Intelligent polyolefin performance prediction and reverse design system and method

Abstract

The invention discloses a polyolefin intelligent performance prediction and reverse design system and method. The system comprises a data acquisition and management module used for acquiring and managing polyolefin production process data, product first characteristic data and product second characteristic data; the intelligent performance prediction module comprises a mechanism model module and a data driving model module; the mechanism model module is used for predicting first characteristic data of the product according to the polyolefin production process data; the data driving model module is used for predicting second characteristic data of the product according to the first characteristic data of the product; the reverse design optimization module is used for generating initial polyolefin production process data by utilizing an intelligent optimization algorithm according to a product first characteristic data target value and / or a product second characteristic data target value input by a user and performing iterative optimization in combination with the intelligent performance prediction module; and outputting the optimized polyolefin production process data meeting the target value.

Description

Technical Field

[0001] This invention relates to the field of polymer material processing, and more specifically to a smart performance prediction and reverse design system and method for polyolefin materials. Background Technology

[0002] Polyolefins (such as polyethylene (PE) and polypropylene (PP) are the most produced and widely used polymer materials, playing a crucial role in packaging, automobiles, home appliances, and medical fields. With increasing downstream demands for material performance and diversification of product varieties, polyolefin production is shifting from large-scale production of standard products to multi-variety customization.

[0003] The polyolefin industry faces numerous technological challenges in its high-quality development: On the one hand, performance control in the production process is lagging. Currently, the industrial acquisition of polyolefin properties (such as melt index MI, density, mechanical properties, etc.) mainly relies on offline laboratory testing, which typically takes several hours to obtain results. This delayed feedback makes it difficult to meet the needs of real-time production control, resulting in performance fluctuations that cannot be corrected in a timely manner. The lag in performance testing not only affects the stability of product quality but also increases the generation of substandard materials and defective products.

[0004] On the other hand, product grade switching is complex and developing new grades is costly. Polyolefin plants need to frequently switch between different grades to meet market demands, but the production conditions for different grades vary significantly, and the switching process is characterized by strong coupling and nonlinearity. Operators often rely on experience to adjust various process parameters for the transition, and slight errors can prolong the switching time or generate a large amount of transition material, resulting in waste of raw materials and energy. Reducing the transition time and material loss during grade switching has been one of the industry's long-standing pain points. How to efficiently develop new grades that meet performance requirements is a significant challenge.

[0005] To address these issues, both domestic and international efforts are gradually exploring the introduction of digital and intelligent technologies into the production and R&D of polyolefins, establishing performance prediction models and reverse design tools. Performance prediction refers to using models to predict the performance of polyolefin products under given raw material, catalyst, and process conditions; while reverse design starts from the target performance and works backward to deduce the required material structure and process parameters. Compared to traditional forward design, which requires repeated experimentation and formulation screening, reverse design directly uses the performance target as input, enabling efficient optimization to find new material solutions that meet the requirements. In the field of polymer materials, reverse design typically relies on a deep understanding of structure-property relationships (the relationship between material microstructure and macroscopic properties) and utilizes optimization algorithms such as machine learning for rapid searching.

[0006] Currently, several exploration paths for intelligent R&D of polyolefins have emerged in China. For example, a hybrid modeling strategy combining mechanistic models and data-driven approaches has been proposed in academia. This soft-sensoring approach, integrating mechanistic mechanisms and artificial intelligence, has been validated in industrial polypropylene plants. By integrating process mechanisms and plant operation data, high-precision soft-sensoring models can be constructed, allowing for real-time virtual simulation of the actual production process. Intelligent optimization algorithms can then be used to solve for the optimal control parameters of quality indicators such as melt flow index and isotacticity. Furthermore, progress has been made in polyolefin materials genomics and machine learning-based reverse design. Materials genomics advocates for the construction of a materials composition-process-structure-performance database and the acceleration of new material development through high-throughput computing and data mining. For polyolefins, high-performance computing combined with machine learning can be used to analyze the influence of polymer molecular structure on performance and establish quantitative structure-activity relationship (QSPR) models. For example, some studies have used machine learning models to correlate catalyst structure with polymerization performance, enabling rapid prediction of catalyst activity and polymer molecular weight. Additionally, for copolymer systems, polymer chain structures are characterized by design descriptors, and machine learning is used to predict the mechanical and electrical properties of materials, achieving performance-oriented reverse design. Overall, the academic community is conducting in-depth research from three levels: kinetic mechanism modeling, structure-property relationship modeling, and artificial intelligence algorithms, to provide theoretical support for the prediction and reverse design of polyolefin performance.

[0007] Despite the above explorations, the existing solutions still have limitations: (1) Many soft measurement models are only for a single performance index (such as MI), which is difficult to cover comprehensive performance prediction such as mechanical performance and processing performance; (2) Data-driven models have limited generalization ability under different devices and different formulations, and are difficult to operate reliably for a long time when there is no adaptive correction mechanism; (3) Systems that truly realize reverse design are not yet mature. Most current solutions are forward prediction + manual trial and error, lacking intelligent reverse calculation algorithms, which results in a long development cycle for new grades.

