Military clothing production scheduling optimization method and system based on intelligent algorithm

Abstract

The invention provides a military clothing production scheduling optimization method and system based on an intelligent algorithm, and the method achieves the intelligent management of a production process through the construction of a dynamic state model and a disturbance cost evaluation system. Firstly, a reference production plan is generated by using a global optimization algorithm, and an event sensing module is deployed to monitor the production process in real time. When a task insertion event is captured, the system automatically collects related production parameters, triggers a disturbance cost evaluation mechanism, and dynamically predicts and quantifies the comprehensive influence of different interruption schemes. And based on the quantification result and a preset decision strategy, the system determines an optimal interrupt execution scheme, calls a rapid local rearrangement algorithm coupled with a disturbance cost index, adaptively adjusts the affected plan segments, and finally generates and issues an optimized production instruction sequence. According to the method, the whole process from disturbance identification to scheme execution is intelligentized, and the problem of dynamic scheduling for emergency tasks in military clothing production is effectively solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of clothing production optimization, and in particular relates to a method and system for optimizing military clothing production scheduling based on an intelligent algorithm. Background Art

[0002] The fundamental reason for the lack of adaptability in existing dynamic scheduling technologies lies in the conflict between traditional production management concepts and the demands of modern manufacturing environments. Traditional production scheduling systems are mostly designed based on static optimization models, which assume a relatively stable production environment, predictable parameters, and known task requirements. However, the real production environment is full of uncertainty and change. Especially in the modern manufacturing model of customization, small batches, and high variety, this static assumption is clearly out of touch with reality.

[0003] On a technical level, traditional scheduling systems typically use a cyclical planning approach. Once a plan is finalized, the system lacks effective mechanisms to assess and respond to changes that arise mid-cycle. When urgent tasks are inserted, managers often face a dilemma: either completely disrupt the current production plan, resulting in significant resource waste and efficiency losses; or postpone the urgent task, potentially delaying important orders and reducing customer satisfaction. This "all-or-nothing" approach makes it difficult to achieve optimal allocation of production resources. Summary of the Invention

[0004] The present invention provides a method and system for optimizing the production scheduling of military clothing based on an intelligent algorithm, so as to solve the problem that it is difficult to achieve the optimal configuration of production resources when urgent tasks are inserted.

[0005] In a first aspect, the present invention provides a method for optimizing production scheduling of military clothing based on an intelligent algorithm, the method comprising the following steps:

[0006] Construct a dynamic state model of the military clothing production process and establish a production disturbance cost assessment system based on real-time production process parameters and prediction models;

[0007] Generate a benchmark production plan for military clothing using a global optimization intelligent algorithm and a dynamic state model;

[0008] Deploy an event perception module to capture task insertion events during real-time production according to the baseline production plan, and simultaneously collect real-time production parameters associated with the task insertion events;

[0009] When the event perception module captures a task insertion event, it triggers the production disturbance cost assessment system, which uses the collected real-time production parameters to dynamically predict and quantify the comprehensive disturbance costs of different interruption scenarios.

[0010] Make interruption response decisions based on the quantified comprehensive disturbance cost and the preset decision-making strategy model to determine the optimal interruption execution plan;

[0011] According to the interruption execution plan, a fast local rescheduling algorithm coupled with a disturbance cost indicator is called to adaptively adjust the affected baseline production plan segments with the goal of minimizing the actual disturbance impact.

[0012] Generate and issue adaptively adjusted executable production instruction sequences.

[0013] Optionally, establishing a production disturbance cost assessment system based on real-time production process parameters and prediction models includes the following steps:

[0014] A machine learning model is used to build a material loss prediction model and a setup time prediction model based on equipment status. The setup time prediction model inputs the current equipment operating status and the process characteristics of the interrupted and inserted tasks. The setup time prediction model outputs the dynamic prediction of the setup time required to execute the interruption switching and subsequent task recovery.

[0015] A quality risk assessment model is constructed based on a pre-trained quality predictor. Both the quality risk assessment model and the material loss prediction model input the current process interrupted by the task insertion event, the elapsed processing time, and real-time production parameters. The quality risk assessment model outputs the dynamic probability of quality defects introduced by the task insertion event, and the material loss prediction model outputs the probability of material scrap caused by the task insertion event.

[0016] A chain delay propagation analysis module based on the baseline production plan topology is constructed. Through the chain delay propagation analysis module and fast graph path analysis, the cascading delay effect of task insertion events on subsequent related tasks and order completion times is estimated.

[0017] The material loss prediction model, preparation time prediction model, quality risk assessment model and chain delay propagation analysis module are integrated to form a production disturbance cost assessment system.

[0018] Optionally, building a quality risk assessment model based on a pre-trained quality predictor includes the following steps:

[0019] Collect historical production data of military clothing and construct a multi-dimensional feature training set. The historical production data includes historical process parameters, historical equipment status, historical interruption event records, and corresponding quality inspection results.

[0020] Extract interruption-sensitive features from the multi-dimensional feature training set through feature engineering;

[0021] A quality risk assessment model based on a two-stream neural network structure is constructed based on the pre-trained quality predictor. The two-stream neural network structure is used to simultaneously process the multi-dimensional feature training set and interruption-sensitive features.

[0022] An attention-based interruption impact assessment unit is designed in the quality risk assessment model. The interruption impact assessment unit is used to dynamically adjust the parameter weights in the two-stream neural network structure according to the time point of the interruption, the current process completion degree and the material status.

[0023] Optionally, a chain delay propagation analysis module based on the baseline production plan topology is constructed. The chain delay propagation analysis module and fast graph path analysis are used to estimate the cascading delay effect of the task insertion event on the subsequent associated tasks and order completion time, including the following steps:

[0024] Convert the baseline production plan into a directed weighted graph structure, where nodes represent production tasks, edges represent dependencies between tasks, and edge weights include time dependency and resource sharing.

[0025] Based on the constructed directed weighted graph, a multi-level task association matrix is established to quantify the association strength between tasks from four dimensions: process dependency, resource occupation, material flow, and personnel deployment.

[0026] By using the task dependency matrix and directed weighted graph and adopting fast graph path analysis to perform forward scanning, the delay propagation path and the critical task set directly affected by the task insertion event are identified;

[0027] The cascading delay effect of task insertion events on subsequent associated tasks and order completion time is predicted based on the key task set.

[0028] Optionally, predicting the cascading delay effect of a task insertion event on subsequent associated tasks and order completion times based on a key task set includes the following steps:

[0029] A time buffer consumption model is introduced to calculate the dynamic delay absorption capacity of the floating time of each critical task set and predict the actual propagation delay amount;

[0030] A resource conflict detection mechanism is developed based on the delay propagation path and delay amount. When delay propagation causes multiple tasks to compete for the same resource, the resource conflict is resolved through preset task priority rules.

[0031] Aggregate the delays of all key task sets to calculate the order delivery delay;

[0032] Combining the delay propagation path and order delivery delay prediction design, the delay prediction results of subsequent associated tasks and order completion time are generated.

[0033] Optionally, using a global optimization intelligent algorithm and generating a baseline production plan for military clothing based on a dynamic state model includes the following steps:

[0034] Construct a multi-objective optimization model for military clothing production. The model objective functions include minimizing total completion time, maximizing equipment utilization, and minimizing delivery delays.

[0035] Extract the current production system constraints from the military clothing production management system, including equipment availability, personnel skill configuration, material inventory status and process route requirements;

[0036] Based on the genetic algorithm framework and combined with the multi-objective optimization model and constraint conditions, a global optimization intelligent algorithm is created, and the chromosome code is generated using the chromosome representation method based on process coding;

[0037] Generate the initial solution population of chromosome encoding using heuristic rules based on key resource bottleneck identification;

[0038] The initial solution population is iteratively optimized through a global optimization intelligent algorithm. When the preset upper limit of iterations is reached, the optimal population is obtained. The optimal population is analyzed to obtain the optimal scheduling plan, and the optimal scheduling plan is converted into a standardized benchmark production plan.

[0039] Optionally, making an interruption response decision based on the quantified comprehensive disturbance cost and a preset decision strategy model to determine the optimal interruption execution plan includes the following steps:

[0040] Determine the disturbance scope and response time limit based on the urgency and importance of the task insertion event;

[0041] Identify all feasible interruption points based on the current production status. Each feasible interruption point corresponds to a process location on the production line that can be paused.

[0042] Calculate the comprehensive disturbance cost index of each feasible interruption point according to the comprehensive disturbance cost, and use the rule-based decision model to apply weights and threshold conditions to different comprehensive disturbance cost indexes;

[0043] The comprehensive disturbance cost index and decision model are combined to select the interruption execution plan with the minimum disturbance cost from the feasible interruption points.

