Tray cycle scheduling system and application method

Abstract

According to the tray cycle scheduling system and the application method, historical data and a production plan are fused, an improved algorithm is adopted to predict tray requirements, and an attention mechanism is introduced to improve precision; the method comprises the following steps: acquiring multi-dimensional state information of a tray through a multi-modal sensor, eliminating noise by using a data fusion algorithm, and constructing a digital twin model to realize state synchronization; a double-layer optimization architecture is constructed, an upper layer solves a global scheme by combining an improved particle swarm and a simulated annealing algorithm, and a lower layer dynamically adjusts a path through reinforcement learning; an instruction is generated based on a digital twin model, an event triggering mechanism is adopted to reduce communication load, and virtual-real interaction closed-loop control is realized; a real-time evaluation index system is established, a meta-learning algorithm is utilized to quickly adapt to a new environment, and system parameters are continuously optimized. Multi-module collaborative innovation is achieved, the tray scheduling efficiency and the intelligent level are remarkably improved, production logistics whole-process collaborative optimization is achieved, and core support is provided for cost reduction and efficiency improvement of enterprises.

Description

Technical Field

[0001] The present invention relates to the technical field of pallet scheduling, and in particular to a pallet circulation scheduling system and an application method. Background Art

[0002] In modern logistics and intelligent manufacturing, pallets serve as the fundamental units of cargo transportation, warehousing, and production flow, and their scheduling efficiency directly impacts the overall effectiveness of the supply chain. Traditional pallet scheduling relies primarily on manual experience and fixed rules, such as allocating pallets according to order sequence and using fixed routes. This approach struggles to cope with complex scenarios such as fluctuating order volumes and shifting production cycles, often leading to both idle and scarce pallet resources. For example, during peak order periods, such as e-commerce promotions, manual scheduling is unable to quickly respond to the surge in pallet demand, resulting in a backlog of goods. During low-order periods, however, a large number of pallets remain idle and wasted, resulting in resource utilization rates of less than 40%.

[0003] With the prevalence of automated logistics equipment (such as AGVs and stacker cranes), some companies have introduced pallet scheduling systems based on a single algorithm, such as a simple genetic algorithm or a rule engine. However, these systems have significant flaws: on the one hand, they lack the ability to accurately predict pallet demand and are unable to plan resources in advance by combining historical data, production plans, and other factors; on the other hand, their perception of pallet status is limited, only acquiring location and load information, making it difficult to grasp deeper conditions such as pallet wear and maintenance needs, resulting in sudden pallet failures that affect production continuity. In addition, traditional systems are mostly independent and have weak data interaction capabilities with enterprise ERP, MES, and other management systems, making it impossible to achieve coordinated scheduling of production, logistics, and warehousing.

[0004] In the context of intelligent upgrades, although some advanced companies are attempting to leverage the Internet of Things and big data technologies to optimize pallet scheduling, they still face numerous challenges. When dealing with multi-objective optimization problems (such as balancing efficiency, cost, and equipment balance), existing systems' algorithms are prone to falling into local optimality and are unable to output a globally optimal scheduling solution. When abnormal events (such as equipment failures and path congestion) occur, they lack the ability to respond quickly and dynamically reschedule, resulting in extended production interruptions. Furthermore, energy consumption issues remain unaddressed, with frequent equipment starts and stops and ineffective handling resulting in significant energy waste, which does not meet the development needs of green manufacturing. Summary of the Invention

[0005] In order to solve the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a pallet circulation scheduling system and application method, which can significantly improve the efficiency and intelligence level of pallet scheduling, realize the coordinated optimization of the entire production logistics process, and provide core support for enterprises to reduce costs and increase efficiency.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A pallet circulation scheduling system, comprising:

[0008] Demand forecasting module: Based on historical order data, an improved seasonal decomposition time series model is used to predict pallet demand. Wavelet transform decomposition is combined to correct prediction errors, and an attention mechanism is introduced to assign weights to data from different historical periods.

[0009] Real-time status perception module: RFID, visual recognition, and vibration sensors are deployed on pallets to collect real-time information on pallet position, load status, surface wear, and internal stress distribution. An improved Kalman filter algorithm is used to fuse data and eliminate measurement noise.

[0010] Scheduling decision module: Constructs a two-layer optimization decision model. The upper layer uses an improved particle swarm algorithm combined with a simulated annealing mechanism to solve the global optimal scheduling solution, while the lower layer dynamically adjusts local path planning through reinforcement learning.

[0011] Execution control module: Builds a virtual mapping model of the physical pallet based on digital twin technology, and controls the virtual and real through real-time data interaction. It uses an event-triggered control mechanism to control the frequency of command transmission, and sends a command when the state error exceeds the trigger threshold.

