A multi-variety mixed production dynamic scheduling system based on a chemical fiber intelligent factory

Abstract

The present application relates to a kind of based on the multi-species mixed production dynamic scheduling system of intelligent factory of chemical fiber, belong to the technical field of chemical fiber production scheduling.The system is by data processing module to the whole production data is preprocessed to generate production dataset;Multi-constraint module is to the whole process core constraint condition of chemical fiber multi-species mixed production quantification characterization and integrated modeling, constructs multi-constraint collaborative scheduling model;Intelligent dynamic scheduling module is based on rolling window mechanism division scheduling period, decomposition scheduling task, combined with reinforcement learning adaptive strategy adjustment algorithm parameter, triggers real-time production scheduling scheme when re-scheduling;Scheduling feedback module generates equipment execution instruction and collects execution data, generates scheduling execution feedback dataset after arrangement.The present application solves the problem of complex constraint, dynamic disturbance response lag in the scheduling of multi-species mixed production of chemical fiber, realizes the accurate, efficient dynamic scheduling of multi-species mixed production, improves the flexibility and reliability of production scheduling.

Description

Technical Field

[0001] This invention belongs to the field of chemical fiber production scheduling technology, specifically involving a dynamic scheduling system for multi-variety mixed production based on a smart chemical fiber factory. Background Technology

[0002] With the intelligent upgrading of the chemical fiber industry and the diversification of market demand, multi-variety, small-batch mixed production has become the mainstream production mode in intelligent chemical fiber factories. The demand for co-production of different varieties and specifications of chemical fiber products such as PET (polyester), PA (polyamide), and PP (polypropylene) is increasing, placing higher demands on the accuracy, flexibility, and real-time nature of production scheduling. Chemical fiber production is characterized by complex processes, long production cycles, strong equipment interoperability, and diverse constraints. Under the multi-variety mixed production mode, the process routes for different varieties of chemical fibers differ significantly, and equipment compatibility requirements are strict. Combined with multiple constraints such as order delivery dates, material supply, and energy consumption, the difficulty of production scheduling is greatly increased. Existing scheduling systems are no longer able to adapt to the actual needs of complex production scenarios, and their main shortcomings severely restrict the production efficiency and market competitiveness of intelligent chemical fiber factories: The constraints involved in the mixed production of multiple chemical fiber varieties are multi-dimensional and strongly coupled, including constraints related to process route differences, equipment capacity and compatibility, order delivery dates and priorities, material supply timing and matching, and energy consumption thresholds. Among these, the process routes for different types of chemical fibers differ significantly. For example, PET spinning requires drying, melting, spinning, cooling, and drawing, while PA spinning requires strict control of the degree of polymerization and spinning temperature. Different process routes place significant differences in the requirements for equipment parameters and process durations. Equipment capacity has a clear upper limit, and different equipment has varying compatibility with different types of chemical fibers. Some equipment can only be used for single-variety production, while others can be used for multiple varieties but require parameter adjustments. These constraints interact and restrict each other, resulting in complex coupling relationships. Most existing scheduling systems only perform simple modeling for single-type constraints, lacking the ability to systematically sort out, quantitatively represent, and integrate the core constraints of the entire process. This results in incomplete consideration of constraints, low modeling accuracy, and an inability to accurately reflect the actual complexity of multi-variety mixed production. Consequently, scheduling schemes are prone to become disconnected from actual production, such as mismatch between equipment allocation and process requirements, and asynchronous material supply and production progress, affecting production continuity.

[0003] The production process of synthetic fibers is characterized by its strong continuity and weak resistance to interference. Various dynamic disturbances can easily occur during production, including emergency order insertions, equipment malfunctions, raw material supply delays, abnormal process parameters, and energy supply fluctuations. Emergency order insertions are a common disturbance scenario in synthetic fiber production, where customers temporarily add orders or adjust order specifications, requiring rapid integration into the existing production plan. Equipment malfunctions, including equipment downtime and decreased precision, can directly lead to production interruptions or substandard product quality. Raw material supply delays may occur due to insufficient supplier capacity or logistical disruptions, resulting in raw material shortages and affecting production progress. Abnormal process parameters and energy supply fluctuations can affect production stability and product qualification rates. Existing scheduling systems mostly adopt a static scheduling mode, which formulates a fixed production schedule based on initial production data and constraints. They lack an efficient dynamic response mechanism, cannot capture dynamic disturbance events in the production process in real time, and cannot quickly adapt to changes in production status caused by disturbance events. They often require manual intervention to readjust the scheduling plan, which not only results in a delayed response but also easily leads to problems such as low scheduling efficiency, delayed order delivery, high equipment idle rate, and energy waste, seriously affecting the company's customer satisfaction and production efficiency.

[0004] Furthermore, existing scheduling systems employ algorithms lacking adaptive adjustment capabilities, mostly relying on traditional algorithms with fixed parameters, making it difficult to optimize algorithm parameters in real time based on changes in dynamic production scenarios. In multi-product mixed production scheduling, the synergy between global and local scheduling is crucial. Existing algorithms often struggle to balance the integrity of global scheduling with the flexibility of local scheduling, resulting in local optimization failing to achieve global optimization, and insufficient optimality and adaptability of scheduling schemes. Simultaneously, algorithm parameters cannot adaptively adjust based on the type and severity of dynamic disturbances and the current production status, making it difficult to balance the accuracy and efficiency of scheduling calculations, failing to meet the high requirements of accuracy, efficiency, and flexibility for multi-product mixed production in the chemical fiber industry. In summary, existing scheduling systems have significant shortcomings in multi-constraint modeling, dynamic response, and algorithm optimization, making them unsuitable for the complex scenarios of multi-product mixed production in intelligent chemical fiber factories. Developing a dynamic scheduling system for multi-product mixed production that can accurately handle multiple constraints, quickly respond to dynamic disturbances, and achieve efficient adaptive scheduling has become a key technical problem to be solved in the current upgrade of intelligent chemical fiber factories, and is of great significance for promoting the intelligent and efficient development of the chemical fiber industry. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a dynamic scheduling system for multi-variety mixed production based on a smart chemical fiber factory. The objective of this invention can be achieved through the following technical solutions: A multi-variety mixed production dynamic scheduling system based on a smart chemical fiber factory includes a data processing module, a multi-constraint module, an intelligent dynamic scheduling module, and a scheduling feedback module; The data processing module acquires all production data during the multi-variety mixed production process of the intelligent chemical fiber factory, and generates a production dataset after preprocessing. The multi-constraint module quantifies and integrates the core constraints of the entire process of multi-variety mixed production of chemical fibers, quantifies various constraints, and constructs a multi-constraint collaborative scheduling model in combination with the optimization objectives of multi-variety mixed production. The intelligent dynamic scheduling module acquires dynamic disturbance event information of the entire process of mixed production of multiple chemical fiber varieties, divides the scheduling cycle based on the rolling window mechanism, and decomposes the global scheduling task into continuous sub-window local scheduling tasks. It performs local optimal scheduling calculation within each sub-window. It introduces a reinforcement learning adaptive strategy to adjust the algorithm parameters in real time. When a dynamic rescheduling instruction is triggered, it calls the production dataset and the multi-constraint collaborative scheduling mathematical model to recalculate and generate the production scheduling scheme in real time. The scheduling feedback module obtains equipment execution instructions based on the production scheduling plan, acquires and organizes scheduling execution process data, and generates a scheduling execution feedback dataset.

[0006] Specifically, the preprocessing process includes: Outlier detection is performed on the full production data, outliers are identified and marked, and interpolation is used for completion processing; redundancy removal processing is performed on the data, and duplicate and invalid redundant data are removed by comparing data similarity. The processed non-standardized data is normalized, and all data is mapped to a preset range. Through structured organization, the data is classified and archived according to data type and production process to generate the production dataset.

[0007] Specifically, the specific process of the quantification representation includes: The core constraints of the entire process of multi-variety mixed production of chemical fibers are systematically classified, divided according to constraint attributes, and the influencing factors corresponding to each type of constraint are clarified. A hierarchical quantification approach is adopted to divide the impact factors of each constraint into two levels: core level and secondary level, according to their importance. Different quantification standards and quantification precision are set for the impact factors of different levels. The quantization conversion algorithm converts the influencing factors of each constraint into quantifiable and computable feature parameters. The quantified feature parameters are then standardized to generate the constraint quantification representation results.

