Search

CN-122021350-A - Smelting process intelligent optimization method and system based on multi-source data fusion

CN122021350ACN 122021350 ACN122021350 ACN 122021350ACN-122021350-A

Abstract

The application relates to a smelting process intelligent optimization method and system based on multi-source data fusion, wherein the method comprises the steps of obtaining target process parameters, real-time sensor data and production process data; and dividing thermodynamic and chemical component parameter sets according to characteristics, constructing corresponding sub-maps and fusing the sub-maps into a process knowledge map. Preprocessing and aligning time sequences, and constructing a comprehensive smelting state space model. And extracting a key state vector, inputting the key state vector into a hierarchical reinforcement learning framework, and coordinating a double-network decoupling decision by the framework according to the parameter priority and the coupling relation to generate a collaborative optimization action set. And calling the knowledge graph to carry out multi-target constraint solving and conflict resolution, and generating an executable instruction set. Executing the instruction, collecting feedback data, calculating a differential rewarding value, and updating frame parameters and map dynamic weight edges. According to the application, through multi-source data fusion and layered reinforcement learning collaborative optimization, dynamic adjustment and accurate control of smelting process parameters are realized.

Inventors

  • YANG JING

Assignees

  • 北京浩德天工新材料科技有限公司

Dates

Publication Date
20260512
Application Date
20260410

Claims (10)

