Method for predicting and grading early warning of safety boundary of runaway of ultrahigh pressure polymerization

Abstract

The method for predicting and grading early warning of runaway safety boundary of ultrahigh pressure polymerization belongs to the technical field of chemical process safety and polymerization reaction process control. The method collects runaway data under the conditions of initial temperature, pressure, ethylene density, oxygen impurities, initiator and comonomer, obtains temperature and pressure peak value and rising time; constructs equivalent temperature rise rate, equivalent pressure rise rate, temperature and pressure coupling slope, temperature and pressure amplification coefficient and coupling strength and other characteristics, represents the coupling evolution of temperature instability to pressure instability; fusion static working condition, induction factor and temperature and pressure dynamic characteristics, establish mechanism perception safety constraint learning model, joint prediction of temperature boundary, pressure boundary, runaway severity and risk grade. By introducing the underestimation penalty of runaway consequences and the consistency constraint of temperature and pressure coupling, the risk of missing dangerous working condition is reduced, and the problems such as the lack of adaptability of fixed threshold early warning to complex working conditions in the prior art are overcome, and the accuracy and reliability of the runaway boundary prediction and grading early warning are improved.

Description

Technical Field

[0001] This invention relates to a technology for predicting the runaway boundary of ultra-high pressure ethylene polymerization reaction, specifically a technology for predicting and classifying the runaway boundary of ultra-high pressure ethylene polymerization reaction based on temperature-pressure coupling characterization and safety constraint learning, which realizes safety early warning of ultra-high pressure ethylene polymerization reaction and belongs to the field of chemical process safety and control technology. Background Technology

[0002] Vinyl polymers such as LDPE and EVA are commonly used and important plastic materials, widely applied in various fields due to their excellent physical and chemical properties. The industrial method for large-scale production of these polymers is high-pressure bulk polymerization. Ultra-high pressure polymerization processes involve extremely harsh conditions, making monomers prone to runaway and subsequent decomposition and explosion. Material leaks or releases of process media can easily cause secondary combustion and explosion accidents.

[0003] Existing ultra-high pressure (100-300MPa) ethylene polymerization units typically rely on temperature and pressure sensors for online monitoring. When the monitored values ​​exceed preset absolute thresholds (e.g., 300℃, 350MPa), a programmable logic controller (PLC) triggers emergency venting or shutdown. This method can only respond when the temperature or pressure is already close to a dangerous state, making it difficult to identify the temperature-pressure coupling acceleration phase in advance. It suffers from problems such as delayed warnings, poor adaptability to complex operating conditions, and a high probability of misjudgment. Especially under different runaway paths such as oxygen impurity induction, initiator induction, and comonomer induction, the variation patterns of temperature peaks, pressure peaks, temperature rise time, and pressure rise time are not the same, making it difficult to accurately characterize the runaway boundary using a single threshold or simple empirical formula.

[0004] Currently, data-driven methods such as artificial neural networks are used for chemical process modeling and fault diagnosis. However, if only initial temperature, initial pressure, oxygen impurity content, or initiator concentration are used as ordinary inputs, and a general feedforward neural network is used to directly predict critical temperature and critical pressure, then this method is essentially still a direct application of existing machine learning algorithms to a specific object. This type of method is not specifically designed for the technical characteristics of runaway processes in ultra-high pressure ethylene polymerization, such as simultaneous temperature and pressure surges, pressure response lag, abrupt changes in temperature rise rate, abrupt changes in pressure rise rate, and differences in inducing factors. Furthermore, it does not reflect the underestimation of the hazards of the runaway boundary under safety warning scenarios during model training. Therefore, it is difficult to fully address the problems of unstable runaway boundary prediction and insufficient engineering interpretability under complex operating conditions. Summary of the Invention

[0005] To address the shortcomings of existing safety early warning methods for ultra-high pressure ethylene polymerization reactions, which mainly rely on fixed temperature thresholds, fixed pressure thresholds, or empirical criteria, making it difficult to accurately characterize the runaway evolution under the coupled effects of multiple key state parameters such as initial temperature, initial pressure, ethylene density, oxygen impurity content, initiator concentration, and comonomer conditions, resulting in insufficient accuracy in predicting runaway boundaries under complex operating conditions, delayed identification of runaway precursors, poor generalization ability under extreme conditions, and a high risk of safety omissions, this invention provides a method for predicting and classifying safety boundaries for ultra-high pressure polymerization runaway.

[0006] The purpose of this invention is to move beyond simply relying on general artificial neural networks to directly fit initial operating parameters. Instead, it addresses the technical characteristics of runaway ethylene polymerization reactions, such as rapid temperature rise, delayed pressure rise, enhanced temperature-pressure coupling, and differences in induced pathways caused by oxygen impurities, initiators, and comonomers. It constructs specialized features including equivalent temperature rise rate, equivalent pressure rise rate, temperature-pressure coupling slope, temperature amplification factor, pressure amplification factor, and temperature-pressure coupling strength index. These features are then combined with a safety constraint loss function to train the model, thereby improving the accuracy, stability, and engineering interpretability of runaway boundary prediction.

[0007] The technical solution adopted in this invention is: a method for predicting and classifying the safety boundary of ultra-high pressure polymerization runaway. This method involves collecting experimental data on runaway ultra-high pressure ethylene polymerization reactions, constructing temperature-pressure coupling characteristics to characterize temperature response, pressure response, and the temperature-pressure coupled instability process, and fusing static operating condition characteristics, inducing factor characteristics, and dynamic equivalent temperature-pressure characteristics. A mechanism-aware safety constraint learning model is used to jointly predict the temperature runaway boundary, pressure runaway boundary, runaway severity, and risk level. Based on the prediction results, safety margin calculations and classified early warnings are performed. The method includes the following steps:

[0008] S1. Experimental Database Construction:

[0009] Experimental data on the runaway reaction of ultra-high pressure ethylene polymerization were collected under different initial temperatures, initial pressures, ethylene densities, oxygen impurity contents, initiator types, initiator concentrations, comonomer types, and comonomer concentrations. The runaway reaction data included peak temperature, peak pressure, temperature rise time, and pressure rise time.

