Garment production intelligent ranking method and system based on Internet of Things

Abstract

The invention belongs to the technical field of industrial Internet of Things and intelligent manufacturing, and particularly relates to an intelligent ranking method and system for garment production based on the Internet of Things. The method comprises the following steps: retrieving matched candidate genotypes from a production mode gene map library and analyzing the candidate genotypes into an initial ranking scheme; calculating a personnel-equipment-process collaborative adaptation degree coefficient of each station in the scheme; an optimization objective function fusing the collaborative adaptation degree, the historical environment adaptation degree and the production constraint is constructed, and an optimal ranking scheme is solved and output by adopting an improved genetic algorithm; and dynamically constructing a process flow potential energy field in production execution, and generating and executing a task shunting scheduling instruction according to a potential energy gradient. According to the method, the whole-process intelligent ranking from historical experience reuse and global multi-objective optimization to dynamic real-time scheduling is realized, the technical problems that production scheduling depends on experience, the optimization objective is single and dynamic response is insufficient in the flexible clothing production line are effectively solved, and the production scheduling efficiency, the production line balance rate and the system adaptive capacity are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of industrial Internet of Things (IoT) and intelligent manufacturing technology, and in particular to an IoT-based intelligent scheduling method and system for garment production. Background Technology

[0002] As the garment manufacturing industry transforms towards a flexible production model of multiple varieties and small batches, the complexity of production line layout and dynamic scheduling is becoming increasingly prominent. Currently, the technical solutions in this field are mainly divided into two categories: The first category is semi-automatic scheduling systems based on fixed rules or human experience. These systems rely heavily on the personal experience of the schedulers and find it difficult to effectively quantify complex process, personnel, and equipment constraints into a calculable model, resulting in highly subjective scheduling results and poor reproducibility. The second category is automatic scheduling systems based on traditional operations research optimization algorithms (such as linear programming and heuristic rules). Although these systems can achieve automation to a certain extent, their core limitation is that each scheduling is an independent optimization process starting from scratch. They fail to effectively utilize the personnel-equipment-process adaptation patterns and scheduling experience verified in production practice contained in historically successful scheduling schemes, and they also cannot pre-evaluate the expected stability and effectiveness of the scheduling scheme under similar production environments.

[0003] The main shortcomings of existing technologies are as follows: First, in the ranking scheme generation stage, there is a lack of a mechanism that can automatically extract, encode, and store reusable ranking patterns (genes) from historical successful cases, resulting in each ranking being an independent exploration, unable to inherit and reuse verified effective experience; Second, in the ranking scheme evaluation stage, only the current static resource matching degree is focused on, failing to combine historical data to predict and quantify the adaptability of the scheme under different production environments (such as order urgency, personnel proficiency distribution, and equipment health status); Third, in the scheme decision-making and execution stage, a scheduling mechanism that can perceive the overall blockage situation of the production line in real time and drive the dynamic reallocation of tasks based on the situation gradient has not been established, resulting in the system's delayed response and rigid adjustment to sudden bottlenecks in the production process.

[0004] Therefore, this invention proposes an intelligent scheduling method and system for garment production based on the Internet of Things. Summary of the Invention

[0005] This invention provides an intelligent ranking method and system for garment production based on the Internet of Things. It achieves the reuse of historical experience by constructing a production pattern gene map library, performs multi-objective optimization by integrating collaborative adaptability and historical environment adaptability, and achieves dynamic scheduling through process flow potential energy field. Thus, the ranking decision is upgraded from a static island that relies on experience to a data-driven intelligent closed-loop system with historical learning and real-time self-balancing capabilities, so as to systematically solve the above-mentioned defects.

[0006] This invention provides an intelligent scheduling method for garment production based on the Internet of Things, comprising: Step S1: Retrieve at least one candidate genotype from the production mode genotype library that matches the process characteristics and production line resource status of the current order; wherein the production mode genotype library stores genotype codes generated based on historical successful ranking schemes, and the genotype codes contain at least personnel skill feature vectors, equipment function feature vectors, and logical relationship topologies between processes; Step S2: Parse the retrieved candidate genotypes into an initial ranking scheme, and calculate the synergy fit coefficient of the personnel-equipment-process combination assigned to each workstation in the initial ranking scheme; wherein the synergy fit coefficient is obtained by fusing the quantitative values ​​of personnel skills, equipment capabilities, and process requirements. Step S3: Construct an optimization objective function that focuses on improving the overall co-fitness coefficient, while also considering the historical environmental fitness associated with candidate genotypes and satisfying delivery time and load constraints; based on the initial ranking scheme, use a genetic algorithm to solve the problem, output and execute the optimal ranking scheme; Step S4: During production execution, the process flow potential energy field is dynamically constructed based on the real-time task load and processing capacity of each workstation, the bottleneck workstation in the process flow potential energy field is identified, and task diversion scheduling instructions are generated and executed based on the potential energy gradient from the bottleneck workstation to the adjacent workstation.

[0007] Preferably, step S1 employs a dynamic evolution and multi-level progressive retrieval mechanism, including: When a new scheduling scheme is verified in production and meets the preset performance standards, the personnel skill feature vector, equipment function feature vector, logical relationship topology between processes, and scheduling decision sequence features of the new scheduling scheme are automatically extracted, and encoded as new genotype codes with the execution environment and performance data, and incrementally integrated into the production mode genotype library in a way that minimizes topological conflicts. Coarse-grained screening is performed based on the topological matching degree of the logical relationship between processes to obtain a primary candidate genotype set; In the primary candidate genotype set, fine-grained matching is performed by combining the compatibility of personnel skill feature vectors and equipment function feature vectors to obtain the secondary candidate genotype set. In the secondary candidate genotype set, the performance is ranked according to the reproducibility of historical scheduling decision sequences under the current resource constraints, and candidate genotypes that match the current order process characteristics and production line resource status are output.

[0008] Preferably, step S2 employs a state-aware dynamic skill evolution and fusion method, including: The personnel skill feature vectors, equipment function feature vectors, and logical relationship topology between processes contained in the candidate genotypes are mapped to specific workstation, personnel, equipment, and process allocation relationships to generate an initial ranking scheme. Based on the personnel allocation in the initial ranking scheme, a process proficiency update algorithm is constructed, which includes a time decay function and a reinforcement learning function. The process proficiency is updated positively based on the operator's task completion data and the process proficiency update algorithm, and then decayed based on the non-operation period to obtain the updated process proficiency. Quantitative values ​​of personnel skills are calculated based on the updated process proficiency level; Based on the equipment allocation and IoT data of the equipment in the initial ranking scheme, calculate the performance reduction factor that reflects the current operating health status, and combine the equipment functional compatibility to calculate the quantitative value of equipment capability; Based on real-time collected operator fatigue data and performance reduction coefficients, the weights of personnel skill quantification values ​​and equipment capability quantification values ​​in the fusion calculation are dynamically adjusted, and then fused with process requirement quantification values ​​to generate a collaborative fit coefficient.

[0009] Preferably, the calculation of the quantitative value of process requirements is based on the process allocation in the initial ranking scheme, and is generated through a multi-dimensional weighted model according to the standard working hours of the process, the process difficulty coefficient, and the dual requirements of the current order for the accuracy and timeliness of the process.

[0010] Preferably, step S3 employs a reinforcement learning-based method for environment fitness prediction and parameter adaptive optimization, including: The optimization objective function is defined as F=w1*C+w2*E+w3*D, where C represents the overall coordination and adaptability coefficient, E represents the historical environmental adaptability prediction value, D represents the delivery urgency factor, and w1, w2, and w3 are dynamic weight coefficients, and w1+w2+w3=1. The optimization objective function solution must meet the preset constraints of workstation load rate, material flow distance, and production changeover time. Construct a production environment feature space that includes order attributes, personnel status, equipment operating conditions, and material supply; Calculate the multidimensional fuzzy similarity between the current production environment and the genotype historical environment in the production environment feature space, and use the multidimensional fuzzy similarity as a weight to perform weighted fusion of historical performance data to obtain the predicted value E of historical environment fitness. Establish a reinforcement learning environment that rewards the actual comprehensive production benefits of the ranking scheme. The weight coefficients w1, w2, and w3 in the optimization objective function F are used as adjustable actions. Through policy learning, the dynamic adjustment strategy of weights that maximizes long-term rewards is output, and the values ​​of w1, w2, and w3 are adjusted. An initial population is constructed based on the initial ranking scheme. The optimization objective function F with adjusted weight coefficients is applied, and the selection, crossover, and mutation iteration process of the genetic algorithm is executed until the convergence condition is met. The optimal ranking scheme is then output and the production system is driven to execute.

[0011] Preferably, the genetic algorithm employs an adaptive improvement mechanism based on population diversity feedback, including: When the initial population is generated, it is screened based on the cofit coefficient, and only schemes with a cofit coefficient higher than a preset threshold are retained to form the initial population. During the iteration process, the crossover rate and mutation rate are dynamically adjusted based on the constraint satisfaction of the current best individual and the population gene diversity index. An elite retention and adaptive generation gap strategy is adopted to maintain global search capability while ensuring convergence speed.

[0012] Preferably, the dynamic scheduling steps of the potential energy field employ a potential energy propagation model and a historical efficiency path selection method, including: Defining potential energy propagation rules based on process topology network; Calculate the original potential energy generated by each workstation due to its own task load and processing capacity, and calculate the additional potential energy generated by the propagation of the original potential energy from upstream to downstream according to the potential energy propagation rules. Superimpose the original potential energy and the additional potential energy to form a global process flow potential energy field that reflects the risk of systemic blockage. Calculate the potential energy gradient from the bottleneck workstation to each adjacent workstation, combine the historical diversion success rate and average efficiency improvement data of each path, predict the secondary bottleneck risks that diversion may cause, perform multi-objective comprehensive scoring, select the optimal path to generate task diversion scheduling instructions and execute them.