[0008] Specifically, the integrated modeling process includes: Obtain the quantization threshold, constraint range, and applicable scenarios of various quantized constraints, clarify the coupling relationship between various constraints, and distinguish between three types of constraints: collaborative constraints, mutually exclusive constraints, and independent constraints. The constraint coupling modeling method is adopted, which establishes an association mapping relationship for collaborative constraints, sets conflict resolution rules for mutually exclusive constraints, and defines separate modeling logic for independent constraints. All quantified constraints are hierarchically integrated according to the process flow sequence and equipment relationships to construct a unified constraint modeling system, clarifying the hierarchy and calling rules of each constraint in the modeling system.

[0009] Specifically, the quantification process for these various constraints includes: Differentiated quantization conversion logic is set for different types of constraints. During the quantization conversion process, the quantization threshold, fluctuation range and allowable deviation of each constraint are defined, and the priority weight of each constraint is set.

[0010] Specifically, the construction process of the multi-constraint cooperative scheduling model includes: The optimization objectives for multi-variety mixed production are divided into core optimization objectives and auxiliary optimization objectives. Each optimization objective is converted into an objective function, and the calculation logic and weight allocation of each objective function are clarified. The quantified constraints are used as constraint boundaries. The constraint boundaries are integrated according to constraint priority, and the effective conditions and triggering mechanisms of the constraint boundaries are clarified. The objective function and constraint boundary are integrated, the initial parameters and solution range of the multi-constraint collaborative scheduling model are set, the initial configuration is performed, and the initial constraint parameters and objective function weights are imported.

[0011] Specifically, the specific process of the scrolling window mechanism includes: Based on the overall production cycle, process duration, and average response time of dynamic disturbance events of multi-variety mixed production of chemical fibers, a fixed duration of the rolling window is set, and the overlap ratio between sub-scheduling windows is preset according to the scheduling accuracy requirements and production flexibility needs. According to the time sequence, the global scheduling cycle is divided into consecutive sub-scheduling windows starting from the initial time point and with fixed durations. Adjacent sub-scheduling windows partially overlap according to the overlap ratio. Define the time range, production processes covered, and scheduling task scope of the sub-scheduling window, record the start and end times, and generate a rolling window division list.

[0012] Specifically, the process of decomposing the global scheduling task into consecutive sub-window local scheduling tasks includes: Based on the overall goal, total number of tasks and time requirements of the global scheduling task, the global task is decomposed into independent production sub-tasks, each sub-task corresponding to a specific product, process and equipment; Obtain the time range, production load limit, and suitable equipment resources for each sub-scheduling window. Based on the duration and production load capacity of each sub-scheduling window, and combined with the process priority, equipment adaptability, and time requirements of the production sub-tasks, allocate each production sub-task to the corresponding sub-scheduling window.

[0013] Specifically, the process of calculating the locally optimal scheduling includes: Invoke the preset improved intelligent scheduling algorithm, load the local scheduling task list, equipment resource information and multi-constraint collaborative scheduling model of the corresponding sub-scheduling window, and determine the local scheduling target; The production subtasks within the sub-scheduling window are sorted by priority. Combined with the constraints in the multi-constraint collaborative scheduling model, the appropriate equipment, execution order, and start and end times of each production subtask are determined. Through iterative calculation, the local optimal scheduling result is obtained. The equipment allocation information, process execution sequence, and time nodes of each task are recorded, and a local scheduling list for the sub-scheduling window is generated.

[0014] Specifically, the specific process of the reinforcement learning adaptive strategy includes: Construct a reinforcement learning model, defining a state space, an action space, and a reward function. The state space is a set of dynamic disturbance event features and current production state parameters, the action space is a set of adjustable core parameters for improving the intelligent scheduling algorithm, and the reward function is a quantitative calculation function for the satisfaction of the scheduling optimization objective. The system acquires dynamic disturbance event characteristics and current production status data in real time, inputs them into the reinforcement learning model as environmental input, and calculates the optimal action under the current state. Based on the optimal action, it calculates the adjustment amount of the core parameters of the improved intelligent scheduling algorithm in real time, and dynamically optimizes the core parameters of crossover probability, mutation probability and iteration number.

[0015] Specifically, the process of recalculating and generating the production scheduling plan in real time includes: Upon receiving a dynamic rescheduling trigger instruction, terminate the currently executing production scheduling scheme, call the latest production dataset and multi-constraint collaborative scheduling model, update constraint parameters and production status parameters, and clarify the scope and degree of the impact of dynamic disturbance events on the current production. The scheduling calculation process is re-executed, the local scheduling results of each sub-scheduling window are integrated, the rationality and constraint satisfaction of the overall scheduling plan are verified, a new production scheduling plan that matches the current production status and constraints is generated, and the dynamic rescheduling calculation is completed.

[0016] Specifically, the process of generating the scheduling execution feedback dataset includes: According to the preset collection frequency, the full execution data during the scheduling process is acquired in real time, and the time node, data type and corresponding production link of the data collection are clearly defined; The full execution data is classified and organized, and archived hierarchically according to equipment type, production task, and time series. Feature extraction is performed on the organized full execution data to extract feature parameters of scheduling execution progress, equipment operating status, and task completion quality. The organized full execution data and the extracted feature parameters are then structured and integrated to generate the scheduling execution feedback dataset.

[0017] The beneficial effects of this invention are as follows: This invention systematically quantifies and integrates the core constraints of the entire process of mixed production of multiple chemical fiber varieties through a multi-constraint module. It clarifies the quantification logic, coupling relationship and priority of various constraints, and constructs a multi-constraint collaborative scheduling model. This effectively solves the problems of incomplete constraint consideration and low modeling accuracy in the prior art. It can describe the complexity of mixed production of multiple varieties and provides an accurate and reliable modeling foundation for subsequent dynamic scheduling, ensuring that the scheduling scheme meets the actual production constraint requirements.

[0018] This invention uses an intelligent dynamic scheduling module as its core, combining a rolling window mechanism with a reinforcement learning adaptive strategy to decompose global scheduling tasks into sub-window local scheduling tasks, achieving a synergy between local and global optima. Simultaneously, through a reinforcement learning model, it adapts to dynamic disturbances in real time, automatically adjusting the core parameters of the algorithm. This enables rapid response to dynamic disturbances such as emergency order insertions and equipment failures, and real-time recalculation to generate the optimal scheduling scheme. This solves the problems of delayed response and excessive manual intervention in existing static scheduling systems, improving the dynamic adaptability and response efficiency of the scheduling system.

[0019] This invention forms a closed-loop scheduling link of "data preprocessing - multi-constraint modeling - dynamic scheduling - execution feedback" through the coordinated operation of four major modules. The data processing module ensures the integrity and accuracy of production data, while the scheduling feedback module provides real-time feedback on the scheduling execution status, providing data support for dynamic scheduling optimization, realizing continuous optimization of scheduling schemes, reducing order delivery delay rate, equipment idle rate and energy consumption, and improving the production efficiency and product quality of multi-variety mixed production of chemical fibers. Attached Figure Description

[0020] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0021] Figure 1 This is a system architecture diagram of a multi-variety mixed production dynamic scheduling system based on a smart chemical fiber factory according to the present invention; Figure 2 This is a flowchart of the intelligent dynamic scheduling decision-making process in this invention; Figure 3 This is a data flow diagram of scheduling execution and feedback in this invention. Detailed Implementation

[0022] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0023] Please see Figures 1-3 A multi-variety mixed production dynamic scheduling system based on a smart chemical fiber factory includes a data processing module, a multi-constraint module, an intelligent dynamic scheduling module, and a scheduling feedback module. The data processing module acquires all production data during the multi-variety mixed production process of the intelligent chemical fiber factory, and generates a production dataset after preprocessing. The multi-constraint module quantifies and integrates the core constraints of the entire process of multi-variety mixed production of chemical fibers, quantifies various constraints, and constructs a multi-constraint collaborative scheduling model in combination with the optimization objectives of multi-variety mixed production. The intelligent dynamic scheduling module acquires dynamic disturbance event information of the entire process of mixed production of multiple chemical fiber varieties, divides the scheduling cycle based on the rolling window mechanism, and decomposes the global scheduling task into continuous sub-window local scheduling tasks. It performs local optimal scheduling calculation within each sub-window. It introduces a reinforcement learning adaptive strategy to adjust the algorithm parameters in real time. When a dynamic rescheduling instruction is triggered, it calls the production dataset and the multi-constraint collaborative scheduling mathematical model to recalculate and generate the production scheduling scheme in real time. The scheduling feedback module obtains equipment execution instructions based on the production scheduling plan, acquires and organizes scheduling execution process data, and generates a scheduling execution feedback dataset.

