A blank intelligent warehousing and scheduling management method and device

Abstract

The application relates to the computer technical field, and in particular to a blank intelligent warehousing and scheduling management method and device, which comprises the following steps: collecting blank data, warehousing environment data and operation equipment data in real time, and constructing a warehousing panoramic digital twin model; based on the warehousing panoramic digital twin model, combining order information and supply chain data, and generating an optimal task allocation scheme, an inventory adjustment scheme and replenishment suggestions and an optimal path planning result through a reinforcement learning algorithm; generating task scheduling instructions, inventory adjustment instructions and equipment path planning instructions through the warehousing panoramic digital twin model; adopting a collaborative control architecture of a centralized global coordination end and a distributed equipment end, realizing multi-device collaborative operation, and completing blank carrying and storage operation; and based on blank warehousing whole-process management data, realizing blank data whole-life cycle tracing through a block chain technology. The application helps to improve the intelligentization and high efficiency level of blank warehousing management.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method and apparatus for intelligent storage and scheduling management of mold blanks. Background Technology

[0002] As the core basic component of mold manufacturing, mold blanks are characterized by diverse specifications, large weight differences (from a few kilograms to tens of tons), high value, frequent turnover, and high maintenance requirements. Their warehousing and scheduling management directly affects mold production efficiency, delivery cycle, and enterprise operating costs.

[0003] Traditional mold blank storage management relies heavily on manual recording and experience-based scheduling, which has many drawbacks: First, scheduling efficiency is low, with manual mold blank locating taking an average of several minutes, which can easily lead to production stoppages during emergency mold changes; second, information traceability is difficult, with mold blank usage status, maintenance records, and location information being managed in a scattered manner, resulting in frequent discrepancies between records and actual inventory; third, coordination is poor, with warehousing and production scheduling disconnected, easily leading to resource waste problems such as "urgently needed mold blanks not being found, and idle mold blanks occupying space".

[0004] Therefore, there is an urgent need for an intelligent warehousing and scheduling management method to solve the above-mentioned technical pain points and improve the intelligence and efficiency of mold blank warehousing management. Summary of the Invention

[0005] Therefore, it is necessary to provide a method and device for intelligent storage and scheduling management of mold blanks that can improve the intelligence and efficiency of mold blank storage management, in order to address the above-mentioned technical problems.

[0006] In a first aspect, this application provides a method for intelligent storage and scheduling management of mold blanks, the method comprising: Real-time data collection of mold base data, storage environment data, and operating equipment data; and construction of a panoramic digital twin model of the storage area based on the mold base data, storage environment data, and operating equipment data. Obtain order information and supply chain data, and based on the warehouse panoramic digital twin model, combine the order information and supply chain data to generate the optimal task allocation scheme, inventory adjustment scheme and replenishment suggestion, as well as the optimal path planning result through reinforcement learning algorithm; Based on the optimal task allocation scheme, the inventory adjustment scheme and replenishment suggestion, and the optimal path planning result, task scheduling instructions, inventory adjustment instructions and equipment path planning instructions are generated through the warehouse panoramic digital twin model; Based on the task scheduling instructions, the inventory adjustment instructions, and the equipment path planning instructions, a collaborative control architecture of centralized global coordination terminal and distributed equipment terminal is adopted to realize multi-device collaborative operation and complete the handling and storage of mold blanks. Based on the full-process management data of mold blank storage, blockchain technology is used to achieve full lifecycle traceability of mold blank data.

[0007] In one embodiment, the step of generating task scheduling instructions, inventory adjustment instructions, and equipment path planning instructions through the warehouse panoramic digital twin model based on the optimal task allocation scheme, the inventory adjustment scheme and replenishment suggestion, and the optimal path planning result includes: The optimal task allocation scheme, the inventory adjustment scheme and replenishment suggestion, and the optimal path planning result are loaded and connected to the real-time warehouse dynamic data stream to provide real-time context for instruction generation; The optimal task allocation scheme, inventory adjustment scheme, replenishment suggestion and optimal path planning results after loading are integrated and analyzed. Potential conflicting operations are identified through preset conflict detection rules and resolved based on dynamic priority strategy to generate a conflict-free primary instruction set. The primary instruction set is input into the warehouse panoramic digital twin model for simulation. Based on the simulation results, the time parameters and resource allocation of the instructions are fine-tuned to generate intermediate instructions. The intermediate-level instructions are converted into specific task scheduling instructions, inventory adjustment instructions, and equipment path planning instructions.

[0008] In one embodiment, the step of inputting the primary instruction set into the warehouse panoramic digital twin model for simulation, and fine-tuning the timing parameters and resource allocation of the instructions based on the simulation results to generate intermediate instructions includes: Based on the aforementioned panoramic digital twin model of the warehouse, an isolated simulation sandbox is generated, which takes the initial instruction set, real-time equipment status data, and warehouse layout data as initial inputs. In the simulation sandbox, each working device is modeled as an intelligent agent with autonomous decision-making capabilities, simulating the process of the intelligent agent cooperating according to the primary instruction set, and simultaneously collecting multi-source simulation data; Based on the multi-source simulation data, the critical path delay in time parameters and the hot spots of competition in resource allocation are identified. Based on the critical path delay in time parameters and the hot spots of competition in resource allocation, the time parameters of the instructions are dynamically adjusted, and the resource allocation is fine-tuned to generate intermediate instructions.

[0009] In one embodiment, the collaborative control architecture of a centralized global coordination terminal and distributed device terminals, based on the task scheduling instruction, the inventory adjustment instruction, and the equipment path planning instruction, enables multi-device collaborative operation to complete the handling and storage of mold blanks, including: Establish a collaborative control architecture between a centralized global coordination terminal and distributed device terminals. The global coordination terminal issues instructions to each device terminal to perform global operation status management and cross-device resource allocation. After receiving the global instructions, each device autonomously negotiates with the work area through point-to-point communication, determines local operation avoidance, resource occupation and process connection strategies, generates a local collaborative solution that meets global constraints and executes it autonomously, so as to realize multi-device collaborative operation and complete the handling and storage of mold blanks. When a local emergency occurs during operation, the equipment will first take local emergency measures and simultaneously report to the global coordination terminal. The global coordination terminal will then dynamically decide whether to optimize the global plan and issue adjustment instructions based on the scope of the emergency.

[0010] In one embodiment, the step of acquiring order information and supply chain data, and based on the warehouse panoramic digital twin model, combining the order information and the supply chain data, generating an optimal task allocation scheme, inventory adjustment scheme and replenishment suggestion, and optimal path planning result through a reinforcement learning algorithm includes: Obtain order information and supply chain data, integrate the order information and supply chain data with the real-time data of the warehouse panoramic digital twin model and perform preprocessing to construct a standardized dataset; Based on the aforementioned warehouse panoramic digital twin model, a Markov decision environment based on reinforcement learning is built, the decision rule elements are clarified, and a multi-objective weighted reward function is designed. The multi-objective weighted reward function includes positive rewards associated with core indicators of operations and supply chain, negative penalties associated with abnormal situations in operations and supply chain, and gradient auxiliary rewards for intermediate decisions that approach the optimal solution. A centralized training and decentralized execution architecture is adopted to build a multi-agent deep reinforcement learning model, and the offline training of the model is completed by combining a digital twin simulation sandbox. Based on the decision rule elements and the multi-objective weighted reward function, the standardized dataset is input into the trained multi-agent deep reinforcement learning model to generate the optimal task allocation scheme, inventory adjustment scheme and replenishment suggestion, as well as the optimal path planning result.