[0010] S2. Construction of out-of-control boundary labels:

[0011] The peak temperature is used as the boundary parameter for temperature runaway, the peak pressure is used as the boundary parameter for pressure runaway, and the temperature rise time and pressure rise time are used as the time scale parameters for runaway evolution. Runaway boundary training labels are constructed based on the peak temperature, peak pressure, temperature rise time and pressure rise time.

[0012] ;

[0013] Among them, T max,k P represents the peak temperature of the k-th experiment. max,k The peak pressure value in the k-th group of experiments. For the time of temperature rise, This refers to the pressure rise time;

[0014] S3. Extraction of equivalent temperature and pressure response features:

[0015] The equivalent temperature rise is calculated based on the difference between the peak temperature and the initial temperature, and the equivalent pressure rise is calculated based on the difference between the peak pressure and the initial pressure.

[0016] ;

[0017] ; in, , T represents the equivalent temperature rise and pressure rise of the k-th experiment, respectively. max,k T represents the peak temperature of the k-th experiment. 0,k Let P be the initial temperature. max,k P represents the peak pressure of the k-th experiment. 0,k Initial pressure;

[0018] Furthermore, by combining the temperature rise time and pressure rise time, the equivalent temperature rise rate and equivalent pressure rise rate are calculated respectively:

[0019] ;

[0020] ;

[0021] Among them, R T,k For the equivalent temperature rise rate, R P,k The equivalent pressure rise rate is given by ε, which is a minimal constant to prevent the denominator from being zero.

[0022] S4. Construction of temperature-pressure coupling strength characteristics:

[0023] The temperature-pressure coupling slope is calculated based on the equivalent temperature rise rate and the equivalent pressure rise rate, and is used to characterize the strength of the transmission from temperature instability to pressure instability during ultra-high pressure ethylene polymerization.

[0024] ;

[0025] Among them, K TP,k R is used to characterize the strength of the transmission from temperature instability to pressure instability during ultra-high pressure ethylene polymerization. P,k R is the equivalent pressure rise rate. T,k The equivalent temperature rise rate is given by ε, which is a minimal constant to prevent the denominator from being zero.

[0026] Further calculations were performed on the temperature amplification factor and the pressure amplification factor, where the temperature amplification factor is the ratio of the peak temperature to the initial temperature, and the pressure amplification factor is the ratio of the peak pressure to the initial pressure.

[0027] ;

[0028] ;

[0029] Among them, A T,k A is the temperature amplification factor for the k-th group of experiments; P,k T is the pressure amplification factor for the k-th experiment. max,k T represents the peak temperature of the k-th experiment. 0,k Let P be the initial temperature. max,k P represents the peak pressure of the k-th experiment. 0,k Initial pressure; The temperature-pressure coupling strength index is jointly determined by the equivalent temperature rise rate, the equivalent pressure rise rate, the temperature amplification factor, and the pressure amplification factor, thus constructing the temperature-pressure coupling strength index:

[0030] ;

[0031] Among them, I CTP,k Z(·) represents the temperature-pressure coupling strength index of the k-th group of experiments, and Z(·) represents the normalization function used to eliminate the differences in dimensions and numerical ranges between various temperature and pressure characteristics. T,k Z(R) P,k ), Z(K TP,k ), Z(A T,k ) and Z(A P,k ) are the normalized values ​​of the equivalent temperature rise rate, equivalent pressure rise rate, temperature-pressure coupling slope, temperature amplification factor, and pressure amplification factor, respectively.

[0032] S5. Characteristic coding of inducing factors:

[0033] Oxygen impurity induction, initiator induction, and comonomer induction are respectively encoded as induction type features. Oxygen impurity content, initiator concentration, and comonomer concentration are used as induction intensity features. Initiator type and comonomer type are categorized and encoded to form an induction factor feature vector.

[0034] S6. Construction of input for multi-source feature fusion:

[0035] The initial temperature, initial pressure, and ethylene density constitute the static operating condition characteristics. The equivalent temperature rise rate, equivalent pressure rise rate, temperature-pressure coupling slope, and temperature-pressure coupling strength index constitute the temperature-pressure dynamic equivalent characteristics. The static operating condition characteristics, inducing factor characteristics, and temperature-pressure dynamic equivalent characteristics are fused to form the model input vector.

[0036] ;

[0037] Among them, F S,k For static operating conditions, F I,k As a characteristic of inducing factors, F D,k This represents the dynamic equivalent characteristics of temperature and pressure.

[0038] S7. Construction of a Mechanism-Aware Safety Constraint Learning Model:

[0039] A mechanism-aware safety constraint learning model is constructed, which includes a static operating condition feature processing module, an inducing factor feature processing module, a temperature-pressure coupling feature processing module, a feature fusion module, and a multi-task output module.

[0040] S8. Training the safety constraint loss function:

[0041] The mechanism-aware safety constraint learning model is trained using a safety constraint loss function, which includes a prediction error term, an underestimation penalty term, a temperature-pressure coupling consistency constraint term, and a risk level error term.

[0042] ;

[0043] Among them, L pred For the prediction error term, L under To underestimate the penalty, L couple L is a temperature-pressure coupling consistency constraint term. risk For the risk level error term, α, β, and γ are weighting coefficients;

[0044] Among them, the underestimation of penalties is:

[0045] ;

[0046] in, This is the predicted value for the temperature runaway boundary. λ is the predicted value of the pressure runaway boundary. T To minimize the penalty weight at the temperature boundary, λ P The penalty weight is underestimated for the pressure boundary, and λ P ≥ λ T ;

[0047] S9. Multi-task prediction of runaway boundaries:

[0048] The static operating condition characteristics, inducing factor characteristics, and temperature and pressure dynamic equivalent characteristics of the operating condition to be predicted are input into the trained mechanism-aware safety constraint learning model, and the predicted values ​​of temperature runaway boundary, pressure runaway boundary, runaway severity, and risk level are output.

