Steel material temperature field time sequence prediction method and system based on multi-objective evolution

Abstract

The invention discloses a steel material temperature field time sequence prediction method and system based on multi-objective evolution, and relates to the field of steel material laser cladding temperature field analysis, and the scheme comprises the steps: obtaining historical temperature field data determined through a finite element simulation strategy; processing the historical temperature field data according to a data set division strategy to obtain a training set and a verification set; carrying out training and parameter adjustment on the deep learning model so as to determine a basic value interval corresponding to each parameter of the model; and determining a target parameter value corresponding to each parameter in combination with a preset multi-target evolutionary optimization strategy so as to retrain the deep learning model to obtain a target deep learning model, and determining predicted temperature field data corresponding to the current process parameter in a future time step based on the target deep learning model. According to the scheme, the deep learning model is introduced into base material laser cladding temperature field data prediction, the prediction precision is kept, meanwhile, the calculation time is greatly shortened, and the overall calculation efficiency is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of laser cladding temperature field analysis technology for steel materials, and in particular to a method and system for time-series prediction of temperature field of steel materials based on multi-objective evolution. Background Technology

[0002] Laser cladding is an advanced surface modification and repair technology. Its core principle is to use the focused energy of a high-energy laser beam to instantly melt metal powder pre-placed on the substrate surface, which then rapidly solidifies at room temperature, ultimately forming a high-performance coating for functional protection on the substrate surface. During laser cladding, the setting of process parameters plays a decisive role in the final process effect. Different process parameters result in significant differences in the temperature field distribution of the cladding area. Dynamic temperature changes directly affect the core quality indicators of the cladding layer, such as density, bonding strength, and microstructure, thus determining the quality and effect of the cladding. Considering the large volume and weight of substrates such as rolling mill rolls, direct experimentation would inevitably waste materials and incur significant trial-and-error costs. Therefore, the current approach primarily involves pre-evaluation using finite element simulation before actual repair. This involves establishing a simulation model based on finite element software to simulate and determine the temperature field data of the laser cladding substrate under different process parameters, thereby obtaining the temperature field distribution. Laser cladding temperature field is a thermal accumulation process, and the temperature field data is time-series. If the transient solution of the temperature field data at each time step is performed using the above method, it would be extremely time-consuming and have very low optimization efficiency. This is especially true when the substrate geometry is large, where the time consumption would be even longer. It is also difficult to efficiently handle the nonlinear relationship between process parameters and the temperature field, and dynamic prediction and optimization cannot be achieved. Summary of the Invention

[0003] In view of this, the present invention provides a method and system for time-series prediction of temperature field of steel materials based on multi-objective evolution. It introduces a deep learning model into the prediction of laser cladding temperature field data of steel materials, which significantly shortens the calculation time while maintaining the prediction accuracy and significantly improves the overall calculation efficiency.

[0004] To address the aforementioned technical problems, this application provides a method for time-series prediction of temperature fields in steel materials based on multi-objective evolution, comprising: Obtain historical temperature field data of the substrate and coating under the current process parameters during laser cladding, determined by the finite element simulation strategy; The historical temperature field data is processed according to the dataset partitioning strategy to obtain a training set for training the deep learning model and a validation set for tuning parameters and evaluating the performance of the deep learning model. The deep learning model is trained and its parameters are tuned based on the training set and the validation set to determine the basic value range of each parameter of the deep learning model. Based on the preset multi-objective evolutionary optimization strategy and the basic value range corresponding to each parameter, the target parameter value corresponding to each parameter is determined; The deep learning model is retrained based on the target parameter values ​​to obtain a target deep learning model, and then the predicted temperature field data corresponding to the current process parameters at future time steps is determined based on the target deep learning model.

[0005] Furthermore, the historical temperature field data is processed according to a dataset partitioning strategy to obtain a training set for training the deep learning model and a validation set for tuning parameters and evaluating the performance of the deep learning model, including: A fixed-length input-output time window is defined, wherein the input portion of the input-output time window includes A time steps and the output portion includes B time steps, where A is an integer greater than 1 and B is an integer not less than 1. Using the input-output time window as the truncation benchmark, the historical temperature field data is processed using the sliding window method to construct multiple sets of input-output mapping pairs; Based on multiple sets of input-output mapping pairs, a training set for training the deep learning model and a validation set for tuning parameters and evaluating the performance of the deep learning model are determined.

[0006] Furthermore, based on multiple sets of input-output mapping pairs, a training set for training the deep learning model and a validation set for tuning parameters and evaluating the performance of the deep learning model are determined, including: Based on a preset ratio, the obtained multiple sets of input-output mapping pairs are divided to obtain a training set, a validation set, and a test set for testing the generalization ability of the deep learning model. After retraining the deep learning model based on the target parameter values ​​to obtain the target deep learning model, the process further includes: The test set is extracted to determine the corresponding test subset; The output terms at the corresponding time step of the input-output mapping in the test subset are compared with the prediction terms at the corresponding time step obtained by the target deep learning model to obtain a comparison result characterizing the generalization ability of the target deep learning model.

