Polyolefin production parameter optimization method and apparatus, device, and storage medium
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- EAST CHINA UNIV OF SCI & TECH
- Filing Date
- 2025-07-25
- Publication Date
- 2026-05-15
Smart Images
Figure CN2025110516_15052026_PF_FP_ABST
Abstract
Description
A method, apparatus, equipment and storage medium for optimizing polyolefin production parameters Technical Field
[0001] This application relates to the field of automation technology, and in particular to a method, apparatus, equipment and storage medium for optimizing polyolefin production parameters. Background Technology
[0002] Polyolefin production parameters refer to the operating parameters of the production equipment during the polyolefin production process, such as PP molecular weight, crystallinity, content, molecular weight, average ethylene content, and copolymer sequence distribution of the two types of ethylene-propylene copolymers. By adjusting the polyolefin production parameters, the microscopic quality indicators and microscopic distribution of polyolefin products will be directly affected.
[0003] However, in current industrial production, due to limitations in production technology and the complexity of production control, when a new polyolefin production plant is put into operation or when producing a new grade of polyolefin product, it is often necessary to rely on a large amount of trial operation and experimental data to conduct inverse optimization studies on polyolefin production parameters in order to find production parameters that meet the standards, which greatly reduces the optimization efficiency of polyolefin production parameters. Summary of the Invention
[0004] This invention provides a method, system, equipment, and storage medium for optimizing polyolefin production parameters, thereby improving the efficiency of polyolefin production parameter optimization.
[0005] In a first aspect, this application provides a method for optimizing polyolefin production parameters, the method comprising:
[0006] For each of the historical production parameter sets corresponding to multiple historical polyolefin categories, a corresponding historical product prediction model is constructed. Each historical production parameter set includes multiple historical production parameters and the molecular weight distribution of polyolefin products for the corresponding historical polyolefin category.
[0007] Based on the prediction models of each historical product, the parameter set to be optimized for the target polyolefin category is iteratively optimized until the parameter set to be optimized meets the production conditions of the target category, thus obtaining the target production parameter set.
[0008] Secondly, this application provides a polyolefin production parameter optimization device, the device comprising:
[0009] The building unit is used to construct corresponding historical product prediction models for each of the historical production parameter sets corresponding to multiple historical polyolefin categories; each historical production parameter set includes multiple historical production parameters and molecular weight distribution of polyolefin products for the corresponding historical polyolefin category.
[0010] The optimization unit is used to iteratively perform parameter optimization processing on the set of parameters to be optimized for the target polyolefin category based on each historical product prediction model, until the set of parameters to be optimized meets the production conditions of the target category, thereby obtaining the target production parameter set.
[0011] Optionally, the optimization unit is specifically used for:
[0012] Based on the aforementioned historical product prediction models, the set of parameters to be optimized is subjected to prediction comparison processing to obtain the similarity weights of each historical product prediction model.
[0013] Based on the obtained similarity weights, recommended production parameters for the target polyolefin category are determined;
[0014] Based on the recommended production parameters, the set of parameters to be optimized is updated, and based on the updated set of parameters to be optimized, the next parameter optimization process is performed.
[0015] Optionally, the optimization unit is specifically used for:
[0016] Based on the set of parameters to be optimized, a corresponding target product prediction model is constructed.
[0017] Based on the model similarity between each historical product prediction model and the target product prediction model, the similarity weight of each historical product prediction model is obtained.
[0018] Optionally, the optimization unit is specifically used for:
[0019] Based on the acquisition function corresponding to each historical production prediction model, and combined with the similarity weights, the recommendation function for the target polyolefin category is obtained.
[0020] Based on the recommendation function, the recommended production parameters are obtained.
[0021] Optionally, if the target product prediction model includes multiple local product prediction models in local search spaces, then the optimization unit is further configured to:
[0022] From the set of parameters to be optimized, obtain multiple subsets of parameters to be optimized, and determine the local search space corresponding to each subset of parameters to be optimized;
[0023] Transfer Bayes processing is performed on each local search space to obtain the local product prediction model corresponding to each local search space.
[0024] Optionally, the optimization unit is further configured to:
[0025] For each local search space, calculate the model similarity between each historical product prediction model and the corresponding local product prediction model;
[0026] Based on the similarity of each model, the local similarity weights of each historical product prediction model are obtained; the local similarity weights represent the local similarity weights of the historical product prediction models in the corresponding local search space.
[0027] Thirdly, this application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the polyolefin production parameter optimization methods described in the first aspect above.
[0028] Fourthly, this application provides a computer storage medium storing computer program instructions, which are executed by a processor using any of the polyolefin production parameter optimization methods described in the first aspect above.
[0029] Fifthly, an embodiment of this application provides a computer program product including computer program instructions, which, when executed by a processor, implement any one of the polyolefin production parameter optimization methods described in the first aspect above.
[0030] The beneficial effects of this invention are as follows:
[0031] This application provides a method for optimizing polyolefin production parameters. The method constructs corresponding historical product prediction models for each of the historical production parameter sets corresponding to multiple historical polyolefin categories. Using these models, iterative parameter optimization is performed on the parameter set to be optimized for the target polyolefin category until the parameter set meets the production conditions of the target category, thus obtaining the target production parameter set. In each iteration, the method compares and predicts the parameter set to be optimized using each historical product prediction model to obtain similarity weights. Based on these weights, recommended production parameters for the target polyolefin category are determined. The parameter set is then updated based on these recommended parameters, and the next parameter optimization process is performed. By fully utilizing historical production data, suitable polyolefin production parameters can be quickly found, reducing testing and debugging time, lowering production cycles and optimization costs, and significantly improving the optimization efficiency of polyolefin production parameters. Attached Figure Description
[0032] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0033] Figure 1 is a schematic diagram of a polypropylene production process provided in an embodiment of this application;
[0034] Figure 2 is a flowchart of a method for optimizing polyolefin production parameters provided in an embodiment of this application;
[0035] Figure 3 is a flowchart of a parameter optimization method provided in an embodiment of this application;
[0036] Figure 4 is a schematic diagram of a polyolefin production parameter optimization process provided in an embodiment of this application;
[0037] Figure 5 is a schematic diagram of the simulation optimization results of production parameters for a multi-stage polymerization process of propylene provided in an embodiment of this application;
[0038] Figure 6 is a schematic diagram of a polyolefin production parameter optimization device provided in an embodiment of this application;
[0039] Figure 7 is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation
[0040] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.