[0008] Therefore, there is an urgent need for an intelligent performance prediction and reverse design system for polyolefins that integrates mechanistic models and machine learning, and possesses self-learning and optimization search capabilities, in order to improve the level of intelligence in polyolefin product development and production. Summary of the Invention

[0009] To address the aforementioned technical problems and shortcomings in the field, this invention provides a smart performance prediction and reverse design system and method for polyolefins, aiming to achieve forward prediction of polyolefin product performance from reaction conditions, and reverse design of process formulations from target performance, thereby improving the performance controllability, design efficiency, and intelligent development level of polyolefin products.

[0010] The main objective of this invention is to address the problems of lagging performance control, difficulty in grade switching optimization, and long new product development cycles in existing technologies for polyolefin products. It provides a smart performance prediction and reverse design system and method for polyolefins. This system and method integrate polyolefin polymerization reaction kinetic modeling, polymer structure-performance relationship modeling, and machine learning and optimization algorithms to achieve accurate prediction of polyolefin material properties and reverse design of process formulations based on target properties. The aim is to accelerate the development of new polyolefin grades and improve the quality control level of the production process.

[0011] The specific technical solution is as follows: In a first aspect, the present invention provides a smart performance prediction and reverse design system for polyolefins, comprising: The data acquisition and management module is used to collect and manage polyolefin production process data, primary product characteristic data, and secondary product characteristic data. Primary product characteristic data includes one or more of the following: density, melt flow index, molecular weight distribution (including molecular weight distribution index PDI, etc.), short-chain branching distribution, molecular sequence distribution, number-average molecular weight, weight-average molecular weight, isotacticity, and crystallinity. Secondary product characteristic data includes one or more of the following: tensile breaking strength, tensile yield strength, elongation at break, tensile modulus, tensile strain, notched impact strength, glass transition temperature, flexural modulus, tensile yield strain, tensile breaking strain, Shore hardness, Rockwell hardness, environmental stress cracking, Vicat softening point, and heat distortion temperature. The intelligent performance prediction module includes a mechanism model module and a data-driven model module; the mechanism model module is used to predict the first characteristic data of the product based on the polyolefin production process data; the data-driven model module is used to predict the second characteristic data of the product based on the first characteristic data. The reverse design optimization module is used to generate initial polyolefin production process data based on the target values ​​of the first and / or second characteristics of the product input by the user, and to perform iterative optimization. The iterative optimization process includes using the intelligent performance prediction module to predict the predicted values ​​of the first and / or second characteristics of the product based on the current polyolefin production process data, and optimizing the polyolefin production process data according to the predicted values ​​and target values ​​for iteration, and outputting optimized polyolefin production process data that meets the target values.

[0012] The data acquisition and management module can acquire multi-source data on polyolefin products during R&D and production, and perform unified storage and processing. The mechanism model module establishes a kinetic mechanism model of the polyolefin polymerization reaction based on this data, used to obtain microstructural parameters such as the primary characteristics of the polyolefin material. The data-driven model module uses historical data to train a predictive model between the microstructure and macroscopic properties of polyolefins, establishing a mapping relationship between the two. Based on polyolefin production process data, the intelligent performance prediction module calls the mechanism model module to calculate microstructural parameters and the data-driven model module to predict the corresponding polyolefin performance indicators, realizing a forward prediction process. The reverse design optimization module receives the target performance requirements, uses an intelligent optimization algorithm to iteratively search the polyolefin production process data, and evaluates its performance through the intelligent performance prediction module, thus achieving the reverse derivation of a polyolefin production formulation that meets the requirements from the target performance.

[0013] The data acquisition and management module can interface with industrial control systems (such as DCS and PLC) and laboratory testing systems to collect and manage polyolefin production process data, primary product characteristic data, and secondary product characteristic data in real time or periodically. This module can clean, standardize, normalize, and archive the data, building a relational database containing polyolefin grades, polyolefin production process data, primary product characteristic data, and secondary product characteristic data to support subsequent model training and system operation.

[0014] In some preferred embodiments, the polyolefin production process data includes one or more of the following: monomer type, monomer ratio, amount of molecular weight regulator (e.g., hydrogen), monomer concentration, reaction temperature, reaction pressure, catalyst type, and catalyst dosage (including feeding rate).

[0015] In some preferred embodiments, the mechanistic model module includes a polymerization reaction model for predicting the first characteristic data of the product.

[0016] The polymerization reaction model may include: a polymerization reaction model based on a multi-active-site catalyst; or one or more combinations of gas-phase polymerization, solution polymerization, and slurry polymerization models; or a single-reactor polymerization model and / or a multi-reactor polymerization model. The polymerization reaction model can support the modeling needs of the three major industrial processes of polyolefins: gas-phase, solution polymerization, and slurry polymerization, and is applicable to different reactor configurations, including single-reactor models, multi-reactor series models, and multi-reactor parallel models. For typical fluidized bed gas-phase polymerization, the system can establish the reaction kinetic equations for each stage in the series reactor, considering factors such as gas circulation, material switching, and heat distribution.