[0011] In one embodiment, the step of inputting the standardized dataset into a multi-agent deep reinforcement learning model trained based on the decision rule elements and the multi-objective weighted reward function to generate an optimal task allocation scheme, an inventory adjustment scheme and replenishment suggestion, and an optimal path planning result includes: The standardized dataset is divided into a training sample set and a validation sample set according to a preset ratio; The training sample set is input into the trained multi-agent deep reinforcement learning model, and the inference framework is built in combination with the decision rule elements. With the multi-objective weighted reward function as the optimization guide, multiple rounds of online inference iteration are carried out to generate multiple sets of candidate task allocation, inventory adjustment and replenishment and path planning schemes. Based on the verification sample set, the effectiveness of multiple candidate solutions is verified, and the candidate solutions are quantitatively scored based on the core evaluation indicators. Based on the scoring results, the optimal task allocation plan, inventory adjustment plan and replenishment suggestions, as well as the optimal route planning results are obtained.

[0012] In one embodiment, obtaining the optimal task allocation scheme, inventory adjustment scheme and replenishment suggestion, and optimal route planning result based on the scoring results includes: Based on the scoring results, the candidate solution with the best score is selected as the basic solution. The basic solution is then input into the warehouse panoramic digital twin model for real-world adaptation simulation. Combined with real-time dynamic data of the warehouse, the key parameters in the basic solution are adjusted for adaptability. The revised plan undergoes a second conflict check. Once the conflict check passes, it is determined as the final optimal task allocation plan, inventory adjustment plan, replenishment suggestion, and optimal path planning result.

[0013] In one embodiment, the step of using blockchain technology to achieve full lifecycle traceability of mold data based on the full-process management data of mold storage includes: Data from the entire process of mold blank storage is collected and structured in stages, and a unique on-chain identifier is assigned to each mold blank. The processed mold blank storage full-process control data is encapsulated in blocks, and after overlaying timestamps and operation node information, hash encryption is completed to generate an immutable on-chain encrypted data unit. Build a dedicated alliance chain for mold storage, synchronize encrypted data units to authorized nodes in warehousing, supply chain, and operation execution, and complete multi-node data verification and on-chain notarization through a consensus mechanism to achieve distributed data storage; A linkage and traceability mechanism is constructed between blockchain and the warehousing panoramic digital twin model. On-chain data is retrieved and synchronously mapped to the entire process operation scenario of the mold blank in the warehousing panoramic digital twin model to form a visual traceability link, so as to realize the full life cycle traceability of the mold blank data.

[0014] Secondly, this application also provides an intelligent storage and scheduling management device for mold blanks. The device includes: The digital twin model building module is used to collect mold base data, warehousing environment data, and operating equipment data in real time, and to build a panoramic digital twin model of the warehouse based on the mold base data, the warehousing environment data, and the operating equipment data. The solution generation module is used to acquire order information and supply chain data, and based on the warehouse panoramic digital twin model, combined with the order information and supply chain data, to generate the optimal task allocation plan, inventory adjustment plan and replenishment suggestion, as well as the optimal path planning result through reinforcement learning algorithm; The instruction generation module is used to generate task scheduling instructions, inventory adjustment instructions, and equipment path planning instructions based on the optimal task allocation scheme, the inventory adjustment scheme and replenishment suggestions, and the optimal path planning results, through the warehouse panoramic digital twin model. The equipment collaborative operation module is used to realize multi-device collaborative operation and complete the handling and storage of mold blanks based on the task scheduling instructions, the inventory adjustment instructions and the equipment path planning instructions, using a collaborative control architecture of centralized global coordination terminal and distributed device terminal. The lifecycle traceability module is used to achieve full lifecycle traceability of mold data based on the data management of the entire mold storage process, using blockchain technology.

[0015] In summary, this application includes the following beneficial technical effects: A panoramic digital twin model of the warehouse was constructed, achieving precise mapping and real-time synchronization of the physical warehouse scene and the virtual model. This breaks through the technical bottlenecks of scattered data collection and lagging scene perception in traditional mold warehousing, enabling a comprehensive and dynamic presentation of the overall warehouse operation status. Based on the panoramic digital twin model, combined with order information and supply chain data, and through reinforcement learning algorithms, optimal task allocation schemes, inventory adjustment schemes, replenishment suggestions, and optimal path planning results are generated. This achieves precise matching between warehouse operation schemes and order demands and the supply chain, effectively improving the rationality of task allocation, the accuracy of inventory control, the adaptability of replenishment strategies, and the optimization of equipment paths, thus achieving optimized allocation of warehouse resources. Based on the generated instructions, a collaborative control architecture of a centralized global coordination terminal and distributed device terminals is adopted to complete mold handling and storage operations, achieving efficient collaborative operation of multiple devices and significantly improving the efficiency of mold handling and storage operations. Through blockchain technology, an intelligent management system for the entire mold warehousing process is formed, thereby improving the intelligence and efficiency of mold warehousing management. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the intelligent warehousing and scheduling management method for mold blanks in one embodiment; Figure 2This is a flowchart illustrating the intelligent warehousing and scheduling management method for mold blanks in another embodiment; Figure 3 This is a structural block diagram of a mold intelligent storage and scheduling management device in one embodiment. Detailed Implementation

[0017] This invention provides a method and apparatus for intelligent storage and scheduling management of mold blanks.

[0018] The embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0019] In the description of the embodiments disclosed in this invention, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0020] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the intelligent warehousing and scheduling management method for mold blanks in this invention includes: The S100 collects mold base data, storage environment data, and operating equipment data in real time, and builds a panoramic digital twin model of the warehouse based on the mold base data, storage environment data, and operating equipment data.

[0021] Specifically, a multi-source heterogeneous sensing network is adopted, integrating high-precision IoT sensors, visual recognition devices, equipment IoT interfaces, and RFID radio frequency identification technology to collect real-time data from mold blanks, storage environment, and operating equipment in a comprehensive and seamless manner. Mold blank data includes basic attributes and dynamic management data such as mold blank model, specifications, material, quantity, storage location, warehousing time, quality inspection information, and maintenance records. Storage environment data includes environmental monitoring data such as temperature, humidity, dust concentration, space occupancy rate, and ventilation status within the warehouse. Operating equipment data includes real-time operational data such as the operating status, operation progress, location information, fault warnings, and energy consumption data of storage operation equipment such as stacker cranes, AGVs, and conveyors. The system intelligently cleans, normalizes, and fuses the collected multi-source heterogeneous data across spatiotemporal dimensions, eliminating data redundancy, errors, and heterogeneity. This provides standardized, high-fidelity data support for digital twin modeling. Breaking through the limitations of traditional single-dimensional mapping in digital twins, it employs multi-scale modeling and bidirectional virtual-physical interaction technology to deeply couple the spatial structure, equipment layout, and mold storage status of the physical warehouse with the fused, multi-dimensional dynamic data. This constructs a panoramic digital twin model of the warehouse, achieving a 1:1 accurate geometric mapping between the physical warehouse and the virtual model, real-time status synchronization, dynamic behavior simulation, and bidirectional virtual-physical driving. This model not only presents the overall operational status of the warehouse comprehensively and dynamically but also enables advance prediction and simulation of mold storage status, equipment operating trends, and changes in the warehouse environment based on real-time data streams. It also supports remote interaction and virtual debugging of the physical warehouse scene, overcoming the technical bottlenecks of separation between physical scene and data management, lagging status perception, and disconnect between simulation and reality in traditional warehouse management. This achieves an intelligent upgrade of warehousing from passive perception to proactive prediction and from static mapping to dynamic interaction.