[0013] Preferably, the execution of task diversion and scheduling instructions requires real-time verification of process divisibility and compatibility with target workstation equipment to ensure the feasibility of scheduling actions. After scheduling is executed, the changes in the potential energy field of the process flow are monitored in real time. If a new imbalance is caused, a local rebalancing or global re-optimization is triggered.

[0014] Preferred options also include: Record the entire process data and final production efficiency indicators of each ranking plan from generation, optimization, execution to dynamic scheduling; Based on full-process data, the genotype codes in the production mode gene map library are updated with performance data, the parameters in the co-fitness coefficient calculation model are calibrated, the strategies in the optimization algorithm are reinforced through reinforcement learning, and the propagation rules in the process flow potential field model are corrected.

[0015] This invention provides an intelligent scheduling system for garment production based on the Internet of Things, comprising: The historical pattern retrieval module is used to retrieve at least one candidate genotype from the production pattern genograph library that matches the current order's process characteristics and production line resource status. The production pattern genograph library stores genotype codes generated based on historical successful ranking schemes, and the genotype codes include at least personnel skill feature vectors, equipment function feature vectors, and logical relationship topologies between processes. The adaptation quantification and scheme initialization module is used to parse the retrieved candidate genotypes into an initial ranking scheme and calculate the collaborative adaptation coefficient of the personnel-equipment-process combination assigned to each workstation in the initial ranking scheme; wherein the collaborative adaptation coefficient is obtained by fusing the quantitative values ​​of personnel skills, equipment capabilities, and process requirements. The multi-objective optimization module is used to construct an optimization objective function that focuses on improving the overall co-fitness coefficient, while also considering the historical environmental fitness associated with candidate genotypes and satisfying delivery time and load constraints. Based on the initial ranking scheme, a genetic algorithm is used to solve the problem, output and execute the optimal ranking scheme. The potential energy field dynamic scheduling module is used to dynamically construct the process flow potential energy field based on the real-time task load and processing capacity of each workstation during production execution, identify the bottleneck workstation in the process flow potential energy field, and generate and execute task diversion scheduling instructions based on the potential energy gradient from the bottleneck workstation to the adjacent workstation.

[0016] The beneficial effects of this invention compared to existing technologies are as follows: By searching the production pattern gene map library, historical successful experiences are transformed into reusable ranking genes, realizing the accumulation and inheritance of knowledge; by quantifying the matching quality of human-machine-process through the co-fitness coefficient, and combining it with the dynamic effectiveness of historical environmental fitness prediction schemes, the evaluation shifts from static to dynamic; by constructing a process flow potential energy field and dynamically scheduling according to the potential energy gradient, the production line possesses a real-time response and adjustment capability similar to fluid self-balancing. Its technical effect is that ranking is transformed from a one-time, isolated, and rigid decision into a inheritable, adaptive, and self-balancing continuous optimization process, thereby significantly improving the scientific nature of production scheduling, the balance rate of the production line, and its robustness to production disturbances.

[0017] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a core flowchart of the IoT-based intelligent ranking method for garment production in this embodiment of the invention. Figure 2 This is a diagram of the production mode gene map library architecture in an embodiment of the present invention; Figure 3 This is a diagram illustrating the collaborative working mechanism of multi-objective optimization and potential energy field scheduling in an embodiment of the present invention. Detailed Implementation

[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] like Figure 1 , Figure 2 As shown, this invention provides an embodiment of an intelligent scheduling method for garment production based on the Internet of Things, including: Step S1: Retrieve at least one candidate genotype from the production mode genotype library that matches the process characteristics and production line resource status of the current order; wherein the production mode genotype library stores genotype codes generated based on historical successful ranking schemes, and the genotype codes contain at least personnel skill feature vectors, equipment function feature vectors, and logical relationship topologies between processes; Step S2: Parse the retrieved candidate genotypes into an initial ranking scheme, and calculate the synergy fit coefficient of the personnel-equipment-process combination assigned to each workstation in the initial ranking scheme; wherein the synergy fit coefficient is obtained by fusing the quantitative values ​​of personnel skills, equipment capabilities, and process requirements. Step S3: Construct an optimization objective function that focuses on improving the overall co-fitness coefficient, while also considering the historical environmental fitness associated with candidate genotypes and satisfying delivery time and load constraints; based on the initial ranking scheme, use a genetic algorithm to solve the problem, output and execute the optimal ranking scheme; Step S4: During production execution, the process flow potential energy field is dynamically constructed based on the real-time task load and processing capacity of each workstation, the bottleneck workstation in the process flow potential energy field is identified, and task diversion scheduling instructions are generated and executed based on the potential energy gradient from the bottleneck workstation to the adjacent workstation.

[0022] In this embodiment, the production pattern gene map library refers to a structured database that stores historically optimal production scheduling schemes, with each scheme encoded as a searchable and reusable gene.

[0023] In this embodiment, the current order process characteristics refer to the specific requirements of the order to be produced, including the process list, the standard operating time for each process, quality requirements, and the sequence of processes.

[0024] In this embodiment, the production line resource status refers to the real-time resource status of the garment production line at the time of production scheduling, including the availability and working status of sewing equipment at each workstation, the on-duty status of operators, and their skill level.

[0025] In this embodiment, candidate genotype refers to one or more historical ranking scheme codes that are most similar to the current production needs and retrieved from the production pattern gene map library through a matching algorithm.

[0026] In this embodiment, the genotype encoding generated based on historically successful spawning schemes refers to transforming historically validated efficient and balanced spawning schemes into a data format that can be stored, analyzed, and matched by computers through feature extraction and structured processing. The genotype encoding uses a structured data object representation and includes at least the following fields: GeneID: A unique identifier for a gene.

[0027] ProcessTopology: A topological graph of logical relationships between processes stored in the form of an adjacency matrix or a node-edge list. Node attributes include process ID, standard working hours, and process difficulty coefficient.

[0028] PersonnelSkillVector: An m×n matrix, where m represents the operator IDs involved in the scheme, n represents the process IDs involved, and the matrix elements P {i,j} This represents the proficiency metric of operator i for process j (calculated based on historical task completion data, with an initial value of ...).

[0029] EquipmentFuncVector: A p×q matrix, where p represents the equipment IDs involved in the scheme, q represents the set of equipment function tags (such as flat seam, buttonhole, fastener), and matrix element E. {k,l} This is a Boolean value or compatibility score, indicating whether device k supports or is compatible with function l.

[0030] SchedulingSequence: An ordered list that records the key decision sequence that triggered dynamic scheduling during the execution of this historical plan. Each record may contain {timestamp, triggering workstation, diversion target, diversion task volume, diversion reason code}.

[0031] EnvironmentContext: The associated historical production environment feature vector, such as order delivery urgency, average skill level of the shift team on the day, and overall equipment uptime, is used to calculate environment similarity.

[0032] PerformanceMetrics: Related historical performance data, such as production line balance rate, order completion time, and defect rate. The production pattern gene map library is stored using a graph database (such as Neo4j) or a relational database. Genotypes are used as nodes and connected by relational edges such as similar process topology and applicable similar equipment to support efficient multi-level progressive retrieval.

[0033] In this embodiment, the genotype encoding includes at least personnel skill feature vectors, equipment function feature vectors, and the logical relationship topology between processes: Personnel skill feature vector: A mathematical vector in which each dimension represents a process and the value represents the operator’s proficiency or efficiency score in that process.

[0034] Equipment Function Feature Vector: A mathematical vector in which each dimension represents a process type or process requirement, and the value (usually 0 or 1) indicates whether the equipment supports performing the process.

[0035] Logical relationship topology between processes: A network structure diagram in which nodes represent processes and directed edges represent the sequential dependencies between processes, clearly defining the process route of the order.

[0036] In this embodiment, the initial scheduling scheme refers to the preliminary production plan generated after decoding and instantiating the retrieved candidate genotypes, which clarifies which operator, which equipment, and which process will be performed at each workstation.

[0037] In this embodiment, the personnel-equipment-process combination's compatibility coefficient is a value between 0 and 1, used to quantitatively evaluate the suitability of assigning a specific personnel and equipment combination to perform a specific process. A higher value indicates that the combination is theoretically more efficient and has more stable quality. The calculation of the compatibility coefficient (C): For a specific personnel-equipment-process combination at workstation j, its compatibility coefficient C... j =(w p ×PSV+w e ×ECV) / PRV. Where w p and w e These are dynamic weights, each initially set to 0.5. The dynamic weight adjustment rule is as follows: if Fatigue... i >θ f (Fatigue threshold) and Health m >θ h (Health threshold), then reduce w p Increase w e For example, w p =0.3,w e=0.7; and vice versa. Overall co-fit coefficient C total For all workstations C j The weighted average or geometric average.

[0038] In this embodiment, the Personnel Skill Quantification Value (PSV) is a specific numerical value calculated based on the operator's historical performance data (such as average efficiency and quality pass rate) in the assigned work process, representing their individual ability level in performing that task. The calculation of the Personnel Skill Quantification Value (PSV) is as follows: For operator i assigned to workstation j and performing work process k, their PSV is... {i,k} =S {i,k} ×(1-α×Fatigue i ). Among them, S {i,k} To update the current proficiency score (range 0-1) obtained from the process proficiency update algorithm, Fatigue i To collect and normalize operator fatigue levels in real time (0-1), α is the fatigue influence coefficient (e.g., 0.2).

