Data prediction method, apparatus, device, storage medium, and program product

Abstract

The application provides a data prediction method and device, equipment, a storage medium and a program product. A plurality of prediction models are obtained, the plurality of prediction models are obtained through machine learning training based on a first sample data set, and the structures of at least two prediction models are different. After receiving a prediction instruction, weight coefficients of the plurality of prediction models are determined, the weight coefficients of the plurality of prediction models are weight values obtained through weight optimization solving of the plurality of prediction models based on a second sample data set. The plurality of prediction models are used to respectively perform target prediction on to-be-predicted data, model prediction values respectively output by the plurality of prediction models are obtained, the model prediction values of the plurality of prediction models are calculated through weighted summation based on the weight coefficients of the prediction models, and the obtained target prediction value is output as a prediction result corresponding to the to-be-predicted data. The prediction ability of the model combination for unknown data can be enhanced, and therefore the prediction accuracy and the decision-making ability of the system are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a data prediction method and device, equipment, a storage medium and a program product. BACKGROUND

[0002] With the development of artificial intelligence and big data technology, machine learning models have been widely used in financial prediction, industrial control, medical diagnosis, intelligent manufacturing and many other fields. In order to improve the understanding and prediction ability of complex data, deep machine learning and other models are widely used to build prediction systems. However, in practical applications, the accuracy of machine learning models often fails to meet the actual prediction requirements, resulting in insufficient prediction accuracy of the model for data, thereby affecting the decision-making ability of the system based on the prediction results. SUMMARY

[0003] The present application provides a data prediction method, device, equipment, storage medium and program product to solve the problem that the model accuracy in related technologies fails to meet the actual prediction requirements, resulting in insufficient prediction accuracy of the model, and affecting the decision-making ability of the system.

[0004] In a first aspect, the embodiments of the present application provide a data prediction method, comprising:

[0005] Obtaining a plurality of prediction models trained in advance, the plurality of prediction models being obtained based on machine learning training of a first sample data set, and structures of at least two prediction models in the plurality of prediction models being different;

[0006] After receiving a prediction instruction for to-be-predicted data, determining weight coefficients of the plurality of prediction models, the weight coefficients of the plurality of prediction models being weight values obtained by performing weight optimization solving on the plurality of prediction models based on a second sample data set;

[0007] Using the plurality of prediction models to respectively perform target prediction on the to-be-predicted data, and obtaining model prediction values respectively output by the plurality of prediction models;

[0008] Based on the weight coefficients of the prediction models, performing weighted summation calculation on the model prediction values respectively output by the plurality of prediction models, and obtaining a target prediction value;

[0009] Outputting the target prediction value as a prediction result corresponding to the to-be-predicted data.

[0010] In a second aspect, the embodiments of the present application provide a data prediction device, comprising:

[0011] An obtaining module is configured to obtain a plurality of prediction models trained in advance, the plurality of prediction models being obtained based on machine learning training of a first sample data set, and structures of at least two prediction models in the plurality of prediction models being different;

[0012] determining a weight coefficient of each of the plurality of prediction models based on the second sample data set, the weight coefficient being a weight value obtained by performing weight optimization on the plurality of prediction models;

[0013] performing target prediction on the to-be-predicted data by using the plurality of prediction models respectively to obtain model prediction values output by the plurality of prediction models respectively;

[0014] performing weighted summation calculation on the model prediction values output by the plurality of prediction models respectively based on the weight coefficient of each of the prediction models to obtain a target prediction value;

[0015] outputting the target prediction value as a prediction result corresponding to the to-be-predicted data, and processing the to-be-predicted data based on the target prediction value.

[0016] In a third aspect, an embodiment of the present application provides a computer device, which comprises a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and the processor implements the steps of the data prediction method when executing the computer program.

[0017] In a fourth aspect, an embodiment of the present application provides a readable storage medium, which stores a computer program, and the computer program implements the steps of the data prediction method when executed by a processor.

[0018] In a fifth aspect, an embodiment of the present application provides a computer program product, which comprises a computer program, and the computer program makes the data prediction method be executed when being run.

[0019] In one solution provided by the aforementioned data prediction method, apparatus, device, storage medium, and program product, multiple pre-trained prediction models are acquired. These multiple prediction models are trained using machine learning on a first sample dataset, and at least two of the prediction models have different structures. Upon receiving a prediction instruction for the data to be predicted, the weight coefficients of the multiple prediction models are determined. These weight coefficients are weight values ​​obtained by optimizing the weights of the multiple prediction models based on a second sample dataset. The multiple prediction models are used to perform target prediction on the data to be predicted, resulting in model prediction values ​​output by each model. Based on the weight coefficients of each prediction model, the model prediction values ​​output by the multiple prediction models are weighted and summed to obtain the target prediction value, which is then output as the prediction result corresponding to the data to be predicted. On the one hand, using multiple heterogeneous prediction models for prediction can complement each other's strengths and weaknesses in prediction performance. By integrating the prediction results of multiple prediction models, the prediction bias and overfitting risk of a single model can be reduced, improving overall prediction performance and accuracy. On the other hand, by optimizing the model weights on the second sample dataset, the importance of each model in the weighted combination can be dynamically adjusted according to the actual situation. The prediction process is more flexible and adaptable, making the prediction results better adapt to new data distributions and various evaluation indicators. This enhances the ability of the model combination to predict unknown data, thereby improving prediction accuracy and the system's decision-making ability. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the structure of a data prediction system in one embodiment of the present invention;

[0022] Figure 2 This is a flowchart illustrating a data prediction method in one embodiment of the present invention;

[0023] Figure 3 yes Figure 2 A schematic diagram of the implementation process of step S20;

[0024] Figure 4 This is another flowchart illustrating the data prediction method in one embodiment of the present invention;

[0025] Figure 5 This is a schematic diagram of the training process of multiple prediction models in one embodiment of the present invention;

[0026] Figure 6 is a curve diagram of the mean square error of the optimization weight method of the plurality of prediction models trained in an embodiment of the present application and other models;

[0027] Figure 7 is a curve diagram of the determination coefficient of the optimization weight method of the plurality of prediction models trained in an embodiment of the present application and other models;

[0028] Figure 8 is a flowchart of the acquisition process of the weight values of the plurality of prediction models in an embodiment of the present application;

[0029] Figure 9 is Figure 8 is a flowchart of the implementation of step S7 in the method;

[0030] Figure 10 is Figure 1 is a structural diagram of the data prediction device in the method;

[0031] Figure 11 is a structural diagram of the computer device in an embodiment of the present application. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0033] It should be understood that when used in the specification and the appended claims of the present application, the term “comprising” indicates the presence of the described features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should also be understood that the term “and / or” refers to any combination of one or more of the associated listed items and all possible combinations thereof, and includes these combinations. In addition, the terms “first”, “second”, “third”, etc. are only used to distinguish descriptions, and cannot be understood as indicating or implying relative importance.

[0034] In the specification of the present application, the reference “one embodiment” means that the specific features, structures or characteristics described in connection with the embodiment are included in one or more embodiments of the present application. The terms “including”, “containing”, “having” and their variants mean “including but not limited to”, unless otherwise specifically emphasized.

[0035] It should be understood that the size of the serial number of each step in the following embodiments does not mean the order of execution, and the execution order of each process should be determined by its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0036] In order to illustrate the technical solutions of the present application, the following will be illustrated by specific embodiments.

[0037] It should be understood that with the development of artificial intelligence and big data technology, machine learning models based on deep learning have been widely used in financial prediction, industrial control, medical diagnosis, intelligent manufacturing and many other fields. In order to improve the understanding and prediction ability of complex data, deep machine learning and other models are widely used to build prediction systems. However, in practical applications, the precision of machine learning models often fails to meet the actual prediction needs, resulting in insufficient prediction accuracy of the model for data, thereby affecting the decision-making ability of the system based on the prediction results.

[0038] Among them, the reasons for the low precision of machine learning models mainly include the following aspects:

[0039] 1. Single model structure: when facing complex data with high nonlinearity, time dynamics or cross-domain characteristics, the modeling ability of single-structure machine learning models has bottlenecks, and it is difficult to capture all effective features in the data, resulting in poor performance of the trained model and insufficient prediction accuracy;

[0040] 2. Limited training data: models usually rely on a large amount of high-quality training data to learn data distribution characteristics, but in actual scenarios, especially in the field of optics, it is often difficult to obtain sufficient measured data as training samples, and the training samples of the model often have noise, imbalance or limited coverage, resulting in insufficient precision of the trained model and inability to accurately generalize to unseen data;

[0041] 3. Overfitting and underfitting problems are common: in order to pursue higher training precision, the model is prone to overfitting, that is, it performs well on the training set, but the prediction ability decreases in the test set or actual data; while a model with too simple structure may lead to underfitting due to insufficient learning ability;

[0042] 4. Difficulty in model updating and maintenance: in the scenario where data distribution changes over time, if the static trained model cannot be updated or adapted to new data in time, its prediction accuracy will also gradually decrease.