[0022] S200 acquires order information and supply chain data, and based on the warehouse panoramic digital twin model, combines order information and supply chain data to generate the optimal task allocation plan, inventory adjustment plan and replenishment suggestions, as well as the optimal path planning results through reinforcement learning algorithms.

[0023] Specifically, by connecting the warehouse management system with the enterprise production management system and supply chain collaboration platform, real-time order information and supply chain-related data for mold blanks are obtained. The obtained order information and supply chain data are then integrated with real-time warehouse operation data in the warehouse panoramic digital twin model. Based on the warehouse panoramic digital twin model, a reinforcement learning algorithm operation and optimization environment is built. Warehouse operation efficiency, inventory turnover rate, supply chain response speed, and equipment resource utilization rate are taken as core optimization objectives. The reinforcement learning algorithm is used to analyze, train, and iteratively optimize the integrated multi-dimensional data. For core warehouse management issues such as task allocation, inventory structure adjustment, mold blank replenishment strategy, and equipment travel path planning, the optimal task allocation scheme, inventory adjustment scheme, replenishment suggestion, and optimal path planning results are generated, which are adapted to the actual warehouse operation status, order demand, and supply chain rhythm.

[0024] The S300 generates task scheduling instructions, inventory adjustment instructions, and equipment path planning instructions through a warehouse panoramic digital twin model, based on the optimal task allocation scheme, inventory adjustment scheme, replenishment suggestions, and optimal path planning results.

[0025] Specifically, the optimal task allocation scheme, inventory adjustment scheme, replenishment suggestion, and optimal path planning results generated by the reinforcement learning algorithm are imported into the warehouse panoramic digital twin model. Simultaneously, the model maintains connectivity with the real-time dynamic data flow of the warehouse, providing a real-time and comprehensive warehouse scenario context for instruction generation. Within the warehouse panoramic digital twin model, various optimal schemes are decomposed, transformed, and quantified. Combining the actual operational capabilities of warehouse equipment, the spatial attributes of warehouse locations, and the storage requirements of different mold blanks, abstract scheme requirements are transformed into standardized instructions that can be directly executed by warehouse equipment and the management system. These include task scheduling instructions for warehouse operations, inventory adjustment instructions for mold blank inventory management, and equipment path planning instructions for mobile equipment such as AGVs and stacker cranes, ensuring that the generated instructions are highly adapted to the actual warehouse operation scenario.

[0026] The S400, based on task scheduling instructions, inventory adjustment instructions, and equipment path planning instructions, adopts a collaborative control architecture of centralized global coordination terminal and distributed equipment terminal to realize multi-device collaborative operation and complete the handling and storage of mold blanks.

[0027] Specifically, a collaborative control architecture is established, consisting of a centralized global coordination terminal and distributed device terminals. The centralized global coordination terminal serves as the global operation control center for the preform storage, responsible for receiving and issuing task scheduling instructions, inventory scheduling instructions, and equipment path planning instructions to each distributed device terminal. Simultaneously, it monitors the overall operational status of the storage facility in real time, coordinating the allocation of storage operation resources across equipment and regions, thus achieving global coordination and control over preform storage operations. The distributed device terminals are local control terminals for each storage operation equipment, such as stacker cranes and AGV transport vehicles. Each device terminal independently receives various execution instructions from the global coordination terminal and possesses autonomous decision-making and inter-device negotiation capabilities. After receiving instructions, each device terminal autonomously negotiates with other devices within the same work area through point-to-point communication, determining collaborative strategies such as local operation avoidance, resource allocation order, and process connection nodes. This generates local collaborative operation plans that meet global operational constraints and executes them autonomously, enabling collaborative operation of multiple devices. During the handling and storage of mold blanks, if the equipment encounters local emergencies such as equipment failure, path obstruction, or abnormal mold blank status, it will first carry out local autonomous emergency response and simultaneously report the relevant information of the emergency to the centralized global coordination terminal in real time. The global coordination terminal will dynamically decide whether to optimize and adjust the global operation plan and instructions based on the scope and severity of the emergency, and promptly issue new adjustment instructions to the relevant equipment terminals to ensure the continuity and stability of warehousing operations, and ultimately complete the entire process of mold blank handling and storage efficiently.

[0028] S500 is based on the full-process management data of mold blank storage and uses blockchain technology to achieve full life cycle traceability of mold blank data.

[0029] Specifically, the process begins by integrating data from the entire process of mold blank warehousing. This data covers operational data for all stages of mold blank handling, from inbound verification and storage to transfer and outbound handover. It also includes basic mold blank attributes, warehousing environment monitoring data, operational equipment execution data, and business data related to orders and the supply chain—all relevant data across the entire mold blank warehousing chain. This achieves integrated data aggregation across all warehousing stages, forming a complete mold blank warehousing data chain. During data traceability, all relevant data for the entire warehousing process can be quickly retrieved from this data chain using the mold blank's unique identifier. This clearly reconstructs the mold blank's status changes, operational procedures, environmental impacts, and business relationships at each stage of warehousing, ensuring that the entire lifecycle data of the mold blank—from inbound to outbound—is verifiable, traceable, and verifiable, creating a complete closed loop for mold blank warehousing data traceability.

[0030] In one embodiment, such as Figure 2 As shown, S300 includes: S310 loads the optimal task allocation scheme, inventory adjustment scheme and replenishment suggestions, as well as the optimal path planning results, and connects to the real-time warehouse dynamic data stream to provide real-time context for instruction generation; S320 integrates and analyzes the optimal task allocation scheme, inventory adjustment scheme, replenishment suggestion, and optimal path planning results after loading. It identifies potential conflicting operations through preset conflict detection rules and resolves them based on dynamic priority strategy to generate a conflict-free primary instruction set. S330 inputs the primary instruction set into the warehouse panoramic digital twin model for simulation, and fine-tunes the time parameters and resource allocation of the instructions based on the simulation results to generate intermediate instructions; S340 converts intermediate instructions into specific task scheduling instructions, inventory adjustment instructions, and equipment path planning instructions.

