Intelligent Scheduling and Optimization Method for Circular Shared Packaging Based on Large-Scale Industrial Model
By using multi-source data acquisition and multi-objective collaborative optimization based on a large industrial model, the problems of insufficient intelligence and single optimization strategy in the scheduling of circular shared packaging are solved. Dynamic adaptation and multi-objective collaborative optimization are achieved, thereby improving the resource utilization efficiency and operational benefits of circular shared packaging.
Patent Information
- Application Number
- CN202611118000.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-27
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies lack sufficient intelligence in the scheduling of cyclic shared packaging, cannot adapt to dynamic variables in real time, and have a single optimization strategy that fails to comprehensively consider multiple constraints, resulting in resource imbalance and low operational efficiency.
The system employs a multi-source data acquisition, feature vector extraction, and multi-objective collaborative optimization scheduling module based on a large industrial model to generate multiple scheduling schemes and select the optimal scheme. It then combines multi-source data for dynamic prediction and multi-objective optimization.
It achieves intelligent and dynamic adaptation of cyclic shared packaging scheduling, avoids resource imbalance, improves the overall operational efficiency, and comprehensively considers multiple constraints to improve the adaptability and efficiency of the scheduling scheme.
Smart Images

Figure CN123117945A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent logistics technology, and in particular to an intelligent scheduling and optimization method for circular shared packaging based on a large industrial model. Background Technology
[0002] Against the backdrop of intelligent manufacturing supply chain upgrades and green logistics development, reusable shared packaging includes standardized turnover boxes, storage pallets, metal tooling racks, and special reusable containers for parts, which are widely used in industries such as automobile manufacturing, 3C electronic components, food and daily chemicals, and industrial parts distribution. Compared with disposable packaging, reusable shared packaging can circulate long-term across suppliers, production plants, logistics transit warehouses, distribution outlets, and repair stations, which can reduce material waste, lower the overall logistics costs of the supply chain, and is also a core carrier for manufacturing enterprises to achieve carbon reduction management. The operation of circular shared packaging covers the entire process of demand forecasting, cross-regional allocation, inventory control, damaged recycling, repair and refurbishment, scrapping and replacement, and asset leasing settlement. The entire circulation link involves many participants, diverse data types, and complex business constraints. The rationality of the scheduling and allocation plan directly affects the continuous operation of the production line, the turnover efficiency of packaging assets, transportation costs, and carbon emission levels. Currently, the industry's scheduling and management of shared packaging relies heavily on manual experience combined with traditional basic optimization algorithms. Digital intelligent scheduling systems are not yet mature and widespread, resulting in the following drawbacks: 1. Insufficient intelligence in scheduling algorithms: Existing circular shared packaging scheduling schemes rely solely on traditional operations research algorithms with fixed parameters or human experience. They are only suitable for stable and simple static supply and demand scenarios and cannot capture dynamic variables such as order fluctuations, packaging loss rates, delivery and receipt times of various logistics nodes, and changes in transportation capacity in real time. This can easily lead to resource imbalance problems such as large backlogs of packaging at local sites and shortages of boxes and shutdowns at downstream production sites. 2. Single dimension of scheduling optimization: Existing optimization strategies for cyclic shared packaging scheduling only take transportation mileage and single scheduling cost as core optimization indicators, without comprehensively considering multiple constraints such as packaging damage and loss rate, asset turnover efficiency, logistics energy consumption, order delivery timeliness, and warehousing and maintenance equipment operation and maintenance costs, and thus cannot achieve multi-objective collaborative equilibrium optimization. Summary of the Invention
[0003] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method for intelligent scheduling and optimization of cyclic shared packaging based on a large industrial model.
[0004] To achieve the above objectives, the present invention adopts the following technical solution: A smart scheduling and optimization system for circular shared packaging based on an industrial big model, the system includes a multi-source data acquisition module, a feature vector extraction module, a multi-objective collaborative optimization scheduling module, and a scheduling execution module; The multi-source data acquisition module is used to collect heterogeneous data from various nodes in the supply chain to generate a standardized dataset. The feature vector extraction module is used to encode the standardized dataset and perform time-series weighted fusion to obtain the overall time-series features. The multi-objective collaborative optimization scheduling module is used to generate multiple scheduling schemes by combining constraints and output the optimal scheduling instruction; The scheduling execution module is used to issue scheduling instructions to drive on-site operations and transmit running data back to the acquisition module.
