Method and device for predicting production state of stainless steel AOD furnace

Abstract

The invention provides a method for predicting the production state of a stainless steel AOD furnace, and the method comprises the following steps: collecting historical AOD furnace production state indexes and multi-modal monitoring indexes, and marking the collected production state indexes and monitoring indexes through timestamps; according to the timestamps, aligning the production state indexes with the monitoring indexes according to a unified time sequence; the production state indexes and the monitoring indexes are used, training is conducted through a Transform model, and an AOD furnace production state prediction model is obtained; and real-time AOD furnace monitoring indexes are collected, and the production state indexes of the AOD furnace are predicted through the AOD furnace production state prediction model. The invention further provides a corresponding device for predicting the production state of the stainless steel AOD furnace. According to the AOD furnace production state prediction method based on multi-modal data fusion, time sequence data can be accurately aligned, and high-precision prediction of the production state is achieved.

Description

Technical Field

[0001] This application relates to a technical field related to stainless steel smelting, and in particular to a method and apparatus for predicting the production status of a stainless steel AOD furnace. Background Technology

[0002] With the rapid development of intelligent and digital processes in the metallurgical industry, the AOD furnace (argon-oxygen decarburization) is a core piece of equipment in stainless steel production. Precise control and optimization of its smelting process have become crucial for improving product quality and reducing energy consumption. The steelmaking process involves numerous physical and chemical reactions, the rates and extent of which are influenced by various factors such as raw material composition, furnace atmosphere, temperature distribution, and cooling rate. The environment inside the steelmaking furnace is extremely harsh; high temperature, high pressure, and high oxidizing properties cause rapidly changing conditions within the furnace. These factors interact, making the steelmaking process exceptionally complex. The AOD furnace is primarily used for the refining of stainless steel, involving even more chemical reactions and more complex control strategies. The strict requirements for stainless steel composition and the need for precise temperature and atmosphere control make prediction and rational operation more challenging, necessitating advanced models and algorithms.

[0003] Existing methods for predicting the production status of AOD furnaces mainly rely on worker experience or use traditional spreadsheets to perform metallurgical calculations based on metallurgical formulas to predict AOD production status indicators. Some methods also use computer information systems to assist in metallurgical calculations. However, due to the complexity of the changes in the production status of stainless steel, these methods are not very effective and therefore have the following drawbacks:

[0004] On the one hand, in the steel and stainless steel production sector, forecasting indicators exhibit strong time-series correlations with various production conditions, equipment, operations, and environments. However, the information sources are diverse, encompassing data from various modalities, such as the initial state of molten steel and slag, air flow rate, pressure, location, temperature measurement, feeding, slag removal, flue gas conditions, flame conditions, and ladle usage status. Existing technologies have significant limitations in multimodal data fusion, typically focusing on forecasting only one or a single modality of indicators.

[0005] On the other hand, insufficient precision in time series alignment limits the ability to extract features from multimodal data. In AOD furnace smelting, there is an inherent time delay in the acquisition and processing of different modal data (such as numerical sensor data, image-based flame status, and text-based equipment logs). If the timestamp alignment strategy does not take these delay differences into account, it will lead to misalignment of data from different modalities on the timeline.

[0006] Therefore, in this context, how to provide a method for predicting the production status of AOD furnaces based on multimodal data fusion, accurately aligning time series data, and achieving high-precision prediction of production status is a technical problem that needs to be solved. Summary of the Invention

[0007] In view of the above-mentioned problems of the prior art, this application provides a method and apparatus for predicting the production status of a stainless steel AOD furnace, which can provide a method for predicting the production status of an AOD furnace based on multimodal data fusion, accurately align time series data, and achieve high-precision prediction of the production status.

[0008] To achieve the above objectives, the first aspect of this application provides a method for predicting the production status of a stainless steel AOD furnace, comprising:

[0009] Historical AOD furnace production status indicators and multimodal monitoring indicators are collected, and the collected production status indicators and monitoring indicators are labeled with timestamps.

[0010] Based on the timestamp, align the production status indicators with the monitoring indicators according to a unified time series;

[0011] Using the production status indicators and the monitoring indicators, a Transformer model is used to train and obtain an AOD furnace production status prediction model.

[0012] Real-time AOD furnace monitoring indicators are collected, and the production status indicators of the AOD furnace are predicted through the AOD furnace production status prediction model.

[0013] As described above, a unified time series alignment mechanism was used to achieve precise synchronization of different modalities such as numerical, image, and text data in the time dimension, avoiding feature semantic misalignment caused by time delay. The Transformer model was used for modeling, and the self-attention mechanism was used to improve the perception of complex correlations in multimodal data, enabling real-time and high-precision prediction of production status.