[0031] Specifically, the optimal task allocation plan, inventory adjustment plan, replenishment suggestions, and optimal path planning results are first fully loaded into the instruction generation module of the warehouse panoramic digital twin model. At the same time, the real-time dynamic data stream of the warehouse is continuously connected, and the real-time on-site information such as the real-time operating status of equipment in the warehouse, the dynamic occupancy of storage locations, and the personalized storage needs of mold blanks are used as the dynamic context for instruction generation, so that the instruction generation is closely aligned with and highly adapted to the actual operating status of the warehouse. Subsequently, a multi-dimensional fusion analysis was conducted on various optimal solutions after loading. Based on preset conflict detection rules, potential conflicting operations, such as overlapping equipment operation paths, simultaneous occupation of warehouse resources, and conflicting task execution sequences, were comprehensively and accurately identified. For the identified potential conflicts, a multi-dimensional dynamic priority strategy was constructed. This strategy does not use a fixed priority ranking, but rather combines core dimensions such as the urgency of order delivery, the importance of mold blank operation, equipment operation efficiency, and supply chain response requirements to assign a dynamic priority score to each task. At the same time, adaptation rules for conflict resolution were set. Specifically, for resource occupation conflicts, warehouse resources corresponding to high-priority tasks were allocated first, and the resource occupation time or resource type of low-priority tasks was adjusted. For spatiotemporal trajectory conflicts, the equipment operation paths and execution times of high-priority tasks were prioritized, and the equipment paths of low-priority tasks were replanned or the execution sequence was postponed. For process connection conflicts, the execution order of work processes was reorganized according to preset operation logic and task priorities to ensure smooth process connection of high-priority tasks. Based on this dynamic prioritization strategy, various potential conflicts are addressed and optimized in a targeted and differentiated manner. The content of the solutions involved in the conflicts is reasonably adjusted, ultimately generating a primary instruction set with no conflicts in tasks, no competition in resource allocation, no collisions in spatiotemporal trajectories, and no contradictions in process connections. This lays a conflict-free foundation for the simulation optimization and implementation of subsequent instructions. Next, the primary instruction set is input into the warehouse panoramic digital twin model for full-process simulation. Based on the feedback from the model simulation, such as equipment operation connections, resource utilization efficiency, and task execution rhythm, the time parameters and resource allocation content in the primary instruction set, such as operation start time, equipment operation duration, and resource allocation ratio, are finely adjusted to generate intermediate instructions with stronger adaptability and feasibility. Finally, according to the communication protocols of the various warehouse operation equipment and the instruction execution format of the warehouse management system, the intermediate instructions are standardized and converted to generate specific task scheduling instructions, inventory adjustment instructions, and equipment path planning instructions that can be directly recognized and executed by the warehouse operation equipment and management system.

[0032] In one embodiment, a primary instruction set is input into a warehouse panoramic digital twin model for simulation. Based on the simulation results, the timing parameters and resource allocation of the instructions are fine-tuned to generate intermediate instructions, including: Based on a panoramic digital twin model of the warehouse, an isolated simulation sandbox is generated. The simulation sandbox uses a primary instruction set, real-time equipment status data, and warehouse layout data as initial inputs. In the simulation sandbox, each operating device is modeled as an intelligent agent with autonomous decision-making capabilities, simulating the process of the intelligent agent cooperating according to the primary instruction set, and simultaneously collecting multi-source simulation data. Based on the multi-source simulation data, the critical path delay in time parameters and the competition hotspots in resource allocation are identified. Based on the critical path delay in time parameters and the competition hotspots in resource allocation, the time parameters of the instructions are dynamically adjusted, and the resource allocation is fine-tuned to generate intermediate instructions.

[0033] Specifically, based on a primary instruction set, an isolated simulation sandbox is constructed using a panoramic digital twin model of the warehouse. Within the simulation sandbox, each warehouse operation equipment is modeled as an independent intelligent agent, endowing each intelligent agent with autonomous decision-making capabilities for operational capability recognition, local operation negotiation, and basic fault handling. The primary instruction set, real-time equipment status data, and warehouse layout data are used as inputs to construct a high-fidelity, interference-free simulation execution environment within the simulation sandbox. This simulates the complete virtual process of multiple intelligent agent devices cooperating according to the primary instruction set, and simultaneously collects equipment operation timing data, resource utilization distribution data, task completion progress data, and interaction behavior data between devices during the simulation execution. A multi-source simulation dataset is constructed. Feature extraction and time-series analysis are performed on the multi-source simulation dataset to accurately identify critical path delay nodes and specific delay durations in the time parameter dimension, as well as equipment competition hotspots and storage location conflict areas in the resource allocation dimension. With the core optimization goal of eliminating path delays and resolving resource competition, the execution start time and operation window of the corresponding tasks in the primary instruction set are dynamically adjusted. At the same time, the resource allocation scheme is finely tuned by time-sharing and resource redundancy allocation, generating intermediate instructions that meet the timing requirements and avoid resource conflicts. This ensures that the instructions complete the virtual simulation and optimization of the entire process before actual implementation, greatly improving the feasibility and execution efficiency of the instructions.

[0034] In one embodiment, based on task scheduling instructions, inventory adjustment instructions, and equipment path planning instructions, a collaborative control architecture combining a centralized global coordination terminal and distributed device terminals is adopted to achieve multi-device collaborative operation and complete the handling and storage of mold blanks, including: A collaborative control architecture is established, consisting of a centralized global coordination terminal and distributed device terminals. The global coordination terminal issues instructions to each device terminal to manage the overall operational status and allocate resources across devices. After receiving the global instructions, each device terminal conducts autonomous negotiation within its own work area through point-to-point communication to determine strategies for local operation avoidance, resource occupation, and process connection. It then generates a local collaborative solution that meets global constraints and executes it autonomously to achieve multi-device collaborative operation and complete the handling and storage of the mold blank. When a device terminal encounters a local emergency during operation, it first performs local autonomous emergency response and simultaneously reports to the global coordination terminal. The global coordination terminal then dynamically decides whether to optimize the global solution and issues adjustment instructions based on the scope of the emergency's impact.

[0035] Specifically, a collaborative control architecture is first established, consisting of a centralized global coordination terminal and distributed device terminals. The centralized global coordination terminal serves as the core control hub for mold storage operations, equipped with a global monitoring module based on a panoramic digital twin model of the storage area. It can receive task scheduling, inventory adjustment, and equipment path planning instructions in real time, and accurately break down and issue these instructions. Each device terminal has independent instruction reception, data acquisition, and autonomous negotiation capabilities. After receiving instructions from the global coordination terminal, it can complete autonomous negotiation of all related devices within the same work area without relying on secondary scheduling by the global terminal, through point-to-point direct communication between devices. During the negotiation process, each device terminal shares information such as its own work tasks, execution sequence, and resource requirements. Combined with the actual storage operation scenario, they jointly determine the priority of local operation avoidance, storage resource occupation rules, and seamless connection nodes between various processes, generating a collaborative solution that meets both global instruction constraints and adapts to the local operation site. Each device terminal then autonomously executes this solution, achieving decentralized collaborative operation of multiple devices and efficiently advancing the entire process of mold handling, storage, and retrieval. Throughout the collaborative operation of multiple devices, each distributed device will collect its own operating status, operation execution status, and on-site environmental information in real time. When encountering local emergencies such as equipment failure, obstructed operation paths, abnormal mold status, or changes in storage space occupancy, the local autonomous emergency response mechanism will be triggered immediately. According to the preset emergency response rules, measures such as operation suspension, local path replanning, preliminary fault self-check, and temporary resource replacement will be automatically taken to control the impact of the emergency as soon as possible and avoid the overall operation from being halted due to a problem with a single device. At the same time, the device will report information such as the type of emergency, location of occurrence, degree of impact, and measures already taken to the global coordination terminal in real time. After receiving the information, the global coordination terminal will simulate and deduce the impact of the emergency on the overall operation through the warehouse panoramic digital twin model and dynamically decide whether to optimize and adjust the original global plan and instructions.