[0024] Specifically, the preprocessing process includes: Outlier detection is performed on all production data. The 3σ principle is used to identify outliers, which are then marked and filled in using interpolation. Next, the data undergoes redundancy removal by comparing data similarity to eliminate duplicate and invalid redundant data. Then, the processed non-standardized data is normalized, mapping all data to a preset range to eliminate dimensional differences between different data types. Finally, all preprocessed data is structured and archived according to data type and production stage, generating a standardized production dataset to ensure its integrity, accuracy, and accessibility.

[0025] Specifically, the specific process of the quantification representation includes: The core constraints of the entire process of multi-variety mixed production of chemical fibers are systematically classified and decomposed, and divided into five categories according to constraint attributes: process, equipment, order, material, and energy. The core influencing factors corresponding to each type of constraint are identified, and redundant factors irrelevant to scheduling modeling are eliminated. Subsequently, a hierarchical quantification method is adopted to classify the core influencing factors of each constraint into two levels according to their importance: core level and secondary level. Differentiated quantification standards and quantification precision are set for the influencing factors of different levels. Finally, through quantification conversion algorithms, the core influencing factors of each constraint are converted into quantifiable and calculable feature parameters one by one. The quantified feature parameters are standardized to eliminate dimensional differences and form standardized constraint quantification representation results, thus completing the quantification representation of various constraints.

[0026] Specifically, the integrated modeling process includes: The quantification thresholds, constraint ranges, and applicable scenarios for various quantified constraints were analyzed to clarify the coupling relationships between them, distinguishing between three types: collaborative constraints, mutually exclusive constraints, and independent constraints. Subsequently, a constraint coupling modeling method was adopted to establish association mapping relationships for collaborative constraints, set conflict resolution rules for mutually exclusive constraints, and define separate modeling logic for independent constraints, eliminating logical conflicts and redundant associations between various constraints. Finally, all quantified constraints were hierarchically integrated according to process flow sequence and equipment association relationships to construct a unified constraint modeling system. The hierarchy and calling rules of each constraint in the modeling system were clarified to ensure that the integrated constraint system can accurately adapt to multi-product mixed production scheduling scenarios.

[0027] Specifically, the quantification process for these various constraints includes: A pre-defined constraint quantization algorithm is employed, with differentiated quantization conversion logic set for different types of constraints. Specifically, process constraints are converted into interval-type mathematical expressions based on process parameter thresholds; equipment constraints are converted into limit-value-type mathematical expressions based on equipment capacity limits; order constraints are converted into time-node-type mathematical expressions based on delivery requirements; material constraints are converted into time-series-type mathematical expressions based on supply sequence; and energy constraints are converted into threshold-type mathematical expressions based on energy consumption thresholds. During the quantization conversion process, the quantization thresholds, fluctuation ranges, and allowable deviations for each constraint are clearly defined, and priority weights are set for each constraint, with core constraints having higher priority than auxiliary constraints. After quantization, all mathematical expressions undergo syntax validation to ensure they are calculable and callable, verifying the consistency between the quantization results and actual production constraints, and ensuring the quantization results are compatible with subsequent scheduling model construction.

[0028] Specifically, the construction process of the multi-constraint cooperative scheduling model includes: The optimization objectives for multi-product mixed production are broken down into core optimization objectives and auxiliary optimization objectives. Core optimization objectives include order delivery compliance rate and equipment load balancing rate, while auxiliary optimization objectives include energy consumption control rate and material utilization rate. Each optimization objective is converted into a model-recognizable objective function, and the calculation logic and weight allocation of each objective function are clarified. Subsequently, the quantified constraints are used as the constraint boundaries of the model, and the constraint boundaries are integrated according to constraint priority, clarifying the effective conditions and triggering mechanisms of the constraint boundaries. Then, mathematical modeling methods are used to integrate the objective functions and constraint boundaries, set the initial parameters and solution range of the model, and complete the construction of the multi-constraint collaborative scheduling model. Finally, the model is initialized and configured, the initial constraint parameters and objective function weights are imported, and the initialization state of the model is verified to ensure that the model can be called and solved normally.

[0029] Specifically, the specific process of the scrolling window mechanism includes: By combining the overall production cycle, process duration, and average response time of dynamic disturbance events in the mixed production of multiple chemical fiber varieties, a fixed duration for the rolling window is set. Simultaneously, based on scheduling accuracy requirements and production flexibility needs, the overlap ratio between sub-windows is set, with the overlap ratio controlled within a preset reasonable range. Then, following the time sequence, the global scheduling cycle is divided into multiple consecutive sub-scheduling windows starting from the initial time point, with fixed durations, ensuring that each sub-window is sequentially connected and that adjacent sub-windows partially overlap according to a preset overlap ratio. After division, the time range, covered production processes, and scheduling task scope of each sub-scheduling window are defined, and the start and end times of each sub-window are recorded to form a rolling window division list, providing a basis for subsequent local scheduling task allocation.

[0030] Specifically, the process of decomposing the global scheduling task into consecutive sub-window local scheduling tasks includes: The overall scheduling task is comprehensively reviewed to clarify the overall goal, total task volume, and time requirements. The overall task is then broken down into multiple independent production sub-tasks, each corresponding to a specific product type, process, and equipment. Subsequently, the time range, production load limit, and suitable equipment resources of each sub-scheduling window are obtained. Based on the duration and production load capacity of each sub-scheduling window, combined with the process priority, equipment compatibility, and time requirements of the production sub-tasks, each production sub-task is assigned to its corresponding sub-scheduling window. After allocation, the local scheduling goals, task list, task priorities, and completion deadlines of each sub-window are clarified. The connection relationship between the local scheduling tasks of each sub-window and other sub-window tasks is defined to avoid task conflicts and connection gaps, thus completing the decomposition of the overall scheduling task.

[0031] Specifically, the process of calculating the locally optimal scheduling includes: The system invokes a pre-defined improved intelligent scheduling algorithm, loads the local scheduling task list, equipment resource information, and multi-constraint collaborative scheduling model for the corresponding sub-window, and clarifies the local scheduling objective for that sub-window. Then, the production sub-tasks within the sub-window are sorted by priority, and the appropriate equipment, execution order, and start and end times for each production sub-task are determined one by one, based on the constraints in the multi-constraint collaborative scheduling model. During the calculation process, the system verifies in real-time whether each task allocation scheme meets the constraints. For issues such as equipment conflicts and process conflicts, conflict resolution strategies are used to adjust and optimize task allocation and time scheduling. Finally, through iterative algorithm calculation, the system obtains the locally optimal scheduling result that satisfies the local scheduling objective of the sub-window and meets all constraints. The system records the equipment allocation information, process execution sequence, and time nodes for each task, forming the local scheduling list for the sub-window.

[0032] Specifically, the specific process of the reinforcement learning adaptive strategy includes: A reinforcement learning model is constructed, defining its state space, action space, and reward function. The state space is a set of dynamic disturbance event features and current production state parameters, the action space is a set of adjustable core parameters for improving the intelligent scheduling algorithm, and the reward function is a quantitative calculation function for the satisfaction of the scheduling optimization objective. Subsequently, dynamic disturbance event features and current production state data are collected in real time and input into the reinforcement learning model as environmental input. The model calculates the optimal action for the current state. Based on the optimal action, the adjustment amount of the algorithm's core parameters is calculated in real time, dynamically optimizing core parameters such as crossover probability, mutation probability, and number of iterations. After adjustment, the effect of parameter adjustment is verified to ensure that the adjusted algorithm can adapt to the current dynamic production scenario, achieving a balance between scheduling calculation accuracy and efficiency, and continuously iterating and optimizing the parameter adjustment strategy.