[0043] To solve the above problems, the embodiment of the present application provides a data prediction method, device, equipment, storage medium and program product. The method comprises the following steps: obtaining a plurality of prediction models pre-trained, the plurality of prediction models are obtained by machine learning training based on a first sample data set, and the structures of at least two prediction models in the plurality of prediction models are different; after receiving a prediction instruction for to-be-predicted data, determining weight coefficients of the plurality of prediction models, the weight coefficients of the plurality of prediction models are weight values obtained by weight optimization solving of the plurality of prediction models based on a second sample data set; using the plurality of prediction models to respectively perform target prediction on the to-be-predicted data, to obtain model prediction values respectively output by the plurality of prediction models; performing weighted summation calculation on the model prediction values respectively output by the plurality of prediction models based on the weight coefficients of the prediction models, to obtain a target prediction value, and outputting the target prediction value as a prediction result corresponding to the to-be-predicted data, so as to perform relevant processing on the to-be-predicted data and a target object corresponding thereto based on the target prediction value. On the one hand, since different structures of models may have different prediction effects on the same data, the embodiment adopts a plurality of heterogeneous prediction models for prediction, can complement the advantages and disadvantages of different models in prediction performance, and can reduce the prediction deviation and overfitting risk of a single model by integrating the prediction results of the plurality of prediction models, so as to improve the overall prediction performance and prediction accuracy. On the other hand, by optimizing the weight of the model on the second sample data set, the importance of each model in the weighted combination can be dynamically adjusted according to the actual situation, the prediction process is more flexible and adaptive, the prediction result better adapts to new data distribution and various evaluation indexes, the prediction ability of the model combination on unknown data can be enhanced, and thus the prediction accuracy and the decision-making ability of the system are improved. In addition, the plurality of heterogeneous prediction models are used for synchronous prediction, and then the final prediction value is obtained by using different weights for weighted summation, which can reduce the dependence of the model on the training sample. In the model training process, a plurality of prediction models can be trained by using a small amount of training sample, and an accurate prediction result can be obtained by using the multi-model prediction and weight weighting method subsequently, the influence of the training sample on the model precision is reduced due to noise, imbalance or limited coverage, and the like, the multi-model set can be accurately generalized to unseen data, and the prediction accuracy and the decision-making ability of the system are improved.

[0044] The data prediction method provided by the embodiment of the present application can be applied to a data prediction system as shown in the figure. Figure 1 The data prediction system includes a server and a terminal device. The terminal device communicates with the server through a network. The terminal device includes but is not limited to various personal computers, notebook computers, smart phones, tablet computers and portable wearable devices, such as an optical data acquisition device. The server can be implemented by an independent server or a server cluster composed of multiple servers.

[0045] In actual application, the server can obtain a plurality of pre-trained prediction models, the plurality of pre-trained prediction models are trained based on a first sample data set, and the plurality of pre-trained prediction models are deployed to the local device for subsequent calling. At least two prediction models in the plurality of pre-trained prediction models are different in structure. In other embodiments, a plurality of pre-trained prediction models can also be trained by other devices, and the plurality of pre-trained prediction models are sent to the server, so that the server deploys the plurality of pre-trained prediction models to the local device for subsequent calling. That is, the server can obtain a plurality of pre-trained prediction models.

[0046] In the model application process, that is, in the process of executing the prediction instruction by the server, the terminal device obtains to-be-predicted data, and sends a prediction instruction for the to-be-predicted data to the server. After receiving the prediction instruction for the to-be-predicted data, the server determines the weight coefficients of the plurality of pre-trained prediction models. The weight coefficients of the plurality of pre-trained prediction models are weight values obtained by performing weight optimization on the plurality of pre-trained prediction models based on a second sample data set. Then, the server uses the plurality of pre-trained prediction models to respectively perform target prediction on the to-be-predicted data, to obtain model prediction values respectively output by the plurality of pre-trained prediction models. The model prediction values respectively output by the plurality of pre-trained prediction models are weighted and summed based on the weight coefficients of the respective prediction models, to obtain a target prediction value. The target prediction value is output as a prediction result corresponding to the to-be-predicted data, so that the to-be-predicted data and a target object corresponding thereto are processed based on the target prediction value.

[0047] In this embodiment, the plurality of heterogeneous prediction models are used for prediction, which can complement the advantages and disadvantages of different models in prediction performance. By integrating the prediction results of the plurality of prediction models, the prediction deviation and overfitting risk of a single model can be reduced, and the overall prediction performance and prediction accuracy can be improved. On the other hand, by optimizing the weights of the models on the second sample data set, the importance of each model in the weighted combination can be dynamically adjusted according to the actual situation. The prediction process is more flexible and adaptive, so that the prediction result better adapts to new data distribution and various evaluation indexes, and the prediction ability of the model combination for unknown data can be enhanced, thereby improving the prediction accuracy and the decision-making ability of the system. In addition, a plurality of prediction models can be trained using a small amount of training samples, and accurate prediction results can be obtained through multi-model prediction and weight weighting in subsequent processes. The influence of training samples on model precision due to noise, imbalance, or limited coverage range can be reduced. The multi-model set can accurately generalize to unseen data, thereby improving the prediction accuracy and the decision-making ability of the system.

[0048] In an embodiment, as shown in FIG. 1, Figure 2 A data prediction method is provided, which is applied to a data prediction device. The data prediction device can be a server. Figure 1The method is performed by a terminal device in the system. The method specifically includes the following steps.

[0049] S10: Obtain a plurality of prediction models pre-trained. The plurality of prediction models are obtained by machine learning training based on a first sample data set.

[0050] In actual application, the server can obtain a plurality of prediction models pre-trained. Structures of at least two prediction models in the plurality of prediction models are different. The server can pre-train a plurality of machine learning models with different structures based on a first sample data set, obtain a plurality of prediction models, and deploy the plurality of prediction models to a local device for subsequent calling. In other embodiments, a plurality of prediction models pre-trained by other devices can be sent to the server, so that the server deploys the plurality of prediction models to a local device for subsequent calling.

[0051] The plurality of prediction models can include at least two types of models such as a support vector regression model, a random forest model, an eXtreme Gradient Boosting (XGBoost) model, and other machine learning models. In other embodiments, the plurality of prediction models can also be a plurality of neural network models with different network structures; or the plurality of prediction models can include at least one of a support vector regression model, a random forest model, and an XGBoost model, and at least one neural network model in a plurality of neural network models with different network structures. The plurality of neural network models with different network structures can include at least two types of models such as a feedforward neural network model, a convolutional neural network model, a recurrent neural network model, and a long short-term memory network model.

[0052] S20: After receiving a prediction instruction for to-be-predicted data, determine weight coefficients of the plurality of prediction models. The weight coefficients of the plurality of prediction models are weight values obtained by weight optimization solving of the plurality of prediction models based on a second sample data set.

[0053] After obtaining the plurality of prediction models pre-trained, the server receives a prediction instruction for to-be-predicted data and obtains the to-be-predicted data. The user can upload the to-be-predicted data through a terminal device and send a prediction instruction for the to-be-predicted data to the server through the terminal device. The server receives the prediction instruction and obtains the to-be-predicted data carried by the prediction instruction.

[0054] In other embodiments, the terminal device can also have a data collection function. In actual application, the terminal device collects data of the target object in real time to obtain to-be-predicted data, and then the terminal device sends the to-be-predicted data and a prediction instruction for the to-be-predicted data to the server, and the server receives the prediction instruction and obtains the to-be-predicted data.

[0055] After receiving the prediction instruction for the to-be-predicted data, the server determines the weight coefficients of the plurality of prediction models. The weight coefficients of the plurality of prediction models are weight values obtained by performing weight optimization solving on the plurality of prediction models based on a second sample data set. The sample data in the second sample data set is different (may also be the same) from the sample data in the first sample data set and is derived from the same historical data sample set.

[0056] By actively solving the optimal weight combination of the plurality of heterogeneous prediction models on the second sample data set using a mathematical programming method, the prediction error caused by a single weight or a fixed weight is reduced, the accuracy of different prediction models is improved, and therefore the prediction accuracy based on weight weighted integration is improved, and the model fitting capability and generalization performance in a small sample scenario are improved.

[0057] S30: Adopting the plurality of prediction models to respectively perform target prediction on the to-be-predicted data to obtain model prediction values respectively output by the plurality of prediction models.

[0058] After receiving the prediction instruction for the to-be-predicted data, the server adopts the plurality of prediction models to respectively perform target prediction on the to-be-predicted data to obtain model prediction values respectively output by the plurality of prediction models. That is, the to-be-predicted data is respectively input into the plurality of prediction models, so that each prediction model respectively performs target prediction on the input to-be-predicted data and outputs a model prediction value of the to-be-predicted data, thereby obtaining the model prediction values output by the plurality of prediction models.

[0059] S40: Based on the weight coefficients of the prediction models, performing weighted summation calculation on the model prediction values respectively output by the plurality of prediction models to obtain a target prediction value.

[0060] After determining the weight coefficients of the plurality of prediction models and the model prediction values output by the prediction models, the server performs weighted summation calculation on the model prediction values respectively output by the plurality of prediction models based on the weight coefficients of the prediction models to obtain a target prediction value.

[0061] The target prediction value is calculated by the following formula:

[0062] ;

[0063] wherein, represents the target prediction value; representing the prediction model outputted model prediction value representing the prediction model weighting coefficient representing the number of prediction models .