[0041] The terms "first" and "second" in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the term "comprising" and any variations thereof are intended to cover non-exclusive protection. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices. The term "multiple" in this application can mean at least two, for example, two, three, or more, and this application does not impose limitations.
[0042] The term "and / or" in the embodiments of this application is merely a description of the association relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.
[0043] It is understood that the following specific embodiments of this application involve data related to polyolefin production parameters. When the various embodiments of this application are applied to specific products or technologies, relevant licenses or consents are required, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, relevant volunteers can be recruited and agreements can be signed to authorize their data, thereby enabling the implementation using the data of these volunteers; alternatively, implementation can be carried out within an authorized organization, using data from members of the organization to implement the following implementation methods for data management; or, the relevant data used in the specific implementation may be simulated data, such as simulated data generated in a virtual scenario.
[0044] The design concept of the embodiments of this application will be briefly introduced below.
[0045] Polyolefin production parameters refer to the operating parameters of the production equipment during the polyolefin production process, such as PP molecular weight, crystallinity, content, molecular weight, average ethylene content, and copolymer sequence distribution of the two types of ethylene-propylene copolymers. By controlling these production parameters, the microscopic quality indicators and microstructure of polyolefin products are directly affected. The microscopic quality and distribution of polymer materials determine their different uses or properties, thus influencing market demand and competitiveness. Controllable production of high-performance polyolefins is a crucial means to increase product added value and meet high-end demands. Therefore, how to stably control the microscopic quality of polyolefin products by optimizing production parameters is a key issue.
[0046] However, in current industrial production, due to limitations in production technology and the complexity of production control, when a new polyolefin production unit is put into operation or produces a new grade of polyolefin product, it is often necessary to rely on a large amount of trial operation and experimental data to conduct inverse optimization studies on polyolefin production parameters in order to find production parameters that meet the standards. This not only leads to low optimization efficiency and long optimization cycle of polyolefin production parameters, resulting in serious waste of capacity and resources, but also the high cost of obtaining production test data, resulting in high economic and time costs for the optimization process.
[0047] In view of the above problems, this application provides a method for optimizing polyolefin production parameters. This method constructs corresponding historical product prediction models for historical production parameter sets corresponding to multiple historical polyolefin categories. Through each historical product prediction model, iterative parameter optimization is performed on the parameter set to be optimized for the target polyolefin category until the parameter set meets the production conditions of the target category, thus obtaining the target production parameter set. Furthermore, in each iteration, this application compares and predicts the parameter set to be optimized using each historical product prediction model to obtain the similarity weights of each model. Based on these similarity weights, recommended production parameters for the target polyolefin category are determined. Then, the parameter set to be optimized is updated based on the recommended production parameters, and the next parameter optimization process is performed. In this way, by fully utilizing historical production data, suitable polyolefin production parameters can be quickly found, reducing experimental debugging time, production cycle and optimization costs, and significantly improving the efficiency of production parameter optimization.
[0048] The following is a brief introduction to the application scenarios to which the technical solutions of the embodiments of this application are applicable. It should be noted that the application scenarios described below are only for illustrating the embodiments of this application and are not intended to limit the scope. In specific implementation, the technical solutions provided by the embodiments of this application can be flexibly applied according to actual needs.
[0049] The solutions provided in this application are applicable to process optimization in most polyolefin production processes, significantly improving the efficiency of polyolefin production parameter optimization. For example, Figure 1 shows a schematic diagram of a polypropylene production process provided in this application. The diagram illustrates the entire production process from prepolymerization and circulation polymerization to gas-phase polymerization, demonstrating how the molecular structure and quality control of the final product are achieved through reaction control at different stages in polypropylene production. As a thermoplastic, polypropylene's physicochemical properties can be altered by adjusting raw material ratios and reaction conditions during production to meet diverse application requirements. Different grades of polypropylene represent different categories of polypropylene products, each with specific performance indicators such as melt index, molecular weight distribution, rigidity, and toughness, used for different applications such as films, fibers, and injection molded products. To prepare the target grade of polypropylene, with product molecular weight distribution as the optimization objective, 11 production parameters are selected as decision variables based on the actual operating conditions of the plant, such as the hydrogen feed rate of four reactors, the propylene feed rate of four reactors, the catalyst feed rate of reactor R-401, the co-catalyst feed rate, and the ethylene feed rate. Accurately selecting target production parameters suitable for the production conditions of the desired polyolefin category will reduce commissioning time and waste generation, thereby improving polyolefin production efficiency. Simultaneously, target production parameters can effectively control key properties such as molecular weight distribution and product purity, which is of great significance for subsequent processing and applications in polyolefins.
[0050] Of course, the methods provided in this application are not limited to the above-described application scenarios, and can also be used in other possible application scenarios. This application does not impose any limitations. The functions that each device can achieve in the above application scenarios will be described in subsequent method embodiments, and will not be elaborated upon here.
[0051] The following describes the methods provided by exemplary embodiments of this application in conjunction with the application scenarios described above and with reference to the accompanying drawings. It should be noted that the application scenarios described above are only shown to facilitate understanding of the spirit and principles of this application, and the embodiments of this application are not limited in any way in this respect.
[0052] Referring to Figure 2, which is a flowchart of a method for optimizing polyolefin production parameters according to an embodiment of this application, the specific implementation process of this method is as follows:
[0053] Step 201: For each of the historical production parameter sets corresponding to multiple historical polyolefin categories, construct corresponding historical product prediction models.