[0012] Evaluation and optimization module: Establish a real-time evaluation model based on digital twins, calculate system performance indicators by comparing the operating status of the physical system and the virtual model; use meta-learning algorithms to adapt to the new production environment and self-optimize the system.

[0013] Furthermore, it also includes:

[0014] Buffer management module: A dynamic adaptive buffer is set up between the production line and the storage area, and a fuzzy logic control algorithm is used to dynamically adjust the buffer capacity. A dynamic priority queue is introduced to manage pallets, and the priority is adjusted in real time based on the production rhythm and order urgency. An early warning is triggered when the buffer utilization rate exceeds the threshold.

[0015] Abnormal processing module: Build an abnormal diagnosis model based on Bayesian network to calculate the probability of abnormal events in real time; when an abnormality is detected, a negotiation mechanism is used to generate the optimal emergency plan.

[0016] Furthermore, it also includes:

[0017] Energy Consumption Optimization Module: Builds an equipment energy consumption prediction model based on deep reinforcement learning, optimizes scheduling strategies based on equipment load, operating speed, and start-stop frequency; introduces an energy recovery mechanism to convert kinetic energy during AGV braking into electrical energy storage;

[0018] System collaboration module: Build a trusted data sharing platform based on blockchain technology, use contracts to automatically execute cross-system business processes, and support enterprises' large-scale production scheduling data interaction needs.

[0019] Furthermore, the sensors in the real-time status perception module use low-power wide-area Internet of Things (LPWAN) technology to transmit data.

[0020] Furthermore, the improved particle swarm algorithm in the scheduling decision module introduces the Levy flight mechanism to search the global environment and optimize the path efficiency.

[0021] Furthermore, the dynamic priority queue in the buffer management module uses a hyperbolic time discounting algorithm to calculate the pallet priority, and automatically adjusts the discount factor when the production line beat fluctuation exceeds a threshold.

[0022] Furthermore, the blockchain in the system collaboration module adopts the practical Byzantine fault-tolerant PBFT consensus algorithm to support enterprise-level data interaction needs.

[0023] An application method of a pallet circulation scheduling system further includes the following steps:

[0024] Demand forecasting steps: Collect historical data and use the improved STL model combined with wavelet transform and attention mechanism to generate pallet demand forecast results;

[0025] State perception step: Real-time pallet status information is collected through the sensor network, and the improved Kalman filter algorithm is used to fuse the data to build a digital twin model synchronization state;

[0026] Scheduling decision-making steps: Construct a two-layer optimization model. The upper layer solves the global solution by improving the PSO algorithm combined with the simulated annealing mechanism, and the lower layer dynamically adjusts the path through reinforcement learning.

[0027] Execution control steps: Generate control instructions based on the digital twin model, use event triggering mechanism to optimize communication load, and control the closed loop through virtual-real interaction;

[0028] Evaluation and optimization steps: Establish a real-time evaluation indicator system, use meta-learning algorithms to adapt to new environments, and continuously optimize system parameters and scheduling strategies.

[0029] Furthermore, when encountering target conflicts in the scheduling decision step, the Nash bargaining solution method is used to balance the target weights, and the priorities are dynamically adjusted through the negotiation mechanism.

[0030] Furthermore, the federated transfer learning method is adopted in the evaluation and optimization step to support cross-enterprise scheduling knowledge sharing while protecting data privacy.

[0031] Compared with the existing technology, the beneficial effects of the present invention are:

[0032] In terms of demand forecasting, by integrating an improved time series model with an attention mechanism, it accurately captures order fluctuations, increasing the accuracy of pallet demand forecasts to over 95%, avoiding resource waste and shortages and reducing the company's inventory costs by 25%. The real-time status perception module, leveraging multimodal sensors and data fusion algorithms, not only detects pallet location and load status, but also monitors hidden information such as wear and stress, providing 72-hour advance warning of pallet failure risks and reducing equipment failure rates by 40%.

[0033] The scheduling decision module, leveraging a two-layer optimization architecture and improved algorithms, rapidly generates globally optimal scheduling solutions in complex scenarios, increasing pallet turnover efficiency by 50% and equipment utilization to 85%. The execution control phase, based on digital twins and event-triggered mechanisms, enables precise control through virtual-real linkage, reducing command transmission latency to under 50 milliseconds and reducing ineffective handling by 30%. The exception handling module, leveraging a Bayesian network, can diagnose anomalies and generate contingency plans within one minute, ensuring that the system maintains over 90% operational efficiency even in the face of anomalies.

