An AI-driven dynamic optimization and scheduling system for a trade supply chain
By leveraging the synergistic effects of the data module, AI model module, and feedback iteration module, the problems of data silos and dynamic adaptation in commercial supply chain scheduling are solved, achieving efficient and reliable supply chain scheduling optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HEBEI YANHE TECHNOLOGY CO LTD
- Filing Date
- 2026-02-28
- Publication Date
- 2026-06-05
AI Technical Summary
Existing AI-driven commercial supply chain scheduling technologies suffer from data silos, lack secure and efficient data aggregation and sharing mechanisms, have poor adaptability to new product scheduling, lack scenario simulation and risk prediction capabilities, and lack dynamic iteration mechanisms for scheduling schemes, making them unable to respond to dynamic changes in the supply chain in real time.
By employing data modules, AI model modules, scheduling and execution modules, and feedback and iteration modules, the system achieves secure aggregation and standardized processing of data from various entities in the supply chain. It integrates federated learning, transfer learning, digital twin, and reinforcement learning modules to construct a full-process scheduling solution, and optimizes the scheduling solution in real time through the feedback and iteration module.
Break down data silos, improve scheduling accuracy, quickly adapt to new products, avoid scheduling risks, achieve dynamic adaptive optimization of the supply chain, and enhance the stability and reliability of scheduling solutions.
Smart Images
Figure CN122155566A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of artificial intelligence and supply chain management technology, and in particular to an AI-driven dynamic optimization and scheduling system for commercial supply chains. Background Technology
[0002] Currently, the integration of artificial intelligence technology and supply chain management is becoming increasingly profound. As the core carrier connecting upstream suppliers, warehousing and logistics and downstream end consumers, the scheduling efficiency and optimization accuracy of the commercial supply chain directly determine the operating costs and market competitiveness of enterprises. Artificial intelligence technology has been gradually applied to all aspects of supply chain scheduling, providing intelligent support for inventory management, transportation capacity allocation, demand forecasting, etc., and has to some extent replaced the traditional manual experience-driven scheduling model, thus improving the operational efficiency of the supply chain.
[0003] However, existing AI-driven commercial supply chain scheduling technologies still have many limitations, making it difficult to meet the dynamic and refined scheduling needs of the commercial supply chain. This results in the failure to fully realize the value of AI technology. Among these limitations, the problem of data silos among multiple entities is prominent. Data from various participants in the supply chain is stored in a scattered manner, lacking a secure and efficient aggregation and sharing mechanism. This leads to incomplete training data for AI models, directly affecting scheduling accuracy. Furthermore, the adaptability to new product scheduling is poor. When new SKUs without historical data appear, traditional AI models cannot adapt quickly, requiring retraining of the entire model, resulting in low response efficiency and potential scheduling deviations. There is a lack of effective scenario simulation and risk prediction capabilities, making it impossible to preview the execution effect of scheduling plans in advance and to avoid scheduling risks caused by dynamic fluctuations in the supply chain. Finally, scheduling plans lack dynamic iteration mechanisms, relying mostly on static planning, and cannot respond in real time to dynamic changes in warehousing, transportation capacity, and end-user demand. Moreover, scheduling errors cannot be promptly fed back and corrected.
[0004] In summary, an AI-driven dynamic optimization and scheduling system for the commercial supply chain is proposed. Summary of the Invention
[0005] This application aims to at least partially solve one of the technical problems in the aforementioned technologies.
[0006] To achieve the above objectives, the first aspect of this application proposes an AI-driven dynamic optimization and scheduling system for the commercial supply chain, including a data module, an AI model module, a scheduling execution module, and a feedback iteration module;
[0007] The data module is used to collect, store, and preprocess data information from all nodes of the commercial supply chain, focusing on achieving secure aggregation and standardized processing of data from various entities in the supply chain, providing reliable data support for subsequent AI model calculations;
[0008] The AI model module integrates federated learning, transfer learning, digital twin, and reinforcement learning modules. This module takes the standardized data output by the data module as input to complete model training, scheduling scenario simulation, and precise tuning of scheduling parameters.
[0009] The scheduling and execution module is responsible for receiving the optimized scheduling plan output by the AI model module, and specifically executing the scheduling operations of the entire supply chain, such as warehouse outbound, transportation capacity allocation, cross-warehouse transfer and terminal distribution, to ensure that the scheduling plan is implemented.
