Casting blank temperature prediction method based on AI model

Abstract

The invention provides a casting blank temperature prediction method based on an AI model, and belongs to the technical field of casting blank temperature intelligent prediction.The casting blank temperature prediction method comprises the steps that initial conditions of a casting blank, real-time data of the cooling process and rear-end casting blank temperature data are collected; performing data cleaning, feature extraction and data scaling on the acquired data; constructing a casting blank temperature prediction AI model; training and verifying the model; performing prediction and reasoning; period management; and result visualization and feedback are carried out. According to the casting blank temperature prediction method based on the AI model, the surface temperature of the continuous casting blank can be accurately known, the solidification process can be controlled, quality defects such as surface cracks and segregation are reduced, and therefore the internal quality and the surface quality of the casting blank are improved. And by monitoring the temperature in real time, a basis is provided for adjusting process parameters such as the cooling speed and the throwing speed, so that the continuous casting process is optimized. And production accidents caused by too high or too low temperature can be avoided, and the stability and safety of the production process are guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent prediction of casting temperature, and in particular to a casting temperature prediction method based on an AI model. Background Art

[0002] The prediction of billet temperature is crucial for converter steelmaking, but the existing billet temperature prediction methods have the following problems:

[0003] Data quality issues:

[0004] Noise and missing values: Sensors may be disturbed during measurement, resulting in noisy or missing data. This affects the accuracy and reliability of the model.

[0005] Limitations of feature selection:

[0006] Some methods may rely on experience in feature selection and ignore key physical phenomena. If the feature selection is inappropriate, the predictive ability of the model will be limited.

[0007] The generalization ability of the model:

[0008] Common AI models may perform well on the training set, but are prone to overfitting when faced with new, unseen data, that is, the model's prediction effect on new data is poor.

[0009] Challenges of real-time prediction:

[0010] Complex models such as deep learning have high computational requirements, which may result in an inability to respond quickly in real-time monitoring. In addition, there may be a delay from data collection to result generation, affecting the timeliness of decision-making.

[0011] Insufficient dynamic adaptability:

[0012] The environment and conditions in the casting process often change, and existing models may lack flexibility and cannot automatically adapt to these changes, thereby reducing the prediction effect.

[0013] Lack of feedback mechanism:

[0014] Some forecasting methods fail to effectively integrate real-time monitoring data and lack a feedback mechanism for adjusting and optimizing the results.

[0015] Lack of explanation:

[0016] AI models, especially deep learning models, are often viewed as “black boxes”, making it difficult for users to understand their internal workings, which may lead to reduced trust and acceptance in industrial applications.

[0017] High implementation and maintenance costs:

[0018] Building and maintaining AI-based predictive systems requires significant resources, including hardware, software, and specialized personnel, which can be a significant expense for many businesses.

[0019] Therefore, there is an urgent need in the art for a technical solution that can solve one or more of the above problems.

[0020] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention and should not be regarded as an acknowledgment or any form of suggestion that the information constitutes the prior art already known to a person skilled in the art. Summary of the invention

[0021] The purpose of the present invention is to provide a technical solution capable of accurately predicting the temperature of a casting slab.

[0022] To achieve the above object, the present invention provides the following solutions:

[0023] A method for predicting billet temperature based on an AI model, comprising:

[0024] Collect the initial conditions of the ingot, real-time data of the cooling process and the back-end ingot temperature data;

[0025] Performing data cleaning, feature extraction and data scaling on the collected data;

[0026] Construct an AI model for predicting billet temperature;

[0027] Model training and validation;

[0028] Make predictions and inferences;

[0029] Cycle management;

[0030] Result visualization and feedback.

[0031] Optionally, the initial conditions of the casting include: superheat, casting speed, secondary cooling water temperature, alloy composition, casting mold temperature and casting process parameters;

[0032] The real-time data of the cooling process includes: sensor signals, flow rate and temperature of the cooling medium and external environmental condition parameters;

[0033] The back-end billet temperature data includes real-time temperature monitoring data of the billet surface and interior.

[0034] Optionally, the data cleaning, feature extraction and data scaling of the collected data specifically include:

[0035] Remove noise and missing values ​​to ensure data quality; extract important features related to billet cooling based on physical models, including time delay, pouring conditions, and material thermal conductivity; standardize or normalize temperature data.

