Blast furnace parameter prediction general modeling method and system fusing domain knowledge

Abstract

The invention provides a blast furnace parameter prediction general modeling method and system fusing domain knowledge, relates to the technical field of steel smelting, and aims at solving the problems that an existing blast furnace parameter prediction method is poor in universality and low in modeling efficiency. The method comprises the steps of performing semantic analysis and mapping on multi-source heterogeneous data based on a pre-constructed domain semantic model to generate standardized structured data; based on a knowledge base containing expert rules and mechanism rules, processing the standardized structured data to generate domain features; and automatically executing a model training process through a configurable automatic modeling tool chain according to the configured target parameters, and outputting a prediction model. According to the method, rapid, accurate and reusable automatic modeling of different blast furnaces and different key parameters can be realized, and the universality, the accuracy and the modeling efficiency of the model are improved.

Description

Technical Field

[0001] This invention relates to the field of iron and steel smelting technology, and in particular to a method and system for predicting blast furnace operating parameters based on data modeling. Background Technology

[0002] Blast furnace ironmaking is a core component of modern steel production. Its internal smelting process is extremely complex, characterized by multiple variables, strong coupling, and large time delays, and is often regarded as a typical "black box" system. To achieve precise monitoring of blast furnace status and prediction of key parameters, thereby guiding production operations and improving production efficiency and stability, data-driven modeling methods have been widely used in the industry.

[0003] In existing technologies, multi-source data from the blast furnace production process is typically collected, and data mining or machine learning algorithms are used to build predictive models. Some solutions establish data platforms to integrate and unify the representation of multi-source heterogeneous data, and construct knowledge systems including expert rules and metallurgical mechanism models to assist in model building. However, these existing technical solutions are often comprehensive application platforms with fixed functions, and still have the following shortcomings in practical applications: First, they lack versatility. Many predictive models are specifically developed for a single or a few specific parameters (such as furnace temperature or gas utilization rate). When production demands change and other key parameters (such as silicon content in molten iron or hearth heat load) need to be predicted, data processing, feature engineering, and model development often need to be redone, resulting in long R&D cycles, high costs, and difficulty in quickly responding to changing production forecasting tasks.

[0004] Secondly, the generalization ability is weak. Different blast furnaces differ significantly in equipment structure, raw material composition, and operating procedures. This makes it difficult to directly apply a model trained on one blast furnace to another, meaning the model's transfer and generalization capabilities are insufficient. Furthermore, existing technologies lack a unified and effective means for systematically processing heterogeneous data from different data sources with varying naming conventions, which further limits the model's versatility and robustness.

[0005] Secondly, the modeling process is inefficient and difficult to reuse. Existing systems largely rely on manual operation by engineers with domain expertise and data science skills, making it impossible to quickly and automatically build and iterate models through simple configuration. This approach is not only inefficient, but the knowledge and experience gained during the modeling process are also difficult to effectively retain and reuse, resulting in a significant amount of repetitive work for different prediction tasks.

[0006] Patent document CN108763550B discloses a blast furnace big data application system. This system has the function of integrating and uniformly representing multi-source data of blast furnaces, and includes a subsystem for extracting operating rules and establishing mechanism models. However, this patent document cannot solve the problem that blast furnaces themselves have "black box" characteristics, complex operating mechanisms and low data interpretability.

[0007] Therefore, how to effectively integrate domain knowledge to improve model accuracy and generalization ability, and to automate, standardize, and rapidly transfer the modeling process, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0008] To address the shortcomings of existing technologies, the present invention aims to provide a general modeling method and system for blast furnace parameter prediction that integrates domain knowledge, thereby solving the technical problems of poor versatility, weak generalization ability, and low modeling efficiency in existing blast furnace parameter prediction methods.

