Prediction method, system and storage medium for ldpe high pressure tubular polymerization process

Abstract

This application discloses a prediction method, system, and storage medium for LDPE high-pressure tubular polymerization process, relating to the field of process industry production process simulation and optimization technology. The method includes: acquiring first process parameters and target product indicators for the LDPE high-pressure tubular polymerization process to be tested; inputting the first process parameters into a trained machine learning model to obtain a first predicted product indicator; constructing an objective function based on the first predicted product indicator and the target product indicator, and solving for the target process parameters that minimize the function value of the objective function. With this technical solution, users can optimize the current production process based on the solved target process parameters, ensuring that the product indicators meet actual industrial needs. This method of solving process parameters from the target product indicator overcomes the functional shortcomings of existing proxy models, and it eliminates the need for manual trial and error, resulting in higher efficiency through machine optimization and derived process indicators that are closer to user needs.

Description

Technical Field

[0001] This application relates to the field of process industry production process simulation and optimization technology, specifically to a prediction method, system and storage medium for LDPE high-pressure tubular polymerization process. Background Technology

[0002] Low-density polyethylene (LDPE) is a core variety of general-purpose synthetic resin with large production volume and wide application, widely used in packaging, building materials, electronics, pharmaceuticals and other fields. Among them, high-pressure tubular polymerization is the mainstream process for LDPE production, accounting for more than 70% of global LDPE production capacity. Its process characteristics include high reaction pressure and reaction temperature, strong nonlinearity, and multi-parameter coupling.

[0003] Aspen Plus is a large-scale general-purpose chemical process simulation platform for calculating process mechanisms, which can be used for mechanism modeling of low-density polyethylene (LDPE) production processes. With the development of artificial intelligence technology, surrogate models built through machine learning can also simulate and generate product indicator data based on process parameters, and are faster and more efficient than Aspen Plus modeling. However, existing machine learning surrogate models still remain at the level of one-way prediction of product indicators from process parameters. Summary of the Invention

[0004] The purpose of this application is to provide a method, system, and storage medium for predicting LDPE high-pressure tubular polymerization processes.

[0005] To achieve the above objectives, the first aspect of this application provides a method for predicting the high-pressure tubular polymerization process of LDPE, comprising: Obtain the first process parameters and target product indicators of the LDPE high-pressure tubular polymerization process to be tested. Both the first process parameters and the target product indicators are given by the user. Input the first process parameters into the trained machine learning model to obtain the first predicted product index. Based on the first predicted product indicator and the target product indicator, an objective function is constructed, defined as follows:

[0006] Where i is a positive integer. For the i-th target product indicator, Let be the i-th type of first-predictive product indicator, and n be the number of target product indicator types. Let be the weight corresponding to the indicator of the i-th target product; Find the target process parameters that minimize the value of the objective function; The method also includes a training step for the machine learning model, which includes: Establish a mechanistic model for the high-pressure tubular polymerization process of LDPE; Sample process parameters are generated by mechanistic model simulation, and Latin hypercube sampling is used to sample the sample process parameters to generate sample data that are uniformly distributed throughout the entire range. Input the sample data into the mechanism model to obtain the sample predicted product indicators; A modeling dataset is generated based on sample data and sample prediction product indicators, and a machine learning model is trained based on the modeling dataset. Among them, the target process parameters that minimize the function value of the objective function are solved, including: A probabilistic model of the objective function is constructed based on a tree-structured Parsons estimator, and the preset number of iterations of the probabilistic model, as well as the feasible region and initial perturbation range of the first process parameter, are obtained. The first process parameters, feasible region, and initial disturbance range are input into the probabilistic model to obtain the optimized first process parameters. The objective function value is calculated based on the optimized first process parameters and target product indicators, and the probability model is updated based on the function value. Repeat the above iterative steps. When the preset number of iterations is reached, the first process parameter that minimizes the function value will be taken as the target process parameter.

