Molten iron operation plan arrangement method based on dynamic constraint modeling hybrid optimization

Abstract

The invention relates to the technical field of steel production plan optimization based on multi-dimensional constraint dynamic coupling, in particular to a molten iron operation plan arrangement method based on dynamic constraint modeling hybrid optimization, which comprises the following steps: acquiring order data, equipment state data and constraint rule data to generate structured input parameters comprising order priority labels; constructing a multi-objective optimization model based on the molten iron distribution constraint and the maintainable time length of the equipment, extracting the maintenance conflict relationship between the equipment groups, constructing a maintenance rule knowledge graph, and generating a dynamic constraint condition set; generating an initial molten iron distribution and maintenance plan through global search by adopting a genetic algorithm, and outputting a molten iron distribution path and an equipment maintenance time sequence through local optimization by combining a linear programming algorithm; according to the method, dynamic collaborative optimization of molten iron distribution and equipment maintenance is achieved, and the resource utilization rate, the order delivery rate and the production system stability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of steel production plan optimization based on multi-dimensional constraint dynamic coupling, and in particular to a molten iron operation plan arrangement method based on dynamic constraint modeling and hybrid optimization. Background Art

[0002] Under conditions of scarce molten iron resources, in the collaborative production of multiple processes at existing steel companies, the molten iron distribution and process connection of equipment such as blast furnaces, continuous casting machines, and rolling mills lack a global optimization strategy based on dynamic demand response, resulting in the risk of an imbalance between supply and demand between processes between molten iron scheduling and production planning. The technical pain point lies in the fact that the existing production planning system lacks a multi-objective real-time collaborative decision-making model that integrates order requirements, equipment status, and molten iron supply constraints, making it difficult to achieve dynamic coupled optimization of molten iron consumption and production line maintenance through intelligent algorithms. For example, when order demand suddenly changes, the existing manual scheduling model cannot quickly establish a mapping relationship between the molten iron supply path and the production sequence of the continuous casting machine, which can easily lead to a mismatch between the blast furnace's iron-making rhythm and the continuous casting pouring times, resulting in production disturbances such as molten iron backlogs or interruptions. Summary of the Invention

[0003] In response to the shortcomings of the existing technology, the present invention provides a molten iron operation planning method based on dynamic constraint modeling and hybrid optimization. The present invention solves the problem in the existing steel production planning system that the dynamic allocation of molten iron and the production line maintenance plan are difficult to synchronously respond to changes in order demand due to the lack of a real-time collaborative optimization model that integrates multiple source constraints.

[0004] In order to solve the above technical problems, the specific technical solutions of the present invention are as follows:

[0005] The present invention provides a method for scheduling molten iron operations based on dynamic constraint modeling and hybrid optimization, comprising:

[0006] Step 1: Collect order data, device status data, and constraint rule data, filter outliers and mark priorities on the order data, and generate structured input parameters including order priority tags;

[0007] Step 2: Based on the molten iron distribution constraints and the equipment maintenance time in the structured input parameters, a multi-objective optimization model is constructed. The molten iron distribution constraints are converted into dynamic constraints including mutually exclusive operation rules of the continuous casting machine, and the equipment maintenance time is converted into a maintenance period continuity constraint. The decision variable set for the molten iron distribution amount, equipment start and stop status, and maintenance period is set;

[0008] Step 3: Input the decision variable set into the genetic algorithm for global search to generate the initial molten iron distribution and maintenance plan. Input the initial molten iron distribution and maintenance plan into the linear programming algorithm for local optimization adjustment, and output the optimized molten iron distribution path and equipment maintenance sequence.

[0009] Step 4: Execute the optimized molten iron distribution path and equipment maintenance sequence, monitor the time deviation between the blast furnace iron-making rhythm and the continuous casting machine casting in real time, and trigger the casting re-arrangement algorithm to generate a revised casting sequence when the time deviation exceeds the preset threshold. The revised casting sequence and scheduling abnormality data are fed back to the data collection step, and the structured input parameters are updated to reconstruct the multi-objective optimization model.

[0010] Furthermore, in the hot metal operation scheduling method based on dynamic constraint modeling and hybrid optimization of the present invention, the order data includes:

[0011] Extract total order volume and delivery cycle data from the ERP system, and perform outlier filtering on the extracted order data to generate a valid order data set;

[0012] Input the valid order dataset into the order priority rule library, and label each order with an urgency tag based on the customer level and contract penalty clauses;

[0013] The annotated urgency labels are used as components of the structured input parameters and fed into the multi-objective optimization model.

[0014] Furthermore, in the molten iron operation scheduling method based on dynamic constraint modeling and hybrid optimization of the present invention, the constraint rule data includes:

[0015] Map the upper and lower limits of molten iron consumption and the mutually exclusive operation rules of the continuous casting machine into a two-dimensional matrix to generate a matrix constraint table;

[0016] Extracting maintenance period conflicts between equipment groups based on equipment group identifiers in equipment maintenance records and pre-stored maintenance plans;

[0017] Based on the conflict relationship between maintenance periods, a maintenance rule knowledge graph including maintenance dependencies between equipment groups is constructed;

[0018] The matrix constraint table and maintenance rule knowledge graph are used as physical rule constraints and input into the multi-objective optimization model.

[0019] Furthermore, in the method for scheduling molten iron operations based on dynamic constraint modeling and hybrid optimization of the present invention, step 2 includes:

[0020] Based on the collected equipment status data, the mutually exclusive operation rules of the continuous casting machine are extracted and converted into 0-1 integer constraints to limit the operation time of the equipment in the group to no overlap;

[0021] Generate a maintenance period continuity constraint function based on the minimum maintenance time requirement in the collected equipment maintenance data;

[0022] Based on the collected real-time data on molten iron production, the molten iron supply fluctuation index is calculated. The molten iron allocation weight factor in the matrix constraint table is dynamically adjusted according to the molten iron supply fluctuation index. When an urgency tag upgrade is detected, the priority coefficient of the corresponding production line is increased.

[0023] The 0-1 integer constraint, the maintenance period continuity constraint function, and the adjusted molten iron distribution weight factor and priority coefficient are combined to generate a dynamic constraint condition set.

[0024] Furthermore, in the method for scheduling molten iron operations based on dynamic constraint modeling and hybrid optimization of the present invention, step 3 includes:

[0025] Real number coding is used to generate the initial population that represents the molten iron distribution amount, equipment start-up and shutdown status, and maintenance period;

[0026] Based on the defined decision variable set of molten iron distribution, equipment start-up and shutdown status, and maintenance period, the initial population is generated using real number coding.

[0027] Based on the order delivery rate weight factor and equipment utilization rate weight factor in the objective function of the multi-objective optimization model, a fitness function is constructed. The fitness function calculates the comprehensive evaluation value of molten iron consumption, order delivery rate and equipment utilization rate through weighted summation;

[0028] For the violation data of the molten iron distribution constraint and the 0-1 integer constraint in the matrix constraint table, set the penalty term and add it to the comprehensive evaluation value to generate the final fitness value;

[0029] The individual parameter values ​​with the best fitness evaluation are converted into the input parameter format of the linear programming algorithm and input into the linear programming algorithm as the initial solution for local optimization adjustment.