[0005] This invention also proposes an intelligent scheduling and optimization method for cyclic shared packaging based on a large industrial model, comprising the following sub-steps: S1: Real-time acquisition of multi-source heterogeneous data; A fixed-duration sliding window is pre-configured and each supply chain node is assigned a unique identifier. The multi-source data acquisition module connects to the business systems of each node in the supply chain and, based on the unique identifier of each supply chain node, collects multi-source heterogeneous data and corresponding timestamps under the corresponding supply chain node within the fixed-duration sliding window. The supply chain nodes include factories, transit warehouses, and maintenance stations; The multi-source heterogeneous data includes cyclically shared packaging basic data, business demand data, logistics resource data, loss and maintenance data, and asset loss data. Each type of data carries parameter values for corresponding business fields. Multi-source heterogeneous data belonging to the same supply chain node with unique identifiers and timestamps are concatenated and merged to obtain a data sample; each data sample contains the unique identifier and timestamp of the supply chain node, and also stores the parameter values corresponding to each business field; Each data sample obtained by splicing is preprocessed, including data cleaning, removal of abnormal data, and filling in missing fields; each preprocessed data sample is then saved into a multi-source heterogeneous dataset.
[0006] S2: Construct a comprehensive feature vector and weighted fuse it to generate overall temporal features; S21: Generate a composite feature vector with fixed dimensions; The feature vector extraction module reads data samples from the multi-source heterogeneous dataset one by one, and generates a numerical sequence and a total category code sequence based on the data samples; the numerical sequence and the total category code sequence are concatenated and fused in sequence to generate a comprehensive feature vector with fixed dimensions; S22: Generate overall temporal features with fixed dimensions; The comprehensive feature vectors are weighted and fused to generate a fixed-dimensional overall time-series feature.
[0007] S3: Outputs predicted data based on a large industrial model; The fixed-dimensional overall temporal features are input into the pre-trained industrial big model. The industrial big model has constructed a large number of historical samples by collecting multi-source heterogeneous data in advance, and the fixed-dimensional overall temporal features generated by each historical sample are used to complete the learning, so as to obtain the change function between business demand, logistics resources, loss and maintenance, asset loss and time corresponding to the cyclic shared packaging in each supply chain node. The industrial big data model receives the overall time-series features of the fixed dimensions and outputs business demand prediction data, logistics resource prediction data, loss and maintenance prediction data, and asset loss prediction data corresponding to the preset time based on the change function.
[0008] S4: Generate scheduling instructions and send them to the scheduling execution module; The generated business demand forecast data, logistics resource forecast data, loss and maintenance forecast data, and asset loss forecast data are input into the multi-objective collaborative optimization scheduling module. The multi-objective collaborative optimization scheduling module receives the above four types of forecast data and generates multiple cyclic shared packaging scheduling schemes that meet the constraints of each supply chain node, such as the inventory threshold, delivery deadline, and vehicle capacity limit, based on the constraints collected in advance. The system retrieves the pre-configured business demand weight, logistics resource weight, loss and maintenance weight, and asset loss weight, extracts the original values of the four dimensions corresponding to each scheduling scheme, multiplies the original values of each dimension with the corresponding weight, and sums them to obtain the comprehensive evaluation value. The comprehensive evaluation value is compared with the comprehensive evaluation value of all cyclic shared packaging scheduling schemes, and the cyclic shared packaging scheduling scheme with the highest comprehensive evaluation value is selected. The scheduling instruction is then generated and sent to the scheduling execution module. The scheduling execution module receives scheduling instructions and distributes them to various business systems to perform operations such as allocation, transportation, and maintenance.