[0007] In one embodiment, the method further includes: when the mechanism model generates sample predicted product indicators in groups, recording the task number of the generated sample predicted product indicators; when the task number reaches a first preset task quantity, saving the generated sample predicted product indicators; when the task number meets a second preset task quantity, restarting the mechanism model, and skipping the generated sample predicted product indicators based on the task number recorded before restarting and continuing to execute the generation task, wherein the first preset task quantity is less than the second preset task quantity.

[0008] In one embodiment, the method further includes: inputting a second process parameter into a trained machine learning model to obtain a second predicted product indicator.

[0009] In one embodiment, the machine learning model includes at least one of six models: Ridge Regression, Random Forest, Gradient Boosting Decision Tree, Extreme Gradient Boosting, Lightweight Gradient Boosting Machine, and Multilayer Perceptron.

[0010] In one embodiment, the target process parameters are input into a machine learning model to obtain a third predicted product index; if the relative deviation between the third predicted product index and the target product index is less than a preset deviation, the target process parameters are output.

[0011] A second aspect of this application provides a prediction system for LDPE high-pressure tubular polymerization process, comprising: The parameter acquisition module is used to acquire the first process parameters and target product indicators of the LDPE high-pressure tubular polymerization process to be tested; the forward prediction module is used to input the first process parameters into the trained machine learning model to obtain the first predicted product indicator; it is also used to input the second process parameters into the trained machine learning model to obtain the second predicted product indicator; the reverse solution module is used to construct an objective function based on the first predicted product indicator and the target product indicator, and solve for the target process parameters that minimize the function value of the objective function.

[0012] In one embodiment, the system further includes a visualization interaction module, which is used to invoke the parameter acquisition module, the forward prediction module, and the reverse solving module in response to user actions.

[0013] A third aspect of this application provides a machine-readable storage medium storing instructions that, when executed by a processor, configure the processor to perform the aforementioned prediction method for the LDPE high-pressure tubular polymerization process.

[0014] The above technical solution involves inputting the user's current process parameters into a trained machine learning model to obtain predicted product indicators. Based on the target product indicators and the predicted product indicators, a target function is constructed, and the target process parameters that minimize the function value are calculated. Users can then optimize their current production process using these calculated target process parameters to ensure that product indicators meet actual industrial needs. This method of solving process parameters from target product indicators overcomes the limitations of existing predictive models. Furthermore, this approach eliminates the need for manual trial and error, making machine optimization more efficient and resulting in process indicators that more closely align with user requirements.

[0015] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the following detailed description to explain the embodiments of this application, but do not constitute a limitation on the embodiments of this application. In the drawings: Figure 1 A schematic diagram of a prediction method for a high-pressure tubular polymerization process of LDPE according to an embodiment of this application is shown. Figure 2 This schematic diagram illustrates the internal structure of a prediction system for a high-pressure tubular polymerization process of LDPE according to an embodiment of this application. Figure 3 This illustration shows a schematic diagram of the types of process parameters and product indicators according to embodiments of this application; Figure 4The schematic diagram illustrates test results for six models according to embodiments of this application; Figure 5 The diagram illustrates the internal structure of a computer device according to an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] Figure 1 A schematic flowchart illustrating a prediction method for a high-pressure tubular polymerization process of LDPE according to an embodiment of this application is shown. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise explicitly stated herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps. For example Figure 1 As shown in one embodiment of this application, a method for predicting the high-pressure tubular polymerization process of LDPE is provided, comprising the following steps: Step 100: Obtain the first process parameters and target product indicators of the LDPE high-pressure tubular polymerization process to be tested.

[0019] The first process parameter is the process parameter that needs to be optimized during the model application phase, and it is given by the user. The target product index is the product index that the user expects to obtain, and it is also given by the user.

[0020] For example, a complete parameter system of 23 primary process parameters and 8 target product indicators is selected for the entire process. The types of primary process parameters and target product indicators are described in [link to relevant documentation]. Figure 3 .