[0064] S50: output the target prediction value as the prediction result corresponding to the to-be-predicted data.

[0065] After obtaining the target prediction value, the server outputs the target prediction value as the prediction result corresponding to the to-be-predicted data, so that the user can know the prediction result of the to-be-predicted data in time, and send a processing instruction for the to-be-predicted data or the target object corresponding to the to-be-predicted data, so that the server processes the to-be-predicted data or the target object corresponding to the to-be-predicted data in response to the processing instruction, and improves the decision-making ability of the system based on the prediction result.

[0066] In this embodiment, multiple heterogeneous prediction models are used for prediction, which can complement the advantages and disadvantages of different models in prediction performance. By integrating the prediction results of multiple prediction models, the prediction deviation and overfitting risk of a single model can be reduced, and the overall prediction performance and prediction accuracy can be improved. On the other hand, by optimizing the weights of the models on the second sample data set, the importance of each model in the weighted combination can be dynamically adjusted according to the actual situation, and the prediction process is more flexible and adaptive, so that the prediction result better adapts to new data distribution and multiple different evaluation indexes, which can enhance the prediction ability of the model combination for unknown data, thereby improving the prediction accuracy and the decision-making ability of the system. In addition, a small amount of training samples can be used to train multiple prediction models, and accurate prediction results can be obtained through multi-model prediction and weight weighting in the future, reducing the influence of training samples on model accuracy due to noise, imbalance or limited coverage, etc. The multi-model set can accurately generalize to unseen data, improve prediction accuracy and system decision-making ability.

[0067] In an embodiment, as shown in Figure 3 Step S20, that is, determining the weighting coefficients of the multiple prediction models, specifically includes the following steps:

[0068] S201: Determine the prediction data type for target prediction of the to-be-predicted data.

[0069] After receiving the prediction instruction for the to-be-detected data, the server can determine a prediction data type for target prediction of the to-be-predicted data, i.e., determine a data type of a prediction value for target prediction of the to-be-detected data. The prediction data type can be determined through the received prediction instruction. The prediction instruction indicates a prediction target (i.e., a prediction value expected to be predicted) for prediction of the to-be-predicted data, and the data type of the prediction target in the prediction instruction can be used to determine the prediction data type.

[0070] For example, the to-be-predicted data can be spectral data of a target object (such as a semiconductor device), and the prediction instruction for the to-be-detected data can indicate that a critical dimension (such as length, width, or height) of the target object or a thickness of the target object is predicted based on the spectral data of the target object, i.e., the prediction target for prediction of the to-be-predicted data is the critical dimension or the thickness, and the prediction data type is a size prediction type, i.e., a first type. Alternatively, the prediction instruction for the to-be-detected data can also indicate that a probability of the target object having a defect is predicted based on the spectral data of the target object, i.e., the prediction target for prediction of the to-be-predicted data is the defect probability, and the prediction data type is a probability prediction type, i.e., a second type. In other embodiments, the to-be-predicted data can be other data, and the prediction instruction for the to-be-detected data and the prediction target can also be other, which will not be described herein.

[0071] S202: Obtain a plurality of weight data sets obtained by pre-solution, and the weight data set includes weight values corresponding to a plurality of prediction models. The plurality of weight values in each weight data set are obtained by weight optimization solution of the plurality of prediction models based on the second sample data set.

[0072] Meanwhile, the server can read weight data for the plurality of prediction models pre-stored in the local database, and the weight data includes one or more weight data sets. Each weight data set includes weight values corresponding to a plurality of prediction models, wherein the plurality of weight values in each weight data set are weight values obtained by weight optimization solution of the plurality of prediction models based on the second sample data set. When the server local database stores a plurality of weight data sets obtained by pre-solution, each weight data set corresponds to a prediction data type. That is, the server uses different weight solution algorithms to perform weight optimization solution of the plurality of prediction models based on the second sample data set for different prediction data types, to obtain weight data sets corresponding to different prediction data types.

[0073] When the server local database stores one weight data set obtained by pre-solution, the server directly assigns the weight values of the prediction models in the weight data set to the corresponding prediction models to obtain weight coefficients of the prediction models.

[0074] S203: Determine the target weight data set corresponding to the prediction data type in the plurality of weight data sets, and determine the corresponding weight value in the target weight data set as the weight coefficient of each prediction model.

[0075] When a plurality of pre-solved weight data sets are stored in the server local database, the server determines the target weight data set corresponding to the prediction data type in the plurality of weight data sets, and determines the corresponding weight value in the target weight data set as the weight coefficient of each prediction model.

[0076] In this embodiment, the prediction data type for target prediction of the to-be-predicted data is determined, and then the target weight data set corresponding to the prediction data type is determined in the plurality of weight data sets, and the corresponding weight value in the target weight data set is determined as the weight coefficient of each prediction model. Each set of weights is an optimal solution optimized based on a specific prediction data type, so that each prediction task can use the most suitable model weight combination method to improve the adaptability and generalization ability of model integration. Moreover, by determining the type of to-be-predicted data, and then selecting the weight data set optimized in advance for the type, the model integration can be more in line with the data characteristics, thereby effectively improving the prediction accuracy. In addition, since the plurality of weight data sets are pre-solved offline, only the corresponding weight set needs to be selected according to the data type during actual prediction, without the need for real-time solving of optimization problems, which significantly improves the efficiency and real-time performance of online prediction.

[0077] In one specific embodiment, the to-be-predicted data can be spectral data of a target object (such as a semiconductor device), and the prediction instruction for the to-be-predicted data can be a prediction instruction for performing critical dimension prediction on the target object based on the spectral data of the target object. For example, as shown in FIG. 1, the data prediction method provided in this embodiment specifically includes the following steps: Figure 4

[0078] S101: Obtain a plurality of prediction models pre-trained, and the plurality of prediction models are obtained by machine learning training based on a first sample data set.

[0079] The first sample data set includes a plurality of spectral image samples. That is, the server trains the support vector regression model, the random forest model, or two models in the plurality of heterogeneous neural network models using the plurality of spectral image samples, to obtain the plurality of prediction models including at least two heterogeneous prediction models. The prediction model is used to predict the critical dimension of the object based on the spectral data (i.e., the spectral image) of the object.

[0080] ​S102: After receiving the prediction instruction of performing critical dimension prediction on the target object, weight coefficients of the plurality of prediction models are determined, and the weight coefficients of the plurality of prediction models are weight values obtained by performing weight optimization solving on the plurality of prediction models based on the second sample data set.

[0081] The server receives the prediction instruction for the to-be-predicted data, that is, receives the prediction instruction of performing critical dimension prediction on the target object based on the spectral data of the target object, and obtains the spectral data of the target object. The user can upload the spectral data of the target object through the terminal device, and send the prediction instruction to the server through the terminal device. The server receives the prediction instruction and obtains the spectral data of the target object carried by the prediction instruction.

[0082] The second sample data set includes a plurality of spectral pattern samples. After obtaining the plurality of spectral pattern samples of the same type of object collected in the history, the plurality of spectral pattern samples are divided into a first sample data set and a second sample data set, and each data set includes at least two different spectral pattern samples. After training the plurality of prediction models based on the plurality of spectral pattern samples in the first sample data set, the server can perform weight optimization solving on the plurality of prediction models based on the plurality of spectral pattern samples in the second sample data set, and obtain at least one weight data group. The weight data group includes weight values corresponding to the plurality of prediction models.

[0083] After receiving the prediction instruction for the to-be-predicted data, that is, receiving the prediction instruction of performing critical dimension prediction on the target object based on the spectral data of the target object, the server determines the weight coefficients of the plurality of prediction models. For example, when a plurality of weight data groups are stored in the server, the prediction data type of the target prediction for the to-be-predicted data is determined according to the prediction instruction, the target weight data group corresponding to the prediction data type is determined in the plurality of weight data groups, and the corresponding weight values in the target weight data group are determined as the weight coefficients of each prediction model.

[0084] S103: The plurality of prediction models are used to perform critical dimension prediction on the spectral data of the target object respectively, and model prediction values output by the plurality of prediction models respectively are obtained.

[0085] After receiving the prediction instruction for the to-be-predicted data, the server uses the plurality of prediction models to perform critical dimension prediction on the spectral data of the target object respectively, and obtains model prediction values output by the plurality of prediction models respectively. That is, the spectral data of the target object is input into the plurality of prediction models respectively, so that each prediction model performs critical dimension prediction on the spectral data of the target object, and outputs a dimension prediction value of the spectral data of the target object, thereby obtaining the model prediction values output by the plurality of prediction models.

[0086] S104: Based on the weight coefficients of each prediction model, the model prediction values respectively output by the plurality of prediction models are weighted and summed to obtain a target prediction value.

[0087] In the determination of the weight coefficients of the plurality of prediction models and the model prediction values output by each prediction model, the server weights and sums the model prediction values respectively output by the plurality of prediction models based on the weight coefficients of each prediction model to obtain a target prediction value, that is, a target prediction value of the critical dimension of the target object.

[0088] S105: Output the target prediction value as the prediction result of the critical dimension of the target object.

[0089] After obtaining the target prediction value, the server outputs the target prediction value as the prediction result of the critical dimension of the target object, so that the user can timely know the prediction result of the critical dimension of the target object, and sends a processing instruction for the target object, so that the server processes the target object in response to the processing instruction, such as adjusting the parameter model of the target object, to improve the decision-making ability of the system based on the prediction result.