[0007] Furthermore, before processing the historical temperature field data according to the dataset partitioning strategy, the process also includes: The historical temperature field data is normalized to obtain normalized historical temperature field data.

[0008] Furthermore, after determining the predicted temperature field data corresponding to the current process parameters at future time steps based on the target deep learning model, the method further includes: Determine whether the predicted temperature field data is greater than a preset temperature threshold; If so, adjust the current process parameters.

[0009] Furthermore, the deep learning model is a long short-term memory network.

[0010] Furthermore, based on the preset multi-objective evolutionary optimization strategy and the basic value range corresponding to each parameter, the target parameter value corresponding to each parameter is determined, including: The dual objective function is to maximize prediction accuracy and minimize model complexity. The search interval is the basic value range corresponding to each parameter. The multi-objective evolutionary optimization algorithm is used for iterative optimization to obtain the final population when the iteration terminates. The final population includes multiple individuals, and each individual is a parameter combination composed of the parameter values ​​corresponding to each parameter. Based on the final population, a Pareto front solution set is determined. Then, according to the weighted sum method and the biobjective function, a Pareto optimal parameter combination is determined from the Pareto front solution set. The Pareto optimal parameter combination is a parameter combination consisting of the objective parameter values ​​corresponding to each parameter.

[0011] Furthermore, the multi-objective evolutionary optimization algorithm is a non-dominated sorting genetic algorithm.

[0012] To address the aforementioned technical problems, this invention also provides a time-series prediction system for the temperature field of steel materials based on multi-objective evolution, comprising: The acquisition module is used to acquire historical temperature field data of the substrate and coating under the current process parameters during laser cladding simulation, which is determined by the finite element simulation strategy. The dataset partitioning module is used to process the historical temperature field data according to the dataset partitioning strategy to obtain a training set for training the deep learning model and a validation set for tuning parameters and evaluating the performance of the deep learning model. The training module is used to train and tune the deep learning model based on the training set and the validation set, so as to determine the basic value range of each parameter of the deep learning model. The parameter optimization module is used to determine the target parameter value corresponding to each parameter based on the preset multi-objective evolutionary optimization strategy and the basic value range corresponding to each parameter. The retraining module is used to retrain the deep learning model based on the target parameter values ​​to obtain a target deep learning model, and then determine the predicted temperature field data corresponding to the current process parameters at future time steps based on the target deep learning model.

[0013] The beneficial effects of this invention are as follows: This application provides a method and system for time-series prediction of temperature field of steel materials based on multi-objective evolution. The scheme includes acquiring historical temperature field data of the substrate and coating under simulated laser cladding under current process parameters, determined by a finite element simulation strategy; processing the historical temperature field data according to a dataset partitioning strategy to obtain a training set for training a deep learning model and a validation set for parameter tuning and performance evaluation of the deep learning model; training and tuning the deep learning model based on the training set and validation set to determine the basic value range of each parameter of the deep learning model; determining the target parameter value of each parameter according to a preset multi-objective evolution optimization strategy and the basic value range of each parameter; retraining the deep learning model based on the target parameter value to obtain a target deep learning model; and then determining the predicted temperature field data corresponding to the current process parameters at future time steps based on the target deep learning model. As can be seen, this scheme utilizes the accuracy of finite element simulation in physical modeling to obtain historical temperature field data. Based on the advantages of deep learning models in time-series data processing, it introduces deep learning models into the prediction of temperature field data for laser cladding of steel materials. Training and parameter tuning are performed using training and validation sets determined from historical temperature field data to obtain the basic value ranges for each parameter. Then, a pre-set multi-objective evolutionary optimization strategy is used to determine the optimal target parameter values ​​within the basic value ranges, thereby determining the target deep learning model. This improves the model's predictive performance, and the target deep learning model enables temperature prediction for future time steps, facilitating subsequent optimization of process parameters. Compared with traditional finite element simulation strategies, this scheme significantly reduces computation time while maintaining prediction accuracy, substantially improving overall computational efficiency and reducing computational resource consumption in engineering applications. It also exhibits strong generalization ability, accurately predicting the dynamic evolution of the temperature field during laser cladding under different process parameters, and effectively capturing subtle temperature fluctuations.

