disturbance state and does not trigger process adjustment. When the similarity result is greater than or equal to the preset confidence threshold and less than the preset execution threshold, it indicates that the actual residual has shown some characteristics of a certain type of anomaly, but the evidence is not enough to support an immediate change in the control execution end. At this time, the system determines the current deviation as a state to be verified and keeps the current control parameters unchanged. Keeping them unchanged does not mean not processing them, but rather avoiding changes to the initiator, feed, or jacket when the evidence is insufficient, and preventing excessive intervention in the continuous polymerization system. When the similarity result is greater than or equal to the preset execution threshold, it means that the actual residual and a certain theoretical residual have reached the executable level in terms of deviation direction, change order and persistence. The system determines the anomaly type represented by the theoretical residual corresponding to the maximum similarity value as the target anomaly state. The above three judgment states correspond to the production conditions that have not formed anomaly features, have preliminary anomaly features to be verified, and have clear anomaly features and allow execution of control and regulation. Furthermore, to ensure consistency between this implementation method and the similarity calculation method used, the similarity result is uniformly defined as a criterion that a larger value indicates a higher degree of matching. For algorithms that directly output similarity scores, these scores can be directly used as the similarity result. For algorithms whose original output is a distance value, the distance value is first converted into a monotonically decreasing similarity score before being compared with a preset confidence threshold and a preset execution threshold. Thus, the threshold judgment in S44 is always based on similarity results in the same direction, and there is no conflict between the judgment criteria that a larger value indicates a higher degree of matching and a smaller value indicates a higher degree of matching due to different underlying measurement methods. Furthermore, the preset confidence threshold and preset execution threshold are preferably obtained by joint calibration of historical normal segments, historically confirmed abnormal segments, and simulated abnormal segments. Specifically, the following method can be adopted: first, the upper bound of the fluctuation of similarity results is calculated in historical normal segments, and then the lower bound of the similarity that can be stably distinguished is calculated in confirmed abnormal segments. A safety interval is left between the two to determine the confidence threshold and execution threshold respectively. For production sections with different formulas, different production loads, or different seasonal cooling conditions, corresponding thresholds can be maintained separately to reduce misjudgments caused by switching operating conditions. In one fault-tolerant implementation: if the similarity result fluctuates around the confidence threshold, the system does not repeatedly switch states, but requires that the same trend be maintained for multiple consecutive time windows before upgrading; if the similarity result reaches the execution threshold but the key execution end reports an anomaly, such as the jacketed valve reporting distortion, the system only issues an alarm or a limiting command, and does not execute closed-loop adjustment exceeding the preset single adjustment step size limit; if the similarity of different anomaly types all reach the execution threshold and the absolute value of the similarity difference is less than the preset judgment margin, the system enters the combined anomaly review process, and prioritizes the adjustment action with lower safety risk and stronger reversibility; In the aforementioned production line, within a certain time window, the actual residual and the theoretical residual of increased thermal resistance only show slight similarity. The system treats this as background disturbance and does not adjust the jacket. In several consecutive time windows, the weak jacket response gradually stabilizes but has not yet met the execution requirements. The system enters a verification state and maintains the valve and pump speed settings. Only when this state continues to strengthen and meets the execution conditions does the system determine the decrease in heat exchange capacity as the target abnormal state. If the underlying calculation uses a distance-based metric, the system will first convert the distance to a similarity score and then make a graded judgment based on the above threshold rules. The purpose of this mechanism is to suppress misadjustments and frequent adjustments through graded judgment, so that continuous production control can both identify early anomalies and maintain process stability when evidence is insufficient.

[0029] Furthermore, similarity calculation employs either the dynamic time warping algorithm or the Mahalanobis distance metric algorithm.