[0090] In the embodiment, a plurality of heterogeneous prediction models are used for critical dimension prediction, which can complement the advantages and disadvantages of different models in prediction performance. By integrating the critical dimension prediction results of the plurality of prediction models, the prediction deviation and overfitting risk of a single model can be reduced, and the overall prediction performance and the prediction accuracy of the critical dimension of the object can be improved. On the other hand, by optimizing the weight of the model on the second sample data set, the importance of each model in the weighted combination can be dynamically adjusted according to the actual situation, the prediction process is more flexible and adaptive, the prediction result of the critical dimension better adapts to the new data distribution, and the prediction ability of the model combination based on spectral data for the critical dimension of the object can be enhanced, thereby improving the prediction accuracy, and the server can adaptively adjust the parameters of the target object based on the critical dimension of the object, thereby improving the decision-making ability of the system.

[0091] In addition, a plurality of prediction models can be trained using a small number of spectral pattern samples, and accurate prediction results can be obtained through multi-model prediction and weight weighting in the future, thereby reducing the influence of training samples on model accuracy due to noise, imbalance, or limited coverage, etc. The multi-model set can accurately generalize to subsequent spectral data, improve the prediction accuracy of the critical dimension of the object, and improve the decision-making ability of the system. At the same time, the embodiment of the present application also solves the problem of poor model accuracy and poor critical dimension prediction effect caused by insufficient samples in the optical field, and improves the accuracy of the model in predicting the critical dimension of the object based on spectral data.

[0092] In an embodiment, the server can train a plurality of initially different models using a first sample dataset to obtain a plurality of prediction models, so as to subsequently use the plurality of prediction models to perform target prediction on the to-be-predicted data. Wherein, as shown in Figure 5 the plurality of prediction models are trained by the following steps:

[0093] S1: Obtain a plurality of historical data samples, the historical data samples being the same type of data as the to-be-predicted data.

[0094] The server obtains a plurality of historical data samples. Wherein, the historical data samples are the same type of data as the to-be-predicted data. For example, the to-be-predicted data is the spectral data of an object, and the historical data samples are also the spectral data (i.e. spectral images) of the same type of object (or the same target object), such as historical spectral data obtained by collecting spectral images of semiconductor devices.

[0095] S2: Perform feature expansion processing on the plurality of historical data samples to obtain a first sample dataset comprising a plurality of training samples.

[0096] After obtaining the plurality of historical data samples, the plurality of historical data samples are divided into a dataset to obtain an initial training set, a generalization set (i.e. a second sample dataset, used for weight solving of the model) and a test set. Wherein, the number of historical data samples in the training set and the prediction set is less than the number of historical data samples in the test set, so as to simulate the small sample machine learning scenario.

[0097] For example, the number of historical data samples obtained this time is 1989, such as 1989 real historical spectral images of a target object. The server divides the first few historical data samples among the 1989 historical data samples into an initial training set, divides the last few historical data samples among the 1989 historical data samples into a generalization set, and divides the remaining historical data samples into a test set, so as to simulate a small sample machine learning scenario, use the initial training set to train the model, and use the test set to test the performance of the trained model, until a plurality of prediction models with performance convergence are obtained. Subsequently, the generalization set is used to solve the weight optimization of the plurality of prediction models to obtain at least one weight data group, so as to obtain the weight coefficients of the plurality of prediction models. Using the same type of data in the same dataset for model training and weight solving can improve the accuracy of the weight, so as to improve the accuracy of the target prediction value obtained by subsequent weighted summation.

[0098] Specifically, after obtaining the plurality of historical data samples and dividing the dataset, the server performs feature expansion processing on the plurality of historical data samples in the initial training set to obtain a plurality of training samples to form a first sample dataset comprising a plurality of training samples.

[0099] S3: performing dataset construction on the plurality of training samples in the first sample dataset to obtain at least one target sample dataset.

[0100] After obtaining the first sample dataset, the server performs dataset construction on the plurality of training samples in the first sample dataset to obtain at least one target sample dataset. The server can use a sampling algorithm in the multi-fold cross-validation method and / or a bootstrap sampling algorithm (such as the Bagging bootstrap sampling algorithm) to sample the plurality of training samples in the first sample dataset to construct a plurality of datasets, thereby obtaining a plurality of target sample datasets.

[0101] S4: training the plurality of heterogeneous initial models using the first sample dataset and the at least one target sample dataset to obtain a plurality of prediction models.

[0102] After obtaining the first sample dataset and the target sample dataset, the server trains the plurality of heterogeneous initial models using the first sample dataset and the at least one target sample dataset to obtain a plurality of prediction models. The plurality of initial models are machine learning models with different structures.

[0103] For example, one or more initial models are trained using the first sample dataset, and the trained initial models are tested using a test set until the performance of the initial models meets the requirements, thereby obtaining one or more converged prediction models. Meanwhile, one or more initial models are trained using the target sample dataset, and the trained initial models are tested using a test set until the performance of the initial models meets the requirements, thereby obtaining one or more converged prediction models. The prediction models obtained by training using the two datasets are output, thereby obtaining a plurality of prediction models.

[0104] In this embodiment, by performing feature expansion processing on the historical data samples, potential data features can be mined, a higher-dimensional input space can be constructed, the learning ability of the model for complex patterns can be enhanced, and the first sample dataset after feature expansion is used to construct the target sample dataset, so that the training data is more comprehensive to cover the data distribution under different situations or states, the robustness and generalization ability of the model are enhanced, and the prediction performance of the model is improved. In this scheme, by performing systematic feature expansion and sample construction on the historical data samples, a plurality of structure-heterogeneous prediction models are trained, thereby significantly improving the learning ability of the model for complex data features and the accuracy, robustness, and scalability of the overall prediction system.

[0105] In an embodiment, step S2, i.e., training the plurality of initial models using the first sample dataset and the at least one target sample dataset to obtain a plurality of prediction models, specifically includes the following steps:

[0106] S21: feature extraction is performed on the plurality of historical data samples to obtain feature vectors of the plurality of historical data samples, and missing value filling is performed on a vector matrix formed by the plurality of feature vectors to obtain a first vector matrix.

[0107] After obtaining the plurality of historical data samples and dividing the data set to obtain the initial training set, the server performs feature extraction on the plurality of historical data samples in the initial training set respectively to obtain feature vectors of the historical data samples. Then, the server converts the feature vectors of the historical data samples into a vector matrix to obtain a vector matrix formed by the feature vectors of the plurality of historical data samples. Each row (i.e., a feature row) in the vector matrix represents a feature vector of a historical data sample, and each feature vector includes a plurality of feature values. The feature values at the same positions in the plurality of feature vectors constitute a feature column. That is, each column (i.e., a feature column) in the vector matrix includes the feature values at the same positions in the plurality of feature vectors, i.e., a feature vector of a historical data sample is composed of a plurality of feature values. The number of rows in the vector matrix is the number of historical data samples, and the number of columns in the vector matrix is the number of feature values in the feature vector.

[0108] For example, the historical data samples can be spectral image samples. The server performs feature extraction on the wavelength points of the plurality of spectral image samples in the initial training set respectively to obtain feature vectors of the spectral image samples, wherein the feature vector of a spectral image sample includes feature values of a plurality of wavelength points in the spectral image sample. The server converts the feature vectors of the plurality of spectral image samples into a vector matrix, wherein each row in the vector matrix represents a feature vector of a spectral image sample, and each column (i.e., a feature column) in the vector matrix includes the values of the wavelength points at the same positions in the plurality of feature vectors; that is, the number of the plurality of spectral image samples in the initial training set is N, the number of wavelength points in a spectral image sample is d, the number of feature values of the wavelength points in the feature vector of a spectral image sample is d, and the feature vectors of the plurality of spectral image samples form an N*d vector matrix.

[0109] After obtaining the vector matrix formed by the feature vectors of the plurality of historical data samples, the server performs missing value filling on the vector matrix to obtain a first vector matrix. The first vector matrix includes the feature vectors of the plurality of historical data samples after missing value filling. Due to data acquisition errors, the collected historical data samples may have missing data of one or more data points, resulting in missing of some feature values in the extracted feature vectors. Therefore, missing value filling needs to be performed on the vector matrix formed by the plurality of feature vectors to obtain the first vector matrix, so as to perform subsequent data calculation and obtain accurate training samples.

[0110] For example, the historical data sample can be a spectrum sample, and there can be some wavelength points in the spectrum sample that do not have actual values, that is, the measurement values of some wavelength points in the spectrum sample can be 0 or missing; the server extracts the feature vector of each spectrum sample and converts it into a vector matrix, and can perform a missing value filling operation on the feature values in the vector matrix, that is, in each row and each column of the vector matrix, the positions of the feature values that are 0 or missing values are filled with a preset feature value, which can be a calibrated fixed value, or the mean or median of the feature values in the column where the position is located, thereby obtaining a first vector matrix after filling the missing values.

[0111] By filling the missing values in the feature vector matrix, the common missing problem in actual data collection is solved, ensuring that all training sample dimensions are complete, and effectively preventing training errors or model instability caused by missing values.