[0014] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application, it can be implemented according to the contents of the specification. In order to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0015] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart of a time-series prediction method for temperature field of steel materials based on multi-objective evolution provided by the present invention; Figure 2 This invention provides a schematic diagram of temperature field data predicted by testing a deep learning model after training and tuning the model using a training set and a validation set. Figure 3 A schematic diagram of the objective function values ​​corresponding to each individual in the final population, each individual in the Pareto front solution set, and the Pareto optimal parameter combination provided by the present invention; Figure 4 A schematic diagram of the predicted temperature field data obtained by testing a target deep learning model provided by the present invention; Figure 5 This is a schematic diagram of the structure of a time-series prediction system for the temperature field of steel materials based on multi-objective evolution, provided by the present invention. Detailed Implementation

[0016] The core of this invention is to provide a method and system for predicting the time series temperature field of steel materials based on multi-objective evolution. By introducing a deep learning model into the prediction of laser cladding temperature field data of steel materials, the calculation time is greatly shortened while maintaining the prediction accuracy, thus significantly improving the overall calculation efficiency.

[0017] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0018] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0019] Please refer to Figure 1 , Figure 1 A flowchart illustrating a time-series prediction method for the temperature field of steel materials based on multi-objective evolution provided by this invention. This time-series prediction method for the temperature field of steel materials based on multi-objective evolution includes: S11: Obtain historical temperature field data of the substrate and coating under the current process parameters during laser cladding, determined by the finite element simulation strategy; S12: Process the historical temperature field data according to the dataset partitioning strategy to obtain the training set for deep learning model training and the validation set for deep learning model parameter tuning and performance evaluation. S13: Train and tune the deep learning model based on the training set and validation set to determine the basic value range of each parameter of the deep learning model. S14: Determine the target parameter value for each parameter based on the preset multi-objective evolutionary optimization strategy and the basic value range of each parameter; S15: Retrain the deep learning model based on the target parameter values ​​to obtain the target deep learning model, and then determine the predicted temperature field data corresponding to the current process parameters at future time steps based on the target deep learning model.

[0020] In this embodiment, considering that traditional finite element simulation strategies are difficult to meet the requirements for efficient solution, this application provides a time-series prediction method for the temperature field of steel materials based on multi-objective evolution. This method fully utilizes historical temperature field data and predicts the temperature field data at future time steps using a target deep learning model. This helps save computational costs, accelerates the calculation process, improves prediction efficiency and accuracy, and facilitates subsequent optimization of process parameters. Specifically, both the substrate and the coating are steel materials, and the substrate includes, but is not limited to, rolling mill rolls. The current process parameters mentioned here can be a single process parameter setting or a combination of several process parameter settings. These process parameters include, but are not limited to, laser power, overlap rate (characterizing the overlap area between two cladding layers), scanning speed, and laser heat source radius.

[0021] Specifically, in step S11, the finite element simulation strategy is used to simulate the temperature field data of the substrate and coating under the current process parameters during laser cladding, at each time step over a period of time (e.g., temperature field data at 54 seconds, a total of 5400 time steps), which is referred to as historical temperature field data. For example, step S11 may include: step S111, establishing a three-dimensional simulation model corresponding to the substrate; specifically, establishing a three-dimensional simulation model of the substrate in three-dimensional modeling software according to the actual size information of the substrate (e.g., the actual diameter of the roll); step S112, importing the three-dimensional simulation model into the finite element software for subsequent mesh generation and physical analysis. In the field analysis, specifically, step S112 may include: step S21, after converting the three-dimensional simulation model into a suitable format, importing it into the finite element software, establishing birth and death elements on the three-dimensional simulation model so that the subsequent process can better simulate the generation of the coating, and meshing the substrate and coating according to the actual application needs to ensure accuracy; step S22, calculating the thermal properties of the material, specifically including selecting 45 Steel as the material of the substrate and T15 powder as the powder. Please refer to Table 1 and Table 2. Table 1 is a schematic table of elemental composition in T15 powder provided by the present invention, and Table 2 is a schematic table of component content in 45 Steel provided by the present invention.

[0022] Table 1

[0023] Table 2

[0024] The J-Matpro software is used to calculate the thermal properties of the material and powder. The temperature range of the thermal properties is determined according to the experimental conditions, and the heating method is also determined according to the experimental conditions. In step S23, the required thermal properties of the substrate and powder, including density, thermal conductivity, coefficient of thermal expansion, specific heat capacity, etc., are exported. In step S24, the exported thermal property data is filtered to select key temperature points that can accurately characterize the trend of thermal properties such as thermal conductivity, specific heat capacity, and coefficient of thermal expansion with temperature. While ensuring accurate characterization of parameter change trends, redundant data is reduced, and the data storage and processing burden is lowered. After the selected key temperature points are arranged in ascending order of temperature, they are entered into the material property library of the finite element software to realize efficient modeling and dynamic calling of the thermal properties of the substrate and cladding layer under different temperature conditions. Through the above settings, the accuracy and feasibility of thermal simulation results can be effectively guaranteed while optimizing finite element calculation resources.