[0014] As one possible implementation of the first aspect, the production status indicators include: molten steel temperature, molten steel composition, and molten steel weight; the monitoring indicators include: initial molten steel temperature, initial molten steel composition, initial molten steel weight, initial slag state, air blowing flow rate, air blowing pressure, air blowing location, feeding amount, slag state, flue gas composition, flame state, and equipment status record; wherein, the initial molten steel temperature, initial molten steel composition, initial molten steel weight, air blowing flow rate, air blowing pressure, feeding amount, and flue gas composition are numerical data; the initial slag state, air blowing location, slag state, and flame state are image data; and the equipment status record is text data.

[0015] The above clarifies the specific content of the production status indicators and monitoring indicators. The monitoring indicators cover all stages of the production process, avoiding inaccurate predictions caused by omissions in monitoring content. The data types of the monitoring indicators have been classified, providing a foundation for the subsequent fusion of multimodal data.

[0016] As one possible implementation of the first aspect, the historical AOD furnace production status indicators and monitoring indicators are collected at different sampling intervals according to different stages of the production process; wherein, the stages of the production process include: the blowing stage, the feeding stage, and the cooling stage.

[0017] As shown above, by setting different sampling intervals for different stages, we can capture key dynamic changes by increasing the sampling frequency in frequently occurring stages; and reduce the sampling frequency in stable stages to reduce redundant data and save storage and computing resources, thereby improving overall data quality and model training efficiency, and enhancing the model's adaptability to stage differences.

[0018] As one possible implementation of the first aspect, after aligning the production status indicators pre-monitoring indicators according to a unified time series, the monitoring indicators at the first moment are pre-processed, including the following steps:

[0019] Features are extracted from the image data, text data, and numerical data, and PCA is used for dimensionality reduction respectively.

[0020] The feature vectors of each modality after dimensionality reduction are merged and normalized to obtain the input feature vector used to train the Transformer model.

[0021] As shown above, by extracting data features from different modalities and performing PCA dimensionality reduction separately, the redundant dimensions of image and text features are effectively reduced, improving computational efficiency. Normalization reduces scale differences between different modalities. Compared to traditional methods such as padding for multimodal data processing, directly merging the feature vectors of each modality and then performing normalization reduces computational redundancy and is more efficient.

[0022] As one possible implementation of the first aspect, the training via the Transformer model includes at least one of the following methods:

[0023] The residual connection mechanism of ResNet is adopted;

[0024] Gating mechanisms are used in the multilayer perceptron layer to filter information.

[0025] As shown above, using residual connections allows for the construction of deeper network structures, thereby improving the model's expressive and learning capabilities. Using gating mechanisms dynamically determines which information should be passed to the next layer and which should be filtered out, enabling the model to focus more on key information and improving overall model performance and prediction accuracy.

[0026] As one possible implementation of the first aspect, the error between the AOD furnace production status index predicted by the AOD furnace production status prediction model and the actual value is calculated as the incremental part of the loss function of the AOD furnace production status prediction model, and the model parameters are optimized and updated.

[0027] As a result, by continuously correcting the model parameters through the error feedback mechanism, the model's adaptability to complex smelting conditions and its long-term stability are enhanced, and the model's fault tolerance is improved.

[0028] The second aspect of this application provides a stainless steel AOD furnace production status prediction device, comprising:

[0029] The data acquisition module is used to collect historical AOD furnace production status indicators and multimodal monitoring indicators, and to annotate the collected production status indicators and monitoring indicators with timestamps.

[0030] The data processing module is used to align the production status indicators and monitoring indicators according to the timestamp and a unified time series.

[0031] The model training module is used to train the AOD furnace production status prediction model using the production status indicators and the monitoring indicators through the Transformer model.

[0032] The real-time prediction module is used to collect real-time AOD furnace monitoring indicators and predict the AOD furnace production status indicators through the AOD furnace production status prediction model.

[0033] A third aspect of this application provides a computing device, including: a processor and a memory storing program instructions thereon, the program instructions, when executed by the processor, causing the processor to perform the stainless steel AOD furnace production status prediction method according to any one of the first aspects.

[0034] The fourth aspect of this application provides a computer-readable storage medium having program instructions stored thereon, which, when executed by a computer, cause the computer to perform the stainless steel AOD furnace production status prediction method according to any one of the first aspects.