[0044] Optionally, a fast local rescheduling algorithm coupled with a disturbance cost indicator is called according to the interruption execution plan to adaptively adjust the affected baseline production plan segments with the goal of minimizing the actual disturbance impact, including the following steps:

[0045] Identify the affected baseline production plan segments based on the interruption execution plan and determine the plan scope that needs to be rescheduled;

[0046] Constructing a local rearrangement objective function based on the comprehensive perturbation cost index and the planning scope;

[0047] Apply a fast scheduling algorithm based on priority rules and determine the reordering of tasks according to task urgency, process requirements and resource utilization efficiency;

[0048] Under the premise of satisfying process constraints, the local rearrangement objective function and rearrangement order are combined and an incremental update method is adopted to insert urgent tasks and rearrange the affected tasks.

[0049] In the second aspect, the present invention also provides a military clothing production scheduling optimization system based on an intelligent algorithm, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the military clothing production scheduling optimization method based on the intelligent algorithm as described in the first aspect.

[0050] In a third aspect, the present invention also provides a computer-readable storage medium having instructions stored thereon, which, when executed by a processor, configures the processor to execute the military clothing production scheduling optimization method based on an intelligent algorithm as described in the first aspect.

[0051] The beneficial effects of the present invention are:

[0052] This invention has achieved significant technological breakthroughs in the field of military clothing production scheduling, bringing remarkable results in many areas. First, by constructing a dynamic state model and a real-time disturbance cost assessment system, the system can accurately quantify the impact of different production interruption scenarios, shifting decision-making from empirical judgment to data-driven decision-making, significantly improving the scientific and accurate response to emergency tasks. Second, the deployment of an event perception module enables real-time monitoring and parameter collection of the production process, reducing response time from traditional hours to minutes, significantly improving the production system's responsiveness to emergencies. Furthermore, a local rescheduling algorithm based on disturbance cost avoids the global recalculation required by traditional methods, ensuring computational efficiency while minimizing disruption to the original production plan, effectively ensuring production continuity. Furthermore, through an intelligent decision-making strategy model, this solution automatically selects the optimal interruption execution plan based on different production scenarios, reducing human intervention and management complexity. Ultimately, the entire system achieves full automation from disturbance identification to plan execution, ensuring efficient and stable military clothing production in the face of dynamic changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of a method for optimizing production scheduling of military clothing based on an intelligent algorithm in one embodiment of the present application.

[0054] Figure 2This is a schematic structural diagram of a production disturbance cost evaluation system in one embodiment of the present application. DETAILED DESCRIPTION

[0055] The following will be combined with the accompanying drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of this application.

[0056] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of this application can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0057] Figure 1 FIG. 1 is a flow chart of a method for optimizing the production schedule of military clothing based on an intelligent algorithm in one embodiment. Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps in the above process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but may be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps. Figure 1 As shown, the present invention discloses a method for optimizing production scheduling of military clothing based on an intelligent algorithm, which specifically includes the following steps:

[0058] S101. Construct a dynamic state model of the military clothing production process and establish a production disturbance cost evaluation system based on real-time parameters and prediction models of the production process.

[0059] The dynamic state model of the military clothing production process is constructed using a multidimensional parameter matrix encompassing four core dimensions: equipment status, staffing, material flow, and process progress. In the equipment status dimension, the real-time operating parameters of each sewing machine, cutting device, and ironing equipment, such as operating time, temperature, and vibration frequency, are recorded. In the staffing dimension, the operator's skill level, fatigue level, and proficiency in the current task are recorded. In the material flow dimension, the inventory level, work-in-progress status, and quality parameters of fabrics and accessories are tracked. In the process progress dimension, the completion of each process, quality inspection results, and deviations from standard working hours are monitored. These parameters are collected in real time by a network of IoT sensors deployed throughout the factory and stored in a time-series database, forming a complete dynamic state model.

[0060] The production disruption cost assessment system, based on real-time production process parameters and prediction models, is constructed through four key modules: First, the material loss prediction model uses a gradient boosting decision tree algorithm as input, taking into account the current process type, elapsed processing time, and material properties, and outputting the probability of material scrap caused by an interruption. Second, the setup time prediction model, built using a deep neural network, considers the current operating state of the equipment (such as temperature and pressure), the process characteristics of the interrupted task (such as sewing type and stitch density), and the process requirements of the inserted task to predict the setup time required to execute the interruption switch and recovery. Third, the quality risk assessment model, based on a two-stream neural network structure, simultaneously processes historical production data and interruption-sensitive characteristics, dynamically adjusts parameter weights through an attention mechanism, and predicts the probability of quality defects that may be introduced by the interruption. Finally, the chain delay propagation analysis module converts the baseline production plan into a directed weighted graph and estimates the cascading delay effect of the task insertion event on subsequent related tasks through fast graph path analysis. The outputs of these four modules are weighted and integrated to form a comprehensive disruption cost indicator, providing a quantitative basis for subsequent decision-making.

[0061] S102. Use a global optimization intelligent algorithm and generate a baseline production plan for military clothing based on a dynamic state model.

[0062] The global optimization intelligent algorithm is built on an improved multi-objective genetic algorithm framework, customized for the specific needs of military clothing production. First, a multi-objective optimization model is constructed, setting three key objective functions: minimizing total makespan, maximizing equipment utilization, and minimizing delivery delays. Total makespan is quantified by calculating the time span from the start of the first task to the completion of the last task; equipment utilization is measured as the ratio of the equipment's actual operating time to its available time; and delivery delays are assessed by calculating a weighted penalty based on the difference between each order's actual completion time and its promised delivery time.

[0063] S103. Deploy an event perception module to capture task insertion events during the real-time production process according to the baseline production plan, and synchronously collect real-time production parameters associated with the task insertion events.

[0064] The event perception module utilizes a distributed architecture, comprised of three layers: a data acquisition layer, an event recognition layer, and a parameter association layer. The data acquisition layer comprises a sensor network deployed at key production line nodes, including RFID readers, photoelectric sensors, temperature and humidity sensors, and data interfaces directly connected to production equipment. These sensors continuously monitor production line status with a sampling frequency of 100 milliseconds. RFID readers track material flow, photoelectric sensors monitor workpiece position, and temperature and humidity sensors monitor environmental parameters. The equipment data interfaces directly collect operating parameters such as speed, temperature, and pressure.

[0065] The event recognition layer uses a hybrid approach combining a rules engine and machine learning to analyze data streams in real time and identify task insertion events. The rules engine processes explicitly defined event patterns, such as urgent order instructions issued directly by the production management system. The machine learning model (primarily an anomaly detection algorithm based on a long-short-term memory network) identifies implicit event patterns, such as inferring possible task insertion needs by detecting material flow anomalies. For example, if an urgent military winter clothing order is entered into the system, or if an unplanned shipment of specialty fabric is detected, the event recognition layer immediately flags it as a task insertion event.

[0066] The parameter association layer immediately triggers a comprehensive parameter collection process after a task insertion event is identified. First, the scope of resources associated with the inserted task is determined, including potentially occupied equipment, personnel, and materials. Then, high-frequency data collection is initiated for these resources, recording the current detailed status parameters. For equipment, the current operating status, continuous operating time, temperature curve, etc. are collected. For personnel, the current work tasks, skill matching, working hours, etc. are recorded. For materials, the current processing status, quality parameters, inventory levels, etc. are collected. At the same time, special attention is paid to the currently executed process, recording key parameters such as its completion rate, quality indicators, and interruption sensitivity.

[0067] S104. When the event perception module captures a task insertion event, it triggers the production disturbance cost assessment system, which uses the collected real-time production parameters to dynamically predict and quantify the comprehensive disturbance costs of different interruption scenarios.

[0068] When the event perception module detects a task insertion event, the production disruption cost assessment system is immediately triggered to perform a multi-dimensional disruption cost analysis. First, the system automatically identifies all possible disruption scenarios, each corresponding to a different interruption point on the production line, such as after a process is completed, when a batch of materials is processed, or during a certain equipment changeover. For a military clothing production line, typical disruption points might include after cutting, between major sewing processes, before and after the addition of auxiliary materials, and before ironing and packaging. For each possible disruption scenario, the system performs a disruption cost assessment across four dimensions. The material loss prediction model first analyzes the interruption sensitivity of the current process. The setup time prediction model analyzes equipment state dependencies. The quality risk assessment model uses a two-stream neural network to analyze the potential impact of the interruption on product quality. The chain delay propagation analysis module analyzes the impact of the interruption on subsequent tasks based on the topological structure of the baseline production plan. All these cost assessment results are combined through a weighted summation to form a comprehensive disruption cost indicator.

