Prediction method, device and equipment of polyolefin melt index and storage medium

Abstract

The invention relates to the technical field of automation, in particular to a polyolefin melt index prediction method and device, equipment and a storage medium. According to the method, a polymerization parameter set of to-be-predicted polyolefin is obtained, and the polymerization parameter set is input into a trained index prediction model, so that a target prediction value of the polyolefin melt index output by the index prediction model is obtained. The index prediction model is obtained by performing self-supervised learning processing on a labeled sample set and an unlabeled sample set to obtain a self-supervised sample set and performing model training on the self-supervised sample set based on semi-supervised learning. Therefore, the data characterization capability of the index prediction model can be improved through the self-supervised learning process, and limited labeled sample data is utilized for supplement through semi-supervised learning, so that when the number of labeled samples is small, a high-precision polyolefin melt index prediction result can still be obtained, and the prediction accuracy of the polyolefin melt index is improved. And the prediction accuracy of the polyolefin melt index is improved.

Description

Technical Field

[0001] The present application relates to the field of automation technology, and in particular to a method, device, equipment and storage medium for predicting a polyolefin melt index. Background Art

[0002] The melt index of polyolefin is an important indicator for evaluating the melt fluidity of polyolefin materials under specific conditions. It is usually expressed as the mass of polyolefin passing through a standard mold in a specific time in a molten state. It is crucial to optimize the polyolefin production process and improve product quality.

[0003] However, the traditional polyolefin melt index is usually measured at the melt extrusion processing position by online measuring equipment such as a melt index meter. Due to the limitations of the equipment installation location and the detection principle, the existing technology for measuring the melt index has a lag and cannot monitor the actual melt index at the outlet of the polymerization reactor in real time, resulting in low prediction accuracy of the melt index. Summary of the invention

[0004] The invention provides a method, device, equipment and storage medium for predicting a polyolefin melt index, which are used to improve the prediction accuracy of the polyolefin melt index.

[0005] In a first aspect, the present application provides a method for predicting a polyolefin melt index, the method comprising:

[0006] Obtaining a polymerization parameter set of a polyolefin to be predicted, and inputting the polymerization parameter set into a trained index prediction model;

[0007] Obtaining a target predicted value of the polyolefin melt index output by the index prediction model;

[0008] The index prediction model is based on self-supervised learning processing of labeled sample sets and unlabeled sample sets to obtain a self-supervised sample set, and model training of the self-supervised sample set is performed based on semi-supervised learning; each labeled sample set includes sample polymerization parameters of multiple polyolefins and corresponding sample melt indexes, and each unlabeled sample set includes sample polymerization parameters of multiple polyolefins.

[0009] In a second aspect, the present application provides a device for predicting a polyolefin melt index, the device comprising:

[0010] A data acquisition unit, used for acquiring a polymerization parameter set of the polyolefin to be predicted, and inputting the polymerization parameter set into a trained index prediction model;

[0011] An index prediction unit, used to obtain a target predicted value of the polyolefin melt index output by the index prediction model;

[0012] The index prediction model is obtained by performing self-supervised learning on labeled sample sets and unlabeled sample sets to obtain a self-supervised sample set, and training the self-supervised sample set based on semi-supervised learning. Each labeled sample set includes sample polymerization parameters of multiple polyolefins and corresponding sample melt indices, and each unlabeled sample set includes sample polymerization parameters of multiple polyolefins.

[0013] Optionally, the device further comprises a model training unit, which is used to:

[0014] Based on the unlabeled sample set, performing comparative learning training on the self-supervised sub-model to obtain a trained self-supervised sub-model;

[0015] Based on the obtained self-supervised sub-model, performing self-supervised learning processing on the labeled sample set and the unlabeled sample set to obtain a self-supervised sample set;

[0016] Based on the self-supervised sample set, iteratively training the semi-supervised sub-model until a preset first convergence condition is met;

[0017] When the set first convergence condition is met, the trained exponential prediction model is output.

[0018] Optionally, the model training unit is specifically used to:

[0019] Based on a plurality of labeled samples in the self-supervised sample set, initializing and training two regressors respectively; wherein the initial parameters of the two regressors are different;

[0020] Based on each trained regressor, confidence calculation is performed on multiple unlabeled sample sets in the self-supervised sample set to obtain the confidence of each unlabeled sample;

[0021] Based on the relative size relationship between the multiple confidences, determine a target unlabeled sample from the unlabeled samples, and obtain a pseudo label corresponding to the target unlabeled sample;

[0022] Based on the target unlabeled samples and the pseudo labels, the self-supervised sample set is updated, and the next iterative training is performed based on the updated self-supervised sample set.

[0023] Optionally, the model training unit is specifically used to:

[0024] Adding the target unlabeled sample and the pseudo label to a plurality of labeled samples in the self-supervised sample set;

[0025] The target unlabeled sample is removed from the multiple unlabeled samples in the self-supervised sample set to obtain an updated self-supervised sample set.

[0026] Optionally, the first convergence condition includes:

[0027] The training times of the semi-supervised sub-model reaches a preset number of iterations; or,

[0028] All unlabeled samples in the self-supervised sample set are removed, and the preset performance indicators of each regressor remain unchanged.