[0054] In this embodiment, each historical production parameter set includes multiple historical production parameters and the molecular weight distribution of polyolefin products corresponding to the relevant historical polyolefin category. The historical production parameter set includes specific production parameters collected from the production processes of multiple historical polyolefin categories, which directly affect the polyolefin production process and may include operational variables such as temperature, pressure, catalyst type and dosage, reaction time, reactor type, and operating mode. The molecular weight distribution of polyolefin products refers to the distribution of molecular chain lengths in the produced polyolefin products. Molecular weight distribution is a crucial factor determining the physical properties of polyolefin materials. Therefore, the ultimate goal of optimizing polyolefin production parameters is to ensure that the molecular weight distribution of the obtained polyolefin products meets the expected material performance requirements. The historical product prediction model is a mathematical model constructed based on the historical production parameter set and the corresponding product molecular weight distribution through machine learning, statistical modeling, and other methods. This model can learn from known historical production parameters and predict the product molecular weight distribution of the target polyolefin category.
[0055] Specifically, taking the preparation of polypropylene of a target polyolefin type (target grade) as an example, this application embodiment will analyze the relevant process variables of the propylene polymerization process, select operating variables such as the reactor hydrogen feed rate, propylene feed rate, catalyst feed rate, co-catalyst feed rate, and ethylene feed rate as production parameters, use the molecular weight distribution of the product as a quality indicator, collect the production parameter values of different polyolefin grades using the plant system, and calculate the molecular weight distribution curve based on the measurement data as a historical production parameter set.
[0056] in, This represents n sets of data samples collected during the polypropylene production process. A large amount of historical data was collected through the factory system, covering production data for multiple polyolefin grades. This includes historically collected production parameter sets and corresponding product quality correlation data, serving as the foundational data for subsequent model training and optimization. k This represents the historical production parameters for the k-th sample, which may specifically include the reactor's hydrogen feed rate, propylene feed rate, catalyst feed rate, etc. k This represents the molecular weight distribution of the product for the k-th sample, i.e., the target variable under these production parameters. n: represents the total number of samples. This data is derived from a large amount of historical data collected through the plant system, covering production data for multiple polyolefin grades.
[0057] In one possible implementation, when data is scarce, embodiments of this application may employ a Gaussian Process (GP) model to construct a historical product prediction model. A Gaussian Process is a non-parametric Bayesian method for regression problems that assumes the relationship between data points can be represented by a kernel function, thereby enabling prediction of unknown functions. The Gaussian Process provides mean and uncertainty estimates for function values; that is, for a given input point, it can not only predict the output value but also estimate the uncertainty of the prediction. The mean of the GP model is denoted as μ(x). * Uncertainty arises from variance σ. 2 (x * The formula is shown below:
[0058] Where, σ 2 (x * ) represents the Gaussian process model at point x * The variance at a given point, i.e., the reliability of the predicted value, reflects the uncertainty of the predicted molecular weight distribution of the product under specific production parameters. A large uncertainty indicates that the Gaussian process model's prediction at that point has low reliability. In practice, a large variance means the model lacks sufficient training data to support that point, while a small variance indicates that the model's prediction at that point is more reliable.
[0059] k * Representative evaluation point x * The covariance vector between the evaluation point x and other historical data points is used to measure the similarity between the evaluation point x and other historical data points; K represents a kernel matrix of size n×n, where each element K i,j Satisfying K i,j =k(x i ,x jThe kernel function is used to calculate the covariance matrix between each data point, measuring the similarity between data points. Through the kernel function, the model can capture the similarity between historical production parameter sets and target production parameter sets. For example, if polyolefin products from two different production conditions have similar molecular weight distributions, the kernel function will assign them a high degree of similarity.
[0060] y contains the labels corresponding to all current data x, which is the known molecular weight distribution of products in the historical production parameter set.
[0061] The noise term represents the measurement noise in the data, where I is the identity matrix. Noise is unavoidable in actual production processes, such as sensor measurement errors or environmental factors. This term is used to represent the impact of this noise on the model's prediction results, thereby improving the robustness of the predictions.
[0062] Thus, by inputting the historical production parameter sets and product quality data of each historical grade into the GP model corresponding to that grade, the product performance under the new combination of production parameters can be predicted, providing support for subsequent similarity assessment. By converting historical data into a predictive model, this embodiment effectively utilizes a large amount of data accumulated during historical production processes, avoiding starting from scratch for each production optimization and saving costs on data collection and model training. Based on the knowledge from multiple historical product prediction models, it can help infer the production parameters of the target polyolefin category, improve the reliability of the model's prediction of production parameters for new grades, and enhance the model's generalization ability.
[0063] Step 202: Based on the prediction models of each historical product, iteratively perform parameter optimization processing on the set of parameters to be optimized for the target polyolefin category until the set of parameters to be optimized meets the production conditions of the target category, and obtain the target production parameter set.
[0064] In this embodiment of the application, the parameters to be optimized for the target polyolefin category are optimized in multiple rounds using a historical product prediction model to obtain the optimal combination of production parameters for the target polyolefin category, i.e., the target production parameter set, so as to maximize the product quality of the target polyolefin category.
[0065] Specifically, the ability of historical product prediction models built from data on different polyolefin grades to generalize to target polyolefin grades is defined as follows: Among them, f i (x) represents the model output for the i-th polyolefin grade. For each different polyolefin grade, there is a corresponding historical product prediction model to predict its product molecular weight distribution. This refers to historical production parameters and product molecular weight information related to polyolefin grade i. Different polyolefin grades in the polyolefin production process correspond to different production parameters such as catalyst feed rate and propylene or ethylene feed rate. Therefore, by using multiple models to predict different grades separately, the production process of different grades can be better optimized.
[0066] In one possible implementation, referring to FIG3, the present application embodiment will specifically perform the following operations in each parameter optimization process when executing step 202:
[0067] Step 2021: Based on the prediction models of each historical product, perform prediction comparison processing on the set of parameters to be optimized to obtain the similarity weight of each historical product prediction model.