[0010] The feedback iteration module collects the actual operating data of the scheduling execution module in real time, accurately calculates the scheduling error, and promptly feeds the error information back to the AI model module, driving the continuous iteration and optimization of the model and scheduling scheme.
[0011] In addition, the AI-driven dynamic optimization and scheduling system for the commercial supply chain proposed in this application may also have the following additional technical features:
[0012] As a further description of the above technical solution:
[0013] The data module includes a multi-entity data acquisition unit, a data preprocessing unit, and a data security storage unit;
[0014] The multi-entity data collection unit comprehensively covers upstream suppliers, warehousing nodes, terminal stores, and cooperative transportation service providers in the commercial supply chain. The types of data collected include SKU attribute data, inventory dynamic data, transportation status data, terminal demand data, and scheduling execution feedback data, comprehensively covering all aspects of the supply chain.
[0015] As a further description of the above technical solution:
[0016] The data preprocessing unit cleans, denoises, standardizes, and extracts features from the collected raw data, converting unstructured data into structured feature vectors that can be directly used for AI calculations. The feature vectors include SKU core attribute features, inventory turnover features, transportation capacity matching features, and terminal demand fluctuation features, which can accurately adapt to the input requirements of the AI model module and ensure the accuracy of model calculations.
[0017] As a further description of the above technical solution:
[0018] The federated learning module in the AI model module adopts a lightweight node deployment mode and works in collaboration with the data module. It aggregates feature data from multiple nodes across the entire supply chain through federated parameter interaction.
[0019] As a further description of the above technical solution:
[0020] The transfer learning module in the AI model module works based on the pre-trained basic AI scheduling model of the system. The basic AI scheduling model is a hybrid model of lightweight XGBoost and Transformer. Its training data comes from the full set of mature SKU scheduling data accumulated by the data module over a long period of time and the multi-node data aggregated by the federated learning module.
[0021] As a further description of the above technical solution:
[0022] The digital twin module in the AI model module is used to build a digital twin scenario for the entire process of commercial supply chain scheduling, which restores the entire process of warehousing and outbound, transportation, terminal distribution and inventory turnover. Its access data module collects dynamic supply chain data in real time, previews the execution effect of scheduling plans, quantifies scheduling errors, and accurately marks abnormal risk points in the scheduling process.
[0023] As a further description of the above technical solution:
[0024] The reinforcement learning module in the AI model module takes maximizing supply chain scheduling efficiency and minimizing scheduling error as its core reward function. Its input data includes simulated data output by the digital twin module and actual scheduling data output by the feedback iteration module.
[0025] As a further description of the above technical solution:
[0026] The scheduling and execution module includes a warehouse scheduling unit, a transportation scheduling unit, a cross-warehouse allocation unit, and a terminal distribution scheduling unit. All of them communicate bidirectionally with the AI model module and the feedback iteration module, accurately receive the optimization scheduling instructions issued by the AI model module, and promptly feed back their own real-time execution data to the feedback iteration module.
[0027] As a further description of the above technical solution:
[0028] The error calculation unit is used to accurately calculate the deviation between the actual running data of the scheduling execution module and the preset target data of the AI model module;
[0029] When the actual scheduling error exceeds the preset threshold, the feedback transmission unit automatically feeds back the error data and related abnormal information to the AI model module.
[0030] As a further description of the above technical solution:
[0031] It also includes a manual intervention module, which works in two directions with the scheduling and execution module and the AI model module. When the scheduling scheme output by the AI model module or the abnormal risk marked by the feedback iteration module exceeds the preset range, it can automatically trigger the manual review and parameter adjustment process.
[0032] Advantages of this invention:
[0033] According to this application, an AI-driven dynamic optimization and scheduling system for the commercial supply chain solves the problem of data silos. Through federated learning and data module collaboration, it achieves secure aggregation of data from multiple entities that is available but not visible, ensuring data comprehensiveness and security, providing high-quality data support for AI scheduling, and improving scheduling accuracy.
[0034] Relying on the transfer learning module, there is no need to retrain the entire model. Through feature transfer and local fine-tuning, it can quickly adapt to new product scheduling scenarios without historical data, and significantly shorten the response cycle.
[0035] By leveraging digital twin modules to construct full-process simulation scenarios, scheduling effects can be previewed in advance, errors can be quantified, and risk points can be marked, effectively avoiding scheduling risks caused by dynamic fluctuations in the supply chain;
[0036] By combining reinforcement learning and feedback iteration modules, scheduling execution data is received in real time, and parameters and iteration schemes are automatically adjusted to adapt to dynamic changes in the supply chain.