[0049] ;

[0050] in, , G(·) represents the predicted value of the severity of runaway and the predicted result of the risk level, respectively. G(·) represents the trained mechanism-aware safety constraint learning model, and F represents the fusion input features of the working condition to be predicted.

[0051] S10. Safety Margin Calculation and Graded Early Warning:

[0052] Based on the difference between the current temperature and the predicted temperature runaway boundary value, the difference between the current pressure and the predicted pressure runaway boundary value, and the predicted runaway severity value, the safety margin for the ultra-high pressure ethylene polymerization reaction is calculated:

[0053] ;

[0054] ;

[0055] Among them, M T For temperature safety margin, M P For pressure safety margin, T cur Given the current temperature, P cur For the current pressure;

[0056] Based on the predicted values ​​of temperature safety margin, pressure safety margin, and runaway severity, output a graded warning result from one of the following: normal, attention, warning, danger, or emergency interlock.

[0057] Furthermore, in step S3, the equivalent temperature rise rate and equivalent pressure rise rate are calculated;

[0058] When the collected data is experimental summary data, the equivalent temperature rise rate, equivalent pressure rise rate, and temperature-pressure coupling strength index are calculated from the peak temperature, peak pressure, temperature rise time, and pressure rise time; when the collected data is online continuous monitoring data, the equivalent temperature rise rate, equivalent pressure rise rate, and temperature-pressure coupling strength index are calculated in real time from the temperature time series and pressure time series.

[0059] Furthermore, when continuous temperature and pressure time series are acquired, the equivalent temperature rise rate and equivalent pressure rise rate are determined by the first-order rate of change of temperature and the first-order rate of change of pressure within the sliding time window, and the temperature-pressure coupling strength index is jointly determined by the rate of change of temperature, the rate of change of pressure, and the pressure-temperature rate of change within the sliding time window.

[0060] Furthermore, when the experimental summary data is collected, the equivalent temperature rise rate is obtained by dividing the difference between the peak temperature and the initial temperature by the temperature rise time, and the equivalent pressure rise rate is obtained by dividing the difference between the peak pressure and the initial pressure by the pressure rise time.

[0061] Furthermore, in step S7, the static condition feature processing module is used to extract the state coupling relationship between the initial temperature, initial pressure and ethylene density, the inducing factor feature processing module is used to extract the influence of oxygen impurities, initiators and comonomers on the runaway reaction path, and the temperature-pressure coupling feature processing module is used to extract the dynamic instability relationship between the equivalent temperature rise rate, equivalent pressure rise rate, temperature-pressure coupling slope and temperature-pressure coupling strength.

[0062] Furthermore, in step S7, the multi-task output module includes a temperature boundary output unit, a pressure boundary output unit, a runaway severity output unit, and a risk level output unit, which are used to output the predicted values ​​of the temperature runaway boundary, the pressure runaway boundary, the runaway severity, and the risk level, respectively.

[0063] Furthermore, in step S8, the temperature-pressure coupling consistency constraint, the temperature-pressure coupling consistency constraint term is used to constrain the predicted pressure boundary to increase with the increase of ethylene density, initial pressure, or temperature-pressure coupling strength, and to constrain the predicted temperature boundary to increase with the increase of equivalent temperature rise rate or inducing factor strength.

[0064] Furthermore, in the runaway severity prediction, the runaway severity prediction value is jointly determined by the temperature runaway boundary prediction value, the pressure runaway boundary prediction value, the equivalent temperature rise rate, the equivalent pressure rise rate, and the temperature-pressure coupling strength index.

[0065] Furthermore, in step S10, the graded early warning output is as follows: when the safety margin is less than the first threshold, the attention level is output; when the safety margin is less than the second threshold, the early warning level is output; when the safety margin is less than the third threshold, the danger level is output; and when the safety margin is less than the fourth threshold, the emergency interlock level is output, wherein the first threshold is greater than the second threshold, the second threshold is greater than the third threshold, and the third threshold is greater than the fourth threshold.

[0066] Furthermore, in step S10, the graded early warning linkage control, the attention level corresponds to increasing the data sampling frequency, the early warning level corresponds to reducing the initiator injection amount or reducing the reaction temperature, the danger level corresponds to starting the venting preparation or restricting the feed, and the emergency interlock level corresponds to performing emergency venting or shutdown operations.

[0067] The advantages of this invention over the prior art are:

[0068] First, this invention no longer simply applies general neural networks to predict runaway reactions in ultra-high pressure ethylene polymerization. Instead, based on the runaway characteristics of the coordinated temperature and pressure rise in the reaction system, it constructs equivalent temperature rise rate, equivalent pressure rise rate, temperature-pressure coupling slope, and temperature-pressure coupling strength indexes, so that the model input features have a clear correspondence with the physical process of the runaway reaction.

[0069] Second, this invention integrates oxygen impurity induction, initiator induction, and comonomer induction into a unified induction factor feature coding system, which can distinguish the influence of different induction paths on temperature runaway boundary, pressure runaway boundary, and runaway evolution time scale, thereby improving the adaptability of the model under complex working conditions.

[0070] Third, by constructing a multi-source fusion input of static operating condition characteristics, inducing factor characteristics, and temperature and pressure dynamic equivalent characteristics, the present invention enables the model to simultaneously learn the coupling relationship between the initial thermodynamic state, the intensity of inducing factors, and the temperature and pressure dynamic response, thus overcoming the problem that a single initial parameter input model is insufficient in characterizing runaway precursors.

[0071] Fourth, this invention introduces an underestimation penalty term and a temperature-pressure coupling consistency constraint term during model training, so that the model not only pursues the minimum prediction error, but also reduces the risk of safety underreporting caused by the underestimation of temperature runaway boundary and pressure runaway boundary, thereby improving the reliability of prediction results in engineering safety early warning.