[0030] Furthermore, the molten iron operation scheduling method based on dynamic constraint modeling and hybrid optimization of the present invention, the optimized molten iron distribution path and equipment maintenance sequence include:

[0031] Based on the generated equipment maintenance time sequence, a time deviation threshold between the blast furnace tapping rhythm and the continuous casting machine pouring time is established, and the time sequence offset is calculated in real time;

[0032] When the risk of molten iron interruption is detected, the backup converter molten iron cache resources in the collected equipment status data are called up, and the yield rate is downgraded for the production line marked with the lowest urgency level;

[0033] The impact weight on order delivery is calculated using the preset yield rate compensation model based on the production capacity loss data after the yield rate degradation operation and the order urgency tag.

[0034] The impact weights are input into the dynamic constraint condition set, and the molten iron distribution weight factors and priority coefficients are updated.

[0035] Furthermore, in the molten iron operation scheduling method based on dynamic constraint modeling and hybrid optimization of the present invention, step 4 includes:

[0036] The generated scheduling exception data is stored in the knowledge base associated with the structured input parameters, triggering the reconstruction of the constraints of the multi-objective optimization model and the update of the dynamic constraint set;

[0037] An incremental optimization algorithm is used to solve the updated multi-objective optimization model. The incremental optimization algorithm dynamically adjusts the crossover probability and mutation probability of the genetic algorithm based on the collected real-time data of the equipment status and the initial population parameters of the genetic algorithm.

[0038] The revised molten iron distribution path and equipment maintenance sequence are generated through the incremental optimization algorithm, and the revised molten iron distribution path and maintenance sequence are input into the preset production line scheduling system.

[0039] Beneficial effects of the present invention:

[0040] The present invention constructs an order priority matrix, a matrix constraint table and a maintenance rule knowledge graph through multi-source data fusion, generates a dynamic constraint condition set and drives a multi-objective optimization model, and realizes the global-local collaborative optimization of the molten iron distribution path and the equipment maintenance sequence by combining the hybrid optimization strategy of the genetic algorithm and the linear programming algorithm; triggers the pouring interruption risk response mechanism by real-time monitoring of the blast furnace iron-tapping rhythm and the continuous casting machine pouring deviation, calls spare resources and performs the yield degradation operation, and dynamically updates the allocation weight factor and priority coefficient in combination with the yield compensation model; adjusts the genetic parameters and reconstructs the constraints based on the incremental optimization algorithm, forming a closed-loop iterative process of data collection, model optimization, and execution feedback, effectively improving the dynamic allocation accuracy of molten iron resources and the adaptability of equipment maintenance plans, solving the problem of imbalance between supply and demand in multiple processes, and enhancing the timeliness of order delivery and the stability of the production system. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, for ordinary technicians in this field, other drawings can be obtained based on the drawings without paying any creative labor.

[0042] Figure 1 A flowchart of a method for scheduling molten iron operations based on dynamic constraint modeling and hybrid optimization provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. The technical solutions provided by each embodiment of the present invention are described in detail below in conjunction with the drawings. In order to better understand the purpose of the present invention, the present invention is further described in detail below.

[0044] See also Figure 1 The present invention provides a method for scheduling molten iron operations based on dynamic constraint modeling and hybrid optimization, comprising:

[0045] Step 1: Collect order data, device status data, and constraint rule data, filter outliers and mark priorities on the order data, and generate structured input parameters including order priority tags;

[0046] Step 2: Based on the molten iron distribution constraints and the equipment maintenance time in the structured input parameters, a multi-objective optimization model is constructed. The molten iron distribution constraints are converted into dynamic constraints including mutually exclusive operation rules of the continuous casting machine, and the equipment maintenance time is converted into a maintenance period continuity constraint. The decision variable set for the molten iron distribution amount, equipment start and stop status, and maintenance period is set;

[0047] Step 3: Input the decision variable set into the genetic algorithm for global search to generate the initial molten iron distribution and maintenance plan. Input the initial molten iron distribution and maintenance plan into the linear programming algorithm for local optimization adjustment, and output the optimized molten iron distribution path and equipment maintenance sequence.

[0048] Step 4: Execute the optimized molten iron distribution path and equipment maintenance sequence, monitor the time deviation between the blast furnace iron-making rhythm and the continuous casting machine casting in real time, and trigger the casting re-arrangement algorithm to generate a revised casting sequence when the time deviation exceeds the preset threshold. The revised casting sequence and scheduling abnormality data are fed back to the data collection step, and the structured input parameters are updated to reconstruct the multi-objective optimization model.

[0049] The molten iron operation scheduling method based on dynamic constraint modeling and hybrid optimization provided by the present invention achieves efficient matching of dynamic scheduling of molten iron resources and production line maintenance plans through multi-stage collaborative optimization. The following is a detailed technical solution and logical relationship description of each step:

[0050] Step 1: Obtain total order volume and delivery cycle data from the ERP system. Use a data cleaning module to filter out outliers and generate a valid order dataset. Establish an order priority rule base based on customer tiers and contract penalty clauses, and label each order with an urgency tag. Combine these urgency tags with equipment status data (blast furnace hot metal production, continuous casting machine operating status, rolling mill capacity utilization), as well as constraint rule data (hot metal consumption limits and equipment group maintenance conflicts) to generate a structured input parameter set including order priority tags. This step, through data standardization, provides a unified data interface for the subsequent optimization model.

[0051] Step 2: Based on the iron distribution constraints in the structured input parameters, the mutually exclusive operating rules of the continuous casting machine are extracted and mapped into a 0-1 integer constraint matrix, restricting the non-overlapping operating periods of equipment within the group. A continuity constraint function for the maintenance period is generated based on the minimum maintenance duration requirements in the equipment maintenance records. The iron supply fluctuation index is calculated based on real-time iron production data, and the iron distribution weight factor is dynamically adjusted. The corresponding production line priority coefficient is increased when an order urgency label upgrade is detected. The above constraints are combined with the weight factor to generate a dynamic constraint set. The decision variable set for iron distribution quantity, equipment start-up and shutdown status, and maintenance period is defined. A multi-objective optimization model is constructed with the goals of minimizing iron consumption, maximizing order delivery rate, and maximizing equipment utilization.

[0052] Step 3: Genetically encode the decision variable set using real number coding to generate an initial population that represents the molten iron distribution amount, equipment start-up and shutdown status, and maintenance period; based on the objective function weight factor of the multi-objective optimization model, construct a fitness function that includes weighted evaluation values ​​of molten iron consumption, order delivery rate, and equipment utilization rate, and impose penalty terms on violations of the molten iron distribution constraints and 0-1 integer constraints in the matrix constraint table to generate the final fitness value; obtain the initial molten iron distribution and maintenance plan through a global search using a genetic algorithm, and input it as the initial solution into the linear programming algorithm for local fine-tuning, outputting the optimized molten iron distribution path and equipment maintenance sequence.

[0053] Step 4: Execute the optimized molten iron distribution path and equipment maintenance sequence, monitor the time deviation between the blast furnace iron-making rhythm and the continuous casting machine pouring in real time, and trigger the pouring re-arrangement algorithm to generate a revised pouring sequence when the deviation exceeds the preset threshold; call the spare converter molten iron cache resources to deal with the risk of pouring interruption, perform the yield rate degradation operation on the low-priority production line, and calculate the impact weight of production capacity loss on order delivery through the yield rate compensation model; feed back the revised pouring sequence, scheduling abnormality data and impact weight to the data collection step, update the structured input parameters to reconstruct the multi-objective optimization model, and dynamically adjust the crossover probability and mutation probability of the genetic algorithm through the incremental optimization algorithm to generate a revised molten iron distribution path and equipment maintenance sequence.