[0037] It boasts high stability and reliability, and features an added manual intervention module to enable human-machine collaboration, avoiding the drawbacks of AI black box decision-making and ensuring a stable and controllable scheduling process.
[0038] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description
[0039] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:
[0040] Figure 1 This is a schematic diagram of the module connections of an AI-driven dynamic optimization and scheduling system for the commercial supply chain according to an embodiment of this application;
[0041] Figure 2 This is a schematic diagram illustrating the system principle of an AI-driven dynamic optimization and scheduling system for the commercial supply chain according to an embodiment of this application. Detailed Implementation
[0042] The embodiments of this application are described in detail below. Examples of these embodiments 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 intended to explain this application, and should not be construed as limiting this application.
[0043] The following description, in conjunction with the accompanying drawings, illustrates an AI-driven dynamic optimization and scheduling system for the commercial supply chain, representing an embodiment of this application.
[0044] like Figure 1-2 As shown in Embodiment 1 of this application, an AI-driven dynamic optimization and scheduling system for the commercial supply chain may include a data module, an AI model module, a scheduling execution module, and a feedback iteration module.
[0045] The data module is used to collect, store, and preprocess data information from all nodes of the commercial supply chain, focusing on achieving secure aggregation and standardized processing of data from various entities in the supply chain, providing reliable data support for subsequent AI model calculations;
[0046] The AI model module integrates federated learning, transfer learning, digital twin, and reinforcement learning modules. This module takes the standardized data output by the data module as input to complete model training, scheduling scenario simulation, and precise tuning of scheduling parameters.
[0047] The scheduling and execution module is responsible for receiving the optimized scheduling plan output by the AI model module, and specifically executing the scheduling operations of the entire supply chain, such as warehouse outbound, transportation capacity allocation, cross-warehouse transfer and terminal distribution, to ensure that the scheduling plan is implemented.
[0048] The feedback iteration module collects the actual operating data of the scheduling execution module in real time, accurately calculates the scheduling error, and promptly feeds the error information back to the AI model module to drive the continuous iteration and optimization of the model and scheduling scheme.
[0049] In the above solution, the data module standardizes and aggregates data across all nodes of the supply chain, providing a foundation for AI model computation. The AI model module integrates multiple AI technologies and leverages algorithmic advantages to intelligently generate and optimize scheduling schemes. The scheduling execution module transforms model outputs into actual scheduling actions, ensuring the implementation of the scheme. The feedback iteration module drives continuous model correction through real-time error collection and feedback, forming a dynamic cycle of data-model-execution-feedback. By utilizing AI technology to overcome the limitations of traditional static scheduling, the solution achieves adaptive optimization of supply chain scheduling through multi-module collaboration, ensuring that scheduling accuracy and efficiency are improved simultaneously.
[0050] like Figure 1-2 As shown:
[0051] The data module comprises a multi-entity data acquisition unit, a data preprocessing unit, and a secure data storage unit. The multi-entity data acquisition unit comprehensively covers upstream suppliers, warehousing nodes, terminal stores, and cooperative transportation service providers in the commercial supply chain. The collected data types include SKU attribute data, inventory dynamic data, transportation status data, terminal demand data, and scheduling execution feedback data, comprehensively covering all aspects of the supply chain. In this solution, data is the core foundation of AI scheduling; its completeness and accuracy directly determine scheduling precision. Therefore, the data module adopts a three-tier architecture of acquisition-preprocessing-storage. The multi-entity data acquisition unit covers all nodes in the supply chain, ensuring that the collected data is comprehensive and fully reflects the operational status of each link in the supply chain. Raw data suffers from noise and inconsistent formats, making it unsuitable for direct use in AI model calculations. Through cleaning, denoising, standardization, and feature extraction, unstructured data is transformed into structured data, eliminating data bias. The secure data storage unit utilizes multi-entity supply chain data, which involves commercial privacy, and requires a secure storage mechanism to ensure data security and provide reliable support for subsequent data aggregation and model calculations.