[0017] In some preferred embodiments, the polymerization reaction model based on multi-active-site catalysts considers the differences in reaction kinetic parameters of multiple active sites and the differences in reaction rate constants of various active sites in different reactions. It can accurately simulate the reaction pathways and contributions of various molecules in the reaction and can be used to predict the first characteristic data of products under different polyolefin production process data. The reaction kinetics include five categories: chain activation, chain initiation, chain growth, chain transfer, and chain termination. Among them, chain transfer includes multiple pathways such as transfer to β-H, transfer to hydrogen, transfer to monomer, transfer to comonomer, and transfer to cocatalyst. It is suitable for the modeling needs of various reaction systems (such as Ziegler-Natta or metallocene catalytic systems).

[0018] In the multi-reactor polymerization model, the reactors can be connected in series and / or in parallel.

[0019] In some preferred embodiments, the data-driven model module employs one or more machine learning algorithms and symbolic learning to establish a predictive model for predicting the second characteristic data of the product. This predictive model can use the first characteristic data of the product as input descriptors. The first characteristic data of the product can be calculated by the mechanistic model module or directly extracted from online or experimental data (which may originate from the data acquisition and management module). The second characteristic data of the product can be partially or fully modeled and predicted according to the product application scenario. For example, for pipe systems, the focus is on environmental stress cracking and modulus; for films, the focus is on fracture strain and flexibility; and for injection molded parts, the focus is on yield strength and heat distortion temperature.

[0020] Machine learning algorithms, including but not limited to random forests, support vector regression (SVR), artificial neural networks (ANN), gradient boosting trees (GBDT), and deep neural networks (DNN), are used to capture complex nonlinear relationships and improve model prediction accuracy. The introduction of symbolic learning methods (such as symbolic regression, algebraic representation trees, or logistic induction learning) aims to obtain a more interpretable model structure, enabling the model not only to predict but also to reveal the physical mechanisms and mathematical relationships between polymer structure and performance. During the training phase, this module can automatically select the optimal model type or, based on requirements, combine machine learning models and symbolic models to construct a hybrid prediction architecture (such as symbolically constrained neural networks) to balance generalization ability and interpretability. Model training can utilize a historical polyolefin product database, establishing high-precision, multi-output prediction capabilities through supervised learning methods to support the rapid evaluation needs of the intelligent performance prediction module and the reverse engineering module.

[0021] To ensure that the structural distribution information input to the model has a clear and computable form, in some preferred embodiments, when the molecular weight distribution and short-chain branch distribution are used as model input descriptors, feature point discretization is employed. The polyolefin intelligent performance prediction and reverse design system discretizes the distribution curve along the molecular weight dimension, using the molecular weight axis as the basic dimension of the distribution function, and extracts no less than 10 and no more than 30 distribution feature points (discrete molecular weight points, corresponding to the distribution value recorded at each point) as input vectors (fixed-length structural feature vectors). The molecular weight sampling points are selected within the distribution range at equal intervals, equal intervals, or logarithmic intervals.

[0022] Proportional interval sampling: Based on the logarithmic trend of molecular weight over the entire range (e.g., 500~500,000 Da), the molecular weight values ​​of the sampling points are increased proportionally in the molecular weight space. Equal-interval sampling: Within a set molecular weight range, several sampling points are directly divided at fixed intervals; Log-interval sampling: Dividing molecular weight ranges on a logarithmic scale (e.g., taking points at equal intervals between log(Mw) from 2 to 6), suitable for situations in polymer systems that are simultaneously sensitive to low and high molecular weights.

[0023] This structured sampling method ensures that when the model faces distribution curves from any source (such as the output of the mechanistic model module, experimental GPC curves, or historical data), it can process information input in a unified dimension and with unified semantics, thereby improving the training stability and actual deployment consistency of the prediction model.

[0024] In some preferred embodiments, the intelligent performance prediction module is configured as an online soft measurement instrument, which has online soft measurement capabilities and can continuously predict the key performance of polyolefins during changes in polyolefin production process data or product switching, thereby achieving real-time quality monitoring. The intelligent performance prediction module includes a model correction unit. When the deviation between the predicted value and the measured value is detected to exceed a preset threshold, the model correction unit is triggered to update the parameters of the mechanism model module and / or the data-driven model module using newly acquired data to ensure prediction accuracy.

[0025] The reverse engineering optimization module can be equipped with a variety of intelligent optimization algorithms, which can be flexibly selected according to the complexity of the problem.

[0026] In some preferred embodiments, the reverse design optimization module employs at least one of the following optimization strategies: trust region algorithm, genetic algorithm, particle swarm optimization, and Bayesian optimization. Using the objective value as the fitness function, it searches for optimal solutions to the polyolefin production process data. In each iteration, the reverse design optimization module feeds candidate variables into the intelligent performance prediction module to obtain corresponding predicted values. Based on these predictions, it evaluates the optimization direction and iteratively generates the optimal combination of process variables that satisfies the target performance. In other words, the optimization process of the reverse design optimization module evaluates the performance of each scheme through iterative calculation and linkage with the intelligent performance prediction module. Based on the evaluation results, it adjusts the search direction and ultimately outputs one or more optimal combinations of process parameters that satisfy the objective. This module supports multi-objective optimization, weighted summation of objective functions, and Pareto front-based optimal solution selection strategies, adapting to the common "multi-index coordinated optimization" requirements in practical development. When used for multi-objective optimization, the reverse design optimization module can obtain a set of optimal solutions through weighted summation or Pareto front selection.

