A paper product packaging web slitting margin optimization and scheduling method

Abstract

This invention provides a method for optimizing and scheduling the slitting surplus of paper rolls used in paper product packaging. The method includes: collecting physical properties and market data of the surplus material to construct economic value nodes and an initial value map; predicting future short-term demand based on historical order data to generate dynamic opportunity windows for specification priorities; integrating the surplus material resource pool with order parameters, using a path graph model to comprehensively consider scheduling costs, value loss, mold change preparation, and production scheduling disturbances; employing an improved Dijkstra algorithm and Shapley value mechanism to select the optimal reuse path; and achieving closed-loop control and risk warning through edge computing nodes and cloud collaboration. This invention can improve the utilization rate of surplus materials, optimize production economics, and reduce management risks.

Description

Technical Field

[0001] This invention relates to the field of waste material management and intelligent scheduling in paper product packaging production, and in particular to a method for optimizing and scheduling the slitting waste of paper rolls in paper product packaging. Background Technology

[0002] In the existing paper packaging industry's slitting and scheduling processes, surplus material management generally adopts the traditional model of static inventory statistics and post-event redistribution. In conventional systems, after roll paper is slitting off the production line, the remaining materials (such as remaining lengths and narrow widths) are usually simply recorded and managed according to the remaining physical quantity and basic material information. Subsequent utilization decisions mainly rely on manual experience or periodic inventory clearance, lacking automated and intelligent surplus material flow and efficient reuse mechanisms in multi-order scenarios. The current mainstream practice is mainly "consumption within the same order" or "centralized clearance upon expiration of inventory age," with only occasional attempts to cross-order surplus material transfer by manually comparing with known future orders. However, this relies on manual searching, is limited by information asymmetry and the uncertainty of dynamic scheduling, resulting in slow scheduling response and limited overall surplus material utilization and production system flexibility. With increasingly diversified customers and fragmented orders, the market demands higher efficiency in raw material utilization, faster production response times, and greater cost control. This has spurred the technological demands for new production models such as "proactive operation of surplus material assets" and "autonomous scheduling across orders." In recent years, some manufacturers in the industry have introduced RFID electronic tags, material location tracking integrated with basic ERP systems, and rule-driven surplus material allocation to improve the timeliness of inventory checks and the ability to automatically issue order parameters. However, most of these technical solutions are limited to the allocation of surplus materials in a single order loop or a local production line. Faced with high-concurrency, large-volume order combinations of heterogeneous materials, they lack a real-time dynamic surplus material call mechanism and globally optimal path planning capabilities. Especially in the face of complex backgrounds such as frequent changes in order product structure, changes in production cycle time, and fluctuations in main material prices, system scheduling strategies often fall into the dilemma of "uncontrolled scheduling complexity and switching costs." Representative application scenarios, such as current MES systems and integrated production planning and inventory management platforms, while enabling automatic maintenance of main material and by-product ledgers and automatic priority use of homogeneous surplus materials, lack intelligent analysis of the coupling between the dynamic value of surplus materials and order demand time windows, and are largely unable to perceive potential scheduling and coordination opportunities between orders. Even if some systems support rule-based matching of surplus materials and demand, the lack of quantification of the physical state of surplus materials, value decay, and substitution and switching costs during scheduling processes leads to scheduling results that are mostly locally optimal but globally unbalanced. Furthermore, some systems adopt a "surplus material priority" strategy, processing logic based on "first-come, first-served" and "similar specifications," but when multiple surplus materials and multiple orders compete across batches, they fail to consider various practical constraints such as opportunity window elimination, stable main production scheduling, and switching costs, easily leading to low efficiency in surplus material reuse or even exacerbating production disruptions. Summary of the Invention

[0003] In order to solve the above-mentioned technical problems, the present invention provides a method for optimizing the slitting allowance and scheduling of paper product packaging rolls.

[0004] The technical solution of this invention is implemented as follows: A method for optimizing and scheduling the slitting allowance of paper product packaging rolls, comprising: S1: After each roll paper slitting operation is completed, the physical characteristic data of each scrap material is collected. The physical characteristic data includes the remaining length, width, material grade, storage location and physical condition score. Combined with the real-time market raw material price and the mapping relationship between alternative use scenarios, the comprehensive value index of each roll of scrap material is calculated to form an initial scrap material value map. S2: Based on historical order patterns, customer delivery cycles and product specification distribution data, construct a short-term order demand forecasting model, output multiple order types that may be triggered in the next 48 hours and their target width and length ranges, and generate a dynamic sequence of surplus material matching opportunity windows. Each opportunity window includes a time interval, matching specification range and priority weight. S3: Input the demand parameters of the current orders to be scheduled into the cross-order scheduling decision engine, and simultaneously input the sequence of surplus material matching opportunity windows within the effective time range and the initial surplus material value map to start the multi-dimensional matching calculation process; S4: In the cross-order scheduling decision engine, based on each leftover material node in the initial leftover material value graph and the target demand node in the opportunity window sequence, a constrained leftover material reuse path graph model is constructed, where the cost function of the edge is a weighted synthesis of leftover material value loss rate, scheduling distance cost, mold change preparation time and disturbance to the main line scheduling. S5: The improved Dijkstra algorithm is used to search for the optimal set of paths that meet the production constraints in the waste material reuse path graph model, and obtain multiple feasible waste material call path schemes. Each scheme is accompanied by a total cost index and a resource saving potential assessment value. S6: Based on the Shapley value allocation mechanism in game theory, the marginal contribution of each feasible waste material utilization path is calculated to obtain the comprehensive benefit score of each path to improve the overall resource efficiency, and the optimal reuse path with the highest comprehensive benefit is selected from the candidate set based on the score. S7: Determine whether the expected cost savings of the surplus material corresponding to the optimal reuse path exceeds 1.8 times its total scheduling cost. If the trigger threshold condition is met, activate the cross-order surplus material call process and generate a preliminary production scheduling instruction containing the main material usage plan and the surplus material call path suggestion. S8: By executing the status update of remaining materials and preliminary path screening through the lightweight edge computing nodes deployed locally, the global strategy after iterative optimization in the cloud center is synchronized to the edge end, ensuring that the production scheduling instruction is output to the production scheduling system with a risk warning label, thus completing closed-loop control.

[0005] The present invention provides a method for optimizing the slitting allowance and scheduling of paper product packaging rolls, which has the following beneficial effects: (1) This invention realizes proactive operation of scrap assets throughout their entire lifecycle by constructing a dynamically updated scrap value map. After each slitting operation, the length, width, material grade, physical state, and storage location of the scrap are automatically collected. The real-time market raw material price and the mapping relationship of alternative use scenarios are integrated to quantify the comprehensive value index of each roll of scrap. This mechanism effectively improves the level of refinement of scrap assessment, can identify high-potential reuse resources, and significantly enhances the responsiveness and adaptation accuracy of scrap in small-batch, multi-variety production environments. It fundamentally changes the previous lagging logic of "first order, then find scrap" to a forward-looking scheduling paradigm of "predicting demand opportunities based on the potential of scrap", which greatly reduces the backlog of ineffective inventory and the phenomenon of repeated material feeding. (2) This invention introduces a demand forecasting model based on historical order patterns and delivery cycle analysis to accurately identify adaptable order types that may appear in the next 48 hours, generate time-sensitive "surplus material adaptation opportunity windows," and assign different window priority weights, thereby establishing a task-driven mechanism oriented towards timeliness matching. On this basis, the system uses an improved constrained Dijkstra algorithm to search for the optimal path in the surplus material map. The cost function comprehensively considers the surplus material value loss rate, scheduling distance, mold change preparation time, and the degree of impact on mainline production, ensuring that the recommended solution is both economical and feasible. Furthermore, when there are multiple optional scheduling paths, a Shapley value allocation mechanism is introduced to evaluate the marginal contribution of each path to the overall resource efficiency, realizing scientific decision-making under multi-objective collaborative optimization. This design significantly improves the intelligence level of cross-order scheduling, maximizing the potential of surplus material utilization while ensuring the stable operation of the mainline. (3) This invention constructs an "edge-cloud" collaborative computing architecture, supporting efficient closed-loop control in dynamic environments. Lightweight edge nodes are deployed locally on the production line, responsible for real-time collection of changes in the status of surplus materials and completing preliminary path selection, greatly reducing data processing latency; the cloud platform focuses on global strategy optimization, model iterative training, and long-term performance evaluation, forming a continuously evolving intelligent scheduling system. At the same time, an adaptive trigger threshold mechanism is set, activating the cross-order call process only when the expected cost savings from surplus materials exceed 1.8 times the total scheduling cost, avoiding unnecessary resource disturbances. The final output production scheduling instruction not only includes the main material configuration scheme, but also integrates surplus material call path suggestions and risk warning indicators to assist in rapid manual decision-making. Attached Figure Description