[0025] Step S113 involves selecting and setting up the heat source model. Specifically, common heat source models include the double ellipsoidal heat source model and the Gaussian heat source model. Considering the energy distribution characteristics and computational efficiency requirements during laser cladding, the Gaussian heat source model is selected in this example to simulate the heat conduction process. Subsequently, based on the dimensions of the 3D simulation model and its position in the Cartesian coordinate system, the heat flow source equation is established. The expression of this heat flow source equation is as follows: ; Where Q represents the intensity of the laser point heat source, and P represents the laser power. This represents the radius of the laser point heat source. The radius of the roll model is represented by , and t represents time. This represents the angular velocity of the laser point heat source rotation. This represents the coordinates of the laser point heat source in the Cartesian coordinate system.

[0026] Subsequently, the heat flow source equations were input into APDL (ANSYS Parametric Design). In the Finite Element Parametric Design Language (FEM), export the code and input it into the finite element software; determine the surface to which the code is applied, set the heat convection exchange coefficient and set the initial temperature to room temperature (e.g., 22℃), set the birth and death element state defined in step S21 to dynamically adjust according to the position of the laser heat source, modify the time and step size in the analysis settings, modify the time step size to 0.01 seconds, and the total solution time for the temperature field data is 54 seconds (5400 time steps are calculated in total). Solve the temperature field data under the current process parameters. For example, the current process parameters set at this time are: laser power selected as 2000W, overlap rate selected as 25%, scanning speed selected as 45mm / s, scanning angular velocity selected as 0.6981rad / s, and laser heat source radius selected as 2mm; export the temperature field data after the calculation is completed. Please refer to Table 3. Table 3 is a statistical schematic table of temperature field data determined by the finite element simulation strategy provided by the present invention. The time step is 0.01 seconds. The data in the table records the temperature field data (i.e., temperature value) corresponding to each time step.

[0027] Table 3

[0028] In step S12, the historical temperature field data is processed according to the dataset partitioning strategy. In addition to obtaining the training set and validation set, a test set for testing the generalization ability of the deep learning model can also be obtained. See the following embodiments for details, which will not be repeated here.

[0029] Step S13 determines the basic value range corresponding to each parameter. This basic value range serves as the basis for subsequent parameter optimization. Then, based on step S14, the optimal target parameter value corresponding to each parameter is determined. Based on step S15, the temperature field data at multiple future time steps is predicted, thus obtaining the predicted temperature field data.

[0030] It is understood that, in some embodiments, after determining the predicted temperature field data corresponding to the current process parameters at future time steps based on the target deep learning model, the method further includes: Determine whether the predicted temperature field data is greater than the preset temperature threshold; If so, adjust the current process parameters.

[0031] Specifically, the preset temperature threshold here can be flexibly set according to the actual application requirements. This allows for reliable prediction of temperature field data at future time steps, enabling early prediction of potential over-temperature anomalies and facilitating adjustments to current process parameters. This allows for rapid adjustment and optimization of process parameters. The specific method for adjusting process parameters can be flexibly adjusted according to the actual application situation to avoid future over-temperature anomalies, and no particular restrictions are imposed here.

[0032] It is understandable that the temperature field data will change when the process parameters change. Therefore, by retraining the deep learning model according to the above steps, reliable prediction of the temperature field data corresponding to the new process parameters in future time steps can be achieved.

[0033] In summary, this application provides a time-series prediction method for the temperature field of steel materials based on multi-objective evolution. This scheme utilizes the accuracy of finite element simulation in physical modeling to obtain historical temperature field data. Based on the advantages of deep learning models in time-series data processing, a deep learning model is introduced into the prediction of temperature field data for laser cladding of steel materials. Training and parameter tuning are performed based on training and validation sets determined by historical temperature field data to obtain the basic value ranges for each parameter. Then, the optimal target parameter values ​​under the basic value ranges are determined by a preset multi-objective evolution optimization strategy, thereby determining the target deep learning model, which helps to improve the predictive performance of the model. Furthermore, based on this target deep learning model, rapid and accurate prediction of temperature field data at future time steps is achieved, which is beneficial for subsequent process parameter optimization. Compared with traditional finite element simulation strategies, this scheme significantly reduces the computation time while maintaining prediction accuracy, significantly improves the overall computational efficiency, reduces the consumption of computational resources in engineering applications, has strong generalization ability and strong time-series modeling ability, and can accurately predict the dynamic evolution of the temperature field during laser cladding under different process parameters, which is conducive to keenly capturing subtle temperature fluctuations.

[0034] Based on the above embodiments: In some embodiments, historical temperature field data is processed according to a dataset partitioning strategy to obtain a training set for training a deep learning model and a validation set for tuning parameters and evaluating performance of the deep learning model, including: Define a fixed-length input-output time window. The input portion of the input-output time window includes A time steps, and the output portion includes B time steps. A is an integer greater than 1, and B is an integer not less than 1. Using the input-output time window as the truncation benchmark, the historical temperature field data is processed using the sliding window method to construct multiple sets of input-output mapping pairs; Based on multiple sets of input-output mapping pairs, determine the training set for training the deep learning model and the validation set for tuning parameters and evaluating the performance of the deep learning model.

