Quality Variable Prediction Method, Device Terminal and Medium under the Assistance Training Framework

Abstract

The present application relates to a method, apparatus, terminal, and storage medium for predicting quality variables under an assistance training framework, and relates to the fields of complex industrial process modeling and fault diagnosis. The method includes: obtaining a dataset to be measured, a labeled sample dataset, and an unlabeled sample dataset; establishing an initial main learning model and an initial auxiliary learning model corresponding to the initial main learning model; training the initial main learning model and the initial auxiliary learning model; establishing a quality variable prediction model; and inputting the dataset to be measured into the quality variable prediction model. During the prediction process, through the combined training of the pre-obtained unlabeled sample dataset and labeled sample dataset, on the basis of selecting the unlabeled sample dataset with high global information content, the quality of the quality variable prediction model is improved, so that in the scenario with a large number of unlabeled samples, there is a stable and specific way for predicting quality variables, and the accuracy and efficiency of predicting quality variables are improved.

Description

Technical Field

[0001] The present application relates to the field of complex industrial process modeling and fault diagnosis, and particularly relates to a method, device, terminal and storage medium for predicting quality variables under an assisted training framework. Background Art

[0002] Complex industrial processes widely exist in fields such as oil refining and chemical engineering, and have characteristics such as multi-variable, strong coupling, strong non-linearity, randomness, large time delay, the output cannot be measured online, and large changes in working conditions, making it difficult to describe with an accurate mathematical model.

[0003] In order to analyze complex industrial processes, when performing corresponding analysis on complex industrial processes, it is necessary to determine the quality variables in the industrial process. Usually, the change situation of the quality variables can reflect whether the working conditions of the complex industrial process are normal. When measuring the quality variables, it is usually necessary to determine the working condition data related to the quality variables, construct a sample set based on the working condition data, and predict the quality variables through the sample set.

[0004] However, in related technologies, in the scenario of automatically measuring the quality variables of complex industrial processes, most sample sets cannot be accompanied by the corresponding measurement results of the quality variables, that is, they are all unlabeled sample data sets. In this case, it is difficult to determine the specific prediction method for the quality variables based on the complex sample set, resulting in low efficiency and accuracy in measuring the quality variables. Summary of the Invention

[0005] The present application relates to a method, device, terminal and storage medium for predicting quality variables under an assisted training framework, which can improve the detection efficiency and accuracy of quality variables. The technical solution is as follows:

[0006] On the one hand, a method for predicting quality variables under an assisted training framework is provided, and the method includes:

[0007] Obtain a data set to be measured, a labeled sample data set and an unlabeled sample data set. The labeled sample data set includes at least two groups of labeled sample data groups and the corresponding sample quality variable values, the unlabeled sample data set includes at least two groups of unlabeled sample data groups, and the data set to be measured includes at least two groups of data groups to be measured;

[0008] Establish an initial main learning model and an initial auxiliary learning model corresponding to the initial main learning model. The initial auxiliary learning model is used for preliminary annotation of unlabeled sample data, and the initial main learning model is used for expanding the labeled sample data set based on the results of the preliminary annotation;

[0009] Train the initial main learning model and the initial auxiliary learning model using an unlabeled sample dataset and a labeled sample dataset to obtain the main learning model and the auxiliary learning model;

[0010] Build a quality variable prediction model based on the main learning model and the auxiliary learning model;

[0011] Input the data group to be measured into the quality variable prediction model and output the quality variable prediction result corresponding to the data group to be measured.

[0012] On the other hand, a quality variable prediction device is provided, and the device includes:

[0013] An acquisition module for acquiring a data set to be measured, a labeled sample data set, and an unlabeled sample data set. The labeled sample data set includes at least two labeled sample data groups and the sample quality variable values corresponding to the labeled sample data groups. The unlabeled sample data set includes at least two unlabeled sample data groups, and the data set to be measured includes at least two data groups to be measured;

[0014] A building module for building an initial main learning model and an initial auxiliary learning model corresponding to the initial main learning model. The initial auxiliary learning model is used for the preliminary annotation of the unlabeled sample data, and the initial main learning model is used for the expansion of the labeled sample data set based on the results of the preliminary annotation;

[0015] A training module for training the initial main learning model and the initial auxiliary learning model using the unlabeled sample data set and the labeled sample data set to obtain the main learning model and the auxiliary learning model;

[0016] The building module is further used to build a quality variable prediction model based on the main learning model and the auxiliary learning model;

[0017] An input module for inputting the data group to be measured into the quality variable prediction model and outputting the quality variable prediction result corresponding to the data group to be measured.

[0018] On the other hand, a computer device is provided. The computer device includes a processor and a memory. At least one instruction, at least one program, a code set, or an instruction set is stored in the memory. The processor can load and execute at least one instruction, at least one program, a code set, or an instruction set to implement the quality variable prediction method in the assistance training framework provided in the embodiments of the present application.

[0019] On the other hand, a computer-readable storage medium is provided, in which at least one instruction, at least one program, a code set or an instruction set is stored. The processor can load and execute at least one instruction, at least one program, a code set or an instruction set to implement the quality variable prediction method under the assistance training framework provided in the embodiments of the present application above.

