WO2021004198A1

WO2021004198A1 - Plate performance prediction method and apparatus

Info

Publication number: WO2021004198A1
Application number: PCT/CN2020/093942
Authority: WO
Inventors: 孙茂杰; 李福存; 王苏扬; 杨波; 姜跃文; 刘小华; 马超; 谢伟建; 杨爱玲
Original assignee: 江苏金恒信息科技股份有限公司
Priority date: 2019-07-10
Filing date: 2020-06-02
Publication date: 2021-01-14
Also published as: CN110321658A; GB202110100D0; GB2600213A; CN110321658B; GB2600213B

Abstract

A plate performance prediction method and apparatus. The method comprises: obtaining production data of a plate to be predicted (S101); for any type of performance index, determining a plurality of candidate prediction models according to the type of the performance index and a pre-stored correspondence between the type of the performance index and prediction models (S102); further determining a combined model according to the plurality of candidate prediction models and preset weight values of the candidate prediction models (S103); and inputting the production data of the plate to be predicted into the combined model to obtain predicted performance data corresponding to the performance index (S104). By using the method to implement performance prediction on a plate, on one hand, frequent sampling operation steps can be reduced and the time of waiting for performance detection can be shortened, so that the production rhythm is accelerated and the production efficiency is improved; on the other hand, in the method, a plurality of prediction models are comprehensively considered, so that the plate performance prediction accuracy can be further improved.

Description

Method and device for predicting sheet performance

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on July 10, 2019, the application number is 201910626875.5, and the invention title is "a method and device for predicting sheet properties", the entire content of which is incorporated herein by reference Applying.

Technical field

This application relates to the technical field of sheet material quality, and in particular to a method and device for predicting sheet properties.

Background technique

With the popularization of artificial intelligence technology, the concept of intelligent manufacturing has penetrated into all walks of life and has had a huge impact on traditional manufacturing. For example, for the steel industry, its production process includes a series of steps from iron ore to molten steel, continuous casting slab, to hot rolling and cold rolling, which is an extremely complex process. In this complex production process, the quality control of steel plates is always the top priority.

Existing technologies usually use physical metallurgy models to predict the performance data of steel plates, and then control the quality of steel plates. The physical metallurgy model is composed of sub-models such as temperature field, recrystallization, flow stress, precipitation and phase change, and can perform qualitative analysis on the chemical composition, process parameters, microstructure and mechanical properties of steel sheets. However, with the diversification of process requirements and the diversification of product customization, and the method of relying on physical metallurgical models to predict the performance data of steel sheets, the accuracy rate continues to decrease, which makes it more and more difficult to control the quality of steel sheets.

Based on this, there is an urgent need for a prediction method of sheet performance to solve the problem of low accuracy caused by the use of physical metallurgy models to predict the performance data of steel sheets in the prior art.

Summary of the invention

The present application provides a method and device for predicting sheet performance, which can be used to solve the technical problem that the use of physical metallurgy models to predict the performance data of steel sheets in the prior art leads to low accuracy.

In the first aspect, an embodiment of the present application provides a method for predicting the performance of a board, and the method includes:

Acquiring production data of the plate to be predicted, where the production data includes at least one of the chemical composition, process parameters, and laboratory data of the plate to be predicted;

For any type of performance index of the plate to be predicted, a plurality of candidate prediction models are determined according to the type of the performance index and the correspondence between the type of the performance index stored in advance and the prediction model; The prediction model and the weight values of the preset candidate prediction models are determined to determine the combined model; the production data of the board to be predicted is input into the combined model to obtain the predicted performance data of the board to be predicted on the performance index.

Optionally, the candidate prediction model includes an output layer, and the combined model includes an input layer;

According to the multiple candidate prediction models and the preset weight values of the candidate prediction models, determining the combined model includes:

After the output layers of the multiple candidate prediction models are respectively multiplied by the preset weight values of the candidate prediction models, they are used as the input layer of the combined model to determine the combined model.

