CN114035529A - ATL-BMA-based low-cost modeling method for nonlinear industrial process - Google Patents

Abstract

The invention provides a low-cost modeling method for nonlinear industrial process based on ATL-BMA, which selects N groups of similar old process modeling data; collects an initial data set for new process modeling; divides the new and old process data into two parts respectively, And normalize them respectively; convert N groups of old process data into N groups of old process data with new process information, and mix them with the corresponding old process data to obtain N groups of mixed data sets, and then train the support vector machine model, Obtain N old process basic models with new process information; map the input variables of the new process training set to the running range of similar old process input variables, and obtain the fusion output of these N prediction models; fuse the old process SVM model to output And the new process input data is used as the input data of the multi-model migration strategy, and the new process model is obtained by training. This method can effectively solve the problems of high modeling cost of complex industrial process, limited modeling data and long modeling period.

Description

A Low-Cost Modeling Method for Nonlinear Industrial Processes Based on ATL-BMA

technical field

The invention belongs to the technical field of industrial process construction performance prediction models, and in particular relates to a low-cost modeling method for nonlinear industrial processes based on ATL-BMA.

Background technique

In order to meet the needs of the market for products with multiple specifications, varieties and high quality, modern industrial processes are moving towards large-scale, high-efficiency and integration. On the one hand, with the gradual expansion of the production scale, new industrial production processes will continue to be added in the actual production process to meet the needs of different products, which also leads to higher and higher complexity of the actual industrial production process. On the other hand, changes in the operating environment and increases in operating time can lead to changes in the characteristics of actual industrial processes. Both of these aspects contribute to the variable nature of process data. A thorny problem needs to be solved when modeling industrial processes with data-driven methods in this case: due to various factors such as cost, the modeling data obtained from new industrial processes is seriously insufficient, and with the support of a small amount of modeling data Data-driven modeling methods cannot be used to establish accurate process prediction models, and the generalization ability of the resulting models is low. Faced with this situation, it is hoped that the existing industrial production process data or knowledge with a long running time can help guide the establishment of predictive models for new industrial processes. Although there are some differences in the characteristics of the old and new industrial process operation data, the physical and chemical mechanisms followed within the process are unchanged or very similar, so the new industrial process data and the old industrial process data have the same or similar feature space and Label space (both input and output data dimensions are the same). As shown in Figure 1, the new industrial process and the old industrial process can be regarded as the target domain and the source domain, respectively, and then the old industrial process data is used to assist in establishing the new industrial process prediction model through the transfer learning method. However, when the source domain data is far more than the target domain data, the phenomenon of "negative transfer" is prone to occur when the source domain data is used to supplement the target domain data under the traditional transfer learning structure.

SUMMARY OF THE INVENTION

Aiming at the problems existing in the above-mentioned prior art, the present invention provides a low-cost modeling method for nonlinear industrial processes based on ATL-BMA, which can effectively solve the problems of high modeling cost of complex industrial processes, limited acquired modeling data, and construction problems. At the same time, it can solve the "negative transfer" phenomenon that may occur when the old process data is much more than the new process data in transfer learning, and it can make full use of the information of existing similar old industrial process models to assist in guiding the establishment of new industrial processes. It can effectively reduce the modeling cost, speed up the modeling speed and improve the modeling accuracy.

In order to achieve the above object, the present invention provides a low-cost modeling method for nonlinear industrial processes based on ATL-BMA, comprising the following steps:

Step 1: Select N groups of similar old industrial process modeling data, and determine the stable operating range of the input variables according to the information of the actual process to be modeled; at the same time, select the Latin hypercube method to sample and collect the target nonlinear industrial process modeling The initial data set; its specific steps are as follows:

根据公式(1)对于第i个旧工业过程进行建模；Step 1.1: Select N groups of similar old industrial process modeling data, denoted as

Model the i-th old industrial process according to formula (1);