[0020] On the other hand, a computer program product or a computer program is provided. The computer program product or the computer program includes computer program instructions, and the computer program instructions are stored in a computer-readable storage medium. The processor reads the computer instructions from the computer-readable storage medium and executes the computer instructions, so that the computer device executes the quality variable prediction method under the assistance training framework provided in the embodiments of the present application.

[0021] The beneficial effects brought by the technical solution provided in the present application at least include:

[0022] In the process of determining the quality variables of a complex industrial process, in the case where the number of labeled sample data groups is small and the number of unlabeled samples is large, before predicting the quality variables of the dataset to be measured, an initial auxiliary learning model and an initial main learning model are established respectively, and while expanding the sample set, they are trained by themselves, and finally a quality variable prediction model is generated to predict the quality variables of the dataset to be measured. In the process of prediction, through the combined training of the pre-established unlabeled sample dataset and the labeled sample dataset, on the basis of selecting the unlabeled sample dataset with high global information content, the quality of the quality variable prediction model is improved, so that in the scenario where the number of unlabeled samples is large, there is a stable and specific way to predict the quality variables, and the accuracy and efficiency of predicting the quality variables are improved. Description of the Drawings

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to these drawings.

[0024] Figure 1 It shows a schematic flowchart of a quality variable prediction method under an assistance training framework provided by an exemplary embodiment of the present application;

[0025] Figure 2 It shows a schematic diagram of the process of a quality variable prediction method under an assistance training framework provided by an exemplary embodiment of the present application;

[0026] Figure 3The process schematic diagram of a method for predicting quality variables under an assisted training framework provided by an exemplary embodiment of the present application is shown;

[0027] Figure 4 The structural block diagram of a prediction device for a quality variable provided by an exemplary embodiment of the present application is shown;

[0028] Figure 5 The structural block diagram of another prediction device for a quality variable provided by an exemplary embodiment of the present application is shown;

[0029] Figure 6 The structural schematic diagram of a computer device that executes a method for predicting quality variables under an assisted training framework provided by an exemplary embodiment of the present application is shown. Detailed implementation manners

[0030] To make the objectives, technical solutions, and advantages of the present application clearer, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

[0031] First, the nouns that appear in the present application are explained:

[0032] Artificial Intelligence (AI) is to use a digital computer or a machine controlled by a digital computer to simulate, extend, and expand human intelligence, a theory, method, technology, and application system that perceives the environment, acquires knowledge, and uses knowledge to obtain the best results. Artificial Intelligence attempts to understand the essence of intelligence and produce an intelligent machine that can react in a way similar to human intelligence. The purpose of Artificial Intelligence is to enable machines to have the functions of perception, reasoning, and decision-making.

[0033] Artificial Intelligence technology is a comprehensive discipline involving a wide range of fields. The basic technologies of Artificial Intelligence include but are not limited to sensor technology, Artificial Intelligence chip technology, cloud computing technology, big data processing technology, and mechatronics technology. The Artificial Intelligence technology applied in the embodiments of the present application is machine learning technology, and this machine learning device is applied in a computer device.

[0034] Machine Learning (ML) is an interdisciplinary subject involving multiple disciplinary fields such as probability theory, statistics, and algorithm complexity theory. The Machine Learning discipline is specifically used to study how a computer simulates or realizes human learning behaviors so that the computer can acquire new knowledge, reorganize the existing knowledge structure, and thus improve its own performance. Machine Learning is usually combined with deep learning. Machine Learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and learning from demonstration.

[0035] Complex industrial processes are a special type of industrial processes that widely exist in fields such as oil refining and chemical engineering. They have characteristics such as multiple variables, strong coupling, strong nonlinearity, randomness, large time delays, output that cannot be measured online, and large changes in operating conditions, making it difficult to describe them with accurate mathematical models.

[0036] For different industrial processes, there are different variables that can best characterize the working state of the industrial process. Such variables are the quality variables of the industrial process. In complex industrial processes, it is difficult to monitor quality variables. In this case, a computer device can obtain operating condition data corresponding to a certain moment. After obtaining at least one parameter, generating a data set, and obtaining the quality variable value corresponding to the data set, this operating condition data is a labeled sample data set; after obtaining at least one parameter but not obtaining the quality variable value corresponding to the data set, this operating condition data is an unlabeled sample data set.

[0037] In this application, the working process of a debutanizer, the penicillin fermentation process, and the sulfur recovery unit which is a process for treating sulfur-containing gases will be listed as three typical examples of complex industrial processes.

[0038] A. Working process of a debutanizer

[0039] The debutanizer is one of the key equipment in the natural gas ethane recovery process, and its main function is to complete the separation of liquefied gas and stable light hydrocarbon components. During the working process of the debutanizer, the working states and working temperatures of its various parts will change. In one example, when the debutanizer is in the working state, its top temperature, top pressure, reflux flow rate, next-stage flow rate, tray temperature, and a total of two temperatures in different areas at the bottom of the tower will change. In the embodiments of this application, to detect the refining quality during the natural gas recovery of ethane, it is necessary to detect the butane content at the bottom outlet of the tower in real time. In this case, the methods for obtaining the butane concentration include but are not limited to the following two:

[0040] (1). By directly setting a butane concentration sensor at the bottom outlet of the tower, for example, and directly obtaining the butane concentration through physical methods.