[0035] The fifth aspect of this application provides a computer program product including program instructions that, when executed by a computer, cause the computer to perform the stainless steel AOD furnace production status prediction method according to any one of the first aspects. Attached Figure Description

[0036] Figure 1 This is a flowchart of the stainless steel AOD furnace production status prediction method provided in the first embodiment of this application;

[0037] Figure 2a This is a flowchart of the stainless steel AOD furnace production status prediction method provided in the second embodiment of this application;

[0038] Figure 2b This is a schematic diagram of the stainless steel AOD furnace production status prediction method provided in the second embodiment of this application;

[0039] Figure 3 This is a schematic diagram of the stainless steel AOD furnace production status prediction device provided in the embodiments of this application;

[0040] Figure 4 This is a schematic structural diagram of a computing device provided in an embodiment of this application.

[0041] It should be understood that the dimensions and shapes of the blocks in the above structural diagrams are for reference only and should not constitute an exclusive interpretation of the embodiments of the present invention. The relative positions and inclusion relationships between the blocks presented in the structural diagrams are only schematic representations of the structural relationships between the blocks, and are not intended to limit the physical connection methods of the embodiments of the present invention. Detailed Implementation

[0042] The technical solutions provided in this application will be further described below with reference to the accompanying drawings and embodiments. It should be understood that the system architecture and business scenarios provided in the embodiments of this application are mainly for illustrating possible implementations of the technical solutions of this application and should not be construed as the sole limitation on the technical solutions of this application. Those skilled in the art will recognize that the technical solutions provided in this application are equally applicable to similar technical problems as system architectures evolve and new business scenarios emerge.

[0043] It should be understood that the stainless steel AOD furnace production status prediction scheme provided in this application includes a method, apparatus, computing device, computer-readable storage medium, and computer program product for predicting the production status of a stainless steel AOD furnace. Since these technical solutions solve problems using the same or similar principles, some repetitive details may not be repeated in the following descriptions of specific embodiments. However, it should be considered that these specific embodiments have mutual references and can be combined with each other.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. In case of any inconsistency, the meaning set forth in this specification or derived from the content described herein shall prevail. Furthermore, the terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application. To accurately describe the technical content of this application and to accurately understand the invention, the following explanations or definitions of the terms used in this specification are provided before describing specific embodiments:

[0045] 1) ResNet (Residual Network): A deep neural network architecture that solves the vanishing and exploding gradient problems in deep network training by introducing residual blocks. ResNet makes it easier for the network to learn identity mappings, allowing the network to be built deeper while maintaining high performance and stability.

[0046] 2) BERT (Bidirectional Encoder Representations from Transformers): This is a pre-trained language model based on the Transformer architecture. It learns deep semantic representations of text through unsupervised pre-training on a large amount of text data. BERT's innovation lies in its bidirectional training mechanism, which simultaneously considers contextual information to understand the meaning of words.

[0047] 3) PCA (Principal Component Analysis): This is a commonly used statistical method to project high-dimensional data into a low-dimensional space while retaining the most important information. Its basic principle is to perform eigenvalue decomposition on the covariance matrix of the data, find the directions of maximum variance (i.e., principal components), and project the data into a low-dimensional subspace composed of these principal components. PCA dimensionality reduction not only reduces the dimensionality of the data and lowers computational complexity, but also removes noise and redundant information, and is widely used in data preprocessing, feature extraction, and visualization.

[0048] 4) Transformer: The Transformer model is a deep learning model based on an attention mechanism, initially proposed by Vaswani et al. in their 2017 paper "Attention is All You Need". The core idea of ​​the Transformer is to completely abandon the traditional recurrent neural network (RNN) structure, relying solely on the attention mechanism to process sequential data, thereby achieving higher parallelism and faster training speed. Large models based on the Transformer structure have led to applications such as ChartGPT in the NLP field. Multimodal large models refer to models involving two or more modalities. Currently, most multimodal models only involve text and image modalities. They have some applications in image and text generation, but are rarely used in industrial production.

[0049] The stainless steel AOD furnace production status prediction scheme provided in this application collects historical AOD furnace production status indicators and multimodal monitoring indicators, annotating them with timestamps. Based on the timestamps, the production status indicators and monitoring indicators are aligned according to a unified time series. The production status indicators and monitoring indicators are then trained using a Transformer model to obtain an AOD furnace production status prediction model for real-time prediction of AOD furnace production status. This method provides an AOD furnace production status prediction approach based on multimodal data fusion, accurately aligning time series data to achieve high-precision prediction of production status. This application embodiment can be applied to production status monitoring, prediction, and production process improvement in various metallurgical production and research fields. The embodiments of this application are described in detail below with reference to the accompanying drawings.

[0050] The first embodiment of this application provides a method for predicting the production status of a stainless steel AOD furnace, which will be described below in conjunction with... Figure 1 The implementation of each step of the method is described in detail, including steps S10-S40.