[0068] In this embodiment, after establishing historical product prediction models using historical data from different polyolefin grades, their ability to generalize to the target polyolefin category is evaluated. The established historical product prediction models predict the set of parameters to be optimized for the target polyolefin category, and the similarity between them is determined based on the accuracy of the observation results, thereby obtaining the similarity weight of each historical product prediction model. Calculating the similarity weight allows for a more accurate assessment of the similarity between historical production parameters and the target polyolefin category, thereby dynamically selecting the most relevant historical model for optimization, improving optimization accuracy, and reducing errors.
[0069] Specifically, in this application, a corresponding loss function will be constructed for each historical product prediction model to measure the set of parameters to be optimized for each historical product prediction model for the target polyolefin category. The accuracy of the ranking measures whether each historical product prediction model correctly ranks the similarity between its production parameters and historical parameters for the target polyolefin category. Through this loss function, the embodiments of this application can continuously adjust the model parameters to better capture the similarity between the target polyolefin category and historical production data. Thus, the ranking-based loss function can improve the model's prediction accuracy for the target production parameter set, reduce trial-and-error costs in the production process, and increase production efficiency. Given n t For target task evaluation with a value greater than 1, we define the loss as the number of misranked observation pairs, and the formula for the loss function is as follows:
[0070] Where i represents the label of the i-th polyolefin category (brand), and when i = t, it represents the target polyolefin brand.
[0071] The loss function represents the historical product prediction model corresponding to the i-th historical polyolefin category, used to measure the accuracy of the corresponding historical product prediction model in predicting the target polyolefin category. The loss function is used to evaluate the performance of different historical prediction models in the ranking task. A lower loss function value indicates that the model's prediction is closer to the actual value.
[0072] This represents the set of production parameters for the t-th sample, such as the feed rate, temperature, and pressure in polyolefin production.
[0073] This indicates that after removing the k-th sample, the model produces the parameter set. The prediction results. This step is similar to cross-validation and aims to prevent the model from overfitting. This represents the prediction model of historical products corresponding to the i-th polyolefin category, and the data of the l-th historical polyolefin sample. The results of the predictions. These two functions represent the predictions for the target production parameters and historical production parameters, respectively. The target production parameters are to be optimized, while the historical production parameters are used to guide model training.
[0074] and These represent the actual observed value of the k-th sample in the target polyolefin category and the actual observed value of the l-th sample in the historical polyolefin category, respectively. These two parameters are used to compare the differences between the model's predicted values and the actual observed values, helping the model to optimize the prediction results.
[0075] 1() represents an indicator function used to assess whether there is an error between the predicted result and the actual observed value. If the order between the predicted and actual values matches the expectation, the indicator function has a value of 1, indicating that the prediction result is correct; otherwise, the value is 0.
[0076] The XOR operator is used to compare predicted and actual values. It determines the difference between the predicted and actual sorting. In sorting tasks, a correct prediction results in a zero XOR value, while a wrong prediction results in a non-zero value.
[0077] In one possible implementation, the embodiments of this application will construct a corresponding target product prediction model based on the set of parameters to be optimized, and obtain the similarity weight of each historical product prediction model based on the model similarity between each historical product prediction model and the target product prediction model.
[0078] Specifically, for a limited amount of known data on the target polyolefin category (target grade), a partial sample of data from the target grades is selected to construct a generalized GP model (GP model) as the target product prediction model. The similarity of the GP model constructed from the selected target grade data to the target task is determined by the accuracy of the GP model in predicting the unselected data. For the target product prediction model, leave-one-out cross-validation can be used in this embodiment to evaluate the predictions of the target product prediction model. The ability to use existing observations. The loss function of the target product prediction model is shown in the following equation:
[0079] in, This represents the model's response after removing the k-th sample. This compares the predicted value *f* with its actual observed value. It determines whether the model's predicted value for the target polyolefin category is smaller than the actual observed value. This term is true if the model's prediction is lower than the actual value.
[0080] This represents a comparison between the k-th sample of the target polyolefin category and the l-th sample of the historical polyolefin category. It is used to assess the relative relationship between the target polyolefin category data and the historical polyolefin category data. This term is true if the predicted value for the target polyolefin category is less than that for the historical polyolefin category.
[0081] This formula complements the previous loss function, further determining the accuracy of the ranking of the target product prediction model across different historical datasets for the target polyolefin category. By comparing the magnitudes of predicted and observed values, the formula effectively optimizes the model's ranking accuracy, making the model's differentiation between different polyolefin samples more reliable, ultimately improving the model's performance in optimizing production parameters. By introducing this loss function, this approach ensures that the predicted results of the target production parameters not only closely approximate the actual values but also accurately reflect the relative superiority or inferiority of product parameters during the production process. This helps to better control quality during production while further improving the accuracy and reliability of the prediction model.
[0082] Based on the loss function constructed in the preceding steps, the loss value of each historical product prediction model and the target product prediction model generalized to the target task can be calculated, thereby obtaining the model similarity between each historical product prediction model and the target product prediction model. Specifically, the smaller the loss value of the model, the higher its similarity to the target task. Therefore, model similarity can be measured by the inverse relationship of loss values, that is, models with smaller loss values have higher similarity.
[0083] In one possible implementation, embodiments of this application can estimate similarity probability weights through sampling, i.e., each time from... A sample dataset is obtained by randomly selecting a portion of the data according to a certain proportion. and replace This is used to measure the loss functions of the historical product prediction model and the target product prediction model, thereby calculating the similarity weight w of each historical product prediction model. i This sampling estimation process will be repeated S times, and the formula for calculating the similarity weight of each model is as follows:
[0084] in, represents the loss value of the i-th historical product prediction model, used to measure the performance of the model on a given dataset, where i represents the i-th model from 1 to t.
[0085] w i The similarity weight represents the prediction model of the i-th historical product. The higher the weight, the more similar the models are.
[0086] S represents the number of times the data was sampled, indicating that the sample was sampled from a subset of the data. Perform multiple samplings;
[0087] This represents the set of models with the smallest loss value; that is, the model with the smallest loss value is selected from all models. This means that if the i-th model belongs to the set of models with the smallest loss value, then the value is 1;
[0088] Represents all that satisfy The models are summed and used as normalization factors.