[0041] (2). Based on the above seven parameters, establish a soft sensor model, input the above seven parameters into the soft sensor model, and output the predicted value of the butane concentration.

[0042] Since the discharged medium composition of the debutanizer is complex, and in actual application, it is difficult for the butane concentration sensor to continuously detect the concentration after being set. That is, during the working process of the debutanizer, the butane concentration is a quality variable.

[0043] In this case, the computer device can obtain the operating condition data corresponding to a certain moment. When obtaining the above seven parameters and the butane concentration data corresponding to the above seven parameters, the operating condition data is the labeled sample data set; when obtaining the above seven parameters but not obtaining the corresponding butane concentration data, the operating condition data is the unlabeled sample data set.

[0044] B. Penicillin fermentation process

[0045] The penicillin fermentation process refers to the metabolic activities of penicillin-producing bacteria for cell growth and antibiotic synthesis in a suitable environment, mainly including the growth stage of the producing bacteria, the penicillin synthesis stage, and the autolysis stage of the producing bacteria. During the penicillin fermentation process, nine parameters such as the mixer power (W), aeration rate (L / h), substrate feeding rate (L / h), feeding temperature (K), dissolved oxygen (mmole / L), pH value, fermenter temperature (K), carbon dioxide concentration (mmole / L), culture volume (L), and generated heat (calories) will all affect the penicillin concentration. In the embodiments of the present application, in order to reduce the production cost of the penicillin fermentation process and improve the product yield at the same time, it is necessary to achieve automatic control and optimization of the process as much as possible. However, these all depend on the on-line accurate measurement of key biological parameters such as the product concentration in the process. In this case, the methods for obtaining the penicillin concentration include but are not limited to the following two:

[0046] (1) Directly obtain the penicillin concentration through a sensor. However, the penicillin concentration obtained through the sensor has a time lag and is the penicillin concentration after fermentation is completed, and it is impossible to regulate the penicillin concentration that will be produced.

[0047] (2) Establish a soft measurement model based on at least one of the above nine parameters, input at least one of the above nine parameters into the soft measurement model, and output the predicted value of the penicillin concentration.

[0048] In this case, the computer device can obtain the operating condition data corresponding to a certain moment. When obtaining the above nine parameters and the penicillin concentration data corresponding to the above nine parameters, the operating condition data is the labeled sample data set; when obtaining the above nine parameters but not obtaining the corresponding penicillin concentration data, the operating condition data is the unlabeled sample data set.

[0049] C. Process of the sulfur recovery unit for treating sulfur-containing gas

[0050] The sulfur recovery unit is a device for treating sulfur-containing gases to prevent air pollution. The main variables in the sulfur recovery process are the H2S concentration and the SO2 concentration. The H2S concentration and the SO2 concentration cannot be directly measured by sensors in the industrial process. Therefore, an accurate soft-sensing model needs to be constructed to monitor the H2S concentration and the SO2 concentration in real time. Five parameters, namely the gas flow, the primary air flow, the secondary air flow, the gas flow in the preset position area, and the air flow in the preset position area, will affect the H2S concentration and the SO2 concentration. In the embodiment of the present application, the preset position indicates the SWS area. A soft-sensing model is established based on the above five parameters, and the above five parameters are input into the soft-sensing model to output the predicted values of the H2S concentration and the SO2 concentration. That is, in the process of treating sulfur-containing gases by the sulfur recovery unit, the quality variables include the H2S concentration and the SO2 concentration.

[0051] In this case, the computer device can obtain the working condition data corresponding to a certain moment. When obtaining the above five parameters and the penicillin concentration data corresponding to the above five parameters, the working condition data is the labeled sample data set; when obtaining the above five parameters but not obtaining the corresponding penicillin concentration data, the working condition data is the unlabeled sample data set.

[0052] Figure 1 The flowchart of a quality variable prediction method under an auxiliary training framework provided by an exemplary embodiment of the present application is shown. Please refer to Figure 1 , the method includes:

[0053] Step 101, obtain the dataset to be measured, the labeled sample dataset, and the unlabeled sample dataset.

[0054] In the embodiment of the present application, the labeled sample dataset includes at least two groups of labeled sample data sets and the sample quality variable values corresponding to the labeled sample data sets. The unlabeled sample dataset includes at least two groups of unlabeled sample data sets. The dataset to be measured includes at least two groups of data to be measured.

[0055] In the embodiment of the present application, the sample refers to the specific value of the quality variable corresponding to the data set. In some cases, the specific value of the quality variable can be obtained by an uncertain measurement method. The present application does not limit the generation method of the sample label.

[0056] In the embodiments of the present application, the dataset to be measured, the labeled sample dataset, and the unlabeled sample dataset all correspond to indicating the working conditions in complex industrial processes. Taking the working conditions of a debutanizer as an example for illustration, each dataset indicates seven parameters corresponding to a certain working condition, that is, the top temperature, the top pressure, the reflux flow rate, the flow rate of the next stage, the tray temperature, and two temperatures in different areas at the bottom of the tower. The embodiments of the present application do not limit the actual implementation form of the dataset in complex industrial processes, but each dataset needs to include parameters characterizing the working state of the chemical process.