[0030] This embodiment provides a similarity measurement mechanism. Specifically, in continuous polymerization production, actual anomalies and theoretical anomalies are not necessarily completely synchronized in time. For example, the reactor has thermal inertia, the jacket heat exchange response has lag, and online viscosity detection may also be delayed due to sampling and measurement processes. If comparisons are made point by point at the same time point, it is easy to misjudge the time scale misalignment caused by the same type of anomaly as feature mismatch. Therefore, this embodiment can use a dynamic time warping algorithm or a Mahalanobis distance metric algorithm for similarity calculation. Dynamic time warping is suitable for scenarios where anomalies have similar forms but slightly different occurrence rates. Its function is to allow two residual sequences to be aligned to a limited extent on the time axis, so that anomalies with the same form but different lags can still be identified. For example, theoretically, an increase in thermal resistance will show a weakening of heat transfer response more quickly after the jacket valve opening changes, but in the field, due to the large thermal inertia of the vessel, the temperature deviation may appear slightly later. Dynamic time warping can align such differences in timing within the allowable range. Mahalanobis distance metric algorithm is suitable for scenarios where multiple variables are correlated; reaction temperature, jacket heat exchange, and stirring load are not independent of each other, and they have a certain synergistic relationship in normal production; Mahalanobis distance metric can consider the fluctuation scale and correlation of different dimensions, and avoid excessive influence of a noisy measurement point on the overall similarity; to illustrate its data flow, the actual residual can be regarded as a three-dimensional state point composed of temperature, heat exchange, and load, and each type of theoretical residual can be regarded as the corresponding anomaly reference area; the system judges which type of reference area the actual state point is closer to, rather than just looking at the size of a single temperature difference; Furthermore, in this embodiment, the use of dynamic time warping algorithm or Mahalanobis distance metric algorithm for similarity calculation means that: dynamic time warping algorithm can directly output similarity evaluation value corresponding to the path matching quality, while Mahalanobis distance metric algorithm first outputs distance value, and then converts it into similarity result compatible with S44 before participating in threshold determination; Preferably, if the Mahalanobis distance metric algorithm is used, the distance value can be converted into an interval similarity score according to a preset monotonic mapping method, so that the result satisfies the unified judgment direction that the smaller the distance, the greater the similarity result; this not only retains the Mahalanobis distance's ability to characterize the correlation of multiple variables, but also avoids the problem of inconsistency with the threshold comparison direction; Furthermore, when using the dynamic time warping algorithm, the system preferably sets a finite warping window for the actual residual sequence and each theoretical residual sequence to avoid non-causal alignment that spans too long a time. The time length of the finite warping window is set based on the maximum physical response lag time of the jacket heat exchange in the production system and the mass transfer delay time of material mixing. When using the Mahalanobis distance metric algorithm, the system preferably estimates the covariance structure of each dimension based on normal operating condition samples or confirmed abnormal samples. When the sample is insufficient or the covariance matrix is ​​ill-conditioned, diagonal approximation or regularization is used to prevent the distance from being abnormally amplified due to the excessively small variance of a certain dimension. Through the above constraints, the similarity calculation can retain the real characteristics of time lag and multivariate coupling in continuous chemical processes, while maintaining the stability in engineering implementation. Furthermore, to ensure consistent terminology throughout the text, in this embodiment, the Dynamic Time Warping Algorithm and its abbreviation (in English) refer to the same time series warping method if they appear in the engineering implementation description; Mahalanobis distance metric algorithm, Mahalanobis distance metric method, and Mahalanobis distance all refer to the same multivariate distance calculation method based on covariance structure; similarity calculation represents the unified matching process from the actual residual vector to the theoretical residual vector set, while the similarity result represents the final judgment quantity obtained directly from the Dynamic Time Warping Algorithm or after Mahalanobis distance conversion; in subsequent embodiments, if matching results, matching scores, or similarity scores appear, they are all used as equivalent expressions of similarity results, and uniformly satisfy the judgment criterion that the larger the value, the higher the degree of matching; In one fault-tolerant implementation: if the detected anomaly is mainly manifested as time lag and similar curve shape, the system prioritizes dynamic time warping; if the detected anomaly is mainly manifested as common offset of multiple variables and good time synchronization, the system prioritizes Mahalanobis distance measurement; if the data window is too short to perform reliable time warping, the system reverts to conservative judgment and extends the observation window; if the measurement quality of a certain dimension deteriorates, the system reduces its impact in the distance measurement or suspends the participation of that dimension. In the aforementioned production line, the lower cooling water temperature in winter causes the jacket response speed to differ slightly from the theoretical state. The system uses dynamic time warping to compare the actual residual with the thermal resistance to increase the theoretical residual, avoiding missed judgments due to a delay of several minutes. In another time period, temperature, viscosity, and stirring load all have coordinated shifts. The system uses Mahalanobis distance measurement to determine whether closer proximity causes a decrease in activity or a fluctuation in feed flow. At this time, the distance is first calculated based on multivariate correlation, and then converted into a unified similarity result for S44 to perform graded judgment. The purpose of this mechanism is to adapt similarity calculation to the thermal inertia, detection delay, and multivariate correlation in continuous chemical processes, thereby improving the stability of residual matching.