[0021] The above 23 process parameters are derived from the entire LDPE high-pressure tubular polymerization process and are key influencing variables of this process. The 8 product indicators represent key quality attributes that need to be controlled and predicted in industrial production and are the core performance indicators of LDPE products. It should be noted that the combinations of the above parameters and indicators are only examples. In practical applications, users can add, subtract, or combine selected process parameters or product indicators according to their needs; no restrictions are imposed here.

[0022] Step 200: Input the first process parameters into the trained machine learning model to obtain the first predicted product index.

[0023] The trained machine learning model serves as a surrogate model for the LDPE high-pressure tubular polymerization process. It can establish a mapping relationship between process parameters and product indicators based on deep learning algorithms, enabling intelligent prediction of product indicators. While this surrogate model used in this solution is not as accurate as Aspen Plus in simulation, it boasts faster computation speed, higher simulation efficiency, and supports large-scale global optimization, enabling reverse engineering of process parameters based on target performance.

[0024] Optionally, the machine learning model is at least one of the following six models: Ridge Regression, Random Forest, Gradient Boosting Decision Tree, Extreme Gradient Boosting, Lightweight Gradient Boosting Machine, and Multilayer Perceptron.

[0025] For example, the first process parameter is input into the multilayer perceptron model to obtain the first predicted product index. The multilayer perceptron model is better at capturing complex nonlinear mapping relationships in industrial production, and its data fitting accuracy and generalization ability are significantly better than the other five models.

[0026] Step 300: Construct an objective function based on the first predicted product index and the target product index, and solve for the target process parameters that minimize the function value of the objective function.

[0027] The objective function is defined as the weighted sum of squares of the differences between the target product indicator and the first predicted product indicator. The specific formula is as follows:

[0028] Where i is a positive integer. For target product indicators, The first predictive indicator is , and n is the number of target product indicators. For example, if 8 target product indicators are selected, n is 8. The weights for each target product indicator can be flexibly set by the user according to industrial needs. In addition, to ensure that the solution results conform to actual working conditions, constraints on the objective function can be set, such as setting the physical value range of 23 process parameters.

[0029] Optionally, global optimization algorithms such as genetic algorithms and particle swarm optimization can be used to solve the objective function.

[0030] Optionally, a tree-structured Parsons estimator algorithm is used to solve the objective function. Starting with the user's current process parameters and the corresponding function values, a probabilistic model of the objective function is constructed based on the tree-structured Parsons estimator, and the target process parameter that minimizes the objective function value is iteratively searched.

[0031] The tree-structured Parsons estimator algorithm significantly outperforms genetic algorithms and particle swarm optimization algorithms in hyperparameter tuning. Furthermore, it is simpler to model, faster in computation, and more suitable for industrial applications. Therefore, employing the tree-structured Parsons estimator algorithm can more quickly search for target process parameters that meet user needs, improving the efficiency of reverse prediction of process parameters.

[0032] The above technical solution inputs the user's current primary process parameters into a proxy model built on a machine learning model to obtain the corresponding predicted product indicators. Based on the target product indicators and the predicted product indicators, a target function is constructed, and the target process parameters that minimize the function value are solved. This allows the user to optimize the current production process based on these target process parameters, ensuring that the product indicators meet actual industrial needs. This method of solving process parameters from the target product indicators overcomes the functional limitations of existing proxy models. Furthermore, this method eliminates the need for manual trial and error; machine optimization not only improves efficiency but also ensures that the derived process indicators are closer to user requirements.

[0033] In one embodiment, the method further includes a machine learning model training step, which includes: establishing a mechanistic model of the LDPE high-pressure tubular polymerization process; simulating and generating sample process parameters through the mechanistic model, and sampling the sample process parameters using Latin hypercube sampling to generate uniformly distributed sample data across the entire range; inputting the sample data into the mechanistic model to obtain sample predicted product indicators; generating a modeling dataset based on the sample data and sample predicted product indicators, and training the machine learning model based on the modeling dataset.