[0029] Optionally, the model training unit is specifically used to:

[0030] Acquire multiple original unlabeled samples from the unlabeled sample set, and perform feature replacement processing on each original unlabeled sample to obtain an updated enhanced unlabeled sample;

[0031] Obtaining a contrastive learning loss function based on similarities between the plurality of enhanced unlabeled samples and the unlabeled sample set;

[0032] Based on the contrastive learning loss function, the parameters of the self-supervised sub-model are adjusted until the self-supervised sub-model meets a preset second convergence condition, thereby obtaining the trained self-supervised sub-model.

[0033] In a third aspect, the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any one of the polyolefin melt index prediction methods described in the first aspect is implemented.

[0034] In a fourth aspect, the present application provides a computer storage medium, wherein the computer-readable storage medium stores computer program instructions, and the computer program instructions are executed by a processor to perform any one of the polyolefin melt index prediction methods described in the first aspect.

[0035] In a fifth aspect, an embodiment of the present application provides a computer program product, comprising computer program instructions, which, when executed by a processor, implement any one of the polyolefin melt index prediction methods described in the first aspect.

[0036] The beneficial effects of the present invention are as follows:

[0037] The embodiment of the present application provides a method for predicting the polyolefin melt index, which obtains a polymerization parameter set of the polyolefin to be predicted and inputs the polymerization parameter set into a trained index prediction model, thereby obtaining a target prediction value of the polyolefin melt index output by the index prediction model. The index prediction model is based on self-supervised learning processing of labeled sample sets and unlabeled sample sets to obtain a self-supervised sample set, and the self-supervised sample set is trained based on semi-supervised learning. In this way, the embodiment of the present application improves the data characterization ability of the index prediction model through the self-supervised learning process, and supplements it with limited labeled sample data through semi-supervised learning. When the number of labeled samples is small, a high-precision prediction result of the polyolefin melt index can still be obtained, thereby improving the prediction accuracy of the polyolefin melt index. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0039] Figure 1 A schematic diagram of a training process of an index prediction model provided in an embodiment of the present application;

[0040] Figure 2 A schematic diagram of a self-supervised learning process provided in an embodiment of the present application;

[0041] Figure 3 A schematic diagram of a semi-supervised collaborative training process provided in an embodiment of the present application;

[0042] Figure 4 A schematic diagram showing the comparison of model performance of three different training methods provided in the embodiments of the present application;

[0043] Figure 5 A schematic flow chart of a polyolefin melt index index method provided in an embodiment of the present application;

[0044] Figure 6 A schematic diagram of the structure of a device for predicting the melt index of polyolefins provided in an embodiment of the present application;

[0045] Figure 7 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme in the embodiment of the present application will be clearly and completely described below in conjunction with the drawings in the embodiment of the present application. Obviously, the described embodiment is only a part of the embodiment of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application. In the absence of conflict, the embodiments in the present application and the features in the embodiments can be arbitrarily combined with each other. In addition, although the logical order is shown in the flow chart, in some cases, the steps shown or described can be performed in an order different from that here.

[0047] The terms "first" and "second" in the specification and claims of the present application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the term "comprising" and any of their variations are intended to cover non-exclusive protection. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices. "Multiple" in the present application can mean at least two, for example, can be two, three or more, and the embodiments of the present application are not limited.

[0048] The term "and / or" in the embodiments of the present application is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0049] It is understood that in the following specific implementations of the present application, related data such as polyolefin production data are involved. When the various embodiments of the present application are applied to specific products or technologies, relevant licenses or consents need to be obtained, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions. For example, relevant volunteers can be recruited and relevant agreements on volunteer authorization data can be signed, and then the data of these volunteers can be used for implementation; or, by implementing within the scope of the authorized organization, the following implementation methods are implemented by using the data of internal members of the organization to manage data; or, the relevant data used in the specific implementation are all simulated data, such as simulated data generated in a virtual scene.

[0050] The embodiments of the present application relate to artificial intelligence and machine learning (ML) technology, and are mainly designed based on machine learning in artificial intelligence.

[0051] Artificial intelligence is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.

[0052] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0053] Machine learning is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0054] Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning and other technologies. Artificial neural networks (ANN) abstract the human brain neural network from the perspective of information processing, establish a simple model, and form different networks according to different connection methods. A neural network is a computational model composed of a large number of nodes (or neurons) connected to each other. Each node represents a specific output function, called an activation function. The connection between each two nodes represents a weighted value for the signal passing through the connection, called a weight, which is equivalent to the memory of the artificial neural network. The output of the network varies depending on the network connection method, weight value and activation function. The network itself is usually an approximation of a certain algorithm or function in nature, or it may be an expression of a logical strategy.

[0055] The following is a brief introduction to the design concept of the embodiments of the present application.

[0056] Polyolefin materials are widely used in many fields such as plastics, rubber, and packaging materials. The industry plays an important role in the national economy. The melt index of polyolefin is an important indicator for evaluating the melt fluidity of polyolefin materials under specific conditions. It is usually expressed as the mass of polyolefin passing through a standard mold in a specific time under a molten state. It is crucial to optimize the polyolefin production process and improve product quality. In order to accurately predict the melt index of polyolefin, the key is to monitor the key parameters in the polymerization reaction process in real time and optimize and control these parameters to regulate the performance of the final product.

[0057] However, the traditional polyolefin melt index is usually measured at the melt extrusion processing position by online measuring equipment such as a melt index meter. Due to the limitations of the equipment installation location and the detection principle, the existing technology for measuring the melt index has a lag and cannot monitor the actual melt index at the outlet of the polymerization reactor in real time, resulting in low prediction accuracy of the melt index.