In the formula, X and x are the input data of the old industrial process, y is the output data of the old industrial process, _ki represents the amount of modeling data of the ith old industrial process, and n is the input variable dimension of the ith old industrial process. There is a certain similarity in the process, so the input variable dimensions for all industrial processes are the same, and they are all n;

Step 1.2: According to the information of the actual process to be modeled, determine the stable operating range of the input variables, and select discrete and sparse data distribution points to sample and collect new industrial process modeling data, and obtain the collected new industrial process according to formula (2). data D _new ;

In the formula, l represents the amount of new industrial process modeling data;

和新工业过程测试数据集

并利用公式(3)将数据映射到[0,1]区间；Step 2: Divide the new industrial process data and the old industrial process data into two parts, which are the training data set and the test data set in the new and old modeling process, respectively, and divide the new industrial process initial data set and the old industrial process modeling data into two parts. Normalization processing is performed separately; among them, for the new industrial process data, it is divided into a new industrial process training data set

and new industrial process test datasets

And use formula (3) to map the data to the [0,1] interval;

In the formula, _zi represents the result of the normalization of the input or output data of the industrial process, _xi is the data before normalization, x _max is the maximum value before the data normalization, and x _min is the minimum value;

其具体步骤如下：Step 3: Use the old and new industrial process data migration algorithm based on Cycle GANs to convert N groups of old industrial process data into N groups of old industrial process data with new industrial process information; among them, the old industrial process training data set is

The specific steps are as follows:

Step 3.1: Initialization parameters: G parameter θ _G , D _o parameter ω _o , F parameter θ _F , D _n parameter ω _n , n _critic =5, α = 0.00005, β ₁ =0, β ₂ =0.7, m = 5 , λ=0.5, Epoch=20000;

Among them: G represents the generator function from the old industrial process to the new industrial process data, D _o represents the discriminator corresponding to the old industrial process, F represents the generator function from the new industrial process to the old industrial process, D _n represents the corresponding Discriminator, n _critic represents the number of times of training the discriminant model after training the generator once, α, β ₁ and β ₂ are the parameters of Adam optimizer, m is the number of samples, and Epoch is the number of training cycles of the model;

中采集的m个样本

转化成m个新工业过程数据，记为X_o→n＝F(X_o)；通过生成器F将从新工业过程数据

中采集的m个样本

转化成m个旧工业过程数据，记为X_n→o＝F(X_n)；Step 3.2: From the i-th old industrial process data by generator G

m samples collected in

Convert into m new industrial process data, denoted as X _o→n =F(X _o ); through the generator F from the new industrial process data

m samples collected in

Convert into m old industrial process data, denoted as X _n→o =F(X _n );

Step 3.3: According to formula (4) and formula (5), the discriminator loss and the two forward loop consistency losses are obtained;

Step 3.4: update the discriminator D _o parameter ω _o and D _n parameter ω _n by formula (6) and formula (7);

Step 3.5: Repeat steps 3.2 to 3.4n _critic times;

Step 3.6: Repeat step 3.2;

Step 3.7: Calculate the consistent loss of the two forward loops by formula (8) and formula (9);

Step 3.8: Calculate the two generator losses by Equation (10) and Equation (11);

Step 3.9: Update the generator G parameter θ _G and F parameter θ _F by formula (12) and formula (13);

Step 3.10: Repeat steps 3.6 to 3.9 Epoch times, and use the trained F to convert the new industrial process data into the jth old industrial process data, denoted as

Step 3.11: Repeat steps 3.1 to 3.9 with each set of old industrial process data, migrate the new industrial process data into the old industrial process domain, and obtain N groups of old industrial process data with new industrial process information through adversarial transfer learning from the new industrial process data. Industrial process data, denoted as

Step 4: After mixing the old industrial process data with the new industrial process information in step 3 and the corresponding old industrial process data, N groups of mixed data sets are obtained;