[0034] The sample process parameters are the process parameters to be optimized used in the model training phase. The mechanistic model is a steady-state model based on Aspen Plus, including ethylene feed, dual initiator injection, four-stage water cooling, four-stage tubular reactor, product separation, and recirculating gas recovery. It can effectively simulate the entire process of LDPE high-pressure tubular polymerization. Aspen Plus can be controlled using Python's COM interface to simulate and generate sample process parameters and sample predicted product indicators.

[0035] For example, establishing a mechanistic model includes the following steps: First, 23 primary process parameter types and 8 target product index types were selected for the mechanistic model. Second, the polymer system property methods for the mechanistic model were set. The mechanistic model uses POLYMER (Polymer Property Method Screening Filter) and POLYSL (Polymer Light Segment Filter) in Aspen Plus as the basic property methods to calculate the property parameters, phase equilibrium, enthalpy, density, and polymer-related properties in the polymer system. STEAM-TA (Steam Surface Property Method), suitable for water systems, was used on the cooling water side. Third, the free radical polymerization reaction set R1 for the mechanistic model was established. Reaction set R1 uses the FREE-RAD (Free Radical Polymerization Reaction Model), with LDPE set as the target polymer, E2 (ethylene) set as the monomer, E2-SEG (Reactor Segmented Feed Ethylene) set as the repeating segment entering the polymer chain, and INI1 (benzoyl peroxide, BPO) and INI2 (di-tert-butyl peroxide, DTBP) set as standard initiators. Reaction set R1 includes reaction steps such as initiator decomposition, chain initiation, chain growth, transfer to monomer chains, transfer to polymer chains, β-cleavage, disproportionation termination, coupling termination, and short-branch generation. Each reaction step is kinetically calculated using pre-factor, activation energy, activation volume, reference temperature, initiator efficiency, and relevant fractional parameters to describe the process of free radical polymerization of ethylene under high pressure to generate LDPE. Fourth, a mechanistic model of the LDPE high-pressure tubular polymerization process flow is established. The main ethylene feed stream is divided into two ethylene branches by a splitter; the first ethylene branch, after preheating by the first heater, enters the first mixer and mixes with the first initiator (1# initiator) to form the first reaction feed; the first reaction feed sequentially enters the first tubular reactor and the second tubular reactor, reacting to form the first reaction effluent. The second ethylene branch, after preheating by the second heater, mixes with the second initiator (2# initiator) and the first reaction effluent in the second mixer to form the second reaction feed; the second reaction feed sequentially enters the third tubular reactor and the fourth tubular reactor. All four tubular reactors utilize the free radical polymerization reaction set R1. The first and third tubular reactors are 250m long, while the second and fourth are 220m long, with a diameter of 0.059m. Both the process flow and the thermal fluid flow are set to liquid phase. Fifth, a four-stage water cooling unit for the mechanistic model is established. The inlet and outlet streams of the cooling water (cooling water #1, #2, #3, and #4) from the first to the fourth stage are coupled to the first, second, third, and fourth tubular reactors, respectively. The inlet stream flow rate, temperature, and pressure of each cooling water stage are pre-set to simulate the thermal coupling between the exothermic reaction of high-pressure ethylene polymerization and the staged water cooling. Sixth, a product separation unit for the mechanistic model is established.The effluent from the fourth tubular reactor enters a high-pressure separator, which operates at 250 bar with a heat load of 0 W. The bottom stream from the high-pressure separator enters a low-pressure separator, which operates at 1 bar with a heat load of 0 W. The bottom stream separated by the low-pressure separator is used as the product stream, and the top stream separated is used as the light component recovery stream. The mechanistic model provides the specific product indicators for this reaction based on the defined product indicator types.

[0036] Specifically, a mechanistic model is established and controlled to simulate and generate sample process parameters. Latin hypercube sampling is then used to sample these parameters, generating multiple sets of sample data uniformly distributed across the entire range. The sample data is then grouped and input into the mechanistic model to simulate and generate the predicted product indicators for each set of sample data. All sample data and their corresponding predicted product indicators are aggregated to obtain a modeling dataset, which is then used to train the machine learning model. Specifically, the sample data in the modeling dataset is used as input to the machine learning model, and the predicted product indicators in the modeling dataset are used as labels for the machine learning model. This allows the machine learning model to learn the mapping relationship between the sample data and the predicted product indicators, ultimately resulting in a trained machine learning model.