[0052] like Figure 1-2 As shown:
[0053] The data preprocessing unit cleans, denoises, standardizes, and extracts features from the collected raw data, transforming unstructured data into structured feature vectors that can be directly used for AI calculations. These feature vectors include SKU core attribute features, inventory turnover features, capacity matching features, and terminal demand fluctuation features, accurately adapting to the input requirements of the AI model module and ensuring the accuracy of model calculations. In this solution, AI model calculations rely on standardized, structured data input. Raw data is often unstructured (e.g., text, scattered numerical values) and contains outliers. Direct input would lead to model calculation deviations and decreased accuracy. Therefore, the data preprocessing unit is necessary. This involves cleaning to remove outliers, denoising to eliminate irrelevant interference, and standardizing data dimensions to ensure consistency. Feature extraction identifies core features strongly correlated with supply chain scheduling, constructing feature vectors that focus on key scheduling influencing factors, reducing redundant data interference with model calculations, and ensuring the data format accurately matches the input requirements of the AI model module. This lays the foundation for accurate model calculations and ensures the reliability of subsequent scheduling schemes.
[0054] like Figure 1-2 As shown:
[0055] The federated learning module in the AI model module adopts a lightweight node deployment mode and works in collaboration with the data module. It aggregates feature data from multiple nodes across the entire supply chain through federated parameter interaction. In this solution, the existing data of multiple entities in the supply chain is stored separately and not shared. Directly aggregating the raw data would pose a privacy risk and would not meet the data security requirements of commercial scenarios. Therefore, the federated learning module is designed in collaboration with the data module. The lightweight node deployment does not require large-scale modification of the existing data systems of each entity, reducing deployment costs. Through federated parameter interaction, each entity only shares data operation parameters and does not disclose the raw data, achieving usability without visibility. This ensures data security while aggregating data from multiple nodes across the entire supply chain. By aggregating data from multiple entities, the shortcomings of incomplete data from a single node are compensated for, providing full-dimensional data for subsequent AI sub-model training. This improves the model scheduling accuracy from the source and solves the scheduling deviation problem caused by data silos.
[0056] like Figure 1-2 As shown:
[0057] The transfer learning module in the AI model module operates based on the pre-trained basic AI scheduling model. This basic AI scheduling model is a lightweight hybrid model of XGBoost and Transformer, and its training data comes from the full set of mature SKU scheduling data accumulated over a long period by the data module and multi-node data aggregated by the federated learning module. In this solution, the basic AI scheduling model has already been trained with full set of mature SKU data, accumulating general logic for supply chain scheduling, eliminating the need for repeated training. New product SKUs and mature SKUs share similar attribute characteristics (such as warehousing and transportation requirements), allowing the reuse of scheduling feature logic from mature SKUs. By fine-tuning the top-level parameters of the model for the specific attributes of new product SKUs, the model can be adapted to the scheduling requirements of new products without retraining the full model, significantly shortening the model adaptation cycle. By balancing the requirements of model computational accuracy and lightweight deployment, the model can be quickly embedded into the main framework of the system, improving the response efficiency of new product scheduling.
[0058] like Figure 1-2 As shown:
[0059] The digital twin module within the AI model module is used to construct a digital twin scenario for the entire business supply chain scheduling process. It recreates the entire process of warehousing and outbound delivery, transportation, terminal distribution, and inventory turnover. Its access data module collects real-time dynamic supply chain data, allowing for pre-simulation of the scheduling plan's execution effect, quantifying scheduling errors, and accurately marking abnormal risk points in the scheduling process. In this solution, digital twin technology uses virtual simulation to recreate the entire supply chain process, mapping the physical world's scheduling scenario into a virtual scenario. This achieves virtual pre-simulation and real-world implementation, ensuring synchronization between the virtual and physical scenarios and that the pre-simulation results closely match reality. By pre-simulating the scheduling plan, the execution effect under different dynamic scenarios can be simulated in advance. By comparing the pre-simulation results with preset targets, deviations can be accurately identified. Virtual simulation captures weak links in the scheduling process, allowing for early optimization of the plan, avoiding scheduling risks after actual implementation, reducing operational losses, and improving the feasibility of the scheduling plan.
[0060] like Figure 1-2 As shown:
[0061] The reinforcement learning module in the AI model module uses maximizing supply chain scheduling efficiency and minimizing scheduling error as its core reward function. Its input data includes simulated data output from the digital twin module and actual scheduling data output from the feedback iteration module. In this scheme, reinforcement learning continuously optimizes scheduling decisions through an input-decision-reward-optimization loop mechanism. By using maximizing scheduling efficiency and minimizing error as the reward function, the model optimization objective is clearly defined, ensuring that model adjustments align with the core needs of the supply chain. Combining virtual simulations with actual operation ensures that parameter adjustments are based on evidence, avoiding blind optimization. It captures dynamic fluctuation signals in the supply chain and iteratively corrects parameters through algorithms, enabling the scheduling scheme to adapt synchronously to fluctuations. This breaks the limitations of static scheduling, achieves dynamic adaptive optimization, and improves the adaptability and accuracy of the scheduling scheme.