[0009] A general modeling method for predicting blast furnace parameters that integrates domain knowledge, provided by the present invention, includes: Data standardization processing steps: Based on a pre-built domain semantic model, semantic parsing and mapping of multi-source heterogeneous data in the blast furnace smelting process are performed to generate standardized structured data; Domain feature generation steps: Based on a pre-built knowledge base containing expert rules and mechanism rules, the standardized structured data is processed to generate domain features; Automated model training steps: Using a configurable automated modeling toolchain, the domain semantic model and the knowledge base are invoked according to the configured target parameters to automatically execute the model training process and output a prediction model for the target parameters.

[0010] Preferably, the domain semantic model is a knowledge graph or a domain ontology.

[0011] Preferably, the target parameters include pressure difference, molten iron temperature, molten iron silicon content, heat load, and gas flow rate.

[0012] Preferably, in the data standardization processing step, the semantic parsing is achieved by performing natural language processing on the original identification information of the data items of the multi-source heterogeneous data.

[0013] Preferably, the expert rules and mechanism rules in the knowledge base are quantified into calculation formulas to generate the domain features.

[0014] Preferably, the model training algorithm used in the automated model training step includes Lightweight Gradient Boosting Machine (LGB), Extreme Gradient Boosting (XGB), Random Forest (RF), Long Short-Term Memory (LSTM), and Multilayer Perceptron (MLP).

[0015] A general modeling system for predicting blast furnace parameters that integrates domain knowledge, provided by the present invention, is characterized by comprising: The data standardization processing module is used to perform semantic parsing and mapping on multi-source heterogeneous data in the blast furnace smelting process based on a pre-built domain semantic model, and generate standardized structured data. The domain feature generation module is used to process the standardized structured data based on a pre-built knowledge base containing expert rules and mechanism rules to generate domain features; An automated model training module is used to form a configurable automated modeling toolchain. Based on the configured target parameters, it calls the domain semantic model and the knowledge base to automatically execute the model training process and output a prediction model for the target parameters.

[0016] Preferably, the domain semantic model is a knowledge graph or a domain ontology; The target parameters include pressure difference, molten iron temperature, molten iron silicon content, heat load, and gas flow rate; The model training algorithms used in the automated model training steps include Lightweight Gradient Boosting Machine (LGB), Extreme Gradient Boosting (XGB), Random Forest (RF), Long Short-Term Memory (LSTM), and Multilayer Perceptron (MLP).

[0017] Preferably, in the data standardization processing step, the semantic parsing is achieved by performing natural language processing on the original identification information of the data items of the multi-source heterogeneous data.

[0018] Preferably, the expert rules and mechanism rules in the knowledge base are quantified into calculation formulas to generate the domain features.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention standardizes multi-source heterogeneous data by constructing a domain semantic model, shielding the differences between different blast furnaces or data sources. Combined with a flexibly configurable automated modeling toolchain, this method can be quickly transferred and applied to prediction tasks of different production lines and multiple parameters, solving the problems of poor versatility and weak generalization ability of existing methods.

[0020] 2. This invention integrates expert experience and knowledge from fields such as physical mechanisms into feature engineering in the form of calculation formulas by constructing a knowledge base. The generated domain features have clear physical meanings, effectively overcoming the "black box" problem of blast furnaces. While improving the accuracy of model prediction, it also enhances the interpretability of the model.

[0021] 3. This invention provides a standardized automated modeling toolchain that transforms the complex modeling process into a simple configuration operation, achieving standardization and automation of model training, greatly shortening the development cycle, avoiding repetitive work for different tasks, and improving modeling efficiency and the reusability of the solution. Attached Figure Description

[0022] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 A flowchart illustrating a general modeling method for predicting blast furnace parameters that integrates domain knowledge, provided in an embodiment of the present invention. Figure 2 A schematic diagram illustrating the structure of a domain semantic model (knowledge graph) provided in an embodiment of the present invention; Figure 3 A schematic diagram illustrating the knowledge base application and update cycle provided in an embodiment of the present invention; Figure 4 The system architecture diagram of the automated modeling toolchain provided in the embodiments of the present invention is shown. Detailed Implementation

[0023] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0024] This invention classifies and maps data items using a blast furnace knowledge graph and semantic segmentation, generates multiple influencing features based on expert experience and blast furnace mechanisms, and uses a toolchain to build a complete training process for the prediction model. This allows for flexible configuration on the toolchain and adjustment and retraining in the prediction of multiple parameters of the blast furnace.