[0027] In some preferred embodiments, the polyolefin intelligent performance prediction and reverse design system further includes a user interface and a visualization module, providing users with an interactive entry point to achieve human-computer interaction decision support. The user interface and visualization module can be used to receive user input and display predicted values ​​and optimized polyolefin production process data.

[0028] The user interface and visualization module can consist of multiple functional sub-interfaces.

[0029] In some preferred embodiments, the user interface and visualization module include: The user input interface is used to receive user input, which may include polyolefin production process data, product primary characteristic data, product secondary characteristic data, etc. The query interface is used to query and display polyolefin grades and their corresponding formulation and performance information; The predictive monitoring interface is used to display the predicted values ​​and their comparison with the measured values ​​and / or target values ​​(in the form of charts or lists), and supports trend analysis and alarm prompts. The results interface is optimized to display (in chart or list form) the optimized polyolefin production process data that meets the target values.

[0030] In some preferred embodiments, the polyolefin intelligent performance prediction and reverse design system further includes a model training and evaluation module, which is used to (periodically or irregularly) train and adjust the intelligent performance prediction module using new data, and evaluate the model prediction error through cross-validation and / or independent test data. When the error exceeds a predetermined range, an alert is sent to the intelligent performance prediction module and / or the intelligent performance prediction module is automatically updated to ensure the long-term stable and reliable operation of the polyolefin intelligent performance prediction and reverse design system, and to ensure that the polyolefin intelligent performance prediction and reverse design system has continuous learning ability and adaptability, and can maintain prediction and design accuracy for a long time under different product and working condition changes.

[0031] In a second aspect, the present invention provides a method for predicting and reverse designing the intelligent performance of polyolefins, using the intelligent performance prediction and reverse design system for polyolefins described in the first aspect; The method for predicting and reverse-engineering the intelligent performance of polyolefins includes: The polyolefin production process data and its corresponding product primary characteristic data from the data acquisition and management module are retrieved and input into the mechanism model module for training, resulting in an optimized mechanism model module. The product's first characteristic data and its corresponding product second characteristic data are retrieved from the data acquisition and management module and input into the data-driven model module for training, resulting in an optimized data-driven model module. The user inputs the target values ​​for the first and / or second characteristics of the product. The reverse design optimization module uses an intelligent optimization algorithm to generate initial polyolefin production process data and performs iterative optimization. The iterative optimization process includes using the optimized mechanism model module to predict the predicted values ​​for the first and / or using the optimized data-driven model module to predict the predicted values ​​for the second characteristics of the product based on the current polyolefin production process data. The polyolefin production process data is then optimized based on the predicted and target values ​​for iteration, and the optimized polyolefin production process data that meets the target values ​​is output.

[0032] In summary, the polyolefin intelligent performance prediction and reverse design system of this invention, through the deep integration of data, models, and algorithms, establishes a digital chain for polyolefin material research and development and production: it can both forward predict various properties of products under given process conditions and reverse design process formulations that meet target performance. In particular, this system integrates mechanistic models and machine learning models, possessing adaptive correction and global optimization capabilities, making performance prediction more accurate and stable, and reverse engineering more efficient and intelligent.

[0033] Compared with the prior art, the beneficial effects of this invention are as follows: Real-time and accurate performance prediction: By replacing manual experiments with soft measurement models, the performance acquisition cycle is significantly shortened from several hours to real-time continuous prediction. Mechanism-based and data-fusion modeling improves the model's extrapolation ability and adaptability to different operating conditions, overcoming the shortcomings of purely empirical models that lack sufficient accuracy or are difficult to generalize.

[0034] Accelerating New Product Development: Traditional development processes for new polyolefin grades, relying on experimental formulation trials, often take months or even longer. The reverse engineering tool provided by this invention can quickly screen millions of formulation combinations on a computer, efficiently identifying candidate solutions that meet target performance requirements. This significantly shortens the R&D cycle, enabling a development model that prioritizes simulation and then verifies experimental results. Companies can respond more agilely to market demands for high-performance or customized polyolefin materials, gaining a competitive edge.

[0035] Knowledge Accumulation and Decision Support: The system's built-in grade database and models have accumulated a wealth of data and knowledge about the structure and properties of polyolefins over many years. Through a user-friendly interface, engineers can gain a deeper understanding of the key factors affecting performance and their mechanisms of action. This is invaluable for developing process optimization capabilities and supporting decision-making. When production encounters anomalies or requires adjustments, the system can provide intelligent suggestions, reducing over-reliance on personal experience and improving the level of intelligent production management.