[0036] Optionally, the temperature prediction AI model architecture specifically includes:

[0037] Input layer, inputting the initial conditions and real-time cooling condition data of the ingot;

[0038] Several hidden layers, including multiple fully connected layers, with different numbers of neurons in each layer; using activation algorithms to increase nonlinearity; using dropout layers to reduce overfitting;

[0039] Temporal layer: when the cooling time is long and the process has temporal characteristics, a long short-term memory network or a gated recurrent unit is embedded in the network to process time series data;

[0040] Output layer, outputs the predicted solidification or front temperature of the ingot.

[0041] Optionally, the model training and verification includes:

[0042] Divide the dataset into training set, validation set and test set; use appropriate batches and iterations to train the model and monitor the changes in the loss function; use the k-fold cross-validation method to verify the generalization ability of the model.

[0043] Optionally, the performing prediction and reasoning includes:

[0044] Input real-time data into the trained model for temperature prediction; predict the front-end solidification temperature based on the back-end temperature, and achieve higher prediction accuracy through the combination of physical models and AI models.

[0045] Optionally, the cycle management includes:

[0046] Continuous training of models: Regularly retrain and optimize the model using new foundry data;

[0047] Performance Evaluation: Regularly evaluate the performance of the model and make improvements.

[0048] Optionally, the result visualization and feedback includes:

[0049] Use visualization tools to show predicted temperatures versus actual temperatures;

[0050] Create a user-friendly interface that allows operators to view real-time data and forecast results.

[0051] Optionally, the temperature prediction AI model also includes:

[0052] Multi-layer network structure: The model adopts a multi-layer architecture, including data preprocessing layer, feature extraction layer, time series prediction layer and generative adversarial layer. Through layered processing, different features in the data are extracted respectively;

[0053] Dynamic timing feature extraction:

[0054] Long short-term memory network: used to process time series data, capture long-term and short-term dependencies, and retain the time series characteristics of the casting process;

[0055] Convolutional neural network: used to process spatial features, extract features from sensor data at different locations in the casting process, and fuse spatial information;

[0056] Generative Adversarial Network Mechanism:

[0057] Generator: Generates predicted distribution of ingot temperature based on input casting conditions, historical data and timing characteristics;

[0058] Discriminator: used to evaluate the difference between the generated temperature prediction and the actual temperature, and improve the prediction accuracy of the model through adversarial training;

[0059] Adaptive loss function:

[0060] Comprehensive loss function: Dynamically adjust the mean square error and adversarial loss to better balance the learning of the generator and the discriminator during training;

[0061] Data augmentation and adaptive learning:

[0062] Incremental learning mechanism: Using transfer learning methods to fine-tune the model with newly collected data to cope with changes in conditions during the casting process;

[0063] Data augmentation technology: Generate additional training samples through simulation experiments to enrich the training data set;

[0064] Real-time feedback and online learning:

[0065] The model can receive real-time monitoring data in real time, perform online predictions and adjustments, and continuously optimize the model using incremental learning.

[0066] Optionally, the transfer learning method includes:

[0067] Pre-trained models:

[0068] Use existing models: first train a deep learning model on a large dataset, and then transfer the model or part of it to the target task;

[0069] Fine-tuning: Load the pre-trained model in the target task, and then perform a small amount of training with the target data to adjust the model parameters;

[0070] Field Adaptation:

[0071] Inter-domain alignment: By aligning the distribution between the source domain and the target domain, the model can be better generalized in new environments.

[0072] Multi-task learning:

[0073] Shared representation: In multi-task scenarios, by training a model to handle multiple related tasks simultaneously, the model shares the underlying features, thereby improving the performance on each single task; different prediction tasks in the casting process can be designed as multi-task simultaneous training;

[0074] Incremental Learning:

[0075] Continuous learning: When new data arrives, update the existing model to avoid retraining from scratch. By fixing some parameters of the model and training only some new layers, the model's capabilities can be gradually improved.

[0076] External knowledge introduction:

[0077] Incorporating expert knowledge: By using engineers’ experience or domain knowledge to guide the model migration process, such as incorporating prior knowledge during feature selection or model structure design;

[0078] Sample reweighting:

[0079] Give new data a higher weight: In transfer learning, the weight of new data in training is adjusted according to the extent to which it affects the performance of the model after conversion, helping the model adapt to the target task faster.