Optionally, inputting the production data of the plate to be predicted into the combined model to obtain the predicted performance data of the plate to be predicted on the performance index includes:

Inputting the production data of the plate to be predicted into the candidate prediction model to obtain a candidate prediction result;

Determine the input data corresponding to the input layer of the combined model according to the candidate prediction result and the weight value corresponding to the preset candidate prediction model;

The input data is input into the input layer of the combined model to obtain the predicted performance data of the board to be predicted on the performance index.

Optionally, the corresponding relationship between the type of the performance indicator and the prediction model is determined in the following manner:

Acquiring a training data set, the training data set including training production data of a plurality of plates to be trained and actual performance data corresponding to multiple types of performance indicators in the plates to be trained;

For the first type of performance indicators, input the training production data into multiple initial prediction models to obtain prediction results corresponding to the initial prediction models; the first type of performance indicators are multiple types of performance Any one of the indicators;

According to the prediction result corresponding to the initial prediction model and the actual performance data corresponding to the performance index of the first type, reverse training is performed to generate a plurality of prediction models of the first type; the first type of prediction The model is used to predict the performance data corresponding to the first type of performance index;

For any prediction model of the first type, input the training production data into the prediction model of the first type to obtain a prediction result corresponding to the prediction model of the first type;

Determine the error rate of the first type of prediction model according to the prediction result corresponding to the first type of prediction model and the actual performance data corresponding to the first type of performance index;

Determining a candidate prediction model from the plurality of prediction models of the first type according to the error rate of each prediction model of the first type;

Establish a correspondence between the first type of performance indicator and the candidate prediction model.

Optionally, the prediction model includes a polynomial regression model, a ridge regression model, a Kernel Ridge Regression (KRR) model, a Support Vector Machine (SVM) model, and a neighboring algorithm (K-Nearest Neighbor, KNN) Model, Random Forest Model, Gradient Boost Regression Tree (GBRT) model, Extreme Gradient Boosting (XGboost) model, Distributed Gradient Boosting Machine (LightGBM) model and iterative algorithm ( Adaboost) any one of the model.

Optionally, the step of inputting the training production data into a plurality of initial prediction models for the first type of performance indicators includes:

Preprocessing the training data set.

In a second aspect, an embodiment of the present application provides a method for predicting the performance of a board, the method including:

Obtain the production data of the plate to be predicted;

Input the production data of the plate to be predicted into the candidate prediction model to obtain the candidate prediction result;

The input data is input into the input layer of the combined model to obtain the predicted performance data of the sheet to be predicted on the performance index. In a third aspect, an embodiment of the present application provides a device for predicting the performance of a board, and the device includes:

An obtaining unit for obtaining production data of the plate to be predicted, the production data including at least one of the chemical composition, process parameters, and laboratory data of the plate to be predicted;

The processing unit is configured to determine a plurality of candidate prediction models according to the type of the performance index and the correspondence between the type of the performance index stored in advance and the prediction model for any type of performance index of the board to be predicted; The multiple candidate prediction models and the weight values of the preset candidate prediction models are determined to determine the combined model; the production data of the board to be predicted is input into the combined model to obtain the performance index of the board to be predicted Predictive performance data.

The processing unit is specifically used for:

Optionally, the processing unit is specifically configured to:

Input the production data of the board to be predicted into the candidate prediction model to obtain a candidate prediction result; and, according to the candidate prediction result and the weight value corresponding to the preset candidate prediction model, determine the input of the combined model Input data corresponding to the layer; and inputting the input data into the input layer of the combined model to obtain the predicted performance data of the board to be predicted on the performance index.

Optionally, the prediction model includes polynomial regression model, ridge regression model, kernel ridge regression KRR model, support vector machine SVM model, proximity algorithm KNN model, random forest model, progressive gradient regression tree GBRT model, extreme gradient boost XGboost Any one of the model, the distributed gradient boosting framework LightGBM model and the iterative algorithm Adaboost model.