[0009] Compared with the prior art, the beneficial effects of the present invention are as follows: This method collects heterogeneous data from multiple sources through a multi-source data acquisition module with a fixed-duration sliding window. After feature vector extraction and processing to generate overall time-series features, the data is input into a pre-trained industrial large-scale model. Based on the business change function learned by the model, it outputs multiple types of prediction data. It abandons the limitations of traditional fixed-parameter operation research algorithms and human experience, and can adapt to dynamic supply and demand scenarios. It can accurately predict the packaging supply and demand of various factories, transit warehouses, and maintenance stations, avoid resource imbalance problems such as local packaging backlog and downstream box shortages and shutdowns, and improve the intelligence and dynamic adaptability of the scheduling algorithm. This invention utilizes a multi-objective collaborative optimization scheduling module to comprehensively incorporate multiple constraints, including packaging damage and loss rate, asset turnover efficiency, logistics energy consumption, order delivery timeliness, and warehousing and maintenance equipment operation and maintenance costs. It combines preset weights to calculate a comprehensive evaluation value for the multi-dimensional indicators of each scheduling scheme, selects the optimal allocation scheme, coordinates multiple objectives, achieves multi-objective collaborative balance optimization, and improves the overall operational efficiency of the entire process of circular shared packaging. Attached Figure Description
[0010] Figure 1 This is a flowchart illustrating the steps of the intelligent scheduling and optimization method for cyclic shared packaging based on a large industrial model proposed in this invention. Detailed Implementation
[0011] To provide a further understanding of the purpose, structure, features, and functions of the present invention, detailed descriptions are provided below with reference to specific embodiments.
[0012] A smart scheduling and optimization system for circular shared packaging based on an industrial big model, the system includes a multi-source data acquisition module, a feature vector extraction module, a multi-objective collaborative optimization scheduling module, and a scheduling execution module; The multi-source data acquisition module is used to collect heterogeneous data from various nodes in the supply chain to generate a standardized dataset. The feature vector extraction module is used to encode the standardized dataset and perform time-series weighted fusion to obtain the overall time-series features. The multi-objective collaborative optimization scheduling module is used to generate multiple scheduling schemes by combining constraints and output the optimal scheduling instruction; The scheduling execution module is used to issue scheduling instructions to drive on-site operations and transmit running data back to the acquisition module.
[0013] Please refer to the reference. Figure 1 This invention provides an intelligent scheduling and optimization method for cyclic shared packaging based on a large industrial model, comprising the following sub-steps: S1: Real-time acquisition of multi-source heterogeneous data; A fixed-duration sliding window is pre-configured and each supply chain node is assigned a unique identifier. The multi-source data acquisition module connects to the business systems of each node in the supply chain and, based on the unique identifier of each supply chain node, collects multi-source heterogeneous data and corresponding timestamps under the corresponding supply chain node within the fixed-duration sliding window. The supply chain nodes include factories, transit warehouses, and maintenance stations; The multi-source heterogeneous data includes cyclically shared packaging basic data, business demand data, logistics resource data, loss and maintenance data, and asset loss data. Each type of data carries parameter values for corresponding business fields. Multi-source heterogeneous data belonging to the same supply chain node with unique identifiers and timestamps are concatenated and merged to obtain a data sample; each data sample contains the unique identifier and timestamp of the supply chain node, and also stores the parameter values corresponding to each business field; Specifically, the multi-source data acquisition module collects basic data on reusable shared packaging from the packaging rental management system. This basic data includes the unique packaging number, packaging category (turnover box, pallet, tooling rack, special reusable container), size specifications, procurement cost, set service life, and packaging usage status (intact, awaiting repair, scrapped, leased out, idle). The multi-source data acquisition module collects business demand data from the Manufacturing Execution System (MES), including packaging demand for each time period and delivery deadline. The multi-source data acquisition module collects logistics resource data from the transportation management system (TMS). The logistics resource data includes transportation mileage, single vehicle load, single transportation cost, number of available vehicles, loading and unloading time, trunk transportation distance, fuel consumption, etc. The multi-source data acquisition module collects loss and maintenance data from the operation and maintenance system. The loss and maintenance data includes damage rate, maintenance hours, scrap quantity, refurbishment cost, etc. The multi-source data acquisition module collects asset loss data from the cost control system. The asset loss data includes daily warehouse rental, rental unit price, and manual dispatching costs. Each data sample obtained by splicing is preprocessed, including data cleaning, removal of abnormal data, and filling in missing fields; each preprocessed data sample is then saved into a multi-source heterogeneous dataset.