[0036] In one embodiment, order information and supply chain data are acquired, and based on a warehousing panoramic digital twin model, combined with the order information and supply chain data, an optimal task allocation plan, inventory adjustment plan and replenishment suggestions, and optimal path planning results are generated through reinforcement learning algorithms, including: The process involves acquiring order information and supply chain data, fusing them with real-time data from a warehouse panoramic digital twin model, preprocessing the data, and constructing a standardized dataset. Based on the warehouse panoramic digital twin model, a Markov decision-making environment using reinforcement learning is built, defining the elements of the decision rules and designing a multi-objective weighted reward function. This function includes positive rewards related to core operational and supply chain indicators, negative penalties related to operational and supply chain anomalies, and gradient-based auxiliary rewards for intermediate decisions approaching the optimal solution. A centralized training and decentralized execution architecture is adopted to construct a multi-agent deep reinforcement learning model, which is then trained offline using a digital twin simulation sandbox. Based on the decision rule elements and the multi-objective weighted reward function, the standardized dataset is input into the trained multi-agent deep reinforcement learning model to generate optimal task allocation schemes, inventory adjustment schemes and replenishment suggestions, as well as optimal path planning results.

[0037] Specifically, order information and supply chain data are acquired in real time. This acquired information is then deeply integrated with the real-time updated model storage data, warehousing environment data, and operational equipment data from the warehousing panoramic digital twin model. The integrated data undergoes preprocessing operations such as cleaning, missing value completion, outlier removal, and data format standardization to filter out invalid data interference and construct a standardized dataset. Based on the physical scene mapping and dynamic data simulation capabilities of the warehousing panoramic digital twin model, a reinforcement learning Markov decision environment adapted to the model storage scheduling scenario is built. The real-time operating state of the warehouse is defined as the environment state space, and various operations such as task allocation, inventory adjustment, replenishment planning, and path planning are defined as the action space. Changes in warehousing efficiency and supply chain response speed after interaction with actions are defined as state transition rules, clarifying the core decision-making rule elements in this decision-making environment. Simultaneously, a multi-objective weighted reward function is designed. This function employs a hierarchical weighting mechanism, including positive rewards that are positively correlated with core indicators such as warehousing efficiency, inventory turnover rate, equipment resource utilization rate, and supply chain response speed, as well as negative penalties associated with operational and supply chain anomalies such as equipment idleness, inventory backlog or stockouts, order delivery delays, and path planning redundancy. Furthermore, a gradient-based auxiliary reward is added for intermediate decision-making behaviors that approach the global optimum. This gradient-based reward value guides the algorithm's iteration direction, making it easier for the algorithm to converge to the global optimum rather than a local optimum during training and inference.

[0038] A multi-agent deep reinforcement learning model is constructed using a centralized training and decentralized execution architecture. Task allocation, inventory adjustment, replenishment planning, and path planning are mapped to different agents in the model. During the centralized training phase, each agent shares simulation data from the warehouse panoramic digital twin model and collaborates to complete model training, ensuring that the decision-making behaviors of each agent are compatible and conflict-free. At the same time, offline training of the model is carried out in conjunction with a digital twin simulation sandbox. In the sandbox, multiple scenarios and working conditions of warehouse operations are simulated, including abnormal scenarios such as sudden increases or decreases in order volume, equipment failures, and supply chain fluctuations. This allows the model to complete sufficient training iterations in a virtual environment, continuously optimize model parameters, and significantly improve the model's generalization ability and decision-making accuracy in real-world scenarios.

[0039] Finally, based on the preset decision rule elements and the designed multi-objective weighted reward function, the constructed standardized dataset is input into the trained multi-agent deep reinforcement learning model. Each agent in the model performs autonomous reasoning and collaborative decision-making for task allocation, inventory adjustment, replenishment suggestions, and path planning in the mold warehousing. Through multiple rounds of simulation iteration and optimization, the model generates the optimal task allocation scheme, inventory adjustment scheme, and replenishment suggestions that are adapted to the real-time operation status of the warehouse, meet order requirements, and match the capabilities of the upstream and downstream of the supply chain, as well as the optimal path planning results for the operating equipment. This realizes the intelligent and scientific generation of the mold warehousing scheduling scheme, making the allocation of warehouse resources and operation planning more reasonable and adaptable.

[0040] In one embodiment, based on decision rule elements and a multi-objective weighted reward function, a standardized dataset is input into a trained multi-agent deep reinforcement learning model to generate optimal task allocation schemes, inventory adjustment schemes and replenishment suggestions, as well as optimal path planning results, including: The standardized dataset is divided into a training sample set and a validation sample set according to a preset ratio. The training sample set is input into a multi-agent deep reinforcement learning model, and an inference framework is built by combining decision rule elements. With a multi-objective weighted reward function as the optimization guide, multiple rounds of online inference iterations are performed to generate multiple sets of candidate task allocation, inventory adjustment and replenishment, and path planning schemes. Based on the validation sample set, the effectiveness of multiple sets of candidate schemes is verified, and the candidate schemes are quantitatively scored based on core evaluation indicators. Based on the scoring results, the optimal task allocation scheme, inventory adjustment scheme and replenishment suggestion, and optimal path planning result are obtained.

[0041] Specifically, the standardized dataset is divided into a training sample set and a validation sample set according to a preset ratio. The training sample set is used for online inference iterative training of the model to ensure that the model can fully learn the decision-making rules of the warehouse scheduling scenario. The validation sample set is used to verify the effectiveness of the candidate solutions generated by the model to ensure the generalization ability and practical adaptability of the final output solution and avoid model overfitting. The divided training sample set is input into the multi-agent deep reinforcement learning model that has completed offline training. A targeted inference framework is built by combining preset decision rule elements to clarify the inference boundaries, interaction logic and decision priorities of each agent, ensuring that each agent makes independent decisions and adapts collaboratively during the inference process. With a multi-objective weighted reward function as the core optimization guide, the model initiates multiple rounds of online inference iterations. In each iteration, each agent autonomously generates a preliminary plan for task allocation, inventory adjustment, replenishment, and path planning based on order information, supply chain data, and real-time warehouse data in the training sample set. The reward value (a weighted sum of positive reward, negative penalty, and gradient auxiliary reward) is calculated for the plan through the reward function, and the model parameters are adjusted in reverse according to the reward value to guide the decisions of each agent to gradually converge toward the global optimal solution. After multiple iterations, multiple sets of candidate task allocation, inventory adjustment and replenishment, and path planning plans covering different decision dimensions and adapted to different scenarios are generated. The validation sample set is then input into the model inference stage. Based on core evaluation indicators such as warehousing operation efficiency, inventory turnover rate, equipment resource utilization rate, supply chain response speed, replenishment cost, and path planning redundancy, a quantitative scoring system is constructed. Reasonable weight coefficients are assigned to each indicator. A comprehensive effectiveness verification and quantitative scoring of multiple candidate solutions is conducted. Specifically, the validation sample set simulates actual warehousing operation scenarios, deduces the execution effect of each candidate solution, statistically analyzes the achievement of each core indicator, and calculates the comprehensive score of each candidate solution according to the scoring system, intuitively presenting the advantages and disadvantages of each solution. Finally, the comprehensive scores of all candidate solutions are ranked, and the candidate solution with the highest comprehensive score is selected as the optimal solution.