The preprocessing unit performs preprocessing on the training data set.

Using the above method, on the one hand, compared with the physical metallurgical model in the prior art, the embodiments of the present application can predict the performance of the plate based on the chemical composition, process parameters and laboratory data. When predicting the performance of the plate, It can reduce the operation steps of frequent sampling and shorten the waiting time for performance testing, thereby speeding up the production rhythm and improving production efficiency; on the other hand, the combination model is used in the embodiments of the present application to predict the performance data of the plate, because the combination model It is determined by combining multiple candidate prediction models. Therefore, the method comprehensively considers multiple prediction models, which can further improve the prediction accuracy of sheet performance.

Description of the drawings

FIG. 1 is a schematic diagram of a process corresponding to a method for predicting sheet performance provided by an embodiment of the application;

2 is a schematic flowchart of a method for determining the correspondence between the type of performance indicator and the prediction model provided by an embodiment of the application;

FIG. 3 is a schematic structural diagram of a prediction device for sheet performance provided by an embodiment of the application;

4 is a schematic diagram of the calculation result of the error rate of the ridge regression model provided by an embodiment of the application;

FIG. 5 is a schematic diagram of the calculation result of the error rate of the SVM model provided by an embodiment of the application;

FIG. 6 is a schematic diagram of a calculation result of a random forest model error rate provided by an embodiment of the application;

FIG. 7 is a schematic diagram of a calculation result of a LightGBM model error rate provided by an embodiment of the application;

FIG. 8 is a schematic diagram of the calculation result of the error rate of the XGboost model provided by an embodiment of the application;

FIG. 9 is a schematic diagram of the error rate calculation result of the Adaboost model provided by an embodiment of the application.

Detailed ways

In order to make the objectives, technical solutions, and advantages of the present application clearer, the following will further describe the embodiments of the present application in detail with reference to the accompanying drawings.

The board performance prediction method of the present application uses a combination model algorithm to predict the performance indicators of the board according to the given board production data. In order to verify the prediction effect of the combined model, the precision method of root mean square error is used for comparative analysis.

Fig. 1 exemplarily shows a schematic diagram of a process corresponding to a method for predicting sheet performance provided by an embodiment of the present application. As shown in Figure 1, it specifically includes the following steps:

Step 101: Obtain the production data of the plate to be predicted.

Step 102: For any type of performance index of the plate to be predicted, a plurality of candidate prediction models are determined according to the type of the performance index and the correspondence between the type of the performance index stored in advance and the prediction model.

Step 103: Determine a combined model according to the multiple candidate prediction models and preset weight values of the candidate prediction models.

Step 104: Input the production data of the plate to be predicted into the combined model to obtain predicted performance data of the plate to be predicted on the performance index.

The combined model in this application is determined by combining multiple candidate prediction models. First, a single machine learning algorithm is selected (the most basic method of machine learning is to use algorithms to parse data, learn from it, and then make changes to events in the real world. Decision-making and prediction. Unlike traditional software programs that are hard-coded to solve specific tasks, machine learning uses a large amount of data to "train" and learns how to complete tasks from the data through various algorithms.), including polynomial regression models, Ridge regression model, KRR model, SVM model, KNN model, random forest model, GBRT model, XGboost model, LightGBM model and Adaboost model are trained on the training set, and a single model is obtained through continuous tuning and optimization. The best parameters on the set. At this time, the predictive ability of each single model on the training set has reached the upper limit, and it is difficult to obtain better results through retraining optimization. Aiming at the problem that a single algorithm model is difficult to improve the prediction of sheet performance, we will select several single models to combine, and combine the prediction results of the single model to form a stronger classifier. At this time, the single model has reached the optimal parameters For the combined model, it is equivalent to the model has been pre-trained. We use the combined model to continuously train on the training set again to obtain the optimal parameters of the combined model.