[0014] S2: Construct a comprehensive feature vector and weighted fuse it to generate overall temporal features; S21: Generate a composite feature vector with fixed dimensions; The feature vector extraction module reads data samples from the multi-source heterogeneous dataset one by one, normalizes business fields with numerical parameter values, and concatenates them into a numerical sequence based on a preset fixed field order. For business fields with string parameters, all enumeration categories of the field under the entire supply chain business process are collected in advance. The total number N of enumeration categories of the field is counted, and a binary vector with a dimension length of N is generated. Unit vectors are assigned sequentially according to the enumeration category sorting order. Only the dimension matching the current category is assigned a value of 1, and all other dimensions are assigned a value of 0. The binary encoding vector corresponding to each category is generated, and the mapping relationship between enumeration categories and binary encoding vectors is constructed. The enumeration category of the business field in a single data sample is read, and the binary encoding vector corresponding to the enumeration category is matched based on the mapping relationship. The binary encoding vectors obtained by matching the string business fields in each data sample are concatenated according to the preset field order to obtain the total sequence of category codes. The numerical sequence and the total category code sequence are sequentially concatenated and merged to generate a comprehensive feature vector of fixed dimensions; each data sample corresponds to a comprehensive feature vector. S22: Generate overall temporal features with fixed dimensions; The feature vector extraction module groups all comprehensive feature vectors within a fixed-duration sliding window based on the unique identifier of the supply chain node, sorts the comprehensive feature vectors in each group in ascending order of timestamp, and assigns continuously increasing position indices based on the sorting order. After normalizing the location index, the corresponding one-dimensional time series feature is obtained. The one-dimensional time series feature is then concatenated with the corresponding comprehensive feature vector to obtain the comprehensive time series feature vector. Using any comprehensive time-series feature vector under the unique identifier of each supply chain node as the reference vector, the similarity value between the reference vector and all other comprehensive time-series feature vectors under the unique identifier of the same supply chain node is calculated using the Euclidean distance similarity calculation algorithm. The Softmax function is used to normalize all the obtained similarity values. The normalized similarity values are the association weights, and the sum of all association weights corresponding to the same reference vector is 1. The remaining comprehensive time-series feature vectors under the unique identifier of the same supply chain node, excluding the baseline vector, are multiplied by their associated weights and then summed to obtain the updated baseline features. Each comprehensive time-series feature vector is used as the baseline vector in turn to complete the weighted update calculation, resulting in multiple updated baseline features. The multiple updated baseline features are arranged in ascending order of timestamps, and the mean value of each feature bit of all updated baseline features is calculated and fused along this arrangement dimension to generate a fixed-dimensional overall time-series feature.
[0015] S3: Outputs predicted data based on a large industrial model; The fixed-dimensional overall temporal features are input into the pre-trained industrial big model. The industrial big model has constructed a large number of historical samples by collecting multi-source heterogeneous data in advance, and the fixed-dimensional overall temporal features generated by each historical sample are used to complete the learning, so as to obtain the change function between business demand, logistics resources, loss and maintenance, asset loss and time corresponding to the cyclic shared packaging in each supply chain node. The industrial big data model receives the overall time-series features of the fixed dimensions and outputs business demand prediction data, logistics resource prediction data, loss and maintenance prediction data, and asset loss prediction data corresponding to the preset time based on the change function.