[0054] Each step is driven by data and coordinated with algorithms to achieve adaptive optimization of dynamic molten iron distribution and production line maintenance plans, solve the imbalance between supply and demand in multiple processes, and improve resource utilization efficiency and order delivery stability.

[0055] Specifically, in the hot metal operation scheduling method based on dynamic constraint modeling and hybrid optimization of the present invention, the order data includes:

[0056] Extract total order volume and delivery cycle data from the ERP system, and perform outlier filtering on the extracted order data to generate a valid order data set;

[0057] Input the valid order dataset into the order priority rule library, and label each order with an urgency tag based on the customer level and contract penalty clauses;

[0058] The annotated urgency labels are used as components of the structured input parameters and fed into the multi-objective optimization model.

[0059] The order data processing process provides structured input for the multi-objective optimization model through multi-stage data integration and rule mapping. After extracting total order volume and delivery cycle data from the ERP system, a data cleansing module is used to filter outliers in the order data. Data with abnormal delivery cycles or orders exceeding production capacity limits is excluded by setting reasonable thresholds, generating a valid order dataset. This dataset is then fed into the order priority rule library, which incorporates customer grading standards and contract penalty clause parsing logic. Each order is dynamically labeled with high, medium, and low urgency levels based on the customer's historical level of cooperation, contractual breach compensation clauses, and the urgency of order delivery. During the labeling process, the rule library uses a weighted calculation based on the customer grading and contract clause priority coefficients using a weight assignment module, outputting a standardized set of urgency labels. Finally, the urgency labels are aligned with the equipment status data and constraint rule data, and their formatting is unified to generate a structured input parameter set consisting of order priority dimensions, equipment operating parameters, and physical constraints, which serves as the data foundation for the multi-objective optimization model.

[0060] The order priority rule library uses a configurable rule engine architecture to support dynamic updates of customer levels and contract penalties. Customer levels are determined based on historical order fulfillment rates, procurement scale, and credit ratings. Contract penalty clauses are parsed through a natural language processing module to analyze key parameters such as liquidated damages ratios and delivery delay tolerance, and converted into quantitative priority coefficients. During the urgency tagging process, a conflict detection algorithm is used to eliminate logical contradictions between rules. For example, when high-level customer orders and high-penalty contract clauses coexist, a priority superposition strategy is used to improve the tag level. The structured input parameter set is connected to the multi-objective optimization model through the data bus to achieve real-time synchronization of urgency tags with molten iron distribution constraints and equipment maintenance rules, providing a decision-making basis for the model to be dynamically updated.

[0061] Structured input parameters are stored in a matrix format, mapping urgency labels to order priority matrices. These matrices are then multi-dimensionally correlated with the blast furnace hot metal production matrix and continuous casting machine operating status matrix in the equipment status data, as well as the upper and lower limit matrices of hot metal consumption in the constraint rule data. In the multi-objective optimization model, the order priority matrix acts as a weighting factor in the objective function calculation, driving the model to prioritize the delivery of high-urgency orders during hot metal allocation. Through the coordinated integration of data preprocessing, rule annotation, and format conversion, order data is integrated end-to-end from the original business system to the optimization model, supporting the multi-objective optimization model's real-time decision-making capabilities under multiple constraints.

[0062] Specifically, the molten iron operation scheduling method based on dynamic constraint modeling and hybrid optimization of the present invention, the constraint rule data includes:

[0063] Map the upper and lower limits of molten iron consumption and the mutually exclusive operation rules of the continuous casting machine into a two-dimensional matrix to generate a matrix constraint table;

[0064] Extracting maintenance period conflicts between equipment groups based on equipment group identifiers in equipment maintenance records and pre-stored maintenance plans;

[0065] Based on the conflict relationship between maintenance periods, a maintenance rule knowledge graph including maintenance dependencies between equipment groups is constructed;

[0066] The matrix constraint table and maintenance rule knowledge graph are used as physical rule constraints and input into the multi-objective optimization model.

[0067] Constraint rule data processing provides physical rule constraints for the multi-objective optimization model through structured modeling and conflict relationship analysis. When mapping the upper and lower limits of molten iron consumption and the mutually exclusive operation rules of the continuous casting machine into a two-dimensional matrix, the matrix elements are populated with the maximum allowable molten iron consumption, the minimum supply threshold, and the mutually exclusive operation identifier of the equipment, with the equipment group as the row dimension and the production period as the column dimension, to generate a matrix constraint table. This matrix constraint table associates equipment groups and production periods through row and column indexes, realizing a spatial and temporal dual-dimensional expression of the molten iron allocation rules. The equipment group identifiers in the equipment maintenance records are matched with the pre-stored maintenance plan. Based on the equipment group topology relationship and the maintenance period overlap detection algorithm, the maintenance period conflict relationships caused by shared energy supply or process connection dependencies between equipment groups are extracted, forming a set of conflict relationship triplets, including the conflicting equipment group identifier, the conflict period start and end time, and the conflict type code.

[0068] Based on a set of conflict relationship triples, a maintenance rule knowledge graph is constructed using a graph database. The knowledge graph uses equipment groups as nodes and conflict relationships as edge attributes. Node attributes include the historical frequency of equipment group maintenance and the shortest duration of a single maintenance. Edge attributes include the length of the conflict period and the weight coefficient of the conflict type. A graph traversal algorithm is used to identify maintenance dependency chains between equipment groups. For example, when the maintenance of equipment group A depends on the shutdown status of equipment group B, a maintenance sequence constraint rule between equipment groups is generated. The matrix constraint table and the maintenance rule knowledge graph are input into the multi-objective optimization model through a unified data interface. The matrix constraint table defines the spatial-temporal boundary conditions for molten iron distribution, and the knowledge graph provides the topological constraint logic for the equipment group maintenance sequence. Together, the two constitute the physical rule constraint conditions of the model.

[0069] During model operation, the upper and lower limits of molten iron consumption in the matrixed constraint table are directly involved in the objective function calculation, limiting the range of feasible solutions for molten iron allocation. The maintenance dependency relationships in the knowledge graph are verified by a directed acyclic graph (DAG) verification algorithm to verify the legitimacy of the equipment group maintenance sequence, avoiding overlapping conflicting periods. When a conflicting period violation is detected, a constraint violation penalty mechanism is triggered, dynamically adjusting the fitness function parameters of the genetic algorithm. Through the dual matrix and graph representation of rule data, efficient coordination between discrete equipment constraints and continuous resource allocation is achieved, supporting the global optimization capabilities of the multi-objective optimization model under multidimensional constraints.