和混合测试数据集

同时，结合N个旧工业过程训练数据集

和新工业过程预测模型y＝f(x)，利用N组混合数据集分别训练支持向量机SVM模型，得到N个带有新工业过程信息的旧工业过程基础模型，记为f₁(·)-f_N(·)；其中，

k_train是训练数据集大小，

k_test是测试数据集大小，任意第i个旧工业过程，

n_i是第i个旧工业过程训练集大小；其具体步骤如下：Step 5: Divide the mixed dataset into a mixed training set

and a mixed test dataset

At the same time, combine N old industrial process training datasets

and the new industrial process prediction model y=f(x), use N groups of mixed data sets to train the support vector machine SVM model respectively, and obtain N old industrial process basic models with new industrial process information, denoted as f ₁ ( ) -f _N ( ); where,

k _train is the training dataset size,

k _test is the test dataset size, any ith old industrial process,

n _i is the training set size of the ith old industrial process; its specific steps are as follows:

Step 5.1: Initialize parameters;

根据公式(14)混合D_n→o和D_o得到N组基础模型训练数据D_Basic；Step 5.2: Convert the new industrial process data into N groups of old industrial process data carrying the new industrial process information through the new and old industrial process data migration algorithm based on Cycle GANs

According to formula (14), D _n→o and D _o are mixed to obtain N groups of basic model training data D _Basic ;

Step 5.3: Use D _Basic to train N SVMs to obtain N old industrial process basic models with new industrial process information, denoted as f ₁ ( )-f _N ( );

通过贝叶斯模型平均算法得到这N个预测模型的融合输出

Step 6: Use the model fusion formula (15) to map the input variables of the new industrial process training set to the operating range of the similar old industrial process input variables, and the transformed input data of the new industrial process training set is recorded as

The fusion output of these N prediction models is obtained through the Bayesian model averaging algorithm

和新工业过程输入数据

作为多模型迁移策略的输入数据，利用最小二乘支持向量机算法训练新工业过程模型，获得新工业过程模型输出

完成新工业过程建模；Step 7: Fusion output from the old industrial process SVM model

and new industrial process input data

As the input data of the multi-model migration strategy, use the least squares support vector machine algorithm to train the new industrial process model, and obtain the output of the new industrial process model

Complete new industrial process modeling;

Step 8: Model verification, using the root mean square error and the coefficient of determination to evaluate the validity of the SVM model according to formula (16) and formula (17) respectively, if the prediction accuracy of the model obtained in step 7 on the test data set meets the experimental setting threshold, the modeling process is completed; otherwise, repeat steps 3 to 7, add new N groups of old industrial process data samples containing new industrial process information to the mixed samples, and continue to train the new industrial process model until the experiment stops condition;

是预测输出的均值，Y_i是新工业过程的真实输出。where N is the number of test data, _yi is the output of the prediction model,

is the mean of the predicted output and _Yi is the true output of the new industrial process.

This method first uses the Latin hypercube method to collect a small sample data set for nonlinear industrial process modeling, combines multiple similar old process data, and learns the conversion mapping function between the new industrial process data and the old industrial process data through an adversarial transfer algorithm. , so as to convert a small amount of new process data into various types of old industrial process data with new industrial process information; The modeling of the new industrial process establishes the foundation; finally, using a multi-model transfer strategy and the Bayesian model averaging theory, several trained "old industrial process prediction models with new industrial process information" are transferred, combined with a small amount of new Industrial process data, resulting in the final new industrial process performance prediction model. The present invention migrates useful information of a plurality of existing similar old industrial processes to help establish a new industrial process performance prediction model and reduce the cost of new industrial process modeling; Negative transfer" problem, using the old and new process data transfer method based on adversarial transfer learning to improve the effect of transfer modeling. This method effectively solves the problems of high cost and long modeling period of complex industrial process modeling, makes full use of the useful information of existing similar old industrial process models, and solves the problem when the old process data is much more than the new process data in transfer learning. The possible phenomenon of "negative transfer" completes the modeling of new industrial processes, reduces modeling costs, speeds up modeling, and improves modeling accuracy.