[0042] In one embodiment, based on the scoring results, the optimal task allocation plan, inventory adjustment plan and replenishment suggestion, as well as the optimal route planning results, include: Based on the scoring results, the candidate solution with the best score is selected as the basic solution. The basic solution is then input into the warehouse panoramic digital twin model for real-world adaptation simulation. Combined with real-time dynamic data of the warehouse, the key parameters in the basic solution are modified for adaptability. The modified solution undergoes a second conflict detection. After the conflict detection is passed, it is determined as the final optimal task allocation solution, inventory adjustment solution, replenishment suggestion, and optimal path planning result.

[0043] Specifically, the quantitative scoring results of multiple candidate solutions are first ranked, and the candidate solution with the highest score is selected as the base solution. The base solution is then input into the warehouse panoramic digital twin model. The model uses the current real-time dynamic data of the warehouse as the simulation boundary conditions to deduce the entire execution process of the base solution in the actual scenario. Through simulation analysis, key parameters in the base solution that do not match the real-time scenario are accurately located, and adaptation corrections are made to these key parameters. For example, the start time of equipment operation is adjusted to avoid busy equipment hours, the target storage location for inventory adjustment is changed to adapt to the current idle resources, the replenishment trigger threshold is optimized to match real-time supply and demand changes, and equipment path nodes are corrected to avoid congested channels, ensuring that the solution is highly consistent with the actual operating state of the warehouse. The revised plan needs to undergo a second conflict detection process. Specifically, the pre-set conflict detection rules are used to focus on identifying potential new resource occupation conflicts, spatiotemporal trajectory conflicts, and process connection conflicts during the real-world adaptation process. In particular, a comprehensive scan is conducted on the changes in the correlation between the elements of each plan after the parameter correction. If a new conflict is detected, it is resolved in real time based on a dynamic priority strategy. If no conflict is detected or the conflict has been completely resolved, the plan is confirmed as the final optimal task allocation plan, inventory adjustment plan, replenishment suggestion, and optimal path planning result.

[0044] In this embodiment, the technical pain point of the algorithm-generated scheme being out of touch with the actual physical warehousing scenario is solved by the dual optimization mechanism of real-world adaptation simulation and secondary conflict detection, ensuring that the final scheme has both theoretical optimality and strong feasibility for implementation.

[0045] In one embodiment, based on the full-process management data of mold blank storage, achieving full lifecycle traceability of mold blank data through blockchain technology includes: The entire process of mold blank warehousing management and control data is collected and structured in stages, and a unique on-chain identifier is assigned to each mold blank. The processed mold blank warehousing management and control data is then encapsulated in blocks, and hashed and encrypted after overlaying timestamps and operation node information to generate tamper-proof on-chain encrypted data units. A dedicated consortium blockchain for mold blank warehousing is built, and the encrypted data units are synchronized to authorized nodes in warehousing, supply chain, and operation execution. Multi-node data verification and on-chain notarization are completed through a consensus mechanism to achieve distributed data storage. A linkage and traceability mechanism between blockchain and the warehousing panoramic digital twin model is constructed, which retrieves on-chain data and synchronously maps it to the entire process operation scenario of the mold blank in the warehousing panoramic digital twin model to form a visualized traceability link, so as to realize the full life cycle traceability of mold blank data.

[0046] Specifically, the data for the entire process of mold blank warehousing is first collected and processed in a structured manner, segmented by stage. Following the operational flow of warehousing, storage, allocation, and outbound, operational data, status change data, and responsible entity data for each stage of the mold blank are categorized and collected. Simultaneously, basic attribute data of the mold blank itself (model, specifications, material, production batch, etc.), warehousing environment monitoring data (temperature, humidity, dust concentration, shelf load-bearing capacity, etc.), operational equipment linkage data (equipment number, operation log, operating parameters, etc.), and order and supply chain related data (order number, supply information, delivery records, etc.) are integrated to form a comprehensive, structured dataset covering mold blanks, environment, equipment, and business operations, ensuring the integrity and relevance of traceability data. A unique blockchain digital identifier is assigned to each mold blank, and this identifier is permanently and uniquely bound to the physical identifier of the mold blank (such as a QR code or RFID tag), achieving a one-to-one mapping between the physical entity of the mold blank and its digital file on the blockchain, establishing a core index for full lifecycle traceability. The processed structured data is encapsulated in blocks. Specifically, the data is divided into several data blocks according to the operation sequence. Each block is overlaid with a timestamp and operation node information (such as operator, equipment number, and operation area). Then, each block is encrypted using hash encryption algorithms such as SHA-256 to generate an immutable on-chain encrypted data unit, ensuring the security and integrity of the data during storage and transmission. A dedicated consortium blockchain architecture for mold warehousing is built. Specifically, multi-dimensional authorized nodes are set up, including warehouse management, operation execution, upstream supply chain, and downstream demand. Each node is configured with a dedicated digital signature and hierarchical data read / write permissions (e.g., the warehouse end has full data read / write permissions, while the supply chain end only has data query permissions related to the mold). The encrypted data unit is synchronized to each authorized node, and multi-node data verification and on-chain notarization are completed through a consensus mechanism to achieve distributed data storage.

[0047] An innovative traceability mechanism linking blockchain and a holistic digital twin model of warehousing has been constructed. When a mold data traceability request is initiated, the entire lifecycle of the stored data can be quickly retrieved on the blockchain through the mold's unique blockchain digital identifier. Simultaneously, the blockchain data is mapped to the holistic digital twin model of warehousing, recreating the complete operational trajectory of the mold from entry verification, in-warehouse storage, transfer and transportation to outbound handover in a virtual scenario. This includes visualized information such as time nodes, operational behaviors, environmental parameters, and equipment linkage status at each stage. This dual traceability model, combining data and scenario, makes the entire lifecycle trajectory of the mold intuitively verifiable and traceable, solving industry pain points such as broken data traceability links, discrepancies between records and actual inventory, and difficulty in defining responsibility in traditional mold data traceability. It also provides reliable data support for warehousing operation review and optimization, and supply chain collaborative accountability.

[0048] In one embodiment, such as Figure 3As shown, a smart warehousing and scheduling management device for mold blanks is provided, including: a digital twin model construction module 10, a solution generation module 20, an instruction generation module 30, an equipment collaborative operation module 40, and a lifecycle traceability module 50, wherein: The digital twin model building module 10 is used to collect mold blank data, warehousing environment data and operating equipment data in real time, and build a panoramic digital twin model of the warehouse based on the mold blank data, warehousing environment data and operating equipment data. The solution generation module 20 is used to acquire order information and supply chain data, and based on the warehouse panoramic digital twin model, combined with order information and supply chain data, it generates the optimal task allocation plan, inventory adjustment plan and replenishment suggestions, as well as the optimal path planning results through reinforcement learning algorithms. The instruction generation module 30 is used to generate task scheduling instructions, inventory adjustment instructions, and equipment path planning instructions based on the optimal task allocation scheme, inventory adjustment scheme, replenishment suggestions, and optimal path planning results, through the warehouse panoramic digital twin model. The equipment collaborative operation module 40 is used to realize multi-device collaborative operation based on task scheduling instructions, inventory adjustment instructions and equipment path planning instructions, and adopts a collaborative control architecture of centralized global coordination terminal and distributed equipment terminal to complete the handling and storage of mold blanks. The lifecycle traceability module 50 is used to achieve full lifecycle traceability of mold data based on the whole process management data of mold storage through blockchain technology.