[0051] S10: Collect historical AOD furnace production status indicators and multimodal monitoring indicators, and label the collected production status indicators and monitoring indicators with timestamps.

[0052] In some embodiments, the monitoring indicators are collected by real-time sensors and transmitted to a computer (or various host or distributed systems) for real-time processing and analysis. These real-time sensors may include temperature sensors, component analyzers (such as spectrometers), pressure sensors, flow meters, position sensors, cameras and image processing systems, flue gas analyzers, weighing sensors, vibration sensors, and text recording systems (such as system logs). The monitoring indicators can also be recorded by the sensor systems integrated into the metallurgical system. The production status indicators can be obtained using temperature sensors, spectrometers, weight sensors, or other analytical instruments.

[0053] In some embodiments, the production status indicators include: molten steel temperature, molten steel composition, and molten steel weight; the monitoring indicators include: initial molten steel temperature, initial molten steel composition, initial molten steel weight, initial slag state, air blowing flow rate, air blowing pressure, air blowing location, feeding amount, slag state, flue gas composition, flame state, and equipment status records (such as ladle usage status), etc.; wherein, the initial molten steel temperature, initial molten steel composition, initial molten steel weight, air blowing flow rate, air blowing pressure, feeding amount, and flue gas composition are numerical data; the initial slag state, air blowing location, slag state, and flame state are image data; and the equipment status records are text data. The equipment usage status may include physical status (such as temperature, wear condition, cleanliness), operational status (such as current task, usage frequency, location information), and historical records (such as maintenance records, usage cycle), etc. The slag state may include slag removal status, such as the specific slag removal operation steps, time points, frequency, and effects.

[0054] In some embodiments, the historical AOD furnace production status indicators and monitoring indicators have different sampling intervals depending on the stage of the production process; wherein, the stages of the production process include: the blowing stage, the charging stage, and the cooling stage. The blowing stage includes at least one of the following monitoring indicators: initial temperature of molten steel, initial composition of molten steel, initial weight of molten steel, initial state of slag, blowing flow rate, blowing pressure, flue gas composition, and flame state; the charging stage includes at least one of the following monitoring indicators: charging amount, slag state, and equipment status record; the cooling stage includes at least one of the following monitoring indicators: flame state and equipment status record.

[0055] In some embodiments, the historical AOD furnace production status indicators and monitoring indicators have different sampling intervals depending on the stage of the production process; wherein, the stages of the production process include: the blowing stage, the feeding stage, and the cooling stage. For example, the sampling frequency is relatively fast during the blowing stage, and a sampling frequency of 1Hz can be used; the sampling frequency is relatively stable during the feeding stage, and a sampling frequency of 0.2Hz can be used; the sampling frequency can be further reduced during the cooling stage.

[0056] In some embodiments, data is filtered and cleaned using methods combined with metallurgical mechanisms to improve data quality and reduce noise and errors. Reasonable upper and lower thresholds are set based on known ranges of metallurgical process parameters, and data exceeding these ranges is discarded. For example, during the blowing stage, the blowing flow rate is typically maintained between 800 and 1500 Nm³. 3 The airflow rate should be between / h. If the airflow rate collected by the sensor exceeds this range, it may be due to sensor drift, communication error, or equipment failure, and should be rejected.

[0057] In some embodiments, a precise timestamp is attached to the collected production status indicators and monitoring indicators to record the specific time of data collection.

[0058] S20: Based on the timestamp, align the production status indicators with the monitoring indicators according to a unified time series.

[0059] In some embodiments, the production status indicators and monitoring indicators are aligned according to a unified time series. When the data acquisition device needs to process the acquired content, resulting in a delay between the timestamp and the actual event, the delayed event is compensated for on the same event sequence, and the acquired data is adjusted according to the actual acquired events.

[0060] In some embodiments, after aligning the production status indicators pre-monitoring indicators according to a unified time series, the monitoring indicators at the first moment are pre-processed, including the following steps: extracting features from the image data, text data, and numerical data, and using PCA to reduce dimensionality respectively; merging the dimensionality-reduced modal feature vectors, and performing normalization processing to obtain the input feature vector for training the Transformer model.

[0061] In some embodiments, ResNet can be used to extract image features from the image data and PCA can be used to reduce the dimensionality to 64 dimensions; BERT can be used to extract text features from the text data and PCA can be used to reduce the dimensionality to 64 dimensions; and a time-series window sliding expansion can be used to extend the dimensionality to 16 dimensions for the numerical data.