[0034] Furthermore, the system innovatively incorporates energy optimization and system collaboration modules. Using deep reinforcement learning to optimize equipment operation strategies, it reduces energy consumption of AGVs and other equipment by 35%. A blockchain-based trusted data sharing platform enables seamless collaboration with ERP and MES systems, increasing the response speed of production planning and logistics scheduling by 60%. Overall, the system effectively addresses the low efficiency, high costs, and poor coordination inherent in traditional pallet scheduling, providing core technical support for cost reduction, efficiency improvement, and intelligent upgrades for enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a schematic block diagram of a pallet circulation scheduling system proposed by the present invention;

[0036] Figure 2 This is a schematic block diagram of an application method of a pallet circulation scheduling system proposed by the present invention;

[0037] Figure 3 This is a diagram comparing the accuracy of demand forecasts;

[0038] Figure 4 Schematic diagram of pallet turnover rate changing over time;

[0039] Figure 5 This is a schematic diagram comparing equipment utilization and energy efficiency;

[0040] Figure 6 This is a schematic diagram of the abnormal response time comparison;

[0041] Figure 7 Schematic diagram of the communication load reduction effect. DETAILED DESCRIPTION

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0043] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like to indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0044] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the said features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined. In addition, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. The present invention will be further described in detail below with reference to the accompanying drawings.

[0045] Reference Figures 1 to 7 : A pallet circulation scheduling system, comprising:

[0046] Demand forecasting module: Extract production order data from the enterprise ERP system for the past three years, covering information such as order time, product model, and number of pallets used. Combined with the production plan data from the MES system, this module forms the original data set. An improved seasonal decomposition time series model (STL) is used to conduct a preliminary analysis of the data, decomposing the time series into trend terms, seasonal terms, and residual terms. The residual terms are then decomposed at multiple scales using wavelet transform to identify and correct abnormal fluctuations. The attention mechanism is introduced to automatically assign weights to data from different historical periods. The calculation formula is: Among them, y tis the pallet demand forecast value at time t, T t is the trend term, S t is the seasonal term, R t is the residual term, E i is the wavelet decomposition coefficient of the i-th scale, w i is the corresponding weight coefficient. After calculation, the prediction accuracy is increased from 82% of the traditional method to 96%.

[0047] Real-time status perception module: A multimodal sensor unit integrating an RFID chip, a micro camera, and a vibration sensor is deployed on each pallet. RFID readers are installed every 50 meters throughout the workshop, and UWB positioning base stations are deployed along critical paths to form a positioning network. As the pallet moves, the RFID reader collects basic information such as the pallet ID and location at a rate of 10 times per second. The camera captures real-time images of pallet surface wear, and the vibration sensor monitors vibration data during pallet handling. An improved Kalman filter algorithm is used to fuse multi-source data. Field testing has demonstrated pallet position accuracy of ±8 cm and wear detection accuracy of 92%.

[0048] Scheduling decision module: Construct a two-layer optimization decision model. The upper layer uses an improved particle swarm algorithm (PSO) combined with a simulated annealing mechanism, sets the particle swarm size to 100, and the number of iterations to 50 to solve the global optimal scheduling solution. The Levy flight mechanism is introduced to enhance the global search capability of the algorithm. The formula is: Where α is the step size control parameter, ⊕ represents point-to-point multiplication, Lévy (λ) is the Lévy distribution, and λ is the exponential parameter. The lower layer dynamically adjusts local path planning through reinforcement learning (DDPG), with the objective function being to maximize pallet turnover efficiency, minimize transportation costs, and balance equipment utilization. Constraints include pallet quantity limits, task priorities, and equipment capacity. In a scheduling network consisting of 120 nodes, the convergence speed of finding the optimal path is 52% faster than traditional algorithms.

[0049] Execution control module: Build a digital twin model of the pallet based on the Unity3D platform, which corresponds one-to-one with the physical pallet. After the scheduling decision module generates a scheduling plan, it sends control instructions to logistics equipment such as AGVs and stackers through the OPCUA protocol. The event-triggered control mechanism is used to reduce the frequency of control instruction transmission. The calculation formula is e k (t) = x(t) = x(t k ) and h(e k (t))=||e k (t)|| 2 -γ 2 ||x(t)|| 2 ≤0. Among them, e k (t) is the state error, x(t) is the system state, x(tk ) is the state at the most recent trigger moment, and γ is the trigger threshold. In actual operation, the communication load is reduced by 40%, and the instruction transmission delay is stabilized within 45 milliseconds.

[0050] Evaluation and Optimization Module: This module establishes a KPI evaluation system encompassing 12 indicators, including pallet turnover rate, equipment utilization, order fulfillment rate, and energy efficiency. A digital twin model compares the operating status of the physical system and the virtual model in real time. If a particular indicator fails to meet the target for three consecutive times, a meta-learning algorithm is triggered to optimize the scheduling model parameters. A / B testing is used to compare the effectiveness of different scheduling strategies, with scheduling strategies automatically switched weekly to continuously optimize system performance. After three months of operation, pallet turnover increased by 58%, equipment utilization increased from 65% to 88%, and energy efficiency improved by 38%.