Description of drawings

Figure 1 is a flow chart of migration modeling;

Figure 2 is a flowchart of a low-cost modeling method for nonlinear industrial processes based on adversarial transfer learning and Bayesian model averaging theory;

Fig. 3 is the graph of the predicted value of ATL-BMA model, BMA model and SVM model on the compressor A test set;

Fig. 4 is the RMSE histogram of ATL-BMA model, BMA model and SVM model predicted value and true value;

Figure ⁵ is a histogram of R2 for the predicted and true values of the ATL-BMA model, the BMA model and the SVM model.

Detailed ways

The present invention will be further described below with reference to the embodiments and accompanying drawings.

As shown in FIGS. 1 to 5 , the present invention provides a low-cost nonlinear industrial process based on ATL-BMA (Adversarial Transform Learning (ATL) and Bayesian Model Averaging (BMA)). Modeling method, including the following steps:

Step 1: Select N groups of similar old industrial process modeling data, and determine the stable operating range of the input variables according to the information of the actual process to be modeled; at the same time, select the Latin Hypercube Design (LHD) method for sampling and collection The initial data set for modeling the target nonlinear industrial process (new industrial process); the information of the actual process to be modeled includes parameter ratings and performance curves, etc. The specific steps are as follows:

根据公式(1)对于第i个旧工业过程进行建模；Step 1.1: Select N groups of similar old industrial process modeling data, denoted as

Model the i-th old industrial process according to formula (1);

where X is the input data set of the old industrial process, x is the input data of the old industrial process, y is the output data of the old industrial process, _ki represents the amount of modeling data of the ith old industrial process, and n is the data of the ith old industrial process. Input variable dimension, because there is a certain similarity between the old and new industrial processes, the input variable dimension for all industrial processes is the same, and it is n;

Step 1.2: According to the information of the actual process to be modeled, determine the stable operating range of the input variables, and select discrete and sparse data distribution points to sample and collect new industrial process modeling data, and obtain the collected new industrial process according to formula (2). data D _new ;

In the formula, l represents the amount of new industrial process modeling data;

和新工业过程测试数据集

并根据公式(3)利用最大值最小值数据归一化方法将数据映射到[0,1]区间；Step 2: Divide the new industrial process data and the old industrial process data into two parts, which are the training data set and the test data set in the new and old modeling process; for the stability of the subsequent training process, and to avoid the difference in data dimensions The adverse effects caused, it is necessary to ensure that the data is normalized, and the initial data set of the new industrial process and the old industrial process modeling data are normalized separately; among them, for the new industrial process data, it is divided into new industrial processes. training dataset

and new industrial process test datasets

And according to formula (3), the data is mapped to the [0,1] interval by using the maximum and minimum data normalization method;

In the formula, _zi represents the result of the normalization of the input or output data of the industrial process, _xi is the data before normalization, x _max is the maximum value before the data normalization, and x _min is the minimum value;

其具体步骤如下：Step 3: Use the old and new industrial process data migration algorithm based on Cycle GANs to convert N groups of old industrial process data into N groups of old industrial process data with new industrial process information; among them, the old industrial process training data set is

The specific steps are as follows:

Step 3.1: Initialization parameters: G parameter θ _G , D _o parameter ω _o , F parameter θ _F , D _n parameter ω _n , n _critic =5, α = 0.00005, β ₁ =0, β ₂ =0.7, m = 5 , λ=0.5, Epoch=20000;

Among them: G represents the generator function from the old industrial process to the new industrial process data, D _o represents the discriminator corresponding to the old industrial process, F represents the generator function from the new industrial process to the old industrial process, D _n represents the corresponding Discriminator, n _critic represents the number of times of training the discriminant model after training the generator once, α, β ₁ and β ₂ are the parameters of Adam optimizer, m is the number of samples, and Epoch is the number of training cycles of the model;