[0039] In this embodiment, the equipment capability quantification value is a specific numerical value that comprehensively considers the equipment's functional compatibility with the process, its current operational health status (such as failure rate and performance degradation), and historical efficiency, representing its equipment support level for performing the task. For equipment m assigned to workstation j to perform process k, its capability quantification value (ECV) is... {m,k} =Comp {m,k} ×Health m ×BaseScore k Among them, Comp {m,k} For the functional compatibility (0 or 1) of equipment m with process k, Health m BaseScore is a performance reduction factor (0-1) calculated based on IoT data (such as spindle vibration amplitude and motor current stability). k The baseline equipment requirement score (preset constant) for process k.

[0040] In this embodiment, the Process Requirements Quantification (PRV) is a specific numerical value calculated based on factors such as the complexity, precision requirements, and standard working hours of the process itself. It represents the technical difficulty and resource requirements of the task. The calculation of the Process Requirements Quantification (PRV) is as follows: k =β1×NormTime k +β2×DiffCoeff k +β3×PrecisionReq k +β4×UrgencyFactor order NormTime kStandard time for each process (normalized), DiffCoeff k PrecisionReq represents the difficulty level of the manufacturing process. k UrgencyFactor is the quantified value for the accuracy level required by the current order. order β1-β4 are configurable weights for order delivery urgency, and Σβ=1.

[0041] In this embodiment, the historical environmental fitness associated with the candidate genotype refers to the actual production efficiency, balance rate, and other performance indicators achieved by the historical ranking scheme represented by the candidate genotype in environments similar to the past and current production scenarios (such as order type and work team composition). It is used to predict the potential effectiveness of adopting a similar ranking scheme in the future.

[0042] In this embodiment, the optimization objective function that satisfies the delivery date and load constraints is a mathematical model or calculation formula. Its core is to find the highest collaborative fit coefficient and the highest historical environment adaptability while ensuring that the entire production plan can meet the final delivery deadline of the order and that the workload allocated to each workstation is within its capacity.

[0043] In this embodiment, the optimal scheduling scheme refers to the production schedule with the highest comprehensive score obtained after iteratively improving the initial scheduling scheme through an optimization algorithm, while satisfying all constraints. This scheme will be determined as the final execution scheme.

[0044] In this embodiment, a process flow potential energy field is dynamically constructed based on the real-time task load and processing capacity of each workstation: During production execution, the system continuously collects data and calculates a potential energy value for each workstation. This value is typically proportional to the amount of backlog of tasks currently awaiting processing at that workstation and inversely proportional to its processing speed. The potential energy values ​​of all workstations collectively form a dynamic and visualized field, intuitively reflecting the congestion status of the production line.

[0045] In this embodiment, bottleneck workstations in the process flow potential energy field are identified. This refers to the automatic identification of critical nodes currently restricting the production line within the dynamically generated process flow potential energy field using a pre-defined bottleneck determination algorithm. The core logic of this algorithm is as follows: the system traverses all workstations in the potential energy field and calculates the ratio of its own potential energy value to the average potential energy value of all its directly adjacent (determined by process topology) upstream and downstream workstations for each workstation. When this ratio for a workstation continuously exceeds a preset bottleneck determination threshold (e.g., 1.5 times) and persists for a preset number of monitoring cycles, the system determines that the workstation is a bottleneck workstation. The threshold value is used to quantify the bottleneck, upstream and downstream adjacency is defined by process topology, and persistence is constrained by time cycles, thus transforming the vague bottleneck concept into a stable and clearly defined logical rule that can be automatically executed.

[0046] In this embodiment, the potential energy gradient from the bottleneck station to adjacent stations refers to the calculated potential energy difference and direction between the bottleneck station and each of its adjacent stations. The larger the difference and the direction pointing towards the station with lower potential energy, the clearer the potential path for pressure release and the more significant the effect may be.

[0047] In this embodiment, based on the potential energy gradient from the bottleneck workstation to the adjacent workstation, a task diversion scheduling instruction is generated: The system automatically analyzes the potential energy gradient in each direction, selects the direction with the largest gradient (i.e., the most effective path for pressure release), and combines the task divisibility, target workstation equipment compatibility and other verification conditions to generate a specific scheduling command (such as transferring X semi-finished products from bottleneck workstation A to workstation B) to guide the rapid relief of the bottleneck on site.

[0048] To achieve intelligent inheritance and efficient reuse of historical best scheduling experience, and avoid the inefficiency of having to explore from scratch for each scheduling, a dynamic evolution and multi-level progressive retrieval mechanism is proposed for step S1, including: When a new scheduling scheme is verified in production and meets the preset performance standards, the personnel skill feature vector, equipment function feature vector, logical relationship topology between processes, and scheduling decision sequence features of the new scheduling scheme are automatically extracted, and encoded as new genotype codes with the execution environment and performance data, and incrementally integrated into the production mode genotype library in a way that minimizes topological conflicts. Coarse-grained screening is performed based on the topological matching degree of the logical relationship between processes to obtain a primary candidate genotype set; In the primary candidate genotype set, fine-grained matching is performed by combining the compatibility of personnel skill feature vectors and equipment function feature vectors to obtain the secondary candidate genotype set. In the secondary candidate genotype set, the performance is ranked according to the reproducibility of historical scheduling decision sequences under the current resource constraints, and candidate genotypes that match the current order process characteristics and production line resource status are output.

[0049] In this embodiment, the preset performance standard refers to a series of quantitative indicator thresholds that are pre-set to determine whether a ranking scheme is successful or excellent, such as the overall production line balance rate reaching more than 85%, the on-time delivery rate of orders reaching more than 95%, and the comprehensive utilization rate of equipment exceeding 80%.

[0050] In this embodiment, the new ranking scheme has been verified in production to meet the preset performance standards: This means that after the system generates and executes a new ranking scheme, the key performance indicators (such as balance rate and delivery rate) calculated by collecting actual production data all meet the above-mentioned preset performance standards, proving that the scheme is effective in practice and thus obtains the qualification to enter the gene bank.

[0051] In this embodiment, the personnel skill feature vector, equipment function feature vector, process topology relationship, and scheduling decision sequence features of the new ranking scheme are extracted and encoded into a new genotype code by associating them with the execution environment and performance data. This means the system automatically analyzes the successful ranking scheme, extracts its personnel allocation pattern (encoded as a personnel skill feature vector), equipment usage pattern (encoded as an equipment function feature vector), and process logical structure (encoded as a process topology relationship), and records the scheduling decision sequence adopted when facing unexpected situations. Simultaneously, the production environment during scheme execution (such as order urgency and personnel status) and the final achieved actual performance data (such as actual working hours and quality data) are associated with this set of features, packaged together, and structured and stored as a new, searchable genotype archive.

[0052] In this embodiment, the production mode gene map library is incrementally integrated in a way that minimizes topological conflicts: This means that when adding newly generated genotype codes to the map library, it is not a simple addition, but rather an analysis of the similarities and differences between its process topology and the existing genes in the library, finding the most suitable classification position, and adjusting the internal associations of the library (such as similarity links) to ensure that the addition of new knowledge does not disrupt the overall structure and retrieval logic of the library, thus achieving smooth accumulation and fusion of knowledge.

[0053] In this embodiment, coarse-grained screening is performed based on the matching degree of process topology relationships to obtain a preliminary candidate genotype set: This means that during the search, the similarity between the process flow diagram (process topology relationship) of the new order and the process flow diagram stored for each gene in the gene bank is first compared. All genes with a similarity exceeding a low threshold are screened out to form a preliminary, relatively large candidate set. This step mainly ensures that the ranking scheme is feasible in terms of overall process logic.

[0054] In this embodiment, a secondary candidate genotype set is obtained by performing fine-grained matching based on the compatibility of personnel skill feature vectors and equipment functional feature vectors within the primary candidate genotype set. This involves further comparison within the set obtained from the initial screening. The skill feature vectors of currently available personnel and the functional feature vectors of available equipment on the production line corresponding to the new order are matched with the personnel and equipment feature vectors recorded in each candidate genotype to assess whether current resources can support the personnel and equipment configuration mode in the genotype scheme. Genotypes with high compatibility are selected to form a more precise and smaller candidate set.

[0055] In this embodiment, current resource constraints refer to the hard conditions that the production site must comply with when making production scheduling decisions. These mainly include: 1) Personnel constraints: which operators are on duty and their skill levels; 2) Equipment constraints: which equipment is available and its functional status; 3) Material / site constraints: material supply status, workstation space layout, etc.

[0056] In this embodiment, the reproducibility of historical scheduling decision sequences under current resource constraints refers to evaluating the historical scheduling decision sequences recorded in the candidate genes (such as transferring the task to device B and operator C when device A fails) to determine whether these decisions can be fully or largely reproduced under current resource constraints (e.g., whether device B is currently idle and whether operator C is on duty and has the necessary skills). Higher reproducibility means that the historical solution remains effective in the current environment when dealing with similar problems.

[0057] In this embodiment, within the secondary candidate genotype set, performance is ranked based on the reproducibility of historical scheduling decision sequences under current resource constraints. The resulting candidate genotypes that match the current order's process characteristics and production line resource status are then output. This means that in the refined set, candidate genes are ultimately ranked based on reproducibility. Genes whose historical scheduling decisions are most easily and completely reproduced under current resource conditions are preferentially selected, as they are considered to be most adaptable to the dynamic changes in the current production environment. This results in the output of one or more candidate genotypes with the highest final matching degree for subsequent analysis steps.