Specifically, in step 101, the production data may include at least one of the chemical composition, process parameters, and laboratory data of the plate to be predicted. Among them, taking steel sheet as an example, the chemical composition may include chemical elements such as carbon, silicon, manganese, chromium, nickel, and copper. The process parameters can refer to the processing process parameters of the steel plate, and can include parameters such as plate thickness, plate width, rolling pressure parameters, and quenching temperature. Laboratory data can refer to the content percentage of each chemical component. In step 102, there are multiple types of performance indicators of the plate to be predicted. Taking steel plates as an example, the types of performance indicators may include at least one of yield strength, tensile strength, elongation, and impact energy.

The prediction model may include any one of polynomial regression model, ridge regression model, KRR model, SVM model, KNN model, random forest model, GBRT model, XGboost model, LightGBM model, and Adaboost model.

The corresponding relationship between the type of performance indicator and the prediction model can be determined in multiple ways. A possible implementation is, as shown in FIG. 2, the corresponding relationship between the type of performance indicator and the prediction model provided in this embodiment of the application. The schematic diagram of the process corresponding to the method of determining includes the following steps:

Step 201: Obtain a training data set.

The training data set may include training production data of multiple plates to be trained and actual performance data corresponding to multiple types of performance indicators in the plates to be trained. The training production data may refer to the production data used to train the model, and the specific data included in the training production data may be consistent with the specific data included in the production data described above.

Further, the way to obtain the training data set can be through ERP, MES and other office systems to obtain production data such as chemical composition, process parameters, and laboratory data, and then perform data integration on these data. For example, you can use data extraction ETL related technology to convert the production data Store to the specified database. Among them, the stored production data may be in the form of a data table.

Step 202: Regarding the first type of performance indicators, input training production data into multiple initial prediction models, respectively, to obtain prediction results corresponding to the initial prediction models.

Wherein, the first type of performance indicator is any one of multiple types of performance indicators. Taking steel sheet as an example, the first type of performance index can be any one of yield strength, tensile strength, elongation and impact energy.

It should be noted that before performing step 202, data preprocessing may be performed on the training data set. Specifically, data cleaning can be performed on the stored production data, including processing of null values and abnormal values, vectorization of category variables, data deduplication, and data de-drying.

The preprocessing of the training data set in this application can not only provide data of specifications for the input of multiple initial prediction models, but also avoid the interference of erroneous data and abnormal values through the preprocessed data, thereby improving the efficiency of data processing . At the same time, the data preprocessing in this application has specific application scenarios. The preprocessing step is performed after the training data set is obtained. Our training data set is the training production data of multiple plates to be trained and the plates to be trained. Multiple types of performance indicators correspond to actual performance data. Therefore, the pre-processing process arrangement has a specific role in a specific link. The specific description is as follows:

It is worth noting that the data preprocessing method of this application is not easy to think of, and is different from ordinary data processing methods. First of all, in daily work applications, most general data processing simply uses ETL or catalog for simple processing. For example, ETL is the process of data extraction (Extract), transformation (Transform), and loading (Load). Catalog is used for data. Operations such as transfer in and out, but based on these two commonly used methods, basically only complete operations such as the removal of null values and outliers of the data, which can meet the needs of daily work applications.

However, in this application, the training data set is preprocessed. The preprocessing includes many aspects. For example, one includes the processing of incomplete data, for example, the processing of missing certain attribute values of interest; One includes the processing of inconsistent data, such as the difference processing for codes or names; also includes the processing of errors or data with outliers, etc. The data set often comes from multiple heterogeneous data sources, so it is solved by preprocessing The problem in the specific scenario above. Specifically, data preprocessing includes data cleaning, data integration, data specification, and data transformation. In this application, the pre-processed data can not only meet the requirement of "input multiple initial prediction models", but also has the following effects: the training data set is preprocessed to greatly avoid the interference of junk data, and can affect the data set. Compression, for example, from the original 2G to 500M, which greatly improves the efficiency of data operation, thereby increasing work efficiency. Step 203: Perform reverse training according to the prediction result corresponding to the initial prediction model and the actual performance data corresponding to the performance index of the first type to generate multiple prediction models of the first type.