[0058] In view of the above problems, an embodiment of the present application provides a method for predicting a polyolefin melt index, which obtains a polymerization parameter set of a polyolefin to be predicted and inputs the polymerization parameter set into a trained index prediction model, thereby obtaining a target prediction value of the polyolefin melt index output by the index prediction model. The index prediction model is based on self-supervised learning processing of labeled sample sets and unlabeled sample sets to obtain a self-supervised sample set, and the self-supervised sample set is trained based on semi-supervised learning. In this way, the embodiment of the present application improves the data characterization ability of the index prediction model through the self-supervised learning process, and supplements it with limited labeled sample data through semi-supervised learning. When the number of labeled samples is small, a high-precision prediction result of the polyolefin melt index can still be obtained, thereby improving the prediction accuracy of the polyolefin melt index.

[0059] The following briefly introduces the application scenarios to which the technical solutions of the embodiments of the present application can be applied. It should be noted that the application scenarios introduced below are only used to illustrate the embodiments of the present application and are not limited. In the specific implementation process, the technical solutions provided by the embodiments of the present application can be flexibly applied according to actual needs.

[0060] The solution provided in the embodiment of the present application can be applied to the application scenarios of process optimization and quality control in most polyolefin production processes, and is used to significantly improve the prediction accuracy of the polyolefin melt index. For example, in the industrial production scenario, the distributed control system is a common automated control system in the industrial production process, which can collect the polymerization parameters of the polyolefin to be predicted (such as temperature, pressure, flow rate, catalyst concentration, etc.) in real time, and predict and feedback the polyolefin melt index according to the index prediction model provided in the embodiment of the present application, and obtain a more accurate melt index prediction result, thereby helping operators to adjust production conditions in real time, optimize process parameters, and improve product quality.

[0061] Of course, the method provided in the embodiment of the present application is not limited to the above application scenarios, but can also be used in other possible application scenarios, which are not limited by the embodiment of the present application. The functions that can be realized by each device in the above application scenarios will be described in the subsequent method embodiments, and no further details will be given here.

[0062] Below, in combination with the application scenarios described above, the method provided by the exemplary embodiment of the present application is described with reference to the accompanying drawings. It should be noted that the above application scenarios are only shown to facilitate understanding of the spirit and principles of the present application, and the implementation methods of the present application are not limited in this regard.

[0063] In the embodiment of the present application, an index prediction model can be used to implement the above-mentioned polyolefin melt index acquisition process. Before the index prediction model is put into use, it is necessary to perform model training in advance so that the index prediction model converges. Next, in order to facilitate the description of the model application process, the training process of the index prediction model is first introduced.

[0064] Please refer to Figure 1 , is a schematic diagram of a training process of an index prediction model provided in an embodiment of the present application, and the specific implementation process of the training process is as follows:

[0065] Step 101: Based on the unlabeled sample set, the self-supervised sub-model is trained by contrastive learning to obtain a trained self-supervised sub-model.

[0066] Step 102: Based on the obtained self-supervised sub-model, self-supervised learning processing is performed on the labeled sample set and the unlabeled sample set to obtain a self-supervised sample set.

[0067] In the embodiment of the present application, the index prediction model may include a self-supervised sub-model and a semi-supervised sub-model. The self-supervised sub-model is trained by comparative learning through an unlabeled sample set to obtain a trained self-supervised sub-model, so that the labeled sample set and the unlabeled sample set are respectively input into the self-supervised sub-model to obtain a self-supervised sample set after self-supervised characterization, thereby effectively utilizing a large amount of unlabeled data and fully exploring the potential characteristics of the polyolefin polymerization process. In the case of fewer labeled samples, the expression ability of the model is enhanced by self-supervised learning, thereby improving the accuracy of melt index prediction.

[0068] In one possible implementation, before model training, the embodiment of the present application will collect labeled samples containing sample polymerization parameters of multiple polyolefins and corresponding sample melt indexes, and unlabeled samples containing only sample polymerization parameters of multiple polyolefins as training sample data for the index prediction model.

[0069] Specifically, each auxiliary variable x∈Rm associated with the melt index y can be selected, where m represents the number of auxiliary variables, and x represents a vector containing m auxiliary variables. In this way, the labeled sample set [XL, YL] contains the auxiliary variables XL and the melt index YL, the unlabeled sample set XU only contains the auxiliary variables XU, and n1 and n2 represent the number of labeled samples and unlabeled samples, respectively. For example, a variable analysis can be performed on the polyolefin polymerization process, and five auxiliary variables including TICL4 (titanium chloride concentration), TEAL (triethylaluminum concentration), donor (electron donor concentration), R201h2 / c3h6 (hydrogen / propylene ratio), and R202h2 / c3h6 (hydrogen / propylene ratio) are selected as the input variables of the model, and the melt index (Melt Index, MI) is selected as the output variable of the model.

[0070] In a possible implementation, in order to effectively capture the time-varying trend of input variables in the model, a suitable time step may be selected to construct input feature vectors XL and XU, and then the labeled sample set [XL, YL] and the unlabeled sample set XU are normalized to map the data to a specified range, such as [0, 1] or [-1, 1], so as to eliminate the influence of different variable dimensions and unify the scales of the variables.