  1. 1. The intelligent optimization method for the smelting process based on multi-source data fusion is characterized by comprising the following steps of: acquiring target technological parameters of steel grades to be smelted, real-time sensor data and production process data in the smelting process; dividing the target process parameters into a thermodynamic parameter set and a chemical component parameter set according to parameter characteristics, respectively constructing a thermodynamic constraint sub-map and a chemical component associated sub-map, and fusing to generate a process knowledge map; preprocessing the real-time sensor data and the production process data and aligning the time sequence, and constructing a comprehensive smelting state space model of the current heat; Extracting a key state vector from the comprehensive smelting state space model, and inputting the key state vector into a preset hierarchical reinforcement learning optimization framework, wherein the hierarchical reinforcement learning optimization framework comprises a thermodynamic strategy network and a chemical composition strategy network; The hierarchical reinforcement learning optimization framework coordinates the thermodynamic strategy network and the chemical component strategy network to carry out decoupling decision based on the parameter priority and the coupling relation in the target process parameters, and generates a collaborative optimization action set; Invoking the process knowledge graph, and carrying out multi-objective constraint solving and conflict resolution on the collaborative optimization action set to generate an executable optimization instruction set; Executing the executable optimized instruction set, and synchronously collecting multi-dimensional feedback detection data in the executing process; And calculating differential rewards of each parameter dimension based on the multi-dimension feedback detection data, and respectively updating network parameters of the hierarchical reinforcement learning optimization framework and dynamic weight edges in a process knowledge graph by using the differential rewards.
  2. 2. The intelligent optimization method for the smelting process based on multi-source data fusion according to claim 1, wherein the target process parameters are divided into thermodynamic parameter sets and chemical component parameter sets according to parameter characteristics, thermodynamic constraint sub-patterns and chemical component associated sub-patterns are respectively constructed, and the process knowledge patterns are generated by fusion comprising the following steps: Classifying parameters related to energy state changes as thermodynamic parameter sets and parameters related to material composition changes as chemical component parameter sets based on physicochemical properties of the target process parameters; according to the thermodynamic parameter set, defining a nonlinear constraint relation between temperature and oxygen activity, and generating a thermodynamic constraint sub-map; Defining a reaction coupling relation among alloy elements according to the chemical component parameter set to generate a chemical component association sub-map; And connecting the thermodynamic constraint sub-map and the chemical component related sub-map by the principle of conservation of material energy, and fusing to generate a process knowledge map.
  3. 3. The intelligent optimization method of a smelting process based on multi-source data fusion according to claim 2, wherein after the step of performing edge connection on the thermodynamic constraint sub-map and the chemical component-associated sub-map by a substance energy conservation rule and generating a process knowledge map by fusion, the method further comprises: acquiring a preset local expert domain knowledge base comprising element reaction coupling relations and safety boundary parameters; Mapping the element reaction coupling relation to the associated side of a chemical component associated sub-graph in the process knowledge graph, and mapping the safety boundary parameter to a constraint node of a thermodynamic constraint sub-graph in the process knowledge graph; And updating the initial connection strength of the dynamic weight edge of the process knowledge graph based on the mapping result, and generating an updated process knowledge graph.
  4. 4. The intelligent optimization method for smelting process based on multi-source data fusion according to claim 1, wherein the steps of preprocessing and time-series aligning the real-time sensor data and the production process data and constructing the comprehensive smelting state space model of the current heat comprise: Identifying the acquisition time stamp of each detection parameter in the real-time sensor data, and establishing a time-lag compensation mapping relation based on response delay characteristics of smelting equipment; performing time axis alignment on the temperature data stream and the oxygen activity data stream according to the time lag compensation mapping relation to generate a synchronous sensor data sequence; extracting dynamic process feature values from the synchronous sensor data sequence, wherein the dynamic process feature values comprise a temperature rise rate gradient, an alloy element diffusion rate and a deoxidization reaction efficiency; carrying out multidimensional tensor fusion on the synchronous sensor data sequence, the dynamic process characteristic quantity and the raw material proportioning information in the production process data; Constructing a smelting state space coordinate system containing time-varying characteristics based on a tensor dimension reduction algorithm, and mapping the feature vector of the multidimensional tensor in the smelting state space coordinate system; And generating a comprehensive smelting state space model of the current heat according to the distribution topological relation of the characteristic vectors.
  5. 5. The intelligent optimization method for a smelting process based on multi-source data fusion according to claim 3, wherein the step of coordinating the thermodynamic strategy network and the chemical composition strategy network to perform decoupling decision based on the parameter priority and the coupling relation in the target process parameters by the hierarchical reinforcement learning optimization framework, and generating the collaborative optimization action set comprises the following steps: Analyzing the priority weights of the thermodynamic parameter and the chemical component parameter of the target technological parameter to generate a parameter decision weight matrix; The thermodynamic action sequences output by the thermodynamic strategy network are subjected to priority ordering according to the parameter decision weight matrix, and the execution sequence of the basic power adjustment action and the basic deoxidization action is determined; Based on the element reaction coupling relation in the chemical component association sub-map, performing conflict marking on the chemical component action sequence output by the chemical component strategy network to obtain a conflict marking result; Invoking a nonlinear constraint relation in the thermodynamic constraint sub-map, calculating interference influence factors of the thermodynamic action sequences after priority ordering on chemical component action sequences with collision marks, and carrying out feasibility verification by combining the interference influence factors with collision mark results; if the interference influence factor exceeds a preset threshold or the feasibility verification is not passed, adjusting the output of the chemical component strategy network by using an attention mechanism, and generating a verified chemical component action sequence; performing space-time dimension matching on the checked chemical component action sequence and the thermodynamic action sequence with the ordered priority, and generating an initial cooperative action set; and according to the physical execution window constraint of the smelting process, carrying out time sequence rearrangement on the initial collaborative action set to generate a collaborative optimization action set.