[0072] Fifth, the output results of this invention include not only the temperature runaway boundary and the pressure runaway boundary, but also the runaway severity and risk level. Furthermore, through safety margin calculation, it realizes graded early warning of normal, attention, warning, danger and emergency interlock, providing technical support for the safe operation, accident early warning and interlock control of ultra-high pressure ethylene polymerization process.

[0073] The above overview is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the invention will become readily apparent from the accompanying drawings and the following detailed description. Attached Figure Description

[0074] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0075] Figure 1 This is a schematic diagram of the structure of an experimental platform for runaway ultra-high pressure ethylene polymerization reaction.

[0076] Figure 2 This is a schematic diagram of the structure of the mechanism-aware safety constraint learning model.

[0077] Figure 3 This is the prediction result for the temperature runaway boundary.

[0078] Figure 4 This is the prediction result for the pressure runaway boundary. Detailed Implementation

[0079] Certain exemplary embodiments are described below. As will be appreciated by those skilled in the art, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are to be considered exemplary in nature and not restrictive.

[0080] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0081] Data acquisition was achieved by constructing an ultra-high pressure ethylene polymerization runaway test platform. This platform mainly includes a polymerization reaction vessel, a gas distribution and pressurization system, a heating system, an initiation system, a control system, a temperature measurement system, a pressure measurement system, a data acquisition system, and a venting system. By adjusting process parameters such as initial temperature, initial pressure, ethylene density, oxygen impurity content, initiator type, initiator concentration, comonomer type, and comonomer concentration, the reaction system was switched between the normal polymerization zone, the critical runaway zone, and the runaway evolution zone. Real-time data on temperature, pressure, and key feed parameters within the reaction vessel were collected, forming a multi-condition runaway experimental database.

[0082] In the data preprocessing stage, the peak temperature, peak pressure, temperature rise time, and pressure rise time collected under each operating condition are used to construct runaway boundary labels. Simultaneously, based on the differences between the peak temperature and the initial temperature, and between the peak pressure and the initial pressure, equivalent temperature rise amplitude, equivalent pressure rise amplitude, equivalent temperature rise rate, equivalent pressure rise rate, temperature-pressure coupling slope, temperature amplification factor, pressure amplification factor, and temperature-pressure coupling strength index are further extracted. These features are used to characterize the temperature and pressure dynamic response of the ultra-high pressure ethylene polymerization reaction as it progresses from the initial state to the runaway state.

[0083] In the model construction steps, a mechanism-aware safety constraint learning model is established to map static operating condition characteristics, inducing factor characteristics, and equivalent temperature and pressure characteristics to runaway boundary parameters and risk levels. The model includes a static operating condition feature processing module, an inducing factor feature processing module, a temperature-pressure coupling feature processing module, a feature fusion module, and a multi-task output module. Specifically, the static operating condition feature processing module receives initial temperature, initial pressure, and ethylene density to characterize the initial thermodynamic state of the reaction system; the inducing factor feature processing module receives oxygen impurity content, initiator parameters, and comonomer parameters to characterize the impact of different inducing factors on the runaway path; and the temperature-pressure coupling feature processing module receives equivalent temperature rise rate, equivalent pressure rise rate, temperature-pressure coupling slope, and temperature-pressure coupling strength index to characterize the dynamic coupling relationship between temperature instability and pressure instability. The features output from the above modules are fused by the feature fusion module, and the multi-task output module outputs predicted values ​​for temperature runaway boundary, pressure runaway boundary, runaway severity, and risk level.

[0084] During the model training phase, a safety constraint loss function is used to train the mechanism-aware safety constraint learning model. This safety constraint loss function includes a prediction error term, an underestimation penalty term, a temperature-pressure coupling consistency constraint term, and a risk level error term. Specifically, the prediction error term constrains the deviation between the model's output value and the experimentally measured value; the underestimation penalty term applies additional penalties when the model's predicted temperature or pressure runaway boundaries are lower than the experimentally measured values, reducing the risk misjudgment caused by underestimating the severity of the runaway consequences; the temperature-pressure coupling consistency constraint term ensures that the model's prediction results maintain consistency with the changing trends of the temperature-pressure coupling strength index; and the risk level error term improves the consistency between the risk level determination and the severity of the runaway. Through this training method, the model not only improves the accuracy of runaway boundary prediction but also enhances the matching between the prediction results and the response instability mechanism and safety early warning requirements.

[0085] Example 1

[0086] This study focuses on the ultra-high pressure ethylene polymerization process. Runaway evolution under different induction conditions is simulated in an ultra-high pressure ethylene polymerization runaway test platform. A runaway reaction experimental database is constructed under different initial temperatures, initial pressures, ethylene densities, oxygen impurity contents, initiator concentrations, and comonomers to verify the effectiveness of the proposed runaway boundary prediction and graded early warning method based on temperature-pressure coupling characterization and safety constraint learning. By changing the oxygen impurity content, initiator concentration, comonomer type, and initial process conditions, the reaction system is switched between normal polymerization, critical runaway, and rapid runaway states. Characteristic parameters such as peak temperature, peak pressure, temperature rise time, and pressure rise time during the transition from a stable to a runaway state under different operating conditions are recorded.

[0087] The data processing and model building process is as follows:

[0088] S1. Experimental Database Construction:

[0089] Experimental data were collected using an ultra-high pressure ethylene polymerization runaway explosion and venting test platform, such as... Figure 1 As shown. Before the experiment, the following procedures were performed in sequence: airtightness check of the reaction vessel, preparation of the gas distribution system, adjustment of the pressurization system, temperature adjustment of the heating system, and calibration of the data acquisition system. Subsequently, the initial temperature, initial pressure, ethylene density, oxygen impurity content, initiator type, initiator concentration, comonomer type, and comonomer concentration were adjusted according to the set operating conditions, and the reaction system was induced to enter the runaway evolution process through the initiation system.

[0090] During the experiment, temperature and pressure changes inside the reaction vessel were collected using temperature and pressure measurement systems, and the peak temperature T was recorded by the data acquisition system. max Peak pressure P max Temperature rise time and pressure rise time After the reaction is complete, the residual medium is discharged through the venting system, and the reaction vessel is purged to prepare for the next set of operating condition tests. An experimental database of runaway reactions in ultra-high pressure ethylene polymerization is constructed from multiple sets of experimental data.