[0035] It is understandable that, through the setting of the input-output time window and the sliding window method, the input items in the obtained input-output mapping pair include A historical temperature field data corresponding to the current position of the input-output time window, and the output items include B historical temperature field data corresponding to the current position of the input-output time window. For example, taking A=5 and B=1 as an example, using the input-output time window as the interception benchmark, the sliding window method is used to process the continuous historical temperature field data, and multiple sets of input-output mapping pairs can be constructed. For example, the first set of mapping pairs is the temperature field data at time steps 1, 2, 3, 4, and 5 - the temperature field data at time step 6; the second set of mapping pairs is the temperature field data at time steps 2, 3, 4, 5, and 6 - the temperature field data at time step 7; the third set of mapping pairs is the temperature field data at time steps 3, 4, 5, 6, and 7 - the temperature field data at time step 8, and so on.

[0036] As can be seen, the above settings enable this scheme to have excellent time series modeling capabilities.

[0037] More specifically, based on multiple sets of input-output mapping pairs, a training set for training the deep learning model and a validation set for tuning parameters and evaluating performance of the deep learning model are determined, including: Based on a preset ratio, the resulting multiple sets of input-output mapping pairs are divided to obtain a training set, a validation set, and a test set for testing the generalization ability of the deep learning model. After retraining the deep learning model based on the target parameter values ​​to obtain the target deep learning model, the following steps are also included: Extract the test set to determine the corresponding test subset; The output terms at the corresponding time step of the input-output mapping in the test subset are compared with the prediction terms at the corresponding time step obtained by the target deep learning model to obtain the comparison results representing the generalization ability of the target deep learning model.

[0038] Specifically, the preset ratio here refers to the ratio between the number of input-output mapping pairs included in the training set, the number of input-output mapping pairs included in the validation set, and the number of input-output mapping pairs included in the test set. No specific value is specified here.

[0039] The test set includes T input-output mapping pairs, and the test subset includes S input-output mapping pairs, where S and T are both integers greater than 1 and S < T. The step of extracting the test subset is beneficial for selecting representative input-output mapping pairs from the test set (such as mapping pairs corresponding to temperature field data in a typical cladding stage) to obtain the test subset. Then, the target deep learning model is tested based on the test subset to compare the output terms with the predicted terms. Specifically, a comparison graph between the predicted terms and the output terms as true values ​​can be drawn to clearly observe the temperature change trend, quickly verify the target deep learning model's ability to capture key temperature features, and verify the model's generalization ability.

[0040] Furthermore, after step S13, the deep learning model can be saved at this point. Predictions can then be made on the test set using this deep learning model. Please refer to [link / reference]. Figure 2 , Figure 2 This is a schematic diagram of temperature field data predicted by testing a deep learning model after training and tuning the model based on a training set and a validation set, as provided by the present invention. The dashed line represents the true value (i.e., the real temperature field data recorded in the test set), and the solid line represents the predicted temperature field data.

[0041] In some embodiments, before processing the historical temperature field data according to the dataset partitioning strategy, the method further includes: The historical temperature field data is normalized to obtain normalized historical temperature field data.

[0042] Specifically, the historical temperature field data is normalized based on a first preset relation to obtain normalized historical temperature field data. The first preset relation is: ; in, This represents the i-th historical temperature field data. This represents the minimum value among all historical temperature field data. This represents the maximum value among all historical temperature field data. This represents the normalized historical temperature field data for the i-th time.

[0043] As can be seen, the above normalization settings can unify the data distribution, reduce the interference of abnormal data on the deep learning model, and help the deep learning model to more accurately capture the potential change patterns in the temperature field data, thereby improving the prediction accuracy of future temperature field data.

[0044] In some embodiments, the deep learning model is a long short-term memory network.

[0045] Specifically, temperature field data under laser cladding is time-series data. Long Short-Term Memory (LSTM) networks, as a special type of neural network, can effectively capture long-term dependencies in time-series data by introducing a gating mechanism. This avoids the gradient vanishing or exploding problems faced by traditional RNNs (Recurrent Neural Networks) in long-term series learning, and thus can effectively predict temperature field data for future time steps. Of course, this deep learning model can also be a Transformer model, etc., without special limitations.