中采集的m个样本

转化成m个新工业过程数据，记为X_o→n＝F(X_o)；通过生成器F将从新工业过程数据

中采集的m个样本

转化成m个旧工业过程数据，记为X_n→o＝F(X_n)；Step 3.2: From the i-th old industrial process data by generator G

m samples collected in

Convert into m new industrial process data, denoted as X _o→n =F(X _o ); through the generator F from the new industrial process data

m samples collected in

Convert into m old industrial process data, denoted as X _n→o =F(X _n );

Step 3.3: According to formula (4) and formula (5), the discriminator loss and the two forward loop consistency losses are obtained;

Step 3.4: update the discriminator D _o parameter ω _o and D _n parameter ω _n by formula (6) and formula (7);

Step 3.5: Repeat steps 3.2 to 3.4n _critic times;

Step 3.6: Repeat step 3.2;

Step 3.7: Calculate the consistent loss of the two forward loops by formula (8) and formula (9);

Step 3.8: Calculate the two generator losses by Equation (10) and Equation (11);

Step 3.9: Update the generator G parameter θ _G and F parameter θ _F by formula (12) and formula (13);

Step 3.10: Repeat steps 3.6 to 3.9 Epoch times, and use the trained F to convert the new industrial process data into the jth old industrial process data, denoted as

Step 3.11: Repeat steps 3.1 to 3.9 with each group of old industrial process data, migrate the new industrial process data to the old industrial process domain, and obtain N groups of old industrial process data with new industrial process information through adversarial transfer learning from the new industrial process data. Industrial process data, denoted as

Step 4: After mixing the old industrial process data with the new industrial process information in step 3 and the corresponding old industrial process data, N groups of mixed data sets are obtained;

和混合测试数据集

同时，结合N个旧工业过程训练数据集

和新工业过程预测模型y＝f(x)，利用N组混合数据集分别训练支持向量机SVM(Support Vector Machine)模型，得到N个带有新工业过程信息的旧工业过程基础模型，记为f₁(·)-f_N(·)；其中，Step 5: Divide the mixed dataset into a mixed training set

and a mixed test dataset

At the same time, combine N old industrial process training datasets

and the new industrial process prediction model y=f(x), use N groups of mixed data sets to train the SVM (Support Vector Machine) model respectively, and obtain N old industrial process basic models with new industrial process information, denoted as f ₁ (·)-f _N (·); where,

k_train是训练数据集大小，

k_test是测试数据集大小，任意第i个旧工业过程，

n_i是第i个旧工业过程训练集大小；其具体步骤如下：

k _train is the training dataset size,

k _test is the test dataset size, any ith old industrial process,

n _i is the training set size of the ith old industrial process; its specific steps are as follows:

Step 5.1: Initialize parameters;

根据公式(14)混合D_n→o和D_o得到N组基础模型训练数据D_Basic；Step 5.2: Convert the new industrial process data into N groups of old industrial process data carrying the new industrial process information through the new and old industrial process data migration algorithm based on Cycle GANs

According to formula (14), D _n→o and D _o are mixed to obtain N groups of basic model training data D _Basic ;

Step 5.3: Use D _Basic to train N SVMs to obtain N old industrial process basic models with new industrial process information, denoted as f ₁ ( )-f _N ( );

通过贝叶斯模型平均算法得到这N个预测模型的融合输出

Step 6: Use the model fusion formula (15) to map the input variables of the new industrial process training set to the operating range of the similar old industrial process input variables, and the transformed input data of the new industrial process training set is recorded as

The fusion output of these N prediction models is obtained through the Bayesian model averaging algorithm

和新工业过程输入数据

作为多模型迁移策略的输入数据，利用最小二乘支持向量机(Least Squares Support VectorMachine，LSSVM)算法训练新工业过程模型，获得新工业过程模型输出

完成新工业过程建模；Step 7: Fusion output from the old industrial process SVM model

and new industrial process input data

As the input data of the multi-model migration strategy, the Least Squares Support Vector Machine (LSSVM) algorithm is used to train the new industrial process model, and the output of the new industrial process model is obtained.