[0069] S105. Make an interruption response decision based on the quantified comprehensive disturbance cost and the preset decision-making strategy model to determine the optimal interruption execution plan.

[0070] S106. Based on the interruption execution plan, a fast local rescheduling algorithm coupled with a disturbance cost indicator is called to adaptively adjust the affected baseline production plan segments with the goal of minimizing the actual disturbance impact.

[0071] S107. Generate and issue an adaptively adjusted executable production instruction sequence.

[0072] After optimization using a local rescheduling algorithm, the system converts the abstract adjustment plan into a concrete, executable sequence of production instructions. This conversion process begins with plan decomposition, refining the rescheduling plan according to three dimensions: equipment, process, and operator. For each piece of equipment, detailed instructions are generated, including start and end times, processing parameters, and quality requirements. For each process, an operating guide is generated, including process key points, material requirements, and quality inspection standards. For each operator, a work instruction is generated, including task assignment, time schedule, and technical requirements. During the instruction sequence generation process, the system performs multiple validation steps to ensure executable execution. First, resource conflict detection ensures that multiple tasks are not assigned to the same piece of equipment or operator within the same time period. Next, process continuity verification ensures that processes requiring continuous processing are not inappropriately fragmented. Next, material availability verification ensures that the necessary materials are readily available before each process begins. Finally, time rationality verification ensures that the allocated time is sufficient to complete the corresponding process and that necessary preparation and adjustment time are accounted for.

[0073] The instruction sequence also includes exception handling plans, providing alternative solutions for various possible abnormal situations. For example, if the actual execution time of a process exceeds expectations, the system will provide alternative solutions for adjusting subsequent processes; if a piece of equipment fails, the system will provide a switch to an alternative device; if material quality does not meet requirements, the system will provide alternative materials or rework solutions. These plans are embedded in the instruction sequence in the form of decision trees, making the production process more adaptable and robust.

[0074] In one embodiment, establishing a production disturbance cost assessment system based on real-time production process parameters and a prediction model includes the following steps:

[0075] A machine learning model is used to build a material loss prediction model and a setup time prediction model based on equipment status. The setup time prediction model inputs the current equipment operating status and the process characteristics of the interrupted and inserted tasks. The setup time prediction model outputs the dynamic prediction of the setup time required to execute the interruption switching and subsequent task recovery.

[0076] A quality risk assessment model is constructed based on a pre-trained quality predictor. Both the quality risk assessment model and the material loss prediction model input the current process interrupted by the task insertion event, the elapsed processing time, and real-time production parameters. The quality risk assessment model outputs the dynamic probability of quality defects introduced by the task insertion event, and the material loss prediction model outputs the probability of material scrap caused by the task insertion event.

[0077] A chain delay propagation analysis module based on the baseline production plan topology is constructed. Through the chain delay propagation analysis module and fast graph path analysis, the cascading delay effect of task insertion events on subsequent related tasks and order completion times is estimated.

[0078] The material loss prediction model, preparation time prediction model, quality risk assessment model and chain delay propagation analysis module are integrated to form a production disturbance cost assessment system.

[0079] In this embodiment, the material loss prediction model is constructed using the gradient boosting decision tree (GBDT) algorithm, which receives the current process information interrupted by the task insertion event, the processed time data and the real-time production parameters as input features. In the specific implementation, historical production data is first collected, including the process interruption point, the interruption duration, the production parameters (such as temperature, pressure, speed) and the corresponding material scrapping situation. The data is then preprocessed, including missing value filling, outlier processing and feature standardization. The model training adopts a five-fold cross-validation method, and the model performance is optimized by adjusting the learning rate, tree depth and regularization parameters. In actual applications, when the production line faces a task insertion request, the model can predict in real time the probability of material scrapping due to the interruption, and output a probability value between 0 and 1 to provide a quantitative basis for decision-making. The model realizes the accurate assessment of material loss risk and effectively reduces the material waste caused by improper task insertion decisions.

[0080] The setup time prediction model, based on a deep learning architecture combining a long short-term memory (LSTM) network with an attention mechanism, is used to dynamically predict the setup time required for task switching. The model inputs include three types of features: the current equipment operating status (such as equipment temperature, operating time, and maintenance status), characteristics of interrupted tasks (such as material type, processing accuracy requirements, and completion rate), and characteristics of inserted tasks (such as process parameters and priority). During implementation, a feature engineering module was first constructed to convert different types of features into vectors suitable for neural network processing. A bidirectional LSTM layer was then designed to capture temporal dependencies, and an attention layer was used to identify key influencing factors. Model training uses root mean square error as the loss function, and the Adam optimizer is used for parameter updates. In practical application scenarios, when assessing the feasibility of task insertion, the model can predict the setup time percentage in real time based on the current production environment. This model significantly improves the accuracy of setup time predictions, making production scheduling decisions more precise and reducing production delays caused by insufficient setup time estimates.

[0081] The quality risk assessment model is built based on a pre-trained convolutional neural network (CNN) and a random forest ensemble method. The model first uses a quality predictor pre-trained on large-scale production data to extract deep features, and then combines the current process status information, processed time, and real-time production parameters to perform quality risk assessment. In specific implementation, the pre-training stage uses a contrastive learning method to enable the model to identify subtle differences between normal production and abnormal production; the fine-tuning stage optimizes the quality defect patterns of specific production lines. The model input includes production parameter time series data, process parameter configuration, and the equipment state vector at the moment of interruption. During the risk assessment process, the model calculates the probability of occurrence of different types of quality defects (such as surface defects, structural defects, and performance deviations), and comprehensively outputs a quality risk index between 0 and 1. The model can accurately identify potential quality problems, provide risk assessment of the quality dimension for task insertion decisions, effectively reduce quality fluctuations caused by production interruptions, and improve the first-time pass rate of products.

[0082] The Chained Delay Propagation Analysis module, built on a directed acyclic graph (DAG) model and a critical path algorithm, assesses the cascading impact of task insertion events on the overall production plan. The module first represents the baseline production plan as a topological graph, where nodes represent production tasks, edges represent inter-task dependencies, and edge weights are task execution times. During implementation, a modified Dijkstra algorithm is used for fast path analysis, assessing delay propagation through the following steps: 1) identifying nodes directly affected by task insertion; 2) calculating the probable delay time for each node; 3) using a breadth-first search algorithm to trace the delay propagation path through the dependency graph; and 4) calculating the cumulative delay time for each affected node. The module also integrates Monte Carlo simulation functionality, generating delay distribution probability maps by randomly sampling different delay scenarios. In practical applications, the module can complete delay analysis of complex production networks within milliseconds, accurately predicting over 90% of order delay risks, providing intuitive decision support for production schedulers and effectively reducing order delivery delays caused by task insertion.

[0083] The production disruption cost assessment system integrates the four aforementioned modules to form a decision support system for comprehensively evaluating the impact of task insertion. This system utilizes a weighted multi-objective optimization framework to transform factors such as material loss, setup time, quality risk, and delay propagation into a unified cost metric. In implementation, cost conversion functions are first defined, such as material loss probability multiplied by material value, quality risk multiplied by rework / scrap cost, and delay time multiplied by a penalty factor. A dynamic weight adjustment mechanism is then designed to adaptively adjust the weights of each factor based on the company's current operational priorities (such as delivery pressure, quality requirements, and cost control). The assessment process utilizes a parallel computing architecture, with each module running simultaneously and inputting the results into an integrated decision layer. Ultimately, a comprehensive disruption cost index and detailed cost component analysis are generated. The system supports analysis of various decision scenarios, such as whether to accept an insertion order, selecting the optimal insertion timing, and determining resource compensation strategies after insertion.

[0084] In one embodiment, constructing a quality risk assessment model based on a pre-trained quality predictor comprises the following steps:

[0085] Collect historical production data of military clothing and construct a multi-dimensional feature training set. The historical production data includes historical process parameters, historical equipment status, historical interruption event records, and corresponding quality inspection results.

[0086] Extract interruption-sensitive features from the multi-dimensional feature training set through feature engineering;

[0087] A quality risk assessment model based on a two-stream neural network structure is constructed based on the pre-trained quality predictor. The two-stream neural network structure is used to simultaneously process the multi-dimensional feature training set and interruption-sensitive features.