[0089] Step 2022: Based on the obtained similarity weights, determine the recommended production parameters for the target polyolefin category.
[0090] In this embodiment of the application, after obtaining the similarity weights of each historical product prediction model, the similarity weights will be used to calculate recommended sampling points, thereby obtaining recommended production parameters for the target polyolefin category, so as to optimize the production process of the target polyolefin category.
[0091] In one possible implementation, the embodiments of this application will obtain a recommendation function for the target polyolefin category based on the acquisition function corresponding to each historical production prediction model and in combination with the similarity weights.
[0092] Specifically, using the mean and uncertainty estimates provided by the GP model, a corresponding acquisition function can be designed for each historical product prediction model. The formula for the acquisition function is as follows:
[0093] α EI Measure at a given data point Given the production parameters, this method considers the potential for improvement at parameter x, helping to determine which new production parameters are most likely to lead to significant output increases. In other words, it takes into account existing data points. The corresponding optimal value f(x) * And the likelihood of improvement expected from the new recommended sampling point x. By maximizing α EI This allows us to find the most promising next recommended sampling point, thereby gradually approaching the global optimal solution and obtaining the target production parameter set for the target polyolefin category.
[0094] f(x * () represents the molecular weight distribution and other product quality indicators of polyolefins under the currently known optimal production parameter conditions.
[0095] x represents a newly selected sampling point x from the set of production parameters to be optimized, and f(x) represents the function value of the newly selected sampling point x, that is, the predictive performance of the model at the new sampling point.
[0096] The aforementioned data collection function is based on the Expected Improvement (EI) criterion. It balances exploration and exploitation by favoring uncertain points with higher improvement potential. By avoiding points too close to the current optimal solution, promising regions can be effectively mined, and convergence to the global optimum rather than just a local optimum can be achieved. Thus, by calculating the data collection function, embodiments of this application can estimate the potential gains from exploring new sampling points. If the f(x) of sampling point x is similar to the historical optimum f(x)... * If the difference between the samples is large, it indicates that there may be more room for improvement at that sampling point.
[0097] In one possible implementation, embodiments of this application may further weight and combine the collection functions of multiple historical product prediction models to form the final recommendation function, ensuring that the better-performing model has a greater weight in the final sampling decision. Considering the different data scales involved in historical production tasks and target production tasks, embodiments of this application provide a probability weighting strategy to combine f... 1 ,…,f t The acquisition functions of multiple historical product prediction models are combined using probability weights, and the weighted combination formula is as follows:
[0098] Wherein, α(x) represents the weighted recommendation function, which integrates the collection functions from different models and obtains the final recommendation result through weighting;
[0099] t represents the number of participating models, which is the total number of historical product prediction models and target product prediction models;
[0100] w iThe similarity weight represents the prediction model for the i-th historical product.
[0101] EI i (x) represents the acquisition function of the hybrid historical product prediction model and the target product prediction model, and then calculates the expected improvement of the integration, which is used to measure the expected improvement that the model can bring given input x.
[0102] In one possible implementation, embodiments of this application can further simplify the acquisition function by combining the cumulative distribution function and the probability density function of the standard normal distribution. To make the calculation more convenient, the specific formula is as follows:
[0103] in, The expected value representing standardization is the standardized value of the difference between the mean and the optimum.
[0104] Φ() represents the cumulative probability density function, which is used to measure the potential gains from exploring new sampling points.
[0105] φ() is the probability function of the standard normal distribution, used to measure the potential of a new sampling point in an unexplored region.
[0106] σ(x) represents the standard deviation of the observation point x, which indicates the uncertainty of that point. The larger the standard deviation, the greater the uncertainty of the model's prediction for that point.
[0107] In one possible implementation, after obtaining the recommendation function, this embodiment of the application can obtain the next recommended sampling point with the greatest expected improvement by maximizing the recommendation function, and then use it to solve for the production parameters of the next experiment, i.e., the recommended production parameters for the target polyolefin category. The specific formula is as follows:
[0108] Where, x best This represents the optimal sampling point, i.e., the finally determined recommended production parameters. These parameters, calculated using the similarity between historical and target production prediction models, are those that bring the best improvement to the production process. Choosing this point means maximizing the expected improvement in the target polyolefin category during production.
[0109] X represents the set of all possible production parameters. During the production process, different combinations of production parameters (such as temperature, pressure, raw material ratio, etc.) will affect the quality and production efficiency of polyolefins. The set X contains various combinations of production parameters, and the embodiments of this application need to find the optimal value for the target polyolefin category from it.
[0110] Thus, from all possible sampling points x∈X, select the one that enables the recommendation function. The point x that reaches the maximum value best This point is the next recommended sampling point, which can bring the greatest expected improvement to the production parameters, i.e., the recommended production parameters for the target polyolefin category.
[0111] In one possible implementation, to further improve the efficiency and accuracy of production parameter optimization, this embodiment employs a parallel strategy. Multiple subsets of parameters to be optimized are obtained from the set of parameters to be optimized, and a local search space is constructed for each subset. Transfer Bayes processing is then applied to each local search space to obtain a local product prediction model, which is used to measure the similarity of each model within its local scope. In this way, each local search space only processes a portion of the parameters, reducing the parameter dimensionality of each subset, lowering computational complexity, and simultaneously optimizing parameters in different local regions, increasing the likelihood of finding the global optimum. This effectively avoids the risk of getting trapped in local optima, thus getting closer to or reaching the global optimum.
[0112] Specifically, the scope of the local search space needs to be determined first. If the local search space is limited, the GP model may not be able to explore a sufficiently broad range of solutions, leading to an incomplete understanding of the target task behavior and a decrease in model performance. Furthermore, if the local search space is too large, the GP model may explore irrelevant or data-inefficient regions, resulting in inaccurate similarity comparisons. Therefore, this embodiment will utilize the principle of sampling without replacement to extract data subsets. Wherein, the hyperparameter m represents the proportion of the selected data portion. The ratio, then use The data constructs a GP agent model, and the recommended next sampling point x is obtained by maximizing the EI acquisition function. best,m And the standard Euclidean distance method is used to measure x. best,m arrive The distance is given by the following formula:
[0113] Where d represents the dimension of the parameter space, that is, the number of features considered in the parameter optimization process. Each dimension in the parameter space corresponds to a production parameter that needs to be optimized.