[0071] Specifically, assuming that each time step t corresponds to a data record at a time point, select information containing n time steps for modeling. For example, set n = 5, that is, each feature vector contains the data of the current time step and the previous 4 time steps, XL = [Xt, Xt - 1. Xt - 2. Xt - 3. Xt -4], where Xt represents the combination of the five auxiliary variables at the current time step. By concatenating the data of the previous time steps into a vector, the changing trend of the input variables can be more comprehensively reflected. Similarly, the construction process of XU is the same as that of XL, and the feature vector can also be constructed based on the information of multiple time steps. Then [XL, YL] and XU are normalized to map the data to the range of [0, 1]. The normalization formula can be shown as follows:

[0072]

[0073] Among them, x new Represents the normalized input feature value;

[0074] x represents the original unnormalized input feature value;

[0075] x min Represents the minimum value of the input feature value in the corresponding sample set.

[0076] x max Represents the maximum value of the input feature value in the corresponding sample set.

[0077] In a possible implementation, the contrastive learning training of the self-supervised sub-model in the embodiment of the present application can be as follows: multiple original unlabeled samples are obtained from the unlabeled sample set, and feature replacement processing is performed on each original unlabeled sample to obtain an updated enhanced unlabeled sample. Then, based on the similarity between the multiple enhanced unlabeled samples and the unlabeled sample set, a contrastive learning loss function is obtained, and then the parameters of the self-supervised sub-model are adjusted according to the contrastive learning loss function until the self-supervised sub-model meets the preset second convergence condition, thereby obtaining a trained self-supervised sub-model.

[0078] Specifically, refer to Figure 2The diagram shows a schematic diagram of a self-supervised learning process provided by an embodiment of the present application. The embodiment of the present application can first determine the number of data samples required to be processed in one training iteration, that is, the batch size N, and randomly select N original unlabeled samples from the unlabeled sample set. For each original unlabeled sample, some of the features contained therein are randomly selected, and the corresponding feature values ​​are replaced with values ​​randomly sampled from the marginal distribution of the feature, thereby obtaining an updated enhanced unlabeled sample to achieve data enhancement. The unlabeled data input (original unlabeled sample) and the 'homogeneous' unlabeled data input (enhanced unlabeled sample) are respectively input into an encoder (such as a deep neural network model) for feature extraction to obtain a data feature representation, that is, a transformed input feature vector. These data feature representations are input into a projection head, which can be a fully connected neural network layer for mapping high-dimensional feature representations to a new low-dimensional space, thereby reducing computational complexity and improving generalization capabilities. After being processed by the projection head, the original data features are projected into a new vector space to obtain a projection vector. The projection vector represents the feature representation of the original data features in the new vector space. In this process, the parameters of the encoder and projection head can be optimized and adjusted by measuring the similarity or difference between the two projection vectors to minimize the loss function. The encoder is optimized by minimizing the loss function so that the encoder can best represent the data. In this way, the above steps are repeated until the loss function is minimized, or when the loss function no longer decreases significantly after multiple iterations, or when the preset maximum number of iterations is reached, the training process stops and the final encoder is obtained, that is, the trained self-supervised sub-model.

[0079] In a possible implementation, the normalized temperature scaled cross-entropy loss (NT-Xent loss) function can be used to optimize the encoder parameters. This loss function is used to measure the similarity of two samples (original unlabeled samples and enhanced unlabeled samples). By adjusting the model parameters, the cosine similarity of the two sample pairs is maximized, while the cosine similarity of the dissimilar sample pairs is minimized. The loss function is as follows:

[0080]

[0081] Where N represents the batch size, that is, the number of data samples that need to be processed in one training iteration;

[0082] T represents the adjustable temperature parameter, which is used to scale the input and expand the range of cosine similarity. It can adjust the similarity distribution between samples to make the similarity of sample pairs clearer.

[0083] si represents the similarity between the i-th sample and its corresponding enhanced sample.

[0084] The calculation formula based on cosine similarity is as follows:

[0085]

[0086] z (i) and z (j) are the encoded identification vectors obtained by projection representation of the features of the i-th original unlabeled sample and the j-th enhanced unlabeled sample;

[0087] ||z (i) ||2 and ||z (j) ||2 is the norm of the vector.

[0088] Step 103: Based on the self-supervised sample set, iteratively train the semi-supervised sub-model until a preset first convergence condition is met.

[0089] In an embodiment of the present application, a semi-supervised sub-model is used to train samples after self-supervised characterization until the semi-supervised sub-model meets a preset first convergence condition, thereby obtaining a trained semi-supervised sub-model.

[0090] In a possible implementation manner, when executing step 103, the embodiment of the present application will specifically perform the following operations in each round of iterative training:

[0091] Step 1031: Based on multiple labeled samples in the self-supervised sample set, initialize and train the two regressors respectively.

[0092] In an embodiment of the present application, two regressors with different initial parameters are used to perform semi-supervised collaborative training on a labeled sample set processed by a self-supervised sub-model.

[0093] In a possible implementation, the two regressors may adopt the same regression algorithm, but may be instantiated using different distance metrics or nearest neighbor parameters to ensure the diversity of the regressors.

[0094] Specifically, regressor h1 and regressor h2 can both use the K-Nearest Neighbors (k-NN) regression algorithm. The k-NN algorithm is a non-parametric method for prediction based on neighbor samples. For each sample, the nearest k neighbors are found and the predicted value is calculated based on the target value (i.e., melting index) of these neighbors. Therefore, the diversity of the regressors can be ensured by instantiating the two regressors with different distance metrics p or different k values. The regressors can be as follows:

[0095] h1=kNN(L1,k1,p1)

[0096] h2=kNN(L2,k2,p2)

[0097] L1=L, L2=L, L={(x1,y1),…,(x |L| ,y |L| )}

[0098] Among them, h1 represents the first regressor, which is trained using k1 nearest neighbors and p1 distance metrics;

[0099] h1 represents the second regressor, trained using k2 nearest neighbors and p2 distance metrics;

[0100] k represents a hyperparameter in the k-NN algorithm, which indicates the number of nearest neighbors selected. A larger K value means that more neighbors have an impact on the prediction results, while a smaller k value means more sensitivity to closer neighbors.