[0062] like Figure 1-2 As shown:
[0063] The scheduling execution module includes a warehouse scheduling unit, a transportation capacity scheduling unit, a cross-warehouse allocation unit, and a terminal distribution scheduling unit. All of these units communicate bidirectionally with the AI model module and the feedback iteration module, accurately receiving optimized scheduling instructions from the AI model module and promptly feeding back their real-time execution data to the feedback iteration module. This solution divides units according to the core links of supply chain scheduling (warehousing, transportation capacity, cross-warehouse, and distribution), achieving segmented control and full-process coverage. This ensures accurate implementation of each scheduling link, avoids process omissions, and ensures that each unit accurately receives optimized instructions from the AI model module, preventing execution errors caused by instruction deviations. Furthermore, it feeds back real-time execution data to the feedback iteration module, providing real data support for error calculation and model optimization. Simultaneously, it enables full-process data traceability in scheduling, facilitating subsequent anomaly investigation and process optimization, and ensuring the efficiency and controllability of scheduling execution.
[0064] like Figure 1-2 As shown:
[0065] The error calculation unit is used to accurately calculate the deviation between the actual running data of the scheduling execution module and the preset target data of the AI model module. When the actual scheduling error exceeds the preset threshold, the feedback transmission unit automatically feeds back the error data and related abnormal information to the AI model module. In this solution, the error calculation unit quantifies the magnitude of the error and clarifies the degree of deviation by comparing the actual execution data with the preset target data, providing direction for optimization. Combined with the actual business needs of the supply chain, it defines the acceptable error range to avoid over-optimization or ignoring key deviations. When the error exceeds the threshold, it automatically feeds back and triggers the optimization process, realizing an automated closed loop of error-feedback-optimization. Deviations can be corrected in a timely manner without manual intervention, ensuring that the scheduling accuracy is always within a reasonable range, guaranteeing the stability and accuracy of system scheduling, and meeting the core requirements of dynamic optimization.
[0066] like Figure 1-2 As shown:
[0067] It also includes a manual intervention module, which works in two directions with the scheduling execution module and the AI model module. When the scheduling plan output by the AI model module or the abnormal risk marked by the feedback iteration module exceeds the preset range, it can automatically trigger the manual review and parameter adjustment process. In this solution, the two-way linkage design ensures that humans can obtain scheduling plans and abnormal information in real time, and can synchronize adjustment instructions to each module, achieving dual protection of automatic AI optimization and manual bottom-line control. When the plan or risk exceeds the preset range, it automatically triggers manual review, accurately identifies extreme scenarios that AI cannot handle, and uses human experience to make up for the limitations of AI, avoids the implementation of unreasonable scheduling plans, ensures that adjustment instructions are implemented quickly, restores normal system scheduling, ensures the stability and reliability of scheduling, and adapts to the high-standard requirements of refined supply chain management.
[0068] Example 2: This example addresses the supply chain scheduling scenario for small and medium-sized commercial enterprises. It establishes an AI-driven dynamic optimization scheduling system for the commercial supply chain. The system is deployed on Alibaba Cloud ECS servers, using TensorFlow 2.8 as the AI model training framework to achieve dynamic scheduling throughout the entire supply chain process. The specific implementation is as follows:
[0069] Data Module Implementation: The data module collects data from all nodes in the supply chain, including 3 upstream suppliers, 2 warehousing nodes, 15 terminal stores, and 2 transportation service providers. Data collection is conducted hourly and includes SKU attribute data, inventory dynamic data, and transportation status data. The data preprocessing unit uses Z-score normalization to standardize the raw data, eliminating the influence of dimensions. The mathematical formula is as follows: ,in This is the original data. This is the mean of this type of data. The standard deviation of this type of data is used to transform all original data into standardized data with a mean of 0 and a variance of 1 using this formula. Simultaneously, through 3... The criteria remove outlier data and extract four types of features, including SKU core attribute features and inventory turnover features. A 128-dimensional structured feature vector is constructed and stored in a MySQL 8.0 database to provide data support for the AI model module.