[0006] Figure 1 This is a flowchart of a method for optimizing the slitting allowance and scheduling of paper product packaging rolls according to the present invention; Figure 2 This is a sub-flowchart of a method for optimizing the slitting allowance and scheduling of paper product packaging rolls according to the present invention; Figure 3 This is another sub-flowchart of the method for optimizing the slitting allowance and scheduling of paper product packaging rolls according to the present invention. Detailed Implementation

[0007] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0008] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0009] like Figure 1 As shown, this invention provides a method for optimizing the slitting allowance and scheduling of paper product packaging rolls, specifically including: S1: After each roll paper slitting operation is completed, the physical characteristic data of each scrap material is collected. The physical characteristic data includes the remaining length, width, material grade, storage location and physical condition score. Combined with the real-time market raw material price and the mapping relationship between alternative use scenarios, the comprehensive value index of each roll of scrap material is calculated to form an initial scrap material value map. S2: Based on historical order patterns, customer delivery cycles and product specification distribution data, construct a short-term order demand forecasting model, output multiple order types that may be triggered in the next 48 hours and their target width and length ranges, and generate a dynamic sequence of surplus material matching opportunity windows. Each opportunity window includes a time interval, matching specification range and priority weight. S3: Input the demand parameters of the current orders to be scheduled into the cross-order scheduling decision engine, and simultaneously input the sequence of surplus material matching opportunity windows within the effective time range and the initial surplus material value map to start the multi-dimensional matching calculation process; S4: In the cross-order scheduling decision engine, based on each leftover material node in the initial leftover material value graph and the target demand node in the opportunity window sequence, a constrained leftover material reuse path graph model is constructed, where the cost function of the edge is a weighted synthesis of leftover material value loss rate, scheduling distance cost, mold change preparation time and disturbance to the main line scheduling. S5: The improved Dijkstra algorithm is used to search for the optimal set of paths that meet the production constraints in the waste material reuse path graph model, and obtain multiple feasible waste material call path schemes. Each scheme is accompanied by a total cost index and a resource saving potential assessment value. S6: Based on the Shapley value allocation mechanism in game theory, the marginal contribution of each feasible waste material utilization path is calculated to obtain the comprehensive benefit score of each path to improve the overall resource efficiency, and the optimal reuse path with the highest comprehensive benefit is selected from the candidate set based on the score. S7: Determine whether the expected cost savings of the surplus material corresponding to the optimal reuse path exceeds 1.8 times its total scheduling cost. If the trigger threshold condition is met, activate the cross-order surplus material call process and generate a preliminary production scheduling instruction containing the main material usage plan and the surplus material call path suggestion. S8: By executing the status update of remaining materials and preliminary path screening through the lightweight edge computing nodes deployed locally, the global strategy after iterative optimization in the cloud center is synchronized to the edge end, ensuring that the production scheduling instruction is output to the production scheduling system with a risk warning label, thus completing closed-loop control.

[0010] Step S1: After each roll of paper is slit, physical characteristic data of each scrap material is collected. This data includes remaining length, width, material grade, storage location, and physical condition score. Combined with the mapping relationship between real-time market raw material prices and alternative use scenarios, a comprehensive value index for each roll of scrap material is calculated to form an initial scrap material value map. Specifically, this includes: S1.1: Acquire real-time scrap output data stream from the end of the roll paper slitting equipment. The data stream includes the remaining length, width, material grade, storage location coordinates, and physical state score evaluated by the machine vision system for each roll of scrap. Based on the industrial sensor network transmission mechanism, perform timestamp alignment and outlier filtering on the original sensor data to generate a structured scrap attribute dataset. At the end of the roll paper slitting equipment, a high-speed industrial data acquisition interface protocol (parameters: sampling period 1s, data packet format JSON Schema definition) is adopted to realize real-time signal capture of residual material length, width, material grade, storage location coordinates, and physical state score generated by the machine vision system; Furthermore, by combining a machine vision system with a roll end surface texture analysis algorithm (parameters: sample window size 64×64 pixels, gray-level co-occurrence matrix orientation angle 0° / 45° / 90° / 135°), the physical state score of the scrap material is automatically generated, and the score value is used as the quality indicator input. Furthermore, through the industrial sensor network transmission mechanism (parameters: IEEE 1588 PTP clock synchronization protocol, PRP data link layer redundancy mechanism), the timestamp alignment of all raw sensor data streams is achieved, ensuring that the data synchronization accuracy between different sensor sources is less than 1ms; Furthermore, an outlier filtering algorithm (parameters: 3σ principle, sliding window length of 10 sampling periods) is used to remove significantly deviating data points in the time series and generate a noise-free and stable data stream. Furthermore, through a structured data mapping method (parameters: scrap attribute field mapping table, data type forced conversion rules), the original sensor data after time-series alignment and outlier filtering is converted into a scrap attribute dataset; during the conversion process, the length and width fields maintain floating-point precision to three decimal places, the material grade field is standardized to a unified coding system, and the storage location coordinate system adopts the global coordinate system of the factory MES system. Through the above algorithm processing method, the raw waste output data stream of the previous step is transformed into a structured waste attribute dataset with complete structure and unified field types, so as to realize high-precision data foundation support for subsequent value calculation and map construction. For example, at the end of a cutting machine on a production line, the remaining length of the scrap material collected in real time is 52.375 meters, the width is 1.248 meters, the material grade code is M03, and the storage location coordinates are (12.54, 8.29). The machine vision system outputs a physical state score of 87. Clock alignment is performed using the IEEE 1588 PTP protocol, and the synchronization deviation of all sensor data timestamps is detected to be 0.64 ms. Outlier filtering based on the 3σ principle is used to detect and remove two data points with abnormal fluctuations in width measurement values ​​within a sliding window of 10 sampling periods, resulting in a smooth length and width sequence. A field mapping table is used to convert the material grade M03 to the standard code C-M03, and the coordinate values ​​are converted to the MES global coordinate format. The final scrap material attribute dataset records include: length 52.375, width 1.248, material C-M03, coordinates (12.54, 8.29), and state score 87. This dataset was directly used as input in the subsequent material-scene matching algorithm call in S1.2, effectively supporting the accurate calculation of the comprehensive value index of surplus materials and significantly improving the reliability of value map construction; S1.2: Call the real-time market raw material price list and multi-level alternative use scenario mapping rule library maintained in the central database. The mapping rules are predefined based on the material compatibility matrix and customer certification level constraints. Use the fuzzy matching algorithm to match the material grade and specification parameters of the current surplus material to the category of alternative main material to determine its potential application scenario set and its corresponding value weight coefficient. S1.3: Based on the structured scrap attribute dataset and potential application scenario set, perform weighted value synthesis calculation: adopt a multi-factor linear combination model to convert the remaining length and width into equivalent usable area, multiply by the market raw material price per unit area in the corresponding scenario, and introduce physical condition score as depreciation decay coefficient to calculate the candidate value items of the scrap in each suitable scenario. S1.4: Select the maximum value from all candidate value items as the benchmark value output for the surplus material, and correct the value by combining it with the estimated storage and scheduling costs to generate a cost-normalized comprehensive value index; this index is bound to the corresponding surplus material identifier as a core attribute to form a single value node with economic dimension description capabilities. S1.5: Aggregate all individual value nodes according to a unified data structure to construct an initial surplus material value map containing surplus material ID, comprehensive value index, spatial location, technical status and suitable scenario tags; this map is stored as a weighted undirected graph data structure in the local cache of the edge computing node for subsequent cross-order scheduling decision engine to support real-time path search and matching calculation.

[0011] Step S2: Based on historical order patterns, customer delivery cycles, and product specification distribution data, a short-term order demand forecasting model is constructed. This model outputs multiple order types that may be triggered within the next 48 hours, along with their target width and length ranges. A dynamic sequence of surplus material matching opportunity windows is then generated, with each opportunity window containing a time interval, a matching specification range, and a priority weight. Specifically, this includes: S2.1: Based on the structured order records in the historical order database, extract the industry category, customer role tag, product specifications (width, length), order time, promised delivery cycle and actual completion time for each order. Use the time series decomposition algorithm to separate the order arrival rate into trend items, seasonal items and residual items to identify the periodic purchasing patterns and sudden demand fluctuations of customers in different sub-sectors and generate an order behavior pattern feature set. Based on the structured order records in the historical order database, the full dataset containing fields such as industry category, customer role tag, product specifications, order time, promised delivery cycle and actual completion time is called. Field parsing and data cleaning rules (parameters: missing value filling, outlier removal, format standardization) are used to achieve structured preprocessing of the input order records. Furthermore, a time series reconstruction algorithm (parameters: timestamp accuracy at the hour level, gap interpolation method: cubic spline) is used to model the continuous time series of order arrival rate, and a trend separation algorithm (method: moving average window period of 7 days) is applied based on the modeling results to obtain trend data reflecting long-term changes. Furthermore, seasonality terms were identified and extracted using a periodicity test algorithm (method: Fourier spectral analysis, parameters: sampling frequency of 1 / day, frequency domain window length of 365 days), separating annual, quarterly, and monthly frequency components to provide a basis for matching periodic patterns at different time scales; Furthermore, using residual analysis (parameters: residual extraction method is the sequence after removing trend and seasonal terms, and noise filtering threshold is ±2 standard deviations), residual terms representing irregular fluctuations are obtained, and sudden peak detection is performed on them to determine possible abnormal large orders or centralized procurement events. The decomposition of the trend term, seasonality term, and residual term is formally described using the following mathematical formulas:

[0012] in, This represents a sequence of order arrival rates. For trend items, It is a seasonal item. For residual terms; Furthermore, an order behavior feature encoding method (parameters: industry category encoding weight, customer role label binarization, specification parameter range processing) is adopted to jointly assemble the above three components with the attribute features of the original order into a high-dimensional feature matrix, forming an order behavior pattern feature set; By normalizing and standardizing features, trend features, seasonal features and sudden fluctuation features are mapped to a unified dimension space, enabling cross-dimensional and cross-field feature comparison capabilities. Through the above algorithm chain, the time series decomposition results and order attribute features of the previous step are transformed into a set of order behavior pattern features that can support the training of short-term order prediction models, so as to accurately identify the periodic purchasing patterns and sudden demand fluctuations of customers in different sub-sectors. For example, in the order database of a paper packaging company, structured order records from the past 36 months are retrieved. Fields include industry category (e.g., food, daily chemicals, electronics), customer role tags (direct sales customer or distributor), product specifications (width range 500mm~1500mm, length range 1000m~5000m), order time (accurate to the minute), promised delivery period (3~10 days), and actual completion time. Based on this dataset, missing value imputation is set to the mean of adjacent orders, and outlier removal threshold is set at three times the historical average of the industry. Cubic spline interpolation is used to interpolate the timestamp series, with a moving average window period of 7 days to extract the trend term. Fourier analysis sampling frequency is 1 / day with a window length of 365 days to capture annual and quarterly seasonality. The threshold for detecting sudden peaks in residual terms is set to ±2 standard deviations. The decomposition reveals that the trend term shows a steady increase in food industry order volume, the seasonality term reveals a significant peak in electronics orders in the fourth quarter of each year, and the residual term shows short-term concentrated purchasing in the daily chemical industry during promotional months. After encoding trend, seasonality, and residual components with industry category, customer role, and specification parameters, a high-dimensional feature matrix is ​​formed and normalized to a unified dimension space. The output order behavior pattern feature set successfully characterizes the stable growth pattern of the food industry, the seasonal fluctuation pattern of the electronics industry, and the promotional peak pattern of the daily chemical industry, providing effective data support for subsequent 48-hour order trigger prediction. S2.2: Based on the order behavior pattern feature set, combined with the market activity index of the customer's sub-industry and the recent promotional activity calendar information, the XGBoost ensemble learning model is used to perform multi-dimensional feature weighted fusion calculation to predict the order trigger probability distribution under each sub-business scenario in the next 48 hours, and output a set of pending order types with high confidence (≥90%) as the initial input condition for short-term demand forecasting; Based on the order behavior pattern feature set generated by S2.1, the market activity index time series table in the customer segment industry database is called and aligned with the recent promotional activity calendar. A sliding time window mechanism (window width of 48 hours) is used to realize the parallel feature loading of industry activity and promotional events, forming a multi-dimensional input matrix containing basic demand driving factors. By using a feature normalization algorithm (Z-score standardization, with the mean and variance taken from the most recent 180 days of historical data), the scale consistency of features with different dimensions in the same feature space is achieved, thereby reducing the bias in model training and providing a stable input distribution for subsequent ensemble learning. The XGBoost ensemble learning model (100 iterations, 0.05 learning rate, and 8 maximum tree depth) is used. The order trigger status is used as the label vector. Multi-dimensional feature weighted fusion calculation is performed. The high-weight features are preferentially split based on the feature importance ranking results to improve the model's prediction sensitivity. Furthermore, based on the predicted output probability of XGBoost, an order trigger probability distribution curve is constructed, and the confidence interval estimation method is used to calculate the confidence index of the prediction result. The confidence index is calculated using the following formula:

[0013] in For credibility, The standard deviation of the predicted probability. To predict the probability mean, high-confidence samples are used for subsequent target set output; By using a threshold determination function, order types with a confidence level greater than or equal to 0.90 are filtered into a set of high-confidence pending order types, and their respective sub-business scenario tags are bound as the initial input conditions for short-term demand forecasting. Through the above-mentioned ensemble learning and confidence filtering, the market activity and promotional event-driven results of the previous step are transformed into a set of pending order types that can be directly called by the S2.3 matching rules, thereby achieving the technical effect of improving the accuracy and stability of short-term order demand forecasting. For example, in the implementation environment of a paper product packaging company, the input order behavior pattern feature set includes two industries: catering takeout packaging and e-commerce express packaging. The market activity index is 0.82 for the catering industry and 0.76 for the e-commerce industry. Recent promotional activities for the catering industry are concentrated between the 12th and 24th hours, while those for the e-commerce industry are concentrated between the 36th and 48th hours. After Z-score standardization, the mean activity index for the two industries is 0 and the variance is 1, respectively. The XGBoost model parameters are set with a learning rate of 0.05, a maximum tree depth of 8, and 100 iterations. The top three features in terms of feature importance are: catering activity index, promotional activity time and location, and e-commerce activity index. The model predicts an order trigger probability of 0.935 for catering packaging and 0.892 for e-commerce packaging. The standard deviation of the predicted probability is 0.03 for catering and 0.06 for e-commerce. Substituting these values ​​into the credibility calculation formula: Catering order credibility. =0.968, compared with the threshold of 0.90, meets the high confidence condition; e-commerce order credibility =0.933, which also meets the condition. The final output set of order types to be dispatched is catering takeaway packaging and e-commerce express packaging, with time windows positioned during the peak promotional period, for use by the specification matching template library of S2.3 to build and call, so as to achieve accurate coverage of order demand in the next 48 hours and generate the opportunity window for surplus material adaptation; S2.3: For the predicted set of order types to be shipped, call the product specification mapping table to obtain the corresponding standard width and standard length parameter ranges. At the same time, introduce a ±5% flexible tolerance mechanism to cover customized fine-tuning requirements, and generate a specification matching template library containing the target width range and target length range as the basic matching rules for subsequent surplus material adaptation judgment. S2.4: Based on the promised delivery time window and production preparation cycle of the predicted orders, calculate the earliest start time and latest completion time of various orders, and determine their effective execution time interval by combining the workshop capacity cycle data; encapsulate the time interval with the corresponding specification matching template to form a primary surplus material adaptation opportunity window unit with time constraints. S2.5: Assign priority weight coefficients to each primary surplus material adaptation opportunity window unit based on customer level weight, order profit margin, and urgency level label; sort all opportunity windows by time axis and remove duplicates and merge overlapping intervals to finally generate a dynamic surplus material adaptation opportunity window sequence with three-dimensional attributes of time, specification, and priority, which can be called by the cross-order scheduling decision engine.