[0036] The intelligent performance prediction and reverse design system for polyolefins of this invention has significant effects on improving the timeliness of quality control in the production process, reducing switching losses, and accelerating the development of new products. It has good prospects for industrial application and economic and social benefits. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of a polyolefin intelligent performance prediction and reverse design system according to the present invention. Detailed Implementation

[0038] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0039] The definitions of relevant terms in this invention are explained below: A smart performance prediction and reverse design system for polyolefins includes a data acquisition and management module, a smart performance prediction module, a reverse design optimization module, and a user interface and visualization module, such as... Figure 1 As shown, this illustrates the various modules and the information interaction relationships between them.

[0040] The data acquisition and management module is used to collect and manage polyolefin production process data, primary product characteristic data, and secondary product characteristic data. Primary product characteristic data includes one or more of the following: density, melt index, molecular weight distribution, short-chain branching distribution, molecular sequence distribution, number-average molecular weight, weight-average molecular weight, isotacticity, and crystallinity. Secondary product characteristic data includes one or more of the following: tensile breaking strength, tensile yield strength, elongation at break, tensile modulus, tensile strain, notched impact strength, glass transition temperature, flexural modulus, tensile yield strain, Shore hardness, Rockwell hardness, environmental stress cracking, Vicat softening point, and heat distortion temperature.

[0041] The intelligent performance prediction module includes a mechanism model module and a data-driven model module; the mechanism model module is used to predict the first characteristic data of the product based on the polyolefin production process data; the data-driven model module is used to predict the second characteristic data of the product based on the first characteristic data.

[0042] The reverse design optimization module is used to generate initial polyolefin production process data based on the target values ​​of the first and / or second characteristics of the product input by the user, and to perform iterative optimization. The iterative optimization process includes using the intelligent performance prediction module to predict the predicted values ​​of the first and / or second characteristics of the product based on the current polyolefin production process data, and optimizing the polyolefin production process data according to the predicted values ​​and target values ​​for iteration, and outputting optimized polyolefin production process data that meets the target values.

[0043] The user interface and visualization modules include: The user input interface is used to receive user input. The query interface is used to query and display polyolefin grades and their corresponding formulation and performance information; The predictive monitoring interface displays the predicted values ​​and their comparison with the measured values ​​and / or target values, and supports trend analysis and alarm prompts. The results interface is optimized to display the optimized polyolefin production process data that meets the target values.

[0044] A method for predicting and reverse designing the intelligent properties of polyolefins, employing, for example... Figure 1 The polyolefin intelligent performance prediction and reverse design system shown.

[0045] The method for predicting and reverse-engineering the intelligent performance of polyolefins includes: The polyolefin production process data and its corresponding product primary characteristic data from the data acquisition and management module are retrieved and input into the mechanism model module for training, resulting in an optimized mechanism model module. The product's first characteristic data and its corresponding product second characteristic data are retrieved from the data acquisition and management module and input into the data-driven model module for training, resulting in an optimized data-driven model module. Users input target values ​​for the first and / or second characteristics of the product through the user interface and visualization module. The reverse design optimization module uses intelligent optimization algorithms to generate initial polyolefin production process data and performs iterative optimization. The iterative optimization process includes using the optimized mechanism model module to predict the predicted value of the first characteristic of the product based on the current polyolefin production process data and / or using the optimized data-driven model module to predict the predicted value of the second characteristic of the product. The polyolefin production process data is then optimized based on the predicted values ​​and target values ​​for iteration, and the optimized polyolefin production process data that meets the target values ​​is output. The polyolefin grades and their corresponding formulations and performance information, predicted values ​​and their comparison with measured values ​​and / or target values, as well as the optimized polyolefin production process data that meets the target values ​​are all presented through the user interface and visualization modules.

[0046] Example 1: Reverse formulation design of a new polyethylene specialty material.

[0047] The goal is to develop a specialized linear low-density polyethylene (LLDPE) material for film processing, requiring a product density of 0.918-0.920 g / cm³. 3The melt index (MI) is approximately 1.0 g / 10 min (2.19 kg), and it also possesses high tensile strength and toughness. The company used the reverse design function of this invention system to accelerate research and development. First, utilizing a large amount of data from the company's existing gas-phase LLDPE production, a mechanistic model and performance prediction model for the polyethylene copolymer were established: the mechanistic model module considered ethylene-1-butene copolymerization, including coordination polymerization kinetic equations, and could simulate the molecular weight and short-chain branch distribution under different copolymer compositions; the data-driven model module used random forest regression, with the melt index, density, and molecular weight distribution of the copolymer as inputs. The molecular weight distribution was obtained uniformly at 15 points using logarithmic intervals, and the outputs were the tensile breaking strength and tensile breaking strain of the film. The model was successfully validated on historical products after training. Next, the optimization objectives were defined in the reverse design optimization module: density 0.919, MI = 1.0, and a desired tensile breaking strength ≥ 35 MPa and tensile breaking strain greater than 300%. The decision variables were selected as follows: ethylene molar fraction in feed (0.2~0.4), hydrogen-to-ethylene ratio (0.01~0.2), butene-to-ethylene ratio (0.3~0.4), polymerization pressure (1.8~2.2 MPa), and polymerization temperature (75~95℃). A genetic algorithm was used for optimization, with a population size of 50 and 100 generations. During optimization, each candidate variable combination was evaluated using a mechanistic + data model to obtain predicted values ​​for density, molecular weight index (MI), and mechanical properties. The genetic algorithm then selected the optimal solutions based on fitness. After approximately 3000 evaluations, the algorithm found several relatively optimal solutions. In the Pareto solution set displayed on the interface, one solution satisfied all objectives: ethylene molar fraction in feed 0.3, hydrogen-to-ethylene ratio 0.10, butene-to-ethylene ratio 0.32, polymerization pressure 2.1 MPa, and temperature 91℃. The model predicted that under these conditions, the resin density would be 0.9185, MI 0.98, tensile strength at break 36 MPa, and tensile strain at break 400%, all meeting the requirements. Based on this formula, the R&D team conducted pilot-scale production and successfully produced a product that met performance standards after only one round of experiments, significantly reducing the time and cost of repeated testing relying on experience. This demonstrates that the reverse engineering module of this invention can effectively explore high-dimensional process space, find polyolefin formulations that meet specific performance requirements, and accelerate new product development.