[0016] S4: Generate scheduling instructions and send them to the scheduling execution module; The generated business demand forecast data, logistics resource forecast data, loss and maintenance forecast data, and asset loss forecast data are input into the multi-objective collaborative optimization scheduling module. The multi-objective collaborative optimization scheduling module receives the above four types of prediction data and generates multiple cyclic shared packaging scheduling schemes that meet the constraints of each supply chain node, such as the inventory threshold, delivery deadline, and vehicle capacity limit, based on the constraints collected in advance. The cyclic shared packaging scheduling scheme contains the original values corresponding to each dimension, including the number of cyclic shared packaging transfers between supply chain nodes, the transfer execution period, the transportation route, the estimated transportation cost, the operation and maintenance loss cost, and the order delivery time. The system retrieves the pre-configured business demand weights, logistics resource weights, loss and maintenance weights, and asset loss weights, extracts the original values corresponding to each scheduling scheme, multiplies the original values of each dimension with their corresponding weights, and sums them to obtain a comprehensive evaluation value. The comprehensive evaluation value is then compared with that of all cyclic shared packaging scheduling schemes, and the cyclic shared packaging scheduling scheme with the highest comprehensive evaluation value is selected. A scheduling instruction is then generated and sent to the scheduling execution module. The scheduling execution module receives scheduling instructions and sends them to various business systems to perform operations such as allocation, transportation, and maintenance, and collects on-site operation data in real time and sends it back to the front-end data acquisition module.
[0017] The present invention has been described in the above-described embodiments; however, these embodiments are merely examples for implementing the present invention. It must be noted that the disclosed embodiments do not limit the scope of the present invention. Conversely, any modifications and refinements made without departing from the spirit and scope of the present invention are within the scope of patent protection of the present invention.
Claims
1. A method for intelligent scheduling and optimization of cyclic shared packaging based on a large industrial model, characterized in that: Includes the following sub-steps: S1: Real-time acquisition of multi-source heterogeneous data; A fixed-duration sliding window is pre-configured and each supply chain node is assigned a unique identifier. The multi-source data acquisition module connects to the business systems of each node in the supply chain and, based on the unique identifier of each supply chain node, collects multi-source heterogeneous data and corresponding timestamps under the corresponding supply chain node within the fixed-duration sliding window. Multi-source heterogeneous data belonging to the same supply chain node with unique identifiers and timestamps are concatenated and merged to obtain a data sample; each preprocessed data sample is saved to a multi-source heterogeneous dataset. S2: Construct a comprehensive feature vector and weighted fuse it to generate overall temporal features; S21: Generate a composite feature vector with fixed dimensions; The feature vector extraction module reads data samples from the multi-source heterogeneous dataset one by one, and generates a numerical sequence and a total category code sequence based on the data samples; the numerical sequence and the total category code sequence are concatenated and fused in sequence to generate a comprehensive feature vector with fixed dimensions; S22: Generate overall temporal features with fixed dimensions; The comprehensive feature vectors are weighted and fused to generate overall temporal features of fixed dimensions. S3: Outputs predicted data based on a large industrial model; The fixed-dimensional overall time series features are input into the pre-trained industrial big data model. The industrial big data model receives the fixed-dimensional overall time series features and outputs business demand prediction data, logistics resource prediction data, loss and maintenance prediction data, and asset loss prediction data corresponding to the preset time. S4: Generate scheduling instructions and send them to the scheduling execution module; The generated business demand forecast data, logistics resource forecast data, loss and maintenance forecast data, and asset loss forecast data are input into the multi-objective collaborative optimization scheduling module. The multi-objective collaborative optimization scheduling module generates scheduling instructions and sends them to the scheduling execution module. The scheduling execution module receives the scheduling instructions and sends them to each business system to execute the scheduling instructions.
2. The intelligent scheduling and optimization method for cyclic shared packaging based on a large industrial model as described in claim 1, characterized in that: In step S1, the multi-source heterogeneous data includes cyclically shared packaging basic data, business demand data, logistics resource data, loss and maintenance data, and asset loss data. Each type of data carries parameter values for the corresponding business fields. Each data sample contains a unique identifier for the supply chain node and a timestamp, while also storing the parameter values corresponding to each business field; Each data sample obtained by splicing is preprocessed by data cleaning, removing abnormal data, and filling in missing fields.