Complete new industrial process modeling;

Step 8: Model verification, using the root mean square error (Root MeanSquare Error, RMSE) and the coefficient of determination (R-Square, R ² ) to evaluate the validity of the SVM model according to formula (16) and formula (17) respectively, if the step 7 The prediction accuracy of the obtained model on the test data set meets the threshold set by the experiment, and the modeling process is completed; otherwise, repeat steps 3 to 7, and add a new N group of old industrial process data samples containing new industrial process information to the In the mixed sample, continue to train the new industrial process model until the experimental stop condition is met;

是预测输出的均值，Y_i是新工业过程的真实输出。where N is the number of test data, _yi is the output of the prediction model,

is the mean of the predicted output and _Yi is the true output of the new industrial process.

This method first uses the Latin hypercube method to collect a small sample data set for nonlinear industrial process modeling, combines multiple similar old process data, and learns the conversion mapping function between the new industrial process data and the old industrial process data through an adversarial transfer algorithm. , so as to convert a small amount of new process data into various types of old industrial process data with new industrial process information; The modeling of the new industrial process establishes the foundation; finally, using a multi-model transfer strategy and the Bayesian model averaging theory, several trained "old industrial process prediction models with new industrial process information" are transferred, combined with a small amount of new Industrial process data, resulting in the final new industrial process performance prediction model. The present invention migrates useful information of a plurality of existing similar old industrial processes to help establish a new industrial process performance prediction model and reduce the cost of new industrial process modeling; Negative transfer" problem, using the old and new process data transfer method based on adversarial transfer learning to improve the effect of transfer modeling. This method effectively solves the problems of high cost and long modeling period of complex industrial process modeling, makes full use of the useful information of existing similar old industrial process models, and solves the problem when the old process data is much more than the new process data in transfer learning. The possible phenomenon of "negative transfer" completes the modeling of new industrial processes, reduces modeling costs, speeds up modeling, and improves modeling accuracy.

In order to verify the effect of the method, experimental data were generated using the laboratory centrifugal compressor mechanism model, and the performance prediction model of the centrifugal compressor was established to verify the validity of the proposed modeling method. By modifying the key geometric parameters of the compressor mechanism model, four different but similar compressor models A, B, C and D are generated for simulation experiments. For the four centrifugal compressors A, B, C, and D, where compressor A is the new compressor to be modeled, generating a small amount of new industrial process modeling data, while the centrifugal compressors B, C, and D are used as the long-running compressor The old compressors generate a large amount of old industrial process modeling data to assist the establishment of new industrial process prediction models. The stable motion range of the old and new compressors is shown in Table 1.

Table 1 Centrifugal compressor A, B, C, D stable operation interval and corresponding One-Hot code

The prediction effect of the established model is compared with the prediction effect of the two sets of comparative experimental models to further demonstrate the superiority of the proposed method. The three comparison methods are as follows:

Method 1: Convert a small amount of new industrial process data into old industrial process data through adversarial transfer learning, and train multiple SVM models after mixing with each group of old industrial process data, and then establish a new compressor prediction model through a multi-model migration strategy, and finally Test model accuracy with new compressor test data. In the analysis of the experimental results, it was recorded as the ATL-BMA method.

Method 2: Use multiple sets of old compressor data to train multiple old compressor SVM models, then use the multi-model migration strategy to combine a small amount of new compressor training data to build a new compressor prediction model, and finally use the new compressor test data to test the model accuracy. In the analysis of the experimental results, it is recorded as the BMA method.

Method 3: Only use a small amount of new compressor training data to build a new compressor SVM model as a new compressor prediction model, and finally use the new compressor test data to test the model accuracy. In the analysis of the experimental results, it is recorded as the SVM method.