[0112] S22: performing feature augmentation processing on the first vector matrix to obtain a second vector matrix, and performing normalization processing on the second vector matrix to obtain a normalized vector matrix.

[0113] After obtaining the first vector matrix, the server performs feature augmentation processing on the first vector matrix to obtain a second vector matrix. Through feature augmentation processing (such as constructing combined features, time window features, etc.), more potential association patterns can be discovered, the expression ability of the data dimension is improved, and it is helpful for the model to learn more complex rules.

[0114] Among them, for the feature vector of each historical data sample in the first vector matrix, the server can calculate the target statistical vector (which can be at least one of the skewness, kurtosis, mean and standard deviation of the feature vector) of the feature vector of the historical data sample, that is, calculate the target statistical vector (such as mean and standard deviation) of all feature values (such as feature values of all wavelength points) in each feature vector; then, the server splices the mean and standard deviation of the feature vector of the historical data sample as feature values to any position (such as the end or a random position of the original feature vector) of the original feature vector, that is, the server calculates the mean and standard deviation of each row of data in the first vector matrix, and then inserts the mean and standard deviation of the row of data into any position of the row of data to form a new row of data (i.e., form a new feature vector of the historical data sample), thereby obtaining the second vector matrix after feature augmentation. The second vector matrix includes the feature vectors of a plurality of historical data samples, and each feature vector of a historical data sample includes the original feature values (and the preset feature values after filling the missing values) extracted from the spectrum sample and the calculated target statistical vector.

[0115] After obtaining the second vector matrix, the server performs normalization processing on the second vector matrix to obtain a normalized vector matrix. Specifically, after obtaining the second vector matrix, the server calculates the mean and standard deviation of each column of data in the second vector matrix; then, for each column in the second vector matrix, the server subtracts each feature value in the column from the mean of the column and divides by the standard deviation of the column to obtain the normalized column data, thereby obtaining the normalized vector matrix after each column of data is normalized.

[0116] S23: Perform dimensionality reduction and data augmentation processing on the normalized vector matrix to obtain a target vector matrix, output each feature vector in the target vector matrix as a training sample, and obtain a first sample dataset.

[0117] After obtaining the normalized vector matrix, the server performs dimensionality reduction and data augmentation processing on the normalized vector matrix to obtain a target vector matrix, and outputs each feature vector in the target vector matrix as a training sample to obtain a first sample dataset. By removing redundant and weakly correlated features through dimensionality reduction, the representation compactness of the data is improved, the risk of model overfitting is reduced, and the training speed is accelerated. Through data augmentation processing (such as noise addition, disturbance, time translation, etc.), more diverse training samples can be generated.

[0118] Wherein, after obtaining the normalized vector matrix, the server can use a preset dimensionality reduction algorithm, which can be a Recursive Feature Elimination (RFE) algorithm, a Principal Component Analysis (PCA) algorithm, etc., to perform dimensionality reduction processing on the normalized vector matrix to obtain a dimensionally reduced normalized vector matrix. Then, the server uses a preset data augmentation algorithm (such as an image data augmentation algorithm, a text data augmentation algorithm) to perform data augmentation processing on each row of data in the dimensionally reduced normalized vector matrix, and outputs each row of augmented data (i.e., the feature vector of each historical data sample) as a training sample to obtain a first sample dataset including multiple training samples.

[0119] In this embodiment, by performing feature extraction, missing data filling, feature expansion, normalization, dimensionality reduction, and data augmentation on historical data samples, multiple training samples are obtained, the diversity and accuracy of the training samples are improved, more accurate and robust input data is provided for subsequent model training, and the problem of underfitting in small sample machine learning due to insufficient number of samples is solved. In the scenario where the number of samples is insufficient, the generalization ability and stability of the model are significantly improved.

[0120] In an embodiment, in step S4, the model training is performed on the plurality of initial models using the first sample dataset and the at least one target sample dataset to obtain a plurality of prediction models, and the model training specifically includes the following steps:

[0121] S41: The plurality of initial models are trained using the first sample dataset to obtain a plurality of first prediction models that have converged in training.

[0122] The server obtains a plurality of initial models with different structures, trains the plurality of initial models using the first sample dataset to obtain a plurality of initial models with updated parameters, and then performs performance testing on the initial models with updated parameters using a test set. When the performance of the initial models with updated parameters meets the requirements, the initial model with updated parameters is output as a first prediction model that has converged in training to obtain a plurality of first prediction models with different structures.

[0123] S42: The plurality of initial models are trained using the target sample dataset to obtain a plurality of second prediction models that have converged in training.

[0124] Meanwhile, after obtaining the at least one target sample dataset, the server trains the plurality of initial models using one target sample dataset to obtain a plurality of initial models with updated parameters, and then performs performance testing on the initial models with updated parameters using a test set. When the performance of the initial models with updated parameters meets the requirements, the initial model with updated parameters is output as a second prediction model that has converged in training to obtain a plurality of second prediction models with different structures. When the number of target sample datasets is K (K is an integer greater than 1), the same initial model is trained for each target sample dataset to obtain K second prediction models with the same structure.

[0125] S43: The plurality of first prediction models and the plurality of second prediction models are output to obtain a plurality of prediction models.

[0126] After the plurality of first prediction models and the plurality of second prediction models are obtained, the server outputs the plurality of first prediction models and the plurality of second prediction models together as a plurality of prediction models, and the plurality of prediction models include a plurality of prediction models with different structures and a plurality of prediction models with the same structure.

[0127] For example, the server obtains N initial models with different structures, the N initial models including a support vector regression model and a random forest model, and at least two models of a plurality of neural network models with different network structures; a plurality of training samples in a first sample data set are subjected to data set construction to obtain K target sample data sets. The server trains the N initial models respectively by using the first sample data set to obtain N first prediction models that are trained to converge; after the plurality of training samples in the first sample data set are subjected to data set construction to obtain the K target sample data sets, the server trains the N initial models respectively by using the target sample data sets to obtain N*K second prediction models that are trained to converge; then, the N first prediction models that are trained to converge and the N*K second prediction models that are trained to converge are collectively output as a prediction model to obtain N*(1+K) prediction models, so as to subsequently use the N*(1+K) prediction models to perform target prediction on to-be-predicted data.

[0128] According to the experimental results, it can be seen that the performance of the data prediction method provided in the embodiment is obviously better than that of a single prediction model trained in a traditional manner. As shown in Figure 6 and Figure 7 , the prediction performance curves of the data prediction by using the plurality of prediction models provided in the embodiment and the traditional prediction model when the training sample quantity is 20, 30, 40, 50, and 60 are shown. Among them, Figure 6 , curve 1, curve 2, and curve 3 respectively represent the mean square error curves of the prediction results of the prediction by using the random forest model, the extreme gradient boosting model (XGBoost model) trained in a traditional manner, and the prediction results of the prediction by using the plurality of prediction models trained in the embodiment and the weighted prediction values. Figure 6 , the abscissa in Figure 6 , the ordinate in Figure 7 , curve 1, curve 2, and curve 3 respectively represent the determination coefficient curves of the random forest model, the extreme gradient boosting model trained in a traditional manner, and the determination coefficient curves of the prediction results of the prediction by using the plurality of prediction models trained in the embodiment and the weighted prediction values. Figure 6 , the abscissa in Figure 6 , the ordinate in

[0129] It should be understood that the smaller the mean square error of the model on a plurality of training samples, the better the performance of the model; the larger the determination coefficient of the model on a plurality of training samples, the better the performance of the model. From Figure 6 and Figure 7It can be seen that the performance of the prediction model trained in the embodiment is obviously better than the performance of the model trained by the traditional scheme alone. Moreover, as the number of training samples increases, the prediction model trained in the embodiment exhibits better performance.

[0130] In the embodiment, the plurality of initial models are trained using the first sample data set to obtain a plurality of first prediction models trained to converge, and the plurality of initial models are trained using the target sample data set to obtain a plurality of second prediction models trained to converge. Finally, the plurality of first prediction models and the plurality of second prediction models are output as a plurality of prediction models. By training a plurality of initial models with different structures on the first sample data set and the target sample data set, respectively, training can enable the model to learn different levels or dimensions of rules, which can reduce the training bias or overfitting that may exist when training all models based on the same training data. The plurality of prediction models generated have differences in data perspective and model structure, thereby enhancing the performance coverage of different prediction models, and further enhancing the generalization ability and robustness of the plurality of prediction models under different data distributions, and improving the prediction accuracy.

[0131] In an embodiment, the server can use the second sample data set to perform weight optimization solving on the plurality of prediction models trained to obtain at least one weight data set including weight values of the plurality of prediction models, so that after using the plurality of prediction models to perform target prediction on the to-be-predicted data, the current weight coefficients of the plurality of prediction models are determined based on the weight data set solved in advance, and then the model prediction values output by each prediction model are weighted and summed based on the corresponding weight coefficients to obtain a target prediction value as a prediction result.

[0132] As shown in Figure 8 After step S4, i.e., after the plurality of prediction models are trained, the weight values corresponding to the plurality of prediction models in the weight data set are solved by the following steps:

[0133] S5: Obtain a second sample data set, the second sample data set including a plurality of historical data samples.