[0049] In one embodiment, the instruction generation module 30 is further configured to load the optimal task allocation scheme, inventory adjustment scheme, replenishment suggestion, and optimal path planning result, and access the real-time warehouse dynamic data stream to provide a real-time context for instruction generation; perform fusion analysis on the loaded optimal task allocation scheme, inventory adjustment scheme, replenishment suggestion, and optimal path planning result, identify potential conflicting operations through preset conflict detection rules, and resolve them based on a dynamic priority strategy to generate a conflict-free primary instruction set; input the primary instruction set into the warehouse panoramic digital twin model for simulation operation, fine-tune the time parameters and resource allocation of the instructions based on the simulation results, and generate intermediate instructions; and convert the intermediate instructions into specific task scheduling instructions, inventory adjustment instructions, and equipment path planning instructions.

[0050] In one embodiment, the instruction generation module 30 is further configured to generate an isolated simulation sandbox based on the warehouse panoramic digital twin model. The simulation sandbox takes a primary instruction set, real-time equipment status data, and warehouse layout data as initial inputs. In the simulation sandbox, each operating device is modeled as an intelligent agent with autonomous decision-making capabilities, simulating the process of the intelligent agent cooperating according to the primary instruction set, and simultaneously collecting multi-source simulation data. Based on the multi-source simulation data, the critical path delay in time parameters and the competition hotspots in resource allocation are identified. Based on the critical path delay in time parameters and the competition hotspots in resource allocation, the time parameters of the instructions are dynamically adjusted, and the resource allocation is fine-tuned to generate intermediate instructions.

[0051] In one embodiment, the equipment collaborative operation module 40 is also used to build a collaborative control architecture between a centralized global coordination terminal and distributed equipment terminals. The global coordination terminal issues instructions to each equipment terminal to perform global operation status management and cross-equipment resource allocation. After receiving the global instructions, each equipment terminal completes autonomous negotiation within the same work area through point-to-point communication, determines local operation avoidance, resource occupation and process connection strategies, generates a local collaborative solution that meets global constraints and executes it autonomously to achieve multi-equipment collaborative operation and complete the handling and storage of the mold blank. When the equipment terminal encounters a local emergency during the operation, it first performs local autonomous emergency handling and simultaneously reports to the global coordination terminal. The global coordination terminal dynamically decides whether to optimize the global solution and issues adjustment instructions based on the impact range of the emergency.

[0052] In one embodiment, the solution generation module 20 is further used to acquire order information and supply chain data, integrate the order information and supply chain data with real-time data from the warehouse panoramic digital twin model, and preprocess the data to construct a standardized dataset; build a Markov decision environment for reinforcement learning based on the warehouse panoramic digital twin model, clarify the decision rule elements, and design a multi-objective weighted reward function. The multi-objective weighted reward function includes positive rewards associated with core indicators of operations and supply chains, negative penalties associated with abnormal situations in operations and supply chains, and gradient auxiliary rewards for intermediate decisions that approach the optimal solution; adopt a centralized training and decentralized execution architecture to construct a multi-agent deep reinforcement learning model, and complete the offline training of the model by combining it with a digital twin simulation sandbox; based on the decision rule elements and the multi-objective weighted reward function, input the standardized dataset into the trained multi-agent deep reinforcement learning model to generate the optimal task allocation scheme, inventory adjustment scheme and replenishment suggestion, and optimal path planning result.

[0053] In one embodiment, the scheme generation module 20 is further configured to divide the standardized dataset into a training sample set and a validation sample set according to a preset ratio; input the training sample set into the trained multi-agent deep reinforcement learning model, build an inference framework by combining decision rule elements, and perform multiple rounds of online inference iteration with a multi-objective weighted reward function as the optimization guide to generate multiple sets of candidate task allocation, inventory adjustment and replenishment, and path planning schemes; verify the effectiveness of multiple sets of candidate schemes based on the validation sample set, and quantitatively score the candidate schemes based on core evaluation indicators; and obtain the optimal task allocation scheme, inventory adjustment scheme and replenishment suggestion, and optimal path planning result based on the scoring results.

[0054] In one embodiment, the solution generation module 20 is further configured to select the candidate solution with the best score as the basic solution based on the scoring results, input the basic solution into the warehouse panoramic digital twin model for real-world adaptation simulation, and combine the real-time dynamic data of the warehouse to make adaptation corrections to the key parameters in the basic solution; perform secondary conflict detection on the corrected solution, and after the conflict detection is passed, determine it as the final optimal task allocation solution, inventory adjustment solution and replenishment suggestion, and optimal path planning result.

[0055] In one embodiment, the lifecycle traceability module 50 is also used to collect and structure the full-process control data of mold blank warehousing in stages, and assign a unique on-chain identifier to each mold blank; to encapsulate the processed full-process control data of mold blank warehousing in blocks, and to perform hash encryption after superimposing timestamps and operation node information to generate an immutable on-chain encrypted data unit; to build a dedicated consortium chain for mold blank warehousing, and to synchronize the encrypted data unit to each authorized node of warehousing, supply chain, and operation execution, and to complete multi-node data verification and on-chain notarization through a consensus mechanism to achieve distributed data storage; to build a linkage traceability mechanism between blockchain and warehousing panoramic digital twin model, and to retrieve on-chain data and synchronously map it to the full-process operation scenario of the mold blank in the warehousing panoramic digital twin model to form a visual traceability link, so as to realize the full lifecycle traceability of mold blank data.

[0056] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for intelligent storage and scheduling management of mold blanks, characterized in that, include: Real-time data collection of mold base data, storage environment data, and operating equipment data; and construction of a panoramic digital twin model of the storage area based on the mold base data, storage environment data, and operating equipment data. Obtain order information and supply chain data, and based on the warehouse panoramic digital twin model, combine the order information and supply chain data to generate the optimal task allocation scheme, inventory adjustment scheme and replenishment suggestion, as well as the optimal path planning result through reinforcement learning algorithm; Based on the optimal task allocation scheme, the inventory adjustment scheme and replenishment suggestion, and the optimal path planning result, task scheduling instructions, inventory adjustment instructions and equipment path planning instructions are generated through the warehouse panoramic digital twin model; Based on the task scheduling instructions, the inventory adjustment instructions, and the equipment path planning instructions, a collaborative control architecture of centralized global coordination terminal and distributed equipment terminal is adopted to realize multi-device collaborative operation and complete the handling and storage of mold blanks. Based on the full-process management data of mold blank storage, blockchain technology is used to achieve full lifecycle traceability of mold blank data.

2. The intelligent warehousing and scheduling management method for mold blanks according to claim 1, characterized in that, The process of generating task scheduling instructions, inventory adjustment instructions, and equipment path planning instructions through the warehouse panoramic digital twin model, based on the optimal task allocation scheme, the inventory adjustment scheme and replenishment suggestions, and the optimal path planning results, includes: The optimal task allocation scheme, the inventory adjustment scheme and replenishment suggestion, and the optimal path planning result are loaded and connected to the real-time warehouse dynamic data stream to provide real-time context for instruction generation; The optimal task allocation scheme, inventory adjustment scheme, replenishment suggestion and optimal path planning results after loading are integrated and analyzed. Potential conflicting operations are identified through preset conflict detection rules and resolved based on dynamic priority strategy to generate a conflict-free primary instruction set. The primary instruction set is input into the warehouse panoramic digital twin model for simulation. Based on the simulation results, the time parameters and resource allocation of the instructions are fine-tuned to generate intermediate instructions. The intermediate-level instructions are converted into specific task scheduling instructions, inventory adjustment instructions, and equipment path planning instructions.