[0088] An attention-based interruption impact assessment unit is designed in the quality risk assessment model. The interruption impact assessment unit is used to dynamically adjust the parameter weights in the two-stream neural network structure according to the time point of the interruption, the current process completion degree and the material status.

[0089] In this embodiment, the military clothing production data collection process covers key information of the entire production process, including four major categories of data: process parameter data (such as sewing speed, stitch setting, thread tension value), sampled every 15 seconds; equipment status data (such as machine temperature, vibration frequency, energy consumption index), recorded once a minute; interruption event records (including interruption cause, duration, and process completion at the time of interruption); and quality inspection results (including stitching strength, appearance rating, and protective performance indicators). Data collection adopts an architecture that combines a distributed sensor network with a central data warehouse to ensure data integrity and real-time performance. When constructing a multi-dimensional feature training set, the original data is first cleaned, outliers are removed, and missing values are filled; then time alignment is performed to unify data collected at different frequencies to the same time scale; finally, feature fusion is performed to associate process parameters, equipment status, and interruption events with corresponding quality inspection results.

[0090] Interruption-sensitive feature extraction utilizes a multi-stage feature engineering approach to identify key features highly correlated with production interruptions from a multidimensional feature training set. First, Pearson correlation coefficient analysis and mutual information calculation are applied to screen a subset of features significantly associated with quality fluctuations. Second, recursive feature elimination (RFE) combined with random forest importance scoring is used to further narrow the feature range. Finally, principal component analysis (PCA) and t-SNE dimensionality reduction techniques are used to map high-dimensional features to a low-dimensional space, enhancing feature interpretability. In practice, given the unique characteristics of military clothing, the following interruption-sensitive features were extracted: the rate of change of thread tension during sewing interruption, the difference in fabric stress state before and after the interruption, the interaction between interruption duration and material properties, and equipment stability indicators after interruption recovery. These features were validated through visual analysis and demonstrated a strong correlation with finished product quality, particularly in key performance indicators such as waterproof seams and tear resistance.

[0091] A two-stream neural network quality risk assessment model based on a pretrained quality predictor utilizes an innovative architectural design to accurately predict the production quality of military clothing. The model comprises two parallel processing streams: the first stream processes the multidimensional feature training set and consists of a three-layer fully connected network, with 128, 64, and 32 neurons in each layer, respectively, using the ReLU activation function. The second stream specifically processes interruption-sensitive features using a one-dimensional convolutional neural network architecture consisting of two convolutional layers (with kernel sizes of 3 and 5, respectively) and a max pooling layer, effectively capturing the temporal patterns of interruption features. The pretrained quality predictor, based on the ResNet-50 architecture, transfers knowledge trained on millions of clothing quality inspection images to the current task through transfer learning. The outputs of the two streams are combined through a feature fusion layer (using an attention-weighted connection) and finally pass through a softmax classification layer to output a quality risk probability distribution. The model is trained using mini-batch gradient descent with a batch size of 64, a learning rate of 0.001, and an early stopping strategy to prevent overfitting.

[0092] The attention-based interruption impact assessment unit is a core innovative component of the quality risk assessment model, specifically designed to dynamically assess the impact of production interruptions on the quality of military clothing. This unit utilizes a multi-head self-attention mechanism (eight attention heads) to adaptively focus on key features in different interruption scenarios. The attention unit receives three key inputs: the interruption's time signature (as a percentage of the total process duration), a completion indicator for the current process (consisting of seven fine-grained completion stage indicators), and material state parameters (such as temperature, humidity, and stress). Attention weights are calculated using a scaled dot-product attention mechanism and normalized using a softmax function. In practice, when an interruption is detected, the assessment unit calculates the attention distribution in real time and dynamically adjusts the parameter weights in the two-stream neural network accordingly, specifically adjusting the fusion ratio of the two streams and the sensitivity of the convolutional layers. For example, an interruption in the early stages of fabric cutting has minimal impact on quality, so the attention unit will reduce the weight of the interruption feature stream. However, an interruption during the critical stitching stage will significantly increase the sensitivity of the relevant parameters.

[0093] In one embodiment, a chain delay propagation analysis module based on a baseline production plan topology is constructed. The chain delay propagation analysis module and fast graph path analysis are used to estimate the cascading delay effect of a task insertion event on subsequent associated tasks and order completion times, including the following steps:

[0094] Convert the baseline production plan into a directed weighted graph structure, where nodes represent production tasks, edges represent dependencies between tasks, and edge weights include time dependency and resource sharing.

[0095] Based on the constructed directed weighted graph, a multi-level task association matrix is established to quantify the association strength between tasks from four dimensions: process dependency, resource occupation, material flow, and personnel deployment.

[0096] By using the task dependency matrix and directed weighted graph and adopting fast graph path analysis to perform forward scanning, the delay propagation path and the critical task set directly affected by the task insertion event are identified;

[0097] The cascading delay effect of task insertion events on subsequent associated tasks and order completion time is predicted based on the key task set.

[0098] In this implementation, the conversion of the baseline production plan into a directed weighted graph is achieved through mathematical modeling, a process that transforms the abstract production plan into a computable network structure. Specifically, each production task is first defined as a node in the graph, assigned a unique identifier and a set of attributes (including planned start time, duration, and required resources). Then, based on the process flow, dependencies between tasks are determined and directed edges are established. Edge weights are designed using a binary tuple (t, r), where t represents the degree of temporal dependency (ranging from 0 to 1, with 1 indicating strict temporal dependency) and r represents the degree of resource sharing (ranging from 0 to 1, with 1 indicating complete sharing of key resources). For example, in a clothing production line, the edge weight between cutting and sewing tasks might be (0.9, 0.3), indicating high temporal dependency but low resource sharing. Once the graph is constructed, its acyclicity is verified through topological sorting to ensure the logical rationality of the production plan. This graph structure representation transforms complex production plans into intuitive and visual network models, facilitating efficient processing by subsequent analysis algorithms while preserving the rich dependency relationships and resource constraints in the production plan.

[0099] The process of constructing a multi-level task association matrix further quantifies and refines the association relationships between tasks in a directed weighted graph. By constructing an association matrix with four dimensions, the complex interactions between tasks are fully captured. Specifically, four n×n matrices are defined (n is the total number of tasks): the process dependency matrix Md records the strength of direct and indirect dependencies within the process flow, using normalized values between 0 and 1; the resource utilization matrix Mr quantifies the degree of resource sharing between tasks, such as equipment and workstations, taking into account resource type and occupancy duration; the material flow matrix Mm represents the quantity and importance of material transfers between tasks; and the personnel allocation matrix Mp reflects the degree of sharing and skill matching requirements between operators. Matrix elements are calculated using a weighted scoring method. For example, the resource utilization matrix element Mr(i,j) = ∑(wk×sk), where wk is the importance weight of the kth resource category and sk is the degree of resource sharing. In practical applications, taking clothing production as an example, cutting and sewing tasks have a high correlation (0.95) in the material flow dimension but a low correlation (0.2) in the personnel allocation dimension. This multi-level matrix representation method provides a refined description of task associations, provides a mathematical basis for subsequent delay propagation analysis, and can accurately reflect the interaction characteristics of tasks in different dimensions.

[0100] Fast graph path analysis and critical task set identification employs an improved breadth-first search algorithm, combined with a task association matrix for forward scanning, to efficiently identify delay propagation paths. Specifically, the algorithm first identifies the point of direct impact of the task insertion event (the interrupted task node) as the starting point for the scan. It then sets propagation threshold vectors T = (td, tr, tm, tp), corresponding to the critical values of the association strength in the four dimensions. Multiple rounds of iterative scanning are then performed: the first round considers only nodes in the process dependency matrix Md with an association strength exceeding td. The second round incorporates nodes in the resource occupancy matrix Mr with an association strength exceeding tr, and so on, completing four rounds of scanning. In each round, a priority queue is used to store nodes to be visited, sorted by descending association strength, and access flags are maintained to avoid duplicate calculations. For example, in a clothing production line, if a cutting task is interrupted, the algorithm first identifies sewing tasks that are strongly related to the interrupted task, followed by other cutting tasks that share cutting equipment. Finally, the algorithm outputs a critical task set K = {k1, k2, ..., km}, where each element contains a task identifier and an impact score. This progressive scanning method significantly improves analysis efficiency. In complex production plans containing thousands of tasks, analysis time is controlled at the millisecond level, while ensuring the accuracy and completeness of the recognition results.