[0134] After the server trains the plurality of prediction models, the second sample data set is obtained, and the second sample data set includes a plurality of historical data samples. The second sample data set is obtained by dividing the plurality of historical data samples into data sets.

[0135] S6: For each historical data sample, use the plurality of prediction models to perform target prediction on the historical data sample, and determine a prediction vector corresponding to the historical data sample based on the sample prediction values output by the plurality of prediction models.

[0136] After obtaining the second sample data set including a plurality of historical data samples, the server respectively performs target prediction on each historical data sample in the second sample data set by using the plurality of prediction models, to obtain a plurality of sample prediction values respectively output by the plurality of prediction models for the historical data sample.

[0137] That is, the historical data sample is respectively input into the plurality of prediction models for target prediction, to obtain sample prediction values respectively output by the plurality of prediction models; and then the server determines a prediction vector corresponding to the historical data sample based on the sample prediction values respectively output by the plurality of prediction models for the historical data sample, to obtain prediction vectors corresponding to the plurality of historical data samples. For example, the server can splice the sample prediction values respectively output by the plurality of prediction models for the historical data sample into a vector, to obtain the prediction vector corresponding to the historical data sample, which is composed of the sample prediction values of the historical data sample by the plurality of prediction models.

[0138] S7: Based on the prediction vectors corresponding to the plurality of historical data samples, the weight values of the plurality of prediction models are optimized and solved, to obtain the weight values corresponding to each prediction model, to obtain a weight data set.

[0139] After obtaining the prediction vectors corresponding to the historical data samples, the server optimizes (such as maximizes or minimizes) the preset index based on the prediction vectors corresponding to the plurality of historical data samples, and uses the sum of the weight values as a constraint of 1, to use the mathematical programming method to optimize and solve the weight values of the plurality of prediction models, to obtain the weight values corresponding to each prediction model, to obtain a weight data set. Each historical data sample corresponds to a real measurement value.

[0140] The server can convert the prediction vectors corresponding to the plurality of historical data samples into a prediction vector matrix, each row of the prediction vector matrix being a prediction vector corresponding to a historical data sample, the prediction vector including sample prediction values of the plurality of prediction models for the historical data sample. The server can also set an initial weight value for each prediction model and splice the weight values of the plurality of prediction models into a weight matrix, each row of the weight matrix being a weight value. Then, the server multiplies the prediction vector matrix and the weight matrix to obtain a target matrix, each row of the target matrix being a target prediction value of a historical data sample, that is, the target matrix includes target prediction values of a plurality of historical data samples. The target prediction value is obtained by weighting and summing the sample prediction values of each prediction model in the prediction vector of the historical data sample with the plurality of weight values in the weight matrix. After obtaining the target matrix, that is, the target prediction values of the plurality of historical data samples, the server determines a preset index according to the target prediction values of the plurality of historical data samples and the real measurement values, the preset index being represented by an optimization objective function composed of the target prediction values of the plurality of historical data samples and the real measurement values. The server iteratively updates the weight value of each prediction model to optimize the weight values of the plurality of prediction models, so as to obtain the weight values of the plurality of prediction models and form a weight data set.

[0141] The prediction vector matrix obtained by converting the prediction vectors corresponding to the plurality of historical data samples can be represented by the following matrix:

[0142]

[0143]

[0144] The weight matrix obtained by splicing the weight values of the plurality of prediction models can be represented by the following matrix:

[0145] ​​​​​​​​​​​​​​​​;

[0146] in, , … These represent the weight values ​​of the first prediction model, the second prediction model, and so on. The weight values ​​of each prediction model, This represents the number of prediction models.

[0147] The target matrix can then be represented by the following formula:

[0148] ;

[0149] in, , … Let represent the target predicted value of the first historical data sample, the target predicted value of the second historical data sample, and so on, in the target matrix. The target predicted value for a sample of historical data; , … Let these represent the prediction vectors for the first historical data sample, the second historical data sample, and so on. A prediction vector for a historical data sample; , … These represent the weight values ​​of the first prediction model, the second prediction model, and so on. The weight values ​​of each prediction model, The number of prediction models; This represents the number of historical data samples in the second sample dataset.

[0150] For example, the preset index can be the coefficient of determination. After determining the objective function of the coefficient of determination based on the target predicted values ​​of multiple historical data samples in the target matrix, the server aims to maximize the coefficient of determination and uses mathematical programming to optimize the weight values ​​of multiple prediction models with the constraint that the sum of the weight values ​​is 1, so as to obtain the weight values ​​corresponding to each prediction model.

[0151] The objective function for the coefficient of determination can be expressed by the following formula:

[0152] ;

[0153] in, Indicates the coefficient of determination; Indicates the first The target predicted value for a sample of historical data; Indicates the first The actual measurement value of a historical data sample. , …; This represents the number of historical data samples in the second sample dataset. This represents the average of the true measurements of all historical data samples in the second sample dataset.

[0154] In this embodiment, by obtaining the prediction output of each prediction model on the second sample dataset and constructing prediction vectors, the performance differences of each prediction model on the actual data can be comprehensively evaluated. This allows for the optimal allocation of weights for each prediction model based on the actual results, improving the accuracy of the weights and significantly enhancing the prediction accuracy after weight fusion. Furthermore, the weight optimization process considers the overall prediction error of multiple samples, helping to suppress the overfitting effect of a particular model on specific samples. This improves the stability of the overall model combination across different input data and enhances the accuracy of the model weights.

[0155] In one embodiment, one weighted data set corresponds to one type of sample prediction value; that is, one weighted data set corresponds to one prediction data type. The prediction targets for the server's target prediction of historical data samples can include multiple categories. Prediction targets can be size prediction targets, such as predicting the key dimensions and thickness of the target object; prediction targets can also be probability prediction targets, such as predicting the probability of a defect in the target object. The server can predict different prediction targets for historical data samples according to requirements to obtain sample prediction values ​​for different types of prediction targets, thereby obtaining prediction vectors for different types of prediction targets corresponding to historical data samples. This allows for subsequent optimization of the weight values ​​of multiple prediction models based on the prediction vectors of different types of prediction targets corresponding to multiple historical data samples. The weighted data sets corresponding to different types of prediction targets are thus obtained as multiple weighted data sets corresponding to different prediction data types. One type of prediction target (i.e., one prediction data type) corresponds to one weighted data set. The process of obtaining each weighted data set is described above and will not be repeated here.

[0156] In one embodiment, such as Figure 9 As shown, step S7, which involves optimizing the weight values ​​of multiple prediction models based on the prediction vectors corresponding to multiple historical data samples, to obtain the weight values ​​corresponding to each prediction model, specifically includes the following steps:

[0157] S71: Determine the optimization index based on the data type of the predicted sample values ​​output by the prediction model and the prediction vectors corresponding to multiple historical data samples.

[0158] The server determines a prediction data type to which the sample prediction value output by the prediction model belongs. If the sample prediction value output by the prediction model is a value of a size type, for example, the sample prediction value is a critical dimension, a thickness, or the like, it is determined that the prediction data type to which the sample prediction value output by the prediction model belongs is a size prediction type, that is, a first type. If the sample prediction value output by the prediction model is a value of a probability type, for example, the sample prediction value is a probability that an object is a defect, or the like, it is determined that the prediction data type to which the sample prediction value output by the prediction model belongs is a probability prediction type, that is, a second type. In addition, if the sample prediction value output by the prediction model is a value other than the value of the size type or the value of the probability type, it is determined that the prediction data type to which the sample prediction value output by the prediction model belongs is an other prediction type, that is, a third type.

[0159] After determining the prediction data type to which the sample prediction value output by the prediction model belongs, the server determines an optimization index according to the prediction data type to which the sample prediction value output by the prediction model belongs and the prediction vectors corresponding to the plurality of historical data samples. Different optimization indexes correspond to different prediction data types. The optimization index can be a coefficient, a mean square error, a cross-entropy loss, or the like.

[0160] S72: Based on the prediction vectors corresponding to the plurality of historical data samples, the weight values of the plurality of prediction models are optimized to obtain the weight values corresponding to the prediction models, with the optimization index being maximized or minimized as the target.

[0161] After determining the optimization index, the server optimizes the weight values of the plurality of prediction models based on the prediction vectors corresponding to the plurality of historical data samples, with the optimization index being maximized or minimized as the target, to obtain the weight values corresponding to the prediction models. The determination process of the weight values corresponding to the prediction models can be referred to in the foregoing, and will not be described here again.

[0162] In this embodiment, the weight values are solved with the optimization index of a certain type being maximized or minimized as the target, to ensure that the overall performance of the fusion result on the historical samples is optimal, and to avoid the local suboptimal or non-optimal problem caused by the artificial setting of the weight values. Moreover, because the evaluation standards suitable for different types of prediction tasks are different, if a certain index is used for optimization, it may not be suitable for some tasks. In this embodiment, by selecting a suitable optimization index according to the prediction data type to which the sample prediction value belongs, the evaluation standards of different task scenarios can be dynamically adapted, and the prediction effect of the fusion model under the target task can be improved.

[0163] In an embodiment, in step S71, the optimization index is determined according to the prediction data type to which the sample prediction value output by the prediction model belongs and the prediction vectors corresponding to the plurality of historical data samples, and specifically includes the following steps:

[0164] S711: When the prediction data type to which the sample prediction value output by the prediction model belongs is the first type, the optimization index is determined as the determination coefficient of the target prediction value of the plurality of historical data samples.