[0033] Specifically, the process of recalculating and generating the production scheduling plan in real time includes: Upon receiving a dynamic rescheduling trigger signal, the system immediately terminates the currently executing production scheduling plan, sends scheduling pause instructions to all production equipment and related units, clears the current scheduling calculation cache, and deletes invalid intermediate scheduling calculation data. Then, it calls the latest production dataset and multi-constraint collaborative scheduling model to update the constraint parameters and production status parameters in the model, clarifying the scope and extent of the impact of dynamic disturbance events on current production. Next, it re-executes the scheduling calculation process, re-dividing the scheduling cycle and decomposing the global scheduling task based on a rolling window mechanism, performing local optimal scheduling calculations within each sub-window, and adjusting algorithm parameters using a reinforcement learning adaptive strategy. Finally, it integrates the local scheduling results from each sub-window, verifies the rationality and constraint satisfaction of the overall scheduling plan, generates a new production scheduling plan adapted to the current production status and constraints, and completes the dynamic rescheduling calculation.

[0034] Specifically, the process of generating the scheduling execution feedback dataset includes: According to the preset collection frequency, all execution data during the scheduling process is acquired in real time, clearly defining the time nodes, data types, and corresponding production stages of data collection. The collected execution data is then categorized and organized, archived hierarchically by equipment type, production task, and time series, and abnormal and invalid data is removed. Next, feature extraction is performed on the organized execution data to extract core feature parameters such as scheduling progress, equipment operating status, and task completion quality. Finally, the organized execution data and extracted core feature parameters are structurally integrated to generate a standardized scheduling execution feedback dataset, clearly defining the dataset's storage format and calling rules to ensure that the feedback data accurately matches the calling requirements of the data processing module and the intelligent dynamic scheduling module.

[0035] In this embodiment, the present invention will be further described in detail below with reference to specific embodiments. This embodiment is only used to explain the present invention and does not constitute a limitation on the scope of protection of the present invention. This embodiment is based on a multi-variety (PET, PA, PP) mixed production scenario in a smart chemical fiber factory. It relies on the system described in the present invention to complete dynamic scheduling, and each module works collaboratively. The specific process is as follows, and all processes are represented by letter parameters without involving specific numerical values.

[0036] 1. Data processing module working process The data processing module acquires the full production data from the multi-variety mixed production process in the intelligent chemical fiber factory and initiates a preprocessing process. The specific process is as follows: First, outlier detection is performed on the full production data. The 3σ principle is used to identify outliers in the data. Outliers are marked and interpolated for completion. The completed outlier data is denoted as X1 (X1: completed outlier production data, covering abnormal equipment operation data, abnormal process parameter data, etc., used to ensure the integrity of production data). Then, redundancy removal is performed. A data similarity comparison algorithm is used to calculate the similarity S between any two sets of data (S: a quantified value of the similarity between any two sets of production data, used to determine if the data is duplicated). When S ≥ S0 (S0: a preset similarity threshold, used to define the criteria for determining duplicate data, with a value range of 0-1), it is determined to be duplicate data, and duplicate and invalid redundant data are removed. The processed non-standardized data is then normalized using a linear normalization method, mapping all data to [X...]. m ᵢ n X max Preset interval (X) m ᵢ n : The minimum normalized data value, X maxThe maximum normalized data value and the normalization interval constitute the normalization interval, used to eliminate the dimensional differences between different types of data. Finally, all preprocessed data are structured and archived according to data type (process parameter data, equipment operation data, order data, etc.) and production process (spinning, drawing, winding, etc.), generating a standardized production dataset D1 (D1: standardized production dataset, integrating all preprocessed production data, for subsequent multi-constraint modules and intelligent dynamic scheduling modules to call, ensuring that the data can be directly used for modeling and calculation).

[0037] 2. Working process of the multi-constraint module The multi-constraint module calls relevant data from production dataset D1 (which includes full preprocessed data such as process, equipment, and orders), and completes quantitative characterization, integrated modeling, and multi-constraint collaborative scheduling model construction based on the core constraints of the entire process of multi-variety mixed production of chemical fibers. The specific process is as follows: 2.1 Quantitative Representation of Constraints First, the core constraints of the entire process of multi-variety mixed production of chemical fibers are systematically classified into five categories according to their constraint attributes: process constraints (C1), equipment constraints (C2), order constraints (C3), material constraints (C4), and energy constraints (C5). (C1: Process constraints, covering process routes and process requirements for different varieties of chemical fibers; C2: Equipment constraints, covering equipment processing capacity and compatibility; C3: Order constraints, covering order delivery dates and priorities; C4: Material constraints, covering material supply and matching; C5: Energy constraints, covering energy consumption thresholds and unit energy consumption.) The influencing factors for each type of constraint are clearly defined: the influencing factor for process constraints C1 is the process sequence and process duration differences for different varieties of chemical fibers; the influencing factor for equipment constraints C2 is equipment processing capacity and equipment compatibility; the influencing factor for order constraints C3 is order delivery dates and order priorities; the influencing factor for material constraints C4 is material supply timing and material matching; and the influencing factor for energy constraints C5 is unit energy consumption of equipment and total energy consumption of the plant.

[0038] A hierarchical quantification method is adopted, dividing the impact factors of each constraint into two levels according to their importance: core level (weight coefficient ω1) and secondary level (weight coefficient ω2). (ω1: weight coefficient of core level impact factor, representing the importance of core impact factor in constraint quantification; ω2: weight coefficient of secondary impact factor, representing the importance of secondary impact factor in constraint quantification), where ω1+ω2=1, and ω1>ω2. Differentiated quantification standards and quantification precision are set for impact factors of different levels. The quantification precision of core level impact factor is P1, and the quantification precision of secondary impact factor is P2 (P1: quantification precision of core level impact factor, representing the accuracy of the quantification result of core impact factor; P2: quantification precision of secondary impact factor, representing the accuracy of the quantification result of secondary impact factor), and P1>P2.

[0039] Through a quantization transformation algorithm, the influencing factors of each constraint are converted into quantifiable and calculable characteristic parameters one by one: Process constraint C1 is quantified into process timing parameter T1 and process correlation parameter R1 (T1: process timing parameter, representing the execution order and duration of each process; R1: process correlation parameter, representing the degree of correlation and dependency between different processes); Equipment constraint C2 is quantified into equipment load parameter L1 and equipment adaptability parameter M1 (L1: equipment load parameter, representing the processing load of the equipment; M1: equipment adaptability parameter, representing the adaptability of the equipment to different types of chemical fibers); Order constraint C3 is quantified into order deadline parameter T2 and order priority parameter P3. (T2: Order deadline parameter, representing the latest delivery time required for an order; P3: Order priority parameter, representing the urgency and priority order of different orders); Material constraint C4 is quantified as material supply cycle parameter T3 and material inventory parameter Q1 (T3: Material supply cycle parameter, representing the supply interval of materials; Q1: Material inventory parameter, representing the current inventory quantity of available materials); Energy constraint C5 is quantified as equipment unit energy consumption parameter E1 and factory total energy consumption threshold parameter E2 (E1: Equipment unit energy consumption parameter, representing the energy consumption of a single piece of equipment processing a unit of product; E2: Factory total energy consumption threshold parameter, representing the upper limit of total energy consumption in the factory production process).

[0040] The quantified feature parameters are standardized using the Z-score standardization method. The standardized value Z of each feature parameter is calculated (Z: the standardized value of the feature parameter, used to eliminate the dimensional differences between different feature parameters, facilitating subsequent modeling calculations). The calculation formula is: Z=(x-μ) / σ, where x is the original value of the feature parameter (x: the original collected value of a feature parameter), μ is the mean of the feature parameter of that type (μ: the average value of the same type of feature parameters, used as the benchmark value for standardization calculations), and σ is the standard deviation of the feature parameter of that type (σ: the dispersion of the same type of feature parameters, used to measure the degree of deviation between the original value and the mean). A constraint quantification representation result Z set is generated, namely Z={Z1, Z2, Z3, Z4, Z5} (Z1-Z5: the standardized quantification results corresponding to process, equipment, order, material, and energy constraints, respectively, used for subsequent integrated modeling), which correspond to the standardized quantification results of the five major categories of constraints.