The prediction model of the first type may be used to predict the performance data corresponding to the performance index of the first type.

Step 204: For any prediction model of the first type, input the training production data into the prediction model of the first type to obtain a prediction result corresponding to the prediction model of the first type.

Step 205: Determine the error rate of the first type of prediction model according to the prediction result corresponding to the first type of prediction model and the actual performance data corresponding to the first type of performance index.

Specifically, the error rate of the first type of prediction model can be calculated by using the root mean square error, and the specific calculation method can refer to formula (1).

In formula (1), RMSE is the error rate of the first type of prediction model; N is the number of samples in the training data set to be trained, N is an integer greater than 1; y _i is the true value of the i-th sample in the training data set ；

It is the prediction result of the first type of prediction model on the i-th sample in the training data set.

The formula (1) is the error rate of the first type of prediction model. The example of this application is the root mean square error, and the root mean square error represents the magnitude of the error between the predicted value of the model and the true value. The smaller the root mean square error, the closer the model prediction value is to the true value. Taking the performance prediction of yield strength as an example, machine learning algorithms are used to train and test on the training data set of the plate to be trained. The calculation results of some prediction models are listed here, and the results are only examples. See Figure 4-9. Specifically:

FIG. 4 is a schematic diagram of the calculation result of the ridge regression model error rate provided by an embodiment of the application, and it can be seen that the calculation result of the ridge regression model error rate is 17.202095;

FIG. 5 is a schematic diagram of the calculation result of the SVM model error rate provided by an embodiment of the application, and it can be seen that the calculation result of the SVM model error rate is 19.115036;

FIG. 6 is a schematic diagram of the error rate calculation result of the random forest model provided by the embodiment of the application, and it can be seen that the error rate calculation result of the random forest model is 13.790967;

FIG. 7 is a schematic diagram of the calculation result of the LightGBM model error rate provided by an embodiment of the application, and it can be seen that the calculation result of the LightGBM model error rate is 7.044598;

FIG. 8 is a schematic diagram of the calculation result of the error rate of the XGboost model provided by an embodiment of the application, and it can be seen that the calculation result of the error rate of the XGboost model is 14.185270;

FIG. 9 is a schematic diagram of the calculation result of the Adaboost model error rate provided by the embodiment of the application, and it can be seen that the calculation result of the Adaboost model error rate is 17.617467.

Among them, in Figure 4-9, black and gray represent the true value of the training data set of the plate to be trained (the yield strength is taken as an example in this embodiment), and black represents the training data set of the plate to be trained (the yield strength is taken as an example in this embodiment) The idea of the combined model is to select the above-mentioned well-performing models for weighted combination. The combined model comprehensively considers multiple prediction models, which can further improve the prediction accuracy of sheet performance.

It should be noted that formula (1) is only one possible calculation method, and those skilled in the art can also select other error calculation methods based on experience and actual conditions, and the specifics are not limited.

Step 206: Determine a candidate prediction model from a plurality of prediction models of the first type according to the error rate of each prediction model of the first type.

Specifically, there can be many ways to determine candidate prediction models from multiple prediction models of the first type. In one example, the error rate can be ranked from small to large, and then the top M prediction models can be used as Candidate prediction model, where M is an integer greater than 1. In another example, a prediction model with an error rate less than a preset threshold may be used as a candidate prediction model, where the preset threshold may be determined by those skilled in the art based on experience and actual conditions, and is not specifically limited.

For example, as shown in Table 1, it is an example of the error rate of the first type of prediction model. The specific content can refer to the content listed in Table 1, which will not be described here.