[0025] Example 1 This embodiment details the complete implementation process of a general modeling method for predicting blast furnace parameters that integrates domain knowledge. Specifically, it aims to predict the "molten iron temperature" of the blast furnace and comprehensively demonstrate the core steps of the technical solution of this application, including data standardization processing, domain feature generation, and automated model training.

[0026] like Figure 1 As shown, the present invention provides a general modeling method for predicting blast furnace parameters that integrates domain knowledge, comprising: Step S101: Collect heterogeneous data from multiple sources. In a specific application scenario, the data comes from the continuous production records of a blast furnace in a steel company over the past year. This data is scattered across different business systems. For example, operational parameters such as air volume, air temperature, oxygen enrichment, and pulverized coal injection volume are obtained from the Manufacturing Execution System (MAS); a large amount of sensor measurement data, such as temperature, pressure, and cooling water temperature difference at different locations including the furnace top, throat, various sections of the furnace body, the furnace belly, and the hearth, are obtained from the Distributed Control System (DCS); and raw material data, such as the load, ash content, and moisture content of different batches of coke, and the chemical composition of sinter and pellets, are obtained from the Laboratory Information Management System (LCS). Understandably, these data sources are different, have different formats, and the accuracy and frequency of timestamps may also vary. For example, sensor data may be at the second or minute level, while raw material testing data may be at the hour or batch level.

[0027] After collecting the raw data, step S201 is executed to construct a data information table. In this step, the raw data needs to be initially integrated and organized. The system aligns and merges all data according to a preset time frequency (e.g., 1 minute). For low-frequency data (such as raw material analysis data), forward padding or interpolation can be used to fill in the gaps; for high-frequency data, averaging or sampling processing can be performed to form a high-dimensional time-series data table. Simultaneously, to effectively manage this heterogeneous data, the system creates a data information table. This data information table serves as a metadata list, recording in detail the original identification information of each column (i.e., each data item) in the high-dimensional time-series data table, such as its original column name in the source system (e.g., "BF01_TE_1101A"), Chinese remarks (e.g., "temperature of thermocouple in the middle of blast furnace body No. 1"), data type (e.g., floating-point), and unit (e.g., degrees Celsius). This data information table lays the foundation for subsequent semantic parsing and automated processes.

[0028] Subsequently, in step S301, a domain semantic model is constructed. To achieve machine-readable understanding of the physical meaning of various data items in the blast furnace domain, this embodiment constructs a domain semantic model, specifically implemented as a blast furnace knowledge graph. (See also...) Figure 2This diagram schematically illustrates the structure of a blast furnace knowledge graph. The knowledge graph centers on the blast furnace and defines related entities across multiple dimensions. For example, it defines a "location information" entity, which includes specific instances such as "furnace throat," "furnace body," "furnace waist," "furnace belly," and "furnace hearth"; a "type" entity, which includes instances such as "temperature," "pressure," "flow rate," "composition," and "weight"; and a "processing method" entity, which includes instances such as "merging similar means," "accumulation," and "difference calculation." The knowledge graph connects these entities through relationships; for example, the location information "furnace body" and the type "temperature" both point to a "standardized name," such as "SHAFT_TEMP." Furthermore, the knowledge graph defines processing methods for similar data items; for example, data from multiple furnace body temperature sensors should undergo "merging similar means" processing after standardization. This knowledge graph can be stored in a file format such as turtle, providing a semantic basis for subsequent data standardization mapping.