[0037] For example, the sample process parameters are generated by controlling Aspen Plus through the COM interface of Python. 8500 sets of sample data are obtained by sampling. These sample data are grouped and input into the mechanism model. The corresponding sample prediction product indicators are generated by the interface control method. Finally, the 8500 sets of sample data and the corresponding sample prediction product indicators are summarized to obtain the modeling dataset.

[0038] Understandably, the above solution uses Python to control Aspen Plus to simulate and generate 8,500 sets of sample process parameters and sample predicted product indicators. The entire generation process does not require manual intervention, and the data integrity rate meets the needs of practical applications.

[0039] In addition, before formal training, a standard scaler can be used to preprocess the modeling dataset to eliminate differences in the dimensions of different parameters. Then, the dataset is divided into training and test sets according to a preset ratio. The training set is used for model training, and the test set is used to verify generalization ability. Data preprocessing can also be accomplished through minimax normalization. Because the standard scaler is less sensitive to outliers in the data, it will not cause serious bias in the entire dataset due to individual data anomalies, making this data processing method more suitable for industrial applications.

[0040] In one embodiment, the method further includes: when the mechanism model generates sample predicted product indicators in groups, recording the task number of the generated sample predicted product indicators; when the task number reaches a first preset task quantity, saving the generated sample predicted product indicators; when the task number meets a second preset task quantity, restarting the mechanism model, and skipping the generated sample predicted product indicators based on the task number recorded before restarting and continuing to execute the generation task, wherein the first preset task quantity is less than the second preset task quantity.

[0041] The first preset task quantity can be set to generate 5 sets of samples to predict product indicators, and the second preset task quantity can be set to generate 50 sets of samples to predict product indicators.

[0042] Specifically, a Python script controls the mechanistic model to generate sample predicted product indicators in groups and records the task numbers of the generated sample predicted product indicators. After every 5 groups of generation tasks are completed, the generated sample predicted product indicators are automatically saved, such as automatically writing them to a temporary CSV file to prevent data loss due to unexpected program crashes. After every 50 groups of sample predicted product indicators are generated, the mechanistic model is actively shut down and restarted to release the memory generated by Aspen Plus running for a long time. After restarting, the saved temporary CSV file is loaded, the completed task numbers are read, and the completed tasks are skipped before continuing to execute the generation tasks.

[0043] Understandably, the above steps establish automatic saving, automatic restart, and resume functionality for the mechanistic model. After an abnormal simulation interruption, the mechanistic model can avoid data loss by saving in batches and resume execution based on the recorded task number. If generated data cannot be saved in a timely manner, in the event of an Aspen Plus crash due to extreme conditions, the entire simulation process must start from scratch, making fully automated, uninterrupted execution impossible. Solving the data loss problem allows the simulation process to be completely independent of human intervention, reduces the probability of program crashes, and ensures the stability of long-term batch simulation execution.

[0044] In one embodiment, the second process parameters are input into a trained machine learning model to obtain a second predicted product index.

[0045] The second process parameter is the process parameter input by the user, and the second predicted product index is the product index obtained based on the process parameter input by the user. The second process parameter has the same parameter type as the first process parameter, and the second predicted product index has the same index type as the first predicted product index.

[0046] It is understood that this embodiment provides a positive prediction method. The trained machine learning model can effectively simulate the mapping relationship between process parameters and product indicators, and outputs the corresponding second predicted product indicator after the user inputs the second process parameter.

[0047] Compared to the mechanistic model built by Aspen Plus, the LDPE high-pressure tubular polymerization process proxy model built based on a machine learning model has a faster computation speed, and its simulation accuracy can be improved with the increase of learning intensity. In addition to predicting process parameters from product indicators, this embodiment also provides a forward prediction method for predicting product indicators from process parameters, which can meet more user needs.