[0080] Compared with the prior art, the present invention has the following beneficial effects:

[0081] The AI ​​model-based ingot temperature prediction method provided by the present invention can accurately understand the surface temperature of the continuous casting ingot, help control the solidification process, reduce quality defects such as surface cracks and segregation, and thus improve the internal and surface quality of the ingot. By real-time monitoring of temperature, a basis is provided for adjusting process parameters such as cooling rate, billet drawing speed, etc. to optimize the continuous casting process. It can also avoid production accidents caused by excessively high or low temperatures and ensure the stability and safety of the production process. Reduce waste and defective products caused by quality problems, thereby improving the yield rate of the final product and reducing production costs. Reasonable temperature control can optimize energy utilization, reduce unnecessary energy consumption, reduce the impact on the environment, and meet the requirements of sustainable development. Appropriate ingot temperature helps to reduce the thermal load of the equipment, extend the service life of the continuous casting equipment, and reduce the maintenance cost of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0083] Figure 1 A schematic flow chart of a method for predicting temperature of a slab based on an AI model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0084] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0085] The purpose of the present invention is to provide a technical solution capable of accurately predicting the temperature of a casting slab.

[0086] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0087] Embodiment 1:

[0088] This embodiment provides a method for predicting the temperature of a casting billet based on an AI model. Figure 1 As shown, including:

[0089] Collect the initial conditions of the ingot, real-time data of the cooling process and the back-end ingot temperature data;

[0090] Performing data cleaning, feature extraction and data scaling on the collected data;

[0091] Construct an AI model for predicting billet temperature;

[0092] Model training and validation;

[0093] Make predictions and inferences;

[0094] Cycle management;

[0095] Result visualization and feedback.

[0096] Furthermore, the initial conditions of the casting include: superheat, casting speed, secondary cooling water temperature, alloy composition, casting mold temperature and casting process parameters;

[0097] The real-time data of the cooling process includes: sensor signals, flow rate and temperature of the cooling medium and external environmental condition parameters;

[0098] The back-end billet temperature data includes real-time temperature monitoring data of the billet surface and interior.

[0099] Furthermore, the data cleaning, feature extraction and data scaling of the collected data specifically include:

[0100] Remove noise and missing values ​​to ensure data quality; extract important features related to billet cooling based on physical models, including time delay, pouring conditions, and material thermal conductivity; standardize or normalize temperature data.

[0101] Furthermore, the temperature prediction AI model architecture specifically includes:

[0102] Input layer, inputting the initial conditions and real-time cooling condition data of the ingot;

[0103] Several hidden layers, including multiple fully connected layers, with different numbers of neurons in each layer; using activation algorithms to increase nonlinearity; using dropout layers to reduce overfitting;

[0104] Temporal layer: when the cooling time is long and the process has temporal characteristics, a long short-term memory network or a gated recurrent unit is embedded in the network to process time series data;

[0105] Output layer, outputs the predicted solidification or front temperature of the ingot.

[0106] Furthermore, the model training and verification includes:

[0107] Divide the dataset into training set, validation set and test set; use appropriate batches and iterations to train the model and monitor the changes in the loss function; use the k-fold cross-validation method to verify the generalization ability of the model.

[0108] Further, the prediction and reasoning include:

[0109] Input real-time data into the trained model for temperature prediction; predict the front-end solidification temperature based on the back-end temperature, and achieve higher prediction accuracy through the combination of physical models and AI models.

[0110] Furthermore, the cycle management includes:

[0111] Continuous training of models: Regularly retrain and optimize the model using new foundry data;

[0112] Performance Evaluation: Regularly evaluate the performance of the model and make improvements.

[0113] Furthermore, the result visualization and feedback include:

[0114] Use visualization tools to show predicted temperatures versus actual temperatures;

[0115] Create a user-friendly interface that allows operators to view real-time data and forecast results.