[0029] To incorporate the experience and mature metallurgical mechanisms of blast furnace smelting experts into the model, a knowledge base was constructed in step S401. (See also...) Figure 3 This diagram illustrates the knowledge base construction and application cycle. The knowledge base construction process includes knowledge acquisition and knowledge integration. Knowledge acquisition refers to collecting domain knowledge about blast furnace ironmaking through interviews with domain experts, consulting metallurgical handbooks and research papers, etc. This knowledge can be divided into two categories: the first is expert rules, which are qualitative or quantitative relationships extracted from the long-term operational experience of metallurgical experts, such as "the temperature of the furnace belly gas is an important indirect indicator for judging the thermal state of the hearth, usually obtained by weighted average of readings from multiple furnace belly temperature sensors"; the second is mechanistic rules, which are mathematical models or calculation formulas summarized based on the physicochemical principles (such as material balance, heat balance, and energy conservation) in the blast furnace ironmaking process, such as the calculation formula for "theoretical combustion temperature" defined according to the principle of heat balance. Knowledge integration involves quantifying and coding these rules, that is, writing each expert rule or mechanistic rule into a specific, executable calculation formula or function (for example, implemented using Python), and storing it in a dedicated tool library, which constitutes the knowledge base. For example, the calculation formula for "theoretical combustion temperature" can be expressed as:

[0030] in, The theoretical combustion temperature, The heat generated by fuel combustion The physical heat brought in by the blower and These represent the average specific heat capacity and volume of the combustion products, respectively. It should be noted that each function in Knowledge Base 43 corresponds to a calculation method with a clearly defined physical characteristic of the domain.

[0031] After completing the above preparations, the core automated modeling process begins. The process proceeds to step S501, where the data is standardized and mapped based on the domain semantic model. This step is performed by... Figure 4 The data standardization processing module in the automated modeling toolchain shown is executed. This module first loads the data information table generated in step S201 and the blast furnace knowledge graph constructed in step S301. For each row in the data information table, the module extracts its "original column name" and "Chinese remarks" information, and uses natural language processing technology to semantically analyze this text information. For example, for a data column with the remarks "absolute pressure in the middle of the furnace body," the system can use a word segmentation tool to divide it into words such as "furnace body," "middle," "absolute," and "pressure," and then use an entity recognition tool to identify that "furnace body" belongs to the "location information" entity and "pressure" belongs to the "type" entity. Next, the system queries the knowledge graph to find a "standardized name" that is associated with both "furnace body" and "pressure." The query result may be "SHAFT_PRESSURE." Considering that there may be multiple sensors measuring furnace body pressure, the system will assign a unique standardized name, such as "SHAFT_PRESSURE_001," according to the numbering rules defined in the knowledge graph. Simultaneously, the system also retrieves recommended "processing methods" for this type of data from the knowledge graph, such as "merging the mean of similar data." Finally, the system writes this standardized name and processing method back into the data information table, completing the semantic mapping of the data item. By performing this process on all data items, the original multi-source heterogeneous data is transformed into unified, standardized, and structured data.

[0032] Accordingly, step S502 is executed, invoking the knowledge base to generate domain features. This step is performed by... Figure 4The feature engineering module, deeply integrated with a knowledge base, executes the calculation. Upon receiving standardized structured data, the feature engineering module iterates through all predefined rule functions in the knowledge base. For each function, the module checks if the input parameters (which are also standardized names) required for its calculation exist in the current standardized structured data. If they do, the module automatically calls the function to perform the calculation, generating a new feature column. For example, the module calls the function "Calculate the furnace gas temperature," which, according to its internal logic, finds all standardized data columns named "BELLY_TEMP_XXX," performs a weighted average, and generates a new feature named "FEATURE_BELLY_GAS_TEMP." Similarly, the module calls the function "Calculate the theoretical combustion temperature," obtains parameters such as air supply temperature and coke calorific value from the standardized data according to the formula, and calculates a new feature named "FEATURE_THEORETICAL_COMBUSTION_TEMP." These new features generated through the knowledge base have clear physical meanings and effectively supplement and deepen the original data. They are merged with the original standardized data to form the final feature set used for model training.