[0048] In one embodiment, the machine learning model includes at least one of six models: Ridge Regression, Random Forest, Gradient Boosting Decision Tree, Extreme Gradient Boosting, Lightweight Gradient Boosting Machine, and Multilayer Perceptron.

[0049] Optionally, the machine learning model is an extreme gradient boosting model. This model is based on the gradient boosting framework, uses a squared loss function, iteratively constructs decision trees to fit the residuals, and determines the optimal configuration through a 5-fold grid search. Specifically, the number of decision trees is 200, the maximum depth is 6, the learning rate is 0.07, the subsampling ratio is 0.8, the column sampling ratio is 0.8, and the default regularization parameters (lambda=1, alpha=0) are used.

[0050] Optionally, the machine learning model is a multilayer perceptron. This model is designed with 23 neurons in the input layer, 8 neurons in the output layer, and 2 hidden layers with 256 and 128 neurons respectively. The hidden layers use the ReLU activation function, and the output layer uses a linear activation function, connected in a fully connected manner. Training uses the Adam optimizer with an initial learning rate of 0.001, an L2 regularization coefficient of 0.01, and an early stopping mechanism specifically enabled to capture the strong nonlinear relationships in the LDPE polymerization process.

[0051] Three metrics were used to evaluate the six models: coefficient of determination, root mean square error (RMSE), and mean absolute error (MAE). The coefficient of determination measures the goodness of fit, while the RMSE and MAE measure the prediction error. 6800 sets of sample data and predicted product indicators were randomly selected from the modeling dataset generated by the mechanistic model as the training set, and each of the six models was trained. From the trained models, the remaining 1700 sets of sample data and predicted product indicators were used as the test set to test each of the six models. The test results are shown below. Figure 4 .

[0052] The test results show that the multilayer perceptron model performs best. The LDPE high-pressure tubular polymerization process is characterized by strong nonlinearity and strong coupling of multiple parameters, and the multilayer perceptron model is better at capturing these complex nonlinear mapping relationships, significantly outperforming the other five models in terms of data fitting accuracy and generalization ability.

[0053] In one embodiment, finding the target process parameter that minimizes the function value of the objective function includes: constructing a probabilistic model of the objective function based on a tree-structured Parsons estimator, and obtaining a preset number of iterations for the probabilistic model, as well as the feasible region and initial perturbation range of the first process parameter; inputting the first process parameter, feasible region, and initial perturbation range into the probabilistic model to obtain the optimized first process parameter; calculating the function value of the objective function based on the optimized first process parameter and the target product index, and updating the probabilistic model based on the function value; repeating the above iterative steps, and when the preset number of iterations is reached, using the first process parameter that minimizes the function value as the target process parameter.

[0054] The Tree-structured Parsons Estimator (TPE) is an intelligent hyperparameter optimization algorithm based on Bayesian optimization. It utilizes historical parameter samples and corresponding target results, constructing a probabilistic model through Parsons kernel density estimation. This model divides existing parameter samples into high-probability and low-probability regions. Based on the density distribution relationship between the two types of samples, it recommends the parameter combination most likely to achieve the optimal result in the next round of iterative evaluation. The feasible region is the physical value range of the first process parameter. The preset number of iterations can be set according to actual conditions, such as 150 iterations. The initial disturbance range is limited to ±10% of the corresponding industrial benchmark value of the parameter, and does not exceed the upper and lower limits of the parameter's own physical value range.

[0055] Specifically, a probabilistic model of the objective function is constructed based on a tree-structured Parsons estimator, and the preset number of iterations, the feasible region of the first process parameter, and the initial perturbation range of the probabilistic model are obtained. During the first iteration of the tree-structured Parsons estimator, the first process parameter is optimized based on the process parameters recommended by the probabilistic model for this iteration. The optimized first process parameter is then input into the machine learning model to update the first predicted product index accordingly. The updated first predicted product index is then used to calculate the function value of the objective function.