[0062] In some embodiments, the merged feature vectors are normalized using methods such as Batch Normalization and Layer Normalization.

[0063] S30: Using the production status indicators and the monitoring indicators, a Transformer model is used to train the AOD furnace production status prediction model.

[0064] In some embodiments, the production status indicators and monitoring indicators processed in step S20 are input into the Transformer model. The input can be divided into a training set, a validation set, and a test set, with the specific proportions depending on the actual situation.

[0065] In some embodiments, a hidden layer containing 512 units can be used to process the input feature vector; depending on computational capabilities, a hidden layer containing more units can be used.

[0066] In some embodiments, training the Transformer model includes at least one of the following methods: employing the residual connection mechanism of ResNet; or using a gating mechanism in the multilayer perceptron layers to filter information. The hidden layers may employ more units depending on computational capabilities.

[0067] In some embodiments, quantization caching optimization techniques are employed to improve model performance and generalization ability. Specifically, intermediate feature vectors are quantized with low bits (such as INT8, FP16, or lower), converting them from high-precision to low-precision representation, thereby reducing memory bandwidth requirements and computational overhead. During multi-round training or multi-stage inference, the quantized key and value vectors are cached in memory or GPU memory to avoid redundant computation.

[0068] S40: Collects real-time AOD furnace monitoring indicators and predicts the AOD furnace production status indicators through the AOD furnace production status prediction model.

[0069] In some embodiments, the production status indicators of the AOD furnace, predicted by the AOD furnace production status prediction model, guide production staff to improve the production process. For example, the oxygen blowing volume can be adjusted in advance based on the predicted molten steel temperature, or the timing of alloy charging can be optimized based on composition trends. This improves the precision of production process control, reduces oxygen consumption, increases alloy material utilization, and reduces decision-making delays caused by reliance on manual experience. The AOD furnace production status prediction model can also provide a preferred range of monitoring indicators.

[0070] In some embodiments, the error between the AOD furnace production status index predicted by the AOD furnace production status prediction model and the actual value is calculated as the incremental part of the loss function of the AOD furnace production status prediction model, and the model parameters are optimized and updated.

[0071] The second embodiment of this application provides a method for predicting the production status of a stainless steel AOD furnace. The following will refer to... Figure 2a The flowchart shown illustrates that the method provided in this second embodiment includes the following steps S200-S230.

[0072] S200: Collects historical AOD furnace production status indicators and multimodal monitoring indicators.

[0073] The system utilizes real-time sensors such as temperature sensors, pressure sensors, cameras, and image processing systems to collect various key production status indicators during the AOD furnace production process, including: molten steel temperature, molten steel composition, and molten steel weight; as well as monitoring indicators, including: initial molten steel temperature, initial molten steel composition, initial molten steel weight, initial slag state, air blowing flow rate, air blowing pressure, air blowing location, feeding amount, slag state, flue gas composition, flame state, and equipment status records.

[0074] Among them, the initial temperature of molten steel, initial composition of molten steel, initial weight of molten steel, air flow rate, air pressure, feed amount, and flue gas composition are numerical data; the initial state of steel slag, air blowing location, slag state, and flame state are image data; and the equipment status record is text data.

[0075] Production status indicators and monitoring indicators should be collected synchronously, with an accurate timestamp appended to record the specific moment of data collection. Based on the timestamp, the production status indicators and monitoring indicators should be aligned according to a unified time series. For example, the air flow rate collected at a certain moment and the air pressure collected at the same moment should be aligned and used together as the data collected at that moment.

[0076] Data that does not conform to the principles of metallurgical processes is cleaned up. For example, if the air flow rate value collected by the sensor exceeds 800-1500 Nm³, it is considered cleaned up. 3 If the error message is / h, it may be due to sensor drift, communication error, or equipment failure, and should be removed.

[0077] S210: Perform preprocessing on the collected multimodal monitoring indicators.

[0078] After cleaning the data from the metallurgical industry source, the monitoring index data of different modalities are standardized. For example, batch normalization is used to standardize numerical data by calculating the mean and variance of small batches of data to standardize the input, normalizing different monitoring index data to a uniform scale and eliminating the influence of dimensional differences.

[0079] Then, for different types of data, data features are extracted, namely... Figure 2b The layering and blending process described herein is as follows:

[0080] For image data, ResNet is used to extract image features. The output is usually a high-dimensional feature vector (e.g., 2048 dimensions). PCA is used to reduce the dimensionality to 64 dimensions to reduce the feature dimension, reduce redundant information, and retain the expressive power of the main features.

[0081] For text-based data, BERT is used to extract text features. Similarly, PCA is used to reduce the dimensionality from high to 64 dimensions.