[0091] S2. Construction of out-of-control boundary labels:

[0092] For each experimental condition that successfully obtained a runaway response, the corresponding peak temperature T is recorded. max,k As a boundary parameter for temperature runaway, the peak pressure P max,k As a boundary parameter for pressure runaway, the temperature rise time and pressure rise time As a parameter for the timescale of runaway evolution, the runaway boundary training labels for the k-th experimental sample are constructed:

[0093] ;

[0094] Among them, T max,k P represents the peak temperature of the k-th experiment. max,k The peak pressure value in the k-th group of experiments. For the time of temperature rise, This refers to the pressure rise time.

[0095] Compared with the common modeling approach that only uses temperature peak and pressure peak as output labels, this embodiment further incorporates temperature rise time and pressure rise time into the runaway boundary labeling system, enabling the model to simultaneously characterize the size of the runaway boundary and the runaway evolution rate.

[0096] S3. Extraction of equivalent temperature and pressure response features:

[0097] For the k-th experimental sample, based on the peak temperature T max,k With initial temperature T 0,k Calculate the equivalent temperature rise by the difference between them. :

[0098] ;

[0099] According to the peak pressure P max,k With initial pressure P 0,k The difference between them is used to calculate the equivalent pressure rise. :

[0100] ;

[0101] Further combining the temperature rise time and pressure rise time Calculate the equivalent temperature rise rate R respectively. T,k and equivalent pressure rise rate R P,k :

[0102] ;

[0103] Where ε is a minimal constant to prevent the denominator from being zero. The equivalent temperature rise rate R... T,k The equivalent pressure rise rate R is used to characterize how quickly the temperature of a reaction system develops from an initial state to a runaway state. P,k It is used to characterize how quickly the pressure of a reaction system develops from an initial state to a runaway state.

[0104] Through the above processing, the temperature peak, pressure peak, temperature rise time, and pressure rise time in the original experimental data are transformed into dynamic equivalent features that can reflect the development speed of runaway precursors, avoiding the problem of static fitting based solely on initial temperature, initial pressure, oxygen impurity content, or initiator concentration.

[0105] S4. Construction of temperature-pressure coupling strength characteristics:

[0106] During the runaway process of ultra-high pressure ethylene polymerization, a significant coupling relationship exists between temperature and pressure increases. To characterize the strength of the transmission from temperature instability to pressure instability, this embodiment uses the equivalent temperature rise rate R... T,k and equivalent pressure rise rate R P,k Calculate the temperature-pressure coupling slope K TP,k :

[0107] ;

[0108] Among them, K TP,kThe larger the value, the greater the pressure rise rate corresponding to the unit temperature rise rate, and the stronger the tendency of the reaction system to transform from temperature instability to pressure instability.

[0109] Further calculate the temperature amplification factor A T,k and pressure amplification factor A P,k :

[0110] ;

[0111] Among them, A T,k Used to characterize the degree of temperature amplification relative to the initial state, A P,k Used to characterize the degree of amplification of pressure relative to the initial state.

[0112] Based on this, a thermo-pressure coupling strength index I is constructed. CTP,k :

[0113] ;

[0114] in, This indicates normalization processing. The temperature-pressure coupling strength index I... CTP,k The combined effects of temperature rise rate, pressure rise rate, temperature-pressure coupling slope, temperature amplification factor, and pressure amplification factor on runaway evolution are comprehensively characterized.

[0115] S5. Characteristic coding of inducing factors:

[0116] Since runaway reactions in ultra-high pressure ethylene polymerization can be induced by various factors such as oxygen impurities, initiators, or comonomers, the peak temperature, peak pressure, temperature rise time, and pressure rise time differ under different induction pathways. To enable the model to identify the influence of different induction pathways on the runaway reaction boundary, this embodiment encodes the induction factors as features.

[0117] Oxygen impurity induction, initiator induction, and comonomer induction are encoded as induction type features; oxygen impurity content, initiator concentration, and comonomer concentration are used as induction intensity features; initiator type and comonomer type are categorized and encoded to form an induction factor feature vector F. I,k :

[0118] ;

[0119] Among them, E type,k For induced type encoding, C O2,k X represents the oxygen impurity content. k For initiator types, C X,k Y is the initiator concentration. k For the type of comonomer, C Y,k H represents the comonomer concentration. X (X k HY (Y k ) represent the unique thermal encoding vectors of the initiator type and the comonomer type in the k-th experiment, respectively.

[0120] S6. Construction of input for multi-source feature fusion:

[0121] The initial temperature T 0,k Initial pressure P 0,k and ethylene density ρ 0,k Constituting static operating condition characteristics F S,k :

[0122] ;

[0123] The equivalent temperature rise rate R T,k Equivalent pressure rise rate R P,k Temperature-pressure coupling slope K TP,k Temperature amplification factor A T,k Pressure amplification factor A P,k Temperature-pressure coupling strength index I CTP,k The dynamic equivalent characteristic of temperature and pressure F D,k :

[0124] ;

[0125] static operating condition characteristics F S,k Inducing factor characteristics F I,k The dynamic equivalent characteristics of temperature and pressure F D,k The data is then fused to form the model input vector:

[0126] ;

[0127] In this way, the model input is no longer limited to a single process parameter, but simultaneously includes the initial thermodynamic state, the intensity of inducing factors, and the dynamic temperature and pressure response characteristics, thereby improving the model's ability to describe multi-factor coupled runaway processes.

[0128] S7. Construction of a Mechanism-Aware Safety Constraint Learning Model:

[0129] Construct a mechanism-aware safety constraint learning model to establish multi-source fusion input features F k The mapping relationship between these parameters and the runaway boundary parameters. For example... Figure 2 As shown, the mechanism-aware safety constraint learning model includes a static operating condition feature processing module, an inducing factor feature processing module, a temperature-pressure coupling feature processing module, a feature fusion module, and a multi-task output module.