[0114] The production parameter set representing the current sampling point contains the values of this parameter set in each dimension and is a sampling point used for local optimization.
[0115] s j The standard deviation represents the j-th dimension, indicating the range of variation of the production parameters in that dimension within historical data.
[0116] x best,m,jThis represents the value of the optimal point of the m-th historical model in the j-th dimension. It is a relatively optimal solution inferred from the historical model, reflecting the optimal parameters of a certain historical model in that dimension, and is close to x. best,m The data is selected to determine the local search space.
[0117] Thus, this distance allows us to assess the proximity of the current sampled point to the historical optimum, guiding subsequent search directions. A smaller distance indicates the current sampled point is close to a local optimum and may be a better candidate solution, while a larger distance indicates the current sampled point deviates from the local optimum and requires further optimization. The standardization process ensures that values from different dimensions can be compared on the same scale, avoiding excessive or insufficient influence from certain dimensions. By using standardized distance to determine the proximity between the sampled point and the optimum, we can reasonably define the range of the local search space, helping the model quickly find the region closest to the optimum in the high-dimensional feature space, thereby guiding further optimization of production parameters.
[0118] In one possible implementation, to obtain the most compact local search space containing the data... Andθ=(l,u)andθ∈R, this application embodiment also constructs a constrained optimization problem, the formula of which is as follows:
[0119] Two constraints define the boundaries of the local search space: all known sampling points x and the current optimal point xbest must fall within the upper and lower bounds of the local search space, ensuring that the local search space contains all currently valuable information. In the polyolefin synthesis scenario, each component of x represents different production parameters, such as reaction temperature, pressure, and feed rate.
[0120] l and u represent the lower and upper bounds of each production parameter, respectively, defining the acceptable range for each parameter, i.e., the boundary of the local search space. In polyolefin synthesis, l and u may represent the minimum and maximum allowable values of process parameters such as temperature and pressure. These values can be set based on actual production experience or theoretical ranges to ensure that the search space includes a reasonable combination of production parameters. This application does not specifically limit these values.
[0121] θ represents the parameter vector, containing upper and lower bounds l and u, and defines the boundary conditions of the local search space. In practical applications, θ is used to control the local search range of the parameter optimization algorithm. By adjusting θ, the size of the search space can be changed, guiding the algorithm to optimize within the effective range of production parameters.
[0122] In this way, by minimizing the range of the local search space and reducing the distance between the upper and lower bounds as much as possible, we can ensure that the production parameters are optimized within a more precise range and avoid invalid searches.
[0123] Furthermore, the compactness of the local search space is achieved through... This is achieved by squaring the terms, which penalizes the maximum range in each dimension. Therefore, the given optimal solution θ can be easily obtained. * =(l * ,u * ):
[0124] Thus, the upper realm u * and lower bound l * These boundary values are determined by maximizing and minimizing the distances between the sampled points and the optimal point, respectively. These boundary values define the range of the search space, ensuring that the optimization algorithm searches only within the most relevant regions, avoiding unnecessary computation and invalid sampling.
[0125] In one possible implementation, after constructing the local search space, this embodiment of the application will calculate the model similarity between each historical product prediction model and the corresponding local product prediction model for each local search space, and obtain the local similarity weight of each historical product prediction model in the corresponding local search space based on the model similarity, thereby calculating the next recommended sampling point and recommended production parameters for each local search space through the local similarity weight.
[0126] Specifically, embodiments of this application can perform a search in each local search space. The process performs Bayesian Optimization (BO) processing using transfer learning. The BO algorithm is an iterative algorithm for global optimization. In this embodiment, the BO algorithm can be used to determine which parameter point should be tested at each sampling. By estimating the potential reward of the sampling point using the GP model, the BO algorithm continuously updates the model based on these estimates, searching for the global optimum. For example, if a historical yield prediction model m1 has a high similarity weight w1 with the current task, the BO optimization process will rely more on the information provided by this model, exploring more frequently in the local search space. Thus, through calculation... The local similarity weight w between each proxy model and the target task t helps the model share information among similar tasks, thereby accelerating the optimization process. Substituting the local similarity weight w into the recommendation function, the most promising next recommendation sampling point is calculated. And record its recommended value α(x) t+1,m′ If a sampling point x in the current local search space... t+1,m′If a point has high potential, the BO optimization algorithm will recommend it as a candidate point for the next trial and record the recommended value. In the process of polyolefin synthesis, this means selecting sampling points that maximize production efficiency or quality. Thus, by optimizing parameters in the local search space, the model explores within a subspace close to the target task, reducing invalid data sampling and making each sampling more likely to approach the optimal solution, thereby improving the accuracy and efficiency of the optimization process.
[0127] Step 2023: Based on the recommended production parameters, update the set of parameters to be optimized, and based on the updated set of parameters to be optimized, perform the next parameter optimization process.
[0128] In this embodiment of the application, after each parameter optimization process to obtain the recommended production parameters for the target polyolefin category, the set of parameters to be optimized for the target polyolefin category will be updated, and the next parameter optimization process will be iterated to improve the posterior probability density of the model and improve the model's ability to describe the polyolefin polymerization process, until the set of parameters to be optimized meets the production conditions of the target category and the target production parameter set is obtained.
[0129] Specifically, in this embodiment, the sampling point with the highest recommended value can be selected from all local search spaces as the optimal recommended sampling point for actual experiments. The contribution to the performance of the target grade polyolefin is verified through experiments, and the experimental results are updated to the target grade's dataset. The parameter optimization process is then repeated until the ideal production parameters that satisfy the target grade polyolefin are found.