[0165] After determining the prediction data type to which the sample prediction value output by the prediction model belongs, when the prediction data type to which the sample prediction value output by the prediction model belongs is the first type (such as the size prediction type), the server determines the optimization index as the determination coefficient of the target prediction value of the plurality of historical data samples.

[0166] The target prediction value of each historical data sample is determined according to the prediction vector of the corresponding historical data sample and the weight value of the model. The determination process of the target prediction value of each historical data sample is described above and will not be repeated here.

[0167] S712: When the prediction data type to which the sample prediction value output by the prediction model belongs is the second type, the optimization index is determined as the mean square error of the target prediction value of the plurality of historical data samples.

[0168] After determining the prediction data type to which the sample prediction value output by the prediction model belongs, when the prediction data type to which the sample prediction value output by the prediction model belongs is the second type (such as the probability prediction type), the server determines the optimization index as the mean square error of the target prediction value of the plurality of historical data samples. In other embodiments, when the prediction data type to which the sample prediction value output by the prediction model belongs is the third type (a type other than the probability prediction type and the size prediction type), the server can determine the cross-entropy loss of the target prediction value of the plurality of historical data samples as the optimization index.

[0169] In this embodiment, the corresponding optimization index is selected according to the prediction data type to which the sample prediction value belongs. The determination coefficient is used for the first type of prediction data, which can measure the explanation ability of the prediction model to data variation. The mean square error is used for the second type of prediction data, which can measure the overall deviation between the prediction value and the true value. The targeted optimization of the multi-model fusion weight is realized, thereby significantly improving the accuracy of the model weight and the adaptability to different types of prediction targets.

[0170] In an embodiment, in step S72, the weight values of the plurality of prediction models are optimized based on the prediction vectors corresponding to the plurality of historical data samples, so as to obtain the weight values corresponding to the plurality of prediction models, with the optimization index being maximized or minimized as the target. Specifically, the following steps are included:

[0171] S721: When the optimization index is the determination coefficient of the target prediction value of the plurality of historical data samples, the weight values of the plurality of prediction models are optimized based on the prediction vectors corresponding to the plurality of historical data samples, so as to obtain the weight values corresponding to the plurality of prediction models, with the maximization of the determination coefficient being taken as the target.

[0172] When the prediction data type to which the sample prediction value output by the prediction model belongs is the first type (e.g., the size prediction type), that is, when the optimization index is the determination coefficient of the target prediction value of the plurality of historical data samples, the server optimizes and solves the weight values of the plurality of prediction models based on the prediction vectors corresponding to the plurality of historical data samples, with the goal of maximizing the determination coefficient and the constraint that the sum of the weight values is equal to 1, to obtain the weight values corresponding to each prediction model, thereby forming a weight data set. The weight data set is the weight data set corresponding to the first type (e.g., the probability prediction type). The solving process of the weight values of each prediction model is described above and will not be repeated here.

[0173] S722: When the optimization index is the mean square error of the target prediction value of the plurality of historical data samples, the server optimizes and solves the weight values of the plurality of prediction models based on the prediction vectors corresponding to the plurality of historical data samples, with the goal of minimizing the mean square error, to obtain the weight values corresponding to each prediction model.

[0174] When the prediction data type to which the sample prediction value output by the prediction model belongs is the second type (e.g., the probability prediction type), that is, when the optimization index is the mean square error of the target prediction value of the plurality of historical data samples, the server can determine the target matrix according to the above, and the target matrix includes the target prediction value of the plurality of historical data samples. After obtaining the target matrix, that is, obtaining the target prediction value of the plurality of historical data samples, the server calculates the mean square error between the target prediction value of the plurality of historical data samples and the actual measurement value, and takes the mean square error as the optimization index. The server optimizes and solves the weight values of the plurality of prediction models based on the mean square error, with the goal of minimizing the mean square error and the constraint that the sum of the weight values is equal to 1, to obtain the weight values of the plurality of prediction models, thereby forming a weight data set. The weight data set is the weight data set corresponding to the second type (e.g., the probability prediction type).

[0175] In this embodiment, when the prediction data type to which the sample prediction value output by the prediction model belongs is the first type (such as the size prediction type), the determination coefficient is used as the optimization index, the target is to maximize the explanation ability of the prediction model to the data variation, to improve the adaptability of the model weight to the size prediction target (i.e. the prediction target of the first type), so as to improve the prediction ability of the prediction model for the size prediction target. When the prediction data type to which the sample prediction value output by the prediction model belongs is the second type (such as the probability prediction type), the mean square error is used to measure the error fluctuation, the target is to minimize the prediction error of the prediction model, to improve the adaptability of the model weight to the probability prediction target (i.e. the prediction target of the first type), so as to improve the prediction ability of the prediction model for the probability prediction target. The present scheme selects the optimization target as needed, and then determines the weight data set corresponding to different prediction data types, which can improve the adaptability of the system to different types of prediction tasks, improve the scientificity and accuracy of the model weight, and can select different weight data sets for prediction result optimization according to actual prediction requirements in the future, improve the accuracy of the prediction result, and make the prediction result more meet the needs of the target scene.

[0176] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0177] In an embodiment, a data prediction device is provided, which corresponds to the data prediction method in the above embodiment. As shown in the figure, the data prediction device comprises an acquisition module 101, a determination module 102, a prediction module 103, a calculation module 104 and an output module 105. The functions of each module are described in detail as follows: Figure 10 The acquisition module 101 is configured to acquire a plurality of prediction models trained in advance, wherein the plurality of prediction models are obtained by machine learning training based on a first sample data set;

[0178] The determination module 102 is configured to determine the weight coefficients of the plurality of prediction models after receiving a prediction instruction for the to-be-predicted data, wherein the weight coefficients of the plurality of prediction models are weight values obtained by weight optimization solving of the plurality of prediction models based on a second sample data set;

[0179] The prediction module 103 is configured to perform target prediction on the to-be-predicted data by using the plurality of prediction models respectively, and obtain model prediction values output by the plurality of prediction models respectively;

[0180] The calculation module 104 is configured to perform weighted summation calculation on the model prediction values output by the plurality of prediction models respectively based on the weight coefficients of each prediction model, and obtain a target prediction value;

[0181] The output module 105 is configured to output the target prediction value.

[0182] The output module 105 is configured to output the target prediction value as a prediction result corresponding to the to-be-predicted data, so as to process the to-be-predicted data based on the target prediction value.

[0183] In an embodiment, the determination module 102 is specifically configured to: determine a prediction data type for which the target prediction is performed on the to-be-predicted data; obtain a plurality of weight data groups that are obtained by pre-solution, wherein each weight data group includes weight values corresponding to a plurality of prediction models, and the weight values in each weight data group are obtained by weight optimization solution of the plurality of prediction models based on a second sample data set; determine a target weight data group corresponding to the prediction data type from the plurality of weight data groups, and determine the weight values corresponding to the target weight data group as weight coefficients of the prediction models.

[0184] In an embodiment, the determination module 102 is further configured to: obtain the second sample data set, wherein the second sample data set includes a plurality of historical data samples; for each historical data sample, perform target prediction on the historical data sample by using the plurality of prediction models respectively, and determine a prediction vector corresponding to the historical data sample based on sample prediction values output by the plurality of prediction models respectively; and perform optimization solution on the weight values of the plurality of prediction models based on the prediction vectors corresponding to the plurality of historical data samples, to obtain the weight values corresponding to the prediction models, so as to obtain a weight data group, and one weight data group corresponds to one type of sample prediction value.

[0185] In an embodiment, the determination module 102 is further configured to: determine an optimization index according to a prediction data type to which the sample prediction value output by the prediction model belongs, and the prediction vectors corresponding to the plurality of historical data samples; and perform optimization solution on the weight values of the plurality of prediction models based on the prediction vectors corresponding to the plurality of historical data samples, to obtain the weight values corresponding to the prediction models, so as to maximize or minimize the optimization index.

[0186] In an embodiment, the data prediction apparatus further includes a training module, which is configured to: obtain a plurality of historical data samples, wherein the historical data samples are of the same type of data as the to-be-predicted data; perform feature expansion processing on the plurality of historical data samples to obtain a first sample data set including a plurality of training samples; perform data set construction on the plurality of training samples in the first sample data set to obtain at least one target sample data set; and perform model training on a plurality of initial models by using the first sample data set and the at least one target sample data set, to obtain the plurality of prediction models, wherein the plurality of initial models are machine learning models that are different in structure.

[0187] In an embodiment, the training module is specifically configured to: perform feature extraction on the plurality of historical data samples to obtain feature vectors of the plurality of historical data samples, perform missing value filling on a vector matrix formed by the plurality of feature vectors to obtain a first vector matrix, perform feature expansion processing on the first vector matrix to obtain a second vector matrix, and perform normalization processing on the second vector matrix to obtain a normalized vector matrix; perform dimension reduction and data enhancement processing on the normalized vector matrix to obtain a target vector matrix, output each feature vector in the target vector matrix as a training sample to obtain a first sample dataset.