Figure 3 shows the predicted values of the models built by the three methods on the compressor A test set. It can be seen from the figure that the predicted value of the model built by the ATL-BMA method has the highest degree of agreement with the test set, which shows that the ATL-BMA method can effectively use the useful information of similar old industrial processes to help the establishment of new industrial process models, and also It is shown that the ATL-BMA method can utilize the information between the old and new industrial processes more effectively than the pure multi-model transfer method.

In order to further compare the accuracy of the three models, Figures 4 and 5 show the RMSE and R ² of the predicted and true values of the three models. It can be seen from the figures that the method proposed in this chapter can make full use of a large number of old industrial Process data and a small amount of new industrial process data can effectively improve the prediction accuracy of the model and reduce the cost of building the model.

It can be seen from the analysis of the appeal that the present invention establishes a performance prediction model for a new industrial process by adopting an adversarial transfer learning method and the Bayesian model averaging theory, and makes full use of the existing performance prediction models of similar old industrial processes in the industry. The new and old industrial process data are migrated, and the support vector machine is used to establish multiple old industrial process prediction models containing new industrial process information. Finally, the Bayesian model average theory is used to train the old industrial process model, thereby accelerating the construction of the new industrial process. The model speed is reduced, the modeling cost is reduced, and the "negative migration" effect caused by the old industrial process data is more than the new industrial process data when the new and old industrial process migration modeling is solved, and the prediction model that meets the accuracy requirements is obtained. It also shows that this method can utilize the information between old and new industrial processes more effectively than the pure multi-model transfer method. Get closer to the actual output, reducing substantial costs for modeling industrial processes.

Claims (1)

1. A low-cost modeling method for a nonlinear industrial process based on ATL-BMA is characterized by comprising the following steps:

step 1: selecting N groups of modeling data similar to the old industrial process, and determining the stable operation range of the input variable according to the information of the actual process to be modeled; simultaneously, selecting a Latin hypercube method for sampling and collecting a target nonlinear industrial process modeling initial data set; the method comprises the following specific steps:

step 1.1: selecting N groups of modeling data of similar old industrial processes and recording the data as

Modeling the ith old industrial process according to formula (1);

wherein X and X are old industrial process input data, y is old industrial process output data, k_iRepresenting the modeling data quantity of the ith old industrial process, wherein n is the input variable dimension of the ith old industrial process, and the input variable dimensions of all the industrial processes are consistent and are n because the new industrial process and the old industrial process have certain similarity;

step 1.2: determining the stable operation range of input variables according to the information of the actual process to be modeled, selecting discrete sparse data distribution points for sampling and collecting new industrial process modeling data, and obtaining the collected new industrial process data D according to the formula (2)_new；

In the formula, l represents the modeling data volume of the new industrial process;

step 2: dividing new industrial process data and old industrial process data into two parts respectively, namely a training data set and a testing data set in a new modeling process and an old modeling process, and respectively carrying out normalization processing on the new industrial process initial data set and the old industrial process modeling data; wherein, for the new industrial process data, the new industrial process data is divided into a new industrial process training data set

And new industrial process test data set

And maps the data to [0,1 ] using equation (3)]An interval;

in the formula, z_iRepresenting the result after normalization of input or output data of an industrial process, x_iIs the data before normalization, x_maxIs the maximum value, x, before data normalization_minIs the minimum value;

and step 3: converting N groups of old industrial process data into N groups of old industrial process data with new industrial process information by using a Cycle GANs-based new and old industrial process data migration algorithm; wherein the old industrial process training data set is

The method comprises the following specific steps:

step 3.1: initializing parameters: g parameter theta_G，D_oParameter omega_oF parameter θ_F，D_nParameter omega_n，n_critic＝5，α＝0.00005、β₁＝0、β₂＝0.7，m＝5，λ＝0.5，Epoch＝20000；