[0046] For Long Short-Term Memory (LSTM) networks, the parameters that need to be trained are the number of neurons, the number of hidden layers, and the learning rate. For example, step S13 can initially determine the range of parameters by trying different numbers of neurons, learning rates, and hidden layers on the training and validation sets to assess their prediction performance. Subsequently, R... 2The evaluation index is one or a combination of R-squared (coefficient of determination), MAE (mean absolute error), and RMSE (root mean square error). Based on the evaluation results, a reasonable basic value range for each parameter is determined, which provides a basis for parameter optimization in the subsequent step S14. More specifically, in trying different numbers of neurons, learning rates, and hidden layers, the Long Short-Term Memory (LSTM) network is trained for 100 epochs (i.e., traversing the training set 100 times) for each parameter combination (i.e., the MSE loss of the training set and the MSE loss of the validation set are calculated in real time to monitor the LSM network's fitting effect on the training set and its generalization trend on the validation set. After 100 epochs of training, the R², MAE, and RMSE on the validation set are used as quantitative indicators to evaluate the model performance under this parameter combination. For example, if the validation set indicators corresponding to the parameter combination are better (e.g., R² is closer to 1, MAE and RMSE are smaller), it is included in the candidate parameter range. Finally, the minimum and maximum values ​​of the candidate parameter range corresponding to a certain parameter form the basic value range corresponding to that parameter. For example, the basic value range for the number of hidden layers is 1-5, the basic value range for the learning rate is 0.001-0.01, and the basic value range for the number of neurons is 10-100.

[0047] In some embodiments, the target parameter values ​​for each parameter are determined based on a preset multi-objective evolutionary optimization strategy and the basic value ranges corresponding to each parameter, including: S31: The dual objective function is to maximize prediction accuracy and minimize model complexity. The basic value range of each parameter is used as the search range. The multi-objective evolutionary optimization algorithm is used for iterative optimization so as to obtain the final population when the iteration terminates. The final population includes multiple individuals, and each individual is a parameter combination composed of the parameter values ​​corresponding to each parameter. S32: Determine the Pareto front solution set based on the final population, and then determine the Pareto optimal parameter combination from the Pareto front solution set according to the weighted sum method and the biobjective function. The Pareto optimal parameter combination is a parameter combination consisting of the objective parameter values ​​corresponding to each parameter.

[0048] In this embodiment, the above method can balance the prediction accuracy and model complexity of the deep learning model, optimize the target parameter values ​​corresponding to each parameter, and then retrain the target deep learning model to achieve a balance between prediction performance and computational efficiency. It should be noted that model complexity is related to the specific number of parameters, such as the number of neurons. To maximize prediction accuracy, it can be specifically set to minimize the validation set MSE value (a smaller MSE value means higher prediction accuracy). In some embodiments, the multi-objective evolutionary optimization algorithm is a non-dominated sorting genetic algorithm, i.e., the NSGA-II algorithm.

[0049] Taking the NSGA-II algorithm as an example of a multi-objective evolutionary optimization algorithm, further explanation is given for step S31: The dual objective functions are minimizing the validation set MSE value and minimizing the model complexity. The basic value ranges corresponding to each parameter are used as the search range. In step S311, population initialization is performed, i.e., based on the search ranges corresponding to each parameter (such as the search range for the number of hidden layers being 1-5, the search range for the learning rate being 0.001-0.01, and the search range for the number of neurons being 10-100), 50 individuals are randomly generated as the initial population. Each individual corresponds to a parameter combination (composed of the parameter values ​​currently taken by each parameter). In step S312, individual evaluation is performed. For each individual, a corresponding Long Short-Term Memory (LSTM) network is constructed, trained on the training set for a preset number of rounds (e.g., 30 rounds), and then validated on the validation set. The MSE value and model complexity corresponding to that individual are calculated as its fitness. Of course, R can also be calculated. 2The value is used as an auxiliary evaluation indicator and is not specifically limited here; Step S313, perform non-dominated sorting and crowding distance calculation, that is, perform fast non-dominated sorting of individuals in the current population based on two objective functions to divide individuals into different frontiers, and sort individuals within the same frontier according to crowding distance to maintain population diversity; Step S314, perform selection operation, that is, instead of tournament selection, merge the parent and offspring populations, and select individuals from high to low according to non-dominated level (i.e., frontier priority). If the current frontier cannot include all of the next generation, select individuals with better dispersion according to crowding distance, and finally retain the population size of individuals to enter the next generation; Step S315, perform crossover and mutation. The process involves using custom crossover (e.g., randomly swapping parent parameters) and custom mutation (e.g., the number of hidden layers ±1, learning rate plus normal noise, number of neurons ±5) to generate offspring individuals. For example, the crossover probability can be set to 0.9, and the mutation probability can be set to 0.1. Step S316 involves elite retention and population update, which means merging the parent population and the offspring population (a total of 100 individuals) and retaining the population size of individuals through non-dominated sorting and crowding distance to form the next generation population. Step S317 involves terminating the optimization and obtaining the final population if the maximum number of iterations is reached (which can be preset, for example, it can be set to 10). If the maximum number of iterations is not reached, the iterative optimization can continue according to the above steps.