[0014] like Figure 2 As shown, step S3 involves inputting the demand parameters of the current pending production orders into the cross-order scheduling decision engine, and simultaneously inputting the sequence of surplus material matching opportunity windows within the effective time range and the initial surplus material value map to initiate the multi-dimensional matching calculation process. Specifically, this includes: S3.1: Based on the production order data to be scheduled output by the customer order management system, obtain the set of demand parameters for the current order. The demand parameters include target width, target length, delivery time window, material grade requirements and production priority identifier, which serve as the benchmark input conditions for multi-dimensional matching calculation. S3.2: Retrieve the dynamic surplus material matching opportunity window sequence generated by S2 from the database, extract the effective opportunity windows whose time intervals cover the current production schedule, and generate a subset of opportunity windows that can be used for this matching based on the matching specification range and priority weight contained therein, as a guiding signal for potential surplus material usage scenarios; S3.3: Obtain the initial surplus material value map constructed by S1, and select available surplus material nodes with normal storage status, physical status score higher than the threshold and not locked to form the current schedulable surplus material resource pool. The nodes include structured attributes such as remaining length, width, material grade, location coordinates and comprehensive value index. S3.4: The demand parameters, effective opportunity window subsets, and schedulable surplus material resource pool of the orders to be scheduled are uniformly injected into the cross-order scheduling decision engine, data alignment and format standardization processing are performed, and multi-source heterogeneous input data packets under a unified spatiotemporal reference framework are generated to support the construction and matching calculation of the subsequent path graph model. The production order demand parameters obtained by step S3.1, the effective opportunity window subset formed by step S3.2, and the schedulable surplus material resource pool selected by step S3.3 are input into the cross-order scheduling decision engine. A multi-source heterogeneous data fusion method is adopted (setting data pattern mapping table parameters: field name, data type, unit system) to achieve standardized dictionary matching of structured information from different sources. Furthermore, by constructing a spatiotemporal joint index (parameters: timestamp precision at the millisecond level, spatial coordinates using a three-dimensional Cartesian system), the order demand, opportunity window, and surplus material node are indexed and bound under a unified spatiotemporal reference framework, and a unified reference time axis and a shared spatial coordinate system are obtained; Furthermore, field correction and unit normalization methods are adopted (parameters: length is uniformly in millimeters, weight is uniformly in grams, and price is uniformly in yuan / square meter) to achieve uniformity of numerical scale for each input data and generate a standardized attribute set to eliminate the impact of measurement differences from different data sources on subsequent calculations; Furthermore, missing value imputation and data consistency verification algorithms are applied (the imputation function uses multiple linear regression, and the verification rules include field type matching, numerical range validity, and time series continuity) to improve data packet integrity and generate a quality-controlled, missing-free, multi-source compatible data structure. Furthermore, by using data packet serialization and compression encoding methods (emphasizing JSON structure serialization format and LZ4 compression encoding), the spatiotemporal fusion data after the above unified processing is transformed into multi-source heterogeneous input data packets that can be directly called by the path modeling module of the scheduling decision engine; By using the above data alignment and format standardization methods, the scattered information from orders, opportunity windows and surplus material nodes is transformed into highly consistent structured data input under a unified spatiotemporal reference, thereby achieving the intensification of the input end of multidimensional matching calculation and the efficient connection of the calculation chain. For example, in a paper product packaging production workshop, the demand parameters for orders awaiting production scheduling include a target width of... millimeters, target length is Millimeters, delivery time window from - - to - - The material grade is A, and the production priority is P1; the effective opportunity window subset includes the matching specification range. to millimeter width and to The millimeter length range has the following priority weight: The remaining length of a certain surplus material node in the schedulable surplus material resource pool. millimeters, width millimeters, material grade A, storage location coordinates are ( , , ) meters, comprehensive value index is After inputting the three types of data, the width field is unified according to the field mapping table. Units, price fields are uniformly set to yuan / Unit; using a temporal-spatial joint index, the surplus material node is aligned with the lead time range of order demand and opportunity windows at the millisecond level, and mapped to a unified reference point in a three-dimensional coordinate system; a multiple linear regression interpolation method is used to supplement the missing profit margin field of the opportunity window. After data consistency verification, it was confirmed that the width field met the tolerance range of the matching template. Finally, the serialization formed a fusion data packet containing the order ID, opportunity window ID, surplus material ID and its multi-dimensional attributes. After compression, it was transmitted to the path modeling module of the decision engine. It was verified that the data parsing latency was significantly shortened in the subsequent path search steps, which supported the rapid identification and execution of the optimal path for surplus material. S3.5: Based on the event triggering mechanism of the cross-order scheduling decision engine, the integrity and consistency of the input data packet are detected. When all necessary fields meet the preset verification rules, the start flag of the multi-dimensional matching calculation process is activated, and the path modeling initialization command is sent to the downstream module to realize the stage transition from static data input to dynamic optimization calculation. Based on the event triggering mechanism of the cross-order scheduling decision engine, a field integrity verification method (parameters: list of required fields, allowable missing rate threshold of 0) is adopted to realize the field existence verification and full comparison of the input data packet; Furthermore, through a consistency verification algorithm (parameters: data type mapping table, cross-source field correspondence matrix, timestamp synchronization tolerance of 5ms), the consistency verification of cross-field values ​​of the same business entity in multi-source heterogeneous input data packets is realized, and a set of consistency verification result flag bits is obtained; Furthermore, a verification rule matching method (parameters: production constraint rule set, specification matching rule set, economic pre-screening rule set) is adopted to calculate the complete matching degree of each necessary field value with the preset rules and generate a rule matching degree index vector. Furthermore, through a logical judgment algorithm (parameters: set of verification result flags, vector of rule matching degree index, vector of trigger threshold), the Boolean comprehensive judgment of multidimensional verification results is realized, and an event trigger Boolean signal stream is generated to determine whether the start conditions of the multidimensional matching calculation process are met. Furthermore, a path modeling initialization instruction generation method (parameters: start flag, unified spatiotemporal reference frame configuration file, node attribute mapping table) is adopted to transform the event-triggered Boolean signal stream into a single start flag and bind the global reference information required for path modeling, thereby generating an initialization instruction package that can be directly executed by downstream modules; Through the above chain processing method, the multi-source heterogeneous input data packets under the unified spatiotemporal reference framework generated in the previous step are transformed into structured startup instructions that meet the requirements of field integrity, value consistency and rule matching, so as to realize the stage transition from static data input to dynamic optimization calculation. For example, in a paper product packaging production workshop, the cross-order scheduling decision engine receives an input data packet containing the following parameters: the demand parameters for orders to be scheduled (target width 1450mm, target length 2300mm, delivery time window from 08:00 on 2024-07-10 to 20:00 on 2024-07-11, material grade requirement B, production priority identifier 7), a subset of effective opportunity windows (covering the time interval from 2024-07-10 to 2024-07-11, matching specification range width 1440-1455mm, length 2280-2320mm, priority weight 0.85), and a schedulable surplus material resource pool (surplus material roll ID: R2024070603, remaining length 2400mm, width 1450mm, material grade B, location coordinates X=12.3, Y=4.7, comprehensive value index 3.25). The field integrity verification method checks the required field list to determine a missing rate of 0. A consistency check algorithm verifies that the width, length, and material grade are consistent across the three data sources, with a timestamp synchronization difference of 3ms, below the tolerance threshold. The rule matching method calculates a specification matching degree of 1.0, a production constraint matching degree of 1.0, and an economic pre-screening matching degree of 0.92. The logical judgment algorithm calculates a Boolean comprehensive judgment of True based on the trigger threshold vectors (0.95, 0.90, and 0.90 respectively) and generates an event trigger signal stream. The path modeling initialization instruction generation method encapsulates the start flag, the unified spatiotemporal reference frame configuration file (containing the position coordinate transformation matrix and material grade compatibility table), and the node attribute mapping table into an initialization instruction package, which is sent to the path modeling module. In this embodiment, the event triggering mechanism ensures that the path modeling stage only begins when all verification conditions are met and the multi-dimensional matching process startup cost is reasonable, effectively guaranteeing the accuracy and stability of subsequent waste material reuse path construction.

[0015] like Figure 3 As shown, step S4 involves constructing a constrained waste material reuse path graph model in the cross-order scheduling decision engine, based on each waste material node in the initial waste material value graph and the target demand node in the opportunity window sequence. The cost function of each edge is a weighted synthesis of waste material value loss rate, scheduling distance cost, mold change preparation time, and disturbance to the mainline scheduling. Specifically, this includes: S4.1: Based on the comprehensive value index and physical characteristic data of each roll of scrap material in the initial scrap material value map generated in S1, obtain the attribute set of each scrap material node. The attribute set includes the remaining length, width, material grade, storage location coordinates and physical state score. The above attributes are used as the basic feature input of the scrap material nodes in the graph model to construct a scrap material resource representation system with quantifiable evaluation dimensions. Based on the comprehensive value index and physical characteristic data of each roll of scrap material contained in the initial scrap material value map formed by step S1, the node attribute parsing method (parameter source: comprehensive value index, remaining length, width, material grade, storage location coordinates, physical state score) is adopted to realize the multi-dimensional attribute deconstruction of individual scrap material nodes and generate a node basic feature set that can be called by the path graph model. Furthermore, by using an attribute standardization algorithm (parameter: minimum-maximum normalization interval set to 0 to 1), the remaining length, width and comprehensive value index of each surplus material node are unbiasedly scaled under a unified dimension, and a structured, dimensionally consistent node numerical feature vector is obtained to ensure the computability and consistency of subsequent multi-node comparisons. Furthermore, a feature encoding strategy is adopted (parameters: material grade uses integer mapping encoding, storage location coordinates maintain native two-dimensional numerical encoding, and physical state score uses continuous floating-point encoding) to achieve unified transformation of non-numerical and hybrid attributes, and generate a complete node encoding matrix that can directly participate in mathematical operations, providing symbolic input data for path graph modeling; Furthermore, by constructing a node evaluation function, the normalized length and width are mapped to area estimation components, which are then multiplied by the comprehensive value index to form a node economic weight indicator. The formula is as follows:

[0016] in, This represents the remaining length after normalization. This represents the normalized width. This represents the comprehensive value index, through which the economic dimension is quantitatively embedded in the single-node attribute; By integrating node features, the numerical feature vectors, coding matrices, and economic weight indicators are jointly bound to the scrap material node identifier, forming a scrap material resource representation system with quantifiable evaluation dimensions. This enables the graph model to have computability, comparability, and multi-dimensional scalability at the node calling level. For example, in the roll paper slitting workshop of a paper packaging factory, the remaining length of scrap node A is 1250mm, the width is 850mm, the material grade is high-strength kraft paper (coded as 3), the storage location coordinates are (12.5, 8.3), the physical condition score is 0.92, and the comprehensive value index is 38.6. Using the minimum-maximum normalization algorithm, within the length range of 1000-1500mm and the width range of 800-1000mm, the normalized length is 0.5 and the normalized width is 0.25, respectively; the material grade code is directly assigned the value of 3, the storage location retains the coordinate value, and the physical condition score code is 0.92. The economic weight index formula is then used to calculate:

[0017] The node's economic weight is 4.825. This weight, along with other normalized attributes of the node, constitutes the complete feature vector of the waste material node A and is integrated into the graph model. In the subsequent path cost calculation process, this weight directly participates in the edge weighting, enabling the model to significantly improve its ability to distinguish different waste material saving potentials in the economic dimension. S4.2: Based on the dynamic surplus material adaptation opportunity window sequence generated in S2, extract the target width and length requirement range corresponding to each time interval, and combine it with customer priority weight to generate a target requirement node set; based on the specification matching relationship between this set and the surplus material nodes, establish potential call connection relationship to form an initial surplus material-demand association edge set, which serves as the topological basis for feasible scheduling paths in the path graph model. S4.3: For each surplus material-demand association edge, calculate its multi-dimensional cost components during the scheduling process: Use the warehouse management system to obtain the scheduling distance from the current storage location of the surplus material to the target production equipment, and combine the energy consumption and time cost per unit distance of AGV transportation to generate the scheduling distance cost; Based on the historical preparation time records of switching between the same material and different widths in the equipment mold change database, estimate the mold change preparation time required for this call through a linear interpolation algorithm, as a proxy indicator for production interruption cost; At the same time, based on the comprehensive value index calculated in S1 and the time window during which the target order allows the use of surplus materials, use a decay function model to calculate the surplus material value loss rate; Based on the attribute set and scheduling execution conditions of the surplus material-demand association edge, the path calculation module of the warehouse management system (input parameters: coordinates of the current storage location of the surplus material and coordinates of the target production equipment station) is used to realize the function of calculating the spatial distance from the surplus material location to the target equipment. This distance is then combined with the unit distance energy consumption and standard running time parameters of the automated guided vehicle (AGV) to obtain the scheduling distance cost component. Furthermore, through the equipment mold change database query interface (parameters: material category, width change range, historical mold change record time series), the historical preparation time interpolation calculation function is realized for the same material but inconsistent width. Based on the linear interpolation algorithm, the expected mold change preparation time index for this call is generated as a proxy quantification result of production interruption cost. Furthermore, through the module that matches the surplus material comprehensive value index with the time window boundary of the target order allowing the use of surplus materials (parameters: surplus material comprehensive value index, start and end time of the order's allowed use time window), the decay function model is called to perform time decay calculation on the value index to obtain the surplus material value loss rate index; the decay function is defined using the following MathML formula:

[0018] in, This represents the loss rate of surplus material value. This is the initial comprehensive value index for surplus materials. The decay coefficient is obtained by fitting historical aging data. This refers to the span between the storage of surplus materials and their recall time. Furthermore, through the multi-dimensional cost component integration module, the scheduling distance cost, mold change preparation time index and residual material value loss rate are encapsulated in a unified format to form a complete multi-dimensional cost description data structure, providing input for the subsequent construction of the weighted cost function; Through the above algorithm processing method, the surplus material-demand association edge attribute in the previous step is transformed into a multi-dimensional cost component containing three types of quantitative indicators: space, time, and economy, so as to achieve a comprehensive quantitative representation of the execution cost of the call path. For example, in a paper roll slitting scenario at a paper packaging factory, a waste material node is stored in storage area A with coordinates (12.5, 34.8). The target production machine's coordinates are (58.2, 102.4). The AGV's travel distance output by the storage management system's path calculation module is 100.5 meters, with an energy consumption of 0.02 kWh per unit distance and a running time of 3.5 seconds per unit distance. Therefore, the scheduling distance cost is calculated as the sum of energy consumption cost and time-converted cost, resulting in a cost value of 0.002 kWh and a time cost of 352 seconds. Historical data from the equipment mold-changing database shows that the average mold-changing time for the same material but with a width difference of 5 cm is 240 seconds, and the mold-changing time for a width difference of 10 cm is 360 seconds. Using a linear interpolation algorithm, the estimated mold-changing time for the current width difference of 7 cm is 300 seconds. The comprehensive value index of surplus materials is 350 yuan, the storage time of surplus materials Δt is 4 hours, and the historical fitting attenuation coefficient λ is 0.05. Then the value loss rate R is calculated using the formula: The calculated cost is approximately 283.56 yuan. All three types of data indicators are encapsulated within the multi-dimensional cost data structure of the surplus material-demand association edge, providing accurate input for subsequent weighted cost modeling and path optimization, significantly improving the credibility and feasibility of the economic assessment. S4.4: Based on four parameters—scheduling distance cost, mold change preparation time, residual material value loss rate, and disturbance to the mainline scheduling—the mainline scheduling disturbance is quantified by comparing the difference between the idle time between adjacent orders in the current master production plan and the expected call duration, and a weighted cost function is constructed. The normalized components are multiplied by their preset strategy weights and summed to generate a comprehensive cost index for each edge, thereby achieving a fusion representation of the dual impact of path economy and production stability. Based on the set of input parameters including scheduling distance cost, mold change preparation time, residual material value loss rate, and mainline scheduling disturbance, a multi-dimensional parameter normalization method (parameter: standardization of the maximum-minimum value interval) is adopted to achieve comparability processing of cost components from different sources and with different dimensions within the same numerical range. Furthermore, by using a disturbance quantification method (parameters: adjacent order idle time period, expected call duration), the mainline scheduling disturbance degree is quantitatively calculated, and the numerical representation of the disturbance degree components is obtained; Furthermore, a cost component weight configuration mechanism (parameter: policy preset weight matrix) is adopted to realize the linear combination operation of each normalized cost component and its corresponding policy weight, and to generate the input term of the comprehensive cost function; Furthermore, a weighted cost function is constructed, and the multi-component cost synthesis is achieved using the following mathematical expression:

[0019] in, As a comprehensive cost indicator, Let i be the policy weight of the i-th cost component. The value of the i-th term after normalization; Furthermore, the comprehensive cost index of each edge is output through weighted summation and mapped to the edge attribute field of the corresponding surplus material-demand node pair, thereby realizing the integrated representation of the dual impact of path economy and production stability. By using a cost function weighting method, the multi-dimensional component results of the previous step are transformed into a single comprehensive cost index, thereby achieving a joint evaluation of the surplus material call path in terms of economic and scheduling stability. For example, in the waste material reuse optimization scenario of a paper product packaging company, the scheduling distance cost is 8.5 (unit: energy consumption time weighted value), the mold change preparation time is 450 seconds, the waste material value loss rate is 0.12, and the main line scheduling disturbance is calculated as the difference between the adjacent order idle period of 900 seconds and the expected call occupation time of 700 seconds, i.e., 200 seconds. Normalization is performed using a maximum value of 10 and a minimum value of 0, resulting in a normalized scheduling distance cost of 0.85, a normalized mold change preparation time value of 0.5, a normalized waste material value loss rate value of 0.12, and a normalized disturbance value of 0.2. The strategy weight matrices are set to 0.4, 0.3, 0.2, and 0.1 respectively. Combined with the weighted cost formula:

[0020] The comprehensive cost index was calculated. This value is stored in the corresponding edge attribute of the graph model and is subsequently used in the Dijkstra path search model to select the cost-optimal path, achieving a dual optimization effect of significantly improving material utilization and significantly reducing production disturbance. S4.5: Integrate the edge set with comprehensive cost index, along with surplus material nodes and demand nodes with complete attributes, into the graph database to construct a constrained directed weighted path graph model. Nodes represent schedulable surplus materials or unmet order demands, while edges represent feasible cross-order call paths and carry cost information synthesized from multi-dimensional constraints. This provides structured input for subsequent improvements to the Dijkstra algorithm to perform optimal path search.

