CN-121983189-A - Multi-target evolution-based steel material temperature field time sequence prediction method and system

CN121983189ACN 121983189 ACN121983189 ACN 121983189ACN-121983189-A

Abstract

The invention discloses a time sequence prediction method and a time sequence prediction system for a temperature field of a steel material based on multi-objective evolution, and relates to the field of analysis of laser cladding temperature fields of the steel material; the method comprises the steps of obtaining a data set, dividing a data set into a plurality of parameters, processing historical temperature field data according to the data set dividing strategy to obtain a training set and a verification set, training and parameter adjustment are carried out on a deep learning model according to the data set dividing strategy to determine a basic value interval corresponding to each parameter of the model, determining target parameter values corresponding to each parameter by combining a preset multi-target evolution optimization strategy, retraining the deep learning model to obtain a target deep learning model, and determining predicted temperature field data corresponding to current technological parameters in future time steps based on the target deep learning model. According to the scheme, a deep learning model is introduced into substrate laser cladding temperature field data prediction, so that the calculation time is greatly shortened while the prediction precision is maintained, and the overall calculation efficiency is remarkably improved.

Inventors

LIU ZHANQI
LIU CHANG
TANG LIXIN

Assignees

东北大学

Dates

Publication Date: 20260505
Application Date: 20251212

Claims (9)

1. A time sequence prediction method for a temperature field of a steel material based on multi-objective evolution is characterized by comprising the following steps: acquiring historical temperature field data determined by means of a finite element simulation strategy when the substrate and the coating are subjected to simulation under the current process parameters; Processing the historical temperature field data according to a data set partitioning strategy to obtain a training set for training a deep learning model and a verification set for optimizing parameters and performance evaluation of the deep learning model; training and parameter adjustment are carried out on the deep learning model based on the training set and the verification set so as to determine a basic value interval corresponding to each parameter of the deep learning model; determining target parameter values corresponding to the parameters according to a preset multi-target evolution optimization strategy and a basic value interval corresponding to the parameters; Retraining the deep learning model based on the target parameter values to obtain a target deep learning model, and determining predicted temperature field data corresponding to the current process parameter at a future time step based on the target deep learning model.
2. The multi-objective evolution-based steel material temperature field time sequence prediction method according to claim 1, wherein the processing of the historical temperature field data according to a data set partitioning strategy to obtain a training set for training a deep learning model, a validation set for tuning parameters and performance evaluation of the deep learning model comprises: determining an input-output time window with a fixed length, wherein an input part under the input-output time window comprises A time steps and an output part comprises B time steps, A is an integer greater than 1, and B is an integer not less than 1; Processing the historical temperature field data by using the input-output time window as a interception benchmark and utilizing a sliding window method to construct a plurality of groups of input-output mapping pairs; And determining a training set for training the deep learning model and a verification set for optimizing parameters and performance evaluation of the deep learning model according to a plurality of groups of the input-output mapping pairs.
3. The multi-objective evolutionary-based steel material temperature field time series prediction method according to claim 2, wherein determining a training set for training a deep learning model, a validation set for tuning parameters and performance assessment of the deep learning model according to a plurality of sets of the input-output mapping pairs comprises: Dividing the obtained multiple groups of input-output mapping pairs based on a preset proportion to obtain a training set, a verification set and a test set for checking generalization capability of the deep learning model; after retraining the deep learning model based on the target parameter values to obtain a target deep learning model, further comprising: extracting the test set to determine a corresponding test subset; And comparing the input-output mapping pair in the test subset with the output item in the corresponding time step and the predicted item in the corresponding time step, which is predicted by the target deep learning model, so as to obtain a comparison result for representing the generalization capability of the target deep learning model.
4. The method for predicting the temperature field time sequence of a steel and iron material based on multi-objective evolution according to claim 1, wherein before processing the historical temperature field data according to a data set partitioning strategy, further comprising: and carrying out normalization processing on the historical temperature field data to obtain normalized historical temperature field data.
5. The multi-objective evolutionary-based steel material temperature field time series prediction method according to claim 1, further comprising, after determining predicted temperature field data corresponding to the current process parameters at a future time step based on the objective deep learning model: Judging whether the predicted temperature field data is larger than a preset temperature threshold value or not; If yes, the current technological parameters are adjusted.
6. The multi-objective evolution-based steel material temperature field time sequence prediction method according to claim 1, wherein the deep learning model is a long-term and short-term memory network.
7. The method for predicting the temperature field time sequence of the steel material based on the multi-objective evolution according to any one of claims 1 to 6, wherein determining the objective parameter value corresponding to each of the parameters according to a preset multi-objective evolution optimization strategy and a basic value interval corresponding to each of the parameters comprises: Taking the maximized prediction precision and the minimized model complexity as a double objective function, taking a basic value interval corresponding to each parameter as a search interval, and performing iterative optimization by utilizing a multi-objective evolutionary optimization algorithm to obtain a final population when iteration is terminated, wherein the final population comprises a plurality of individuals, and the individuals are parameter combinations formed by parameter values corresponding to each parameter; and determining a pareto front solution set based on the final population, and further determining a pareto optimal parameter combination from the pareto front solution set according to a weighted sum method and the double objective function, wherein the pareto optimal parameter combination is a parameter combination formed by target parameter values corresponding to the parameters.
8. The method for predicting the temperature field time sequence of the steel material based on the multi-objective evolution according to claim 7, wherein the multi-objective evolutionary optimization algorithm is a non-dominant ordering genetic algorithm.
9. A multi-objective evolution based steel material temperature field time sequence prediction system, comprising: The acquisition module is used for acquiring historical temperature field data determined by means of a finite element simulation strategy when the substrate and the coating are subjected to simulation under the current process parameters; The data set dividing module is used for processing the historical temperature field data according to a data set dividing strategy to obtain a training set for training a deep learning model and a verification set for optimizing parameters and performance evaluation of the deep learning model; The training module is used for training and adjusting parameters of the deep learning model based on the training set and the verification set so as to determine a basic value interval corresponding to each parameter of the deep learning model; the parameter optimization module is used for determining target parameter values corresponding to the parameters according to a preset multi-target evolution optimization strategy and a basic value interval corresponding to the parameters; And the retraining module is used for retraining the deep learning model based on the target parameter value to obtain a target deep learning model, and further determining predicted temperature field data corresponding to the current process parameter in a future time step based on the target deep learning model.