[0031] Further, S5 includes the following sub-steps: S51. Call the pre-established abnormal state-control parameter mapping strategy library according to the target abnormal state; S52. Generate control instructions based on the abnormal state-control parameter mapping strategy library; S53. Output the control instructions to the control execution terminal to adjust the initiator dropping parameters, heat exchange control parameters, feed drive parameters and / or stirring drive parameters.

[0032] This embodiment provides a mapping mechanism from abnormal states to control commands. Specifically, after determining the target abnormal state, if the system directly modifies the control parameters without constraints, it may cause excessive intervention in the reaction system. For example, misinterpreting a decrease in heat exchange capacity as insufficient initiation and increasing the initiator may exacerbate the subsequent exothermic risk. Therefore, this embodiment generates constrained control commands through a pre-established abnormal state-control parameter mapping strategy library. The abnormal state-control parameter mapping strategy library stores the correspondence between various target abnormal states and executable control parameters. For the state of reduced heat exchange capacity, the control command prioritizes the jacket control parameters, such as adjusting the jacket valve opening, cooling medium flow rate or heat exchange feedforward compensation, while limiting the feed load to reduce instantaneous heat release. For initiator activity decrease, the control command prioritizes the initiator dropping parameters, which can fine-tune the dropping rate or trigger initiator pipeline checks within the process allowable range; for flow pulsation, the control command prioritizes the feed drive parameters, which can adjust the feed pump operating frequency, switch to the standby pump, or limit rapid valve action; for abnormal viscosity increase leading to insufficient mixing, the stirring drive parameters can be adjusted to improve mixing uniformity. The essence of the abnormal state-control parameter mapping strategy library is a data table that maps the abnormal state type to the control execution action. If the target abnormal state is a decrease in heat exchange capacity, the system will not call the main action of increasing initiator dripping, but will call the action group that enhances heat exchange and limits load fluctuations. If the target abnormal state is a decrease in initiator activity, the system will prioritize calling the action group that compensates for initiator dripping or checks pipelines. If the target abnormal state is a pulsation in feed flow, the system will prioritize calling the action group that smooths the feed pump speed or checks valves. Each action group includes the allowable adjustment range, adjustment rate, and post-execution observation window to avoid a large change in continuous reaction conditions at once. The handling procedures for abnormal situations are as follows: If the target abnormal state is at the pending verification level, the system will not call the key control action, but will only record the trend and maintain the current control parameters; if there is no corresponding abnormal type in the policy library, the system will not generate an automatic execution instruction, but will generate an operator confirmation prompt; if the execution end is in manual mode or maintenance mode, the system will only output suggested values ​​and alarm information, and will not forcibly override the field control; if a safety interlock is triggered, the policy library control priority is lower than that of the safety interlock, and the basic safety system will take over. In the aforementioned production line, after the system confirms a decrease in heat exchange capacity, it calls the strategy library to generate instructions for fine-tuning the jacket valve opening and compensating for cooling water flow, and limits the changes in monomer feed to a stable range; when another batch is confirmed to have decreased initiator activity, the system calls the initiator dripping compensation strategy, and prompts the system to check the initiator storage tank temperature and dripping pipeline; when feed pulsation is confirmed, the system outputs instructions for smooth control of feed pump speed or switching to standby pump. The purpose of this step is to transform the anomaly identification results into executable control actions constrained by the process, thereby avoiding the disconnect between anomaly identification and on-site execution, and reducing the impact of misoperation on polymerization stability.

[0033] Furthermore, the control execution end consists of a feed pump, an initiator dripping pump, a jacket control valve, and a stirring drive controller. The abnormal state-regulation parameter mapping strategy library stores the correspondence between abnormal types and the adjustment ranges of initiator dripping amount, jacket valve opening, feed pump speed, and stirring drive speed.