[0021] Step S5: Using an improved Dijkstra algorithm, the optimal set of paths satisfying production constraints is searched in the waste material reuse path graph model to obtain multiple feasible waste material utilization path schemes. Each scheme is accompanied by a total cost index and a resource saving potential assessment value. Specifically, this includes: S5.1: A constrained path graph model for the reuse of surplus materials, constructed based on the preceding steps, is provided. The nodes include all surplus material nodes in the initial surplus material value graph and the target demand nodes in the dynamic surplus material adaptation opportunity window. The edges are weighted by a multidimensional cost function, whose inputs include surplus material value loss rate, scheduling distance cost, mold change preparation time, and disturbance to the mainline scheduling. A graph structure modeling method is used to map these elements into a directed weighted graph to express the possibility and comprehensive cost of different surplus materials flowing to specific orders, generating a path topology structure with physical feasibility. Based on the constrained waste material reuse path graph model constructed in the previous steps, the complete attribute set of waste material nodes and demand nodes is collected. The comprehensive value index, remaining length, width, material grade, storage location coordinates and physical status score are used as node description vectors and input into the path modeling module to ensure that the nodes of the graph model have computable characteristics in both the logical layer and the data layer. A multi-dimensional cost function mapping method is adopted (parameters: surplus material value loss rate, scheduling distance cost, mold change preparation time, main line scheduling disturbance degree). Each surplus material-demand matching relationship is weighted and synthesized at the edge level. Normalization processing and weight parameter configuration are used to make each cost component have a unified dimension, so as to construct a comprehensive cost index as the edge weight value. Furthermore, through the directed weighted graph structure modeling method (parameters: node type label, edge weight matrix, constraint set), directional connections are established between surplus material nodes and demand nodes according to the matching relationship. The direction is from the surplus material node to the demand node, and the edge weight is taken from the output of the multi-dimensional cost function, so as to realize the joint expression of the possibility and comprehensive cost of different surplus materials flowing to specific orders. Furthermore, a topology verification algorithm is used to check the connectivity of the directed weighted graph, identify isolated nodes and invalid edges, and mark their status in the node or edge attribute table to prevent subsequent algorithms from performing redundant calculations on unreachable paths and ensure the effectiveness of path search. By storing the above directed weighted topology in a graph database, the node attributes, edge weights, and constraint sets are indexed to form a path graph that is physically feasible and computationally feasible. This provides a standardized input data structure for the improved Dijkstra algorithm, achieving the expected technical effects of subsequent optimal path search and resource saving potential assessment. For example, on three production lines of a paper packaging factory, the number of surplus material nodes is configured as 100, and the number of demand nodes is configured as 40. The node vectors include a comprehensive value index ranging from 20 to 120, a remaining length ranging from 200 to 1200 mm, a width ranging from 300 to 800 mm, five material grades, and a physical condition score ranging from 0.6 to 0.95. The surplus material value loss rate in the multidimensional cost function is determined by a decay model. (in The aging coefficient, The scheduling distance cost is calculated based on the storage time and the AGV transportation unit distance cost. The unit price per meter and real-time distance measurement are linearly multiplied. The mold change preparation time is obtained from the historical dataset using a linear interpolation algorithm. The mainline scheduling disturbance is calculated by normalizing the difference between the master production plan and the time occupied by the calling path. After graph structure modeling, connectivity is checked, and 15 invalid edges are removed, ultimately forming 85 directed connections from feasible surplus materials to demand nodes. After being loaded into the graph database, the improved Dijkstra algorithm calls this topology structure to achieve initialization during resource-saving path search. Verification results show that this structured input significantly improves the algorithm's initialization time and reduces the dimensionality of the subsequent search space, greatly improving computational efficiency and path evaluation accuracy. S5.2: Perform node reachability analysis on the generated directed weighted path topology, and perform pre-pruning based on production constraints, including width matching threshold, remaining length satisfaction, material grade compatibility rules, and storage location reachability restrictions; use Boolean logic judgment mechanism to filter out candidate edge sets that meet basic process requirements, and eliminate infeasible transfer paths to reduce the search space dimension and improve the efficiency of subsequent path search. S5.3: On the path topology optimized by pre-pruning, an improved Dijkstra algorithm engine is deployed for iterative calculation. Its priority queue is initialized starting from the target demand node corresponding to the current order to be scheduled for production, and the adjacent nodes are expanded layer by layer. During the algorithm iteration process, a dynamic relaxation mechanism is introduced to update the cumulative cost of each path. The cumulative cost is the sum of the output values ​​of the cost function of each edge, forming a path-level total cost index, which is used to quantify the overall implementation cost of different calling paths. On the path topology optimized by pre-pruning, an improved Dijkstra algorithm engine (parameters: comprehensive cost weight coefficient, priority queue capacity, relaxation iteration step size) is used to realize the priority queue initialization function starting from the target demand node corresponding to the current order to be scheduled. Furthermore, by using the node attribute parsing method (parameters: node type label, spatial location coordinates, remaining length, width and material grade), the adjacent node set is expanded layer by layer, and the cost accumulation calculation is performed on each expansion path to obtain the path cost data based on the accumulation of the edge cost function. Furthermore, a dynamic relaxation mechanism (parameters: relaxation decision threshold δ, cost update function f) is employed to achieve real-time updates of the cumulative cost for each path, where the cumulative cost formula is:

[0022] in Output the cost function for all edges on path P. The cumulative result, where n is the number of path edges, is used to quantify the overall implementation cost of different calling paths; Furthermore, by using a cost comparison and queue reordering algorithm (parameters: comprehensive cost index, initial value of resource saving potential), the priority of path expansion is dynamically adjusted, and the upper and lower bounds of the cost of the current optimal path are maintained during the search process to reduce redundant expansion calculations. By using the improved Dijkstra algorithm calculation process, the candidate edge set and node attributes from the previous step are transformed into a path-level total cost index, thereby achieving a quantitative representation of the implementation cost of different calling paths. For example, in a topology containing 6 surplus material nodes and 4 demand nodes, the initial priority queue starts with demand node D1, and the cost weighting coefficients are set to scheduling distance cost 0.4, mold change preparation time 0.3, surplus material value loss rate 0.2, and scheduling disturbance degree 0.1. Node attributes include surplus material nodes R1 (position coordinates (12,8), remaining length 2000mm, width 800mm, material grade A) and R2 (position coordinates (5,3), remaining length 1500mm, width 750mm, material grade B), etc. The specification matching range of demand nodes is set to width 760~820mm and length ≥1500mm. During algorithm execution, whenever an edge is expanded to a new adjacent node, the cost function is called to calculate the cost of a single edge. For example, the cost from R1 to D1 is: scheduling distance cost 8 × 0.4 = 3.2, mode change preparation time 10 min × 0.3 = 3, value loss rate 0.05 × 0.2 = 0.01, scheduling disturbance degree 0.02 × 0.1 = 0.002, which are normalized to a total edge cost of 6.202. The cumulative cost is calculated using the formula... The calculation is performed, where n is the number of path edges. When the current path P contains two edges, R1-D1 and D1-D2, the cumulative cost is updated to (6.202+5.876) / 2=6.039. A dynamic relaxation mechanism determines whether to update the path cost based on δ=0.05. If the current cumulative cost decreases by more than the δ threshold compared to the old value, an update is performed and the queue is reordered. Finally, the two candidate paths with the lowest total cost are obtained and associated with R1 and R2 respectively, for subsequent resource saving potential assessment. S5.4: During the path expansion process, the resource saving potential assessment value corresponding to each feasible path is calculated simultaneously. This value is determined based on the difference between the comprehensive value index of the activated surplus material and the procurement cost of its substitute main material, and is converted into equivalent economic benefits in combination with the mold change time savings. The total cost index and resource saving potential assessment value of each complete path are used as dual-objective output parameters and encapsulated into structured path scheme units to form a preliminary set of feasible path candidates. S5.5: Perform non-dominated sorting filtering on the generated feasible path candidate set, identify Pareto front path subsets based on two dimensions: minimizing total cost index and maximizing resource saving potential; retain the optimal path set that satisfies multi-objective optimization balance, and attach standardized label information, including path number, related scrap ID sequence, associated opportunity window identifier, and key performance index field, for subsequent Shapley value allocation mechanism to call.