[0130] Please refer to Figure 4, which is a schematic diagram of a polyolefin production parameter optimization process provided in this embodiment of the application. In this parameter optimization process, this embodiment of the application collects data samples from multiple historical grades, such as "Historical Grade 1 Data," "Historical Grade 2 Data," and so on, up to "Historical Grade m Data" in Figure 4. These data contain historical data related to different historical production parameters and product quality indicators. The data for each historical grade is used to construct a corresponding historical GP model. Each historical GP model reflects the relationship between the historical production parameters and the molecular weight distribution of the product for that grade, preparing for subsequent transfer learning. Simultaneously, known sample data for the target grade is collected. Initially, the target grade data may be relatively limited, with only a portion of the sample data used for modeling. Similar to the historical grades, a target GP model for the target grade is constructed based on the partial sample data to predict the performance of new data points. For each historical grade's GP model, a local search space is constructed, such as "Local Search Space 1," "Local Search Space 2," etc., as shown in Figure 4. The local search space is the scope for evaluating and exploring the local similarity of each historical grade model on the target grade data. The generalization ability of historical model pools is evaluated using target model data to determine the similarity between historical GP models and the target model. This step involves ranking and evaluating the target model data using historical GP models to determine which historical models are best suited for the target model optimization task. Next, a sampling method is used to randomly sample a portion of the target model data proportionally, constructing a loss function to measure the accuracy of the historical models in ranking the target observations. Each calculation assigns a similarity weight to each historical model. Through repeated sampling and evaluation, the similarity weight of each historical GP model is finally determined. A higher weight indicates a greater similarity between the historical model and the target model task. Based on the similarity weights of each historical model, an ensemble sampling function is used to calculate the recommended sampling point for the next step. This recommended point represents the production parameter most likely to improve optimization efficiency. Simultaneously, a parallel strategy is employed to perform Bayesian optimization in different local search spaces. Parallel processing accelerates the finding of the optimal sampling point and reduces reliance on a single historical model. The recommended sampling points obtained from multiple local search spaces are sorted, and the sampling point with the largest recommended value is selected as the optimal point for actual experiments. After the actual experiments are completed, the actual experimental data is acquired and fed back into the model to update the target grade dataset. The previous optimization steps are repeated (as shown in Figure 4, the iterative process from "building a Gaussian model" to "determining recommended sampling locations and recommended values"), continuously adjusting and optimizing the process parameters until the combination of production parameters that best meets the production requirements of the target grade is found, ensuring that the production unit can stably produce compliant polyolefin products under the new production parameters.
[0131] In one possible implementation, using 100 data points uniformly distributed on the molecular weight distribution curve as the target task, this embodiment of the application, based on the optimization method for deleting polyolefin production parameters, can calculate the values of production parameters that conform to the target molecular weight distribution, such as hydrogen feed rate, ethylene and propylene feed rate, and catalyst and co-catalyst feed rate. Figure 5 shows a schematic diagram of the simulation optimization results of production parameters for a multi-stage polymerization process of propylene provided by this embodiment of the application. With 10 repeated experiments, Figure 5 shows the average loss and 95% confidence interval for each iteration step. As can be seen from Figure 5, the transfer learning BO algorithm provided by this embodiment of the application has good optimization performance, especially in the early stages of iteration where it can quickly obtain a small loss value. This indicates that, especially in the early stages of iteration, this embodiment of the application can accelerate the optimization solution through the information provided by the source task.
[0132] Please refer to Figure 6. Based on the same inventive concept, this application also provides a polyolefin production parameter optimization device 60, which includes:
[0133] Construction unit 601 is used to construct corresponding historical product prediction models for each of the historical production parameter sets corresponding to multiple historical polyolefin categories; each historical production parameter set includes multiple historical production parameters and molecular weight distribution of polyolefin products for the corresponding historical polyolefin category.
[0134] The optimization unit 602 is used to iteratively perform parameter optimization processing on the set of parameters to be optimized for the target polyolefin category based on the prediction models of each historical product until the set of parameters to be optimized meets the production conditions of the target category, thereby obtaining the target production parameter set.
[0135] Optionally, the optimization unit 602 is specifically used for:
[0136] Based on the prediction models of each historical product, a prediction comparison process is performed on the parameter set to be optimized to obtain the similarity weight of each historical product prediction model.
[0137] Based on the obtained similarity weights, recommended production parameters for the target polyolefin category are determined.
[0138] Based on the recommended production parameters, update the set of parameters to be optimized, and then perform the next parameter optimization process based on the updated set of parameters to be optimized.
[0139] Optionally, optimization unit 602 is specifically used for:
[0140] Based on the set of parameters to be optimized, construct the corresponding target product prediction model;
[0141] Based on the model similarity between each historical product prediction model and the target product prediction model, the similarity weight of each historical product prediction model is obtained.
[0142] Optionally, optimization unit 602 is specifically used for:
[0143] Based on the acquisition functions corresponding to each historical production prediction model, and combined with the similarity weights, a recommendation function for the target polyolefin category is obtained.
[0144] Recommended production parameters are obtained based on the recommendation function.
[0145] Optionally, if the target product prediction model includes multiple local product prediction models in the local search space, then the optimization unit 602 is further used for:
[0146] From the set of parameters to be optimized, obtain multiple subsets of parameters to be optimized, and determine the local search space corresponding to each subset of parameters to be optimized;
[0147] Transfer Bayes processing is performed on each local search space to obtain the local product prediction model corresponding to each local search space.
[0148] Optionally, the optimization unit 602 is also used for:
[0149] For each local search space, calculate the model similarity between each historical product prediction model and the corresponding local product prediction model;
[0150] Based on the similarity of each model, the local similarity weights of each historical product prediction model are obtained; the local similarity weights represent the local similarity weights of the historical product prediction models in the corresponding local search space.
[0151] For ease of description, the above sections are divided into functional units (or modules) and described separately. Of course, in implementing this application, the functions of each unit (or module) can be implemented in one or more software or hardware components. Those skilled in the art will understand that various aspects of this application can be implemented as systems, methods, or program products. Therefore, various aspects of this application can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, collectively referred to herein as "circuit," "module," or "system."