[0116] Furthermore, the temperature prediction AI model also includes:

[0117] Multi-layer network structure: The model adopts a multi-layer architecture, including data preprocessing layer, feature extraction layer, time series prediction layer and generative adversarial layer. Through layered processing, different features in the data are extracted respectively;

[0118] Dynamic timing feature extraction:

[0119] Long Short-Term Memory Network (LSTM): used to process time series data, capture long-term and short-term dependencies, and retain the time series characteristics of the casting process;

[0120] Convolutional Neural Network (CNN): used to process spatial features, extract features from sensor data at different locations in the casting process, and fuse spatial information;

[0121] Generative Adversarial Network Mechanism:

[0122] Generator: Generates predicted distribution of ingot temperature based on input casting conditions, historical data and timing characteristics;

[0123] Discriminator: used to evaluate the difference between the generated temperature prediction and the actual temperature, and improve the prediction accuracy of the model through adversarial training;

[0124] Adaptive loss function:

[0125] Comprehensive loss function: Combines mean square error (MSE) and adversarial loss (AD) for dynamic adjustment to better balance the learning of the generator and discriminator during training;

[0126] Data augmentation and adaptive learning:

[0127] Incremental learning mechanism: Using transfer learning methods to fine-tune the model with newly collected data to cope with changes in conditions during the casting process;

[0128] Data augmentation technology: Generate additional training samples through simulation experiments to enrich the training data set;

[0129] Real-time feedback and online learning:

[0130] The model can receive real-time monitoring data in real time, perform online predictions and adjustments, and continuously optimize the model using incremental learning.

[0131] Furthermore, the transfer learning method includes:

[0132] Pre-trained models:

[0133] Use existing models: First, train a deep learning model (e.g., CNN, LSTM, etc.) on a large dataset (e.g., historical data of the casting process), and then migrate the model or parts of it (e.g., weights and feature extraction layers) to the target task (temperature prediction in a specific casting environment);

[0134] Fine-tuning: Load the pre-trained model in the target task, and then perform a small amount of training with the target data to adjust the model parameters;

[0135] Field Adaptation:

[0136] Inter-domain alignment: By aligning the distribution between the source domain (existing dataset) and the target domain (new dataset), the model can be better generalized in the new environment. Common methods include maximum mean difference (MMD) and adversarial training.

[0137] Multi-task learning:

[0138] Shared representation: In multi-task scenarios, by training a model to handle multiple related tasks simultaneously, the model shares the underlying features, thereby improving the performance on each single task; different prediction tasks in the casting process (such as temperature, composition, etc.) can be designed as multiple tasks for simultaneous training;

[0139] Incremental Learning:

[0140] Continuous learning: When new data arrives, update the existing model to avoid retraining from scratch. By fixing some parameters of the model and training only some new layers, the model's capabilities can be gradually improved.

[0141] External knowledge introduction:

[0142] Incorporating expert knowledge: By using engineers’ experience or domain knowledge to guide the model migration process, such as incorporating prior knowledge during feature selection or model structure design;

[0143] Sample reweighting:

[0144] Give new data a higher weight: In transfer learning, the weight of new data in training is adjusted according to the extent to which it affects the performance of the model after conversion, helping the model adapt to the target task faster.

[0145] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0146] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for predicting billet temperature based on an AI model, characterized in that: include: Collect the initial conditions of the ingot, real-time data of the cooling process and the back-end ingot temperature data; Performing data cleaning, feature extraction and data scaling on the collected data; Construct an AI model for predicting billet temperature; Model training and validation; Make predictions and inferences; Cycle management; Result visualization and feedback.

2. The method for predicting temperature of a casting billet based on an AI model according to claim 1, characterized in that: The initial conditions of the casting include: superheat, casting speed, secondary cooling water temperature, alloy composition, casting mold temperature and casting process parameters; The real-time data of the cooling process includes: sensor signals, flow rate and temperature of the cooling medium and external environmental condition parameters; The back-end billet temperature data includes real-time temperature monitoring data of the billet surface and interior.

3. The method for predicting temperature of a casting billet based on an AI model according to claim 1, characterized in that: The data cleaning, feature extraction and data scaling of the collected data specifically include: Remove noise and missing values ​​to ensure data quality; extract important features related to billet cooling based on physical models, including time delay, pouring conditions, and material thermal conductivity; standardize or normalize temperature data.