[0058] To accurately quantify and dynamically evaluate the real-time matching degree among personnel, equipment, and processes, and to ensure that the ranking scheme is based on objective data rather than subjective experience, a state-aware dynamic skill evolution and fusion method is proposed for step S2, including: The personnel skill feature vectors, equipment function feature vectors, and logical relationship topology between processes contained in the candidate genotypes are mapped to specific workstation, personnel, equipment, and process allocation relationships to generate an initial ranking scheme. Based on the personnel allocation in the initial ranking scheme, a process proficiency update algorithm is constructed, which includes a time decay function and a reinforcement learning function. The process proficiency is updated positively based on the operator's task completion data and the process proficiency update algorithm, and then decayed based on the non-operation period to obtain the updated process proficiency. Quantitative values ​​of personnel skills are calculated based on the updated process proficiency level; Based on the equipment allocation and IoT data of the equipment in the initial ranking scheme, calculate the performance reduction factor that reflects the current operating health status, and combine the equipment functional compatibility to calculate the quantitative value of equipment capability; Based on real-time collected operator fatigue data and performance reduction coefficients, the weights of personnel skill quantification values ​​and equipment capability quantification values ​​in the fusion calculation are dynamically adjusted, and then fused with process requirement quantification values ​​to generate a collaborative fit coefficient.

[0059] In this embodiment, the personnel skill feature vectors, equipment function feature vectors, and logical relationship topologies between processes contained in the candidate genotypes are mapped to specific workstation, personnel, equipment, and process allocation relationships to generate an initial ranking scheme: this refers to instantiating the retrieved abstract genes. The system determines the execution order and workstation layout of each process based on the process logical relationship topology, and then assigns suitable operators to each process (or workstation) based on the personnel skill feature vectors and suitable sewing equipment to each process (or workstation) based on the equipment function feature vectors, thereby generating a preliminary production plan that contains specific personnel, equipment, workstation, and process correspondences and can be directly executed or optimized.

[0060] In this embodiment, based on the personnel allocation in the initial ranking scheme, a process proficiency update algorithm is constructed, which includes a time decay function and a reinforcement learning function: Time decay function: A mathematical model that defines how an operator's skills will become rusty over time. For example, if an operator does not perform a certain procedure for N consecutive days, their proficiency score for that procedure will gradually decrease according to a preset rule.

[0061] Reinforcement learning function: Defines a mathematical model that shows how an operator's skills improve through successful practice. When an operator successfully completes a task, the system provides a positive reward based on their efficiency and quality, thereby increasing their proficiency score for that task according to preset rules.

[0062] In this embodiment, operator task completion data refers to the data related to the operator automatically collected by the system through IoT terminals (such as sewing machine data interfaces and quality inspection systems) after each production task is completed. This data mainly includes: actual operation time (compared with standard working hours), output, defect rate, number of equipment start-ups and shutdowns, etc., which are used to objectively evaluate the operator's task completion status.

[0063] In this embodiment, the process proficiency is updated forward based on the operator's task completion data and the process proficiency update algorithm, and decayed according to the inactivity period to obtain the updated process proficiency. This is a process of dynamically maintaining an individual's skill profile. The system periodically (e.g., daily or per shift) performs the following operations: 1) For the process that the operator has just completed, using their task completion data as input, a reinforcement learning function is run; if the performance is excellent, the corresponding proficiency is increased; 2) For all processes that the operator has not recently performed, a time decay function is run, decreasing the corresponding proficiency according to a set period and decay rate. Finally, an updated process proficiency reflecting the operator's latest true skill level is obtained.

[0064] In this embodiment, the inactivity period refers to the length of time an operator has continuously failed to perform a specific procedure, typically calculated in days or shifts. This parameter is a key input to the time decay function; the longer the period, the greater the potential for skill decay.

[0065] In this embodiment, based on the device allocation and IoT data of the devices in the initial ranking scheme, a performance reduction factor reflecting the current operational health status is calculated, and a quantitative value of device capability is calculated in combination with device functional compatibility: Equipment allocation: refers to the plan of which equipment will be used for which process.

[0066] Device IoT data: refers to the real-time collection of device operating parameters.

[0067] Performance degradation factor calculation: The system maintains a device health rating table, comparing IoT data (such as temperature and current) with preset normal value ranges and deducting points based on the degree of deviation. The performance degradation factor is calculated using formula 1 - (total deductions / maximum allowable deductions), with the result limited to between 0 and 1 (1 represents complete health).

[0068] Calculation of equipment capability quantification value: First, query the equipment's functional compatibility identifier for this process (1 for compatible, 0 for incompatible). Then, use the formula: Equipment capability quantification value = Functional compatibility identifier × Performance reduction factor × Standard process baseline score of the equipment. The standard process baseline score is a pre-calibrated score reflecting the inherent capability of the equipment to perform this process under ideal conditions.

[0069] In this embodiment, the performance degradation factor for the current operating health status is a value between 0 and 1 (1 represents complete health, 0 represents complete failure). It is calculated by analyzing the device's IoT data and is used to quantify the degree of degradation of the device's current performance relative to its brand-new or standard state. For example, a device that frequently reports errors or has unstable speed will have a lower coefficient.

[0070] In this embodiment, device functional compatibility is defined as a Boolean value, typically 0 or 1, or a compatibility score between 0 and 1, indicating whether a device possesses the basic functionality required to perform a specific process. For example, a standard flatbed sewing machine cannot perform the buttonhole process, therefore its compatibility with the buttonhole process is 0.

[0071] In this embodiment, the real-time operator fatigue data refers to data that reflects the operator's current physiological and psychological fatigue state, such as continuous working hours, movement frequency, and heart rate variability, obtained through wearable devices, workstation terminal interaction, or estimation based on working hours models. This data is used to assess the potential negative impact on the operator's current work performance.

[0072] In this embodiment, based on real-time collected operator fatigue data and performance reduction coefficients, the weights of personnel skill quantification values ​​and equipment capability quantification values ​​in the fusion calculation are dynamically adjusted, and then fused with process requirement quantification values ​​to generate a collaborative fit coefficient. This means that when calculating the final collaborative fit coefficient, real-time status is introduced to dynamically correct the basic capabilities. The system determines whether the human state or the equipment state has a greater impact based on the current operator fatigue data and equipment performance reduction coefficients. If the personnel are very fatigued, the weight of personnel skill quantification values ​​in the fusion formula is reduced, and the weight of equipment capability quantification values ​​is increased; conversely, the weighting is reversed. Then, using the adjusted weights, the dynamically corrected personnel skill quantification values ​​and equipment capability quantification values ​​are weighted and fused with the process requirement quantification values ​​to finally obtain a collaborative fit coefficient that reflects the real-time matching status.

[0073] To transform abstract process requirements into calculable and comparable quantitative indicators and provide a basis for accurate matching, a method for calculating the quantitative value of process requirements is proposed. Based on the process allocation in the initial ranking scheme, and taking into account the standard working hours of the process, the process difficulty coefficient, and the dual requirements of the current order for the accuracy and timeliness of the process, a multi-dimensional weighted model is used to generate the quantitative value.

[0074] In this embodiment, the calculation of the quantitative value of process requirements is based on the process allocation in the initial ranking scheme. It is generated through a multi-dimensional weighted model, taking into account the standard working hours of the process, the technological difficulty coefficient, and the current order's dual requirements for the accuracy and timeliness of the process. Specifically, for each workstation assigned a specific process, a quantitative value representing the process's requirements for the executor's (personnel and equipment) capabilities is calculated. This value calculation process comprehensively considers multiple factors and uses a weighted model to integrate them.

[0075] In this embodiment, the process allocation in the initial scheduling scheme refers to the specific designation in the preliminary production plan of which one or more specific garment processing processes will be performed at each workstation.

[0076] In this embodiment, the standard working time for a process refers to the time required for a qualified operator to complete a unit output (e.g., one piece) of a specific process under standard working conditions using appropriate equipment. It is a fundamental indicator for measuring the workload of a process.

[0077] In this embodiment, the process difficulty coefficient refers to a pre-set numerical value used to quantify the technical complexity of a certain process. This coefficient comprehensively considers factors such as the skill level required for the process, the complexity of the operation steps, and the requirements for operational stability. For example, simple straight-line sewing has a lower difficulty coefficient, while complex curved stitching or special embroidery has a higher difficulty coefficient.

[0078] In this embodiment, the current order imposes dual requirements on the process: precision and timeliness. Precision requirements refer to the specific requirements of a particular order on a particular process. Precision requirements refer to the quality standard level that the product must achieve in terms of dimensions, stitching, and appearance, such as ordinary tolerances versus high-precision tolerances. Timeliness requirements refer to the urgency of the process or the entire order, such as whether it is an expedited order or whether the process is on the critical path directly affecting the delivery date. These two factors together constitute the contextualized requirements of the process within the specific order context.

[0079] In this embodiment, the multidimensional weighted model refers to a mathematical framework used to integrate the quantitative indicators of multiple different dimensions (standard working hours, process difficulty, accuracy requirements, and timeliness requirements) into a single quantitative value for process requirements. This model assigns a configurable weight to each dimension and calculates the value through weighted summation or other fusion methods (such as multiplication or nonlinear combination). The model aims to ensure that the final quantitative value reflects both the inherent technical difficulty and workload of the process, as well as the specific pressure and requirements brought about by the current order.