  6. 6. The intelligent optimization method for a smelting process based on multi-source data fusion according to claim 5, wherein before the step of coordinating the thermodynamic strategy network and the chemical composition strategy network to perform decoupling decision based on the parameter priority and the coupling relation in the target process parameters by the hierarchical reinforcement learning optimization framework to generate a collaborative optimization action set, the method further comprises: Inputting the key state vector into a preset working condition identification classifier, identifying the type of the smelting stage where the current heat is located, and generating a smelting stage type label comprising a melting stage, an oxidation stage and a reduction stage; based on the smelting stage type label, searching matched historical optimization case nodes in the process knowledge graph, and acquiring a historical state feature vector and a historical optimization action sequence associated with the historical optimization case nodes to form a historical successful optimization case set; Calculating cosine similarity of the key state vector and each historical state feature vector in the historical successful optimization case set, and selecting a historical optimization case node with the maximum cosine similarity as a target reference case; Extracting the historical optimization action sequence from the target reference case, and converting the historical optimization action sequence into a priori strategy parameter matrix; performing initial parameter configuration on a strategy network of the hierarchical reinforcement learning optimization framework by using the prior strategy parameter matrix to generate an initialized hierarchical reinforcement learning optimization framework; and executing the step of coordinating the thermodynamic strategy network and the chemical composition strategy network to carry out decoupling decision based on the initialized hierarchical reinforcement learning optimization framework.
  7. 7. The intelligent optimization method of smelting process based on multi-source data fusion according to claim 1, wherein the step of calling the process knowledge graph, performing multi-objective constraint solving and conflict resolving on the collaborative optimization action set, and generating an executable optimization instruction set comprises: Extracting rule sides of thermodynamic constraint sub-graphs and associated sides of chemical component associated sub-graphs from the process knowledge graph to generate a multi-target constraint rule set; based on the multi-objective constraint rule set, carrying out feasibility evaluation on each action unit in the collaborative optimization action set, and identifying conflict action units; Carrying out resolution strategy matching on the conflict action units according to the dynamic weight edge priority in the process knowledge graph; redefining or eliminating the conflict action units by applying the resolution strategy to generate a conflict resolved action sequence; and carrying out suitability check on the action sequence subjected to conflict resolution and the physical execution capacity of smelting equipment to generate an executable optimization instruction set.
  8. 8. The intelligent optimization method of smelting process based on multi-source data fusion according to claim 1, wherein the step of calculating differential rewards values of each parameter dimension based on the multi-dimensional feedback detection data and using the differential rewards values to update the network parameters of the hierarchical reinforcement learning optimization framework and the dynamic weight edges in the process knowledge graph respectively comprises the following steps: Extracting actual measured values of each parameter dimension and expected values of target process parameters from the multi-dimensional feedback detection data, and calculating parameter deviation values; generating a differential rewarding value of each parameter dimension based on the parameter deviation and a preset rewarding function template; Performing counter propagation gradient calculation on a thermodynamic strategy network and a chemical component strategy network of the hierarchical reinforcement learning optimization framework according to the differential rewarding value, and updating network weight parameters; Carrying out importance evaluation on dynamic weight edges in the process knowledge graph based on the differential rewards value to generate a weight adjustment coefficient; And updating the connection strength of the dynamic weight edge by using the weight adjustment coefficient to generate an updated process knowledge graph.
  9. 9. A smelting process intelligent optimization system based on multi-source data fusion is characterized in that the optimization system comprises: The multi-source data acquisition module is used for acquiring target technological parameters of the steel grade to be smelted, and real-time sensor data and production process data in the smelting process; The process knowledge graph construction module is used for dividing the target process parameters into a thermodynamic parameter set and a chemical component parameter set according to parameter characteristics, respectively constructing a thermodynamic constraint sub-graph and a chemical component associated sub-graph, and fusing to generate a process knowledge graph; The smelting state space modeling module is used for preprocessing the real-time sensor data and the production process data and aligning the time sequence to construct a comprehensive smelting state space model of the current heat; the strategy network driving module is used for extracting a key state vector from the comprehensive smelting state space model and inputting the key state vector into a preset hierarchical reinforcement learning optimization framework, wherein the hierarchical reinforcement learning optimization framework comprises a thermodynamic strategy network and a chemical composition strategy network; The multi-strategy collaborative decision module is used for controlling the hierarchical reinforcement learning optimization framework to coordinate the thermodynamic strategy network and the chemical composition strategy network to carry out decoupling decision based on the parameter priority and the coupling relation in the target process parameters so as to generate a collaborative optimization action set; the multi-objective constraint solving module is used for calling the process knowledge graph, and carrying out multi-objective constraint solving and conflict resolution on the collaborative optimization action set to generate an executable optimization instruction set; The execution feedback module is used for executing the executable optimization instruction set and synchronously collecting multidimensional feedback detection data in the execution process; And the optimization updating module is used for calculating differential rewards of each parameter dimension based on the multi-dimension feedback detection data, and respectively updating the network parameters of the hierarchical reinforcement learning optimization framework and the dynamic weight edges in the process knowledge graph by utilizing the differential rewards.
  10. 10. A computer-readable storage medium, characterized in that a computer program is stored which can be loaded by a processor and which performs the method according to any one of claims 1 to 8.