[0057] Step 102, establish an initial main learning model and an initial auxiliary learning model corresponding to the initial main learning model.

[0058] In the embodiments of the present application, the initial auxiliary learning model is used for the preliminary annotation of unlabeled sample data, and the initial main learning model is used to expand the labeled sample dataset based on the results of the preliminary annotation of the sample set.

[0059] Optionally, both the initial auxiliary learning model and the initial main learning model are models constructed based on the initial parameters defined in the computer device, which are used for data collation and feature extraction of the labeled sample dataset and the unlabeled sample dataset. In the embodiments of the present application, the initial main learning model and the initial auxiliary learning model need to be used in series to collate the labeled sample dataset and the unlabeled sample dataset.

[0060] Step 103, train the initial main learning model and the initial auxiliary learning model through the unlabeled sample dataset and the labeled sample dataset to obtain the main learning model and the auxiliary learning model.

[0061] During the process of the initial auxiliary learning model and the initial main learning model performing their corresponding functions, the computer device will simultaneously train the models based on the accuracy and working speed of the sample division results. After the training is completed, the initial main learning model is trained to generate the main learning model, and the initial auxiliary learning model is trained to generate the auxiliary learning model. During the training process of the main learning model and the auxiliary learning model, multiple rounds of parameter adjustments have been carried out.

[0062] Step 104, establish a quality variable prediction model based on the main learning model and the auxiliary learning model.

[0063] In the embodiments of the present application, the quality variable prediction model is formed by connecting the main learning model and the auxiliary learning model in series. In other embodiments of the present application, the structural parameters corresponding to the quality variable prediction model are selected and constituted by the main learning model and the auxiliary learning model. The present application does not limit the source form of the structural parameters of the quality variable prediction model.

[0064] Step 105: Input the data set to be measured into the quality variable prediction model, and output the quality variable prediction result corresponding to the data set to be measured.

[0065] After the construction of the quality variable prediction model is completed, the data set to be measured is input into the quality variable prediction model, and the quality variable prediction result corresponding to the data set to be measured in the data set to be measured is output. During the working process of the debutanizer, the quality variable is the butane concentration value.

[0066] In the embodiment of the present application, after obtaining the prediction result of the quality variable, the data set corresponding to the quality variable can be determined according to the predicted quality variable, or a quality variable curve can be generated according to the predicted quality variable to determine the working state of the corresponding chemical process.

[0067] In summary, the method provided by the embodiment of the present application, in the process of determining the quality variable of a complex industrial process, for the situation where the number of labeled sample data sets is small and the number of unlabeled samples is large, before predicting the quality variable of the data set to be measured, an initial auxiliary learning model and an initial main learning model are respectively established, and while expanding the sample set, they are trained on their own, and finally a quality variable prediction model is generated to predict the quality variable of the data set to be measured. In the process of prediction, through the combined training of the pre - determined unlabeled sample data set and the labeled sample data set, based on the selection of the unlabeled sample data set with high global information content, the quality of the quality variable prediction model is improved, so that in the scenario where the number of unlabeled samples is large, there is a stable and specific way to predict the quality variable, and the accuracy and efficiency of predicting the quality variable are improved.

[0068] Figure 2 The flow chart of a quality variable prediction method under a co - training framework provided by an exemplary embodiment of the present application is shown. Please refer to Figure 2 This process includes:

[0069] Step 201: Obtain the data set to be measured, the labeled sample data set, and the unlabeled sample data set.

[0070] This process is the same as the process shown in step 101 and will not be elaborated here.

[0071] Step 202: Establish an initial main learning model and an initial auxiliary learning model corresponding to the initial main learning model.

[0072] This process is the same as the process shown in step 102 and will not be elaborated here.

[0073] It should be noted that in the embodiments of the present application, the auxiliary learning model is a machine learning model based on the K-Nearest Neighbor (KNN) algorithm, and the main learning model is a machine learning model based on the Twin Support Vector Regression (TSVR) algorithm.

[0074] Step 203: Randomly select at least two unlabeled sample data groups from the unlabeled sample dataset to obtain an unlabeled training dataset.

[0075] In the embodiments of the present application, after constructing the initial main learning model and the initial auxiliary learning model, steps 203 to 215 illustrate the adjustment process for the initial main learning model and the initial auxiliary learning model.

[0076] It should be noted that the construction process of the initial main learning model and the initial auxiliary learning model is a loop process. In the embodiments of the present application, when the training of the initial main learning model and the initial auxiliary learning model is not completed, the training process will be executed in a loop. In the embodiments of the present application, let the number of loops, that is, the number of iterations, be P.

[0077] Optionally, the process shown in step 203 determines the process of constructing a sample set for the initial auxiliary learning model based on the unlabeled sample dataset. In this process, a random selection method is used to generate a new sample set, that is, the unlabeled training dataset.

[0078] Step 204: Based on the unlabeled sample data groups in the unlabeled training set, select at least two nearest neighbor sample data groups of the unlabeled sample data groups from the labeled sample dataset.

