CN-122020047-A - Industrial process soft measurement method based on dynamic tuning attention neural network

CN122020047ACN 122020047 ACN122020047 ACN 122020047ACN-122020047-A

Abstract

The application relates to the technical field of measurement, in particular to an industrial process soft measurement method based on a dynamic tuning attention neural network, which comprises the steps of constructing a dynamic tuning attention back propagation neural network model based on a preset dynamic discarding layer, a preset dynamic residual error connection module and a preset attention mechanism module, training the dynamic tuning attention back propagation neural network model according to training data until a preset iteration stop condition is reached, constructing a prediction model, inputting input data of an industrial process to be measured into the prediction model, and outputting a yield predicted value and/or a quality index predicted value of an industrial product. Therefore, the problems that the stability and reliability of the model in a complex industrial scene are affected due to the fact that the representation capability of the model to dynamic characteristics and time-varying rules is insufficient due to the fact that complex dependency relations between input variables and complex relations between the input variables and a plurality of output variables are difficult to capture in a related technology are solved.

Inventors

GENG ZHIQIANG
WANG XINTIAN
HAN YONGMING
LIU MIN
HU XUAN

Assignees

北京化工大学

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (10)

1. An industrial process soft measurement method based on a dynamic tuning attention neural network is characterized by comprising the following steps: Collecting training data in an industrial process; based on a preset dynamic discarding layer, a preset dynamic residual error connection module and a preset attention mechanism module, constructing a dynamic tuning attention back propagation neural network model; and training the dynamic tuning attention back propagation neural network model according to the training data until reaching a preset iteration stop condition so as to construct a prediction model, inputting input data of the industrial process to be measured into the prediction model, and outputting a yield predicted value and/or a quality index predicted value of the industrial product.
2. The method of claim 1, wherein the constructing a dynamic tuning attention back propagation neural network model based on a preset dynamic discard layer, a preset dynamic residual connection module, and a preset attention mechanism module comprises: calculating discarding probability according to training turns based on the preset dynamic discarding layer; calculating residual weights based on the preset dynamic residual connection module; based on the preset attention mechanism module, respectively carrying out linear transformation on the feature matrix by utilizing a query matrix, a key matrix and a value matrix to calculate the attention score of at least one feature, and obtaining attention output according to the attention score; And constructing the dynamic tuning attention back propagation neural network model based on the discarding probability, the residual error weight and the attention output.
3. The method of claim 2, wherein the calculation formula of the discard probability is: , Wherein, the And the discarding probability is used for determining whether the current training round is the current training round, and the total_ epochs is the total training round.
4. The method of claim 2, wherein the residual weights are calculated as: , Wherein, the The residual weights are; Total_ epochs is the total training round.
5. The method of claim 2, wherein the attention score is calculated by the formula: , Wherein, the Is the number of input features; Representing the feature matrix; 、 representing the query matrix and the key matrix, respectively.
6. An industrial process soft measurement device based on a dynamic tuning attention neural network, comprising: the acquisition module is used for acquiring training data in an industrial process; The construction module is used for constructing a dynamic tuning attention back propagation neural network model based on a preset dynamic discarding layer, a preset dynamic residual error connection module and a preset attention mechanism module; And the measurement module is used for training the dynamic tuning attention back propagation neural network model according to the training data until reaching a preset iteration stop condition so as to construct a prediction model, input data of the industrial process to be measured into the prediction model, and output a yield predicted value and/or a quality index predicted value of the industrial product.
7. The apparatus of claim 6, wherein the build module comprises: the first calculation unit is used for calculating the discarding probability according to training rounds based on the preset dynamic discarding layer; The second calculation unit is used for calculating residual weights based on the preset dynamic residual connection module; The third calculation unit is used for respectively carrying out linear transformation on the feature matrix by utilizing the query matrix, the key matrix and the value matrix based on the preset attention mechanism module so as to calculate the attention score of at least one feature and obtain the attention output according to the attention score; And the construction unit is used for constructing the dynamic tuning attention back propagation neural network model based on the discarding probability, the residual error weight and the attention output.
8. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor executing the program to implement the dynamic dominant eye neural network based industrial process soft measurement method of any one of claims 1-5.
9. A computer readable storage medium having stored thereon a computer program, the program being executable by a processor for implementing the dynamic tuning attention neural network based industrial process soft measurement method according to any one of claims 1-5.
10. A computer program product comprising a computer program, characterized in that the computer program is executed for implementing the industrial process soft measurement method based on a dynamically tuned attention neural network according to any of claims 1-5.