[0070] Specifically, the method for scheduling molten iron operations based on dynamic constraint modeling and hybrid optimization of the present invention, step 2 includes:

[0071] Based on the collected equipment status data, the mutually exclusive operation rules of the continuous casting machine are extracted and converted into 0-1 integer constraints to limit the operation time of the equipment in the group to no overlap;

[0072] Generate a maintenance period continuity constraint function based on the minimum maintenance time requirement in the collected equipment maintenance data;

[0073] Based on the collected real-time data on molten iron production, the molten iron supply fluctuation index is calculated. The molten iron allocation weight factor in the matrix constraint table is dynamically adjusted according to the molten iron supply fluctuation index. When an urgency tag upgrade is detected, the priority coefficient of the corresponding production line is increased.

[0074] The 0-1 integer constraint, the maintenance period continuity constraint function, and the adjusted molten iron distribution weight factor and priority coefficient are combined to generate a dynamic constraint condition set.

[0075] Step 2 constructs the core rule system of the multi-objective optimization model through the dynamic coupling of multi-dimensional constraints. Based on the continuous casting machine operation logs in the collected equipment status data, the mutually exclusive operation rules of the equipment in the group are extracted, specifically including the conflicting operation time periods of multiple continuous casting machines in the same equipment group due to energy supply restrictions or process connection requirements; the conflicting relationship is mapped to a 0-1 integer constraint, and the start and stop status of the equipment in a specific time period is represented by a binary variable, and a hard constraint condition is set that the operation time periods of the equipment in the group do not overlap. The minimum maintenance time requirement in the equipment maintenance data is analyzed through the time window continuity rule, and the shortest duration of a single maintenance operation is converted into a continuity constraint function of the start and end time of the maintenance period, which limits the violations of time period segmentation or interval in the maintenance plan, such as generating prohibitive conditions for segmented maintenance.

[0076] Real-time molten iron production data is statistically calculated using a sliding window method to calculate the molten iron supply fluctuation index, reflecting the real-time stability of molten iron production capacity. Combined with the molten iron allocation weight factors in the matrix constraint table, the molten iron allocation ratio of each production line is dynamically adjusted based on the fluctuation index. For example, when the supply fluctuation index exceeds a threshold, the weight allocation of the high-volatility production line is reduced. Furthermore, when an order urgency tag is detected to have increased, the molten iron allocation weight of the corresponding production line is increased through a priority coefficient superposition mechanism, prioritizing the delivery of high-urgency orders. 0-1 integer constraints, continuity constraint functions, and dynamically adjusted weight factors and priority coefficients are logically integrated through a constraint combiner to generate a dynamic constraint set consisting of discrete equipment operating rules and continuous resource allocation strategies.

[0077] In a multi-objective optimization model, a set of dynamic constraints serves as boundary conditions for objective function calculation and decision variable search. 0-1 integer constraints define the feasible solution space for equipment start-up and shutdown states through mixed integer programming. Continuity constraints eliminate illegal maintenance plans through a time window validation algorithm. Dynamic weight factors and priority coefficients balance the economic efficiency of hot metal distribution with the timeliness of order delivery through a weighted summation. Through the synergistic effect of these constraints, dynamic coupled optimization of hot metal distribution and equipment maintenance under multi-dimensional constraints is achieved, enabling the model to generate feasible solutions that balance resource utilization and production stability.

[0078] Specifically, the method for scheduling molten iron operations based on dynamic constraint modeling and hybrid optimization of the present invention, step 3 includes:

[0079] Real number coding is used to generate the initial population that represents the molten iron distribution amount, equipment start-up and shutdown status, and maintenance period;

[0080] Based on the defined decision variable set of molten iron distribution, equipment start-up and shutdown status, and maintenance period, the initial population is generated using real number coding.

[0081] Based on the order delivery rate weight factor and equipment utilization rate weight factor in the objective function of the multi-objective optimization model, a fitness function is constructed. The fitness function calculates the comprehensive evaluation value of molten iron consumption, order delivery rate and equipment utilization rate through weighted summation;

[0082] For the violation data of the molten iron distribution constraint and the 0-1 integer constraint in the matrix constraint table, set the penalty term and add it to the comprehensive evaluation value to generate the final fitness value;

[0083] The individual parameter values ​​with the best fitness evaluation are converted into the input parameter format of the linear programming algorithm and input into the linear programming algorithm as the initial solution for local optimization adjustment.

[0084] Step 3 uses a hybrid optimization algorithm to achieve global-local coordinated optimization of hot metal distribution and equipment maintenance planning. Based on a defined set of decision variables for hot metal distribution, equipment start / stop status, and maintenance period, a real-number encoding strategy is used to genetically encode each decision variable. Hot metal distribution is mapped to a continuous real-number variable in tons, equipment start / stop status is represented by a 0-1 switch variable, and maintenance periods are encoded as real-number pairs of start time and duration. This generates an initial population consisting of multidimensional parameters. The size of the initial population is dynamically adjusted based on the number of equipment groups and the span of the production period. Uniform distribution and random perturbation strategies are used to ensure population diversity, providing a feasible solution space for the genetic algorithm's global search.

[0085] A fitness function is constructed based on the order delivery rate and equipment utilization weighting factors defined in the multi-objective optimization model. This function integrates the three objective dimensions of molten iron consumption, order delivery rate, and equipment utilization through weighted summation. Molten iron consumption is calculated as a negative indicator, while order delivery rate and equipment utilization are weighted and accumulated as positive indicators to generate a comprehensive evaluation value. A penalty calculation module is implemented to address violations of the upper and lower limits of molten iron allocation and the 0-1 integer constraints in the matrixed constraint table. Dynamic penalty coefficients are generated based on the degree of constraint violation (e.g., excessive molten iron quantity or overlapping equipment operating periods). These coefficients are then added to the comprehensive evaluation value to generate the final fitness value. The fitness score of the degraded solution guides the population to converge toward the feasible domain.

[0086] After the fitness assessment is complete, an elite retention strategy is used to select the optimal individual parameter set. The resulting parameters, including the hot metal distribution quantity, equipment start / stop periods, and maintenance sequence, are then converted into the standard input format for a linear programming algorithm. During this conversion, the hot metal distribution quantity is retained as a continuous variable, the equipment start / stop status is converted from a 0-1 variable into a time period activation identifier, and the maintenance period parameters are mapped into time window boundary conditions. These converted parameters serve as the initial solution for the linear programming algorithm. By combining the hot metal distribution rules from the matrixed constraint table with the maintenance sequence dependencies from the knowledge graph, localized fine-tuning is performed to optimize the continuity of the hot metal distribution path and the time period integrity of the equipment maintenance plan, ultimately outputting a final solution that satisfies multiple objective trade-offs.

[0087] Through a hybrid iterative mechanism combining genetic algorithms and linear programming algorithms, a coordinated optimization of global exploration and local development is achieved. The diversity of the initial population ensures the comprehensiveness of the global search, the penalty term in the fitness function strengthens the hard limits of the constraints, and the local optimization of the linear programming algorithm improves the accuracy and feasibility of the solution. This hybrid strategy effectively overcomes the shortcomings of a single algorithm, which can easily lead to local optimality or insufficient convergence speed, and provides a highly robust solution for dynamic production scheduling under conditions of limited molten iron resources.