[0079] In the embodiments of the present application, for each unlabeled sample data group in the unlabeled training set, it is necessary to select nearest neighbor samples to assign values to the samples in the unlabeled training set. In the embodiments of the present application, compared with the unlabeled samples in the unlabeled training set, at least two nearest neighbor sample data groups are determined through the labeled sample dataset in the embodiments of the present application.

[0080] Step 205: Generate a labeled training dataset based on at least two nearest neighbor sample data groups.

[0081] In the embodiments of the present application, the labeled training dataset is a sample dataset generated based on at least two nearest neighbor sample data groups.

[0082] Step 206: Train the initial auxiliary learning model through the labeled sample dataset and the unlabeled training dataset to obtain an auxiliary learning model.

[0083] Step 207: In response to the completion of the training of the auxiliary learning model, input the unlabeled sample data set into the auxiliary learning model, and output the confidence values corresponding to the unlabeled sample data set.

[0084] In the embodiments of the present application, the training of the auxiliary learning model can be performed by inputting a combination of a labeled sample data set and an unlabeled sample data set. Optionally, input one data set of the labeled sample data set or the unlabeled sample data set into the auxiliary learning model, and according to the output result of the auxiliary learning model, that is, the confidence value, determine the sample set to be input into the subsequent main training model.

[0085] Optionally, in the embodiments of the present application, the acquisition formula of the confidence value is as shown in Formula 1 below:

[0086] Formula 1:

[0087] In the formula, y i represents the true label value of the input x i , U indicates the set of neighboring samples of x u in the labeled sample data set M, k is the number of neighboring samples, h(x i ) and h′(x i ) are the predicted values of the initial auxiliary learning model for x u respectively, and the predicted value of the trained auxiliary learning model for x u .

[0088] Step 208: Screen at least two high-confidence unlabeled sample data sets from the unlabeled sample training set based on the confidence values.

[0089] After the output of the progressive confidence value, in the embodiments of the present application, at least two high-confidence unlabeled sample data sets with higher confidence values can be selected from the confidence results of the auxiliary learning model based on the descending order of the confidence values as part of the subsequent training sample set.

[0090] Step 209: Generate a main learning model training sample set based on the high-confidence unlabeled sample data sets.

[0091] In the embodiments of the present application, the high-confidence unlabeled sample data sets can be used as part of the main learning model training sample set. Optionally, the sample data sets in the main learning model training sample set include all the high-confidence unlabeled sample data sets, or the main learning model training sample set includes at least two high-confidence unlabeled sample data sets.

[0092] Step 210: Input the main learning model training sample set into the initial main learning model, and output a pseudo-label training sample set.

[0093] In the embodiment of the present application, the initial main learning model generates a pseudo-label training sample set for the main learning model training sample set.

[0094] Step 211: Expand the labeled data set with the pseudo-label training sample set.

[0095] In the embodiment of the present application, the sample set for training the main learning model includes, in addition to the pseudo-label training sample set, a labeled data set. Therefore, after obtaining the pseudo-label training sample set, the labeled data set is expanded.

[0096] Step 212: Input the labeled data groups in the expanded labeled data set into the initial main learning model, and output the mean squared error corresponding to the labeled data groups.

[0097] In the embodiment of the present application, the acquisition formula of the mean squared error is as shown in Formula 2 below:

[0098] Formula 2:

[0099] In Formula 2 shown above, (x i , y i ) is a sample in the labeled sample data set M, and y i represents the true label value of the input x i ; y i ′ is the pseudo-label added by the new main learner to x i , that is, y i ′ = g′(x i ).

[0100] Step 213: Train the initial main learning model based on the mean squared error to obtain the main learning model.

[0101] This process is the process of training the initial main learning model.

[0102] Step 214: Determine the first iteration number corresponding to the training times of the main learning model and the second iteration number corresponding to the training times of the auxiliary learning model.

[0103] Step 215: Responding to the first iteration number reaching the first number threshold, determine that the training of the main learning model is completed.

[0104] Step 216: Responding to the second iteration number reaching the second number threshold, determine that the training of the auxiliary learning model is completed.

[0105] The process shown in steps 214 to 216 is the process of determining whether the main learning model and the auxiliary learning model are sufficiently trained. In the embodiment of the present application, the computer device sets a first number threshold corresponding to the training situation of the main learning model, and sets a second number threshold corresponding to the training situation of the auxiliary learning model. When the training times of the main learning model reach the first number threshold and the training times of the auxiliary learning model reach the second number threshold, it is determined that the two models are trained.

[0106] It should be noted that the number threshold can be a value pre-stored in the computer device, or a value determined based on the received signal during the training of the model by the computer device. The present application does not limit the actual determination form of the number threshold.

[0107] Step 217, establish a quality variable prediction model based on the main learning model and the auxiliary learning model.

[0108] This process is the process of constructing a quality variable prediction model based on the parameters of the main learning model and the auxiliary learning model after the main learning model and the auxiliary learning model are trained.

[0109] Step 218, input the data group to be measured into the quality variable prediction model, and output the quality variable prediction result corresponding to the data group to be measured.

[0110] This process corresponds to the process shown in step 105 and will not be elaborated here.