[0101] p represents the distance metric of the regressor, and different metrics can be selected to calculate the distance between samples. For example, p1 can be Manhattan distance (L1), and p2 can be Euclidean distance (L2).

[0102] L1 and L2 are the sample sets used by regressor h1 and regressor h2 respectively, both of which are L, that is, both regressors use labeled sample sets.

[0103] Specifically, the two regressors are trained using the labeled sample set XL. For each labeled sample, the distance can be calculated based on other samples in the labeled sample set to determine the k1 or k2 nearest neighbors. The predicted melting index value corresponding to the sample is calculated using the neighbor's label YL value, for example, the predicted melting index value is obtained by the weighted average of the neighbor's labels. When the predicted values ​​of all labeled samples are calculated, the regressors h1 and h2 have completed the initial training based on the labeled sample set and can predict the unlabeled samples.

[0104] Step 1032: Based on the trained regressors, confidence calculations are performed on multiple unlabeled sample sets in the self-supervised sample set to obtain the confidence of each unlabeled sample.

[0105] In the embodiment of the present application, after the initial training of the regressor is completed, the collaborative training phase will be entered, and the confidence of each unlabeled sample will be calculated using regressors h1 and h2 to measure the improvement effect of adding the pseudo-label of the unlabeled sample to the training set on the model performance.

[0106] In one possible implementation, for an unlabeled sample x u , calculate its confidence score The confidence evaluation formula can be shown as follows:

[0107]

[0108] in, Represents the confidence score of the unlabeled sample, which is a measure of the improvement in the model's prediction ability after adding the pseudo-label to the self-supervised sample set. The higher the confidence score, the greater the improvement in the model's performance due to the pseudo-label.

[0109] Ω u Represents an unlabeled sample x u The neighboring sample set in the labeled sample set XL, that is, the neighboring sample set in the labeled sample with the unlabeled sample x u The k closest samples.

[0110] h represents the regressor, h(x i ) represents the predicted value of the neighboring sample xi.

[0111] h′ represents the value of the unlabeled sample x u After updating the self-supervised sample set, the new regressor obtained by training, h ′ (x i ) represents the predicted value of the new regressor for the neighboring sample xi.

[0112] According to the confidence score, it can be determined which pseudo-labels will have a greater impact on the neighbor prediction after being added to the self-supervised sample set. That is, the greater the improvement in the model prediction accuracy caused by adding the pseudo-label, the more likely the pseudo-label is to be correct.

[0113] Step 1033: Based on the relative size relationship between the multiple confidences, determine the target unlabeled sample from the unlabeled samples, and obtain the pseudo label corresponding to the target unlabeled sample.

[0114] In the embodiment of the present application, after obtaining the confidence scores of each unlabeled sample, the target unlabeled sample with the highest confidence can be selected according to the relative size relationship of each confidence score, and its melting index prediction value can be used as the corresponding pseudo label.

[0115] In one possible implementation, since the confidence of each regressor on all unlabeled samples is calculated in each round of training, which consumes more unnecessary computing resources, the embodiment of the present application may use an unlabeled small sample pool U′ that is smaller than the unlabeled sample set U, and screen the target unlabeled samples by calculating the confidence of each unlabeled sample in the unlabeled small sample pool.

[0116] In one possible implementation, if but

[0117]

[0118] in, represents the unlabeled sample with the highest confidence score, that is, the target unlabeled sample selected to generate pseudo labels.

[0119] U′ represents a subset of the unlabeled sample set, that is, a small pool of unlabeled samples. In each iterative training, pseudo labels can be selected only from a smaller subset, thereby reducing the consumption of computing resources.

[0120] π j Representative target unlabeled samples The pseudo-label of Representative model for target unlabeled samples The predicted value of .

[0121] In this way, the above formula can be used to select the sample with the highest confidence score from the unlabeled sample set: as target unlabeled samples.

[0122] Step 1034: Based on the target unlabeled samples and pseudo labels, the self-supervised sample set is updated, and the next iterative training is performed based on the updated self-supervised sample set.

[0123] In the embodiment of the present application, after the target unlabeled samples and pseudo labels are selected, the self-supervised sample set is updated, and the two regressors are continuously trained using the new self-supervised sample set.

[0124] In a possible implementation, after obtaining the target unlabeled sample and the pseudo label, the embodiment of the present application may add the target unlabeled sample and the pseudo label to multiple labeled samples in the self-supervised sample set, and remove the target unlabeled sample from the multiple unlabeled samples in the self-supervised sample set, thereby obtaining an updated self-supervised sample set, as shown below:

[0125] L1=L1∪π j

[0126] L2=L2∪π j

[0127] U′=U′-π j

[0128] Among them, L1 and L2 are the sample sets used by regressor h1 and regressor h2 respectively, and U′ represents the target unlabeled sample π removed. j Unlabeled sample set.

[0129] Step 104: When the set first convergence condition is met, the trained exponential prediction model is output.