[0041] 2.2 Constraint Integration Modeling Obtain the quantification thresholds, constraint ranges, and applicable scenarios for each type of constraint after quantification. Clarify the coupling relationships between various constraints and distinguish between three types: collaborative constraints, mutually exclusive constraints, and independent constraints. Process constraint C1 and equipment constraint C2 are collaborative constraints (collaborative constraints refer to constraints that cooperate and influence each other and need to be considered simultaneously); equipment constraint C2 and energy constraint C5 are collaborative constraints; order constraint C3 and material constraint C4 are mutually exclusive constraints (mutually exclusive constraints refer to constraints that conflict with each other and need to be selected based on priority) (when material supply is insufficient, order priority needs to be adjusted); each type of constraint is an independent constraint (independent constraints refer to constraints that only affect their own corresponding constraint type and do not have a coupling relationship with other constraints).

[0042] A constraint coupling modeling method is adopted. For collaborative constraints, an association mapping relationship is established, denoted as A1 (A1: collaborative constraint association matrix, used to quantify the degree of association between various collaborative constraints). A1 is a 5×5 matrix, where A1ᵢⱼ represents the degree of collaborative association between the i-th type of constraint and the j-th type of constraint (A1ᵢⱼ: a quantified value of collaborative association, ranging from 0 to 1, where 1 indicates complete collaboration between the two types of constraints, and 0 indicates no collaboration between the two types of constraints), 0≤A1ᵢⱼ≤1, A1ᵢⱼ=1 indicates complete collaboration, and A1ᵢⱼ=0 indicates no collaboration. For mutually exclusive constraints, conflict resolution rules are set. When an order-type constraint C3 conflicts with a material-type constraint C4, they are sorted according to the order priority parameter P3 (P3: the order priority parameter mentioned above, representing the urgency of the order), prioritizing the constraint corresponding to the order with the higher P3 value. For independent constraints, separate modeling logic is defined, and each type of independent constraint constructs its own constraint expression according to its own quantification threshold and range.

[0043] All quantified constraints are hierarchically integrated according to the process flow sequence and equipment relationships to construct a unified constraint modeling system. The hierarchy and calling rules of each constraint in the modeling system are clearly defined: the first layer is process-related constraint C1, which serves as the basic constraint; the second layer is equipment-related constraint C2 and energy-related constraint C5, which serve as the core collaborative constraints; the third layer is order-related constraint C3 and material-related constraint C4, which serve as the dynamic adjustment constraints. When calling, the constraints are verified in sequence according to the hierarchy to ensure the rationality and callability of the constraint system.

[0044] 2.3 Construction of Multi-Constraint Cooperative Scheduling Model First, the optimization objectives for multi-product mixed production are divided into core optimization objectives and auxiliary optimization objectives. The core optimization objectives are order delivery compliance rate F1 and equipment load balancing rate F2, while the auxiliary optimization objectives are energy consumption control rate F3 and material utilization rate F4 (F1: order delivery compliance rate, representing the proportion of on-time delivered orders to total orders; F2: equipment load balancing rate, representing the degree of load balancing among various equipment; F3: energy consumption control rate, representing the degree of compliance with energy consumption control standards; F4: material utilization rate, representing the degree of effective utilization of materials). Each optimization objective is converted into an objective function. The core optimization objective function is F = α1F1 + α2F2 (F: the core optimization objective function value, representing the overall achievement degree of the core optimization objective; α1: the weighting coefficient of order delivery compliance rate; α2: the weighting coefficient of equipment load balancing rate, which are used to allocate the two core optimization objectives). The importance of the objectives is determined by the auxiliary optimization objective function f = β1F3 + β2F4 (f: auxiliary optimization objective function value, representing the overall degree of achievement of the auxiliary optimization objectives; β1: weight coefficient of energy consumption control rate, β2: weight coefficient of material utilization rate, which are used to allocate the importance of the two auxiliary optimization objectives), where α1 and α2 are the weight coefficients of the core objectives, β1 and β2 are the weight coefficients of the auxiliary objectives, and α1 + α2 = 1, β1 + β2 = 1; the calculation logic of each objective function is clarified as follows: F1 = (number of orders delivered on time / total number of orders) × 100%, F2 = 1 - (maximum load of each equipment - minimum load of each equipment) / average load of each equipment, F3 = (standard energy consumption - actual energy consumption) / standard energy consumption × 100%, F4 = (actual material consumption / planned material consumption) × 100%.

[0045] The quantified constraints are used as the model's constraint boundaries. These boundaries are then integrated according to constraint priority, and their effective conditions and triggering mechanisms are clarified: Core constraints (C1, C2, C5) are effective throughout the entire production process, triggered by changes in production status; dynamic adjustment constraints (C3, C4) are effective when order or material status changes, triggered by dynamic disturbance events. The mathematical expressions for the constraint boundaries are as follows: Process-related constraint boundaries: T1m ᵢ n ≤T1≤T 1max (T) 1m ᵢ n The minimum value of the process timing parameter T1, i.e., the shortest execution time of the process; T 1max The maximum value of the process timing parameter T1, i.e., the longest execution time of the process, defines the reasonable range of the process timing. R1∈[R 1m ᵢ n R 1max ](R 1m ᵢ n The minimum value of the process correlation parameter R1, i.e., the lowest degree of correlation between processes; R 1max The maximum value of the process association parameter R1 represents the highest degree of association between processes, and the two define the reasonable range of process association.

[0046] Equipment class constraint boundary: L1≤L 1max (L) 1max : The maximum value of equipment load parameter L1, i.e., the upper limit of the maximum processing load of the equipment), M1∈{M 11 M 12 M 1n} (M 11 ~M 1n The specific values ​​of the equipment adaptation parameter M1 correspond to different product types that the equipment can adapt to, and are used to define the equipment's compatibility range. 11 ~M 1n (Types of products compatible with the equipment).

[0047] Order class boundary constraints: T2≤T 2max (T) 2max : The maximum value of the order deadline parameter T2, i.e., the upper limit of the latest delivery time of the order), P3∈[P 3m ᵢ n P 3max ](P 3m ᵢ n The minimum value of the order priority parameter P3, i.e., the lowest priority; P 3max The maximum value of the order priority parameter P3, i.e., the highest priority, defines the reasonable range of order priority.

[0048] Material class boundary constraints: Q1≥Q 1m ᵢ n (Q) 1m ᵢ n The minimum value of the material inventory parameter Q1, i.e., the minimum material inventory level to ensure normal production operation, is T3≤T. 3max (T) 3maxThe maximum value of the material supply cycle parameter T3, i.e., the longest material supply interval; exceeding this value will cause production interruption.

[0049] Energy-related boundary constraints: E1≤E 1max (E) 1max E1 is the maximum value of the unit energy consumption parameter E1 of the equipment, which is the maximum allowable energy consumption of a single unit of product of a single piece of equipment. ΣE1≤E2 (ΣE1: the sum of the unit energy consumption of all equipment, which is the actual value of the total energy consumption of the factory; E2: the threshold parameter of the total energy consumption of the factory, which is the upper limit of the total energy consumption of the factory. The two ensure that the total energy consumption does not exceed the standard).

[0050] The objective function and constraint boundary are integrated to set the initial parameters and solution range of the multi-constraint collaborative scheduling model. The initial parameters include weight coefficients α1, α2, β1, and β2 (explained earlier, used to assign the importance of each optimization objective during initial modeling), and constraint threshold T. 1m ᵢ n T 1max (As explained above, this is used to define the reasonable range of each constraint). The solution range is the set of all feasible production task allocation schemes X1 (X1: the set of production task allocation schemes, which includes all task allocation methods that satisfy the constraint boundaries, used for subsequent solution of the optimal scheme); perform initial configuration, import initial constraint parameters and objective function weights, calculate the initial objective function value F0 (F0: the initial objective function value, i.e. the degree of achievement of the core optimization objective during initial modeling, used to verify the model initialization state), verify the model initialization state, ensure that the model can be called and solved normally, and complete the construction of the multi-constraint collaborative scheduling model M1 (M1: multi-constraint collaborative scheduling model, which integrates the objective function and constraint boundaries, used for subsequent scheduling calculations).