[0056] In the next iteration, the probabilistic model can divide the feasible region of the process parameters into high-probability and low-probability regions based on the function value calculated in the previous iteration. It can continue to search for possible process parameters in the high-probability region and optimize the first process parameter again. The optimized first process parameter is then input into the machine learning model to update the first predicted product index. The updated first predicted product index is then substituted into the objective function to calculate the function value again.

[0057] Repeating the above iterative steps allows the probabilistic model to efficiently search for optimal process parameters. After reaching the preset number of iterations, the first process parameter that minimizes the function value is selected as the target process parameter.

[0058] A tree-structured Parsons estimator is used for parameter optimization, avoiding the tedious process of manually inputting data for testing and optimization. This algorithm can automatically search for the global optimum without easily getting trapped in local optima, and it solves the problem quickly, thus significantly improving the efficiency of finding target process parameters.

[0059] In one embodiment, the method further includes: inputting the target process parameters into a machine learning model to obtain a third predicted product index; and outputting the target process parameters if the relative deviation between the third predicted product index and the target product index is less than a preset deviation.

[0060] The third predicted product indicator verifies the target process parameters and has the same indicator type as the first predicted product indicator. The preset deviation can be set by the user; for example, the preset deviation can be set to 0.5%.

[0061] The target process parameters are input into the machine learning model to obtain the third predicted product index, which can be used to evaluate whether the obtained target process parameters meet user requirements. If the relative deviation between the third predicted product index and the target product index is less than 0.5%, the obtained target process parameters are output.

[0062] In one embodiment, a prediction system for LDPE high-pressure tubular polymerization is provided, comprising: a parameter acquisition module for acquiring first process parameters and target product indicators of the LDPE high-pressure tubular polymerization process to be detected; a forward prediction module for inputting the first process parameters into a trained machine learning model to obtain a first predicted product indicator; and for inputting second process parameters into a trained machine learning model to obtain a second predicted product indicator; and a reverse solution module for constructing an objective function based on the first predicted product indicator and the target product indicator, and solving for the target process parameters that minimize the function value of the objective function.

[0063] In one embodiment, the system further includes a visualization interaction module, which is used to invoke the parameter acquisition module, the forward prediction module, and the reverse solving module in response to user actions.

[0064] The visualization and interaction module can be built on existing graphical user interface frameworks, such as Python GUI libraries, and achieve data communication between modules through an event-driven mechanism. Algorithms and modeling processes such as dataset construction, model training, forward prediction, and inverse solving are encapsulated in the backend. Users can operate the parameter acquisition module, forward prediction module, and inverse solving module through the visualization and interaction module without needing to understand the underlying system code.

[0065] In one embodiment, a machine-readable storage medium is provided, on which instructions are stored, characterized in that, when executed by a processor, the instructions cause the processor to be configured to perform the prediction method for the LDPE high-pressure tubular polymerization process described above.

[0066] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor A01, a network interface A02, a display screen A04, an input device A05, and a memory (not shown) connected via a system bus. The processor A01 provides computing and control capabilities. The memory includes internal memory A03 and a non-volatile storage medium A06. The non-volatile storage medium A06 stores an operating system B01 and a computer program B02. The internal memory A03 provides an environment for the operation of the operating system B01 and the computer program B02 stored in the non-volatile storage medium A06. The network interface A02 is used for communication with external terminals via a network connection. When the computer program is executed by the processor A01, it implements a predictive method for LDPE high-pressure tubular polymerization. The display screen A04 can be a liquid crystal display (LCD) or an e-ink display. The input device A05 can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0067] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0068] This application provides a computer (electronic) device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the above-mentioned prediction method for LDPE high-pressure tubular polymerization process.

[0069] This application also provides a computer program product that, when executed on a data processing device, is suitable for performing a program that initializes a predictive method step involving a high-pressure tubular polymerization process for LDPE.

[0070] 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.