Table 1: An example of the error rate of the first type of prediction model

第一类型的预测模型The first type of predictive model	误差率(RMSE)Error rate (RMSE)
多项式回归模型Polynomial regression model	17.20317.203
岭回归模型Ridge Regression Model	17.20217.202
KRR模型KRR model	15.08515.085
SVM模型SVM model	19.11519.115
KNN模型KNN model	13.79413.794

随机森林模型Random forest model	13.79113.791
GBRT模型GBRT model	7.0457.045
XGboost模型XGboost model	11.97611.976
LightGBM模型LightGBM model	14.18514.185
Adaboost模型Adaboost model	17.61717.617

Among them, the first column of Table 1 is the prediction model of the first type, and the second column of Table 1 is the corresponding error rate corresponding to the specific category of the first type of prediction model, that is, for the different prediction models in the first column Corresponding values of different error rates. Table 1 clearly lists the specific error rate of each model. Based on this, the candidate prediction model is determined from multiple first-type prediction models.

Further, according to the content shown in Table 1, combined with the above-provided method of using the top M prediction models as candidate prediction models, assuming M=5, the candidate prediction models include random forest model, KNN model, GBRT model, XGboost model and LightGBM model.

The confirmation method of the above-mentioned candidate prediction model is to assume that M=5, rank the error rate from small to large, and then use the top 5 prediction models as candidate prediction models.

Step 207: Establish a correspondence between the first type of performance index and the candidate prediction model.

It should be noted that, in other possible implementation manners, the corresponding relationship between the type of performance index and the prediction model may also be determined by a person skilled in the art based on experience or actual conditions, and is not specifically limited.

Step 101 is mainly for data collection, selecting data that has an impact on model performance, and using these data for model training; step 102 is mainly for selecting which models to fuse, and performing operations such as data integration and data preprocessing through the data provided in step 101 Obtain a training data set, and use the obtained data set to train multiple models to evaluate the accuracy of each model.

In step 103, there are many ways to set the weight value of the candidate prediction model. In one example, the weight value of the candidate prediction model can be determined according to the error rate of the candidate prediction model. For example, the smaller the error rate of the candidate prediction model, the corresponding The higher the weight value.

In other possible examples, the weight value of the candidate prediction model can also be determined by those skilled in the art based on experience and actual conditions. For example, if the importance of the random forest model is higher by those skilled in the art based on experience tasks, it can be random. The forest model sets a higher weight value.

Further, the candidate prediction model may include an output layer, and the combined model may include an input layer. Specifically, there can be multiple ways to determine the combination model. Two possible ways are described below.

Manner 1: The output layers of multiple candidate prediction models can be respectively multiplied by the preset weight values of the candidate prediction models and used as the input layer of the combined model to determine the combined model.

Method 2: The candidate prediction model can be used as a sub-model. After the sub-model is multiplied by the preset weight value of the candidate prediction model, the large model obtained can be regarded as a combined model. In step 104, taking the combination model determination method as the above-mentioned method 1 as an example, the production data of the plate to be predicted can be input into the candidate prediction model to obtain the candidate prediction result; then, the candidate prediction result can be obtained according to the candidate prediction result and the preset candidate prediction The weight value corresponding to the model determines the input data corresponding to the input layer of the combined model; finally, the input data can be input into the input layer of the combined model to obtain the predicted performance data of the sheet to be predicted on the performance index. The following are device embodiments of this application, which can be used to implement the method embodiments of this application. For details not disclosed in the device embodiment of this application, please refer to the method embodiment of this application.

Fig. 3 exemplarily shows a schematic structural diagram of a device for predicting sheet performance provided by an embodiment of the application. As shown in Fig. 3, the device has the function of realizing the above prediction method of sheet performance, and the function can be realized by hardware, or by hardware executing corresponding software. The device may include: an acquiring unit 301 and a processing unit 302.