[0082] To enhance the ability of numerical data to capture dynamic changes over time (numerical data typically has a higher sampling frequency and needs to reflect dynamic characteristics), a time-series window sliding expansion dimension is used to extend the numerical data to 16 dimensions.

[0083] Since each timestamp may contain data with different modalities—for example, the first time point may contain both numerical and image data, while the second time point may contain only numerical data—a layered fusion approach is used to process the feature data.

[0084] For example: M ni For the nth modal data at time i, F ni For the features of the nth modal data at time i, F ni =net(M ni The feature F obtained at time i is obtained (different feature extraction methods can be used). i for That is, normalization is performed after merging.

[0085] Finally, the input feature data used for training the Transformer model is obtained.

[0086] S220: A prediction model for AOD furnace production status is obtained by training a Transformer model.

[0087] The data used for training was divided into a 3 / 4 training set and a 1 / 4 test set, and then trained using a Transformer model.

[0088] Since this application adopts a multimodal data stacking and fusion method, in order to prevent problems such as inconsistent feature dimensions, longer information transmission paths, gradient vanishing, and unstable training caused by the method not limiting the amount of modal data at a certain time, it is necessary to optimize the structure of the Transformer model.

[0089] Specifically, a hidden layer with 512 units is first constructed to process the input feature vector.

[0090] Secondly, the residual connection mechanism of ResNet is introduced into each sub-layer of Transformer, that is, the input vector is superimposed on the output of the sub-layer, and layer normalization is performed after superposition.

[0091] Furthermore, a gating mechanism is used in the Transformer multilayer perceptron layer, that is, a learnable gating function is added after the linear transformation of the multilayer perceptron to perform weighted filtering of the activated features, thereby realizing dynamic filtering control of information.

[0092] Training can also improve model performance and generalization ability by employing quantization caching optimization techniques. For example, low-bit quantization (such as INT8, FP16, or lower bit-level quantization) can be used to convert intermediate feature vectors from high precision to low precision representation, thereby reducing memory bandwidth requirements and computational overhead. During multi-round training or multi-stage inference, the quantized key and value vectors can be cached in memory or GPU memory to avoid redundant computation.

[0093] When the prediction rate of production status indicators on the test set reaches the expected standard, the AOD furnace production status prediction model for real-time prediction is obtained.

[0094] S230: Real-time collection and monitoring of indicators and prediction using the AOD furnace production status prediction model.

[0095] The same acquisition method as step S200 is used to collect monitoring indicators in real time, and the AOD furnace production status prediction model is used for prediction.

[0096] The production status indicators of the furnace predicted by the AOD furnace production status prediction model can guide production staff to improve the production process. For example, the oxygen blowing amount can be adjusted in advance based on the predicted molten steel temperature, or the timing of alloy charging can be optimized based on the composition trend, thereby improving the precision of production process control, reducing oxygen consumption, increasing alloy material utilization, and reducing decision-making delays caused by reliance on manual experience.

[0097] The prediction results obtained from the AOD furnace production status prediction model can also be used to optimize and update model parameters, thereby improving the model's stability and predictive ability.

[0098] The embodiments of this application are developed and designed for the production control of stainless steel AOD, and their main advantages are as follows:

[0099] 1) Advantages of multimodal data fusion: By adopting an innovative multimodal data fusion method, it can comprehensively integrate multi-dimensional information such as the initial state of molten steel and slag, gas flow rate, pressure and location. Compared with a single data source, it can obtain richer and more comprehensive production process characteristics, deeply explore the potential relationships behind the data, provide a solid data foundation for model prediction, and thus more accurately reflect the actual production status of AOD furnace.

[0100] 2) Advantages of model structure improvement: The large model structure has been improved in a targeted manner. The data has been screened and cleaned in combination with metallurgical mechanism. This not only improves the data quality, but also makes the model more in line with the professional knowledge and actual needs of AOD furnace production. The optimized model is more efficient and accurate in learning and processing production data, which greatly enhances the model's performance and generalization ability.

[0101] 3) High-precision prediction advantages: The computational model has high prediction accuracy and can predict the AOD furnace production process status in real time and accurately, including key parameters such as steel temperature, composition, and weight. Production personnel can make decisions and adjustments in advance based on accurate predictions, avoiding product quality problems caused by parameter deviations, effectively reducing scrap rates, and improving product quality stability and consistency.

[0102] 4) Strong fault tolerance and adaptability: It has strong fault tolerance and can adapt to the complex and ever-changing working conditions of AOD production. Whether it is raw material fluctuations, equipment performance changes, or environmental conditions change, the system can operate stably and maintain high prediction accuracy, reduce the interference of external factors on production, and ensure the continuity and stability of the production process.