[0051] The present invention also includes the following modules:

[0052] Buffer management module: 10 dynamic adaptive buffers are set up between the production line and the storage area, and the buffer capacity is dynamically adjusted using a fuzzy logic control algorithm. Emergency queues and normal queues are set up, and the order of pallets entering and leaving the buffer is determined by a priority algorithm based on factors such as production rhythm and order urgency. When the buffer utilization rate exceeds 80%, the system automatically sends an early warning to the dispatch center. Dynamic priority queue management is introduced to manage pallets, and the priority is adjusted in real time based on multiple factors such as production rhythm and order urgency. The calculation formula is: Where V(t) is the value at time t, V0 is the initial value, and k is the discount factor. When the production line's cycle time fluctuates by more than 20%, the system automatically adjusts the discount factor to ensure stable buffer management. In practice, the average buffer utilization rate remains stable at 75%, reducing material supply delays by 60%.

[0053] Abnormal processing module: Based on the Bayesian network, an abnormal diagnosis model is built to calculate the probability of abnormal events such as equipment failure and path congestion in real time. The formula is When an anomaly is detected, a negotiation mechanism is used to generate the optimal contingency plan. In one AGV failure incident, the system diagnosed the anomaly within 45 seconds and generated an alternative scheduling plan within one minute, limiting production disruption to under five minutes—an 80% reduction compared to traditional methods.

[0054] The present invention also includes the following modules:

[0055] Energy consumption optimization module: Establish an equipment energy consumption prediction model based on deep reinforcement learning, and optimize the scheduling strategy by combining factors such as equipment load, operating speed, start-stop frequency, etc. Introduce an energy recovery mechanism to convert the kinetic energy during the AGV braking process into electrical energy storage. The formula is Among them, E recis the recovered energy, η is the energy conversion efficiency, m is the AGV mass, and v is the speed before braking. Tests have shown that AGV energy consumption is reduced by 37%.

[0056] Multi-system collaboration module: Build a trusted data sharing platform based on blockchain technology, and use contracts to execute cross-system business processes. The formula is SC=f(I1,I2,…,I n ,O1,O2,…,O m ). Among them, SC is the contract, I i is the input parameter, O j is the output result, and f is the contract execution function. Data interoperability with enterprise ERP, MES, and other systems allows the system to calculate pallet demand and generate a scheduling plan within 10 seconds after an order is placed, increasing the response speed of production planning and logistics scheduling by 65%.

[0057] In this invention, multimodal sensors serve as the core data acquisition device, leveraging Low-Power Wide Area Network (LPWAN) technology for data transmission. LPWAN technology leverages its unique narrowband modulation and spread spectrum communication techniques to ensure data reliability while effectively reducing energy consumption. Modulation schemes such as linear frequency modulation spread spectrum (LoRa) are employed to enhance signal interference resistance by spreading the signal over a wider frequency band. This ensures accurate transmission of sensor-collected data, such as pallet position, weight, and environmental parameters, in complex industrial environments subject to significant electromagnetic noise interference. An adaptive transmission power adjustment strategy is incorporated to further optimize transmission performance. Sensor nodes dynamically adjust transmit power based on communication distance and signal strength. When communicating at close range, power is reduced to minimize energy consumption. When transmitting at long distances and with weak signals, power is moderately increased to ensure data reachability, ensuring stable transmission over distances ≥ 3 km. A node wakeup scheduling mechanism based on deep reinforcement learning is introduced. This mechanism determines node wakeup and sleep timing by deeply analyzing multiple sources of information, including changes in the sensor node's surrounding environment and data update frequency. In scenarios where data changes slowly, node sleep time is extended to keep node power consumption ≤5mA, reducing unnecessary energy consumption. In critical scenarios where data changes frequently, nodes are promptly awakened for data collection and transmission to ensure real-time data. Wavelet transforms are used to preprocess transmitted data, filtering out noise and redundant components through multi-scale decomposition to improve data quality. Incorporating an attention mechanism, different weights are assigned to data with different timestamps, prioritizing data at key time points and critical states. This further improves the effectiveness of data transmission and the system's accurate perception of the pallet's real-time status, thereby ensuring the long-term stable operation of the sensor network and providing reliable data support for the pallet circulation scheduling system.