[0033] Subsequently, in step S503, the toolchain is configured and run for model training. (See also...) Figure 4 Users interact with the automated modeling toolchain through a configuration interface. In this embodiment, the user performs the following configuration: 1. In the data access module 31, specify the input data as the feature set containing domain features processed in the previous steps; 2. In the configuration items, specify the prediction target (i.e., prediction label) for this task as the standardized name of "molten iron temperature," such as "HM_TEMP"; 3. In the algorithm selection module, select the "Lightweight Gradient Boosting Machine (LGB)" algorithm from the list of supported algorithms. It is understood that this toolchain supports multiple algorithms, including but not limited to Extreme Gradient Boosting (XGB), Random Forest (BF), and Long Short-Term Memory (LSTM), allowing users to flexibly choose according to task characteristics.

[0034] After configuration, the user starts the toolchain, which then automatically executes the subsequent processes. The data preprocessing module performs standard operations such as missing value imputation and outlier handling. In addition to generating domain features, the feature engineering module may also perform feature filtering to remove redundant or irrelevant features. Afterward, the core model training process begins. The toolchain uses the configured lightweight gradient boosting machine algorithm and the prepared feature and label sets to train the model, automatically performing hyperparameter tuning during the process.

[0035] After model training is complete, the prediction model is output in step S504. The evaluation module uses a reserved test set to evaluate the performance of the trained model, calculating multiple metrics such as mean absolute error and root mean square error, and displays the evaluation results to the user. If the user is satisfied with the model performance, the toolchain will package the trained optimal model and related metadata (such as the list of features used, model parameters, etc.) as model output and save it to the model library for subsequent online prediction calls. At this point, a high-precision prediction model for "molten iron temperature" has been automatically built.

[0036] Example 2 This embodiment aims to demonstrate the versatility and efficiency of the proposed solution. Specifically, it illustrates how, based on Embodiment 1, a completely new model for predicting "silicon content in molten iron" can be quickly constructed simply by modifying the configuration, and the core training algorithm has been replaced.

[0037] As an optional implementation, the basic environment of this embodiment is exactly the same as that of Embodiment 1. That is to say, the basic resources used, such as blast furnace production data, constructed data information tables, blast furnace knowledge graph, and knowledge base, can be directly reused without any repetitive development work. This reflects the advantages of this application in terms of knowledge accumulation and reuse.

[0038] The main difference from Example 1 lies in step S503, which is the stage of configuring and running the toolchain for model training. See again... Figure 4 Users can make the following two simple modifications in the toolchain configuration interface: First, modify the prediction target parameters. Users change the prediction label configuration item from "HM_TEMP" (molten iron temperature) to a standardized name representing "molten iron silicon content," such as "HM_SI_CONTENT." Molten iron silicon content is another crucial parameter for blast furnace production, and its variation pattern differs from that of molten iron temperature.

[0039] Secondly, the model training algorithm was changed. Considering that changes in the silicon content of molten iron often exhibit stronger long-term temporal dependencies, traditional tree models may not be able to fully capture these dependencies. Therefore, in the algorithm selection module, users can change the model algorithm from "Lightweight Gradient Boosting Machine" to a deep learning algorithm more suitable for processing time-series data—"Long Short-Term Memory Network." The toolchain in this application incorporates several mainstream algorithms, making such switching operations extremely convenient.

[0040] After completing the above configuration modifications, the user restarts the toolchain. The toolchain's execution flow is highly similar to that of Example 1, but the internal processing details differ. The data standardization mapping in step S501 and the knowledge base generation of domain features in step S502 are completely identical, as these steps are task-independent. The knowledge base may already contain expert rules or mechanistic rules related to the silicon content of molten iron (e.g., calculation formulas based on the silicon reduction reaction kinetic model). These rules will be automatically activated and invoked in this task, thereby generating domain features strongly correlated with silicon content prediction.