[0130] The static operating condition feature processing module receives static operating condition feature F. S,kThis is used to extract the state coupling relationship between initial temperature, initial pressure, and ethylene density. The inducing factor feature processing module receives the inducing factor feature F. I,k This is used to extract the effects of oxygen impurities, initiators, and comonomers on the runaway reaction pathway. The thermo-pressure coupling feature processing module receives the thermo-pressure dynamic equivalent feature F. D,k It is used to extract the dynamic instability relationship between the equivalent temperature rise rate, equivalent pressure rise rate, temperature-pressure coupling slope, and temperature-pressure coupling strength.

[0131] The outputs of the three feature processing modules are fused by the feature fusion module to obtain the fused latent feature H. k The multi-task output module is based on the fusion of latent features H. k Output predicted values ​​for temperature runaway boundary, pressure runaway boundary, runaway severity, and risk level:

[0132] ;

[0133] in, This is the predicted value for the temperature runaway boundary. This is the predicted value for the pressure runaway boundary. This is a predicted value for the severity of the loss of control. For risk level prediction results, This represents the mechanism-aware safety constraint learning model after training.

[0134] In specific implementations, the mechanism-aware safety constraint learning model can be implemented using a multi-branch artificial neural network, an ensemble learning model, a gradient boosting model, a random forest model, or a combination of the above models. Preferably, a multi-branch structure is used to process static operating condition features, inducing factor features, and dynamic equivalent features of temperature and pressure respectively, and then joint learning is performed through a feature fusion module.

[0135] S8. Training the safety constraint loss function:

[0136] The mechanism-aware safety constraint learning model is trained using a safety constraint loss function. The safety constraint loss function includes a prediction error term, an underestimation penalty term, a temperature-pressure coupling consistency constraint term, and a risk level error term, and its form is as follows:

[0137] ;

[0138] Among them, L pred For the prediction error term, L under To underestimate the penalty, L couple L is a temperature-pressure coupling consistency constraint term. risk For the risk level error term, α, β, and γ are weighting coefficients.

[0139] The underestimation penalty term is used to impose an additional penalty on the prediction results when the model-predicted temperature or pressure runaway boundaries are lower than the experimentally measured temperature or pressure runaway boundaries. Its form is as follows:

[0140] ;

[0141] Where N is the sample size. The penalty weight for the temperature boundary is underestimated. The penalty weight for underestimating the pressure boundary is set. Since underestimating the pressure boundary can lead to delayed venting, explosions, or interlocking actions, it is preferable to set a specific penalty weight. .

[0142] The temperature-pressure coupling consistency constraint term is used to ensure that the model prediction results are consistent with the trend of temperature-pressure coupling strength. When the temperature-pressure coupling strength index, equivalent pressure rise rate, or pressure amplification factor increases, the pressure runaway boundary or runaway severity output by the model should not decrease unreasonably; when the equivalent temperature rise rate or temperature amplification factor increases, the temperature runaway boundary or runaway severity output by the model should not decrease unreasonably.

[0143] Through the above training method, the model not only strives to minimize the prediction error, but also reduces the risk of safety underreporting caused by the underestimation of the runaway boundary.

[0144] S9. Multi-task prediction and validation of runaway boundary conditions

[0145] After training, the static operating condition characteristics, inducing factor characteristics, and dynamic equivalent temperature and pressure characteristics under the test conditions are input into the trained mechanism-aware safety constraint learning model to perform predictive calculations and output predicted values ​​for temperature runaway boundary, pressure runaway boundary, runaway severity, and risk level.

[0146] ;

[0147] Where F represents the fusion input features of the working condition to be predicted.

[0148] The predicted temperature runaway boundary value Compared with the experimentally measured peak temperature T max By comparison, the predicted values ​​of the pressure runaway boundary are obtained. Compared with the experimentally measured peak pressure P max A comparison was conducted. Mean squared error, mean absolute error, relative error, and underestimation ratio were used as evaluation indicators to verify the model's predictive ability for runaway boundaries.

[0149] like Figure 3As shown, the model's predicted temperature runaway boundary shows good consistency with the experimental temperature peak, and most predicted points are close to the ideal reference line, indicating that the model can accurately capture the changing trend of the temperature runaway boundary. Figure 4 As shown, the model's predicted pressure runaway boundary has good consistency with the experimental pressure peak value, indicating that the model can effectively reflect the changing law of pressure runaway boundary under different operating conditions.

[0150] Compared with ordinary models that only use initial temperature, initial pressure, oxygen impurity content, and initiator concentration as inputs, the method of this invention further introduces ethylene density, inducing factor encoding, equivalent temperature rise rate, equivalent pressure rise rate, temperature-pressure coupling slope, temperature amplification factor, pressure amplification factor, and temperature-pressure coupling strength index. By training with an underestimation of the penalty term and the temperature-pressure coupling consistency constraint term, it can more fully characterize the coupling relationship between the initial thermodynamic state, the intensity of inducing factors, and the temperature-pressure dynamic response.

[0151] S10. Safety Margin Calculation and Graded Early Warning:

[0152] After the model completes the runaway boundary prediction, based on the current temperature T cur Predicted value of temperature runaway boundary The difference between them is used to calculate the temperature safety margin M. T :

[0153] ;

[0154] Based on the current pressure P cur Compared with the predicted value of the pressure runaway boundary The difference between them is used to calculate the pressure safety margin M. P :

[0155] ;

[0156] Among them, M T The smaller the value, the closer the current temperature is to the temperature runaway boundary; M P The smaller the value, the closer the current pressure is to the pressure runaway boundary.

[0157] Based on temperature safety margin M T Pressure safety margin M P and the predicted severity of loss of control It outputs a graded warning result from one of the following levels: Normal, Attention, Warning, Danger, or Emergency Interlock. When the system is at the Normal level, it maintains the current operating state; when the system is at the Attention level, it increases the sampling frequency of temperature and pressure; when the system is at the Warning level, it reduces the initiator injection amount, lowers the reaction temperature, or adjusts the feed conditions; when the system is at the Danger level, it initiates venting preparation, restricts feed, or performs depressurization operations; when the system is at the Emergency Interlock level, it performs emergency venting or shutdown operations.