4. The method for predicting temperature of a casting billet based on an AI model according to claim 1, characterized in that: The temperature prediction AI model architecture specifically includes: Input layer, inputting the initial conditions and real-time cooling condition data of the ingot; Several hidden layers, including multiple fully connected layers, with different numbers of neurons in each layer; using activation algorithms to increase nonlinearity; using dropout layers to reduce overfitting; Temporal layer: when the cooling time is long and the process has temporal characteristics, a long short-term memory network or a gated recurrent unit is embedded in the network to process time series data; Output layer, outputs the predicted solidification or front temperature of the ingot.

5. The method for predicting temperature of a casting billet based on an AI model according to claim 1, characterized in that: The model training and verification includes: Divide the dataset into training set, validation set and test set; use appropriate batches and iterations to train the model and monitor the changes in the loss function; use the k-fold cross-validation method to verify the generalization ability of the model.

6. The method for predicting temperature of a casting billet based on an AI model according to claim 1, characterized in that: The prediction and inference include: Input real-time data into the trained model for temperature prediction; predict the front-end solidification temperature based on the back-end temperature, and achieve higher prediction accuracy through the combination of physical models and AI models.

7. The method for predicting temperature of a casting billet based on an AI model according to claim 1, characterized in that: The cycle management includes: Continuous training of models: Regularly retrain and optimize the model using new foundry data; Performance Evaluation: Regularly evaluate the performance of the model and make improvements.

8. The method for predicting temperature of a casting billet based on an AI model according to claim 1, characterized in that: The result visualization and feedback include: Use visualization tools to show predicted temperatures versus actual temperatures; Create a user-friendly interface that allows operators to view real-time data and forecast results.

9. The method for predicting temperature of a casting billet based on an AI model according to claim 4, characterized in that: The temperature prediction AI model also includes: Multi-layer network structure: The model adopts a multi-layer architecture, including data preprocessing layer, feature extraction layer, time series prediction layer and generative adversarial layer. Through layered processing, different features in the data are extracted respectively; Dynamic timing feature extraction: Long short-term memory network: used to process time series data, capture long-term and short-term dependencies, and retain the time series characteristics of the casting process; Convolutional neural network: used to process spatial features, extract features from sensor data at different locations in the casting process, and fuse spatial information; Generative Adversarial Network Mechanism: Generator: Generates predicted distribution of ingot temperature based on input casting conditions, historical data and timing characteristics; Discriminator: used to evaluate the difference between the generated temperature prediction and the actual temperature, and improve the prediction accuracy of the model through adversarial training; Adaptive loss function: Comprehensive loss function: Dynamically adjust the mean square error and adversarial loss to better balance the learning of the generator and the discriminator during training; Data augmentation and adaptive learning: Incremental learning mechanism: Using transfer learning methods to fine-tune the model with newly collected data to cope with changes in conditions during the casting process; Data augmentation technology: Generate additional training samples through simulation experiments to enrich the training data set; Real-time feedback and online learning: The model can receive real-time monitoring data in real time, perform online predictions and adjustments, and continuously optimize the model using incremental learning.

10. The method for predicting temperature of a casting billet based on an AI model according to claim 9, characterized in that: The transfer learning method comprises: Pre-trained models: Use existing models: first train a deep learning model on a large dataset, and then transfer the model or part of it to the target task; Fine-tuning: Load the pre-trained model in the target task, and then perform a small amount of training with the target data to adjust the model parameters; Field Adaptation: Inter-domain alignment: By aligning the distribution between the source domain and the target domain, the model can be better generalized in new environments. Multi-task learning: Shared representation: In multi-task scenarios, by training a model to handle multiple related tasks simultaneously, the model shares the underlying features, thereby improving the performance on each single task; different prediction tasks in the casting process can be designed as multi-task simultaneous training; Incremental Learning: Continuous learning: When new data arrives, update the existing model to avoid retraining from scratch. By fixing some parameters of the model and training only some new layers, the model's capabilities can be gradually improved. External knowledge introduction: Incorporating expert knowledge: By using engineers’ experience or domain knowledge to guide the model migration process, such as incorporating prior knowledge during feature selection or model structure design; Sample reweighting: Give new data a higher weight: In transfer learning, the weight of new data in training is adjusted according to the extent to which it affects the performance of the model after conversion, helping the model adapt to the target task faster.

Images