3. The intelligent scheduling and optimization method for cyclic shared packaging based on a large industrial model as described in claim 1, characterized in that: In step S21, the method for generating the numerical sequence and the total category code sequence is as follows: Normalize business fields with numeric parameter values and concatenate them into a numeric sequence based on a preset fixed field order; For business fields with string parameters, all enumeration categories of the field under the entire supply chain business process are collected in advance. The total number of enumeration categories N is counted, and a binary vector with a dimension length of N is generated. Unit vectors are assigned sequentially according to the enumeration category sorting order. Only the dimension matching the current category is assigned a value of 1, and all other dimensions are assigned a value of 0. The binary encoding vector corresponding to each category is generated, and the mapping relationship between enumeration categories and binary encoding vectors is constructed. The enumeration category of the business field in a single data sample is read, and the binary encoding vector corresponding to the enumeration category is matched based on the mapping relationship. The binary encoding vectors obtained by matching the string business fields in each data sample are concatenated according to the preset field order to obtain the total sequence of category codes.
4. The intelligent scheduling and optimization method for cyclic shared packaging based on a large industrial model as described in claim 1, characterized in that: The specific details of step S22 are as follows: The feature vector extraction module groups all comprehensive feature vectors within a fixed-duration sliding window based on the unique identifier of the supply chain node, sorts the comprehensive feature vectors in each group in ascending order of timestamp, and assigns continuously increasing position indices based on the sorting order. After normalizing the location index, the corresponding one-dimensional time series feature is obtained. The one-dimensional time series feature is then concatenated with the corresponding comprehensive feature vector to obtain the comprehensive time series feature vector. Using any comprehensive time-series feature vector under the unique identifier of each supply chain node as the reference vector, the similarity value between the reference vector and all other comprehensive time-series feature vectors under the unique identifier of the same supply chain node is calculated using the Euclidean distance similarity calculation algorithm. The Softmax function is used to normalize all the obtained similarity values. The normalized similarity values are the association weights, and the sum of all association weights corresponding to the same reference vector is 1. The remaining comprehensive time-series feature vectors under the unique identifier of the same supply chain node, excluding the baseline vector, are multiplied by the association weights and then summed to obtain the updated baseline features; Each comprehensive temporal feature vector is used in turn as a reference vector to complete the weighted update calculation, resulting in multiple corresponding updated reference features; Multiple updated baseline features are arranged in ascending order of timestamps. The mean of each feature position of all updated baseline features is calculated along this arrangement dimension and fused to generate an overall time series feature with a fixed dimension.
5. The intelligent scheduling and optimization method for cyclic shared packaging based on a large industrial model as described in claim 1, characterized in that: The industrial big data model was constructed in advance by collecting a large number of historical samples from multiple sources of heterogeneous data. The model then learned by generating fixed-dimensional overall time-series features from each historical sample, and obtained the variation function between business demand, logistics resources, loss and maintenance, asset loss and time corresponding to the cyclic shared packaging in each supply chain node.
6. The intelligent scheduling and optimization method for cyclic shared packaging based on a large industrial model as described in claim 1, characterized in that: In step S4, the multi-objective collaborative optimization scheduling module receives the above four types of prediction data and generates multiple cyclic shared packaging scheduling schemes that meet the constraints of the inventory threshold, delivery deadline, and vehicle capacity limit of each supply chain node collected in advance. The circular shared packaging scheduling scheme includes the original values corresponding to each dimension; The system retrieves the pre-configured business demand weights, logistics resource weights, loss and maintenance weights, and asset loss weights, extracts the original values corresponding to each scheduling scheme, multiplies the original values of each dimension with their corresponding weights, and sums them to obtain a comprehensive evaluation value. The comprehensive evaluation value is then compared with that of all cyclic shared packaging scheduling schemes, and the cyclic shared packaging scheduling scheme with the highest comprehensive evaluation value is selected. A scheduling instruction is then generated and sent to the scheduling execution module.
7. A smart scheduling and optimization system for circular shared packaging based on an industrial large-scale model for implementing the method of any one of claims 1-6, characterized in that: The system includes a multi-source data acquisition module, a feature vector extraction module, a multi-objective collaborative optimization scheduling module, and a scheduling execution module.