[0101] Cascading delay effect prediction is based on an identified set of critical tasks. A probability propagation model is used to quantify the time impact of task insertion events on the overall production plan. Specifically, a delay propagation function, D(ki), is constructed to calculate the expected delay of critical task ki. The function takes the form D(ki) = d0 × f(si) × g(pi), where d0 is the initial delay (the duration of the task insertion), f(si) is an adjustment factor related to the task's completion stage, and g(pi) is an adjustment factor related to the task's priority. Monte Carlo simulation is then applied to each task in the critical task set, generating multiple possible delay scenarios and calculating the cumulative delay distribution. Next, the critical path method (CPM) is used to assess the impact of delays on order completion times based on the earliest start time (EST) and latest finish time (LFT) of each task. Finally, a cascading delay effect report is generated, containing three key metrics: order delay probability, expected delay time and its 95% confidence interval, and delay sensitivity ranking (identifying the tasks that contribute most to the total delay). For example, in a clothing production scenario, the system might predict that "due to a 20-minute interruption in cutting, there is an 85% probability that order A will be delayed by 1.5-2.3 hours." This probability-based delay prediction method provides production managers with a scientific basis for decision-making, enabling them to assess the time risk of task insertion and implement targeted mitigation measures.

[0102] In one embodiment, predicting the cascading delay effect of a task insertion event on subsequent associated tasks and order completion times based on a key task set includes the following steps:

[0103] A time buffer consumption model is introduced to calculate the dynamic delay absorption capacity of the floating time of each critical task set and predict the actual propagation delay amount;

[0104] A resource conflict detection mechanism is developed based on the delay propagation path and delay amount. When delay propagation causes multiple tasks to compete for the same resource, the resource conflict is resolved through preset task priority rules.

[0105] Aggregate the delays of all key task sets to calculate the order delivery delay;

[0106] Combining the delay propagation path and order delivery delay prediction design, the delay prediction results of subsequent associated tasks and order completion time are generated.

[0107] In this implementation, the time buffer consumption model accurately predicts the actual propagation delay by precisely calculating the dynamic delay absorption capacity of each task's float time within the critical task set. This model, based on elastic buffer theory, considers each task's float time as a consumable delay absorption resource. In specific implementation, the total float time of a task, TF(i), is first calculated as: LFT(i) - EST(i) - Duration(i), where LFT is the latest finish time and EST is the earliest start time. The buffer consumption function, B(i, d) = min(TF(i), d × α(i)), is then introduced, where d is the input delay and α(i) is the buffer consumption coefficient (ranging from 0 to 1), reflecting the task's sensitivity to delay. The buffer consumption coefficient, α(i), is calculated using task characteristic parameters: α(i) = w1 × C1 + w2 × C2 + w3 × C3, where C1 represents task criticality, C2 represents resource scarcity, C3 represents process complexity, and w1, w2, and w3 represent weight coefficients. The actual propagation delay is calculated as: PD(i) = max(0, dB(i, d)), which represents the delay that cannot be absorbed by the buffer. For example, in clothing production, the cutting task has a 30-minute float time and a buffer consumption coefficient of 0.8. When the upstream task is delayed by 40 minutes, 24 minutes of delay can be absorbed, and the actual propagation delay is 16 minutes.

[0108] The resource conflict detection mechanism combines delay propagation paths and delay amounts to identify and resolve conflicts among multiple tasks competing for the same resource in real time. This mechanism employs a resource-time window mapping technique to construct a resource occupancy schedule, RT(r,t), which records the occupancy status of each resource r at time t. In implementation, a new time window [EST'(i), LFT'(i)] is calculated for each critical task based on delay propagation. Resource occupancy overlap is then detected. When RT(r,t) > 1, a conflict exists for resource r at time t. A multi-level task prioritization rule is then applied to resolve the conflict. The prioritization rule consists of four levels: P1 (order priority) (based on customer importance and contract penalties), P2 (process criticality) (based on the impact on subsequent processes), P3 (resource specificity) (based on the task's dependence on specific resources), and P4 (remaining duration) (based on the task's remaining completion time). When a conflict is detected, the system compares task priorities in the order P1 → P2 → P3 → P4. High-priority tasks remain on their original schedules, while low-priority tasks are deferred. For example, on a clothing production line, when the sewing tasks of high-end customized orders and the sewing tasks of ordinary batch orders compete for the same dedicated sewing machine due to delays, the system will give priority to ensuring the resource needs of high-end customized orders.

[0109] The order delivery delay aggregation calculation accurately predicts the change in the final order delivery time by comprehensively analyzing the delays of a set of critical tasks. This calculation uses the critical path delay propagation model to map task-level delays to order-level delays. In its implementation, the order-task dependency graph G(O) is first constructed, containing all tasks in order O and their dependencies. Then, the order critical path CP'(O) is calculated based on the modified critical path method (MCPM), accounting for delays. The order delivery delay OD(O) = max{PD(i)|i∈CP'(O)} is then calculated, representing the maximum cumulative value of task delays on the critical path. To improve calculation accuracy, a delay decay factor λ(i,j) is introduced to represent the degree of attenuation of task i's delay when it is propagated to task j. The calculation formula is λ(i,j) = exp(-β×d(i,j)), where d(i,j) is the network distance between tasks and β is the decay coefficient. The final order delay is calculated as OD(O) = max{PD(i)×∏λ(i,j)|i∈CP'(O)}. For example, in a clothing order that includes three key tasks: cutting, sewing, and inspection, if cutting is delayed by 30 minutes, sewing is delayed by 15 minutes, and inspection is not delayed, after considering delay attenuation, the order delivery delay may be 38 minutes.

[0110] The delay prediction results generation design combines delay propagation paths and order delivery delays to create a comprehensive and intuitive delay prediction reporting system. This design utilizes a multi-level information display architecture, presenting delay impacts from a macro to micro perspective. In implementation, an order-level delay summary is first generated, including the order number, original planned completion time, predicted delay time, and delay probability distribution (generated through Monte Carlo simulation). Next, a task-level delay detail is constructed, listing all affected key tasks, including task ID, original planned time window, adjusted time window, delay amount, and buffer consumption. A delay propagation path visualization is then created, using color shading to indicate delay severity and line thickness to indicate propagation probability. Finally, a resource conflict analysis report is provided, identifying all resource conflict points caused by delays and their resolution. Prediction results can be output in a variety of formats, including interactive dashboards, PDF reports, and structured data interfaces. For example, for a clothing production order, the system generates a comprehensive report stating, "Order #A2103 is expected to be delayed by 45±8 minutes (95% confidence level), primarily due to the delay in cutting task T105. Adjustments to the schedules of the five subsequent associated tasks are recommended."

[0111] In one embodiment, generating a baseline production plan for military clothing using a global optimization intelligent algorithm and a dynamic state model includes the following steps:

[0112] Construct a multi-objective optimization model for military clothing production. The model objective functions include minimizing total completion time, maximizing equipment utilization, and minimizing delivery delays.

[0113] Extract the current production system constraints from the military clothing production management system, including equipment availability, personnel skill configuration, material inventory status and process route requirements;

[0114] Based on the genetic algorithm framework and combined with the multi-objective optimization model and constraint conditions, a global optimization intelligent algorithm is created, and the chromosome code is generated using the chromosome representation method based on process coding;

[0115] Generate the initial solution population of chromosome encoding using heuristic rules based on key resource bottleneck identification;

[0116] The initial solution population is iteratively optimized through a global optimization intelligent algorithm. When the preset upper limit of iterations is reached, the optimal population is obtained. The optimal population is analyzed to obtain the optimal scheduling plan, and the optimal scheduling plan is converted into a standardized benchmark production plan.

[0117] In this embodiment, the multi-objective optimization model for military clothing production is constructed using a weighted comprehensive evaluation method, integrating the three key production indicators into a single objective function. The mathematical expression of the model is: MinF = w1·C max +w2·(1-U)+w3·∑T j , where C max represents the total completion time (i.e., the time span from the start of production to the completion of all orders), U represents the average equipment utilization (calculated as the ratio of actual processing time to available time), ∑T j The weight coefficients w1, w2, and w3 reflect the relative importance of each objective and are determined by the analytic hierarchy process (AHP). For military equipment, w1 = 0.3, w2 = 0.2, and w3 = 0.5 are usually set to reflect the priority of delivery accuracy. Each indicator in the objective function is normalized to eliminate dimensional differences, such as C max '=(C max -C min ) / (C max -C min This multi-objective optimization model is particularly suitable for the characteristics of military clothing production. It can balance the contradiction between production efficiency and delivery accuracy, and provide clear evaluation criteria for subsequent optimization algorithms.