[0158] Based on the above steps, this invention further transforms the runaway boundary prediction results into safety margin and graded early warning results, realizing the expansion from simple numerical prediction to runaway risk identification, graded early warning and interlocking control, providing technical support for the safe operation and accident prevention of ultra-high pressure ethylene polymerization process.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting and classifying the safety boundary of ultra-high voltage polymer runaway, characterized in that: By collecting experimental data on the runaway of ultra-high pressure ethylene polymerization reaction, a temperature-pressure coupling characteristic is constructed to characterize the temperature response, pressure response, and temperature-pressure coupling instability process. Static operating condition characteristics, inducing factor characteristics, and dynamic equivalent temperature-pressure characteristics are fused together. A mechanism-aware safety constraint learning model is used to jointly predict the temperature runaway boundary, pressure runaway boundary, runaway severity, and risk level. Based on the prediction results, safety margin calculations and graded early warnings are performed. The method includes the following steps: S1. Experimental Database Construction: Experimental data on the runaway reaction of ultra-high pressure ethylene polymerization were collected under different initial temperatures, initial pressures, ethylene densities, oxygen impurity contents, initiator types, initiator concentrations, comonomer types, and comonomer concentrations. The runaway reaction data included peak temperature, peak pressure, temperature rise time, and pressure rise time. S2. Construction of out-of-control boundary labels: The peak temperature is used as the boundary parameter for temperature runaway, the peak pressure is used as the boundary parameter for pressure runaway, and the temperature rise time and pressure rise time are used as the time scale parameters for runaway evolution. Runaway boundary training labels are constructed based on the peak temperature, peak pressure, temperature rise time and pressure rise time. ； in, The peak temperature of the k-th experiment is... The peak pressure value in the k-th group of experiments. For the time of temperature rise, This refers to the pressure rise time; S3. Extraction of equivalent temperature and pressure response features: The equivalent temperature rise is calculated based on the difference between the peak temperature and the initial temperature. ； ； in, , These represent the equivalent temperature rise and pressure rise for the k-th experiment, respectively. The peak temperature of the k-th experiment is... The initial temperature, The peak pressure value in the k-th group of experiments. The initial pressure is used; further, combining the temperature rise time and pressure rise time, the equivalent temperature rise rate and equivalent pressure rise rate are calculated respectively: ； ； in, For the equivalent temperature rise rate, This is the equivalent pressure rise rate. To prevent extremely small constants with a denominator of zero; S4. Construction of temperature-pressure coupling strength characteristics: The temperature-pressure coupling slope is calculated based on the equivalent temperature rise rate and the equivalent pressure rise rate, and is used to characterize the strength of the transmission from temperature instability to pressure instability during ultra-high pressure ethylene polymerization. ； in, Used to characterize the strength of the transmission from temperature instability to pressure instability during ultra-high pressure ethylene polymerization. This is the equivalent pressure rise rate. For the equivalent temperature rise rate, To prevent extremely small constants with a denominator of zero; Further calculations were performed on the temperature amplification factor and the pressure amplification factor, where the temperature amplification factor is the ratio of the peak temperature to the initial temperature, and the pressure amplification factor is the ratio of the peak pressure to the initial pressure. ； ； in, The temperature amplification factor for the k-th group of experiments; Let be the pressure amplification factor for the k-th group of experiments. The peak temperature of the k-th experiment is... The initial temperature, The peak pressure value in the k-th group of experiments. Initial pressure; The thermo-pressure coupling strength index is jointly determined by the equivalent temperature rise rate, the equivalent pressure rise rate, the temperature amplification factor, and the pressure amplification factor. The thermo-pressure coupling strength index is constructed as follows: ； Among them, I CTP,k Z(·) represents the temperature-pressure coupling strength index of the k-th group of experiments, and Z(·) represents the normalization function used to eliminate the differences in dimensions and numerical ranges between various temperature and pressure characteristics. T,k Z(R) P,k ), Z(K TP,k ), Z(A T,k ) and Z(A P,k ) are the normalized values ​​of equivalent temperature rise rate, equivalent pressure rise rate, temperature-pressure coupling slope, temperature amplification factor, and pressure amplification factor, respectively. S5. Characteristic coding of inducing factors: Oxygen impurity induction, initiator induction, and comonomer induction are encoded as induction type features, while oxygen impurity content, initiator concentration, and comonomer concentration are used as induction intensity features. Furthermore, initiator type and comonomer type are categorized and encoded to form the induction factor feature vector F. I,k : ； Among them, E type,k For induced type encoding, C O2,k X represents the oxygen impurity content. k For initiator types, C X,k Y is the initiator concentration. k For the type of comonomer, C Y,k H represents the comonomer concentration. X (X k H Y (Y k ) represent the unique thermal encoding vectors of the initiator type and the comonomer type in the k-th experiment, respectively; S6. Construction of input for multi-source feature fusion: The initial temperature, initial pressure, and ethylene density constitute the static operating condition characteristics. The equivalent temperature rise rate, equivalent pressure rise rate, temperature-pressure coupling slope, and temperature-pressure coupling strength index constitute the temperature-pressure dynamic equivalent characteristics. The static operating condition characteristics, inducing factor characteristics, and temperature-pressure dynamic equivalent characteristics are fused to form the model input vector. ； Among them, F S,k For static operating conditions, F I,k As a characteristic of inducing factors, F D,k This represents the dynamic equivalent characteristics of temperature and pressure. S7. Construction of a Mechanism-Aware Safety Constraint Learning Model: A mechanism-aware safety constraint learning model is constructed, which includes a static operating condition feature processing module, an inducing factor feature processing module, a temperature-pressure coupling feature processing module, a feature fusion module, and a multi-task output module. S8. Training the safety constraint loss function: The mechanism-aware safety constraint learning model is trained using a safety constraint loss function, which includes a prediction error term, an underestimation penalty term, a temperature-pressure coupling consistency constraint term, and a risk level error term. ； Among them, L pred For the prediction error term, L under To underestimate the penalty, L couple L is a temperature-pressure coupling consistency constraint term. risk For the risk level error term, α, β, and γ are weighting coefficients; Among them, the underestimation of penalties is: ； in, This is the predicted value for the temperature runaway boundary. λ is the predicted value of the pressure runaway boundary. T To minimize the penalty weight at the temperature boundary, λ P The penalty weight is underestimated for the pressure boundary, and λ P ≥ λ T ; S9. Multi-task prediction of runaway boundaries: The static operating condition characteristics, inducing factor characteristics, and temperature and pressure dynamic equivalent characteristics of the operating condition to be predicted are input into the trained mechanism-aware safety constraint learning model, and the predicted values ​​of temperature runaway boundary, pressure runaway boundary, runaway severity, and risk level are output. ； in, , These represent the predicted severity of the loss of control and the predicted risk level, respectively. This represents the trained mechanism-aware safety constraint learning model, and F represents the fused input features of the working condition to be predicted. S10. Safety Margin Calculation and Graded Early Warning: Based on the difference between the current temperature and the predicted temperature runaway boundary value, the difference between the current pressure and the predicted pressure runaway boundary value, and the predicted runaway severity value, the safety margin for the ultra-high pressure ethylene polymerization reaction is calculated: ； ； Among them, M T For temperature safety margin, M P For pressure safety margin, T cur Given the current temperature, P cur For the current pressure; Based on the predicted values ​​of temperature safety margin, pressure safety margin, and runaway severity, output a graded warning result from one of the following: normal, attention, warning, danger, or emergency interlock.