[0070] AI Model Module Implementation: The AI model module integrates four sub-modules, all deployed based on the existing server framework. The federated learning module uses the FedAvg federated averaging algorithm, with three lightweight nodes corresponding to the supplier, warehouse, and terminal store respectively. Model parameters are trained locally on each node, and secure data aggregation is achieved through parameter aggregation. The parameter aggregation formula is as follows: ,in For global aggregation parameters, For the first Data volume per node This represents the total data volume of all nodes. For the first The local training parameters of each node are used to aggregate available but invisible data through this formula, breaking down data silos. The basic AI scheduling model of the transfer learning module is a lightweight hybrid model of XGBoost and Transformer. XGBoost is set with 100 decision trees and a learning rate of 0.1. Transformer is set with 2 encoder layers and 64 hidden layer dimensions. This model is pre-trained using 10,000 mature SKU scheduling data accumulated by the data module. For new SKUs, feature transfer and local parameter fine-tuning mode are adopted. The fine-tuning loss function is:
[0071] ,in This is the actual scheduling value. To predict scheduling values for the model, This is the regularization coefficient (value 0.01). As a parameter regularization term, by minimizing this loss function, the top 20% of the model parameters are fine-tuned to achieve rapid adaptation to new products, with a fine-tuning cycle of no more than 2 days;
[0072] The digital twin module uses Unity 3D to build a virtual scenario of the entire supply chain, recreating the entire process of warehousing, outbound delivery, and transportation. It integrates real-time dynamic data from the data module and calculates scheduling errors using the mean square error formula:
[0073] ,in For virtual rehearsal scheduling values, To preset the target value, For the data sample size, when When the value is greater than 0.05, it is marked as a risk point and fed back to the transfer learning module for optimization;
[0074] The reinforcement learning module employs the DQN algorithm, with the reward function being the maximization of supply chain scheduling efficiency and the minimization of scheduling error. The reward function formula is as follows: ,in Efficiency weight (value 0.6). This represents the scheduling efficiency value. This is the error weight (value 0.4). To mitigate scheduling errors, the reward function is used to adjust parameters such as capacity allocation ratio and replenishment cycle in real time, achieving dynamic adaptive optimization. The learning rate is set to 0.001, and the model parameters are updated once per hour.
[0075] The scheduling execution module is divided into four units, all of which communicate bidirectionally with the AI model module. After receiving the optimized scheduling scheme output by the model, the warehouse scheduling unit executes the outbound operation according to the outbound priority, the transportation capacity scheduling unit allocates transportation vehicles according to the transportation capacity allocation ratio output by the model (with an error of no more than 5%), the cross-warehouse transfer unit executes the transfer operation according to the dynamic inventory data, and the terminal distribution unit completes the terminal delivery according to the distribution rhythm. Each unit will feed back the real-time execution data to the feedback iteration module every 30 minutes.
[0076] Feedback Iteration Module Implementation: The error calculation unit uses the absolute error formula to calculate the actual scheduling error. The error formula is as follows: The preset error threshold is 0.03. When the error value is greater than 0.03, the feedback transmission unit automatically feeds back the error data and abnormal information to the AI model module, triggering the parameter tuning process of the reinforcement learning module. After tuning, the scheduling scheme is re-output to ensure that the scheduling accuracy meets the business requirements.
[0077] Manual intervention module implementation: When the capacity allocation deviation in the scheduling plan output by the AI model module exceeds 8%, or when the risk points marked by the feedback iteration module appear more than 3 times consecutively, the manual review process is automatically triggered. After manually adjusting the parameters, the data is synchronously sent back to each module to ensure the stable operation of the system.
[0078] This embodiment has been tested and found that the system scheduling error can be controlled within 3%, the new product adaptation cycle is shortened to 1.5 days, and the supply chain scheduling efficiency is improved by more than 25%. It can be stably applied to the supply chain scheduling scenarios of small and medium-sized commercial enterprises.
[0079] In the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.
[0080] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0081] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.