[0130] In the embodiment of the present application, the first convergence condition of the semi-supervised sub-model may include: the number of training times of the semi-supervised sub-model reaches a preset number of iterations, or all unlabeled samples in the self-supervised sample set are removed, and the preset performance indicators such as the mean squared error (MSE) of each regressor remain unchanged. When the preset first convergence condition is met, the trained exponential prediction model can be output.

[0131] Specifically, in the iterative training process of updating the self-supervised sample sets L1 and L2, the two regressors are trained using the new self-supervised sample sets L1 and L2 until the self-supervised sample sets L1 and L2 no longer change, that is, all unlabeled samples in the self-supervised sample sets are removed, and the MSE or other evaluation indicators of the model no longer change, indicating that the semi-supervised sub-model has been trained, or the number of training times of the semi-supervised sub-model reaches the preset number of iterations, and the trained regressor and index prediction model are output as shown below:

[0132] h1=kNN(L1,k,p1)

[0133] h2=kNN(L2,k,p2)

[0134]

[0135] h1 and h2 are two trained regressors;

[0136] h * (x) represents the target predicted value of the polyolefin melt index obtained by inputting the polymerization parameter set x of the polyolefin to be predicted into the trained index prediction model.

[0137] In one possible implementation, reference Figure 3The figure shows a collaborative training flow chart of a semi-supervised sub-model provided by an embodiment of the present application. The embodiment of the present application selects labeled data from a labeled sample set and inputs it into a trained encoder (deep neural network) to extract the data feature representation xl of the labeled data and the labeled data output y. At the same time, the unlabeled sample set also extracts the data feature representation xu through the same encoder (deep neural network). Unlabeled data is mainly used for self-supervised learning. By training the model together with labeled data, the problem of insufficient labeled data is compensated. Then, according to the labeled data, learners knn1 and knn2 are constructed respectively. Knn1 uses the labeled data set L1, and knn2 uses the labeled data set L2. The preliminary model training is completed through supervised learning. For the unlabeled data feature xu, a neighboring sample set xi∈Ωu will be selected from the labeled data set. Neighboring samples can help the model learn the potential output information of unlabeled data. Before adding unlabeled data to the training set of the learner, it is necessary to determine whether the confidence δxu of the unlabeled data is greater than zero. If the confidence of the unlabeled data is high, that is, the prediction result of the data is more reliable, it will be added to the training set, and the two learners will continue to be trained using the updated training set. When the convergence conditions are met, such as when the training set no longer changes or the performance evaluation index remains unchanged, the final trained exponential prediction model is output.

[0138] refer to Figure 4 Shown is a schematic diagram of the comparison of model performance of different training methods provided in the embodiments of the present application. The figure shows the three different training methods of supervised training, semi-supervised training and self-supervised combined with semi-supervised training adopted in the present application, and compares the prediction results of polyolefin melt index (MI). The horizontal axis of each figure is the serial number of the sample data, and the vertical axis is the value of the melt index (MI). The red line represents the actual value (real) of the melt index, and the green line represents the predicted value (pred) output by the model. If the two lines can overlap closely on most of the data points, it means that the prediction effect of the model is good and can effectively fit the trend of the real data. MSE represents the mean square error between the model prediction value and the true value. The lower the MSE, the more accurate the model prediction. R2 represents the model fitting effect. The closer it is to 1, the better the model fitting effect. Through this Figure 4 It can be clearly seen that only using supervised training and semi-supervised training will lead to obvious deviations in the model prediction results at some points. The green predicted value and the red true value are quite different, especially the prediction deviation of some peak or trough points is large, which means that the fitting ability is weak at these complex changing points. The training method combining self-supervised learning and semi-supervised learning adopted in the embodiment of the present application can significantly improve the prediction accuracy of the exponential prediction model, reduce the mean square error, and improve the goodness of fit.

[0139] After obtaining the trained index prediction model, the embodiment of the present application may use the index prediction model to perform melt index prediction processing on the polymerization parameter set of the polyolefin to be predicted, so as to obtain a target prediction value of the polyolefin melt index output by the index prediction model.

[0140] refer to Figure 5 As shown, it is a flow chart of a method for predicting a polyolefin melt index provided in an embodiment of the present application, and the specific implementation process of the method is as follows:

[0141] Step 501: Obtain a polymerization parameter set of a polyolefin to be predicted, and input the polymerization parameter set into a trained index prediction model.

[0142] In an embodiment of the present application, after self-supervised learning is performed on a labeled sample set and an unlabeled sample set to obtain a self-supervised sample set, and model training is performed on the self-supervised sample set based on semi-supervised learning to obtain a trained index prediction model, the index prediction model can be applied to the actual melt index prediction of polyolefin polymerization parameters, that is, the polymerization parameter data of the polyolefin whose melt index is to be predicted is input into the index prediction model.

[0143] Specifically, in the actual polyolefin production process, the concentrations of chemical substances and reaction conditions in the reactors such as TICL4, TEAL, donor, R201h2 / c3h6, and R202h2 / c3h6 can be collected in real time through monitoring equipment in the polymerization reactor, and the real-time data can be aggregated through data collection to form a set of polymerization parameters of the polyolefin to be predicted. Then, the acquired time series data of the five variables can be preprocessed, for example, each variable can be normalized to ensure the consistency and stability of the data, and then the processed data can be input into the trained index prediction model.

[0144] Step 502: Obtain a target predicted value of the polyolefin melt index output by the index prediction model.