[0118] The constraint extraction of the military clothing production system adopts multi-source data fusion technology to obtain and formally express four key constraints from the production management system in real time. The equipment availability constraint is represented by the equipment state matrix M, where M ij represents the availability status of device i in time period j (1 means available, 0 means unavailable), including the planned maintenance time and the current fault status; the personnel skill configuration constraint is represented by the skill matching matrix S, where S ij∈[0,1] represents the skill proficiency of operator i in completing process j. The production requirement of military clothing is S ij ≥0.8 to arrange the corresponding work; the material inventory status constraint is expressed by the material available time vector V, where V i represents the earliest available time point of material i, taking into account the current inventory and expected arrival time; the process route requirement constraints are expressed by the priority relationship matrix P, where P ij =1 means that process i must be completed before process j, especially for military clothing, which has special process requirements (for example, the lamination process of bulletproof vests must be completed before sewing). The constraint extraction process uses ETL (Extract-Transform-Load) technology and is automatically updated every 15 minutes from the ERP, MES, and WMS systems.

[0119] This intelligent global optimization algorithm is based on the NSGA-II (Non-Dominated Sorting Genetic Algorithm II) framework and has been customized for the specifics of military clothing production. The algorithm uses a chromosome representation based on process encoding. Each chromosome consists of two parts: a process priority sequence and a machine assignment sequence. The process priority sequence has a length equal to the total number of processes, n, and its elements are process numbers. The machine assignment sequence also has a length of n, and its elements represent the selected machine numbers. For example, the chromosome [3,1,5,2,4|2,1,3,2,1] represents the process execution order as 3-1-5-2-4, with each process executed on machines 2, 1, 3, 2, and 1, respectively. The algorithm incorporates a special crossover operator: position-based crossover (POX) for the process sequence to maintain priority relationships between processes, and two-point crossover for the machine assignment sequence to ensure reasonable allocation. Mutation operations include exchange mutation of the process sequence and random mutation of the machine assignment sequence, with a mutation probability of 0.1. Fitness is evaluated using fast non-dominated sorting and crowding distance calculations, and a penalty function is introduced to address constraint violations. The algorithm parameters are set as: population size 100, maximum number of iterations 500, and crossover probability 0.8.

[0120] The initial solution population generation method based on critical resource bottleneck identification employs hierarchical heuristic rules to generate a high-quality initial solution population by identifying and prioritizing critical resource bottleneck processes. First, critical bottleneck resources are identified through resource load analysis. The load ratio (LR) of each resource type is calculated as ∑(processing time) / available time. The resource with the highest load ratio is labeled as the bottleneck resource. Four heuristic rules are then constructed based on bottleneck resources: R1, the bottleneck resource priority rule (which prioritizes processes using bottleneck resources); R2, the critical order priority rule (based on deadline urgency and military importance scores); R3, the shortest processing time rule; and R4, the minimum inter-process waiting time rule. When generating the initial population, 40% of the individuals use the R1 rule, 30% use the R2 rule, 15% use the R3 rule, and 15% use the R4 rule. A 10% random perturbation is introduced to ensure population diversity. For each individual generated by a rule, a greedy algorithm is used to gradually construct the complete schedule: first, the next process to be scheduled is selected according to the rule, then the optimal resource that satisfies the constraints is assigned to that process. This process is repeated until all processes are scheduled. This bottleneck resource-based initial solution generation method significantly improves the starting point quality and convergence speed of the algorithm, laying a good foundation for subsequent optimization.

[0121] The optimal scheduling solution acquisition and conversion process transforms the abstract chromosome encoding into an executable production plan through iterative optimization and analytical transformation. During the iterative optimization phase, an elite retention strategy and an adaptive mutation rate adjustment mechanism are employed to ensure stable algorithm convergence. During each iteration, 20% of the elite individuals are retained and directly advanced to the next generation. The mutation rate is dynamically adjusted based on population diversity, calculated as μ = μ0 × (1-d / D), where d is the current iteration number, D is the maximum iteration number, and μ0 is the initial mutation rate of 0.1. The algorithm terminates when the preset upper limit of 500 generations is reached or when the optimal solution remains unchanged for 50 consecutive generations, outputting a Pareto-optimal solution set. From this optimal solution set, the final solution is selected based on the decision maker's preference weights w1, w2, and w3, typically selecting the individual with the smallest weighted objective function value. The analytical phase decodes the selected chromosome into a specific scheduling solution: first, the execution order of each process is determined based on the process priority sequence; then, the resource allocation for each process is determined based on the machine allocation sequence; and finally, the precise start and completion times of each process are calculated. The final scheduling plan is converted into a standardized baseline production plan, which includes four key data tables: process-timetable (recording the start and end time of each process), resource allocation table (recording the usage arrangements of each resource), material requirement plan and personnel schedule.

[0122] In one embodiment, making an interruption response decision based on the quantified comprehensive disturbance cost and a preset decision strategy model to determine the optimal interruption execution plan includes the following steps:

[0123] Determine the disturbance scope and response time limit based on the urgency and importance of the task insertion event;

[0124] Identify all feasible interruption points based on the current production status. Each feasible interruption point corresponds to a process location on the production line that can be paused.

[0125] Calculate the comprehensive disturbance cost index of each feasible interruption point according to the comprehensive disturbance cost, and use the rule-based decision model to apply weights and threshold conditions to different comprehensive disturbance cost indexes;

[0126] The comprehensive disturbance cost index and decision model are combined to select the interruption execution plan with the minimum disturbance cost from the feasible interruption points.

[0127] In this implementation, the disturbance scope and response timeframe for task insertion events are determined using a two-dimensional assessment matrix. This approach achieves precise classification through quantitative analysis of urgency and importance. Urgency (U) is based on time pressure assessment and employs a five-level quantitative scale: U1 (extremely urgent, <2 hours), U2 (highly urgent, 2-6 hours), U3 (moderately urgent, 6-24 hours), U4 (lowly urgent, 1-3 days), and U5 (routine, >3 days). Importance (I) is based on task value assessment and also employs a five-level quantitative scale: I1 (critical, impacting national defense security), I2 (important, impacting major military operations), I3 (moderate, impacting conventional military support), I4 (general, impacting non-core support), and I5 (routine, impact can be postponed). These two dimensions combine to create 25 scenarios, corresponding to different disturbance scopes and response timeframes. The disturbance scope is categorized into four levels: R1 (full production line shutdown), R2 (related production line shutdown), R3 (local process adjustment), and R4 (no interruption required). Response time limits are divided into four levels: T1 (immediate, <30 minutes), T2 (rapid, <2 hours), T3 (routine, <8 hours), and T4 (planned, <24 hours). For example, when an emergency chemical protective clothing order (U1, I2) is inserted, the system automatically determines the disturbance range as R2 (related production line shutdown) and the response time limit as T1 (immediate response).

[0128] The identification of feasible interruption points is based on real-time production status monitoring and process flow analysis, and a multi-level screening mechanism is used to identify all process locations that can be paused on the production line. First, the current status data of all processes, including completion, equipment status, material status, and personnel configuration, are collected in real time through the Manufacturing Execution System (MES). Then, based on the process knowledge base, technically feasible interruption points are screened, and three-level filtering criteria are applied: (1) process safety filtering, excluding processes that may cause product quality problems after interruption, such as ongoing hot pressing and chemical treatment processes; (2) equipment safety filtering, excluding processes that may cause equipment damage after interruption, such as sudden shutdown of high-temperature equipment; (3) material preservation filtering, excluding processes that may cause deterioration or damage of semi-finished products after interruption. For processes that pass the safety filtering, their interruption feasibility index (IFI) is further evaluated, and the calculation formula is IFI = w1×CP+w2×ES+w3×MS, where CP is the process completion degree (0-100%), ES is the equipment status score (1-10), MS is the material status score (1-10), and w1, w2, and w3 are weight coefficients. When IFI≥7.5, the process is marked as a feasible interruption point.