[0034] This embodiment provides a control execution end and adjustment range mapping mechanism. Specifically, in the previous embodiment, the strategy library was able to map abnormal states to control actions. However, if the execution end and adjustment range are not clearly defined, the system may adjust the wrong object or adjust excessively. Therefore, this embodiment limits the control execution end to the feed pump, initiator dripping pump, jacket control valve and stirring drive controller, and stores the correspondence between abnormal types and adjustment ranges of each execution end in the strategy library. The feed pump is used to regulate the flow rate of monomers or raw materials into the reactor, and its adjustment directly affects the reactant concentration and residence time; the initiator drop pump is used to regulate the amount of initiator added, and its adjustment directly affects the polymerization initiation strength and the exothermic process; the jacket control valve is used to regulate the flow rate or heat transfer intensity of the heat exchange medium, and its adjustment affects the heat removal rate; the stirring drive controller is used to regulate the stirring speed or torque output, and its adjustment affects the uniformity of material mixing and the local heat and mass transfer state. The strategy library does not store unlimited control actions, but rather the correspondence between anomaly types and adjustment ranges; for example, a decrease in heat exchange capacity corresponds to the adjustment range of the jacket valve opening and, if necessary, the feed pump speed limit; a decrease in initiator activity corresponds to the fine adjustment range of the initiator dropping amount, and can be combined with stirring intensity checks to ensure uniform mixing; flow pulsation corresponds to the smooth adjustment range of the feed pump speed or valve action limits; an abnormal increase in material viscosity corresponds to the adjustment range of the stirring drive speed, while avoiding shear or temperature rise problems caused by excessive stirring; the above ranges can be pre-configured by process engineers according to the equipment's rated capacity, formula window, and safety boundaries, and are limited by the system during execution; In one fault-tolerant implementation: if a certain execution end reaches the mechanical or process limit, the system will not continue to add commands in the same direction, but will trigger load reduction, switch to backup equipment, or manually confirm the process; if the feedback from the initiator drip pump is inconsistent with the setting, the system will not compensate by continuing to increase the setting value, but will prompt to check the pump head, pipeline and liquid storage status; if the jacket control valve opening has reached the limit of the preset opening threshold and the temperature still deviates, the system will prioritize limiting the feed load and prompt the heat exchange system for maintenance; if the stirring drive current is close to the protection threshold, the system will no longer increase the stirring speed to prevent motor overload; In the aforementioned production line, after the system confirms an increase in thermal resistance, the strategy library issues a combination of instructions: the jacket control valve increases its opening step by a preset increment, the cooling water side maintains a stable flow rate, and the feed pump speed is smoothly limited. After the system confirms a decrease in initiator activity, the strategy library issues a fine-tuning instruction for the initiator dripping pump and sets an observation window. If the viscosity increase does not recover, it prompts the system to replace the initiator batch. After the system confirms flow pulsation, the strategy library issues instructions for smooth adjustment of the feed pump speed or switching to a standby pump, rather than changing the jacket valve as the main control action. The purpose of this mechanism is to establish a clear mapping between anomaly types, execution objects, and adjustment ranges, so that automatic control actions are kept within the boundaries of equipment capacity and process safety, thereby achieving controllability, traceability, and verifiability of continuous production processes.

[0035] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A big data analysis and control method for continuous production of high-temperature stabilizer for cementing slurry, characterized in that, Includes the following steps: S1. Collect process status timing data from the reactor, jacketed heat exchange circuit, stirring drive component and online viscosity detection component, and collect control command data from the feed drive component, initiator dropping component and control valve that are synchronized with the process status timing data according to the timestamp; and perform alignment processing on the process status timing data and control command data. S2. Based on the mechanism constraint rules including polymerization kinetics constraints and jacket heat exchange thermal balance constraints, a benchmark model is constructed to characterize the reference evolution relationship of the continuous production process of the cement slurry high temperature stabilizer under normal operating conditions. S3. Load the perturbation parameters corresponding to the preset perturbation operator set into the heat transfer parameters, reaction kinetic parameters and / or feed flow parameters of the benchmark model, and perform forward deduction to generate a theoretical abnormal state set; S4. Extract the multidimensional deviation value between the actual collected values ​​of the process state time series data in the same physical dimension under the same timestamp and the reference state value output by the benchmark model. Determine the actual residual vector based on the multidimensional deviation value. Determine the theoretical residual vector set based on the theoretical abnormal state set and the benchmark model. Perform similarity calculation to determine the target abnormal state. S5. Generate control instructions based on the target abnormal state and output them to the control execution terminal for adjusting the feed flow rate, initiator dripping amount, heat exchange intensity and / or stirring intensity.