[0145] In the embodiment of the present application, the above-mentioned polymerization parameter set is processed by a trained index prediction model to obtain a target prediction value of the melt index of the polyolefin to be predicted, and the predicted value of the melt index is output to an operator or a related process control system to adjust the production conditions in real time, optimize the process parameters, and improve the quality of the polyolefin product.

[0146] Specifically, the trained index prediction model can be quickly calculated based on the input polymerization parameter information such as TICL4, TEAL, donor, R201h2 / c3h6, R202h2 / c3h6, etc., combined with historical data and learned experience. The self-supervised sub-model in the model first processes the unlabeled data to extract potential features, and the semi-supervised sub-model can be calibrated and optimized based on a small amount of labeled data. The output result of the model is the melt index of polyolefins under current production conditions. This melt index is a predicted value obtained based on the input polymerization parameter set and the parameter mapping relationship within the model. The index prediction model can ensure the accuracy and reliability of the prediction results by capturing the interaction between the time step information and each variable.

[0147] In the embodiment of the present application, the prediction process of the polyolefin melt index during the training process is the same as the prediction process during the actual application process, so the process can refer to the detailed introduction of the aforementioned training process, and no further details will be given here.

[0148] Please refer to Figure 6 Based on the same inventive concept, the embodiment of the present application further provides a prediction device 60 for a polyolefin melt index, the device comprising:

[0149] A data acquisition unit 601 is used to acquire a polymerization parameter set of the polyolefin to be predicted, and input the polymerization parameter set into a trained index prediction model;

[0150] An index prediction unit 602, used to obtain a target prediction value of the polyolefin melt index output by the index prediction model;

[0151] Among them, the index prediction model is based on self-supervised learning processing of labeled sample sets and unlabeled sample sets to obtain self-supervised sample sets, and model training of the self-supervised sample sets is performed based on semi-supervised learning. Each labeled sample set includes sample polymerization parameters of multiple polyolefins and corresponding sample melt indices, and each unlabeled sample set includes sample polymerization parameters of multiple polyolefins.

[0152] Optionally, the device further includes a model training unit 603, which is used to:

[0153] Based on the unlabeled sample set, the self-supervised sub-model is trained by contrastive learning to obtain a trained self-supervised sub-model;

[0154] Based on the obtained self-supervised sub-model, self-supervised learning is performed on the labeled sample set and the unlabeled sample set to obtain a self-supervised sample set;

[0155] Based on the self-supervised sample set, the semi-supervised sub-model is iteratively trained until the preset first convergence condition is met;

[0156] When the set first convergence condition is met, the trained exponential prediction model is output.

[0157] Optionally, the model training unit 603 is specifically used for:

[0158] Based on multiple labeled samples in the self-supervised sample set, two regressors are initialized and trained respectively; wherein the initial parameters of the two regressors are different;

[0159] Based on the trained regressors, the confidence of multiple unlabeled sample sets in the self-supervised sample set is calculated to obtain the confidence of each unlabeled sample.

[0160] Based on the relative size relationship between multiple confidence levels, a target unlabeled sample is determined from each unlabeled sample, and a pseudo label corresponding to the target unlabeled sample is obtained;

[0161] Based on the target unlabeled samples and pseudo labels, the self-supervised sample set is updated, and the next iterative training is performed based on the updated self-supervised sample set.

[0162] Optionally, the model training unit 603 is specifically used for:

[0163] Add the target unlabeled samples and pseudo labels to multiple labeled samples in the self-supervised sample set;

[0164] From multiple unlabeled samples in the self-supervised sample set, remove the target unlabeled sample to obtain an updated self-supervised sample set.

[0165] Optionally, the first convergence condition includes:

[0166] The semi-supervised sub-model is trained for a preset number of iterations; or,

[0167] All unlabeled samples in the self-supervised sample set are removed, and the preset performance indicators of each regressor remain unchanged.

[0168] Optionally, the model training unit 603 is specifically used for:

[0169] Obtain multiple original unlabeled samples from the unlabeled sample set, and perform feature replacement processing on each original unlabeled sample to obtain an updated enhanced unlabeled sample;

[0170] Obtaining a contrastive learning loss function based on the similarity between multiple enhanced unlabeled samples and the unlabeled sample set;

[0171] Based on the contrastive learning loss function, the parameters of the self-supervised sub-model are adjusted until the self-supervised sub-model meets the preset second convergence condition, thereby obtaining a trained self-supervised sub-model.

[0172] For the convenience of description, the above parts are divided into each unit module (or module) according to the function and described separately. Of course, when implementing the present application, the functions of each unit (or module) can be implemented in the same one or more software or hardware. It can be understood by those skilled in the art that various aspects of the present application can be implemented as a system, method or program product. Therefore, various aspects of the present application can be specifically implemented in the following forms, namely: a complete hardware implementation method, a complete software implementation method (including firmware, microcode, etc.), or an implementation method combining hardware and software, which can be collectively referred to as "circuit", "module" or "system" here.

[0173] The device can be used to execute the methods shown in the various embodiments of the present application. Therefore, for the functions that can be implemented by the various functional modules of the device, reference can be made to the description of the aforementioned embodiments and no further details will be given.

[0174] See also Figure 7 As shown, based on the same technical concept, an embodiment of the present application further provides a computer device 70 , which may include a memory 701 and a processor 702 .