[0058] In this invention, the scheduling decision module uses an improved particle swarm algorithm (PSO) to optimize paths. To address the limitations of traditional PSO algorithms, which are prone to local optima in complex network environments, a Lévy flight mechanism is introduced. From an algorithmic perspective, the step size of Lévy flight follows a power-law distribution, enabling particles to perform long-distance random jumps during the search process. In complex network scenarios involving pallet cyclic scheduling, this allows particles to break free from the constraints of local optima and explore potential optimal paths within a broader search space. During algorithm iterations, particles enter Lévy flight mode based on a set probability, enabling a large-scale, cross-region search. Incorporating an adaptive weight adjustment strategy, weights are dynamically adjusted based on the number of algorithm iterations and changes in the search space. In the early stages of the algorithm, the algorithm focuses on global search, exploring possible paths using the Lévy flight mechanism; in the later stages, it focuses on local search and refines the optimization process. A chaotic map is used to initialize the particle swarm, leveraging the randomness and ergodicity of chaotic sequences to achieve a more uniform and random initial distribution of particles, laying the foundation for efficient search. Furthermore, an information entropy-based diversity assessment mechanism is introduced. The particle swarm's information entropy is monitored in real time to assess the diversity of the particles. When information entropy falls below a threshold, indicating particle convergence and the algorithm's tendency to fall into a local optimum, the Levy flight parameters and search strategy are dynamically adjusted to stimulate particle search activity and maintain the algorithm's global search capability. The improved particle swarm algorithm significantly improves path optimization efficiency in complex networks, effectively shortening pallet transport routes, reducing transportation costs, and enhancing the overall efficiency and adaptability of the pallet circulation scheduling system.

[0059] In this invention, the buffer management module constructs a dynamic priority queue based on the hyperbolic time discounting algorithm to calculate pallet priorities. This algorithm deeply considers the temporal nature of pallet tasks and assigns a priority value to each pallet based on multiple factors, such as the urgency of the production process and the strictness of delivery deadlines for the materials it carries. A dynamic adjustment strategy based on adaptive feedback is introduced. When the production line cycle fluctuation exceeds 20%, the system quickly detects the change and automatically triggers fine-tuning of the discount factor in the hyperbolic time discounting algorithm. Leveraging real-time dynamic data from monitoring the production line cycle, a data analysis model is used to assess the impact of current production rhythm changes on pallet scheduling. The adjustment range and direction of the discount factor are then intelligently determined. To further improve the accuracy of priority calculation and the stability of buffer management, multi-source data fusion technology is employed. This technology integrates production planning and scheduling data, equipment operating status data, and material supply progress data. Deep neural networks are used to extract features and perform correlation analysis on the multivariate data, accurately capturing the potential connections and influencing patterns between different data sets. Dynamic calibration of the hyperbolic time discounting algorithm ensures that priority calculation more comprehensively and in real time reflects actual production needs. A pallet priority prediction and evaluation model is constructed. Model training utilizes historical production data and pallet dispatch records, and continuously optimizes model parameters using reinforcement learning algorithms. This model proactively predicts trends in pallet priorities under different scenarios, allowing scheduling strategies to be adjusted in advance, ensuring that buffer management remains highly stable and efficient in complex and volatile production environments.

[0060] In the present invention, blockchain technology adopts the Practical Byzantine Fault Tolerance (PBFT) consensus algorithm. A PBFT consensus optimization model is established, and indicators such as transaction confirmation time, TPS processing capacity, and node communication overhead are incorporated into the utility function. Automatic negotiation of participating nodes is achieved through contracts. When it is detected that the Pareto optimal solution of the consensus process deviates from the threshold range, the system automatically triggers the multi-objective optimization algorithm and dynamically adjusts parameters such as the message verification mechanism, voting weight distribution, and block generation cycle. The A / B testing method is used to compare the effects of different parameter configuration schemes, and the test strategy is automatically switched every 2 hours to continuously optimize system performance. Zero-knowledge proof technology is introduced to achieve privacy protection of transaction data, and a hierarchical consensus mechanism is used to improve system scalability. Through the federated learning mechanism, scheduling knowledge sharing and collaborative optimization are achieved while protecting the data privacy of each participant. When the system recognizes a new business scenario, it automatically calls the transfer learning algorithm to migrate and apply historical scheduling experience to the new scenario, and quickly generate an initial consensus parameter configuration scheme.

[0061] An application method of a pallet circulation scheduling system includes the following steps:

[0062] Demand forecasting steps: We collected historical order data from the past three years and production plans for the next three months to construct a multidimensional dataset. We used an improved STL model combined with wavelet transforms for time series analysis to generate initial forecasts. We introduced an attention mechanism to dynamically weight data from different time periods, correcting for forecast errors and ultimately generating a 72-hour pallet demand forecast with 96% accuracy.

[0063] State perception: RFID, cameras, and vibration sensors deployed on the pallet collect real-time multi-dimensional information, including position, load status, surface wear, and internal stress. An improved Kalman filter algorithm is used to fuse multi-source data, eliminate measurement noise, and construct a digital twin model. This synchronizes the physical pallet's state with the virtual model in real time, achieving a position accuracy of ±8 cm.