[0041] The core difference occurs during the model training phase. The general model training module builds and trains a Long Short-Term Memory (LSTM) network model based on the new configuration, instead of a gradient boosting tree model. This may involve serializing the data to fit the LTM network's input format. After training, the evaluation module also performs performance evaluation on the newly generated iron-silicon content prediction model and finally outputs a model file specifically for predicting iron-silicon content in step S504.

[0042] As can be seen from this embodiment, the technical solution provided in this application has extremely high flexibility and modeling efficiency. Faced with a completely new prediction task (from predicting temperature to predicting silicon content), users do not need to write any new code, nor do they need to perform tedious data processing and feature engineering design again. They can automatically generate a completely new, high-performance prediction model in a short time simply by making simple configuration switches on a graphical interface. This greatly shortens the model development cycle and enables rapid migration and reuse of the solution.

[0043] Example 3 This embodiment aims to illustrate another optional implementation of the "domain semantic model" in the technical solution of this application, to demonstrate the broad applicability of its concept, thereby supporting a wider scope of protection. In Embodiment 1, the domain semantic model is specifically implemented as a knowledge graph; while in this embodiment, a domain ontology based on a web ontology language will be used to achieve the same data standardization goal.

[0044] The overall process in this embodiment still follows... Figure 1 The steps shown are mostly the same as in Example 1. The core difference lies in the specific implementation of steps S301 (constructing a domain semantic model) and S501 (standardizing and mapping data based on the domain semantic model).

[0045] In step S301, instead of constructing a knowledge graph, this embodiment uses a professional ontology editing tool to construct a formalized and logically rigorous ontology of the blast furnace domain. This ontology defines a series of "classes," such as "sensor," "physical location," "measurement parameter," and "equipment component." Simultaneously, it defines "object attributes" between classes, such as "located in," "measured," and "belongs to." Furthermore, it defines "data attributes" to describe the specific characteristics of instances, such as "tag name" and "unit."

[0046] Specifically, a particular temperature sensor located in the middle of the blast furnace can be described in the ontology as an instance of the "Thermometer" class. This instance is associated with an instance of the "Physical Location" class (located in the middle of the furnace) through the object attribute "located in," and with an instance of the "Temperature" class (measurement parameter) through the object attribute "measured." In this way, the domain ontology can rigorously describe the concepts, terminology, and complex relationships between them in a more reasoning-oriented manner than a knowledge graph.

[0047] Accordingly, the internal mechanism changes when performing data standardization mapping in step S501. The system no longer performs graph queries based on string matching as in Example 1, but instead performs logical reasoning and instance lookup in the domain ontology by executing a formal ontology query language.

[0048] When processing a data item labeled "absolute pressure in the middle of the furnace body," the system first extracts key entities such as "furnace body" and "pressure" using natural language processing technology. Then, the system generates an ontology query, the logic of which can be expressed as: "Search for a 'sensor' instance in the ontology that simultaneously meets the following two conditions: 1. It is associated with a 'physical location' instance related to 'furnace body' via the 'located' attribute; 2. It is associated with a 'measurement parameter' instance related to 'pressure' via the 'measurement' attribute." After the ontology inference engine executes this query, it returns a specific sensor instance that meets the conditions. The system then extracts a predefined normalized name (e.g., "SHAFT_PRESSURE_001") from the data attributes of that instance, thus completing the mapping.

[0049] Once the mapping from data items to standardized names is completed through the domain ontology, subsequent steps, such as step S502 calling the knowledge base to generate domain features and step S503 configuring and running the toolchain for model training, are completely consistent with Example 1. The dedicated knowledge application module of the automated modeling toolchain is adapted to interact with the query interface of the domain ontology, but for the rest of the toolchain, its input (standardized data) and processing logic remain unchanged.