[0088] Specifically, the molten iron operation scheduling method based on dynamic constraint modeling and hybrid optimization of the present invention, the optimized molten iron distribution path and equipment maintenance sequence include:

[0089] Based on the generated equipment maintenance time sequence, a time deviation threshold between the blast furnace tapping rhythm and the continuous casting machine pouring time is established, and the time sequence offset is calculated in real time;

[0090] When the risk of molten iron interruption is detected, the backup converter molten iron cache resources in the collected equipment status data are called up, and the yield rate is downgraded for the production line marked with the lowest urgency level;

[0091] The impact weight on order delivery is calculated using the preset yield rate compensation model based on the production capacity loss data after the yield rate degradation operation and the order urgency tag.

[0092] The impact weights are input into the dynamic constraint condition set, and the molten iron distribution weight factors and priority coefficients are updated.

[0093] The optimized molten iron distribution path and equipment maintenance sequence achieve real-time adaptation of resource scheduling and constraints through a dynamic monitoring and feedback mechanism. Based on the generated equipment maintenance sequence and combined with historical data on blast furnace tapping rhythms and continuous casting machine pours, a sliding time window algorithm is used to calculate the theoretical time intervals between adjacent processes and set a time deviation threshold. The blast furnace tapping timestamps and continuous casting machine pour start times are collected in real time, and a timing alignment algorithm is used to calculate the offset between the actual production sequence and the planned sequence, generating a timing offset matrix. This offset matrix is ​​displayed through a visual monitoring interface, and a warning signal is triggered when the offset exceeds the threshold.

[0094] When a hot metal outage risk is detected, the system calls upon the backup converter hot metal cache resources from the equipment status data, including the backup converter's real-time hot metal inventory, pipeline status, and activation priority parameters. Based on the backup resource scheduling strategy, the backup converter closest to the outage point is prioritized for activation, and the feasibility of hot metal delivery is verified using pipeline pressure sensor data. Simultaneously, the system downgrades the yield rate of production lines marked with the lowest urgency level. This is achieved by adjusting rolling parameters to reduce the yield rate and free up hot metal supply resources. The scope and duration of this downgrade are dynamically adjusted based on the outage risk level to avoid impacting the delivery of high-priority orders.

[0095] The yield rate compensation model calculates impact weights based on capacity loss data and order urgency tags. The model receives real-time capacity loss data after a yield rate downgrade, including the percentage of production reduction and the amount of hot metal saved. Combined with the remaining delivery cycle days and the penalty coefficient from the order urgency tag, the impact weight value for each order is calculated through weighted accumulation. During the weight calculation process, the penalty coefficient and delivery cycle urgency of high-urgency orders are given higher weight factors, tilting the allocation of capacity loss toward lower-priority orders. The calculated impact weight set is transmitted to the dynamic constraint condition set via the data bus, triggering real-time updates of the hot metal allocation weight factor and priority coefficient.

[0096] During the update process, the hot metal allocation weight factors within the dynamic constraint set are adjusted proportionally based on the impact weights. For example, the allocation weights of production lines with high impact weights are reduced to compensate for their capacity losses. The priority coefficients are dynamically increased based on the updated status of order urgency tags, strengthening the decision weight of high-urgency orders in subsequent optimization models. The updated constraints are synchronized to the multi-objective optimization model through an incremental optimization algorithm. This drives the model to generate hot metal allocation routes and equipment maintenance schedules adapted to the latest production status in the next iteration, achieving a dynamic balance between hot metal supply and order demand.

[0097] Specifically, the method for scheduling molten iron operations based on dynamic constraint modeling and hybrid optimization of the present invention, step 4 includes:

[0098] The generated scheduling exception data is stored in the knowledge base associated with the structured input parameters, triggering the reconstruction of the constraints of the multi-objective optimization model and the update of the dynamic constraint set;

[0099] An incremental optimization algorithm is used to solve the updated multi-objective optimization model. The incremental optimization algorithm dynamically adjusts the crossover probability and mutation probability of the genetic algorithm based on the collected real-time data of the equipment status and the initial population parameters of the genetic algorithm.

[0100] The revised molten iron distribution path and equipment maintenance sequence are generated through the incremental optimization algorithm, and the revised molten iron distribution path and maintenance sequence are input into the preset production line scheduling system.

[0101] Step 4 implements dynamic iteration of the production plan through closed-loop feedback and incremental optimization. The generated scheduling exception data, including blast furnace tapping timing deviations, continuous casting machine pouring interruption events, and molten iron distribution overrun records, are standardized through the data cleaning module and stored in a knowledge base associated with structured input parameters. The knowledge base uses a graph structure to store the spatiotemporal correlations of the exception data, for example, mapping timing deviations to time node attributes and associating molten iron distribution overrun records to equipment group topology relationships. When the exception data reaches the preset trigger threshold, the constraint reconstruction process of the multi-objective optimization model is initiated, calling the weight factor and priority coefficient update module in the dynamic constraint set to recalculate the molten iron distribution weight and equipment maintenance timing dependency.

[0102] The incremental optimization algorithm uses real-time data collected from equipment status, including blast furnace hot metal production, continuous casting machine operating status, and spare converter buffer capacity. Combined with the genetic algorithm's initial population parameters, it dynamically adjusts crossover and mutation probabilities. The crossover probability is dynamically adjusted based on population diversity metrics (such as the Hamming distance between individuals), while the mutation probability is adaptively adjusted based on the frequency of constraint violations. For example, as the number of constraint violations increases in successive iterations, the mutation probability is increased to enhance search breadth. During the incremental optimization process, the genetic algorithm's elite retention strategy prioritizes individuals with the highest fitness. A taboo search mechanism is also introduced to prevent repeated searches for local optimal solutions, improving the algorithm's convergence efficiency.

[0103] The revised hot metal distribution path and equipment maintenance sequence generated by the incremental optimization algorithm are converted into the input protocol format of the production line scheduling system via the data format conversion module. During this conversion process, the hot metal distribution path is mapped into a hot metal supply matrix for each equipment group, and the equipment maintenance sequence is converted into Gantt chart data with time period dependencies. The revised data is transmitted to the preset production line scheduling system via the industrial bus protocol, driving the physical equipment to execute the updated production plan. During execution, the scheduling system monitors the equipment response status in real time and feeds back data on execution delays or resource conflicts to the knowledge base, achieving dynamic coordinated optimization of hot metal distribution and equipment maintenance.

[0104] Embodiment of the present invention: In a steel production scenario, first, the total monthly order volume of general steel, high-quality steel, and special steel is divided according to the order type, and the theoretical molten iron demand is calculated in combination with the yield rate parameters of each steel grade. The process connection rules of each converter corresponding to the continuous casting machine are determined through the equipment association relationship, a converter-continuous casting machine mapping table is established, and the process constraints of the molten iron supply path are clarified. For blast furnaces, converters, and rolling mill equipment, the total monthly allowable maintenance time is set according to the equipment maintenance plan, and the continuous maintenance mode and the single-day irregular maintenance mode are distinguished to generate the maintenance period continuity constraint condition. Based on the upper and lower limits of molten iron consumption for general steel, high-quality steel, and special steel, a molten iron allocation constraint matrix is ​​constructed, in which the rows of the matrix represent the production date, the columns represent the steel grade category, and the elements are filled with the daily molten iron consumption threshold. According to the priority rule of the continuous casting machine, the priority weight coefficient is set, and the equipment operation time activation identifier is generated in combination with the daily quota production capacity of the continuous casting machine to limit the mutually exclusive operation time periods of the continuous casting machines in the same group.