[0080] like Figure 3 As shown, to find the global optimal balance point among multiple conflicting optimization objectives and adaptively adjust the optimization direction to adapt to dynamic production needs, step S3 is proposed to employ a reinforcement learning-based environment fitness prediction and parameter adaptive optimization method, including: The optimization objective function is defined as F=w1*C+w2*E+w3*D, where C represents the overall coordination and adaptability coefficient, E represents the historical environmental adaptability prediction value, D represents the delivery urgency factor, and w1, w2, and w3 are dynamic weight coefficients, and w1+w2+w3=1. The optimization objective function solution must meet the preset constraints of workstation load rate, material flow distance, and production changeover time. Construct a production environment feature space that includes order attributes, personnel status, equipment operating conditions, and material supply; Calculate the multidimensional fuzzy similarity between the current production environment and the genotype historical environment in the production environment feature space, and use the multidimensional fuzzy similarity as a weight to perform weighted fusion of historical performance data to obtain the predicted value E of historical environment fitness. Establish a reinforcement learning environment that rewards the actual comprehensive production benefits of the ranking scheme. The weight coefficients w1, w2, and w3 in the optimization objective function F are used as adjustable actions. Through policy learning, the dynamic adjustment strategy of weights that maximizes long-term rewards is output, and the values ​​of w1, w2, and w3 are adjusted. An initial population is constructed based on the initial ranking scheme. The optimization objective function F with adjusted weight coefficients is applied, and the selection, crossover, and mutation iteration process of the genetic algorithm is executed until the convergence condition is met. The optimal ranking scheme is then output and the production system is driven to execute.

[0081] In this embodiment, the optimization objective function is defined as F = w1*C + w2*E + w3*D, where C represents the overall collaborative adaptability coefficient, E represents the predicted historical environmental adaptability value, D represents the delivery urgency factor, and w1, w2, and w3 are dynamic weight coefficients, with w1 + w2 + w3 = 1. For example, w1 can take values ​​of 0.4~0.7, w2 can take values ​​of 0.2~0.4, and w3 can take values ​​of 0.1~0.2. This refers to establishing a comprehensive mathematical standard for evaluating and comparing the merits of different ranking schemes. This standard consists of a weighted sum of three core indicators: the overall synergy and adaptability coefficient (C), which measures the static matching quality of human-machine-process; the historical environmental adaptability prediction value (E), which predicts the stability of the scheme under the current dynamic production environment; and the delivery urgency factor (D), which reflects the urgency of order delivery. The system will dynamically adjust the importance weights (i.e., weights w1, w2, w3) of these three indicators in the final decision based on actual production needs. The sum of the weights of the three indicators is always 1, ensuring the consistency and comparability of the evaluation standard.

[0082] In this embodiment, the optimization objective function must satisfy preset constraints regarding workstation load rate, material flow distance, and changeover time. This means that in the process of finding the optimal ranking scheme using the aforementioned objective function, all considered or generated schemes must first meet a series of rigid production feasibility conditions. These conditions include: the total workload allocated to each workstation cannot exceed its maximum capacity (workstation load rate upper limit); the material transfer path between processes cannot be too long and must be within a reasonable handling distance (material flow distance upper limit); and when the production line switches to produce different products, the required preparation and adjustment time must be controlled within an allowable range (changeover time upper limit). Only schemes that meet all these constraints are eligible for the final comparison of merits.

[0083] In this embodiment, a production environment feature space is constructed, encompassing order attributes, personnel status, equipment operating conditions, and material supply. This refers to the system's multi-dimensional descriptive framework, defined to quantify a specific production scenario at a given moment. This framework breaks down the production environment into several key dimensions, such as: the characteristics of the order itself (e.g., style complexity, batch size), the real-time status of personnel on the production line (e.g., average skill level, attendance rate), the operating status of equipment (e.g., failure rate, utilization rate), and the adequacy and timeliness of material preparation. Each dimension can be represented numerically, and all these numerical values ​​together constitute a feature point representing the production environment. The set of all possible environmental feature points forms the feature space, used to systematically compare the similarity between different environments. A feature vector Env=[f1,f2,...,fn] is defined, where features may include: order batch size (normalized), style complexity score, average process difficulty, average skill level of the work team, overall equipment utilization rate, material availability rate, etc. Each feature is normalized.

[0084] In this embodiment, the multidimensional fuzzy similarity between the current production environment and the historical environment of the genotype in the production environment feature space is calculated. This means that to predict whether a historical ranking scheme (genotype) is applicable in the current situation, the system needs to evaluate the similarity between the environment in which it was successful in the past and the current environment. Specifically, the values ​​of each feature dimension of the current environment are compared one by one with the values ​​of each feature dimension of the historical environment recorded by the genotype. This comparison is not a simple yes or no, but rather calculates a fuzzy similarity value between 0 and 1, where 1 represents complete similarity and 0 represents complete dissimilarity. Since multiple dimensions are involved (such as orders, personnel, equipment, etc.), the final judgment is based on a comprehensive calculation of the similarity of all these dimensions, hence the name multidimensional fuzzy similarity. Current environment feature vector Env current With a certain genotype's historical environmental feature vector Env history The similarity Sim between them is calculated using cosine similarity or the reciprocal of a weighted Euclidean distance. For example, using weighted Euclidean distance: Distance=sqrt(Σ(λ t *(Env current [t]-Env history [t]) 2 )), where λ t Let be the weight of the t-th feature. Then the similarity Sim = 1 / (1 + Distance).

[0085] In this embodiment, historical performance data is weighted and fused using multidimensional fuzzy similarity as the weight to obtain the historical environment fitness prediction value E. This refers to the system using the multidimensional fuzzy similarity between the current environment and each historical environment of a certain genotype as a weight to calculate a weighted average of multiple historical actual production performance data (such as production efficiency and balance rate) associated with that genotype. The principle is: the more similar a historical environment is to the current environment, the greater its actual performance data at that time has reference value for predicting the current possible performance, and the higher its weight. Through this weighted fusion, a single value reflecting the expected performance of the genotype in the current environment is finally calculated, namely the historical environment fitness prediction value E. For a candidate genotype, N associated historical performance data Perf1, Perf2, ..., Perf... are taken. N (e.g., production line balance rate) and their corresponding historical environments Env1,...,Env N Calculate Env separately. current With each Env i Similarity Sim i Then the predicted value E = (Σ(Sim) i *Perf i )) / ΣSim i If historical data for this genotype is insufficient, E can be set as a conservative baseline value.

[0086] In this embodiment, a reinforcement learning environment is established, with the actual comprehensive production benefits of the ranking scheme as the reward: the system optimizes its own decisions by simulating a learning-feedback loop. In this loop, each ranking decision made by the system (outputting a scheme and executing it) is equivalent to taking an action in the environment. Subsequently, the system collects real production result data after the scheme is executed (such as actual production cycle, achievement rate, and quality indicators), and calculates a reward value based on these results to quantify the quality of the decision. The higher the reward value, the better the comprehensive production benefits brought by the decision. The goal of the system is to learn to adjust its own decision-making strategy in order to obtain higher long-term cumulative rewards in the future. Reinforcement learning environment setup: State: The feature vector Env of the current production environment. current Discretized or normalized representations of the data, and category labels for the current production stage (such as regular production scheduling or emergency order insertion).

[0087] Action: Adjustment of the weight coefficients [w1, w2, w3]. The action space can be designed as discrete, for example, with 9 preset combinations: [0.7, 0.2, 0.1], [0.6, 0.3, 0.1], ..., corresponding to strategies emphasizing adaptation, stability, and delivery time, respectively; or it can be designed as continuous, with fine-tuning values ​​of Δw1, Δw2, and Δw3 output by a neural network.

[0088] Reward: After a production scheduling-production cycle, the reward R is calculated based on actual production efficiency: R = η1 * ActualBalanceRate + η2 * (1 - TardinessPenalty) + η3 * (1 - QualityDefectRate) - η4 * ReschedulingCost. Where TardinessPenalty is the delay penalty function, ReschedulingCost is the additional cost incurred due to dynamic scheduling (such as material handling and changeover time), and η1-η4 are the reward weights.

[0089] Learning algorithms: Deep deterministic policy gradient (DDPG) or proximal policy optimization (PPO) algorithms suitable for continuous or high-dimensional discrete action spaces are employed. The agent trains offline or learns online based on historical scheduling-production data (state, action, reward), aiming to maximize long-term cumulative discount rewards.

[0090] In this embodiment, the weight coefficients w1, w2, and w3 in the objective function F are treated as adjustable actions. Through policy learning, a dynamic weight adjustment strategy that maximizes long-term rewards is output. This means that in the reinforcement learning environment described above, the system learns to optimize the specific settings of the three weight coefficients (w1, w2, w3) in the objective function F. The system tries different weight combinations (i.e., different actions) and observes the rewards obtained after executing the final ranking scheme under each combination. Through continuous trial and learning, the system gradually discovers and summarizes patterns: under what production needs or scenario characteristics (e.g., when delivery is particularly urgent, or when personnel skills fluctuate greatly), which weight allocation ratio (i.e., assigning different importance to C, E, and D) can more stably obtain high rewards. Ultimately, the system learns an intelligent strategy that can dynamically adjust weights according to actual conditions, rather than a fixed set of weight values.

[0091] In this embodiment, an initial population is constructed based on the initial ranking scheme. An optimization objective function F with adjusted weight coefficients is applied, and a genetic algorithm's selection, crossover, and mutation iterative process is executed. Specifically, in each round of specific ranking scheme optimization, the system uses a genetic algorithm for searching. First, the initial ranking scheme generated in the previous steps is used as a high-quality seed. By introducing some random variations, a set of similar but slightly different schemes are generated, forming the initial scheme population. Then, the objective function F, composed of the latest weight coefficients (w1, w2, w3) determined through reinforcement learning, is used to evaluate the quality of each scheme in the population (i.e., calculate fitness). Next, the algorithm simulates the biological evolution process: prioritizing schemes with high fitness (selection), allowing them to combine to generate new schemes that integrate the advantages of both (crossover), and allowing small-probability random changes to explore new possibilities (mutation). This process is repeated multiple times (iteration).