[0051] 3. Working process of the intelligent dynamic scheduling module The intelligent dynamic scheduling module acquires dynamic disturbance event information for the entire process of multi-variety mixed production of chemical fibers, calls the production dataset D1 and the multi-constraint collaborative scheduling model M1, and combines the rolling window mechanism and reinforcement learning adaptive strategy to complete the scheduling cycle division, task decomposition, local optimum calculation and dynamic rescheduling. The specific process is as follows: 3.1 Rolling window mechanism for dividing scheduling cycles Based on the overall production cycle T4 of mixed production of multiple chemical fiber varieties, the process duration T5, and the average response time T6 of dynamic disturbance events (T4: overall production cycle, i.e., the total time to complete all production tasks; T5: process duration, i.e., the average execution time of a single process; T6: average response time of dynamic disturbance events, i.e., the average time from the discovery of a disturbance to the initial response, the three are used to set the rolling window duration), the fixed duration of the rolling window is set to T7 (T7: fixed duration of the rolling window, i.e., the time length of each sub-scheduling window, used to divide the scheduling cycle). According to the scheduling accuracy requirements and production flexibility requirements, the overlap ratio between sub-scheduling windows is preset to λ1 (λ1: sub-window overlap ratio, the value range is 0<λ1<1, used to ensure the continuity of scheduling and avoid task connection gaps) (0<λ1<1). Following the time sequence, the global scheduling cycle starts from the initial time point t1 and is divided into consecutive sub-scheduling windows of fixed duration T7, i.e., W1=[t1, t1+T7], W2=[t1+λ1T7, t1+λ1T7+T7], ..., W n =[t1+(n-1)λ1T7,t1+(n-1)λ1T7+T7](t1: global scheduling start time, i.e., the base time for scheduling to start; W1~W n Each sub-scheduling window corresponds to a continuous time period and the scheduling tasks within that time period. Adjacent sub-scheduling windows partially overlap according to an overlap ratio λ1. The time range, covered production processes, and scheduling task range of each sub-scheduling window are clearly defined. The start and end times of each sub-window are recorded, and a rolling window partitioning list W={W1, W2, ..., W...} is generated. n} (W: A scrolling window partition list containing information about all sub-scheduling windows for subsequent task decomposition).

[0052] 3.2 Global Scheduling Task Decomposition Based on the overall goal F5, total task N1, and time requirement T4 of the global scheduling task (F5: overall global scheduling goal, i.e., the degree of achievement of the core optimization goal of the entire production process; N1: total task, i.e., the total number of production sub-tasks to be completed; T4: global scheduling time requirement), the global task is decomposed into N1 independent production sub-tasks. Each production sub-task corresponds to a specific product type, process, and equipment. The set of production sub-tasks is denoted as S1 = {S 11 S 12 S 1n} (S1: Set of production subtasks, containing all independent production subtasks; S 11 ~S 1n: A single production subtask, each subtask corresponds to a specific product, process and equipment), each subtask S1ᵢ corresponds to product type V1ᵢ, process sequence G1ᵢ and matching equipment K1ᵢ (V1ᵢ: chemical fiber product type corresponding to subtask S1ᵢ; G1ᵢ: process execution sequence corresponding to subtask S1ᵢ; K1ᵢ: matching equipment corresponding to subtask S1ᵢ).

[0053] Obtain the time range, production load limit L2ᵢ, and suitable equipment resource set K2ᵢ for each sub-scheduling window Wᵢ (L2ᵢ: production load limit of sub-window Wᵢ, i.e., the maximum amount of tasks that can be carried within this window; K2ᵢ: suitable equipment resource set of sub-window Wᵢ, i.e., the set of equipment that can be used for production within this window). Based on the duration T7 of each sub-scheduling window (the fixed duration of the rolling window mentioned above) and the production load carrying capacity L2ᵢ, combined with the process priority P4ᵢ, equipment adaptability M2ᵢ, and time requirement T8ᵢ of the production sub-task S1ᵢ (P4ᵢ: process priority of sub-task S1ᵢ, representing the urgency of the sub-task; M2ᵢ: ... The device adaptability of subtask S1ᵢ represents the degree of matching between the subtask and the device; T8ᵢ: the time requirement of subtask S1ᵢ, i.e. the planned execution time of the subtask), uses a task allocation algorithm to allocate each production subtask to the corresponding sub-scheduling window. During the allocation process, the following constraints are met: ΣS1ᵢ≤L2ᵢ (the total number of tasks in each sub-window does not exceed the load limit), M2ᵢ∈K2ᵢ (the device adapted by the subtask belongs to the device resource of the sub-window). The global scheduling task is decomposed to obtain the local scheduling task set S2ᵢ of each sub-window (S2ᵢ: the local scheduling task set of sub-window Wᵢ, which contains all production subtasks allocated to this window).

[0054] 3.3 Local Optimal Scheduling Calculation The pre-defined improved intelligent scheduling algorithm is invoked, loading the local scheduling task list S2ᵢ (the aforementioned set of local scheduling tasks for the sub-window), equipment resource information K2ᵢ (the aforementioned set of equipment resources adapted to the sub-window), and multi-constraint collaborative scheduling model M1 (the aforementioned multi-constraint collaborative scheduling model) for the corresponding sub-scheduling window Wᵢ. The local scheduling target F6ᵢ (F6ᵢ: the local scheduling target of sub-window Wᵢ, i.e., the decomposition target of global target F5 in this sub-window, used to measure the quality of scheduling schemes within this window) (i.e., the decomposition target of global target F5 in this sub-window) is determined. The production sub-tasks S2ᵢ within the sub-scheduling window are sorted according to priority P4ᵢ (the aforementioned sub-task process priority), resulting in the sorted task sequence S3ᵢ=[S2ᵢ1, S2ᵢ2, ..., S2ᵢ] k (S3ᵢ: The sorted sequence of subtasks, arranged from highest to lowest priority, used to determine the execution order of subtasks).

[0055] Based on the constraint boundaries in the multi-constraint collaborative scheduling model M1, the adaptable device K3ᵢⱼ, execution order O1ᵢⱼ, start time t2ᵢⱼ, and end time t3ᵢⱼ for each production subtask S2ᵢⱼ are determined one by one (K3ᵢⱼ: the specific adaptable device for subtask S2ᵢⱼ, selected from the sub-window device resource K2ᵢ; O1ᵢⱼ: the execution order of subtask S2ᵢⱼ, determined based on priority sorting results; t2ᵢⱼ: the start execution time of subtask S2ᵢⱼ; t3ᵢⱼ: the end execution time of subtask S2ᵢⱼ, which define the execution time range of the subtask). During the calculation process, it is verified in real time whether each task allocation scheme meets all constraint conditions. If a task allocation scheme violates the constraint boundaries, an adjustment mechanism is triggered to reallocate devices or adjust the execution order.

[0056] By improving the intelligent scheduling algorithm through iterative calculation, the number of iterations is set to K1 (K1: the number of algorithm iterations, i.e., the maximum number of iterations the algorithm needs to find a local optimum, used to control computational accuracy and efficiency). Each iteration calculates the objective function value F6ᵢ of the current task allocation scheme. k (F6ᵢ) k The local objective function value in the k-th iteration (representing the degree to which the optimization objective of the iteration scheme is achieved) is compared with the difference ΔF1=F6ᵢ between the objective function values ​​of two adjacent iterations. k -F6ᵢ k-1 (ΔF1: the difference in objective function values ​​between two adjacent iterations, used to determine whether the algorithm has converged), when ΔF1 < ΔF 01 (ΔF) 01 : A preset convergence threshold is used to determine whether the algorithm has reached a convergence state. If the difference is less than this value, the algorithm is considered to have converged and a local optimum has been obtained (ΔF). 01 When the preset convergence threshold is reached, the iteration stops, the local optimal scheduling result is obtained, the equipment allocation information K3ᵢⱼ, the process execution sequence O1ᵢⱼ and the time node [t2ᵢⱼ, t3ᵢⱼ] of each task are recorded, and the local scheduling list P1ᵢ of the sub-scheduling window is generated (P1ᵢ: the local scheduling list of the sub-window Wᵢ, which contains the execution information of all sub-tasks in the window, and is used for subsequent global scheduling integration).