[0105] A multi-objective optimization model was established using linear programming, with minimizing monthly hot metal consumption as the core objective function. The hot metal allocation constraint matrix, maintenance period continuity constraints, and continuous casting machine priority weights were used as hard constraints. 0-1 variables were introduced to represent equipment start / stop status and maintenance period occupancy indicators. Matrix operations were used to integrate the daily capacity constraints of each production line to determine the minimum hot metal consumption plan that meets order delivery requirements. Specific maintenance plans for the blast furnace, converter, and rolling mill were simultaneously output, including the start and end times of continuous maintenance periods and the distribution of daily maintenance windows. To address mutually exclusive operation scenarios for continuous casting machines, constraints were implemented to ensure that the operating periods of continuous casting machines within the same equipment group did not overlap, enabling dynamic matching of hot metal supply paths and equipment status. Finally, daily hot metal consumption allocation tables for standard steel, high-quality steel, and special steel were generated, accurate to the hourly hot metal supply volume for each production line. These tables were updated in conjunction with maintenance plans, achieving global optimization of production plans under resource-scarce conditions.

[0106] Order data processing and priority labeling algorithm: After order data is extracted from the enterprise resource planning system, outlier filtering is performed. The filtering process sets dynamic thresholds based on historical production data to exclude orders that exceed production capacity limits or have unreasonable delivery cycles, thereby generating a valid order set. Valid orders are input into the priority rule base, which includes a customer grading module and a contract parsing module. The customer grading module comprehensively evaluates parameters such as the customer's historical cooperation level and procurement scale to generate a grade coefficient; the contract parsing module identifies terms such as the liquidated damages ratio and delivery grace period to generate a risk coefficient. The two use weighted calculations to output three levels of urgency: high, medium, and low. The conflict detection mechanism automatically handles rule inconsistencies.

[0107] Constraint rule structuring algorithm: Hot metal consumption rules are mapped into a two-dimensional matrix structure: rows correspond to equipment group numbers, columns correspond to production period numbers, and matrix elements contain maximum hot metal consumption, minimum supply thresholds, and equipment mutual exclusion identifiers. A maintenance conflict detection algorithm, based on the topological relationships of equipment groups, scans pre-stored maintenance plans for overlapping periods, and outputs conflicting equipment group pairs and conflicting period parameters. A graph database constructs equipment group nodes and conflict relationship edges. Nodes store equipment attributes, and edges store conflict types and weight coefficients. A graph traversal algorithm identifies multi-level maintenance dependency chains.

[0108] Dynamic constraint generation algorithm: Equipment mutual exclusion rules are converted into a binary decision mechanism. This mechanism defines the start and stop state variables for equipment during specific time periods. Constraint equations enforce that the sum of the operating state variables within the same equipment group does not exceed 1, achieving zero overlap in the operating periods of equipment within the group. Maintenance continuity constraints are based on the minimum maintenance duration required for equipment, constructing a time window function to limit the difference between maintenance start and end times. The dynamic hot metal allocation weight adjustment module proportionally adjusts the allocation weights for each production line based on the real-time production fluctuation index. As order urgency increases, a coefficient amplifier increases the priority of the corresponding production line.

[0109] A hybrid optimization algorithm uses real-number encoding to generate the initial population of decision variables: hot metal distribution is represented as a continuous value, equipment status is represented as a binary value, and maintenance periods are encoded as pairs of start and duration values. The fitness function comprehensively calculates weighted evaluations of hot metal consumption, order delivery rate, and equipment utilization, and applies dynamic penalties for constraint violations. After the genetic algorithm completes the global search, the elitist solution parameters are converted to a linear programming input format. With the hot metal distribution path continuity as the optimization objective, local refinement is performed within the matrix constraints and graph dependency boundaries.

[0110] Real-time risk response algorithm: The timing deviation monitoring module sets dynamic thresholds based on historical data, collects blast furnace tapping and continuous casting pour timestamps in real time, and calculates minute-level deviations between planned and actual timings. When deviations exceed limits or the risk of a hot metal outage is detected, the backup converter hot metal cache resource selection algorithm is activated: an optimal call plan is generated based on spatial distance, pipeline status, and inventory data. The yield degradation module adjusts rolling parameters for the least urgent production line, reallocating released hot metal resources to high-priority production lines. The capacity loss impact weight calculator comprehensively evaluates the production loss value and order urgency of the degradation operation, outputting weight coefficients to update the constraint condition set.

[0111] Incremental optimization and closed-loop control algorithm: Scheduling anomaly data is stored in a graph-structured knowledge base, with nodes associated with equipment group attributes and edges associated with the spatiotemporal characteristics of anomaly events. The accumulation of anomaly data triggers a constraint reconstruction process, recalculating hot metal distribution weights and equipment dependencies. The incremental optimization algorithm receives real-time equipment status data and dynamically adjusts crossover and mutation probabilities based on the initial genetic algorithm population parameters. The algorithm increases the mutation probability when population diversity decreases and expands the search range when the frequency of constraint violations increases. The optimization results are converted into executable instructions for the equipment through the industrial protocol conversion module, driving the production line scheduling system to complete closed-loop control.

[0112] Multi-objective optimization model:

[0113] The multi-objective optimization model's core objectives are minimizing hot metal consumption, maximizing order delivery rates, and maximizing equipment utilization. The model's input integrates an order priority matrix (labeled with high / medium / low urgency), real-time equipment status parameters (blast furnace hot metal output, continuous casting machine operating status), and a matrixed constraint table (high and low limits for hot metal allocation). The objective function balances three optimization objectives using weighted coefficients:

[0114] Hot metal consumption target: Minimize total hot metal consumption, with the weight coefficient dynamically adjusted based on the hot metal supply fluctuation index;

[0115] Order delivery goal: maximize the completion rate of high-urgency orders, with the weight coefficient positively correlated with the order urgency label;

[0116] Equipment utilization goal: maximize the effective operating time of the continuous casting machine, and the weight coefficient is associated with the equipment group production capacity threshold.

[0117] Dynamic constraint model:

[0118] The dynamic constraint model consists of physical rule constraints and real-time adjustment mechanisms:

[0119] Physical rules layer:

[0120] Equipment mutual exclusion constraints are implemented through 0-1 integer programming, forcing the continuous casting machines in the same equipment group to operate without overlap;

[0121] Maintenance continuity constraints use a time window function to limit the minimum duration of a single maintenance and prohibit segmented maintenance.

[0122] Dynamic Adjustment Layers:

[0123] The molten iron allocation weight factor is scaled proportionally based on the real-time production fluctuation index (e.g., the weight of inefficient production lines is reduced when the fluctuation index rises);

[0124] The production line priority coefficient is automatically amplified (e.g., coefficient × 1.2) when it detects that the order urgency label has been upgraded, giving priority to ensuring the supply of resources for high-urgency orders.

[0125] Order priority model:

[0126] The order priority model generates quantitative evaluation results based on the rule base:

[0127] Customer rating sub-model: Input historical order fulfillment rate, procurement scale, and credit rating parameters, and output customer rating coefficients from 1 to 10;

[0128] Contract risk sub-model: Analyzes the liquidated damages ratio and delivery grace period in the contract terms and generates risk coefficients of 1-5 levels;

[0129] Urgency synthesis module: The customer level coefficient and the contract risk coefficient are weighted and accumulated, and divided into three levels of high / medium / low according to the threshold range. Conflicting orders adopt a coefficient superposition strategy.