2. The method for predicting and classifying the safety boundary of ultra-high pressure polymerization runaway according to claim 1, characterized in that: In step S3, the equivalent temperature rise rate and equivalent pressure rise rate are calculated. When the collected data is experimental summary data, the equivalent temperature rise rate, equivalent pressure rise rate, and temperature-pressure coupling strength index are calculated from the peak temperature, peak pressure, temperature rise time, and pressure rise time; when the collected data is online continuous monitoring data, the equivalent temperature rise rate, equivalent pressure rise rate, and temperature-pressure coupling strength index are calculated in real time from the temperature time series and pressure time series.

3. The method for predicting and classifying the safety boundary of ultra-high pressure polymerization runaway according to claim 2, characterized in that: When continuous temperature and pressure time series are acquired, the equivalent temperature rise rate and equivalent pressure rise rate are determined by the first-order rate of change of temperature and the first-order rate of change of pressure within the sliding time window, and the temperature-pressure coupling strength index is jointly determined by the rate of change of temperature, the rate of change of pressure, and the pressure-temperature rate of change within the sliding time window.

4. The method for predicting and classifying the safety boundary of ultra-high pressure polymerization runaway according to claim 2, characterized in that: When the experimental summary data is collected, the equivalent temperature rise rate is obtained by dividing the difference between the peak temperature and the initial temperature by the temperature rise time, and the equivalent pressure rise rate is obtained by dividing the difference between the peak pressure and the initial pressure by the pressure rise time.

5. The method for predicting and classifying the safety boundary of ultra-high pressure polymerization runaway according to claim 1, characterized in that: In step S7, the static condition feature processing module is used to extract the state coupling relationship between the initial temperature, initial pressure and ethylene density, the inducing factor feature processing module is used to extract the influence of oxygen impurities, initiators and comonomers on the runaway reaction path, and the temperature-pressure coupling feature processing module is used to extract the dynamic instability relationship between the equivalent temperature rise rate, equivalent pressure rise rate, temperature-pressure coupling slope and temperature-pressure coupling strength.

6. The method for predicting and classifying the safety boundary of ultra-high pressure polymerization runaway according to claim 1, characterized in that: In step S7, the multi-task output module includes a temperature boundary output unit, a pressure boundary output unit, a runaway severity output unit, and a risk level output unit, which are used to output the predicted values ​​of the temperature runaway boundary, the pressure runaway boundary, the runaway severity, and the risk level, respectively.

7. The method for predicting and classifying the safety boundary of ultra-high pressure polymerization runaway according to claim 1, characterized in that: In step S8, the temperature-pressure coupling consistency constraint term is used to constrain the predicted pressure boundary from increasing with the increase of ethylene density, initial pressure, or temperature-pressure coupling strength, and to constrain the predicted temperature boundary from increasing with the increase of equivalent temperature rise rate or the strength of inducing factors.

8. The method for predicting and classifying the safety boundary of ultra-high pressure polymerization runaway according to claim 1, characterized in that: In the runaway severity prediction, the runaway severity prediction value is jointly determined by the temperature runaway boundary prediction value, the pressure runaway boundary prediction value, the equivalent temperature rise rate, the equivalent pressure rise rate, and the temperature-pressure coupling strength index.

9. The method for predicting and classifying the safety boundary of ultra-high pressure polymerization runaway according to claim 1, characterized in that: In step S10, the graded early warning output is as follows: when the safety margin is less than the first threshold, the attention level is output; when the safety margin is less than the second threshold, the early warning level is output; when the safety margin is less than the third threshold, the danger level is output; and when the safety margin is less than the fourth threshold, the emergency interlock level is output. The first threshold is greater than the second threshold, the second threshold is greater than the third threshold, and the third threshold is greater than the fourth threshold.

10. The method for predicting and classifying the safety boundary of ultra-high pressure polymerization runaway according to claim 9, characterized in that: In step S10, the graded early warning linkage control, the attention level corresponds to increasing the data sampling frequency, the early warning level corresponds to reducing the initiator injection amount or reducing the reaction temperature, the danger level corresponds to starting the venting preparation or restricting the feed, and the emergency interlock level corresponds to performing emergency venting or shutdown operations.

Images