[0050] This embodiment demonstrates that the core idea of ​​this application—using a structured semantic model to automatically parse and standardize heterogeneous data—is not limited to the specific technology of knowledge graphs. Whether using knowledge graphs or domain ontology, the same technical effect can be achieved: shielding the heterogeneity of the underlying data and providing a standardized data foundation for upper-level general modeling. Therefore, this embodiment provides strong evidence for the universality and advancement of the present invention.

[0051] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0052] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features of the present invention can be arbitrarily combined with each other.

Claims

1. A general modeling method for predicting blast furnace parameters that integrates domain knowledge, characterized in that, include: Data standardization processing steps: Based on a pre-built domain semantic model, semantic parsing and mapping of multi-source heterogeneous data in the blast furnace smelting process are performed to generate standardized structured data; Domain feature generation steps: Based on a pre-built knowledge base containing expert rules and mechanism rules, the standardized structured data is processed to generate domain features; Automated model training steps: Using a configurable automated modeling toolchain, the domain semantic model and the knowledge base are invoked according to the configured target parameters to automatically execute the model training process and output a prediction model for the target parameters.

2. The general modeling method for predicting blast furnace parameters by integrating domain knowledge as described in claim 1, characterized in that, The domain semantic model is a knowledge graph or a domain ontology.

3. The general modeling method for predicting blast furnace parameters by integrating domain knowledge as described in claim 1, characterized in that, The target parameters include pressure difference, molten iron temperature, molten iron silicon content, heat load, and gas flow rate.

4. The general modeling method for predicting blast furnace parameters by integrating domain knowledge as described in claim 1, characterized in that, In the data standardization process, the semantic parsing is achieved by performing natural language processing on the original identification information of the data items in the multi-source heterogeneous data.

5. The general modeling method for predicting blast furnace parameters by integrating domain knowledge according to claim 1, characterized in that, The expert rules and mechanism rules in the knowledge base are quantified into calculation formulas to generate the domain features.

6. The general modeling method for predicting blast furnace parameters by integrating domain knowledge according to claim 1, characterized in that, The model training algorithms used in the automated model training steps include Lightweight Gradient Boosting Machine (LGB), Extreme Gradient Boosting (XGB), Random Forest (RF), Long Short-Term Memory (LSTM), and Multilayer Perceptron (MLP).

7. A general modeling system for predicting blast furnace parameters that integrates domain knowledge, characterized in that, include: The data standardization processing module is used to perform semantic parsing and mapping on multi-source heterogeneous data in the blast furnace smelting process based on a pre-built domain semantic model, and generate standardized structured data. The domain feature generation module is used to process the standardized structured data based on a pre-built knowledge base containing expert rules and mechanism rules to generate domain features; An automated model training module is used to form a configurable automated modeling toolchain. Based on the configured target parameters, it calls the domain semantic model and the knowledge base to automatically execute the model training process and output a prediction model for the target parameters.

8. The general modeling system for predicting blast furnace parameters by integrating domain knowledge according to claim 7, characterized in that, The domain semantic model is a knowledge graph or a domain ontology; The target parameters include pressure difference, molten iron temperature, molten iron silicon content, heat load, and gas flow rate; The model training algorithms used in the automated model training steps include Lightweight Gradient Boosting Machine (LGB), Extreme Gradient Boosting (XGB), Random Forest (RF), Long Short-Term Memory (LSTM), and Multilayer Perceptron (MLP).

9. The general modeling system for predicting blast furnace parameters by integrating domain knowledge according to claim 7, characterized in that, In the data standardization process, the semantic parsing is achieved by performing natural language processing on the original identification information of the data items in the multi-source heterogeneous data.

10. The general modeling system for predicting blast furnace parameters by integrating domain knowledge according to claim 7, characterized in that, The expert rules and mechanism rules in the knowledge base are quantified into calculation formulas to generate the domain features.

Images