[0050] Step S32 may specifically include: selecting Pareto front solutions (including multiple parameter combinations) that are not dominated by other individuals in the final population; introducing a weighted comprehensive evaluation function according to the weighted sum method, that is, multiplying the two objective functions (minimizing the validation set MSE value and minimizing the model complexity) by a weight and then summing them (the weights corresponding to the first objective function and the second objective function are summed to 1) to merge the two objective functions into a single objective function to calculate the comprehensive score, which facilitates sorting and selection (for example, the weight corresponding to the first objective function can be 0.5 and the weight corresponding to the second objective function can be 0.5); finally, determining the parameter combination with the highest score as the Pareto optimal parameter combination.

[0051] Please refer to Figure 3 , Figure 3 This invention provides a schematic diagram of the objective function values ​​corresponding to each individual in the final population, each individual in the Pareto front solution set, and the Pareto optimal parameter combination. A target deep learning model can be obtained by retraining the deep learning model based on the Pareto optimal parameter combination. The target deep learning model is then tested on a test subset. Please refer to [reference needed]. Figure 4 , Figure 4This diagram illustrates the predicted temperature field data obtained from testing a target deep learning model provided by the present invention. The dashed lines represent the true values ​​(i.e., the actual temperature field data recorded in the test subset), and the solid lines represent the predicted temperature field data. It can be seen that the predicted temperature field data is highly consistent with the true values, which proves the effectiveness of the technical solution in this application. It is evident that this solution provides an efficient and accurate data-driven auxiliary tool for laser cladding process optimization, parameter adjustment, and quality control.

[0052] Furthermore, the target deep learning model and the deep learning model obtained after training and parameter tuning based on the training set and validation set in step S13 are tested on the test set. Please refer to Table 4, which is a test result comparison table provided by the present invention, to prove the effectiveness of the setting of obtaining the target deep learning model by optimizing based on the preset multi-objective evolutionary optimization strategy.

[0053] Table 4

[0054] Furthermore, comparing the time required to calculate the predicted temperature field data for a certain number of future time steps using the target deep learning model with the time required to calculate the same number of time steps using the finite element simulation strategy, the calculation time required using the finite element simulation strategy is 5750 seconds, while the calculation time required using the target deep learning model is 3610 seconds. This demonstrates that the solution provided in this application is beneficial for improving computational efficiency. It is evident that this application can guarantee prediction accuracy and improve computational efficiency.

[0055] Furthermore, different historical temperature field data correspond to different process parameters. Based on this, a corresponding target deep learning model can be obtained according to the method in this application, thereby achieving reliable prediction of temperature field data at future time steps. For example, corresponding to the historical temperature field data under the first process parameter setting (laser power of 2800W, laser heat source radius of 2.5mm, scanning angular velocity of 0.6981rad / s), the target deep learning model is obtained using the scheme in this application, with a corresponding MAE value of 3.4217, RMSE value of 4.5423, and R... 2 The value is 0.9155; corresponding to the historical temperature field data under the second process parameter setting (laser power of 2500W, laser heat source radius of 2.5mm, scanning angular velocity of 0.6981rad / s), the target deep learning model is obtained based on the scheme in this application, with a corresponding MAE value of 3.2271, RMSE value of 4.3138, and R... 2 The value of 0.8351 further confirms the effectiveness of the technical solution in this application in predicting temperature field data under laser cladding.

[0056] Please refer to Figure 5 , Figure 5 This is a schematic diagram of the structure of a time-series prediction system for the temperature field of steel materials based on multi-objective evolution, provided by the present invention.

[0057] This time-series prediction system for the temperature field of steel materials based on multi-objective evolution includes: The acquisition module 21 is used to acquire historical temperature field data of the substrate and coating under the current process parameters during laser cladding, which is determined by the finite element simulation strategy. The dataset partitioning module 22 is used to process historical temperature field data according to the dataset partitioning strategy to obtain a training set for deep learning model training and a validation set for deep learning model parameter tuning and performance evaluation. Training module 23 is used to train and tune the deep learning model based on the training set and validation set in order to determine the basic value range of each parameter of the deep learning model. The parameter optimization module 24 is used to determine the target parameter value of each parameter based on the preset multi-objective evolutionary optimization strategy and the basic value range of each parameter. The retraining module 25 is used to retrain the deep learning model based on the target parameter values ​​to obtain the target deep learning model, and then determine the predicted temperature field data corresponding to the current process parameters at future time steps based on the target deep learning model.

[0058] For an introduction to the time series prediction system for temperature field of steel materials based on multi-objective evolution provided in this application, please refer to the embodiments of the time series prediction method for temperature field of steel materials based on multi-objective evolution described above, which will not be repeated here.