Description

Smelting process intelligent optimization method and system based on multi-source data fusion Technical Field The application relates to the technical field of intelligent optimization control of industrial processes, in particular to an intelligent optimization method and system for a smelting process based on multi-source data fusion. Background With the continuous development of intelligent manufacturing technology, the steel industry is accelerating transformation towards the direction of greenization, high efficiency and intellectualization as an important foundation of national economy. In the process, the accurate control and optimization of the smelting core process are realized, and the method is a key for improving the quality of products, reducing the production cost and ensuring the production safety and stability. In recent years, the sensing and connecting technology represented by an online detection sensor, the Internet of things and big data provides a hardware basis for acquiring key parameters such as bath temperature, oxygen activity, chemical components and the like in real time, and the artificial intelligence technology represented by a machine learning and optimizing algorithm provides a new theoretical tool for mining knowledge from massive data and searching an optimal process path. The development of these technologies together constitute an external condition that drives the evolution of the smelting process from empirical to model and data driven. However, in the practical application process, because the smelting process is a dynamic system with multiple variables, strong coupling, large time lags and complex physicochemical reactions, the traditional optimization model based on a single mechanism or single data drive is often difficult to coordinate and process the intricate and complex interaction influence between thermodynamic state control and chemical component control, and the deep coupling mechanism between the thermodynamic state control and the chemical component control is ignored, so that the phenomenon that the thermodynamic state control and the chemical component control are out of consideration easily occurs in the control process, and global optimization is difficult to realize. In addition, the optimization suggestions generated by the existing data-driven optimization model lack interpretability, and are difficult to strictly ensure to meet the process safety specifications, so that the floor application of the optimization suggestions in high-safety-requirement industrial sites is limited. Disclosure of Invention In order to solve the technical problems, the application provides an intelligent optimization method and system for a smelting process based on multi-source data fusion. In a first aspect, the application provides an intelligent optimization method for a smelting process based on multi-source data fusion, which adopts the following technical scheme: acquiring target technological parameters of steel grades to be smelted, real-time sensor data and production process data in the smelting process; dividing the target process parameters into a thermodynamic parameter set and a chemical component parameter set according to parameter characteristics, respectively constructing a thermodynamic constraint sub-map and a chemical component associated sub-map, and fusing to generate a process knowledge map; preprocessing the real-time sensor data and the production process data and aligning the time sequence, and constructing a comprehensive smelting state space model of the current heat; Extracting a key state vector from the comprehensive smelting state space model, and inputting the key state vector into a preset hierarchical reinforcement learning optimization framework, wherein the hierarchical reinforcement learning optimization framework comprises a thermodynamic strategy network and a chemical composition strategy network; The hierarchical reinforcement learning optimization framework coordinates the thermodynamic strategy network and the chemical component strategy network to carry out decoupling decision based on the parameter priority and the coupling relation in the target process parameters, and generates a collaborative optimization action set; Invoking the process knowledge graph, and carrying out multi-objective constraint solving and conflict resolution on the collaborative optimization action set to generate an executable optimization instruction set; Executing the executable optimized instruction set, and synchronously collecting multi-dimensional feedback detection data in the executing process; And calculating differential rewards of each parameter dimension based on the multi-dimension feedback detection data, and respectively updating network parameters of the hierarchical reinforcement learning optimization framework and dynamic weight edges in a process knowledge graph by using the differential rewards. By adopting the technical scheme