2. The big data analysis and control method for continuous production of high-temperature stabilizer for cement slurry according to claim 1, characterized in that, S1 includes the following sub-steps: S11. Collect process status time-series data output by the distributed control system; wherein, the process status time-series data includes reaction temperature data, jacket heat exchange data, stirring load data, and material rheology data; S12. Collect control command data synchronized with the process status timing data; wherein, the control command data includes feed drive data, initiator dripping data and valve opening feedback data; S13. Perform timestamp alignment, missing value compensation, and outlier removal on the process status timing data and control command data.

3. The method for continuous production big data analysis and control of high-temperature stabilizer for cement slurry according to claim 2, characterized in that, S2 includes the following sub-steps: S21. Generate a reference temperature evolution relationship based on preset polymerization kinetic constraints; S22. Generate a reference heat transfer evolution relationship based on the preset jacket heat transfer thermal balance constraint; S23. Generate a reference baseline trajectory based on the reference temperature evolution relationship, the reference heat transfer evolution relationship, and the stirring load data; wherein, the reference baseline trajectory is used to characterize the reference temperature gradient trajectory and the reference load evolution trajectory of the target continuous production process under normal operating conditions.

4. The method for big data analysis and control of continuous production of high-temperature stabilizer for cement slurry according to claim 1, characterized in that, S3 includes the following sub-steps: S31. Obtain a preset set of perturbation operators; wherein, the set of perturbation operators includes a thermal resistance increase operator, a reaction rate decay operator, and a flow pulsation operator; S32. Load the perturbation parameters corresponding to the perturbation operator set into the baseline model respectively; S33. Perform forward deduction on the model after loading the perturbation parameters to generate a theoretical set of anomalous states.

5. The method for continuous production big data analysis and control of high-temperature stabilizer for cement slurry according to claim 4, characterized in that, The thermal resistance increase operator is used to characterize the state of decreased heat exchange capacity, the reaction rate decay operator is used to characterize the state of decreased activity, and the flow pulsation operator is used to characterize the flow pulsation state caused by wear of the feed pump and / or feed control valve.

6. The method for big data analysis and control of continuous production of high-temperature stabilizer for cement slurry according to claim 1, characterized in that, S4 includes the following sub-steps: S41. Based on the process state time series data and the benchmark model, determine the actual residual vector; S42. Based on the theoretical abnormal state set and the benchmark model, determine the theoretical residual vector set; S43. Perform similarity calculation between the actual residual vector and the theoretical residual vector set to generate similarity results; S44. Based on the similarity results, determine the target abnormal state.

7. The method for continuous production big data analysis and control of high-temperature stabilizer for cement slurry according to claim 6, characterized in that, S44 includes: When the similarity result is less than the preset confidence threshold, the current deviation represented by the actual residual vector is determined as a background disturbance state; when the similarity result is greater than or equal to the preset confidence threshold and less than the preset execution threshold, the current deviation represented by the actual residual vector is determined as a state to be verified, and the current control parameters remain unchanged. When the similarity result is greater than or equal to the preset execution threshold, the anomaly type represented by the theoretical residual vector corresponding to the maximum similarity value in the theoretical residual vector set is determined as the target anomaly state; wherein, the preset execution threshold is greater than the preset confidence threshold.

8. The method for big data analysis and control of continuous production of high-temperature stabilizer for cement slurry according to claim 6, characterized in that, The similarity calculation employs either the dynamic time warping algorithm or the Mahalanobis distance metric algorithm.

9. The method for big data analysis and control of continuous production of high-temperature stabilizer for cement slurry according to claim 1, characterized in that, S5 includes the following sub-steps: S51. Invoke the pre-established abnormal state-regulation parameter mapping strategy library according to the target abnormal state; S52. Generate control instructions based on the abnormal state-control parameter mapping strategy library; S53. Output the control command to the control execution terminal to adjust the initiator dropping parameters, heat exchange control parameters, feed drive parameters and / or stirring drive parameters.

10. The method for continuous production big data analysis and control of high-temperature stabilizer for cement slurry according to claim 9, characterized in that, The control execution end includes a feed pump, an initiator dripping pump, a jacket control valve, and a stirring drive controller. The abnormal state-regulation parameter mapping strategy library stores the correspondence between abnormal types and the adjustment ranges of initiator dripping amount, jacket valve opening, feed pump speed, and stirring drive speed.

Images