[0129] The comprehensive disruption cost calculation utilizes a multi-factor weighted assessment method, conducting a comprehensive quantitative analysis of each feasible disruption point and intelligently screening them through a rule-based decision-making model. The comprehensive disruption cost (TDC) calculation formula is: TDC = α1 × TC + α2 × QC + α3 × RC + α4 × SC, where TC represents the time cost (including restart time, adjustment time, and delay time), QC represents the quality cost (including increased scrap rate and potential rework), RC represents the resource cost (including material loss and energy consumption), and SC represents the chain reaction cost (including the impact on other orders). α1, α2, α3, and α4 are weight coefficients that are dynamically adjusted based on the product type. For military equipment products, the typical weights are α1 = 0.3, α2 = 0.4, α3 = 0.1, and α4 = 0.2, reflecting the principle of quality first. The rule-based decision model uses an IF-THEN structure and applies three levels of rules: (1) hard constraint rules, such as "IF QC > 8 THEN reject the interruption point," to ensure product quality; (2) soft constraint rules, such as "IF TC > 6 AND SC > 5 THEN increase the TDC by 20%," to reflect the interactive influence of multiple factors; and (3) optimization rules, such as "IF the current process completion rate > 90% THEN prioritize completing the current process," to improve production efficiency. For example, on a military camouflage uniform production line, the TDC of sewing process A is calculated to be 6.8, while the TDC of cutting process B is 4.2. After the decision model applies the rules, the adjusted TDC of process A is 8.2 (due to the large chain reaction), while process B remains at 4.2, clearly indicating that the cutting process is the more optimal interruption point.

[0130] The optimal interruption execution plan is selected by comprehensively analyzing the disturbance cost indicators of each feasible interruption point and combining multi-objective optimization technology to determine the interruption execution plan with the minimum total disturbance cost. First, all feasible interruption points are sorted in ascending order according to the adjusted comprehensive disturbance cost indicator (TDC') to form a candidate plan set; then, considering the combined effect of the interruption points, the combined disturbance cost (CDC) is calculated, and the formula is CDC = ∑TDC' i +β×∑I(i,j), where I(i,j) represents the interaction coefficient between interruption points i and j, and β is the interaction weight. Then, the Pareto optimization principle is applied to screen out the non-dominated solution set under the response time constraint. Finally, based on the decision maker's preference (usually minimizing the total disturbance cost), the optimal interruption execution plan is selected from the non-dominated solution set. After the plan is selected, the system automatically generates detailed interruption execution instructions, including: (1) interruption time point and location; (2) work-in-process storage method; (3) equipment safety shutdown steps; (4) personnel deployment arrangements; (5) restart preparations. This optimal interruption plan selection method based on comprehensive disturbance cost significantly improves the scientific and economic efficiency of production scheduling, while ensuring timely response to emergency orders and orderly resumption of regular production.

[0131] In one embodiment, invoking a fast local rescheduling algorithm coupled with a disturbance cost indicator according to an interruption execution plan to adaptively adjust the affected baseline production plan segments with the goal of minimizing the actual disturbance impact includes the following steps:

[0132] Identify the affected baseline production plan segments based on the interruption execution plan and determine the plan scope that needs to be rescheduled;

[0133] Constructing a local rearrangement objective function based on the comprehensive perturbation cost index and the planning scope;

[0134] Apply a fast scheduling algorithm based on priority rules and determine the reordering of tasks according to task urgency, process requirements and resource utilization efficiency;

[0135] Under the premise of satisfying process constraints, the local rearrangement objective function and rearrangement order are combined and an incremental update method is adopted to insert urgent tasks and rearrange the affected tasks.

[0136] In this implementation, a spatiotemporal dual-dimensional analysis method is used to identify the affected baseline production plan segments based on the interruption execution plan, accurately determining the scope of the plan that requires rescheduling. First, in the temporal dimension, the impact boundary is determined through forward and backward propagation analysis: the forward analysis starts from the interruption point and calculates the time delay impact of the insertion of the urgent task on subsequent tasks until the delay is less than a preset threshold (usually 15 minutes); the backward analysis traces back from the interruption point to identify tasks that may need to be completed early to free up resources. Second, in the spatial dimension, the impact scope is determined through resource association network analysis: a resource dependency graph (RDG) is constructed, with nodes representing production resources (equipment, personnel, workstations, etc.) and edges representing the relationships between resources. A breadth-first search algorithm is used, starting from the resource nodes involved in the interruption point, to mark all reachable nodes as potentially affected resources. Then, applying the spatiotemporal intersection rule, the affected plan segment is defined as the set of tasks that meet both the temporal impact boundary and the resource impact scope.

[0137] The local rearrangement objective function is constructed using a multi-objective weighted combination method, integrating the comprehensive disturbance cost index and the key performance indicators within the planning range into a unified optimization goal. The mathematical expression of the objective function is: MinF = λ1×TDC+λ2×Cmax+λ3×∑w i T i +λ4×∑S ij , where TDC is the comprehensive disturbance cost index, reflecting the overall impact of the interruption on the production system; Cmax is the maximum completion time within the rescheduled plan range, reflecting production efficiency; ∑w i T i is the sum of weighted task delays, reflecting the timeliness of delivery, w i is the importance weight of task i, T i is the delay time of task i; ∑S ij is the sum of the number of equipment switching times, reflecting production stability. The weight coefficients λ1, λ2, λ3, and λ4 are dynamically adjusted based on different application scenarios. For military emergency tasks, the typical settings are λ1 = 0.3, λ2 = 0.2, λ3 = 0.4, and λ4 x 0.1, prioritizing timely delivery. The objective function also includes multiple constraints: process path constraints, resource capacity constraints, delivery constraints, and quality constraints. To improve solution efficiency, piecewise linearization is used to handle nonlinear terms, and slack variables are introduced to handle hard constraints. This multi-objective fusion of the locally rearranged objective function allows the optimization process to consider both minimizing interruption costs and balancing production efficiency and timely delivery.

[0138] The application of a fast scheduling algorithm based on priority rules adopts a multi-level dynamic priority mechanism to determine the reordering of tasks according to task urgency, process requirements, and resource utilization efficiency. First, a three-level priority assessment system is established: the first level is task urgency assessment, which uses the weighted urgency index (WUI) calculated using the formula WUI = (Dt) / PT × IW, where D is the due date, t is the current time, PT is the remaining processing time, and IW is the importance weight. The second level is process requirement assessment, which considers process continuity requirements (CS), quality sensitivity (QS), and process complexity (PC) to comprehensively calculate the process priority index (TPI) using the formula TPI = γ1 × CS + γ2 × QS + γ3 × PC. The third level is resource utilization efficiency assessment, which calculates the resource matching degree (RM) and load balance degree (LB) to form the resource efficiency index (REI) using the formula REI = δ1 × RM + δ2 × LB. Then, by weighting and combining the three levels of indicators, a comprehensive priority score (CPS) is calculated using the formula: CPS = η1 × WUI + η2 × TPI + η3 × REI. All tasks within the planned scope are ranked according to the CPS value. In practice, differentiated weights are assigned to different types of tasks. For example, for urgent military equipment maintenance tasks, the typical weights are set to η1 = 0.6, η2 = 0.3, and η3 = 0.1, prioritizing urgency.

[0139] Incremental task rescheduling and updating employs a progressive adjustment strategy. While satisfying process constraints, it combines the local rescheduling objective function and rescheduling order to efficiently insert urgent tasks and reschedule affected tasks. First, a sliding time window approach is used to divide the rescheduling interval, dividing the planning horizon into multiple consecutive time windows (typically 2-4 hours), prioritizing the near-term window. Then, within each time window, tasks are scheduled sequentially using a forward scheduling approach according to the predefined rescheduling order. Urgent tasks are prioritized for the earliest available suitable resource. For affected routine tasks, a task dependency graph (TDG) is constructed based on process constraints to ensure that subsequent tasks are scheduled only after the completion of the predecessor process. A resource conflict detection and resolution mechanism is also implemented. When multiple tasks compete for the same resource, resource allocation is determined based on priority scores. During the rescheduling process, the objective function value is continuously evaluated, and a greedy strategy is used to move between local optima. To maintain plan stability, a "minimum perturbation principle" is introduced, requiring adjustments to be made only when the improvement in the objective function resulting from an adjustment exceeds a preset threshold (typically 5%). After completing the rescheduling of a time window, the scheduling results for that window are solidified and the process continues with the next window until the entire plan is covered. This incremental task rescheduling and updating method ensures the timely insertion of urgent tasks while maximizing the stability of the original plan, while also improving the reliability of plan execution and the responsiveness of the production system.

[0140] The present invention also discloses a military clothing production scheduling optimization system based on an intelligent algorithm, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the military clothing production scheduling optimization method based on the intelligent algorithm as described in any one of the above embodiments is implemented.

[0141] Among them, the processor can adopt a central processing unit (CPU). Of course, according to actual usage, other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be adopted. The general-purpose processor can adopt a microprocessor or any conventional processor, etc., and this application does not impose any restrictions on this.

[0142] Among them, the memory can be an internal storage unit of a computer device, such as a hard disk or memory of a computer device, or an external storage device of a computer device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD) or flash memory card (FC) equipped on the computer device. In addition, the memory can also be a combination of an internal storage unit and an external storage device of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or is to be output. This application does not impose any restrictions on this.