The obtaining unit 301 is configured to obtain production data of the plate to be predicted, the production data including at least one of the chemical composition, process parameters, and laboratory data of the plate to be predicted;

The processing unit 302 is configured to determine multiple candidate prediction models according to the type of the performance index and the corresponding relationship between the type of the performance index and the prediction model stored in advance for any type of performance index of the plate to be predicted; Determine the combined model according to the multiple candidate prediction models and the weight values of the preset candidate prediction models; input the production data of the to-be-predicted board into the combined model to obtain the performance index of the to-be-predicted board Predictive performance data on

The processing unit 302 is specifically configured to:

Optionally, the processing unit 302 is specifically configured to:

The preprocessing unit performs preprocessing on the training data set.

In an exemplary embodiment, a computer-readable storage medium is also provided, the storage medium stores a computer program or smart contract, and the computer program or smart contract is loaded and executed by a node to implement the above-mentioned embodiments. Transaction processing method. Optionally, the above-mentioned computer-readable storage medium may be a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, etc. .

Those skilled in the art can clearly understand that the technology in the embodiments of the present application can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solutions in the embodiments of the present application can be embodied in the form of a software product in essence or a part that contributes to the prior art. The computer software product can be stored in a storage medium, such as ROM/RAM , Magnetic disks, optical disks, etc., including a number of instructions to enable a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the application or some parts of the embodiment.

Those skilled in the art will easily think of other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptive changes of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or conventional technical means in the technical field not disclosed in the present disclosure . The description and the embodiments are to be regarded as exemplary only, and the true scope and spirit of the present disclosure are pointed out by the following claims.

It should be understood that the present disclosure is not limited to the precise structure that has been described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present disclosure is only limited by the appended claims.

Claims

A method for predicting sheet performance, characterized in that, the method includes:

Acquiring production data of the plate to be predicted, where the production data includes at least one of the chemical composition, process parameters, and laboratory data of the plate to be predicted;

For any type of performance index of the plate to be predicted, a plurality of candidate prediction models are determined according to the type of the performance index and the correspondence between the type of the performance index stored in advance and the prediction model; The prediction model and the weight values of the preset candidate prediction models are determined to determine the combined model; the production data of the board to be predicted is input into the combined model to obtain the predicted performance data of the board to be predicted on the performance index.
The method according to claim 1, wherein the candidate prediction model includes an output layer, and the combined model includes an input layer;

According to the multiple candidate prediction models and the preset weight values of the candidate prediction models, determining the combined model includes:

After the output layers of the multiple candidate prediction models are respectively multiplied by the preset weight values of the candidate prediction models, they are used as the input layer of the combined model to determine the combined model.
The method according to claim 2, wherein inputting the production data of the plate to be predicted into the combined model to obtain the predicted performance data of the plate to be predicted on the performance index comprises:

Inputting the production data of the plate to be predicted into the candidate prediction model to obtain a candidate prediction result;

Determine the input data corresponding to the input layer of the combined model according to the candidate prediction result and the weight value corresponding to the preset candidate prediction model;

The input data is input into the input layer of the combined model to obtain the predicted performance data of the board to be predicted on the performance index.
The method according to claim 1, wherein the corresponding relationship between the type of the performance indicator and the prediction model is determined in the following manner:

Acquiring a training data set, the training data set including training production data of a plurality of plates to be trained and actual performance data corresponding to multiple types of performance indicators in the plates to be trained;

For the first type of performance indicators, input the training production data into multiple initial prediction models to obtain prediction results corresponding to the initial prediction models; the first type of performance indicators are multiple types of performance Any one of the indicators;

According to the prediction result corresponding to the initial prediction model and the actual performance data corresponding to the performance index of the first type, reverse training is performed to generate a plurality of prediction models of the first type; the first type of prediction The model is used to predict the performance data corresponding to the first type of performance index;

For any prediction model of the first type, input the training production data into the prediction model of the first type to obtain a prediction result corresponding to the prediction model of the first type;

Determine the error rate of the first type of prediction model according to the prediction result corresponding to the first type of prediction model and the actual performance data corresponding to the first type of performance index;

Determining a candidate prediction model from the plurality of prediction models of the first type according to the error rate of each prediction model of the first type;