3. The intelligent warehousing and scheduling management method for mold blanks according to claim 2, characterized in that, The process involves inputting the primary instruction set into the warehouse panoramic digital twin model for simulation, fine-tuning the timing parameters and resource allocation of the instructions based on the simulation results, and generating intermediate instructions, including: Based on the aforementioned panoramic digital twin model of the warehouse, an isolated simulation sandbox is generated, which takes the initial instruction set, real-time equipment status data, and warehouse layout data as initial inputs. In the simulation sandbox, each working device is modeled as an intelligent agent with autonomous decision-making capabilities, simulating the process of the intelligent agent cooperating according to the primary instruction set, and simultaneously collecting multi-source simulation data; Based on the multi-source simulation data, the critical path delay in time parameters and the hot spots of competition in resource allocation are identified. Based on the critical path delay in time parameters and the hot spots of competition in resource allocation, the time parameters of the instructions are dynamically adjusted, and the resource allocation is fine-tuned to generate intermediate instructions.

4. The intelligent warehousing and scheduling management method for mold blanks according to claim 1, characterized in that, The method, based on the task scheduling instructions, the inventory adjustment instructions, and the equipment path planning instructions, employs a collaborative control architecture combining a centralized global coordination terminal and distributed device terminals to achieve multi-device collaborative operation and complete the handling and storage of mold blanks, including: Establish a collaborative control architecture between a centralized global coordination terminal and distributed device terminals. The global coordination terminal issues instructions to each device terminal to perform global operation status management and cross-device resource allocation. After receiving the global instructions, each device autonomously negotiates with the work area through point-to-point communication, determines local operation avoidance, resource occupation and process connection strategies, generates a local collaborative solution that meets global constraints and executes it autonomously, so as to realize multi-device collaborative operation and complete the handling and storage of mold blanks. When a local emergency occurs during operation, the equipment will first take local emergency measures and simultaneously report to the global coordination terminal. The global coordination terminal will then dynamically decide whether to optimize the global plan and issue adjustment instructions based on the scope of the emergency.

5. The intelligent warehousing and scheduling management method for mold blanks according to claim 1, characterized in that, The process of acquiring order information and supply chain data, and based on the warehouse panoramic digital twin model, combining the order information and supply chain data, generating optimal task allocation schemes, inventory adjustment schemes and replenishment suggestions, as well as optimal path planning results through reinforcement learning algorithms, includes: Obtain order information and supply chain data, integrate the order information and supply chain data with the real-time data of the warehouse panoramic digital twin model and perform preprocessing to construct a standardized dataset; Based on the aforementioned warehouse panoramic digital twin model, a Markov decision environment based on reinforcement learning is built, the decision rule elements are clarified, and a multi-objective weighted reward function is designed. The multi-objective weighted reward function includes positive rewards associated with core indicators of operations and supply chain, negative penalties associated with abnormal situations in operations and supply chain, and gradient auxiliary rewards for intermediate decisions that approach the optimal solution. A centralized training and decentralized execution architecture is adopted to build a multi-agent deep reinforcement learning model, and the offline training of the model is completed by combining a digital twin simulation sandbox. Based on the decision rule elements and the multi-objective weighted reward function, the standardized dataset is input into the trained multi-agent deep reinforcement learning model to generate the optimal task allocation scheme, inventory adjustment scheme and replenishment suggestion, as well as the optimal path planning result.

6. The intelligent warehousing and scheduling management method for mold blanks according to claim 5, characterized in that, The process of inputting the standardized dataset into a multi-agent deep reinforcement learning model trained based on the decision rule elements and the multi-objective weighted reward function to generate optimal task allocation schemes, inventory adjustment schemes and replenishment suggestions, and optimal path planning results includes: The standardized dataset is divided into a training sample set and a validation sample set according to a preset ratio; The training sample set is input into the trained multi-agent deep reinforcement learning model, and the inference framework is built in combination with the decision rule elements. With the multi-objective weighted reward function as the optimization guide, multiple rounds of online inference iteration are carried out to generate multiple sets of candidate task allocation, inventory adjustment and replenishment and path planning schemes. Based on the verification sample set, the effectiveness of multiple candidate solutions is verified, and the candidate solutions are quantitatively scored based on the core evaluation indicators. Based on the scoring results, the optimal task allocation plan, inventory adjustment plan and replenishment suggestions, as well as the optimal route planning results are obtained.

7. The intelligent warehousing and scheduling management method for mold blanks according to claim 6, characterized in that, The optimal task allocation plan, inventory adjustment plan and replenishment suggestion, and optimal route planning results obtained based on the scoring results include: Based on the scoring results, the candidate solution with the best score is selected as the basic solution. The basic solution is then input into the warehouse panoramic digital twin model for real-world adaptation simulation. Combined with real-time dynamic data of the warehouse, the key parameters in the basic solution are adjusted for adaptability. The revised plan undergoes a second conflict check. Once the conflict check passes, it is determined as the final optimal task allocation plan, inventory adjustment plan, replenishment suggestion, and optimal path planning result.

8. The intelligent warehousing and scheduling management method for mold blanks according to claim 1, characterized in that, The method of achieving full lifecycle traceability of mold data based on the full-process management data of mold storage includes: Data from the entire process of mold blank storage is collected and structured in stages, and a unique on-chain identifier is assigned to each mold blank. The processed mold blank storage full-process control data is encapsulated in blocks, and after overlaying timestamps and operation node information, hash encryption is completed to generate an immutable on-chain encrypted data unit. Build a dedicated alliance chain for mold storage, synchronize encrypted data units to authorized nodes in warehousing, supply chain, and operation execution, and complete multi-node data verification and on-chain notarization through a consensus mechanism to achieve distributed data storage; A linkage and traceability mechanism is constructed between blockchain and the warehousing panoramic digital twin model. On-chain data is retrieved and synchronously mapped to the entire process operation scenario of the mold blank in the warehousing panoramic digital twin model to form a visual traceability link, so as to realize the full life cycle traceability of the mold blank data.

9. A mold blank intelligent storage and scheduling management device, characterized in that, include: The digital twin model building module is used to collect mold base data, warehousing environment data, and operating equipment data in real time, and to build a panoramic digital twin model of the warehouse based on the mold base data, the warehousing environment data, and the operating equipment data. The solution generation module is used to acquire order information and supply chain data, and based on the warehouse panoramic digital twin model, combined with the order information and supply chain data, to generate the optimal task allocation plan, inventory adjustment plan and replenishment suggestion, as well as the optimal path planning result through reinforcement learning algorithm; The instruction generation module is used to generate task scheduling instructions, inventory adjustment instructions, and equipment path planning instructions based on the optimal task allocation scheme, the inventory adjustment scheme and replenishment suggestions, and the optimal path planning results, through the warehouse panoramic digital twin model. The equipment collaborative operation module is used to realize multi-device collaborative operation and complete the handling and storage of mold blanks based on the task scheduling instructions, the inventory adjustment instructions and the equipment path planning instructions, using a collaborative control architecture of centralized global coordination terminal and distributed device terminal. The lifecycle traceability module is used to achieve full lifecycle traceability of mold data based on the data management of the entire mold storage process, using blockchain technology.

Images