[0092] In this embodiment, the optimal ranking scheme is output and the production system is driven to execute until the convergence condition is met: this means that the iterative process of the genetic algorithm will not proceed indefinitely. The system sets convergence conditions for stopping iterations, such as: the fitness of the optimal scheme no longer significantly improves in several consecutive generations of the population, or the preset maximum number of iterations is reached. When any convergence condition is met, the algorithm terminates, and the ranking scheme with the highest fitness found in all iterations is determined as the optimal ranking scheme for this iteration. Subsequently, the system converts this scheme into specific production instructions and issues them to the corresponding workstations, personnel, and equipment on the production line, driving the actual production to begin.

[0093] like Figure 3 As shown, to improve the optimization efficiency and convergence speed of the genetic algorithm while effectively avoiding getting trapped in local optima and ensuring a high-quality global ranking scheme, an adaptive improvement mechanism based on population diversity feedback is proposed for the genetic algorithm, including: When the initial population is generated, it is screened based on the cofit coefficient, and only schemes with a cofit coefficient higher than a preset threshold are retained to form the initial population. During the iteration process, the crossover rate and mutation rate are dynamically adjusted based on the constraint satisfaction of the current best individual and the population gene diversity index. An elite retention and adaptive generation gap strategy is adopted to maintain global search capability while ensuring convergence speed.

[0094] In this embodiment, the preset threshold refers to a numerical standard pre-set by the system for judgment or screening. Different thresholds (e.g., similarity threshold in gene retrieval, fit threshold in algorithm screening, potential energy threshold in bottleneck judgment, etc.) are set at different stages of the invention, depending on specific functional requirements, as demarcation points for automated decision-making.

[0095] In this embodiment, during the initial population generation, a selection process is performed based on the cofitness coefficient, retaining only schemes with a cofitness coefficient higher than a preset threshold to form the initial population. This means that before the genetic algorithm begins its search, a batch of randomly generated or rule-based candidate ranking schemes undergoes preliminary quality filtering. The system calculates the cofitness coefficient of each scheme and then selects only those schemes with coefficient values ​​greater than a preset acceptable threshold as the starting population for the genetic algorithm's evolutionary process. This ensures that the algorithm searches within a high-quality solution space from the outset, improving optimization efficiency.

[0096] In this embodiment, the constraint satisfaction of the current optimal individual refers to the degree to which the solution with the highest objective function value in the current generation of ranked solutions adheres to all hard production constraints (such as workstation load rate, material flow distance, etc.). Satisfaction can be measured by the number or severity of constraint violations; the higher the satisfaction, the more feasible the optimal solution.

[0097] In this embodiment, the population genetic diversity index is a measure used to quantify the degree of difference between all ranking schemes in the current generation. Low diversity occurs when all schemes are very similar in personnel allocation, equipment arrangement, or process sequence; high diversity occurs when the schemes are distinct. This index reflects the breadth of the solution space currently being explored by the algorithm.

[0098] In this embodiment, the crossover rate refers to the proportion of individuals selected for crossover in one iteration of the genetic algorithm. Crossover simulates gene recombination, exchanging parts of the structure of two parent schemes to generate a new scheme. It is a primary means of maintaining population diversity and combining desirable traits.

[0099] In this embodiment, the mutation rate refers to the probability of performing a random mutation operation on an individual in the scheme during one iteration of the genetic algorithm. The mutation operation simulates gene mutation, randomly changing a detail of the scheme (such as changing personnel at a certain workstation), and is a key mechanism for introducing new features and avoiding the algorithm from getting trapped in local optima.

[0100] In this embodiment, during the iteration process, the crossover rate and mutation rate are dynamically adjusted based on the constraint satisfaction of the current best individual and the population gene diversity index. This means that the genetic algorithm does not use fixed parameters during runtime, but intelligently adjusts its behavior according to the current search state. The specific rules are: 1) If the constraint satisfaction of the current best individual is very low (i.e., there is a serious violation), the mutation rate is increased in the hope of escaping the infeasible region through more random changes; 2) If the population gene diversity index is too low (i.e., all schemes are too similar), the crossover rate and mutation rate are increased to promote the generation of new scheme structures and expand the search range; conversely, if the constraint satisfaction is high and the diversity is moderate, the mutation rate is appropriately reduced to facilitate convergence.

[0101] In this embodiment, elite preservation and adaptive generation gap strategies are two mechanisms used to improve standard genetic algorithms. Elite preservation means that in each generation of evolution, the few best solutions from the current generation are forcibly retained to the next generation to prevent the loss of excellent genes. Adaptive generation gap strategy refers to dynamically controlling the proportion of individuals replaced by newly generated offspring in each generation. When the generation gap is large, the update speed is fast and convergence is quick, but it may be unstable; when the generation gap is small, the update is more gradual, which helps maintain diversity. The system automatically adjusts the size of this generation gap according to the evolutionary process (such as the rate of population fitness improvement).

[0102] In this embodiment, an elite retention and adaptive generation gap strategy are employed to maintain global search capability while ensuring convergence speed. This means that by combining these two strategies, the genetic algorithm can find the optimal solution more efficiently and robustly. Elite retention ensures that the algorithm does not degenerate and accelerates the approach to the optimal solution (ensuring convergence speed). The adaptive generation gap strategy balances the need to utilize existing good solutions and explore new possible solutions. When the current generation evolution is progressing smoothly, the generation gap is narrowed for intensive refinement; when it stagnates, the generation gap is widened to introduce more new individuals, thereby effectively maintaining the ability to search the entire solution space (maintaining global search capability) and avoiding premature entrapment in local optima.

[0103] like Figure 3 As shown, to transform the complex production line congestion problem into an intuitive physical field model and intelligently determine the most efficient and lowest-risk real-time scheduling path based on historical data, a dynamic scheduling step using the potential energy field adopts a potential energy propagation model and a historical efficiency path selection method, including: Defining potential energy propagation rules based on process topology network; Calculate the original potential energy generated by each workstation due to its own task load and processing capacity, and calculate the additional potential energy generated by the propagation of the original potential energy from upstream to downstream according to the potential energy propagation rules. Superimpose the original potential energy and the additional potential energy to form a global process flow potential energy field that reflects the risk of systemic blockage. Calculate the potential energy gradient from the bottleneck workstation to each adjacent workstation, combine the historical diversion success rate and average efficiency improvement data of each path, predict the secondary bottleneck risks that diversion may cause, perform multi-objective comprehensive scoring, select the optimal path to generate task diversion scheduling instructions and execute them.

[0104] In this embodiment, a potential energy propagation rule is defined based on the process topology network: This refers to setting mathematical rules for the transfer of potential energy between workstations based on the directed network graph formed by the logical connections such as the sequence and parallelism between various processes on the production line. This rule explicitly specifies how much of the potential energy value of a workstation will propagate along the network lines to its subsequent (downstream) workstations when congestion (high potential energy) occurs, or whether it will be affected by the state of the preceding (upstream) workstations, thus simulating the diffusion effect of production congestion on the assembly line from a mechanistic perspective. Defined on the process topology network, the original potential energy of an upstream workstation u will propagate to its directly downstream workstation d with an attenuation coefficient γ (e.g., γ=0.3), as part of d's additional potential energy. That is, workstation d receives potential energy from all upstream workstations U. d The additional potential energy received is P addedd =γ*Σ(P primaryu ),u∈U d This rule can be applied recursively to multiple levels of propagation, but usually only one or two levels of propagation are considered to control complexity.

[0105] In this embodiment, the inherent potential energy generated by each workstation due to its own task load and processing capacity is calculated: For each specific workstation on the production line, a basic potential energy value is calculated based on its real-time state. This value is primarily proportional to the current amount of tasks awaiting processing at that workstation (e.g., the number of workpieces waiting in queue) and inversely proportional to its standard processing rate (e.g., capacity per unit time). The more tasks accumulate or the slower the processing speed, the higher the inherent potential energy generated by that workstation, indicating that it is already in or nearing a congested state. For workstation j, its inherent potential energy P... primaryj =QueueLength j / ProcessRate j QueueLength j The ProcessRate represents the number of tasks pending processing at this workstation (including those in process). j This represents the current actual or standard processing rate (number of tasks / unit of time) for this workstation.

[0106] In this embodiment, the additional potential energy generated by the propagation of upstream primary potential energy to downstream is calculated according to the potential energy propagation rule. This means that after calculating the primary potential energy of all workstations, the system simulates the additional pressure on downstream workstations caused by congestion at upstream workstations, based on predefined potential energy propagation rules. Specifically, for each downstream workstation, the system iterates through all its directly or indirectly upstream workstations, and calculates how much of the primary potential energy of these upstream workstations will be transmitted to or affect the downstream workstation based on the network connection distance and the attenuation coefficient in the propagation rule. The sum of these transmitted potential energy values ​​is the additional potential energy borne by the downstream workstation. It reflects the potential future congestion risk caused by upstream bottlenecks.

[0107] In this embodiment, the primary potential energy and the additional potential energy are superimposed to form a global process flow potential energy field reflecting the systemic blockage risk: For each workstation, its own generated primary potential energy is added to all the additional potential energy from upstream, resulting in the total potential energy of that workstation. The total potential energy value of all workstations on the entire production line at this moment constitutes a complete and dynamic potential energy field. This field not only shows where there is currently a blockage (high primary potential energy), but also warns of where blockages are about to occur due to upstream issues (high additional potential energy), thus reflecting the blockage risk distribution of the entire production system more comprehensively and proactively. This is represented by P. totalj =P primaryj +P addedj .