[0057] 3.4 Reinforcement Learning Adaptive Strategy for Adjusting Algorithm Parameters A reinforcement learning model is constructed, defining its state space S2, action space A2, and reward function R1 (S2: the state space of the reinforcement learning model, used to describe the current production environment state; A2: the action space of the reinforcement learning model, used to describe the parameter adjustment actions that the algorithm can execute; R1: the reward function of the reinforcement learning model, used to evaluate the quality of the actions): The state space S2 is a set of dynamic disturbance event features E3 and current production state parameters S1, i.e., S2 = {E3, S1} (E3: dynamic disturbance event features, representing the type, scope of influence, etc. of the disturbance event; S1: current production state parameters, representing the current equipment operation, task progress, and other production states), where E3 represents the type and degree of influence of the dynamic disturbance event, and S1 represents parameters such as equipment operation status and task completion progress; the action space A2 is a set of adjustable core parameters for improving the intelligent scheduling algorithm, i.e., A2 = {A2, A2, R1}. 11 A 12 A 13} (A 11 : Improve the crossover probability of the intelligent scheduling algorithm to control the search range of the algorithm; A 12 Improve the mutation probability of the intelligent scheduling algorithm to avoid getting trapped in local optima; A 13 : Improve the number of iterations in the intelligent scheduling algorithm to control the computational accuracy of the algorithm, where A 11 For crossover probability, A 12 For the mutation probability, A 13 The iteration number is denoted as γ1. The reward function R1 is a quantitative calculation function for the satisfaction of the scheduling optimization objective, R1=γ1F6ᵢ+γ2F5 (γ1: weight coefficient of local objective F6ᵢ, γ2: weight coefficient of global objective F5, used to allocate the importance of local and global objectives in the reward evaluation), where γ1 and γ2 are weight coefficients, and γ1+γ2=1.

[0058] The system acquires dynamic disturbance event features E3 (the dynamic disturbance event features mentioned above) and current production state data S1 (the current production state parameters mentioned above) in real time. These are input into the reinforcement learning model as environmental input. The model calculates the optimal action A2*∈A2 under the current state (A2*: the optimal action under the current state, i.e., the algorithm parameter adjustment scheme most suitable for the current production state). Based on the optimal action A2*, the system calculates in real time the adjustment amount ΔA1 of the core parameters of the improved intelligent scheduling algorithm (ΔA1: the adjustment amount of the core algorithm parameters, i.e., the specific value that the parameters need to be adjusted, used to optimize the algorithm parameters), and dynamically optimizes the crossover probability A. 11 Probability of mutation A 12 and the number of iterations A 13 The adjusted formula is: A k+1 =A k +ΔA1(A k: Core parameter values ​​of the algorithm before adjustment; A k+1 : Adjusted core algorithm parameter values ​​(used to update algorithm parameters), where A k To adjust the parameter values ​​before adjustment, A k+1 These are the adjusted parameter values. After adjustment, verify the effect of parameter adjustment by calculating the objective function value F7 of the algorithm (F7: the objective function value after parameter adjustment, used to evaluate the effect of parameter adjustment). If F7 > F5 (F5: the global scheduling target value mentioned above, i.e., the baseline value of the objective function before adjustment) (objective function value before adjustment), then retain the adjusted parameters; otherwise, recalculate the adjustment amount and continuously iterate to optimize the parameter adjustment strategy.

[0059] 3.5 Dynamic Rescheduling Calculation When a dynamic rescheduling trigger instruction is received (such as an emergency order insertion or equipment failure), the currently executing production scheduling plan is immediately terminated. The latest production dataset D2 (D2: the standardized production dataset after dynamic disturbance, containing all production data after the disturbance event, used for rescheduling calculation) and the multi-constraint collaborative scheduling model M1 are called to update the constraint parameters and production status parameters in the model. The scope and degree of impact of the dynamic disturbance event on the current production are clarified. The scope of impact is denoted as S3 shadow and the degree of impact is denoted as D1 shadow (S3 shadow: the scope of impact of the dynamic disturbance event, i.e. the scope of production processes, equipment and tasks involved in the disturbance event; D1 shadow: the degree of impact of the dynamic disturbance event, i.e. the quantitative value of the impact of the disturbance on production progress, quality, etc.).

[0060] The scheduling calculation process is re-executed: the scheduling cycle is re-divided based on the rolling window mechanism to obtain a new set of sub-scheduling windows W2 (W2: the set of sub-scheduling windows after the disturbance, the window division is adjusted according to the impact of the disturbance to ensure adaptation to the new production state); the global scheduling task is re-decomposed, and the sub-task allocation scheme is adjusted based on the influence range S3 shadow and the influence degree D1 shadow to obtain a new set of local scheduling tasks for each sub-window S4ᵢ (S4ᵢ: the set of local scheduling tasks for sub-windows Wᵢ after the disturbance, the adjusted sub-task allocation scheme to adapt to the production state after the disturbance); local optimal scheduling calculation is performed within each sub-window, and the algorithm parameters are adjusted based on the reinforcement learning adaptive strategy to obtain a new local optimal scheduling result P2ᵢ (P2ᵢ: the set of sub-scheduling windows after the disturbance, the adjusted sub-scheduling task allocation scheme to adapt to the production state after the disturbance); local optimal scheduling calculation is performed within each sub-window, and the algorithm parameters are adjusted based on the reinforcement learning adaptive strategy to obtain a new local optimal scheduling result P2ᵢ (P2ᵢ: the set of sub-scheduling windows after the disturbance, the adjusted sub-scheduling task allocation scheme to adapt to the production state after the disturbance). The local scheduling list of window Wᵢ is the optimal scheduling result obtained based on the adjusted parameters and tasks; the local scheduling results P2ᵢ of each sub-window are integrated to verify the rationality and constraint satisfaction of the overall scheduling scheme, and the overall objective function value F8 is calculated (F8: the global objective function value after perturbation, used to evaluate the optimization effect of the new scheduling scheme). If F8≥F5 (F5: the global scheduling total objective value mentioned above, i.e., the preset target threshold) (preset target threshold), then a new production scheduling scheme P1 matching the current production status and constraints is generated (P1: the new production scheduling scheme after perturbation, used to guide subsequent production execution); otherwise, the scheduling parameters are readjusted and the calculation is repeated until a production scheduling scheme that meets the requirements is generated, completing the dynamic rescheduling calculation.

[0061] 4. Working process of the scheduling feedback module The scheduling feedback module, based on the production scheduling scheme P1 generated by the intelligent dynamic scheduling module, decomposes it into the equipment execution instructions I1={I} for each production equipment, material supply unit, and energy control unit. 11 I 12 , ..., I 1n} (I1: The set of equipment execution instructions, containing the specific execution requirements of all production units; I 11 ~I 1n The instruction I1 is issued to the corresponding unit to ensure the scheduling plan is implemented. At a preset collection frequency f1 (f1: execution data collection frequency, i.e. how often to collect scheduled execution data for real-time monitoring), the full execution data D3 (D3: full scheduled execution data, including all execution-related data such as equipment operation, task progress, and quality) is acquired in real time during the scheduled execution process. The data collection time node t4, data type, and corresponding production stage are clearly defined (t4: data collection time node, i.e., the specific time of each data collection to ensure data timeliness).

[0062] The full execution data D3 is categorized and organized, and then archived hierarchically according to equipment type K4, production task S1, and time series t5 (K4: equipment type classification, used to organize data by equipment type; t5: data time series, used to organize data in chronological order to ensure data order). Abnormal and invalid data in the execution data are removed. Feature extraction is performed on the organized full execution data D3 to extract feature parameters of schedule execution progress P5, equipment operating status S2, and task completion quality Q2 (P5: schedule execution progress parameter, representing the completion rate of the current schedule; S2: equipment operating status parameter). Q1: Characterizes the current operating status of the equipment; Q2: Task completion quality parameters, characterizing the degree to which the completed tasks meet quality standards); The processed full execution data D3 and the extracted feature parameters are structurally integrated to generate a standardized scheduling execution feedback dataset D4 (D4: scheduling execution feedback dataset, containing all execution-related data and feature parameters, used to feed back to the data processing module and the intelligent dynamic scheduling module), clarifying the storage format and calling rules of the dataset, and feeding D4 back to the data processing module and the intelligent dynamic scheduling module to provide data support for subsequent data preprocessing and dynamic scheduling optimization, forming a closed-loop scheduling link.

