CN-121997218-A - Yield interval prediction method based on working condition transfer learning

CN121997218ACN 121997218 ACN121997218 ACN 121997218ACN-121997218-A

Abstract

The invention relates to the field of industrial data processing, and discloses a yield interval prediction method based on working condition transfer learning, which comprises the steps of obtaining historical production time sequence data with working condition labels, taking normal working condition data as a source domain and abnormal working condition data as a target domain, and training a base model of yield interval prediction based on the source domain; and in the prediction stage, the gating time sequence attention fusion network is utilized to sense the real-time working condition, and if the working condition is abnormal, the target domain model is called to predict the yield interval. According to the invention, through information granulation processing of time sequence data, design of a composite loss function and improvement of a network structure, accuracy and stability of yield interval prediction under abnormal working conditions are effectively improved, the problem of weak generalization capability of a traditional method under a rare sample working condition is solved, and reliable support is provided for intelligent monitoring and optimization of a complex industrial process.

Inventors

DU SHENG
MA XIAN

Assignees

中国地质大学（武汉）

Dates

Publication Date: 20260508
Application Date: 20251215

Claims (9)

1. A yield interval prediction method based on working condition transfer learning is characterized by comprising the following steps: S1, acquiring historical production time sequence data with a working condition label, taking normal working condition data as source domain data and abnormal working condition data as target domain data, and training a source domain base model of yield interval prediction based on the source domain data; S2, transferring and self-adapting knowledge of a source domain base model to target domain data to obtain a target domain yield interval prediction model; S3, in a prediction stage, sensing the working condition type of a sample to be predicted by using time sequence data of a production field sensor; And S4, according to the working condition sensing result of the sample to be predicted, if the sample to be predicted is an abnormal working condition, a target domain interval prediction model obtained through transfer learning is called to predict a yield interval.
2. The method for predicting the yield interval based on the working condition transfer learning according to claim 1, wherein the step S1 is specifically as follows: S11, carrying out information granulation processing on historical production time sequence data with working condition labels according to fixed sampling windows, constructing the lower limit and the upper limit of information particles in each window, and forming interval information particles; S12, dividing a source domain and a target domain data set according to the working condition label, wherein normal working condition data are the source domain, and over-burning and under-burning working condition data are the target domain; s13, respectively carrying out normalization processing on the source domain data and the target domain data; s14, constructing a yield interval prediction base model based on a gating time sequence attention fusion network by using source domain data, inputting the model into a historical multi-granularity information grain sequence, outputting the model into a lower limit or an upper limit of the next grain yield, and storing model parameters.
3. The method for predicting the yield interval based on the working condition transfer learning according to claim 1, wherein the step S2 is specifically as follows: S21, taking the parameters of the source domain base model as the initialization parameters of the target domain model; S22, performing fine tuning training on the model by adopting supervised learning on a target domain sample, and optimizing by using a composite loss function; s23, saving the trained target domain model and a normalizer corresponding to the target domain model.
4. The method for predicting a yield interval based on condition transition learning according to claim 1, wherein in step S22, the composite loss function is formed by weighting interval coverage loss and interval width loss, and the expression is: Wherein: And The lower and upper limits of the prediction interval, And Is an adjustable weight coefficient, and N is the total number of samples.
5. The method for predicting a yield interval based on working condition transfer learning according to claim 1, wherein step S3 specifically includes: s31, constructing a gating time sequence attention fusion network as a working condition sensing model, wherein the network comprises a time sequence feature extraction layer, a time sequence attention fusion layer and a working condition classification layer; S32, in a prediction stage, inputting sensor time sequence data into the working condition sensing model, and outputting the class probability distribution of future working conditions; And S33, taking the class with the highest probability as a working condition sensing result.
6. The method for predicting a yield interval based on condition transition learning of claim 5, wherein the structure of the gating timing attention fusion network comprises: The time sequence feature extraction layer is used for extracting multi-scale local time sequence features by adopting a plurality of time sequence convolution modules with different convolution kernel sizes in parallel; The time sequence attention fusion layer is used for receiving the multi-scale time sequence characteristics, calculating the characteristic weights of all time steps through a gating attention mechanism and carrying out dynamic weighted fusion; the attention weight is controlled by the gate control mechanism through a learnable gate control unit, and the expression is as follows: Wherein the method comprises the steps of As a k-th timing characteristic of the signal, In order to be able to train the gating weight matrix, To activate the function, output Fusion weight vectors for the features; And the working condition classification layer inputs the fused characteristics into a fully-connected network and outputs probability distribution of working condition categories.
7. The method for predicting the yield interval based on the working condition transfer learning of claim 1, wherein step S4 specifically comprises: s41, if the working condition sensing result is an abnormal working condition, calling a target domain interval prediction model and a corresponding normalizer thereof; s42, carrying out normalization processing on the input information grain sequence, and respectively inputting a lower limit prediction model and an upper limit prediction model of a target domain to obtain two endpoint prediction values of a next grain yield interval; s43, taking the minimum value as the lower limit and the maximum value as the upper limit for the two endpoints to form a yield prediction interval.
8. A computer device is characterized by at least comprising one or more processors and a memory storing one or more computer programs, wherein the processors call the computer programs to realize the steps of the yield interval prediction method based on the working condition migration learning according to any one of claims 1-7.
9. A computer storage device storing a computer program which is called by a processor to implement the steps of the yield interval prediction method based on the condition shift learning according to any one of claims 1 to 7.