[0064] Scheduling decision-making steps: Construct a two-layer optimization decision-making model. The upper layer uses the improved PSO algorithm combined with the simulated annealing mechanism to solve the global optimal scheduling solution; the lower layer uses the DDPG algorithm to dynamically adjust the local path planning. When encountering multiple target conflicts, the Nash bargaining solution method is used to balance the weights of each target. The formula is Among them, u i is the utility value of the i-th goal, d i The controversial point is w i The negotiation mechanism dynamically adjusts the priority of each target, improving the overall system performance by more than 15%.

[0065] Execution control steps: Generate physical system control instructions based on the digital twin model, using an event-triggered control mechanism to reduce the frequency of instruction transmission. Monitor device execution status via the CAN bus network, adjust control parameters in real time, and ensure accurate execution of the scheduling plan with instruction transmission latency ≤45 milliseconds.

[0066] Evaluation and optimization steps: Establish a real-time evaluation index system to calculate KPI indicators such as pallet turnover rate and equipment utilization rate. Adopt the federated transfer learning method to achieve cross-enterprise scheduling knowledge sharing under the premise of protecting the data privacy of each enterprise. The formula is Among them, θ is the global model parameter, θ0 is the pre-training parameter, L i is the loss function of the i-th enterprise, λ i is the weight coefficient, and μ is the regularization parameter. This reduces the system’s adaptation time in a new environment by more than 60%.

[0067] In this invention, the scheduling decision-making process constructs a multi-objective utility function, encompassing a KPI evaluation system encompassing 12 indicators: pallet turnover rate, equipment utilization rate, order fill rate, energy efficiency, inventory turnover rate, equipment failure rate, personnel utilization rate, transportation cost rate, space utilization rate, order processing timeliness, resource idleness rate, and scheduling response speed. A digital twin model compares the operating status of the physical system and the virtual model in real time. When a particular indicator fails to meet the target for three consecutive times, a meta-learning algorithm is triggered to optimize the scheduling model parameters. Simultaneously, a Nash bargaining decision model is established, incorporating indicators such as pallet turnover rate, equipment utilization rate, and order fill rate into the utility function. Contracts enable automated negotiation among all parties involved. When the Pareto optimal solution for a scheduling solution deviates from a threshold, the system automatically triggers a multi-objective optimization algorithm to dynamically adjust the weight coefficients of each indicator. A / B testing is used to compare the effectiveness of different weighting schemes, with the testing strategy automatically switching weekly to continuously optimize system performance. Blockchain technology is introduced to ensure full traceability of the scheduling decision-making process, and contracts are used to automatically execute the optimal scheduling solution. Through the federated learning mechanism, scheduling knowledge sharing and collaborative optimization are achieved while protecting the data privacy of all participants. When the system identifies a new scheduling scenario, it automatically calls the transfer learning algorithm to transfer and apply historical scheduling experience to the new scenario, quickly generating an initial scheduling plan.

[0068] In the present invention, the evaluation and optimization step adopts the federated transfer learning method. Based on the federated transfer learning architecture, a cross-enterprise scheduling knowledge federation sharing platform is established, and the full traceability and tamper-proofing of the model parameter transmission process are achieved through blockchain technology. When it is detected that the "cross-enterprise knowledge sharing accuracy" indicator is lower than the 92% threshold for 5 consecutive times, the federated learning parameter optimization algorithm is automatically triggered, and the model parameters are perturbed using differential privacy technology to improve the accuracy of knowledge sharing while ensuring data privacy. A knowledge distillation mechanism is introduced to refine the complex scheduling knowledge shared across enterprises into a lightweight model, which is deployed on edge computing nodes to reduce response delays. A federated learning aggregation algorithm is used to update the global model parameters after every 10 rounds of local model training, and the fairness of the contribution of each enterprise model is ensured through an adaptive weight adjustment strategy. During the model training process, the domain adaptation technology in transfer learning is used to align the scheduling features that differ between different enterprises, further improving the efficiency of cross-enterprise knowledge transfer.

[0069] The present invention further comprises the following steps:

[0070] The framework for the regional scheduling implementation and its relationship with existing systems: Existing multimodal intelligent pallet systems have achieved global state perception and path planning. This implementation, through a regional decomposition-collaboration architecture, divides the warehouse logistics network into several scheduling sub-areas (such as warehouse sorting areas, loading and unloading areas, and transshipment areas), forming a three-level linkage mechanism with the existing system: "global planning-regional execution-cross-regional collaboration." The key linkage points are as follows: Data layer: Reuses the UWB positioning data (accuracy ±10cm) of the existing pallet state perception module, divides the data collection unit by region, and maintains a refresh rate of 10 times / second; Algorithm layer: Embeds regional priority parameters into the existing LSTM demand forecasting and Hungarian task allocation algorithms; Execution layer: Based on the existing digital twin model, generates an independent topology map for each region, forming a hierarchical collaboration with the global path planning. II. Design of the core modules of regional scheduling and linkage mechanism 1. Regional division and state perception unit functional positioning: Divide the logistics network into M dynamic regions (such as storage shelf partitions and berth partitions), deploying edge nodes and local sensor networks in each region.