[0152] This device can be used to execute the methods shown in the various embodiments of this application. Therefore, the functions that each functional module of this device can achieve can be referred to the description of the foregoing embodiments, and will not be repeated here.
[0153] Please refer to Figure 7. Based on the same technical concept, this application embodiment also provides a computer device 70, which can be the master system or slave system shown in Figure 1. The computer device 70 may include a memory 701 and a processor 702.
[0154] The memory 701 is used to store computer programs executed by the processor 702. The memory 701 mainly includes a program storage area and a data storage area. The program storage area can store the operating system, application programs required for at least one function, etc.; the data storage area can store data created based on the use of the computer device, etc. The processor 702 can be a central processing unit (CPU), a digital processing unit, etc. This application embodiment does not limit the specific connection medium between the memory 701 and the processor 702. In this application embodiment, the memory 701 and the processor 702 are connected via a bus 703 in Figure 4. The bus 703 is represented by a thick line in Figure 4. The connection methods between other components are only illustrative and not intended to be limiting. The bus 703 can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is used in Figure 4, but this does not indicate that there is only one bus or one type of bus.
[0155] Memory 701 may be volatile memory, such as random-access memory (RAM); memory 701 may also be non-volatile memory, such as read-only memory, flash memory, hard disk drive (HDD), or solid-state drive (SSD); or memory 701 may be any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. Memory 701 may be a combination of the above-described memories.
[0156] The processor 702 is used to execute the polyolefin production parameter optimization method performed by the device in the various embodiments of this application when calling the computer program stored in the so-called memory 701.
[0157] In some possible implementations, various aspects of the polyolefin production parameter optimization method provided in this application can also be implemented in the form of a program product, which includes program code. When the program product is run on a computer device, the program code is used to cause the computer device to perform the steps in the polyolefin production parameter optimization method according to various exemplary embodiments of this application described above. For example, the computer device can perform the steps of the various embodiments.
[0158] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.
[0159] The program product of the embodiments of this application may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a computing device. However, the program product of this application is not limited thereto. In this application, the readable storage medium may be any tangible medium that contains or stores a program that may be used by or in conjunction with a command execution system, apparatus, or device.
[0160] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying readable program code. This propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting a program for use by or in conjunction with a command execution system, apparatus, or device.
[0161] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.
[0162] Program code for performing the operations of this application can be written in any combination of one or more programming languages, including object-oriented programming languages such as Java and C++, and conventional procedural programming languages such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).
[0163] It should be noted that although several units or sub-units of the device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units.
[0164] Furthermore, although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.
[0165] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0166] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.
[0167] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.
Claims
1. A method for optimizing polyolefin production parameters, characterized in that, The method includes: For each of the historical production parameter sets corresponding to multiple historical polyolefin categories, a corresponding historical product prediction model is constructed. Each historical production parameter set includes multiple historical production parameters and the molecular weight distribution of polyolefin products for the corresponding historical polyolefin category. Based on the prediction models of each historical product, the parameter set to be optimized for the target polyolefin category is iteratively optimized until the parameter set to be optimized meets the production conditions of the target category, thus obtaining the target production parameter set.
2. The method as described in claim 1, characterized in that, Each parameter optimization process includes: Based on the aforementioned historical product prediction models, the set of parameters to be optimized is subjected to prediction comparison processing to obtain the similarity weights of each historical product prediction model. Based on the obtained similarity weights, recommended production parameters for the target polyolefin category are determined; Based on the recommended production parameters, the set of parameters to be optimized is updated, and based on the updated set of parameters to be optimized, the next parameter optimization process is performed.
3. The method as described in claim 2, characterized in that, The step of performing prediction comparison processing on the parameter set to be optimized based on each historical product prediction model to obtain the similarity weights of each historical product prediction model includes: Based on the set of parameters to be optimized, a corresponding target product prediction model is constructed. Based on the model similarity between each historical product prediction model and the target product prediction model, the similarity weight of each historical product prediction model is obtained.
4. The method as described in claim 2, characterized in that, The process of determining recommended production parameters for the target polyolefin category based on the obtained similarity weights includes: Based on the acquisition function corresponding to each historical production prediction model, and combined with the similarity weights, the recommendation function for the target polyolefin category is obtained. Based on the recommendation function, the recommended production parameters are obtained.
5. The method as described in claim 2, characterized in that, The target product prediction model includes multiple local product prediction models under local search spaces. Therefore, the step of constructing the corresponding target product prediction model based on the set of parameters to be optimized further includes: From the set of parameters to be optimized, obtain multiple subsets of parameters to be optimized, and determine the local search space corresponding to each subset of parameters to be optimized; Transfer Bayes processing is performed on each local search space to obtain the local product prediction model corresponding to each local search space.
6. The method as described in claim 5, characterized in that, The method of obtaining the similarity weights of each historical product prediction model based on the model similarity between each historical product prediction model and the target product prediction model also includes: For each local search space, calculate the model similarity between each historical product prediction model and the corresponding local product prediction model; Based on the similarity of each model, the local similarity weights of each historical product prediction model are obtained; the local similarity weights represent the local similarity weights of the historical product prediction models in the corresponding local search space.
7. A device for optimizing polyolefin production parameters, characterized in that, The device includes: The building unit is used to construct corresponding historical product prediction models for each of the historical production parameter sets corresponding to multiple historical polyolefin categories; each historical production parameter set includes multiple historical production parameters and molecular weight distribution of polyolefin products for the corresponding historical polyolefin category. The optimization unit is used to iteratively perform parameter optimization processing on the set of parameters to be optimized for the target polyolefin category based on each historical product prediction model, until the set of parameters to be optimized meets the production conditions of the target category, thereby obtaining the target production parameter set.
8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.
9. A computer storage medium storing computer program instructions thereon, characterized in that, When executed by a processor, the computer program instructions implement the steps of the method according to any one of claims 1 to 6.
10. A computer program product comprising computer program instructions, characterized in that, When executed by a processor, the computer program instructions implement the steps of the method according to any one of claims 1 to 6.