Description

Yield interval prediction method based on working condition transfer learning Technical Field The invention belongs to the field of industrial data processing, and particularly relates to a yield interval prediction method based on working condition transfer learning. Background The yield is one of the core indexes for measuring the production process stability, economy and product quality level of the iron ore sintering process. The change trend of the yield is predicted in advance, so that an operator can be guided to adjust the process parameters in time, and an important basis can be provided for equipment operation and batching scheduling. The sintering process has the characteristics of strong coupling, strong nonlinearity, multiple working conditions and the like, and the data distribution difference between different working conditions is obvious. Under normal conditions, a stable production environment can be formed, the obtained sinter yield is high, and the yield fluctuation under abnormal conditions is strong and low. In addition, the number of samples under the abnormal working condition is limited, and the problems of insufficient model training, weak generalization capability and the like easily occur by directly aiming at independent modeling under the abnormal working condition, so that the prediction deviation is larger. The transfer learning provides an effective technical approach for solving the problem, and model knowledge obtained through training under a normal working condition is transferred to an abnormal working condition, so that the prediction capability of the model under the abnormal working condition can be improved while the training data requirement is reduced. Therefore, under the complex sintering environment with changeable working conditions, the existing yield prediction method is difficult to meet the production requirements in terms of adaptability and prediction accuracy under abnormal working conditions. Therefore, the yield interval prediction method based on the working condition transfer learning has important significance for improving the quality and the yield of the sinter. Disclosure of Invention The invention aims to solve the technical problem that the existing yield prediction method is difficult to meet production requirements in terms of adaptability and prediction accuracy under abnormal working conditions. In order to solve the technical problems, the invention provides a yield interval prediction method based on working condition transfer learning. The invention provides a yield interval prediction method based on working condition transfer learning, which specifically comprises the following steps: S1, acquiring historical production time sequence data with a working condition label, taking normal working condition data as source domain data and abnormal working condition data as target domain data, and training a source domain base model of yield interval prediction based on the source domain data; S2, transferring and self-adapting knowledge of a source domain base model to target domain data to obtain a target domain yield interval prediction model; S3, in a prediction stage, sensing the working condition type of a sample to be predicted by using time sequence data of a production field sensor; And S4, according to the working condition sensing result of the sample to be predicted, if the sample to be predicted is an abnormal working condition, a target domain interval prediction model obtained through transfer learning is called to predict a yield interval. The computer equipment at least comprises one or more processors and a memory storing one or more computer programs, wherein the processors call the computer programs to realize the step of the yield interval prediction method based on the working condition migration learning. A computer storage device stores a computer program that is invoked by a processor to implement the steps of the yield interval prediction method based on operating mode shift learning. The technical scheme provided by the invention has the beneficial effects that the accuracy and the stability of the yield interval prediction under abnormal working conditions are effectively improved through the information granulation processing of time sequence data, the design of the composite loss function and the improvement of the network structure, the problem of weak generalization capability of the traditional method under the rare working conditions of samples is solved, and the reliable support is provided for intelligent monitoring and optimization of complex industrial processes. Drawings The invention will be further described with reference to the accompanying drawings and examples, in which: FIG. 1 is a general flow diagram of a yield interval prediction method based on working condition transfer learning; FIG. 2 is a yield interval prediction result under abnormal conditions. Detailed Description For the purpose of ma