[0103] 5) Production control optimization advantages: Effectively improves the precision of current AOD production control. Through accurate prediction and intelligent decision-making, it achieves refined control of the production process, such as precise control of feeding amount, blowing time and pressure, etc., thereby improving production efficiency, reducing energy and raw material consumption, reducing production costs, and enhancing the company's competitiveness in the stainless steel production field.

[0104] 6) Professional customization and adaptation advantages: Specifically developed and designed for stainless steel AOD production control, with in-depth understanding of the process characteristics, technical requirements and actual pain points of this production process. The system's functions and parameter settings are closely based on the production needs of AOD furnace. Compared with general systems, it is more in line with the actual production scenario and has stronger practicality and adaptability.

[0105] The embodiments of this application can improve the accuracy and stability of production process prediction. Based on the prediction results, the system can optimize the control strategy. This technology breaks through the limitations of traditional AOD furnace control that relies on human experience and single-parameter monitoring, and provides an intelligent solution for the stainless steel smelting industry.

[0106] The third embodiment of this application provides a stainless steel AOD furnace production status prediction device, which can be used to implement the stainless steel AOD furnace production status prediction method in the above embodiments, such as... Figure 3 As shown, the stainless steel AOD furnace production status prediction device includes:

[0107] The data acquisition module is used to collect historical AOD furnace production status indicators and multimodal monitoring indicators, and to annotate the collected production status indicators and monitoring indicators with timestamps; specifically, the data acquisition module can be used to implement step S10 in the first embodiment and its optional embodiments.

[0108] The data processing module is used to align the production status indicators and monitoring indicators according to the timestamp and a unified time series; specifically, the data processing module can be used to implement step S20 in the first embodiment and its optional embodiments.

[0109] The model training module is used to train the AOD furnace production status prediction model using the production status indicators and the monitoring indicators through the Transformer model; specifically, the model training module can be used to implement step S30 in the first embodiment and its optional embodiments.

[0110] The real-time prediction module is used to collect real-time AOD furnace monitoring indicators and predict the AOD furnace production status indicators through the AOD furnace production status prediction model. Specifically, the real-time prediction module can be used to implement step S40 in the first embodiment and its optional embodiments.

[0111] Figure 4 This is a schematic structural diagram of a computing device 900 provided in an embodiment of this application. This computing device can execute various optional embodiments of the methods described above. The computing device can be a terminal, or a chip or chip system within the terminal. Figure 4 As shown, the computing device 900 includes: a processor 910, a memory 920, and a communication interface 930.

[0112] It should be understood that Figure 4 The communication interface 930 in the computing device 900 shown can be used to communicate with other devices, and may specifically include one or more transceiver circuits or interface circuits.

[0113] The processor 910 can be connected to the memory 920. The memory 920 can be used to store the program code and data. Therefore, the memory 920 can be a storage unit inside the processor 910, an external storage unit independent of the processor 910, or a component that includes both the storage unit inside the processor 910 and the external storage unit independent of the processor 910.

[0114] Optionally, the computing device 900 may also include a bus. The memory 920 and communication interface 930 can be connected to the processor 910 via the bus. The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 The symbol is represented by a line without an arrow, but this does not mean that there is only one bus or one type of bus.

[0115] It should be understood that in the embodiments of this application, the processor 910 may be a central processing unit (CPU). The processor may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. Alternatively, the processor 910 may employ one or more integrated circuits to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0116] The memory 920 may include read-only memory and random access memory, and provides instructions and data to the processor 910. A portion of the processor 910 may also include non-volatile random access memory. For example, the processor 910 may also store device type information.

[0117] When the computing device 900 is running, the processor 910 executes computer execution instructions stored in the memory 920 to perform any of the operational steps of the above method and any of the optional embodiments thereof.

[0118] It should be understood that the computing device 900 according to the embodiments of this application can correspond to the corresponding subject in executing the methods according to the various embodiments of this application, and the above and other operations and / or functions of each module in the computing device 900 are respectively for implementing the corresponding processes of the methods of this embodiment. For the sake of brevity, they will not be described in detail here.

[0119] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0120] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0121] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0122] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0123] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0124] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0125] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is used to perform the above-described method, which includes at least one of the schemes described in the above embodiments.