[0111] In summary, the method provided by the embodiment of the present application, in the process of determining the quality variables of complex industrial processes, in view of the situation where the number of labeled sample data groups is small and the number of unlabeled samples is large, before predicting the quality variables of the data set to be measured, an initial auxiliary learning model and an initial main learning model are established respectively, while expanding the sample set and training itself, and finally a quality variable prediction model is generated to predict the quality variables of the data set to be measured. In the process of prediction, through the combined training of the pre-prepared unlabeled sample data set and the labeled sample data set, on the basis of selecting the unlabeled sample data set with high global information content, the quality of the quality variable prediction model is improved, so that in the scenario where the number of unlabeled samples is large, there is a stable and specific way to predict the quality variables, and the accuracy and efficiency of predicting the quality variables are improved.

[0112] Figure 3 The process diagram of a quality variable prediction method under an assisted training framework provided by an exemplary embodiment of the present application is shown. Please refer to Figure 3 , and this process includes:

[0113] Step 301, initialization.

[0114] This process is to perform corresponding suggestions and parameter initialization on the initial main learning model and the initial auxiliary learning.

[0115] Step 302: Obtain the unlabeled sample set N.

[0116] Step 303: Obtain the labeled sample set M.

[0117] Step 304: Obtain the sample H.

[0118] The processes shown in steps 302 to 304 are respectively the processes of obtaining the labeled sample data set, the unlabeled sample data set, and the sample set to be tested.

[0119] Step 305: Randomly select n unlabeled sample data groups to form the sample set N'.

[0120] This process is the process of generating unlabeled sample data groups.

[0121] Step 306: Train the auxiliary learner.

[0122] This process is to train the initial auxiliary learning model to obtain the auxiliary learning model.

[0123] Step 307: Evaluate the confidence of the auxiliary learner.

[0124] When the training of the auxiliary learning model is completed, or when the training of the auxiliary learning model reaches a certain training level, determine the confidence related to the sample input to itself through the auxiliary learning model.

[0125] Step 308: Train the main learner.

[0126] This process is to train the main learning model to obtain the main learning model.

[0127] Step 309: Evaluate the confidence of the main learner.

[0128] When the training of the main learning model is completed, or when the training of the main learning model reaches a certain training level, determine the confidence related to the sample input to itself through the main learning model.

[0129] Step 310: Obtain the prediction model.

[0130] This process is to suggest and improve the quality variable prediction model.

[0131] Step 311: Predict the sample H.

[0132] This process is to input the data group to be tested into the quality variable prediction model and output the quality variable prediction result corresponding to the data group to be tested.

[0133] Step 312: Exclude the sample data group corresponding to the prediction sample H from the unlabeled sample set N.

[0134] Step 313: Add the sample data group corresponding to the prediction sample H to the labeled sample set M.

[0135] The processes shown in Step 312 and Step 313 are the processes of adjusting the data set content in the labeled sample data set and the unlabeled sample data set based on the prediction results of the samples.

[0136] Step 314: Determine whether the number of iterations is greater than the iteration number threshold.

[0137] In the embodiments of the present application, this process is the process of determining whether the adjustment of the quality variable prediction model is perfect. In one example, when the number of iterations is greater than the iteration number threshold, the process ends; when the number of iterations is less than the iteration number threshold, the training process described in the embodiments of this process is repeatedly executed.

[0138] In summary, in the quality variable prediction process shown in the embodiments of the present application, in the process of determining the quality variables in a complex industrial process, in the case where the number of labeled sample data groups is small and the number of unlabeled samples is large, before predicting the quality variables of the data set to be measured, an initial auxiliary learning model and an initial main learning model are respectively established, and while expanding the sample set, self-training is performed, and finally a quality variable prediction model is generated to predict the quality variables of the data set to be measured. In the prediction process, through the combined training of the pre-prepared unlabeled sample data set and the labeled sample data set, based on the selection of the unlabeled sample data set with high global information content, the quality of the quality variable prediction model is improved, so that in the scenario where the number of unlabeled samples is large, there is a stable and specific way to predict the quality variables, and the accuracy and efficiency of predicting the quality variables are improved.

[0139] Figure 4 The block diagram of the structure of a quality variable prediction device under an assisted training framework provided by an exemplary embodiment of the present application is shown. The device includes:

[0140] An acquisition module 401, configured to acquire a data set to be measured, a labeled sample data set, and an unlabeled sample data set. The labeled sample data set includes at least two groups of labeled sample data groups and the corresponding sample quality variable values, the unlabeled sample data set includes at least two groups of unlabeled sample data groups, and the data set to be measured includes at least two groups of data to be measured;

[0141] The establishment module 402 is used to establish an initial main learning model and an initial auxiliary learning model corresponding to the initial main learning model. The initial auxiliary learning model is used for the preliminary annotation of unlabeled sample data, and the initial main learning model is used to expand the labeled sample data set based on the results of the preliminary annotation.

[0142] The training module 403 is used to train the initial main learning model and the initial auxiliary learning model through the unlabeled sample data set and the labeled sample data set to obtain the main learning model and the auxiliary learning model.

[0143] The establishment module 402 is also used to establish a quality variable prediction model based on the main learning model and the auxiliary learning model.