[0071] Linkage design: Dynamic regional division algorithm: Based on K-means clustering (existing visualization platform technology), the high-frequency use area of ​​pallets is analyzed. The formula is: Region i =K-means (UWB, M) regional state fusion: Each regional edge node summarizes pallet location and load data (existing sensor data) and uploads it to the global gateway via LoRaWAN, maintaining a data compression ratio of 10:1. 2. Regional collaborative scheduling engine functional positioning: Execute independent task allocation and path planning within the region and interact with the global scheduling system.

[0072] Core algorithm linkage: Regional task allocation: Improve the Hungarian algorithm and introduce the regional load balancing factor η i (value 0-1, η i The higher the value, the busier the area), the cost function is expanded to: where c ij Assign a cost to the existing task, η i Calculated based on indicators such as the AGV load rate and pallet idle rate within the region (data from the existing intelligent maintenance module). Regional path planning: Based on the existing improved A* algorithm, a local topology map is generated for each region. The dynamic weight factor α increases the regional congestion coefficient. The formula is: f(n) = g(n) + α·(h(n) + β·congestion_index), where congestion_index is calculated in real time based on the AGV density within the region.

[0073] Cross-regional resource pool management function positioning: Dynamically allocate pallet resources in each region, complementing the resource pool model of the existing intelligent maintenance module.

[0074] Linkage mechanism: Regional resource warning: When the idle rate of pallets in a certain area is less than 10% (existing threshold), cross-regional allocation is triggered, and the optimal allocation quantity is calculated based on the queuing theory formula (existing): where λ i / μ i is the demand rate / service rate of region i, ρ i is the resource utilization rate of region i. Cross-region transport coordination: Invoke the existing multi-AGV game obstacle avoidance algorithm to plan exclusive routes for cross-region pallet transport, ensuring that the inter-region transfer time is less than 30.

[0075] Regionalized scheduling execution process

[0076] Region initialization: Based on the historical pallet trajectory data (existing state perception module), the initial region division is generated by K-means clustering (such as Figure 1 module linkage);

[0077] Intra-regional scheduling: Demand forecasting: Each region independently runs the LSTM model (existing), and the forecast cycle is shortened to 10 minutes; Task allocation: Call the improved Hungarian algorithm to prioritize the allocation of idle pallets in the region (such as Figure 4 Task completion rate linkage); Cross-regional collaboration: When regional resources are insufficient, cross-regional deployment is triggered, and the existing path planning module is called to generate cross-regional transportation routes (such as Figure 2 Step 3 linkage).

[0078] Dynamic regional adjustment mechanism

[0079] Real-time monitoring: The existing edge server analyzes the pallet flow data of each area in real time. When the pallet turnover rate in the area fluctuates by more than 20%, the area is redivided. Adaptive optimization: The existing YOLO algorithm is used to identify equipment anomalies in the area (such as AGV failures) and dynamically adjust the area boundaries and resource allocation strategies.

[0080] Implementation effect and data linkage analysis

[0081]

[0082] Data association description:

[0083] When the number of tasks exceeds 30, regional scheduling increases the Hungarian algorithm completion rate from 93% to 96%, as regionalized allocation reduces the complexity of global scheduling; labor costs are further reduced by 10%, as regionalized scheduling reduces the need for cross-regional manual intervention and is linked to the existing visualization platform's smart dashboard (decision response time is shortened by 40%).

[0084] Technological innovation and linkage value

[0085] Layered collaborative architecture: Through "global-regional" dual-layer scheduling, the centralized decision-making of the existing system is transformed into distributed collaboration, improving scheduling robustness in large-scale scenarios;

[0086] Dynamic zone adaptation: Dynamically adjust zone divisions based on real-time data (existing sensors and edge computing), solving the problem that traditional fixed zones cannot adapt to fluctuations in logistics flow;

[0087] Energy consumption collaborative optimization: Regional scheduling and existing energy optimization modules are linked (such as AGV wireless charging strategy), which reduces the energy consumption of equipment in the region by 22% (such as Figure 6 data linkage).

[0088] in conclusion

[0089] This embodiment, through the regional scheduling concept, forms the following complementary relationships with the existing multimodal intelligent pallet system:

[0090] Spatial dimension: Decomposing global scheduling into efficient execution within a region and cross-regional coordination fills the gap of the existing system's "lack of local optimization" in large-scale logistics networks;

[0091] Time dimension: Combining high-frequency forecasts within a region (10 minutes) with global long-term planning (hourly) improves the response speed to sudden demands (from 500ms to 200ms);

[0092] Resource dimension: The cross-regional resource pool is linked with the existing pallet life prediction model to reduce the overall idle rate of pallets from 15% to 9%, achieving the scheduling goal of "regional self-balancing + global optimization".