Description

Industrial process soft measurement method based on dynamic tuning attention neural network Technical Field The application relates to the technical field of measurement, in particular to an industrial process soft measurement method based on a dynamic tuning attention neural network. Background In the related art, a BPNN (Backpropagation Neural Network ) is a typical feedforward neural network, and the nonlinear mapping relationship between input and output can be realized by continuously adjusting the network weight and bias through an error back propagation algorithm, so that the method is widely applied to soft measurement modeling of an industrial process. However, when the related technology has the characteristics of strong nonlinearity, dynamic property, noise interference and the like in the face of an industrial process, complex dependency relationship between input variables and complex association between the input variables and a plurality of output variables are difficult to capture effectively, and the characterization capability of a model on dynamic property and time-varying rule is insufficient easily, so that the stability and reliability of the model in a complex industrial scene are affected, and the problem is to be solved. Disclosure of Invention The application provides an industrial process soft measurement method based on a dynamic tuning attention neural network, which aims to solve the problems that in the related art, the characterization capability of a model on dynamic characteristics and time-varying rules is insufficient easily caused by difficult effective capture of complex dependency relations between input variables and complex association between the input variables and a plurality of output variables, so that the stability and reliability of the model in a complex industrial scene are affected. An embodiment of the first aspect of the application provides an industrial process soft measurement method based on a dynamic tuning attention neural network, which comprises the following steps of collecting training data in an industrial process, constructing a dynamic tuning attention back propagation neural network model based on a preset dynamic discarding layer, a preset dynamic residual error connection module and a preset attention mechanism module, training the dynamic tuning attention back propagation neural network model according to the training data until a preset iteration stop condition is reached, constructing a prediction model, inputting input data of the industrial process to be measured into the prediction model, and outputting a yield predicted value and/or a quality index predicted value of an industrial product. Through the technical means, the embodiment of the application can construct the dynamic tuning attention back propagation neural network model according to the preset dynamic discarding layer, the preset dynamic residual error connecting module and the preset attention mechanism module, further carry out industrial process soft measurement, improve the nonlinear modeling capacity and dynamic response characteristic of the industrial process soft measurement model, and enhance the stability and robustness of the industrial process soft measurement model under complex and time-varying working conditions, thereby more accurately predicting the yield and quality index of industrial products and meeting the application requirements of modern industrial processes on the soft measurement technology in the aspects of high precision and high reliability. Optionally, in one embodiment of the present application, the building a dynamic tuning attention back propagation neural network model based on a preset dynamic discarding layer, a preset dynamic residual error connection module and a preset attention mechanism module includes calculating a discarding probability according to training rounds based on the preset dynamic discarding layer, calculating residual error weights based on the preset dynamic residual error connection module, performing linear transformation on a feature matrix based on the preset attention mechanism module by using a query matrix, a key matrix and a value matrix to calculate an attention score of at least one feature and obtain an attention output according to the attention score, and building the dynamic tuning attention back propagation neural network model based on the discarding probability, the residual error weights and the attention output. Through the technical means, the embodiment of the application can construct the dynamic tuning attention back propagation neural network model based on the discarding probability, the residual error weight and the attention output, and can realize the self-adaptive feature extraction and nonlinear modeling of the multivariable features in the industrial process, thereby carrying out high-precision prediction on the yield and quality indexes of industrial products. Optionally, in an embo