[0059] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section. Relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0060] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A time-series prediction method for the temperature field of steel materials based on multi-objective evolution, characterized in that, include: Obtain historical temperature field data of the substrate and coating under the current process parameters during laser cladding, determined by the finite element simulation strategy; The historical temperature field data is processed according to the dataset partitioning strategy to obtain a training set for training the deep learning model and a validation set for tuning parameters and evaluating the performance of the deep learning model. The deep learning model is trained and its parameters are tuned based on the training set and the validation set to determine the basic value range of each parameter of the deep learning model. Based on the preset multi-objective evolutionary optimization strategy and the basic value range corresponding to each parameter, the target parameter value corresponding to each parameter is determined; The deep learning model is retrained based on the target parameter values ​​to obtain a target deep learning model, and then the predicted temperature field data corresponding to the current process parameters at future time steps is determined based on the target deep learning model.

2. The time-series prediction method for temperature field of steel materials based on multi-objective evolution as described in claim 1, characterized in that, The historical temperature field data is processed according to a dataset partitioning strategy to obtain a training set for training the deep learning model and a validation set for tuning parameters and evaluating the performance of the deep learning model, including: A fixed-length input-output time window is defined, wherein the input portion of the input-output time window includes A time steps and the output portion includes B time steps, where A is an integer greater than 1 and B is an integer not less than 1. Using the input-output time window as the truncation benchmark, the historical temperature field data is processed using the sliding window method to construct multiple sets of input-output mapping pairs; Based on multiple sets of input-output mapping pairs, a training set for training the deep learning model and a validation set for tuning parameters and evaluating the performance of the deep learning model are determined.

3. The time-series prediction method for temperature field of steel materials based on multi-objective evolution as described in claim 2, characterized in that, Based on multiple sets of input-output mapping pairs, a training set for training the deep learning model and a validation set for tuning parameters and evaluating the performance of the deep learning model are determined, including: Based on a preset ratio, the obtained multiple sets of input-output mapping pairs are divided to obtain a training set, a validation set, and a test set for testing the generalization ability of the deep learning model. After retraining the deep learning model based on the target parameter values ​​to obtain the target deep learning model, the process further includes: The test set is extracted to determine the corresponding test subset; The output terms at the corresponding time step of the input-output mapping in the test subset are compared with the prediction terms at the corresponding time step obtained by the target deep learning model to obtain a comparison result characterizing the generalization ability of the target deep learning model.

4. The time-series prediction method for temperature field of steel materials based on multi-objective evolution as described in claim 1, characterized in that, Before processing the historical temperature field data according to the dataset partitioning strategy, the process also includes: The historical temperature field data is normalized to obtain normalized historical temperature field data.

5. The method for time-series prediction of temperature field of steel materials based on multi-objective evolution as described in claim 1, characterized in that, After determining the predicted temperature field data corresponding to the current process parameters at future time steps based on the target deep learning model, the method further includes: Determine whether the predicted temperature field data is greater than a preset temperature threshold; If so, adjust the current process parameters.

6. The time-series prediction method for temperature field of steel materials based on multi-objective evolution as described in claim 1, characterized in that, The deep learning model is a long short-term memory network.

7. The method for time-series prediction of temperature field of steel materials based on multi-objective evolution as described in any one of claims 1 to 6, characterized in that, Based on the preset multi-objective evolutionary optimization strategy and the basic value range corresponding to each parameter, the target parameter value corresponding to each parameter is determined, including: The dual objective function is to maximize prediction accuracy and minimize model complexity. The search interval is the basic value range corresponding to each parameter. The multi-objective evolutionary optimization algorithm is used for iterative optimization to obtain the final population when the iteration terminates. The final population includes multiple individuals, and each individual is a parameter combination composed of the parameter values ​​corresponding to each parameter. Based on the final population, a Pareto front solution set is determined. Then, according to the weighted sum method and the biobjective function, a Pareto optimal parameter combination is determined from the Pareto front solution set. The Pareto optimal parameter combination is a parameter combination consisting of the objective parameter values ​​corresponding to each parameter.

8. The method for time-series prediction of temperature field of steel materials based on multi-objective evolution as described in claim 7, characterized in that, The multi-objective evolutionary optimization algorithm is a non-dominated sorting genetic algorithm.

9. A time-series prediction system for the temperature field of steel materials based on multi-objective evolution, characterized in that, include: The acquisition module is used to acquire historical temperature field data of the substrate and coating under the current process parameters during laser cladding simulation, which is determined by the finite element simulation strategy. The dataset partitioning module is used to process the historical temperature field data according to the dataset partitioning strategy to obtain a training set for training the deep learning model and a validation set for tuning parameters and evaluating the performance of the deep learning model. The training module is used to train and tune the deep learning model based on the training set and the validation set, so as to determine the basic value range of each parameter of the deep learning model. The parameter optimization module is used to determine the target parameter value corresponding to each parameter based on the preset multi-objective evolutionary optimization strategy and the basic value range corresponding to each parameter. The retraining module is used to retrain the deep learning model based on the target parameter values ​​to obtain a target deep learning model, and then determine the predicted temperature field data corresponding to the current process parameters at future time steps based on the target deep learning model.

Images