[0048] Example 2: Reverse formulation design of a new special polyethylene material.

[0049] This embodiment is the same as Example 1, except that it uses a kinetic model of ethylene-1-hexene copolymerization to find a solution that satisfies all objectives: ethylene molar fraction in feed 0.3, hydrogen-to-ethylene ratio 0.13, hexene-to-ethylene ratio 0.13, polymerization pressure 2.1 MPa, and temperature 87°C. The model predicts that under these conditions, the resin density is 0.9182, MI 1.02, tensile strength at break 38 MPa, and tensile strain at break 400%, all of which meet the requirements.

[0050] Example 3: Development of high ESCR linear low-density polyethylene materials for outdoor storage tank applications.

[0051] To meet the requirements of environmental stress cracking resistance (ESCR) and structural rigidity for medium and large outdoor water tanks and chemical storage tanks, this embodiment aims to develop a special LLDPE material suitable for rotational molding processes. The performance target is a density ≥ 0.935 g / cm³. 3 The desired properties are: melt flow index 3.0 ± 0.3 g / 10 min, ESCR F50 ≥ 1000 hours, flexural modulus ≥ 800 MPa, and tensile yield strength ≥ 19 MPa. The development process relied on the reverse design function of the system of this invention. First, a polymerization kinetic model based on ethylene and hexene copolymerization was established to simulate the short-chain branching distribution and molecular weight regulation mechanism. A data-driven model based on symbolic learning was then trained, using melt flow index, polymer molecular weight distribution, and density as inputs to predict its ESCR and rigidity properties. Decision variables were set for the feed ethylene molar fraction, hydrogen-to-ethylene ratio, hexene-to-ethylene ratio, and polymerization temperature. The polymerization pressure was set to a constant value of 2.1 MPa. The optimal formulation combination under the target performance conditions was searched using Bayesian optimization. After 1000 model evaluations, a preferred scheme was selected: ethylene molar fraction 0.40, hydrogen-to-ethylene ratio 0.25, hexene-to-ethylene ratio 0.07, and polymerization temperature 97°C. The predicted material properties are: density 0.936, MI 3.05, ESCR > 1200 hours, flexural modulus 820 MPa, and tensile yield strength 19.8 MPa. Validation shows that this scheme possesses good resistance to stress cracking and structural support, making it suitable for long-term use in high-strength outdoor storage tank products.

[0052] Example 4: Design of high-flow LLDPE material for injection molded products.

[0053] To meet the requirements of high flowability, good hardness, and dimensional stability for everyday injection-molded products (such as thin-walled caps and household goods), this embodiment develops an LLDPE product suitable for high-speed injection molding. The target values ​​are a melt flow index (MI) of approximately 20 g / 10 min and a density of approximately 0.925 g / cm³. 3The desired material has a Shore D hardness ≥ 55 and an elastic modulus between 400-500 MPa, balancing flowability and strength. Using this invention, a kinetic model of the polymerization reaction of ethylene and butene is established to simulate the polymerization behavior under copolymerization conditions. Simultaneously, a symbolic learning model is trained to predict the elastic modulus and Shore D hardness of the material. The inputs are the melt index, density, molecular weight distribution, and short-chain branching of the copolymer. Both the molecular weight distribution and short-chain branching distribution are obtained using logarithmic intervals of 10 points each. A genetic algorithm is used to search the parameter space for: ethylene molar fraction (0.2-0.4), hydrogen / ethylene ratio (0.3-0.5), butene / ethylene ratio (0.3-0.4), polymerization pressure (1.8-2.2 MPa), and polymerization temperature (75-95℃). Optimization results show that when the ethylene molar fraction is 0.3, the hydrogen / ethylene molar ratio is 0.48, the butene / ethylene ratio is 0.32, and the polymerization temperature is 88℃, the resulting material has a melt index (MI) of approximately 20.2 g / 10 min and a density of 0.925 g / cm³. 3 The Shore D hardness is 56, the elastic modulus is approximately 450 MPa, and the tensile yield strength is 15 MPa. This formulation's performance is comparable to existing injection molding materials and is suitable for high-efficiency injection molding processes requiring high molding stability.