[0130] Maintenance rule knowledge graph model:

[0131] The knowledge graph model constructs the topological structure of maintenance dependencies between equipment groups:

[0132] Node attributes: storage device group number, historical maintenance frequency, and minimum single maintenance duration;

[0133] Edge attributes: define equipment group conflict types (such as energy dependency, process connection), and conflict period overlap weights;

[0134] Dependency chain reasoning: Identify multi-level dependency paths (such as the dependency chain formed by equipment group A→B→C) through graph traversal algorithms and generate a maintenance timing constraint rule base.

[0135] Yield rate compensation model:

[0136] The yield rate compensation model quantifies the impact of production capacity loss on order delivery:

[0137] Capacity loss calculation module: inputs the rolling parameter adjustment value of the yield rate degradation operation, and outputs the output reduction ratio per unit time;

[0138] Order Impact Assessment Module: This module combines the remaining delivery days of the order urgency tag with the penalty coefficient to calculate the risk weight of delivery delays caused by production capacity loss.

[0139] Weight feedback mechanism: risk weights are input into the dynamic constraint model to trigger iterative updates of the molten iron allocation weight factors and priority coefficients.

[0140] Closed-loop control model:

[0141] The closed-loop control model implements the data flow closed loop of "monitoring, optimization, and execution":

[0142] Real-time monitoring layer: collects blast furnace tapping timestamps and continuous casting start times, and calculates minute-level deviations between planned and actual timings;

[0143] Abnormal response layer: When the deviation exceeds the threshold, the backup resource selection algorithm and the yield rate degradation operation are called to generate an emergency plan;

[0144] Incremental optimization layer: Dynamically adjusts crossover / mutation probability based on the initial population parameters of the genetic algorithm (e.g., increases mutation rate when population diversity decreases) and quickly outputs a revised plan;

[0145] Execution feedback layer: Optimization results are converted into production line scheduling instructions, and execution exception data is sent back to the knowledge base to reconstruct constraint rules.

[0146] The specific implementation method of the present invention in the steel production planning system is as follows: Under the working condition of scarce molten iron resources, the total order volume and delivery cycle data are first collected through the ERP system, and the threshold filtering algorithm is used to exclude invalid data with abnormal delivery cycle or order volume exceeding the production capacity limit, so as to generate a valid order data set. The valid data set is input into the order priority rule library, and according to the customer's historical cooperation level, contract breach compensation ratio and delivery urgency, high, medium and low urgency labels are dynamically marked to form a standardized order priority matrix. At the same time, the real-time molten iron output of the blast furnace, the operating status parameters of the continuous casting machine and the rolling line capacity utilization rate are extracted from the equipment status data. Combined with the upper and lower limits of molten iron consumption in the constraint rule data and the mutually exclusive operation rules of the continuous casting machine, a two-dimensional matrix constraint table is constructed with equipment groups as rows and production time periods as columns. The elements include maximum molten iron consumption, minimum supply threshold and equipment mutually exclusive operation identifier. The equipment group identifiers in the equipment maintenance records and the pre-stored maintenance plans are used to extract conflict relationships through a period overlap detection algorithm to generate a set of conflict period triplets. A maintenance rule knowledge graph is then constructed based on a graph database. The nodes represent the equipment groups and their maintenance history frequencies, and the edge attributes describe the conflict period length and type weight.

[0147] A multi-objective optimization model was constructed based on the order priority matrix, matrix constraint table, and maintenance rule knowledge graph in the structured input parameters. The mutually exclusive operation rules of the continuous casting machine were converted into 0-1 integer constraints to ensure that the equipment operating periods within the group did not overlap. A continuity constraint function was generated based on the minimum maintenance duration requirement for the equipment to limit the segmentation of maintenance periods. In combination with real-time molten iron production data, a sliding window statistical method was used to calculate the molten iron supply fluctuation index. The molten iron allocation weight factors in the matrix constraint table were dynamically adjusted, and a priority coefficient was added when an order urgency label upgrade was detected. These constraints were integrated into a dynamic constraint set, defining decision variables for molten iron allocation, equipment start / stop status, and maintenance period. Real number encoding was used to generate the initial population of the genetic algorithm. The molten iron allocation was encoded as a continuous variable, the equipment start / stop status was represented by a 0-1 variable, and the maintenance period was encoded as the start time and duration. A fitness function was constructed based on the weights of order delivery rate and equipment utilization in the objective function. This weighted calculation of molten iron consumption, delivery rate, and utilization was applied. Dynamic penalties were applied to constraint violations to generate a fitness value and select the optimal individual. The optimal parameters are converted into linear programming input format, and local optimization is performed by combining matrix constraints and knowledge graph dependencies to output the molten iron distribution path and equipment maintenance timing.

[0148] When executing the optimization plan, the blast furnace tapping timestamp and continuous casting machine pouring start sequence are monitored in real time, and the deviation between the actual and planned values ​​is calculated. When the deviation exceeds a threshold or the risk of a hot metal outage is detected, the backup converter hot metal cache resources are activated, prioritizing the nearest backup unit. Yield degradation is performed on the production line marked as the least urgent, freeing up hot metal resources by adjusting rolling parameters. The yield compensation model calculates impact weights based on capacity loss data and order urgency tags, and updates the allocation weights and priority coefficients in the dynamic constraint set. Scheduling anomaly data is stored in a knowledge base, triggering an incremental optimization algorithm to dynamically adjust the genetic algorithm's crossover and mutation probabilities, for example, by increasing the mutation probability based on population diversity. The corrected hot metal distribution path and maintenance sequence are transmitted to the production line scheduling system via industrial protocols, driving equipment execution and continuously monitoring feedback. This complete process combines multi-source data fusion, dynamic constraint coupling, and closed-loop iterative optimization, enabling real-time coordinated response between hot metal distribution and equipment maintenance.

[0149] The present invention solves the real-time collaborative optimization problem of molten iron dynamic allocation and production line maintenance in the steel production planning system through multi-source data fusion, dynamic constraint modeling and closed-loop feedback mechanism. The specific technical solution is as follows:

[0150] By collecting order data, equipment status data, and constraint rule data, we construct structured input parameters, including order priority tags, upper and lower limits for hot metal distribution, mutually exclusive equipment operation rules, and maintenance conflict relationships. After filtering outliers and labeling the order data with urgency, we generate an order priority matrix. Equipment status data is mapped into a continuous casting machine operating status matrix and a hot metal production fluctuation index. Constraint rule data is converted into a matrix constraint table and a maintenance rule knowledge graph. This data is integrated through a unified interface to support a multi-objective optimization model for dynamic modeling of hot metal distribution paths and equipment maintenance sequences, achieving a multi-dimensional coupling of order demand, resource supply, and equipment constraints.