[0175] The so-called memory 701 is used to store computer programs executed by the processor 702. The memory 701 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the computer device, etc. The processor 702 may be a central processing unit (CPU), or a digital processing unit, etc. The specific connection medium between the above-mentioned memory 701 and the processor 702 is not limited in the embodiments of the present application. The embodiments of the present application are Figure 4 In the embodiment, the memory 701 and the processor 702 are connected via a bus 703. The bus 703 is connected to the processor 702 via a bus 703. Figure 4 The connection between other components is shown by bold lines, and is not intended to be limiting. The so-called bus 703 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0176] The memory 701 may be a volatile memory, such as a random-access memory (RAM); the memory 701 may also be a non-volatile memory, such as a read-only memory, a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD), or the memory 701 may be any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 701 may be a combination of the above memories.

[0177] The processor 702 is used to call the computer program stored in the so-called memory 701 to execute the polyolefin melt index prediction method executed by the device in each embodiment of the present application.

[0178] In some possible embodiments, various aspects of the method for predicting the polyolefin melt index provided in the present application may also be implemented in the form of a program product, which includes a program code. When the program product is run on a computer device, the program code is used to enable the computer device to execute the steps of the method for predicting the polyolefin melt index according to various exemplary embodiments of the present application described above in this specification. For example, the computer device can execute the steps of each embodiment.

[0179] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0180] The program product of the embodiment of the present application may adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a computing device. However, the program product of the present application is not limited thereto. In the present application, the readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with a command execution system, device or device.

[0181] The readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, wherein the readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with a command execution system, apparatus, or device.

[0182] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0183] Program code for performing the operations of the present application may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user equipment, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device via any type of network including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0184] It should be noted that, although several units or subunits of the device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present 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 into multiple units to be embodied.

[0185] In addition, although the operations of the method of the present application are described in a specific order in the drawings, this does not require or imply that the operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0186] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0187] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0188] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for predicting a polyolefin melt index, characterized in that: The method comprises: Obtaining a polymerization parameter set of a polyolefin to be predicted, and inputting the polymerization parameter set into a trained index prediction model; Obtaining a target predicted value of the polyolefin melt index output by the index prediction model; The index prediction model is based on self-supervised learning processing of labeled sample sets and unlabeled sample sets to obtain a self-supervised sample set, and model training of the self-supervised sample set is performed based on semi-supervised learning; each labeled sample set includes sample polymerization parameters of multiple polyolefins and corresponding sample melt indexes, and each unlabeled sample set includes sample polymerization parameters of multiple polyolefins.

2. The method according to claim 1, characterized in that The index prediction model is trained in the following way: Based on the unlabeled sample set, performing comparative learning training on the self-supervised sub-model to obtain a trained self-supervised sub-model; Based on the obtained self-supervised sub-model, performing self-supervised learning processing on the labeled sample set and the unlabeled sample set to obtain a self-supervised sample set; Based on the self-supervised sample set, iteratively training the semi-supervised sub-model until a preset first convergence condition is met; When the set first convergence condition is met, the trained exponential prediction model is output.

3. The method according to claim 2, characterized in that Each training iteration includes the following steps: Based on a plurality of labeled samples in the self-supervised sample set, initializing and training two regressors respectively; wherein the initial parameters of the two regressors are different; Based on each trained regressor, confidence calculation is performed on multiple unlabeled sample sets in the self-supervised sample set to obtain the confidence of each unlabeled sample; Based on the relative size relationship between the multiple confidences, determine a target unlabeled sample from the unlabeled samples, and obtain a pseudo label corresponding to the target unlabeled sample; Based on the target unlabeled samples and the pseudo labels, the self-supervised sample set is updated, and the next iterative training is performed based on the updated self-supervised sample set.

4. The method according to claim 3, characterized in that The updating of the self-supervised sample set based on the target unlabeled sample and the pseudo label comprises: Adding the target unlabeled sample and the pseudo label to a plurality of labeled samples in the self-supervised sample set; The target unlabeled sample is removed from the multiple unlabeled samples in the self-supervised sample set to obtain an updated self-supervised sample set.

5. The method according to any one of claims 2 to 4, characterized in that: The first convergence condition includes: The training times of the semi-supervised sub-model reaches a preset number of iterations; or, All unlabeled samples in the self-supervised sample set are removed, and the preset performance indicators of each regressor remain unchanged.

6. The method according to claim 2, characterized in that The step of performing comparative learning training on the self-supervisory sub-model based on the unlabeled sample set to obtain a trained self-supervisory sub-model includes: Acquire multiple original unlabeled samples from the unlabeled sample set, and perform feature replacement processing on each original unlabeled sample to obtain an updated enhanced unlabeled sample; Obtaining a contrastive learning loss function based on similarities between the plurality of enhanced unlabeled samples and the unlabeled sample set; Based on the contrastive learning loss function, the parameters of the self-supervised sub-model are adjusted until the self-supervised sub-model meets a preset second convergence condition, thereby obtaining the trained self-supervised sub-model.

7. A device for predicting polyolefin melt index, characterized in that: The device comprises: A data acquisition unit, used for acquiring a polymerization parameter set of the polyolefin to be predicted, and inputting the polymerization parameter set into a trained index prediction model; An index prediction unit, used to obtain a target predicted value of the polyolefin melt index output by the index prediction model; The index prediction model is obtained by performing self-supervised learning on labeled sample sets and unlabeled sample sets to obtain a self-supervised sample set, and training the self-supervised sample set based on semi-supervised learning. Each labeled sample set includes sample polymerization parameters of multiple polyolefins and corresponding sample melt indices, and each unlabeled sample set includes sample polymerization parameters of multiple polyolefins.

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, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising computer program instructions, characterized in that When the computer program instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images