Description

Multi-target evolution-based steel material temperature field time sequence prediction method and system Technical Field The invention relates to the technical field of steel material laser cladding temperature field analysis, in particular to a multi-target evolution-based steel material temperature field time sequence prediction method and system. Background The laser cladding is an advanced surface modification and repair technology, and the core principle is that the metal powder preset on the surface of the substrate is instantaneously melted by utilizing the focusing energy of a high-energy laser beam, then is quickly solidified in a room temperature environment, and finally forms a coating with excellent performance and used for functional protection on the surface layer of the substrate. In the laser cladding process, the setting of the technological parameters plays a decisive role in the final technological effect, the temperature field distribution difference of the cladding area under different technological parameters is obvious, and the dynamic change of the temperature can directly influence the density, the bonding strength, the microstructure and other core quality indexes of the cladding layer, so that the cladding quality and the effect are determined. Considering that the substrate such as a roller has the characteristics of large volume and large weight, the experiment is liable to waste materials and the trial-and-error cost is relatively high, the method mainly adopts a finite element simulation strategy to carry out pre-evaluation before actual repair, namely, a simulation model is established based on finite element software to simulate and determine the temperature field data of the laser cladding substrate under different process parameters to obtain the temperature field distribution condition, the laser cladding temperature field is a heat accumulation process, the temperature field data has time sequence, the transient solution of the temperature field data under each time step is very time-consuming if the method is adopted, the optimization efficiency is very low, the time consumption is relatively long especially when the geometrical size of the substrate is relatively large, the nonlinear relation between the process parameters and the temperature field is difficult to efficiently process, and the dynamic prediction and optimization cannot be realized. Disclosure of Invention In view of the above, the invention provides a time sequence prediction method and a time sequence prediction system for a steel material temperature field based on multi-objective evolution, which introduce a deep learning model into steel material laser cladding temperature field data prediction, and the calculation time is greatly shortened while the prediction precision is maintained, so that the overall calculation efficiency is remarkably improved. In order to solve the technical problems, the application provides a steel material temperature field time sequence prediction method based on multi-objective evolution, which comprises the following steps: acquiring historical temperature field data determined by means of a finite element simulation strategy when the substrate and the coating are subjected to simulation under the current process parameters; Processing the historical temperature field data according to a data set partitioning strategy to obtain a training set for training a deep learning model and a verification set for optimizing parameters and performance evaluation of the deep learning model; training and parameter adjustment are carried out on the deep learning model based on the training set and the verification set so as to determine a basic value interval corresponding to each parameter of the deep learning model; determining target parameter values corresponding to the parameters according to a preset multi-target evolution optimization strategy and a basic value interval corresponding to the parameters; Retraining the deep learning model based on the target parameter values to obtain a target deep learning model, and determining predicted temperature field data corresponding to the current process parameter at a future time step based on the target deep learning model. Further, processing the historical temperature field data according to a data set partitioning strategy to obtain a training set for training a deep learning model, a verification set for tuning parameters and performance evaluation of the deep learning model, including: determining an input-output time window with a fixed length, wherein an input part under the input-output time window comprises A time steps and an output part comprises B time steps, A is an integer greater than 1, and B is an integer not less than 1; Processing the historical temperature field data by using the input-output time window as a interception benchmark and utilizing a sliding window method to construct a plurality of groups of input