[0093] The above are only preferred specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A pallet circulation scheduling system, characterized in that: Includes the following modules: Demand forecasting module: Based on historical order data, an improved seasonal decomposition time series model is used to predict pallet demand. Wavelet transform decomposition is combined to correct prediction errors, and an attention mechanism is introduced to assign weights to data from different historical periods. Real-time status perception module: RFID, visual recognition, and vibration sensors are deployed on pallets to collect real-time information on pallet position, load status, surface wear, and internal stress distribution; Use improved Kalman filter algorithm to fuse data and eliminate measurement noise; Scheduling decision module: Constructs a two-layer optimization decision model. The upper layer uses an improved particle swarm algorithm combined with a simulated annealing mechanism to solve the global optimal scheduling solution, while the lower layer dynamically adjusts local path planning through reinforcement learning. Execution control module: Builds a virtual mapping model of the physical pallet based on digital twin technology, and controls the virtual and real through real-time data interaction. It uses an event-triggered control mechanism to control the frequency of command transmission, and sends a command when the state error exceeds the trigger threshold. Evaluation and optimization module: Establish a real-time evaluation model based on digital twins, calculate system performance indicators by comparing the operating status of the physical system and the virtual model; use meta-learning algorithms to adapt to the new production environment and self-optimize the system.

2. The pallet circulation scheduling system according to claim 1, characterized in that: Also includes: Buffer management module: A dynamic adaptive buffer is set up between the production line and the storage area, and a fuzzy logic control algorithm is used to dynamically adjust the buffer capacity. A dynamic priority queue is introduced to manage pallets, and the priority is adjusted in real time based on the production rhythm and order urgency. An early warning is triggered when the buffer utilization rate exceeds the threshold. Abnormal processing module: Build an abnormal diagnosis model based on Bayesian network to calculate the probability of abnormal events in real time; when an abnormality is detected, a negotiation mechanism is used to generate the optimal emergency plan.

3. The pallet circulation scheduling system according to claim 2, characterized in that: The dynamic priority queue in the buffer management module uses a hyperbolic time discounting algorithm to calculate the pallet priority, and automatically adjusts the discount factor when the production line beat fluctuation exceeds the threshold.

4. The pallet circulation scheduling system according to claim 1, characterized in that: Also includes: Energy consumption optimization module: Establishes an equipment energy consumption prediction model based on deep reinforcement learning, and optimizes the scheduling strategy based on equipment load, operating speed, and start-stop frequency; Introducing an energy recovery mechanism to convert kinetic energy during AGV braking into electrical energy storage; System collaboration module: Build a trusted data sharing platform based on blockchain technology, use contracts to automatically execute cross-system business processes, and support enterprises' large-scale production scheduling data interaction needs.

5. The pallet circulation scheduling system according to claim 4, characterized in that: The blockchain in the system collaboration module adopts the practical Byzantine fault-tolerant PBFT consensus algorithm to support enterprise-level data interaction needs.

6. The pallet circulation scheduling system according to claim 1, characterized in that: The sensors in the real-time status perception module use low-power wide-area Internet of Things (LPWAN) technology to transmit data.

7. The pallet circulation scheduling system according to claim 1, characterized in that: In the scheduling decision module, the improved particle swarm algorithm introduces the Levy flight mechanism to search the global environment and optimize the path efficiency.

8. An application method of the pallet circulation scheduling system according to claim 1, characterized in that: The following steps are involved: Demand forecasting steps: Collect historical data and use the improved STL model combined with wavelet transform and attention mechanism to generate pallet demand forecast results; State perception step: Real-time pallet status information is collected through the sensor network, and the improved Kalman filter algorithm is used to fuse the data to build a digital twin model synchronization state; Scheduling decision-making steps: Construct a two-layer optimization model. The upper layer solves the global solution by improving the PSO algorithm combined with the simulated annealing mechanism, and the lower layer dynamically adjusts the path through reinforcement learning. Execution control steps: Generate control instructions based on the digital twin model, use event triggering mechanism to optimize communication load, and control the closed loop through virtual-real interaction; Evaluation and optimization steps: Establish a real-time evaluation indicator system, use meta-learning algorithms to adapt to new environments, and continuously optimize system parameters and scheduling strategies.

9. The application method of the pallet circulation scheduling system according to claim 8, characterized in that: When encountering target conflicts in the scheduling decision step, the Nash bargaining solution method is used to balance the target weights, and the priority is dynamically adjusted through the negotiation mechanism.

10. The application method of the pallet circulation scheduling system according to claim 8, characterized in that: The federated transfer learning method is used in the evaluation and optimization step to support cross-enterprise scheduling knowledge sharing while protecting data privacy.

Images