[0071] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0072] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0073] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0074] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0075] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0076] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0077] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0078] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method of prediction of an LDPE high pressure tubular polymerization process, characterized in that, The method includes: The first process parameters and target product indicators of the high-pressure tubular polymerization process of LDPE to be tested are obtained, and both the first process parameters and the target product indicators are given by the user. The first process parameters are input into the trained machine learning model to obtain the first predicted product index. An objective function is constructed based on the first predicted product indicator and the target product indicator. The objective function is defined as follows: wherein i is a positive integer, is the ith target product index, is the ith first predicted product index, and n is the number of target product indexes, is the weight corresponding to the ith target product index. Find the target process parameters that minimize the function value of the objective function; The method further includes a training step for the machine learning model, the training step comprising: Establish a mechanistic model for the LDPE high-pressure tubular polymerization process; The process parameters of the sample are generated by simulating the mechanism model, and the process parameters of the sample are sampled by Latin hypercube sampling to generate sample data that are uniformly distributed throughout the range. The sample data is input into the mechanism model to obtain the sample predicted product indicators; A modeling dataset is generated based on the sample data and the sample predicted product indicators, and the machine learning model is trained based on the modeling dataset; The process parameters that minimize the value of the objective function include: A probabilistic model of the objective function is constructed based on a tree-structured Parsons estimator, and the preset number of iterations of the probabilistic model, as well as the feasible region and initial perturbation range of the first process parameter, are obtained. The first process parameters, the feasible region, and the initial disturbance range are input into the probability model to obtain the optimized first process parameters. The function value of the objective function is calculated based on the optimized first process parameters and the target product index, and the probability model is updated based on the function value. Repeat the above iterative steps. When the preset number of iterations is reached, the first process parameter that minimizes the function value is taken as the target process parameter.

2. The method of prediction of LDPE high pressure tubular polymerization process according to claim 1, characterized in that, The method further includes: When the sample predicted product indicators are generated by the mechanism model grouping, the task number of the generated sample predicted product indicators is recorded. When the task number reaches the first preset task quantity, the generated sample predicted product indicators are saved. If the task number meets the second preset task quantity, the mechanism model is restarted, and the generated sample prediction product index is skipped based on the task number recorded before the restart, and the generation task is continued, wherein the first preset task quantity is less than the second preset task quantity.

3. The method of prediction of LDPE high pressure tubular polymerization process according to claim 1, characterized in that, The method further includes: The second process parameter is input into the trained machine learning model to obtain the second predicted product index.

4. The method of prediction of LDPE high pressure tubular polymerization process according to claim 1, characterized in that, The machine learning model includes at least one of the following: Ridge Regression, Random Forest, Gradient Boosting Decision Tree, Extreme Gradient Boosting, Lightweight Gradient Boosting Machine, and Multilayer Perceptron.

5. The method of prediction of LDPE high pressure tubular polymerization process according to claim 1, characterized in that, The method further includes: The target process parameters are input into the machine learning model to obtain the third predicted product index; If the relative deviation between the third predicted product indicator and the target product indicator is less than a preset deviation, the target process parameter is output.

6. A prediction system for an LDPE high pressure tubular polymerization process, characterized in that, include: The parameter acquisition module is used to acquire the first process parameters and target product indicators of the LDPE high-pressure tubular polymerization process to be tested. The positive prediction module is used to input the first process parameter into the trained machine learning model to obtain the first predicted product index; it is also used to input the second process parameter into the trained machine learning model to obtain the second predicted product index. The reverse solution module is used to construct an objective function based on the first predicted product index and the target product index, and solve for the target process parameters that minimize the function value of the objective function.

7. The prediction system for LDPE high pressure tubular polymerization process of claim 6, wherein, The system also includes a visualization interaction module, which is used to invoke the parameter acquisition module, the forward prediction module and the reverse solution module in response to user operations.

8. A machine-readable storage medium having stored thereon instructions, the instructions being executable by a machine to cause the machine to: When executed by a processor, this instruction causes the processor to be configured to perform a predictive method for the LDPE high-pressure tubular polymerization process according to any one of claims 1 to 5.

Images