[0143] The present invention also discloses a computer-readable storage medium, which stores instructions. When the instructions are executed by a processor, the processor is configured to execute the military clothing production scheduling optimization method based on intelligent algorithms described in any of the above embodiments.

[0144] Among them, the computer program can be stored in a machine-readable medium, the computer program includes computer program code, the computer program code can be in the form of source code, object code, executable file or certain middleware, etc. The machine-readable medium includes any entity or device that can carry computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the machine-readable medium includes but is not limited to the above-mentioned components.

[0145] Among them, the power transmission line comprehensive fault detection method in the above embodiment is stored in the computer-readable storage medium through the computer-readable storage medium, and is loaded and executed on the processor to facilitate the storage and application of the above method.

[0146] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0147] The one or more embodiments of this application are intended to encompass all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of this application should be included in the scope of protection of this application.

Claims

1. A method for optimizing production scheduling of military clothing based on intelligent algorithms, characterized in that: The steps include: Construct a dynamic state model of the military clothing production process and establish a production disturbance cost assessment system based on real-time production process parameters and prediction models; Generate a benchmark production plan for military clothing using a global optimization intelligent algorithm and a dynamic state model; Deploy an event perception module to capture task insertion events during real-time production according to the baseline production plan, and simultaneously collect real-time production parameters associated with the task insertion events; When the event perception module captures a task insertion event, it triggers the production disturbance cost assessment system, which uses the collected real-time production parameters to dynamically predict and quantify the comprehensive disturbance costs of different interruption scenarios. Make interruption response decisions based on the quantified comprehensive disturbance cost and the preset decision-making strategy model to determine the optimal interruption execution plan; According to the interruption execution plan, a fast local rescheduling algorithm coupled with a disturbance cost indicator is called to adaptively adjust the affected baseline production plan segments with the goal of minimizing the actual disturbance impact. Generate and issue adaptively adjusted executable production instruction sequences.

2. The method for optimizing production scheduling of military clothing based on intelligent algorithms according to claim 1, characterized in that: The establishment of a production disturbance cost assessment system based on real-time production process parameters and prediction models includes the following steps: A machine learning model is used to build a material loss prediction model and a setup time prediction model based on equipment status. The setup time prediction model inputs the current equipment operating status and the process characteristics of the interrupted and inserted tasks. The setup time prediction model outputs the dynamic prediction of the setup time required to execute the interruption switching and subsequent task recovery. A quality risk assessment model is constructed based on a pre-trained quality predictor. Both the quality risk assessment model and the material loss prediction model input the current process interrupted by the task insertion event, the elapsed processing time, and real-time production parameters. The quality risk assessment model outputs the dynamic probability of quality defects introduced by the task insertion event, and the material loss prediction model outputs the probability of material scrap caused by the task insertion event. A chain delay propagation analysis module based on the baseline production plan topology is constructed. Through the chain delay propagation analysis module and fast graph path analysis, the cascading delay effect of task insertion events on subsequent related tasks and order completion times is estimated. The material loss prediction model, preparation time prediction model, quality risk assessment model and chain delay propagation analysis module are integrated to form a production disturbance cost assessment system.

3. The method for optimizing production scheduling of military clothing based on intelligent algorithm according to claim 2, characterized in that: Building a quality risk assessment model based on pre-trained quality predictors includes the following steps: Collect historical production data of military clothing and construct a multi-dimensional feature training set. The historical production data includes historical process parameters, historical equipment status, historical interruption event records, and corresponding quality inspection results. Extract interruption-sensitive features from the multi-dimensional feature training set through feature engineering; A quality risk assessment model based on a two-stream neural network structure is constructed based on the pre-trained quality predictor. The two-stream neural network structure is used to simultaneously process the multi-dimensional feature training set and interruption-sensitive features. An attention-based interruption impact assessment unit is designed in the quality risk assessment model. The interruption impact assessment unit is used to dynamically adjust the parameter weights in the two-stream neural network structure according to the time point of the interruption, the current process completion degree and the material status.

4. The method for optimizing production scheduling of military clothing based on intelligent algorithm according to claim 2, characterized in that: A chain delay propagation analysis module based on the baseline production plan topology is constructed. The following steps are involved in estimating the cascading delay effects of task insertion events on subsequent associated tasks and order completion times using fast graph path analysis: Convert the baseline production plan into a directed weighted graph structure, where nodes represent production tasks, edges represent dependencies between tasks, and edge weights include time dependency and resource sharing. Based on the constructed directed weighted graph, a multi-level task association matrix is established to quantify the association strength between tasks from four dimensions: process dependency, resource usage, material flow, and personnel deployment. By using the task dependency matrix and directed weighted graph and adopting fast graph path analysis to perform forward scanning, the delay propagation path and the critical task set directly affected by the task insertion event are identified; The cascading delay effect of task insertion events on subsequent associated tasks and order completion time is predicted based on the critical task set.

5. The method for optimizing production scheduling of military clothing based on intelligent algorithm according to claim 4, characterized in that: The cascading delay effect of task insertion events on subsequent associated tasks and order completion times is predicted based on the key task set, including the following steps: A time buffer consumption model is introduced to calculate the dynamic delay absorption capacity of the floating time of each critical task set and predict the actual propagation delay amount; A resource conflict detection mechanism is developed based on the delay propagation path and delay amount. When delay propagation causes multiple tasks to compete for the same resource, the resource conflict is resolved through preset task priority rules. Aggregate the delays of all key task sets to calculate the order delivery delay; Combining the delay propagation path and order delivery delay prediction design, the delay prediction results of subsequent associated tasks and order completion time are generated.

6. The method for optimizing production scheduling of military clothing based on intelligent algorithm according to claim 1, characterized in that: Generating a baseline production plan for military clothing using a global optimization intelligent algorithm and a dynamic state model includes the following steps: Construct a multi-objective optimization model for military clothing production. The model objective functions include minimizing total completion time, maximizing equipment utilization, and minimizing delivery delays. Extract the current production system constraints from the military clothing production management system, including equipment availability, personnel skill configuration, material inventory status and process route requirements; Based on the genetic algorithm framework and combined with the multi-objective optimization model and constraint conditions, a global optimization intelligent algorithm is created, and the chromosome code is generated using the chromosome representation method based on process coding; Generate the initial solution population of chromosome encoding using heuristic rules based on key resource bottleneck identification; The initial solution population is iteratively optimized through a global optimization intelligent algorithm. When the preset upper limit of iterations is reached, the optimal population is obtained. The optimal population is analyzed to obtain the optimal scheduling plan, and the optimal scheduling plan is converted into a standardized benchmark production plan.

7. The method for optimizing production scheduling of military clothing based on intelligent algorithm according to claim 1, characterized in that: Based on the quantified comprehensive disturbance cost and the preset decision-making strategy model, the interruption response decision is made and the optimal interruption execution plan is determined, which includes the following steps: Determine the disturbance scope and response time limit based on the urgency and importance of the task insertion event; Identify all feasible interruption points based on the current production status. Each feasible interruption point corresponds to a process location on the production line that can be paused. Calculate the comprehensive disturbance cost index of each feasible interruption point according to the comprehensive disturbance cost, and use the rule-based decision model to apply weights and threshold conditions to different comprehensive disturbance cost indexes; The comprehensive disturbance cost index and decision model are combined to select the interruption execution plan with the minimum disturbance cost from the feasible interruption points.

8. The method for optimizing production scheduling of military clothing based on intelligent algorithm according to claim 1, characterized in that: The interruption execution plan calls a fast local rescheduling algorithm coupled with a disturbance cost indicator to adaptively adjust the affected baseline production plan segments with the goal of minimizing the actual disturbance impact. The steps include: Identify the affected baseline production plan segments based on the interruption execution plan and determine the plan scope that needs to be rescheduled; Constructing a local rearrangement objective function based on the comprehensive perturbation cost index and the planning scope; Apply a fast scheduling algorithm based on priority rules and determine the reordering of tasks according to task urgency, process requirements and resource utilization efficiency; Under the premise of satisfying process constraints, the local rearrangement objective function and rearrangement order are combined and an incremental update method is adopted to insert urgent tasks and rearrange the affected tasks.

9. A military clothing production scheduling optimization system based on an intelligent algorithm, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for optimizing the production scheduling of military clothing based on an intelligent algorithm as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instruction is executed by a processor, the processor is configured to execute the military clothing production scheduling optimization method based on an intelligent algorithm according to any one of claims 1 to 8.

Images