[0151] Based on the 0-1 integer constraints and continuity constraint functions in a matrixed constraint table, the allocation weights are dynamically adjusted in conjunction with the hot metal supply fluctuation index to generate a dynamic constraint set. A genetic algorithm is used to perform a global search of the decision variable set to generate an initial hot metal allocation and maintenance plan. A fitness function integrates hot metal consumption, order delivery rate, and equipment utilization targets, and constraint violation penalties are applied to guide convergence of feasible solutions. The optimal individual parameters are then input into a linear programming algorithm for local refinement, optimizing hot metal allocation continuity and maintenance period integrity. This hybrid algorithm achieves a synergistic effect between global exploration and local development, overcoming the vulnerability of single algorithms to local optimality.

[0152] During the execution of the optimization plan, real-time monitoring of deviations between blast furnace tapping and continuous casting pouring sequences triggers a pouring interruption risk response strategy: activating backup converter resources and downgrading the yield rate of low-priority production lines. A yield rate compensation model calculates the impact of capacity loss on order delivery, dynamically updating the hot metal allocation weight and priority coefficients. Scheduling anomaly data is stored in a knowledge base, driving an incremental optimization algorithm to adjust the crossover and mutation probabilities of the genetic algorithm and reconstruct the constraints of the multi-objective model. The revised hot metal allocation path and maintenance sequence are then fed into the production line scheduling system for execution, forming a closed loop and achieving a dynamic balance between hot metal supply and order demand.

[0153] The above solution solves the problem of real-time coordination of multi-process resource allocation and equipment maintenance through data-driven modeling, dynamic constraint coupling and adaptive optimization mechanism, and improves molten iron utilization efficiency and production stability.

Claims

1. A hot metal operation scheduling method based on dynamic constraint modeling and hybrid optimization, characterized in that: include: Step 1: Collect order data, device status data, and constraint rule data, filter outliers and mark priorities on the order data, and generate structured input parameters including order priority tags; Step 2: Based on the molten iron distribution constraints and the equipment maintenance time in the structured input parameters, a multi-objective optimization model is constructed. The molten iron distribution constraints are converted into dynamic constraints including mutually exclusive operation rules of the continuous casting machine, and the equipment maintenance time is converted into a maintenance period continuity constraint. The decision variable set for the molten iron distribution amount, equipment start and stop status, and maintenance period is set; Step 3: Input the decision variable set into the genetic algorithm for global search to generate the initial molten iron distribution and maintenance plan. Input the initial molten iron distribution and maintenance plan into the linear programming algorithm for local optimization adjustment, and output the optimized molten iron distribution path and equipment maintenance sequence. Step 4: Execute the optimized molten iron distribution path and equipment maintenance sequence, monitor the time deviation between the blast furnace iron-making rhythm and the continuous casting machine casting in real time, and trigger the casting re-arrangement algorithm to generate a revised casting sequence when the time deviation exceeds the preset threshold. The revised casting sequence and scheduling abnormality data are fed back to the data collection step, and the structured input parameters are updated to reconstruct the multi-objective optimization model.

2. The method for arranging molten iron operation plan based on dynamic constraint modeling and hybrid optimization according to claim 1, characterized in that: Order data includes: Extract total order volume and delivery cycle data from the ERP system, and perform outlier filtering on the extracted order data to generate a valid order data set; Input the valid order dataset into the order priority rule library, and label each order with an urgency tag based on the customer level and contract penalty clauses; The annotated urgency labels are used as components of the structured input parameters and fed into the multi-objective optimization model.

3. The method for arranging molten iron operation plan based on dynamic constraint modeling and hybrid optimization according to claim 2, characterized in that: Constraint rule data includes: Map the upper and lower limits of molten iron consumption and the mutually exclusive operation rules of the continuous casting machine into a two-dimensional matrix to generate a matrix constraint table; Extracting maintenance period conflicts between equipment groups based on equipment group identifiers in equipment maintenance records and pre-stored maintenance plans; Based on the conflict relationship between maintenance periods, a maintenance rule knowledge graph including maintenance dependencies between equipment groups is constructed; The matrix constraint table and maintenance rule knowledge graph are used as physical rule constraints and input into the multi-objective optimization model.

4. The method for arranging molten iron operation plan based on dynamic constraint modeling and hybrid optimization according to claim 3, characterized in that: Step 2 includes: Based on the collected equipment status data, the mutually exclusive operation rules of the continuous casting machine are extracted and converted into 0-1 integer constraints to limit the operation time of the equipment in the group to no overlap; Generate a maintenance period continuity constraint function based on the minimum maintenance time requirement in the collected equipment maintenance data; Based on the collected real-time data on molten iron production, the molten iron supply fluctuation index is calculated. The molten iron allocation weight factor in the matrix constraint table is dynamically adjusted according to the molten iron supply fluctuation index. When an urgency tag upgrade is detected, the priority coefficient of the corresponding production line is increased. The 0-1 integer constraint, the maintenance period continuity constraint function, and the adjusted molten iron distribution weight factor and priority coefficient are combined to generate a dynamic constraint condition set.

5. The method for scheduling molten iron operations based on dynamic constraint modeling and hybrid optimization according to claim 4, characterized in that: Step 3 includes: Real number coding is used to generate the initial population that represents the molten iron distribution amount, equipment start-up and shutdown status, and maintenance period; Based on the defined decision variable set of molten iron distribution, equipment start-up and shutdown status, and maintenance period, the initial population is generated using real number coding. Based on the order delivery rate weight factor and equipment utilization rate weight factor in the objective function of the multi-objective optimization model, a fitness function is constructed. The fitness function calculates the comprehensive evaluation value of molten iron consumption, order delivery rate and equipment utilization rate through weighted summation; For the violation data of the molten iron distribution constraint and the 0-1 integer constraint in the matrix constraint table, set the penalty term and add it to the comprehensive evaluation value to generate the final fitness value; The individual parameter values ​​with the best fitness evaluation are converted into the input parameter format of the linear programming algorithm and input into the linear programming algorithm as the initial solution for local optimization adjustment.

6. The method for arranging molten iron operation plan based on dynamic constraint modeling and hybrid optimization according to claim 5, characterized in that: The optimized molten iron distribution path and equipment maintenance sequence include: Based on the generated equipment maintenance time sequence, a time deviation threshold between the blast furnace tapping rhythm and the continuous casting machine pouring time is established, and the time sequence offset is calculated in real time; When the risk of molten iron interruption is detected, the backup converter molten iron cache resources in the collected equipment status data are called up, and the yield rate is downgraded for the production line marked with the lowest urgency level; The impact weight on order delivery is calculated using the preset yield rate compensation model based on the production capacity loss data after the yield rate degradation operation and the order urgency tag. The impact weights are input into the dynamic constraint condition set, and the molten iron distribution weight factors and priority coefficients are updated.

7. The method for scheduling molten iron operations based on dynamic constraint modeling and hybrid optimization according to claim 6, characterized in that: Step 4 includes: The generated scheduling exception data is stored in the knowledge base associated with the structured input parameters, triggering the reconstruction of the constraints of the multi-objective optimization model and the update of the dynamic constraint set; An incremental optimization algorithm is used to solve the updated multi-objective optimization model. The incremental optimization algorithm dynamically adjusts the crossover probability and mutation probability of the genetic algorithm based on the collected real-time data of the equipment status and the initial population parameters of the genetic algorithm. The revised molten iron distribution path and equipment maintenance sequence are generated through the incremental optimization algorithm, and the revised molten iron distribution path and maintenance sequence are input into the preset production line scheduling system.

Images