[0126] The computer storage medium in this application embodiment can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0127] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0128] The program code contained on a computer-readable medium may be transmitted using any suitable medium, including, but not limited to, wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0129] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0130] Furthermore, the terms "first, second, third, etc." or similar terms such as module A, module B, and module C used in the specification and claims are only used to distinguish similar objects and do not represent a specific ordering of objects. It is understood that, where permissible, a specific order or sequence may be interchanged so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0131] In the above description, the labels of the steps involved, such as S110, S120, etc., do not mean that the steps will necessarily be executed. The order of the steps can be interchanged or executed simultaneously if permitted.

[0132] The term "comprising" as used in the specification and claims should not be construed as limiting itself to what follows; it does not exclude other elements or steps. Therefore, it should be interpreted as specifying the presence of the mentioned feature, integral, step, or component, but does not exclude the presence or addition of one or more other features, integrals, steps, or components, or groups thereof. Thus, the statement "device comprising means A and B" should not be limited to a device consisting solely of components A and B.

[0133] The terms "an embodiment" or "an embodiment" as used in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in at least one embodiment of this application. Therefore, the terms "in one embodiment" or "in an embodiment" appearing throughout this specification do not necessarily refer to the same embodiment, but may refer to the same embodiment. Furthermore, in one or more embodiments, the particular features, structures, or characteristics can be combined in any suitable manner, as will be apparent to those skilled in the art from this disclosure.

[0134] Note that the above are merely preferred embodiments and the technical principles employed in this application. Those skilled in the art will understand that this application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of this application, all of which fall within the scope of protection of this application.

Claims

1. A method for predicting the production status of a stainless steel AOD furnace, characterized in that, Includes the following steps: Historical AOD furnace production status indicators and multimodal monitoring indicators are collected, and the collected production status indicators and monitoring indicators are labeled with timestamps. Based on the timestamp, align the production status indicators with the monitoring indicators according to a unified time series; Using the production status indicators and the monitoring indicators, a Transformer model is used to train and obtain an AOD furnace production status prediction model. Real-time AOD furnace monitoring indicators are collected, and the production status indicators of the AOD furnace are predicted through the AOD furnace production status prediction model.

2. The method according to claim 1, characterized in that, The production status indicators include: molten steel temperature, molten steel composition, and molten steel weight; the monitoring indicators include: initial molten steel temperature, initial molten steel composition, initial molten steel weight, initial slag blowing flow rate, blowing pressure, blowing location, feeding amount, slag condition, flue gas composition, flame condition, and equipment condition record. Among them, the initial temperature of molten steel, initial composition of molten steel, initial weight of molten steel, air flow rate, air pressure, feeding amount, and flue gas composition are numerical data; The initial state of the steel slag, the air blowing position, the slag state, and the flame state are image-type data; The device status record is text-based data.

3. The method according to claim 2, characterized in that, The historical AOD furnace production status indicators and monitoring indicators are collected at different sampling intervals according to different stages of the production process; wherein, the stages of the production process include: the blowing stage, the feeding stage, and the cooling stage.

4. The method according to claim 3, characterized in that, After aligning the production status indicators with the pre-monitoring indicators according to a unified time series, the monitoring indicators at the first moment are pre-processed, including the following steps: Features are extracted from the image data, text data, and numerical data, and PCA dimensionality reduction is applied to each. The feature vectors of each modality after dimensionality reduction are merged and normalized to obtain the input feature vector used to train the Transformer model.

5. The method according to claim 1, characterized in that, The training using the Transformer model includes at least one of the following methods: The residual connection mechanism of ResNet is adopted; Gating mechanisms are used in the multilayer perceptron layer to filter information.

6. The method according to claim 1, characterized in that, The error between the AOD furnace production status index predicted by the AOD furnace production status prediction model and the actual value is calculated as the incremental part of the loss function of the AOD furnace production status prediction model, and the model parameters are optimized and updated.

7. A production status prediction device for a stainless steel AOD furnace, characterized in that, include: The data acquisition module is used to collect historical AOD furnace production status indicators and multimodal monitoring indicators, and to annotate the collected production status indicators and monitoring indicators with timestamps. The data processing module is used to align the production status indicators and monitoring indicators according to the timestamp and a unified time series. The model training module is used to train the AOD furnace production status prediction model using the production status indicators and the monitoring indicators through the Transformer model. The real-time prediction module is used to collect real-time AOD furnace monitoring indicators and predict the AOD furnace production status indicators through the AOD furnace production status prediction model.

8. A computing device, characterized in that, include: processor, and A memory storing program instructions that, when executed by the processor, cause the processor to perform the stainless steel AOD furnace production status prediction method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, It stores program instructions that, when executed by a computer, cause the computer to perform the stainless steel AOD furnace production status prediction method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes program instructions that, when executed by a computer, cause the computer to perform the stainless steel AOD furnace production status prediction method according to any one of claims 1 to 6.

Images