[0054] Example 5: Reverse optimization design of high-performance materials for high-density polyethylene (HDPE) pipes.

[0055] To develop high-density polyethylene (HDPE) pipe materials that meet the PE100 standard, this embodiment proposes a design scheme for HDPE materials with a bimodal molecular weight distribution, targeting properties including a density ≥0.960 g / cm³. 3 MI5 = 0.3 g / 10 min, MI 21 =8~10 g / 10 min, tensile breaking strength ≥30 MPa, elongation at break ≥700%, elastic modulus ≥1200 MPa. A kinetic model based on the bimodal polymerization mechanism is constructed in the system, using melt index, density, and molecular weight distribution (10 points at equal intervals) as inputs, and tensile breaking strength and elongation at break as outputs, to build a symbolic learning model. This process obtains a product with a bimodal molecular weight distribution by adding two different catalysts A and B in the same reactor. The A / B ratio of the catalyst system (i.e., the mass flow ratio of catalyst A and catalyst B) and the time of different polymerization stages are set as process variables; a multi-objective genetic algorithm is used to perform a global search of the variable space. The optimal parameter combination is finally obtained as a catalyst A:B ratio of 0.6:0.4. The predicted performance result is a density of 0.961 g / cm³. 3 MI5=0.29, MI 21=8.9, tensile strength at break 33 MPa, elongation at break >700%, elastic modulus 1300 MPa. The overall performance of the material reaches the level of existing PE100 products, verifying that this system has the reverse design capability for high-performance HDPE materials.

[0056] Example 6: Development of high-rigidity heat-resistant polypropylene materials for injection-molded parts in the automotive industry.

[0057] To meet the demands of the automotive and industrial sectors for high-rigidity, high-heat-resistance, and excellent processability polypropylene injection molding materials, this embodiment aims to develop a PP material with rapid injection molding cycles and structural support performance. Performance requirements include a melt flow index of approximately 30 g / 10 min (230°C, 2.16 kg), tensile yield strength ≥30 MPa, flexural modulus ≥1800 MPa, Vicat softening point ≥150°C, heat distortion temperature ≥130°C, and certain low-temperature impact toughness. The system first constructs a polymerization mechanism model for block copolymer polypropylene, focusing on simulating the effects of crystallinity and molecular sequence distribution on rigidity and thermal properties. Then, a multivariate neural network model is used, with input descriptors including isotacticity, molecular weight distribution, and molecular sequence distribution, and the output being predicted mechanical and thermal properties. A trust region algorithm is used for optimization search, with the objective function set as the comprehensive minimum performance deviation and the maximum rigidity-thermal stability score. After approximately 1200 optimization iterations, a set of parameter combinations matching the design requirements was obtained, predicting the material properties as follows: MI = 30.2 g / 10min, tensile yield strength 31 MPa, flexural modulus 1900 MPa, Vicat softening point 152℃, and heat distortion temperature 135℃. This example demonstrates the industrial application potential of this system in the design of high-performance injection-molded polypropylene.

[0058] In summary, through the description of the above embodiments, those skilled in the art can clearly understand the system architecture and working principle of the present invention. The intelligent performance prediction and reverse design system and method for polyolefins provided by the present invention organically integrates data, models, and optimization algorithms, and can be widely applied to process control and new product development in polyolefin production plants such as polyethylene and polypropylene. It has significant effects in improving product quality consistency, reducing production costs, and shortening the R&D cycle. The specific embodiments of the present invention have been described above, but the present invention is not limited to the above embodiments. Various modifications and improvements can be made by those skilled in the art without departing from the spirit of the present invention, and these should also be considered within the scope of protection of the present invention.

Claims

1. A smart performance prediction and reverse design system for polyolefins, characterized in that, include: The data acquisition and management module is used to collect and manage polyolefin production process data, primary product characteristic data, and secondary product characteristic data. Primary product characteristic data includes one or more of the following: density, melt index, molecular weight distribution, short-chain branching distribution, molecular sequence distribution, number-average molecular weight, weight-average molecular weight, isotacticity, and crystallinity. Secondary product characteristic data includes one or more of the following: tensile breaking strength, tensile yield strength, elongation at break, tensile modulus, tensile strain, notched impact strength, glass transition temperature, flexural modulus, tensile yield strain, tensile breaking strain, Shore hardness, Rockwell hardness, environmental stress cracking, Vicat softening point, and heat distortion temperature. The intelligent performance prediction module includes a mechanism model module and a data-driven model module; the mechanism model module is used to predict the first characteristic data of the product based on the polyolefin production process data; the data-driven model module is used to predict the second characteristic data of the product based on the first characteristic data. The reverse design optimization module is used to generate initial polyolefin production process data based on the target values ​​of the first and / or second characteristics of the product input by the user, and to perform iterative optimization. The iterative optimization process includes using the intelligent performance prediction module to predict the predicted values ​​of the first and / or second characteristics of the product based on the current polyolefin production process data, and optimizing the polyolefin production process data according to the predicted values ​​and target values ​​for iteration, and outputting optimized polyolefin production process data that meets the target values.