[0108] In this embodiment, the potential energy gradient from the bottleneck workstation to each adjacent workstation is calculated. Combined with historical diversion success rates and average efficiency improvement data for each path, the secondary bottleneck risk that diversion may trigger is predicted. This refers to the system calculating the potential energy difference and direction from the bottleneck workstation (the point with the highest total potential energy) to each of its connectable workstations (usually physically adjacent or directly related to processes) after determining the bottleneck workstation (the point with the highest total potential energy). The gradient indicates the potential path for pressure release. Simultaneously, the system queries the knowledge base for historical performance data of each such path when diversion scheduling was performed in the past, including success rate (whether the scheduling was successfully executed and achieved the expected results) and average efficiency improvement (the degree of improvement in the overall production cycle time after scheduling). The system uses this historical data and the current potential energy state to simulate and predict whether, if a certain path is selected for diversion, new congestion points will be triggered at or around the target workstation due to insufficient capacity of the target workstation, material path conflicts, etc., i.e., secondary bottleneck risk. Specifically, the potential energy gradient calculation and path scoring are as follows: For the bottleneck workstation b to each of its adjacent connectable workstations n, the potential energy gradient Grad... {b->n} =P totalb -P totaln The overall score for each path. n =ω1*Grad {b->n} +ω2*HistSuccessRate n +ω3*HistEfficiencyGain n -ω4*PredictedRisk n Where ω1-ω4 are the weights, and HistSuccessRate is... n and HistEfficiencyGain n PredictedRisk retrieves past performance data for this path from the historical scheduling record database. n For risk prediction values ​​(e.g., PredictedRisk) based on the current load and remaining capacity of the target workstation. n =(QueueLengthn +DivertTaskSize) / Capacity n ).

[0109] In this embodiment, the bottleneck station identification algorithm calculates the total potential energy of each station every Δt time interval (e.g., 5 minutes). A threshold ratio R is set. threshold (e.g., 1.5) and the number of continuous monitoring cycles N threshold (e.g., 3). For workstation j, calculate the ratio R of its potential energy to the average potential energy of all its directly adjacent (upstream and downstream) workstations. j =P totalj / Avg(P totaln (eighbors). If R j >R threshold If a workstation j is continuously N, then workstation j is marked as a potential bottleneck. threshold If a monitoring cycle is marked as a potential bottleneck, it will eventually be determined as the current bottleneck workstation, triggering a traffic diversion and scheduling analysis.

[0110] In this embodiment, a multi-objective comprehensive scoring is performed to select the optimal path, generate task diversion scheduling instructions, and execute them. This involves the system establishing an evaluation system to score each available diversion path originating from the bottleneck workstation. The scoring comprehensively considers multiple objectives: 1) the current potential energy gradient (the larger the better, indicating a more direct pressure reduction effect); 2) the historical diversion success rate of the path (the higher the better, indicating strong reliability); 3) the historical average efficiency improvement of the path (the higher the better, indicating significant effect); and 4) the predicted secondary bottleneck risk (the lower the better). The system assigns weights to these objectives and calculates a comprehensive score for each path. For example, the potential energy gradient weight is 0.4, the historical success rate weight is 0.3, the efficiency improvement weight is 0.2, and the secondary risk weight is 0.1. Finally, the path with the highest comprehensive score is selected, generating detailed task diversion scheduling instructions (including which part of the task to divert, when to divert, and to which workstation), and driving the production system to execute them after feasibility verification.

[0111] like Figure 3 As shown, in order to ensure the feasibility of real-time scheduling instructions and prevent new production bottlenecks caused by improper scheduling, and to ensure the stability and reliability of the dynamic adjustment process, it is proposed that the execution of task diversion scheduling instructions should be verified for real-time process divisibility and compatibility with target workstation equipment to ensure the feasibility of scheduling actions. After scheduling is executed, the changes in the potential energy field of the process flow are monitored in real time. If a new imbalance is caused, a local rebalancing or global re-optimization is triggered.

[0112] In this embodiment, the execution of task diversion scheduling instructions requires real-time process divisibility verification and target workstation equipment compatibility verification to ensure the feasibility of the scheduling action. This means that after the system calculates the resulting diversion scheduling instructions based on the potential energy gradient, it does not execute them immediately but first performs two key feasibility verifications. First, it verifies process divisibility, determining whether the task to be diverted (usually a batch of semi-finished products or a process unit) can be split into different times or different workstations for completion, or whether splitting it will affect product quality and process continuity. Second, it verifies target workstation equipment compatibility, confirming whether the workstation planned to receive the diverted task has the necessary functions and precision to execute the diverted task. Only when both verifications pass does the system consider the scheduling instruction safe and feasible and issue it for execution.

[0113] In this embodiment, after scheduling is executed, the changes in the process flow potential energy field are monitored in real time. If a new imbalance is triggered, a local rebalancing or global reoptimization is initiated: This means that after the diversion scheduling command is executed, the system does not stop monitoring but continues to observe the dynamic changes in the process flow potential energy field. It needs to assess whether the scheduling effectively alleviated the original bottleneck, merely shifted the bottleneck to another location, or created multiple new minor congestion points. Once a new imbalance exceeding the allowable range is detected in the potential energy field, the system will immediately activate the response mechanism. If it is only a local, small-scale problem, a local rebalancing is triggered, such as fine-tuning only between a few relevant workstations; if it causes a large-scale disturbance or systemic bottleneck, it is determined that a global reoptimization is required. At this time, the current ranking scheme may be suspended, and the entire multi-objective optimization process starting from step S3 may be restarted to generate a completely new global ranking scheme to thoroughly solve the problem.

[0114] To enable the system's ranking and scheduling capabilities to continuously evolve and optimize with production practice, and to break free from long-term reliance on manual parameter tuning, the following additional measures are proposed: Record the entire process data and final production efficiency indicators of each ranking plan from generation, optimization, execution to dynamic scheduling; Based on full-process data, the genotype codes in the production mode gene map library are updated with performance data, the parameters in the co-fitness coefficient calculation model are calibrated, the strategies in the optimization algorithm are reinforced through reinforcement learning, and the propagation rules in the process flow potential field model are corrected.

[0115] In this embodiment, the entire process data and final production efficiency indicators of each scheduling plan are recorded, from generation, optimization, execution to dynamic scheduling. This means the system collects and embeds data across the entire production chain, taking a complete scheduling and production cycle as the unit. This includes multiple versions of the scheduling plan during the generation phase, parameters and intermediate results during the optimization phase, real-time status and operation records of each workstation during the execution phase, and every instruction triggered during the dynamic scheduling phase and the on-site response. Finally, the system collects efficiency indicators reflecting the final results, such as the quantity, quality, total time, and energy consumption of finished products produced under this plan. These process data and result indicators are then structured and stored according to timelines and causal relationships, forming a complete data package that can be traced and analyzed.

[0116] In this embodiment, based on full-process data, the genotype encoding in the production mode gene map library is updated with performance data, the parameters in the co-fitness coefficient calculation model are calibrated, the strategies in the optimization algorithm are reinforced through learning, and the propagation rules in the process flow potential field model are corrected. This means the system uses the complete data package recorded above to iteratively optimize its core model and algorithm, forming a self-improving closed loop. Specifically: 1) Genotype encoding performance data update: Based on the actual production performance indicators, the historical performance records associated with the corresponding genotype in the map library (i.e., the gene from which the current ranking scheme originates) are updated, or a new data point with performance X in a certain environment is added, making the gene's environment-performance profile more accurate. 2) Co-fitness model parameter calibration: The model's predicted values, such as personnel skill quantification values ​​and equipment capability quantification values, are compared with the actual performance of personnel and equipment in this production. These differences are used to adjust the parameters within the model (such as weights and coefficients in the calculation formula) to make future predictions closer to reality. 3) Optimize Algorithm Strategy Reinforcement Learning: The ranking decision (such as objective function weight selection and genetic algorithm parameters) and the final reward (i.e., comprehensive production efficiency) are used as a new set of state-action-reward samples and input into the reinforcement learning agent that controls the weight adjustment. This is used to train and update its decision-making strategy, enabling it to make better weight selections in similar scenarios in the future. 4) Potential Energy Field Propagation Rule Correction: The differences between the actual changes in the potential energy field and the model's predicted changes during the production process are analyzed, especially the differences between the actual paths of bottleneck generation, propagation, and dissipation and the model simulation. These data are used to adjust the coefficients (such as attenuation factors) in the potential energy propagation rule, allowing the potential energy field model to more accurately simulate the congestion dynamics of a real production line.

[0117] This invention provides an embodiment of an intelligent scheduling system for garment production based on the Internet of Things, comprising: The historical pattern retrieval module is used to retrieve at least one candidate genotype from the production pattern genograph library that matches the current order's process characteristics and production line resource status. The production pattern genograph library stores genotype codes generated based on historical successful ranking schemes, and the genotype codes include at least personnel skill feature vectors, equipment function feature vectors, and logical relationship topologies between processes. The adaptation quantification and scheme initialization module is used to parse the retrieved candidate genotypes into an initial ranking scheme and calculate the collaborative adaptation coefficient of the personnel-equipment-process combination assigned to each workstation in the initial ranking scheme; wherein the collaborative adaptation coefficient is obtained by fusing the quantitative values ​​of personnel skills, equipment capabilities, and process requirements. The multi-objective optimization module is used to construct an optimization objective function that focuses on improving the overall co-fitness coefficient, while also considering the historical environmental fitness associated with candidate genotypes and satisfying delivery time and load constraints. Based on the initial ranking scheme, a genetic algorithm is used to solve the problem, output and execute the optimal ranking scheme. The potential energy field dynamic scheduling module is used to dynamically construct the process flow potential energy field based on the real-time task load and processing capacity of each workstation during production execution, identify the bottleneck workstation in the process flow potential energy field, and generate and execute task diversion scheduling instructions based on the potential energy gradient from the bottleneck workstation to the adjacent workstation.