Claims
1. An AI-driven dynamic optimization and scheduling system for commercial supply chains, characterized in that, It includes a data module, an AI model module, a scheduling and execution module, and a feedback and iteration module; The data module is used to collect, store, and preprocess data information from all nodes of the commercial supply chain, focusing on achieving secure aggregation and standardized processing of data from various entities in the supply chain, providing reliable data support for subsequent AI model calculations; The AI model module integrates federated learning, transfer learning, digital twin, and reinforcement learning modules. This module takes the standardized data output by the data module as input to complete model training, scheduling scenario simulation, and precise tuning of scheduling parameters. The scheduling and execution module is responsible for receiving the optimized scheduling plan output by the AI model module, and specifically executing the scheduling operations of the entire supply chain, such as warehouse outbound, transportation capacity allocation, cross-warehouse transfer and terminal distribution, to ensure that the scheduling plan is implemented. The feedback iteration module collects the actual operating data of the scheduling execution module in real time, accurately calculates the scheduling error, and promptly feeds the error information back to the AI model module, driving the continuous iteration and optimization of the model and scheduling scheme.
2. The AI-driven dynamic optimization and scheduling system for the commercial supply chain according to claim 1, characterized in that, The data module includes a multi-entity data acquisition unit, a data preprocessing unit, and a data security storage unit; The multi-entity data collection unit comprehensively covers upstream suppliers, warehousing nodes, terminal stores, and cooperative transportation service providers in the commercial supply chain. The types of data collected include SKU attribute data, inventory dynamic data, transportation status data, terminal demand data, and scheduling execution feedback data, comprehensively covering all aspects of the supply chain.
3. The AI-driven dynamic optimization and scheduling system for the commercial supply chain according to claim 2, characterized in that, The data preprocessing unit cleans, denoises, standardizes, and extracts features from the collected raw data, converting unstructured data into structured feature vectors that can be directly used for AI calculations. The feature vectors include SKU core attribute features, inventory turnover features, transportation capacity matching features, and terminal demand fluctuation features, which can accurately adapt to the input requirements of the AI model module and ensure the accuracy of model calculations.
4. The AI-driven dynamic optimization and scheduling system for the commercial supply chain according to claim 1, characterized in that, The federated learning module in the AI model module adopts a lightweight node deployment mode and works in collaboration with the data module. It aggregates feature data from multiple nodes across the entire supply chain through federated parameter interaction.
5. The AI-driven dynamic optimization and scheduling system for the commercial supply chain according to claim 1, characterized in that, The transfer learning module in the AI model module works based on the pre-trained basic AI scheduling model of the system. The basic AI scheduling model is a hybrid model of lightweight XGBoost and Transformer. Its training data comes from the full set of mature SKU scheduling data accumulated by the data module over a long period of time and the multi-node data aggregated by the federated learning module.
6. The AI-driven dynamic optimization and scheduling system for the commercial supply chain according to claim 1, characterized in that, The digital twin module in the AI model module is used to build a digital twin scenario for the entire process of commercial supply chain scheduling, which restores the entire process of warehousing and outbound, transportation, terminal distribution and inventory turnover. Its access data module collects dynamic supply chain data in real time, previews the execution effect of scheduling plans, quantifies scheduling errors, and accurately marks abnormal risk points in the scheduling process.
7. The AI-driven dynamic optimization and scheduling system for the commercial supply chain according to claim 1, characterized in that, The reinforcement learning module in the AI model module takes maximizing supply chain scheduling efficiency and minimizing scheduling error as its core reward function. Its input data includes simulated data output by the digital twin module and actual scheduling data output by the feedback iteration module.
8. The AI-driven dynamic optimization and scheduling system for the commercial supply chain according to claim 1, characterized in that, The scheduling and execution module includes a warehouse scheduling unit, a transportation scheduling unit, a cross-warehouse allocation unit, and a terminal distribution scheduling unit. All of them communicate bidirectionally with the AI model module and the feedback iteration module, accurately receive the optimization scheduling instructions issued by the AI model module, and promptly feed back their own real-time execution data to the feedback iteration module.
9. The AI-driven dynamic optimization and scheduling system for the commercial supply chain according to claim 1, characterized in that, The error calculation unit is used to accurately calculate the deviation between the actual running data of the scheduling execution module and the preset target data of the AI model module; When the actual scheduling error exceeds the preset threshold, the feedback transmission unit automatically feeds back the error data and related abnormal information to the AI model module.
10. The AI-driven dynamic optimization and scheduling system for the commercial supply chain according to claim 1, characterized in that, It also includes a manual intervention module, which works in two directions with the scheduling and execution module and the AI model module. When the scheduling scheme output by the AI model module or the abnormal risk marked by the feedback iteration module exceeds the preset range, it can automatically trigger the manual review and parameter adjustment process.