[0188] In an embodiment, the training module is specifically further configured to: train the plurality of initial models respectively using the first sample dataset to obtain a plurality of first prediction models trained to converge, train the plurality of initial models respectively using the target sample dataset to obtain a plurality of second prediction models trained to converge, and output the plurality of first prediction models and the plurality of second prediction models to obtain the plurality of prediction models.

[0189] It should be noted that the information interaction, execution process and the like between the above apparatus / units are based on the same concept as the method embodiments of the present application, and specific functions and brought technical effects can be referred to the method embodiments part, which will not be described herein.

[0190] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional units and modules is exemplified, and in actual application, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the apparatus is divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of software functional unit. In addition, the specific names of the functional units and modules are for the purpose of mutual distinction, and do not limit the protection scope of the present application. The specific working process of the units and modules in the system can refer to the corresponding process in the foregoing method embodiments, which will not be described herein.

[0191] The embodiments of the present application also provide a computer device, which can be a terminal device or a server, as shown in Figure 11 The computer device includes at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, and the processor implements the steps in any of the method embodiments described above when executing the computer program, or the processor implements the functions of the modules / units in the apparatus embodiments described above when executing the computer program.

[0192] For example, the computer program can be divided into one or more modules / units, one or more modules / units are stored in the memory and executed by the processor to complete the present application. One or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the computer device. Those skilled in the art can understand, Figure 11 Only an example of the computer device does not constitute a limitation to the computer device, which can include more or fewer components than those shown, or combine some components, or different components, such as the computer device can also include input / output devices, network access devices, buses, etc.

[0193] The above processor can be a central processing unit, and can also be other general-purpose processors, digital signal processors, application-specific integrated circuits, field programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The memory can be an internal storage unit of the computer device, such as a hard disk or memory of the computer device. The memory can also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card, a secure digital card, a flash memory card, etc. The memory can also include both the internal storage unit and the external storage device of the computer device.

[0194] The embodiments of the present application also provide a readable storage medium, the readable storage medium stores a computer program, and the computer program is executed by the processor to realize the steps in the above various method embodiments.

[0195] The embodiments of the present application provide a computer program product, when the computer program product runs on the electronic device, so that the electronic device executes to realize the steps in the above various method embodiments.

[0196] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the computer program can be stored in a computer-readable storage medium. When the processor executes the computer program, the steps of the above-mentioned various method embodiments can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms. The computer-readable medium can at least include any entity or device capable of carrying the computer program code to the photographing device / terminal equipment, recording medium, computer memory, read-only memory, random access memory, electrical carrier signal, telecommunication signal and software distribution medium. For example, U disk, mobile hard disk, magnetic disk or optical disk and the like. In some jurisdictions, according to legislation and patent practice, the computer-readable medium can not be an electrical carrier signal and a telecommunication signal.

[0197] In the above embodiments, the description of each embodiment has its own focus, and the parts not described or recorded in detail in a certain embodiment can be referred to the related description of other embodiments. Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in the present text can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered beyond the scope of the present application.

[0198] In the embodiments provided in the present application, it should be understood that the disclosed devices / apparatuses and methods can be implemented in other ways. For example, the device / apparatus embodiments described above are merely schematic, for example, the division of modules or units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed ones can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms. The units described as separate components can be or can not be physically separated, and the components displayed as units can be or can not be physical units, i.e. they can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0199] The above examples are only used to illustrate the technical solutions of the present application, but not limit the same; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can be modified, or some technical features can be replaced by equivalent ones; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A data prediction method characterized by, The method comprises the following steps: obtaining a plurality of prediction models pre-trained, the plurality of prediction models are obtained based on machine learning training of a first sample data set, and structures of at least two prediction models in the plurality of prediction models are different; after receiving a prediction instruction for to-be-predicted data, determining weight coefficients of the plurality of prediction models, the weight coefficients of the plurality of prediction models being weight values obtained based on weight optimization solving of the plurality of prediction models on a second sample data set; using the plurality of prediction models to respectively perform target prediction on the to-be-predicted data, to obtain model prediction values respectively output by the plurality of prediction models; based on the weight coefficients of the respective prediction models, performing weighted summation calculation on the model prediction values respectively output by the plurality of prediction models, to obtain a target prediction value; outputting the target prediction value as a prediction result corresponding to the to-be-predicted data; wherein the to-be-predicted data is spectral data of a target object; a prediction target for predicting the to-be-predicted data is a critical dimension or a thickness; or a prediction target for predicting the to-be-predicted data is a defect probability; the determining of the weight coefficients of the plurality of prediction models comprises: determining a prediction data type for target prediction of the to-be-predicted data, the prediction data type being a size prediction type or a probability prediction type; obtaining a plurality of weight data groups pre-solved, the weight data groups comprising weight values corresponding to the plurality of prediction models, and the plurality of weight values in each weight data group being obtained based on weight optimization solving of the plurality of prediction models on the second sample data set; determining a target weight data group corresponding to the prediction data type in the plurality of weight data groups, and determining a corresponding weight value in the target weight data group as the weight coefficient of each prediction model.

2. The data prediction method of claim 1, wherein, The plurality of prediction models are obtained through the following method: obtaining a plurality of historical data samples, the historical data samples being the same type of data as the to-be-predicted data; performing feature expansion processing on the plurality of historical data samples, to obtain the first sample data set comprising a plurality of training samples; performing data set construction on the plurality of training samples in the first sample data set, to obtain at least one target sample data set; using the first sample data set and at least one target sample data set to perform model training on a plurality of initial models, to obtain the plurality of prediction models, and the plurality of initial models being machine learning models with different structures.

3. The data prediction method of claim 2, wherein, The using of the first sample data set and at least one target sample data set to perform model training on a plurality of initial models, to obtain the plurality of prediction models, comprises: using the first sample data set to respectively train the plurality of initial models, to obtain a plurality of first prediction models that have converged in training; using the target sample data set to respectively train the plurality of initial models, to obtain a plurality of second prediction models that have converged in training; outputting the plurality of first prediction models and the second prediction models, to obtain the plurality of prediction models.

4. The data prediction method of claim 2, wherein, The feature expansion processing is performed on the plurality of historical data samples to obtain the first sample data set including a plurality of training samples, which comprises: The feature extraction is performed on the plurality of historical data samples to obtain the feature vectors of the plurality of historical data samples, and the vector matrix formed by the plurality of feature vectors is filled with missing values to obtain a first vector matrix; The feature expansion processing is performed on the first vector matrix to obtain a second vector matrix, and the second vector matrix is normalized to obtain a normalized vector matrix; The dimension reduction and data enhancement processing are performed on the normalized vector matrix to obtain a target vector matrix, each feature vector in the target vector matrix is output as a training sample, and the first sample data set is obtained.

5. The data prediction method of claim 1, wherein, The weight values corresponding to the plurality of prediction models in the weight data set are determined by the following method: Obtain the second sample data set, which includes a plurality of historical data samples; For each historical data sample, a plurality of prediction models are used to perform target prediction on the historical data sample respectively, and based on the sample prediction values output by the plurality of prediction models respectively, a prediction vector corresponding to the historical data sample is determined; Based on the prediction vectors corresponding to the plurality of historical data samples, the weight values of the plurality of prediction models are optimized and solved to obtain the weight values corresponding to each prediction model, so as to obtain one weight data set.

6. A data prediction apparatus, characterized by comprising: Comprise: The acquisition module is used for acquiring a plurality of prediction models trained in advance, the plurality of prediction models are obtained by machine learning training based on a first sample data set, and the structures of at least two prediction models in the plurality of prediction models are different; The determination module is used for determining the weight coefficients of the plurality of prediction models after receiving a prediction instruction for to-be-predicted data, the weight coefficients of the plurality of prediction models are weight values obtained by weight optimization and solving of the plurality of prediction models based on a second sample data set; The to-be-predicted data is spectrum data of a target object; the prediction target of the to-be-predicted data is a critical dimension or a thickness; or the prediction target of the to-be-predicted data is a defect probability; The determination of the weight coefficients of the plurality of prediction models comprises: Determining a prediction data type for target prediction of the to-be-predicted data, the prediction data type is a size prediction type or a probability prediction type; Obtaining a plurality of weight data sets solved in advance, the weight data set includes weight values corresponding to the plurality of prediction models, and the plurality of weight values in each weight data set are obtained by weight optimization and solving of the plurality of prediction models based on the second sample data set; Determining a target weight data set corresponding to the prediction data type in the plurality of weight data sets, and determining the weight values corresponding to the target weight data set as the weight coefficients of each prediction model; The prediction module is used for performing target prediction on the to-be-predicted data by using the plurality of prediction models respectively to obtain model prediction values output by the plurality of prediction models respectively. The computing module is configured to perform weighted summation calculation on the model prediction values respectively output by the plurality of prediction models based on the weight coefficients of the prediction models, to obtain a target prediction value. The output module is configured to output the target prediction value as a prediction result corresponding to the to-be-predicted data.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the data prediction method according to any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, the computer-readable storage medium comprising: The computer program is executed by the processor to implement the steps of the data prediction method according to any one of claims 1 to 5.

9. A computer program product comprising a computer program, characterized in that, The computer program is executed to cause the data prediction method according to any one of claims 1 to 5 to be performed.

Images