Wherein: g denotes a generator function of old to new industrial process data, D_oA discriminator representing the correspondence of the old industrial process, F a generator function representing the new industrial process to the old industrial process, D_nDiscriminators, n, corresponding to new industrial processes_criticRepresenting the number of times of training the discriminant model after training the generator once, alpha, beta₁And beta₂The parameters of the Adam optimizer are shown, m is the sampling number, and Epoch is the number of times of model cyclic training;

step 3.2: will be derived from the ith old industrial process data by the generator G

M samples collected in

Is converted into m new industrial process data,is marked as X_o→n＝F(X_o) (ii) a Will be derived from new industrial process data by the generator F

M samples collected in

Converted into m old industrial process data, denoted X_n→o＝F(X_n)；

Step 3.3: obtaining the loss of the discriminator and the consistent loss of the two forward circulations according to a formula (4) and a formula (5);

step 3.4: updating the discriminator D by the formula (6) and the formula (7)_oParameter omega_oAnd D_nParameter omega_n；

Step 3.5: repeating the step 3.2 to the step 3.4n_criticSecondly;

step 3.6: repeating the step 3.2;

step 3.7: calculating two forward cyclic consistent losses through formula (8) and formula (9);

step 3.8: calculating two generator losses by equation (10) and equation (11);

step 3.9: updating generator G parameter θ by equation (12) and equation (13)_GAnd F parameter theta_F；

Step 3.10: repeating the steps 3.6 to 3.9Epoch times, converting the new industrial process data into the jth old industrial process data by using the trained F, and recording the jth old industrial process data as

Step 3.11: repeating the steps 3.1-3.9 by using each group of old industrial process data, transferring the new industrial process data into the old industrial process domain, obtaining N groups of old industrial process data with new industrial process information by the new industrial process data through antagonistic transfer learning, and recording the N groups of old industrial process data with the new industrial process information as

And 4, step 4: mixing the old industrial process data with the new industrial process information in the step 3 with the corresponding old industrial process data to obtain N groups of mixed data sets;

and 5: splitting a hybrid data set into a hybrid training set

And mixing the test data sets

At the same time, N old industrial process training data sets are combined

And a new industrial process prediction model y ═ f (x), training a Support Vector Machine (SVM) model respectively by utilizing N groups of mixed data sets to obtain N old industrial process basic models with new industrial process information, and recording the N old industrial process basic models as f₁(·)-f_N(·); wherein,

k_trainis the size of the training data set and,

k_testis the test data set size, any ith old industrial process,

n_iis the ith old industrial process training set size; the method comprises the following specific steps:

step 5.1: initializing parameters;

step 5.2: new industrial process data are converted into N groups of old industrial process data carrying new industrial process information through a Cycle GANs-based new and old industrial process data migration algorithm

Mixing D according to formula (14)_n→oAnd D_oObtaining N groups of basic model training data D_Basic；

Step 5.3: by using D_BasicTraining N SVM to obtain N old industrial process basic models with new industrial process information, and recording as f₁(·)-f_N(·)；

Step 6: mapping input variables of the new industrial process training set to the operation interval of the input variables of the similar old industrial process through a model fusion formula (15), and recording input data of the converted new industrial process training set as

Obtaining the fusion output of the N prediction models by the Bayesian model average algorithm

And 7: fusing and outputting an old industrial process SVM model

And new industrial process input data

As input data of a multi-model migration strategy, training a new industrial process model by utilizing a least square support vector machine algorithm to obtain output of the new industrial process model

Completing new industrial process modeling；

And 8: model verification, namely evaluating the effectiveness of the SVM model by utilizing the root mean square error and the determination coefficient according to a formula (16) and a formula (17), and finishing the modeling process if the prediction precision of the model obtained in the step 7 on the test data set meets an experiment set threshold; otherwise, repeating the step 3 to the step 7, adding N new groups of old industrial process data samples containing new industrial process information into the mixed sample, and continuing training the new industrial process model until the experiment stopping condition is met;

where N is the number of test data, y_iIs the output of the predictive model and is,

is the mean of the predicted outputs, Y_iIs the real output of the new industrial process.

Images