[0118] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A smart scheduling method for garment production based on the Internet of Things, characterized in that, include: Step S1: Retrieve at least one candidate genotype from the production mode gene map library that matches the current order process characteristics and production line resource status; The production mode gene map library stores genotype codes generated based on historical successful ranking schemes, and the genotype codes include at least personnel skill feature vectors, equipment function feature vectors, and logical relationship topologies between processes. Step S2: Parse the retrieved candidate genotypes into an initial ranking scheme, and calculate the synergistic fit coefficient of the personnel-equipment-process combination assigned to each workstation in the initial ranking scheme; The collaboration adaptability coefficient is obtained by integrating the quantitative values ​​of personnel skills, equipment capabilities, and process requirements. Step S3: Construct an optimization objective function that focuses on improving the overall co-fitness coefficient, while also considering the historical environmental fitness associated with candidate genotypes and satisfying delivery time and load constraints; based on the initial ranking scheme, use a genetic algorithm to solve the problem, output and execute the optimal ranking scheme; Step S4: During production execution, the process flow potential energy field is dynamically constructed based on the real-time task load and processing capacity of each workstation, the bottleneck workstation in the process flow potential energy field is identified, and task diversion scheduling instructions are generated and executed based on the potential energy gradient from the bottleneck workstation to the adjacent workstation.

2. The intelligent scheduling method for garment production based on the Internet of Things according to claim 1, characterized in that, Step S1 employs a dynamic evolution and multi-level progressive retrieval mechanism, including: When a new scheduling scheme is verified in production and meets the preset performance standards, the personnel skill feature vector, equipment function feature vector, logical relationship topology between processes, and scheduling decision sequence features of the new scheduling scheme are automatically extracted, and encoded as new genotype codes with the execution environment and performance data, and incrementally integrated into the production mode genotype library in a way that minimizes topological conflicts. Coarse-grained screening is performed based on the topological matching degree of the logical relationship between processes to obtain a primary candidate genotype set; In the primary candidate genotype set, fine-grained matching is performed by combining the compatibility of personnel skill feature vectors and equipment function feature vectors to obtain the secondary candidate genotype set. In the secondary candidate genotype set, the performance is ranked according to the reproducibility of historical scheduling decision sequences under the current resource constraints, and candidate genotypes that match the current order process characteristics and production line resource status are output.

3. The intelligent scheduling method for garment production based on the Internet of Things according to claim 1, characterized in that, Step S2 employs a state-aware dynamic skill evolution and fusion method, including: The personnel skill feature vectors, equipment function feature vectors, and logical relationship topology between processes contained in the candidate genotypes are mapped to specific workstation, personnel, equipment, and process allocation relationships to generate an initial ranking scheme. Based on the personnel allocation in the initial ranking scheme, a process proficiency update algorithm is constructed, which includes a time decay function and a reinforcement learning function. The process proficiency is updated positively based on the operator's task completion data and the process proficiency update algorithm, and then decayed based on the non-operation period to obtain the updated process proficiency. Quantitative values ​​of personnel skills are calculated based on the updated process proficiency level; Based on the equipment allocation and IoT data of the equipment in the initial ranking scheme, calculate the performance reduction factor that reflects the current operating health status, and combine the equipment functional compatibility to calculate the quantitative value of equipment capability; Based on real-time collected operator fatigue data and performance reduction coefficients, the weights of personnel skill quantification values ​​and equipment capability quantification values ​​in the fusion calculation are dynamically adjusted, and then fused with process requirement quantification values ​​to generate a collaborative fit coefficient.

4. The intelligent scheduling method for garment production based on the Internet of Things according to claim 3, characterized in that, The calculation of the quantitative value of process requirements is based on the process allocation in the initial ranking scheme, and is generated through a multi-dimensional weighted model according to the standard working hours of the process, the process difficulty coefficient, and the dual requirements of the current order for the accuracy and timeliness of the process.

5. The intelligent scheduling method for garment production based on the Internet of Things according to claim 1, characterized in that, Step S3 employs a reinforcement learning-based method for environment fitness prediction and parameter adaptive optimization, including: The optimization objective function is defined as F=w1*C+w2*E+w3*D, where C represents the overall coordination and adaptability coefficient, E represents the historical environmental adaptability prediction value, D represents the delivery urgency factor, and w1, w2, and w3 are dynamic weight coefficients, and w1+w2+w3=1. The optimization objective function solution must meet the preset constraints of workstation load rate, material flow distance, and production changeover time. Construct a production environment feature space that includes order attributes, personnel status, equipment operating conditions, and material supply; Calculate the multidimensional fuzzy similarity between the current production environment and the genotype historical environment in the production environment feature space, and use the multidimensional fuzzy similarity as a weight to perform weighted fusion of historical performance data to obtain the predicted value E of historical environment fitness. Establish a reinforcement learning environment that rewards the actual comprehensive production benefits of the ranking scheme. The weight coefficients w1, w2, and w3 in the optimization objective function F are used as adjustable actions. Through policy learning, the dynamic adjustment strategy of weights that maximizes long-term rewards is output, and the values ​​of w1, w2, and w3 are adjusted. An initial population is constructed based on the initial ranking scheme. The optimization objective function F with adjusted weight coefficients is applied, and the selection, crossover, and mutation iteration process of the genetic algorithm is executed until the convergence condition is met. The optimal ranking scheme is then output and the production system is driven to execute.

6. The intelligent scheduling method for garment production based on the Internet of Things according to claim 5, characterized in that, Genetic algorithms employ an adaptive improvement mechanism based on population diversity feedback, including: When the initial population is generated, it is screened based on the cofit coefficient, and only schemes with a cofit coefficient higher than a preset threshold are retained to form the initial population. During the iteration process, the crossover rate and mutation rate are dynamically adjusted based on the constraint satisfaction of the current best individual and the population gene diversity index. An elite retention and adaptive generation gap strategy is adopted to maintain global search capability while ensuring convergence speed.

7. The intelligent scheduling method for garment production based on the Internet of Things according to claim 1, characterized in that, The dynamic scheduling steps of the potential energy field employ a potential energy propagation model and a historical efficiency path selection method, including: Defining potential energy propagation rules based on process topology network; Calculate the original potential energy generated by each workstation due to its own task load and processing capacity, and calculate the additional potential energy generated by the propagation of the original potential energy from upstream to downstream according to the potential energy propagation rules. Superimpose the original potential energy and the additional potential energy to form a global process flow potential energy field that reflects the risk of systemic blockage. Calculate the potential energy gradient from the bottleneck workstation to each adjacent workstation, combine the historical diversion success rate and average efficiency improvement data of each path, predict the secondary bottleneck risks that diversion may cause, perform multi-objective comprehensive scoring, select the optimal path to generate task diversion scheduling instructions and execute them.

8. The intelligent scheduling method for garment production based on the Internet of Things according to claim 7, characterized in that, The execution of task diversion and scheduling instructions requires real-time verification of process divisibility and compatibility with target workstation equipment to ensure the feasibility of scheduling actions. After scheduling is executed, the changes in the potential energy field of the process flow are monitored in real time. If a new imbalance is caused, a local rebalancing or global re-optimization is triggered.

9. The intelligent scheduling method for garment production based on the Internet of Things according to claim 1, characterized in that, Also includes: Record the entire process data and final production efficiency indicators of each ranking plan from generation, optimization, execution to dynamic scheduling; Based on full-process data, the genotype codes in the production mode gene map library are updated with performance data, the parameters in the co-fitness coefficient calculation model are calibrated, the strategies in the optimization algorithm are reinforced through reinforcement learning, and the propagation rules in the process flow potential field model are corrected.

10. An intelligent scheduling system for garment production based on the Internet of Things, characterized in that, include: The historical pattern retrieval module is used to retrieve at least one candidate genotype from the production pattern genograph library that matches the current order's process characteristics and production line resource status. The production mode gene map library stores genotype codes generated based on historical successful ranking schemes, and the genotype codes include at least personnel skill feature vectors, equipment function feature vectors, and logical relationship topologies between processes. The adaptation quantification and scheme initialization module is used to parse the retrieved candidate genotypes into an initial ranking scheme and calculate the synergistic adaptation coefficient of the personnel-equipment-process combination assigned to each workstation in the initial ranking scheme. The collaboration adaptability coefficient is obtained by integrating the quantitative values ​​of personnel skills, equipment capabilities, and process requirements. The multi-objective optimization module is used to construct an optimization objective function that focuses on improving the overall co-fitness coefficient, while also considering the historical environmental fitness associated with candidate genotypes and satisfying delivery time and load constraints. Based on the initial ranking scheme, a genetic algorithm is used to solve the problem, output and execute the optimal ranking scheme. The potential energy field dynamic scheduling module is used to dynamically construct the process flow potential energy field based on the real-time task load and processing capacity of each workstation during production execution, identify the bottleneck workstation in the process flow potential energy field, and generate and execute task diversion scheduling instructions based on the potential energy gradient from the bottleneck workstation to the adjacent workstation.

Images