[0023] Step S6: Based on the Shapley value allocation mechanism in game theory, the marginal contribution of each feasible waste material utilization path is calculated to obtain a comprehensive benefit score for each path in improving overall resource efficiency. Based on this score, the optimal reuse path with the highest comprehensive benefit is selected from the candidate set. Specifically, this includes: S6.1: Based on the multiple feasible waste material utilization path schemes and their corresponding total cost indicators and resource saving potential assessment values ​​output from the previous steps, a path collaborative contribution analysis model is constructed. Each path acts as a game participant, and its inputs are the waste material resources consumed by the path, the raw material cost saved, the scheduling disturbance introduced, and the mold change time overhead. Using the cooperative game set defined under the game theory framework, all feasible paths are included in a unified utility evaluation space to generate a path alliance utility matrix, which serves as the basic input for Shapley value calculation, in order to support the subsequent quantitative decomposition of the marginal contribution of each path. Under the conditions of feasible surplus material utilization path schemes and their dual-objective evaluation data input, the collaborative contribution analysis method (parameters: set of structured path scheme units, total cost index, set of resource saving potential evaluation values) is adopted to realize the cooperative modeling of each path; Furthermore, by using a feature reduction method (parameters: waste material consumption, saved raw material cost, scheduling disturbance, and mold changing time overhead), the path input features are mapped to utility space coordinates, and the consistent dimensional data required for matrix construction is obtained. Furthermore, a cooperative game set construction method (parameters: N of all feasible path numbers, input feature vectors of each path) is used to realize a formal mapping from the path set to the game participant set, and to generate a participant index structure; Furthermore, by using the alliance utility function modeling algorithm (parameters: set of path participants, utility function v(S) definition rules), the overall system utility of all possible path combinations is calculated, and the calculation results are filled into matrix cells according to the combination relationship to generate a path alliance utility matrix with row and column symmetry. By constructing and processing the path collaboration contribution analysis model and the alliance utility matrix, the candidate path set from the previous step is transformed into structured game data that can be used for marginal contribution calculation, thus providing the basic input for subsequent Shapley value calculation. For example, in a batch of production data, the feasible path solution unit set contains 5 paths, where the residual material resource consumption of each path ranges from 0.8 to 1.5 unit area, the saved raw material cost ranges from 120 to 220 yuan, the scheduling disturbance is between 0.05 and 0.15, and the mold changeover time is between 3 and 8 minutes. The game participants are represented by path numbers P1 to P5, and the weighted utility function v(S) is defined as the weighted sum of resource saving benefits and scheduling stability, as shown in the following formula: in, The total resource savings benefit for path combination S, As the cost normalization factor, Let S be the overall scheduling disturbance degree of combination S. The utility value of combination P1, P2, and P3 is calculated as follows: The utility value of the combination P2 and P4 is calculated as follows: This process is repeated to fill the coalition utility matrix, ensuring its symmetry is maintained during row and column swaps. The resulting matrix can then be directly used for subsequent subset enumeration and Shapley value marginal contribution calculations, achieving a quantitative decomposition of the overall benefits of each path. S6.2: Based on the path alliance utility matrix, perform subset enumeration operation, traverse all possible path combination subsets, and for each subset S containing the target path i, calculate its overall system utility difference with and without path i. This difference reflects the marginal gain of path i in a specific collaborative environment. Use a weighted permutation method to calculate the average incremental utility brought by path i in all possible sequences. The weighting factors are dynamically adjusted according to the order urgency, production line load balancing index and surplus material aging risk coefficient to ensure that the marginal gain calculation fits the actual production constraints. S6.3: Based on the above average incremental utility data, apply the Shapley value formula for mathematical modeling: For each feasible path i, calculate its expected marginal contribution in all path permutations and combinations, i.e., the Shapley value; through this calculation, assign a unique comprehensive benefit score to each path, reflecting its fair value allocation result under the global optimization objective; S6.4: The Shapley value corresponding to each path is used as the core evaluation index. Combined with the preset normalization processing algorithm, it is converted into a standardized comprehensive benefit score in the range of 0 to 1 to eliminate the influence of different dimensions and orders of magnitude on the decision. At the same time, a priority correction factor is introduced to dynamically adjust the score according to the current master schedule stability requirements and inventory turnover pressure index, and generate a final comparable path optimization ranking sequence as the direct basis for selecting the optimal path. S6.5: Based on the sorted path optimization sequence, select the path with the highest comprehensive benefit score as the optimal reuse path, and output the surplus material call instruction package associated with the path, including the selected surplus material number, scheduling start point and target workstation, expected execution window and expected net resource benefit value; if there are ties for the highest score path, further compare the rate of change of the second derivative of their disturbance to the main line scheduling, select the one with the smallest disturbance to ensure the continuity of production cycle, and complete the deterministic convergence from the multi-path candidate set to the unique execution scheme.

[0024] Step S7: Determine whether the expected cost savings from surplus materials corresponding to the optimal reuse path exceed 1.8 times its total scheduling cost. If this trigger threshold condition is met, activate the cross-order surplus material call process and generate a preliminary production scheduling instruction containing the main material usage plan and surplus material call path suggestions. Specifically, this includes: S7.1: Based on the optimal reuse path with the highest comprehensive benefits output by S6, obtain the expected cost savings of the corresponding scrap materials. The expected cost savings of the scrap materials is calculated based on the difference between the comprehensive value index of the scrap material rolls reused in the path and the equivalent raw material substitution cost, so as to quantify the potential benefits of the call scheme in terms of material savings. S7.2: Perform cost modeling for each scheduling link involved in the optimal reuse path, calculate the scheduling distance cost based on the spatial coordinates of the surplus material storage location and the target machine, generate the mold change time cost by combining the equipment mold change preparation time standard database, and convert it into production delay cost according to the main line scheduling disturbance evaluation results. Then, the total scheduling cost of the path is obtained by weighted summation, which serves as the benchmark input for economic decision-making. S7.3: Construct an economic ratio index based on the expected surplus material saving cost and the total scheduling cost, calculate the cost-benefit ratio between the two, and compare the ratio with the preset adaptive trigger threshold of 1.8 to determine whether the path is economically feasible, and ensure that the subsequent activation process is only entered when the resource saving potential significantly exceeds the execution cost. S7.4: If the cost-benefit ratio is greater than or equal to 1.8, the triggering condition is met. The system generates a cross-order surplus material call permission signal and starts the call process initialization program, marking the relevant surplus material roll as 'pending scheduling status' to prevent other production tasks from concurrently calling and causing resource conflicts. S7.5: Generate preliminary production scheduling instructions based on the call permission signal, including the main material usage plan, the surplus material call path suggestion and the execution sequence of each node, and embed risk warning indicators. The indicators are dynamically generated based on the mold change frequency change rate and the main line cycle offset, and are used to remind the scheduler to pay attention to the production fluctuations that may be caused, and complete the transformation from optimization decision to executable instructions. Based on the input of the permission signal to the production scheduling instruction generation module, the full attribute data of the optimal reuse path output by S6 and S7.4 is loaded, including the sequence of surplus material numbers, the coordinates of the scheduling start point and the target workstation, the expected execution window and the net resource benefit index, and the main material usage plan of the current order is bound as the core basis for instruction compilation; A path instruction orchestration algorithm (parameters: path node sequence, workstation process constraint matrix, production cycle configuration table) is adopted to perform sequential scheduling decomposition of each surplus material call link involved in the optimal path and generate the corresponding node execution sequence. Furthermore, by using the workstation process mapping method (parameters: material grade, width range, equipment capacity table), the execution sequence of nodes and the main material usage plan are arranged in parallel, ensuring the scheduling correlation of main materials and surplus materials in the instruction package on the time axis. Furthermore, through a dynamic calculation method for risk indicators (parameters: historical data on mold change frequency, mainline beat record), the real-time calculation of the mold change frequency change rate and the mainline beat offset is achieved, and risk warning labels are generated, among which the mold change frequency change rate... This indicates the proportion of newly added mold changes within the call path to the originally planned number of mold changes, and the main timeline offset. This indicates the proportion of the original production cycle time that the call-in time represents; Furthermore, through a tag injection mechanism, risk warning indicators are embedded into the metadata fields of the production scheduling instruction package to provide advance warnings to schedulers during the execution phase. The above chain processing integrates the optimal reuse path with the main material usage plan, node execution sequence and risk warning labels into a structured preliminary production scheduling instruction package, realizing the transformation from optimization decision to executable instructions; For example, in a production order, the main material usage plan is a 1200mm wide and 2000m long roll of main body material. The optimal waste material reuse path involves calling two rolls of waste material numbered R001 and R005, located in the east area of ​​warehouse A and the west area of ​​warehouse B, respectively, with corresponding target workstation coordinates of (12,5) and (3,8). The node execution time windows are 08:00-09:15 and 09:20-10:00, respectively, and the main material scheduling time is 08:00-12:00. The original plan for the mold change frequency was 5 times. The calling path introduces 1 additional mold change, so the mold change frequency change rate is calculated as follows: The original main storyline beat was 240 minutes. An additional 12 minutes was introduced to extend it, resulting in a main storyline beat offset of [value missing]. The risk warning label is marked as "medium risk" in the metadata. The instruction package output includes a main material usage plan table, a surplus material call path table, the start and end time of each execution node, and a risk warning list. The task order is generated by calling the MES system. During the execution process, the scheduler reserves equipment idle time in advance to buffer possible cycle disturbances, ultimately achieving the dual goals of efficient material utilization and stable production cycle.

[0025] Step S8: By using locally deployed lightweight edge computing nodes to update the status of remaining materials and perform preliminary path filtering, the global strategy iteratively optimized by the cloud center is synchronized to the edge, ensuring that production scheduling instructions are output to the production scheduling system with risk warning indicators, thus completing closed-loop control. Specifically, this includes: S8.1: Based on locally deployed lightweight edge computing nodes, real-time status data of surplus materials from various storage areas is obtained. The real-time status data of surplus materials includes storage location coordinates, physical state score changes, environmental temperature and humidity influencing factors, and entry and exit operation timestamps. The above data is filtered using an incremental state observer to eliminate sensor noise interference and generate a high-confidence dynamic surplus material status vector, which serves as the input basis for subsequent path selection. S8.2: Based on the generated dynamic surplus material state vector, combined with the latest global optimization strategy model parameters synchronously distributed from the cloud center, perform local surplus material adaptation capability assessment, and use a fast matching algorithm based on the rule engine to perform preliminary path feasibility screening on surplus material nodes within the current effective time window, eliminate invalid call paths caused by physical state degradation or unreachable location, output candidate surplus material call subset, and reduce uplink communication load and cloud decision-making burden; S8.3: Perform scheduling conflict prediction analysis on the output candidate surplus material call subset. Based on the production line mold change preparation time matrix, the current main line scheduling Gantt image segment, and the availability status of material handling equipment, construct a local resource competition detection model, calculate the production line disturbance increment index that may be caused by each candidate path during execution, and embed this index as a risk weighting factor into the local path sorting function to generate a preliminary call sequence with disturbance priority label; S8.4: While uploading the initial call sequence with the perturbation priority label to the cloud center for global benefit re-verification, receive the integrated risk warning label from the cloud. The risk warning label is generated based on the joint determination of the marginal contribution volatility of the Shapley value, the multi-order coupling strength index and the opportunity window decay slope. The label is used to dynamically mark the local initial production scheduling instructions to form an enhanced production scheduling instruction package that includes the main material usage plan, the suggestion of the surplus material call path and multi-level risk warnings. S8.5: The enhanced production scheduling instruction package is output to the MES production scheduling system through the industrial standard communication protocol to trigger the production scheduling task creation process. At the same time, the instruction execution feedback interface is stored locally on the edge node to collect data on scheduling confirmation status, material outbound action response delay and actual mold change time in real time. This constitutes the execution feedback sample in the closed-loop control loop, which is used to drive the reinforcement learning update of the cloud strategy model in the next cycle.