[0144] The input module 404 is used to input the data group to be measured into the quality variable prediction model and output the quality variable prediction result corresponding to the data group to be measured.

[0145] In an optional embodiment, please refer to Figure 5 , the device further includes a selection module 405, which is used to randomly select at least two unlabeled sample data groups from the unlabeled sample data set to obtain an unlabeled training data set.

[0146] The training module 403 is also used to train the initial auxiliary learning model through the labeled sample data set and the unlabeled training data set to obtain the auxiliary learning model.

[0147] The device further includes a determination module 406, which is used to, in response to the completion of the training of the auxiliary learning model, determine the main learning model training sample set from the unlabeled training data set through the auxiliary learning training model.

[0148] Train the initial main learning model through the main learning model training sample set to obtain the main learning model.

[0149] In an optional embodiment, the selection module 405 is also used to select a labeled training data set corresponding to the unlabeled training data set from the labeled sample data set. In the labeled training data set, for each unlabeled sample data group in the unlabeled training data set, there is a labeled sample data subset, and the labeled sample data subset includes the training sample quality variable values.

[0150] The training module 403 is also used to train the initial auxiliary learning model based on the sample quality variable values and the unlabeled sample data groups to obtain the auxiliary learning model.

[0151] In an optional embodiment, the selection module 405 is also used to select at least two neighboring sample data groups of the unlabeled sample data group in the labeled sample data set based on the unlabeled sample data groups in the unlabeled training set.

[0152] The device further includes a generation module 407, configured to generate a labeled training data set based on at least two neighboring sample data groups.

[0153] In an optional embodiment, the input module 404 is further configured to, in response to the completion of the training of the auxiliary learning model, input unlabeled sample data into the auxiliary learning model and output a confidence value corresponding to the unlabeled sample data group.

[0154] The device further includes a screening module 408, configured to screen at least two high-confidence unlabeled sample data groups from the unlabeled sample training set based on the confidence value.

[0155] The generation module 407 is further configured to generate a main learning model training sample set based on the high-confidence unlabeled sample data groups.

[0156] In an optional embodiment, the input module 404 is further configured to input the main learning model training sample set into the initial main learning model and output a pseudo-labeled training sample set.

[0157] The device further includes an expansion module 409, configured to expand the labeled data set through the pseudo-labeled training sample set.

[0158] Input the labeled data groups in the expanded labeled data set into the initial main learning model, and output a mean square error corresponding to the labeled data groups.

[0159] The training module 403 is further configured to train the initial main learning model based on the mean square error to obtain the main learning model.

[0160] In an optional embodiment, the determination module 406 is further configured to determine a first iteration number corresponding to the training times of the main learning model and a second iteration number corresponding to the training times of the auxiliary learning model.

[0161] In response to the first iteration number reaching the first number threshold, determine that the training of the main learning model is completed.

[0162] In response to the second iteration number reaching the second number threshold, determine that the training of the auxiliary learning model is completed.

[0163] In summary, in the process of determining the quality variables of a complex industrial process, for the device provided in the embodiment of the present application, in the case where the number of labeled sample data sets is small and the number of unlabeled samples is large, before predicting the quality variables of the data set to be measured, an initial auxiliary learning model and an initial main learning model are established respectively. While expanding the sample set, the models are trained on their own, and finally a quality variable prediction model is generated to predict the quality variables of the data set to be measured. In the process of prediction, through the combined training of the pre-prepared unlabeled sample data set and the labeled sample data set, based on the selection of the unlabeled sample data set with high global information content, the quality of the quality variable prediction model is improved, so that in the scenario where the number of unlabeled samples is large, there is a stable and specific way to predict the quality variables, and the accuracy and efficiency of predicting the quality variables are improved.

[0164] It should be noted that: for the quality variable prediction device under the assisted training framework provided in the above embodiment, only the above-mentioned division of each functional module is used as an example for illustration. In actual application, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above.

[0165] Figure 6 The figure shows a schematic structural diagram of a computer device that executes the quality variable prediction method under the assisted training framework provided by an exemplary embodiment of the present application. The computer device includes:

[0166] The processor 601 includes one or more processing cores. The processor 601 executes various functional applications and data processing by running software programs and modules.

[0167] The receiver 602 and the transmitter 603 can be implemented as a communication component, and the communication component can be a communication chip. Optionally, the communication component can be implemented to include a signal transmission function. That is, the transmitter 603 can be used to transmit control signals to the image acquisition device and the scanning device, and the receiver 602 can be used to receive corresponding feedback instructions.

[0168] The memory 604 is connected to the processor 601 through the bus 605.

[0169] The memory 604 can be used to store at least one instruction, and the processor 601 is used to execute the at least one instruction to implement each step in the above method embodiment.

[0170] The embodiment of the present application also provides a computer-readable storage medium, in which at least one instruction, at least one segment of program, code set or instruction set is stored, and is loaded and executed by a processor to implement the quality variable prediction method under the above-mentioned assisted training framework.

[0171] The present application also provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the quality variable prediction method under the assistance training framework described in any one of the foregoing embodiments.