Establish a correspondence between the first type of performance indicator and the candidate prediction model.
The method according to claim 1, wherein the prediction model includes a polynomial regression model, a ridge regression model, a kernel ridge regression KRR model, a support vector machine SVM model, a neighboring algorithm KNN model, a random forest model, and progressive gradient regression Any one of tree GBRT model, extreme gradient boost XGboost model, distributed gradient boosting framework LightGBM model and iterative algorithm Adaboost model.
The method according to claim 4, characterized in that, before the step of inputting the training production data into a plurality of initial prediction models for the first type of performance indicators, comprises: pre-processing the training data set deal with.
A method for predicting sheet performance, characterized in that, the method includes:

Obtain the production data of the plate to be predicted;

Input the production data of the plate to be predicted into the candidate prediction model to obtain the candidate prediction result;

Determine the input data corresponding to the input layer of the combined model according to the candidate prediction result and the weight value corresponding to the preset candidate prediction model;

The input data is input into the input layer of the combined model to obtain the predicted performance data of the sheet to be predicted on the performance index.
A prediction device for sheet performance, characterized in that the device comprises:

An obtaining unit for obtaining production data of the plate to be predicted, the production data including at least one of the chemical composition, process parameters, and laboratory data of the plate to be predicted;

The processing unit is configured to determine a plurality of candidate prediction models according to the type of the performance index and the correspondence between the type of the performance index stored in advance and the prediction model for any type of performance index of the board to be predicted; The multiple candidate prediction models and the weight values of the preset candidate prediction models are determined to determine the combined model; the production data of the board to be predicted is input into the combined model to obtain the performance index of the board to be predicted Predictive performance data.
The device according to claim 8, wherein the candidate prediction model comprises an output layer, and the combined model comprises an input layer;

The processing unit is specifically used for:

After the output layers of the multiple candidate prediction models are respectively multiplied by the preset weight values of the candidate prediction models, they are used as the input layer of the combined model to determine the combined model.
The device according to claim 9, wherein the processing unit is specifically configured to:

Input the production data of the board to be predicted into the candidate prediction model to obtain a candidate prediction result; and, according to the candidate prediction result and the weight value corresponding to the preset candidate prediction model, determine the input of the combined model Input data corresponding to the layer; and inputting the input data into the input layer of the combined model to obtain the predicted performance data of the board to be predicted on the performance index.
The device according to claim 8, wherein the corresponding relationship between the type of the performance indicator and the prediction model is determined in the following manner:

Acquiring a training data set, the training data set including training production data of a plurality of plates to be trained and actual performance data corresponding to multiple types of performance indicators in the plates to be trained;

For the first type of performance indicators, input the training production data into multiple initial prediction models to obtain prediction results corresponding to the initial prediction models; the first type of performance indicators are multiple types of performance Any one of the indicators;

According to the prediction result corresponding to the initial prediction model and the actual performance data corresponding to the performance index of the first type, reverse training is performed to generate a plurality of prediction models of the first type; the first type of prediction The model is used to predict the performance data corresponding to the first type of performance index;

For any prediction model of the first type, input the training production data into the prediction model of the first type to obtain a prediction result corresponding to the prediction model of the first type;

Determine the error rate of the first type of prediction model according to the prediction result corresponding to the first type of prediction model and the actual performance data corresponding to the first type of performance index;

Determining a candidate prediction model from the plurality of prediction models of the first type according to the error rate of each prediction model of the first type;

Establish a correspondence between the first type of performance indicator and the candidate prediction model.
The device according to claim 8, wherein the prediction model comprises a polynomial regression model, a ridge regression model, a kernel ridge regression KRR model, a support vector machine SVM model, a neighboring algorithm KNN model, a random forest model, and a progressive gradient regression Any one of tree GBRT model, extreme gradient boost XGboost model, distributed gradient boosting framework LightGBM model and iterative algorithm Adaboost model.
The device according to claim 8, wherein the device further comprises:

The preprocessing unit performs preprocessing on the training data set.