[0026] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0027] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and rules of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for optimizing the slitting allowance and scheduling of paper product packaging rolls, characterized in that, Includes the following steps: S1: After each roll of paper is slit, collect the physical characteristics data of each scrap material, and combine the real-time market raw material price with the mapping relationship between alternative use scenarios to calculate the comprehensive value index of each roll of scrap material and form an initial scrap material value map. S2: Based on historical order patterns, customer delivery cycles, and product specification distribution data, construct a short-term order demand forecasting model, output multiple order types that may be triggered in the next 48 hours and their target width and length ranges, and generate a sequence of surplus material matching opportunity windows accordingly. S3: Input the demand parameters of the current orders to be scheduled into the cross-order scheduling decision engine, and simultaneously input the sequence of surplus material matching opportunity windows within the effective time range and the initial surplus material value map to start the multi-dimensional matching calculation process; S4: In the cross-order scheduling decision engine, a path graph model for the reuse of surplus materials is constructed based on each surplus material node in the initial surplus material value graph and the target demand node of the surplus material adaptation opportunity window sequence. S5: The improved Dijkstra algorithm is used to search for the optimal set of paths that meet the production constraints in the scrap material reuse path graph model, and obtain multiple feasible scrap material call path schemes. S6: Based on the Shapley value allocation mechanism in game theory, the marginal contribution of each feasible waste material utilization path is calculated to obtain the comprehensive benefit score of each path to improve the overall resource efficiency, and the optimal reuse path with the highest comprehensive benefit is selected from the candidate set according to the comprehensive benefit score.

2. The method for optimizing and scheduling the slitting allowance of paper product packaging rolls according to claim 1, characterized in that, Following step S6, the following is also included: S7: Determine whether the expected cost savings of the surplus material corresponding to the optimal reuse path exceeds 1.8 times its total scheduling cost. If so, activate the cross-order surplus material call process and generate a preliminary production scheduling instruction containing the main material usage plan and the surplus material call path suggestion. S8: By performing residual material status updates and preliminary path filtering on locally deployed edge computing nodes, the global strategy iteratively optimized by the cloud center is synchronized to the edge.

3. The method for optimizing and scheduling the slitting allowance of paper product packaging rolls according to claim 1, characterized in that, The physical characteristic data includes remaining length, width, material grade, storage location, and physical condition score.

4. The method for optimizing and scheduling the slitting allowance of paper product packaging rolls according to claim 1, characterized in that, The construction of the short-term order demand forecasting model includes: The industry category, customer role tag, product specifications, order time, promised delivery cycle and actual completion time of each order are extracted from the historical order database. The trend, seasonality and residual items are separated by time series decomposition algorithm to generate an order behavior pattern feature set. Combined with the market activity index of the customer's industry segment and the promotion calendar, the XGBoost ensemble learning model is used to predict the order trigger probability of each scenario in the next 48 hours, and the types of pending orders with a confidence level of greater than or equal to 0.9 are selected.

5. The method for optimizing and scheduling the slitting allowance of paper product packaging rolls according to claim 1, characterized in that, Step S3 specifically includes: Based on the production order data to be scheduled output by the customer order management system, obtain the set of demand parameters for the current order as the benchmark input conditions for multi-dimensional matching calculation; The database is retrieved from the dynamic surplus material matching opportunity window sequence generated in step S2. The effective opportunity windows whose time intervals cover the current production schedule are extracted. Based on the matching specification range and priority weight contained therein, a subset of effective opportunity windows that can be used for this matching is generated. Obtain the initial surplus material value map constructed in step S1, and select available surplus material nodes with normal storage status, physical status score higher than the threshold and not locked to form the current schedulable surplus material resource pool. The set of demand parameters, the subset of effective opportunity windows, and the schedulable surplus material resource pool are uniformly injected into the cross-order scheduling decision engine, and data alignment and format standardization processing are performed to generate multi-source heterogeneous input data packets. Based on the event triggering mechanism of the cross-order scheduling decision engine, the integrity and consistency of the multi-source heterogeneous input data packets are detected. When all necessary fields meet the preset verification rules, the start flag of the multi-dimensional matching calculation process is activated, and a path modeling initialization command is sent to the downstream module.

6. The method for optimizing and scheduling the slitting allowance of paper product packaging rolls according to claim 5, characterized in that, The cross-order scheduling decision engine performs heterogeneous data fusion on the input current order parameters, opportunity window, and surplus material resource pool. It uses multiple linear regression to impute missing values ​​and unit normalization to generate a standardized attribute set. It then generates multi-source heterogeneous input data packets under a unified spatiotemporal reference framework through data packet serialization and compression encoding.

7. The method for optimizing and scheduling the slitting allowance of paper product packaging rolls according to claim 1, characterized in that, Step S4 specifically includes: Based on the comprehensive value index and physical characteristic data of each roll of scrap material in the initial scrap material value map generated in step S1, the attribute set of each scrap material node is obtained, and the attribute set is used as the basic feature input of the scrap material node in the graph model to construct a scrap material resource representation system. Based on the surplus material adaptation opportunity window sequence generated in step S2, the target width and length requirement range corresponding to each time interval is extracted, and combined with the customer priority weight, a target requirement node set is generated. Based on the specification matching relationship between the target requirement node set and the surplus material node, a potential call connection relationship is established to form an initial surplus material-demand association edge set. For each surplus material-demand association edge, calculate its multi-dimensional cost component during the scheduling process, obtain the scheduling distance from the current storage location of the surplus material to the target production equipment, and generate the scheduling distance cost by combining the energy consumption and time cost per unit distance of AGV transportation; based on the historical preparation time records of switching between the same material and different widths in the equipment mold change database, estimate the mold change preparation time required for this call; based on the comprehensive value index calculated in step S1 and the time window during which the target order allows the use of surplus materials, calculate the surplus material value loss rate using a decay function model; Based on the scheduling distance cost, the mold change preparation time, the residual material value loss rate, and the disturbance to the main line scheduling, a weighted cost function is constructed. The normalized components are multiplied by their preset strategy weights and summed to generate a comprehensive cost index for each edge. The edge set with the comprehensive cost index is integrated with the surplus material node and demand node with complete attributes into the graph database to construct a constrained directed weighted path graph model.

8. The method for optimizing the slitting allowance and scheduling of paper product packaging rolls according to claim 7, characterized in that, The disturbance degree of the mainline scheduling is quantified by comparing the difference between the idle time between adjacent orders in the current master production plan and the expected call duration.

9. The method for optimizing and scheduling the slitting allowance of paper product packaging rolls according to claim 1, characterized in that, Step S5 specifically includes: Based on the constrained waste material reuse path graph model constructed in step S4, the nodes and edges of the waste material reuse path graph model are mapped to a directed weighted graph using a graph structure modeling method to generate a directed weighted path topology. Node reachability analysis is performed on the directed weighted path topology, pre-pruning is performed based on production constraints, and candidate edge sets that meet basic process requirements are selected through Boolean logic judgment mechanism, while infeasible transfer paths are eliminated. On the pre-pruned and optimized path topology, an improved Dijkstra algorithm engine is deployed for iterative calculation. During the iteration process, a dynamic relaxation mechanism is introduced to update the cumulative cost of each path. The cumulative cost is the sum of the output values ​​of the cost functions of each edge, forming a path-level total cost index. During the path expansion process, the resource saving potential assessment value corresponding to each feasible path is calculated simultaneously. The total cost index and resource saving potential assessment value of each complete path are used as dual-objective output parameters and encapsulated into structured path scheme units to form a preliminary set of feasible path candidates. Non-dominated ranking filtering is performed on the feasible path candidate set to retain the optimal path set that satisfies the balance of multi-objective optimization, and standardized label information is added.

10. The method for optimizing and scheduling the slitting allowance of paper product packaging rolls according to claim 9, characterized in that, The production constraints include width matching threshold, remaining length satisfaction, material grade compatibility rules, and storage location accessibility restrictions.

Images