[0172] Optionally, the computer-readable storage medium may include: read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), solid state drives (SSD, Solid State Drives), or optical discs, etc. Among them, the random access memory may include resistive random access memory (ReRAM, Resistance Random Access Memory) and dynamic random access memory (DRAM, Dynamic Random Access Memory). The serial numbers of the foregoing embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments.

[0173] Those of ordinary skill in the art can understand that all or part of the steps for implementing the foregoing embodiments can be completed by hardware, or can be completed by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and the storage medium mentioned above can be read-only memory, a magnetic disk, or an optical disc, etc.

[0174] The foregoing are only optional embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting quality variables under an assistance training framework, characterized in that, The method is applied to a computer device, and the method includes: Obtain a dataset to be measured, a labeled sample dataset, and an unlabeled sample dataset. The labeled sample dataset includes at least two groups of labeled sample data groups, and corresponding sample quality variable values. The unlabeled sample dataset includes at least two groups of unlabeled sample data groups. The dataset to be measured includes at least two groups of data to be measured; Establish an initial main learning model and an initial auxiliary learning model corresponding to the initial main learning model. The initial auxiliary learning model is used for preliminary annotation of the unlabeled sample data, and the initial main learning model is used to expand the labeled sample dataset based on the results of the preliminary annotation; Randomly select at least two unlabeled sample data groups from the unlabeled sample dataset to obtain an unlabeled training dataset; Train the initial auxiliary learning model through the labeled sample dataset and the unlabeled training dataset to obtain the auxiliary learning model; In response to the completion of the training of the auxiliary learning model, input the unlabeled sample data into the auxiliary learning model, and output the confidence value corresponding to the unlabeled sample data group; Screen at least two high-confidence unlabeled sample data groups from the unlabeled training dataset based on the confidence value; Generate a main learning model training sample set based on the high-confidence unlabeled sample data groups; Input the main learning model training sample set into the initial main learning model, and output a pseudo-label training sample set; Expand the labeled sample dataset through the pseudo-label training sample set; Input the labeled sample data groups in the expanded labeled sample dataset into the initial main learning model, and output the mean square error corresponding to the labeled sample data groups; Train the initial main learning model based on the mean square error to obtain the main learning model; Establish a quality variable prediction model based on the main learning model and the auxiliary learning model; Input the data group to be measured into the quality variable prediction model, and output the quality variable prediction result corresponding to the data group to be measured.

2. The method according to claim 1, wherein The training of the initial auxiliary learning model through the labeled sample dataset and the unlabeled training dataset to obtain the auxiliary learning model includes: Select a labeled training dataset corresponding to the unlabeled training dataset from the labeled sample dataset. In the labeled training dataset, for each unlabeled sample data group in the unlabeled training dataset, there is a labeled sample data subset, and the labeled sample data subset includes training sample quality variable values; Train the initial auxiliary learning model based on the sample quality variable values and the unlabeled sample data groups to obtain the auxiliary learning model.

3. The method according to claim 2, wherein The selection of the labeled training dataset corresponding to the unlabeled training dataset from the labeled sample dataset includes: Based on the unlabeled sample data groups in the unlabeled training set, select at least two neighboring sample data groups corresponding to the unlabeled sample data groups in the labeled sample data set; Generate the labeled training data set based on the at least two neighboring sample data groups.

4. The method according to claim 1, characterized in that, The method further includes: Determine a first iteration number corresponding to the number of training times of the main learning model, and a second iteration number corresponding to the number of training times of the auxiliary learning model; In response to the first iteration number reaching a first number threshold, determine that the training of the main learning model is completed; In response to the second iteration number reaching a second number threshold, determine that the training of the auxiliary learning model is completed.

5. A prediction device for the quality variable prediction method under the assistance training framework described in claim 1, characterized in that, The device includes: An acquisition module, configured to acquire a data set to be measured, a labeled sample data set, and an unlabeled sample data set. The labeled sample data set includes at least two groups of labeled sample data groups, and sample quality variable numerical values corresponding to the labeled sample data groups. The unlabeled sample data set includes at least two groups of unlabeled sample data groups, and the data set to be measured includes at least two groups of data groups to be measured; A building module, configured to build an initial main learning model and an initial auxiliary learning model corresponding to the initial main learning model. The initial auxiliary learning model is used for preliminary annotation of the unlabeled sample data, and the initial main learning model is used for expanding the labeled sample data set based on the results of the preliminary annotation; A training module, configured to train the initial main learning model and the initial auxiliary learning model through the unlabeled sample data set and the labeled sample data set to obtain a main learning model and an auxiliary learning model; The building module is further configured to build a quality variable prediction model based on the main learning model and the auxiliary learning model; An input module, configured to input the data group to be measured into the quality variable prediction model, and output a quality variable prediction result corresponding to the data group to be measured.

6. A computer device, characterized in that, The computer device includes a processor and a memory. At least one instruction, at least one program, a code set, or an instruction set is stored in the memory. The at least one instruction, at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the quality variable prediction method under the assisted training framework according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, At least one instruction, at least one program, a code set, or an instruction set is stored in the readable storage medium. The at least one instruction, at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the quality variable prediction method under the assisted training framework according to any one of claims 1 to 4.

Images