2. The intelligent performance prediction and reverse design system for polyolefins according to claim 1, characterized in that, The polyolefin production process data includes one or more of the following: monomer type, monomer ratio, molecular weight regulator dosage, monomer concentration, reaction temperature, reaction pressure, catalyst type, and catalyst dosage.

3. The intelligent performance prediction and reverse design system for polyolefins according to claim 1, characterized in that, The mechanism model module includes a polymerization reaction model for predicting the primary characteristic data of a product; The polymerization reaction model includes: a polymerization reaction model based on a multi-active-center catalyst, or one or more combinations of gas-phase polymerization reaction model, solution polymerization reaction model, slurry polymerization reaction model, or single-reactor polymerization reaction model and / or multi-reactor polymerization reaction model. The polymerization reaction model based on multi-active-site catalysts takes into account the differences in reaction kinetic parameters of multiple active sites. The reaction kinetics include five categories: chain activation, chain initiation, chain growth, chain transfer, and chain termination. Chain transfer includes transfer to β-H, transfer to hydrogen, transfer to monomer, transfer to comonomer, and transfer to cocatalyst. In the multi-reactor polymerization model, the reactors are connected in series and / or in parallel.

4. The intelligent performance prediction and reverse design system for polyolefins according to claim 1, characterized in that, The data-driven model module uses one or more of machine learning algorithms and symbolic learning to establish a predictive model for predicting the second characteristic data of a product.

5. The intelligent performance prediction and reverse design system for polyolefins according to claim 1 or 4, characterized in that, When molecular weight distribution and short-chain branch distribution are used as model input descriptors, feature point discretization is employed. The polyolefin intelligent performance prediction and reverse design system discretizes the distribution curve along the molecular weight dimension, using the molecular weight axis as the basic dimension of the distribution function, and extracts no less than 10 and no more than 30 distribution feature points as input vectors. Molecular weight sampling points are selected within the distribution range at equal intervals, equal intervals, or logarithmic intervals.

6. The intelligent performance prediction and reverse design system for polyolefins according to claim 1, characterized in that, The intelligent performance prediction module is configured as an online soft measurement instrument. The intelligent performance prediction module includes a model correction unit. When the deviation between the predicted value and the measured value exceeds a preset threshold, the model correction unit is triggered to update the parameters of the mechanism model module and / or the data-driven model module using newly acquired data to ensure prediction accuracy.

7. The intelligent performance prediction and reverse design system for polyolefins according to claim 1 or 2, characterized in that, The reverse design optimization module employs at least one of the following optimization strategies: trust region algorithm, genetic algorithm, particle swarm optimization, and Bayesian optimization. It uses the objective value as the fitness function to perform optimization search on polyolefin production process data. When used for multi-objective optimization, the reverse design optimization module obtains the optimal solution set through weighted summation or Pareto front screening.

8. The intelligent performance prediction and reverse design system for polyolefins according to claim 1, characterized in that, The intelligent performance prediction and reverse design system for polyolefins also includes a user interface and a visualization module; The user interface and visualization modules include: The user input interface is used to receive user input. The query interface is used to query and display polyolefin grades and their corresponding formulation and performance information; The predictive monitoring interface displays the predicted values ​​and their comparison with the measured values ​​and / or target values, and supports trend analysis and alarm prompts. The results interface is optimized to display the optimized polyolefin production process data that meets the target values.

9. The intelligent performance prediction and reverse design system for polyolefins according to claim 1, characterized in that, The polyolefin intelligent performance prediction and reverse design system also includes a model training and evaluation module, which is used to train and adjust the intelligent performance prediction module using new data, and evaluate the model prediction error through cross-validation and / or independent test data. When the error exceeds a predetermined range, an alert is sent to the intelligent performance prediction module and / or the intelligent performance prediction module is automatically updated to ensure the long-term stable and reliable operation of the polyolefin intelligent performance prediction and reverse design system.

10. A method for predicting and reverse designing the intelligent properties of polyolefins, characterized in that, The polyolefin intelligent performance prediction and reverse design system according to any one of claims 1 to 9 is adopted; The method for predicting and reverse-engineering the intelligent performance of polyolefins includes: The polyolefin production process data and its corresponding product primary characteristic data from the data acquisition and management module are retrieved and input into the mechanism model module for training, resulting in an optimized mechanism model module. The product's first characteristic data and its corresponding product second characteristic data are retrieved from the data acquisition and management module and input into the data-driven model module for training, resulting in an optimized data-driven model module. The user inputs the target values ​​for the first and / or second characteristics of the product. The reverse design optimization module uses an intelligent optimization algorithm to generate initial polyolefin production process data and performs iterative optimization. The iterative optimization process includes using the optimized mechanism model module to predict the predicted values ​​for the first and / or using the optimized data-driven model module to predict the predicted values ​​for the second characteristics of the product based on the current polyolefin production process data. The polyolefin production process data is then optimized based on the predicted and target values ​​for iteration, and the optimized polyolefin production process data that meets the target values ​​is output.

Images