[0063] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A dynamic scheduling system for multi-variety mixed production based on a smart chemical fiber factory, characterized in that, It includes a data processing module, a multi-constraint module, an intelligent dynamic scheduling module, and a scheduling feedback module; The data processing module acquires all production data during the multi-variety mixed production process of the intelligent chemical fiber factory, and generates a production dataset after preprocessing. The multi-constraint module quantifies and integrates the core constraints of the entire process of multi-variety mixed production of chemical fibers, quantifies various constraints, and constructs a multi-constraint collaborative scheduling model in combination with the optimization objectives of multi-variety mixed production. The intelligent dynamic scheduling module acquires dynamic disturbance event information of the entire process of mixed production of multiple chemical fiber varieties, divides the scheduling cycle based on the rolling window mechanism, and decomposes the global scheduling task into continuous sub-window local scheduling tasks. It performs local optimal scheduling calculation within each sub-window. It introduces a reinforcement learning adaptive strategy to adjust the algorithm parameters in real time. When a dynamic rescheduling instruction is triggered, it calls the production dataset and the multi-constraint collaborative scheduling mathematical model to recalculate and generate the production scheduling scheme in real time. The scheduling feedback module obtains equipment execution instructions based on the production scheduling plan, acquires and organizes scheduling execution process data, and generates a scheduling execution feedback dataset.

2. The system according to claim 1, characterized in that, The specific process of the preprocessing includes: Outlier detection is performed on the full production data, outliers are identified and marked, and interpolation is used for completion processing; redundancy removal processing is performed on the data, and duplicate and invalid redundant data are removed by comparing data similarity. The processed non-standardized data is normalized, and all data is mapped to a preset range. Through structured organization, the data is classified and archived according to data type and production process to generate the production dataset.

3. The system according to claim 1, characterized in that, The specific process of the quantitative characterization includes: The core constraints of the entire process of multi-variety mixed production of chemical fibers are systematically classified, divided according to constraint attributes, and the influencing factors corresponding to each type of constraint are clarified. A hierarchical quantification approach is adopted to divide the impact factors of each constraint into two levels: core level and secondary level, according to their importance. Different quantification standards and quantification precision are set for the impact factors of different levels. The quantization conversion algorithm converts the influencing factors of each constraint into quantifiable and computable feature parameters. The quantified feature parameters are then standardized to generate the constraint quantification representation results.

4. The system according to claim 1, characterized in that, The specific process of the integrated modeling includes: Obtain the quantization threshold, constraint range, and applicable scenarios of various quantized constraints, clarify the coupling relationship between various constraints, and distinguish between three types of constraints: collaborative constraints, mutually exclusive constraints, and independent constraints. The constraint coupling modeling method is adopted, which establishes an association mapping relationship for collaborative constraints, sets conflict resolution rules for mutually exclusive constraints, and defines separate modeling logic for independent constraints. All quantified constraints are hierarchically integrated according to the process flow sequence and equipment relationships to construct a unified constraint modeling system, clarifying the hierarchy and calling rules of each constraint in the modeling system.

5. The system according to claim 1, characterized in that, The specific process for quantifying the various constraints includes: Differentiated quantization conversion logic is set for different types of constraints. During the quantization conversion process, the quantization threshold, fluctuation range and allowable deviation of each constraint are defined, and the priority weight of each constraint is set.

6. The system according to claim 1, characterized in that, The specific construction process of the multi-constraint cooperative scheduling model includes: The optimization objectives for multi-variety mixed production are divided into core optimization objectives and auxiliary optimization objectives. Each optimization objective is converted into an objective function, and the calculation logic and weight allocation of each objective function are clarified. The quantified constraints are used as constraint boundaries. The constraint boundaries are integrated according to constraint priority, and the effective conditions and triggering mechanisms of the constraint boundaries are clarified. The objective function and constraint boundary are integrated, the initial parameters and solution range of the multi-constraint collaborative scheduling model are set, the initial configuration is performed, and the initial constraint parameters and objective function weights are imported.

7. The system according to claim 1, characterized in that, The specific process of the scrolling window mechanism includes: Based on the overall production cycle, process duration, and average response time of dynamic disturbance events of multi-variety mixed production of chemical fibers, a fixed duration of the rolling window is set, and the overlap ratio between sub-scheduling windows is preset according to the scheduling accuracy requirements and production flexibility needs. According to the time sequence, the global scheduling cycle is divided into consecutive sub-scheduling windows starting from the initial time point and with fixed durations. Adjacent sub-scheduling windows partially overlap according to the overlap ratio. Define the time range, production processes covered, and scheduling task scope of the sub-scheduling window, record the start and end times, and generate a rolling window division list.

8. The system according to claim 1, characterized in that, The specific process of decomposing the global scheduling task into continuous sub-window local scheduling tasks includes: Based on the overall goal, total number of tasks and time requirements of the global scheduling task, the global task is decomposed into independent production sub-tasks, each sub-task corresponding to a specific product, process and equipment; Obtain the time range, production load limit, and suitable equipment resources for each sub-scheduling window. Based on the duration and production load capacity of each sub-scheduling window, and combined with the process priority, equipment adaptability, and time requirements of the production sub-tasks, allocate each production sub-task to the corresponding sub-scheduling window.

9. The system according to claim 1, characterized in that, The specific process of calculating the local optimal scheduling includes: Invoke the preset improved intelligent scheduling algorithm, load the local scheduling task list, equipment resource information and multi-constraint collaborative scheduling model of the corresponding sub-scheduling window, and determine the local scheduling target; The production subtasks within the sub-scheduling window are sorted by priority. Combined with the constraints in the multi-constraint collaborative scheduling model, the appropriate equipment, execution order, and start and end times of each production subtask are determined. Through iterative calculation, the local optimal scheduling result is obtained. The equipment allocation information, process execution sequence, and time nodes of each task are recorded, and a local scheduling list for the sub-scheduling window is generated.

10. The system according to claim 1, characterized in that, The specific process of the reinforcement learning adaptive strategy includes: Construct a reinforcement learning model, defining a state space, an action space, and a reward function. The state space is a set of dynamic disturbance event features and current production state parameters, the action space is a set of adjustable core parameters for improving the intelligent scheduling algorithm, and the reward function is a quantitative calculation function for the satisfaction of the scheduling optimization objective. The system acquires dynamic disturbance event characteristics and current production status data in real time, inputs them into the reinforcement learning model as environmental input, and calculates the optimal action under the current state. Based on the optimal action, it calculates the adjustment amount of the core parameters of the improved intelligent scheduling algorithm in real time, and dynamically optimizes the core parameters of crossover probability, mutation probability and iteration number.

11. The system according to claim 1, characterized in that, The specific process of recalculating and generating the production scheduling plan in real time includes: Upon receiving a dynamic rescheduling trigger instruction, terminate the currently executing production scheduling scheme, call the latest production dataset and multi-constraint collaborative scheduling model, update constraint parameters and production status parameters, and clarify the scope and degree of the impact of dynamic disturbance events on the current production. The scheduling calculation process is re-executed, the local scheduling results of each sub-scheduling window are integrated, the rationality and constraint satisfaction of the overall scheduling plan are verified, a new production scheduling plan that matches the current production status and constraints is generated, and the dynamic rescheduling calculation is completed.

12. The system according to claim 1, characterized in that, The process of generating the scheduling execution feedback dataset includes: According to the preset collection frequency, the full execution data during the scheduling process is acquired in real time, and the time node, data type and corresponding production link of the data collection are clearly defined; The full execution data is classified and organized, and archived hierarchically according to equipment type, production task, and time series. Feature extraction is performed on the organized full execution data to extract feature parameters of scheduling execution progress, equipment operating status, and task completion quality. The organized full execution data and the extracted feature parameters are then structured and integrated to generate the scheduling execution feedback dataset.

Images