Intelligent factory logistics automatic scheduling system and method thereof
By combining adaptive workshop collaboration, robust dynamic scheduling and multi-grained knowledge migration algorithms in the intelligent factory logistics automation scheduling system, the problems of low scheduling efficiency and poor ability to deal with changes are solved, intelligent closed-loop scheduling is realized, and scheduling efficiency is improved. and robustness to adapt to the needs of intelligent manufacturing.
Patent Information
- Application Number
- CN202411213139.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-30
- Publication Date
- 2024-11-12
AI Technical Summary
Traditional factory logistics scheduling relies on manual labor, has low efficiency, lacks a global perspective and poor ability to respond to changes. The existing intelligent scheduling algorithms are not complete enough in modeling and characterization, multi-task collaboration and end-to-end control, and it is difficult to meet the needs of intelligent manufacturing.
Design an intelligent factory logistics automation scheduling system, integrating adaptive workshop collaborative algorithm, robust dynamic scheduling algorithm and multi-grained knowledge migration algorithm to form an end-to-end intelligent closed-loop scheduling solution, including data acquisition, scheduling optimization, planning execution, monitoring and early warning and system management modules to achieve global collaborative and robust control through machine learning and reinforcement learning.
It has achieved cross-workshop coordination, dynamic interference resistance and cross-scenario generalization, improved scheduling efficiency and robustness, formed a complete scheduling optimization solution, which can quickly adapt to changes in complex logistics scenarios, and significantly improved the execution of production plans. and the robustness of the system.
Smart Images

Figure CN118941058A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of automated dispatching systems, and more specifically, to an automated dispatching system and method for intelligent factory logistics. Background Art
[0002] Smart manufacturing is the core of the new round of industrial revolution, and smart logistics scheduling is the key to ensuring efficient production. Traditional factory logistics scheduling mainly relies on manual experience. Schedulers manually make material distribution plans and scheduling decisions based on production plans, inventory levels, equipment status and other information. This manual scheduling model has many limitations:
[0003] (1) High labor costs and low efficiency. Manual scheduling relies on the personal experience and business capabilities of the scheduler, and the decision-making process lacks standards. The scheduling efficiency is low and it is difficult to cope with the sharp increase in scheduling tasks caused by the expansion of factory scale.
[0004] (2) Lack of global perspective and one-sided scheduling decisions. Dispatchers often only focus on a single or local scheduling goal, such as workshop material distribution, and it is difficult to fully consider cross-workshop collaboration and the linkage and coordination of different logistics nodes. The decision-making perspective is limited and it is difficult to achieve the global optimum.
[0005] (3) Poor ability to cope with changes and insufficient execution. In the face of emergencies such as production fluctuations and equipment failures, it is difficult for dispatchers to quickly re-plan and adjust. The scheduling plan lacks robustness and flexibility, resulting in frequent logistics chaos and poor execution of production plans.
[0006] In view of the limitations of manual scheduling, the industry has introduced computer technology to assist scheduling decisions. The mainstream is to use mathematical modeling methods of operations research optimization to abstract scheduling problems into optimization models such as integer programming and linear programming, and then solve the model through a solver to obtain the optimal scheduling decision. For example, for the scheduling problem of flexible job shops, a multi-objective optimization model is established, and an improved genetic algorithm is designed for solution; considering the collaboration of multiple AGVs, a nonlinear integer programming model is constructed, and the column generation method is used for optimization and solution.
[0007] However, this type of scheduling method based on precise optimization algorithm still has shortcomings:
[0008] (1) The modeling capability is limited, making it difficult to accurately describe the complex constraints of the scheduling scenario. The scheduling model is out of touch with actual problems, and the optimization results are not practical enough.
[0009] (2) Low optimization efficiency and difficulty in handling large-scale scheduling problems. Factory logistics scheduling is an NP-hard problem with high solution complexity. Heuristic algorithms are prone to premature stagnation, and precise optimization algorithms take a long time to solve, making it difficult to meet the needs of real-time scheduling.
[0010] (3) Poor generalization ability, difficult to adapt to scenario changes. Different products, processes, layouts, optimization objectives, and constraints vary greatly. Each change requires re-modeling and solving, which has high development and maintenance costs. The optimization algorithm lacks versatility and scalability.
[0011] In recent years, the development of artificial intelligence technology has brought new ideas to intelligent scheduling optimization. Among them, reinforcement learning gradually learns the optimal decision-making strategy through the interactive trial and error between the intelligent agent and the environment, providing a new way to solve complex scheduling problems. For example, deep reinforcement learning is used for production scheduling of discrete manufacturing, and the adaptive optimization of scheduling decisions is achieved by approximating the value function through neural networks; some scholars have proposed an intelligent workshop scheduling model based on state-action-reward-state-action (SARSA), which optimizes scheduling performance by exploring and utilizing balance dynamics.
[0012] However, existing intelligent scheduling algorithms still have limitations:
[0013] (1) The use of single-agent reinforcement learning algorithms such as tabular Q-learning limits the modeling of environmental states and action spaces, and the dimensionality curse of state representation makes it difficult to characterize the feature space of complex scheduling problems.
[0014] (2) Most of them focus on a single scheduling problem and lack overall consideration of multi-workshop coordination and human-machine collaboration, and have not formed a universal system solution.
[0015] (3) Most of them rely on simulation environments for offline training and optimization, lack end-to-end real-time closed-loop control capabilities, and are unable to cope with the dynamic changes and uncertain interference in complex logistics scenarios.
[0016] In summary, the low efficiency and limited optimization capabilities of traditional manual scheduling can no longer meet the development needs of intelligent manufacturing, and it is urgent to introduce intelligent scheduling methods. Although operations research optimization can assist in solving scheduling decisions, its modeling capabilities and optimization efficiency are difficult to meet actual needs. Intelligent scheduling optimization based on reinforcement learning is the general trend, but the existing algorithms are not yet complete and mature in terms of modeling representation, multi-task collaboration, and end-to-end control. Therefore, it is urgent to develop an intelligent factory logistics automation scheduling system and method. Summary of the invention
[0017] It is in this context that the present invention is proposed, and aims to address the limitations of the existing technology by providing an intelligent factory logistics automation scheduling system and method thereof, which integrates a variety of intelligent optimization algorithms, possesses capabilities such as global coordination, robust control, and knowledge reuse, forms an end-to-end intelligent closed-loop scheduling solution, and fully enables factory logistics scheduling.
[0018] The present invention provides an intelligent factory logistics automation scheduling system, comprising: a data acquisition module, a scheduling optimization module, a plan execution module, a monitoring and early warning module and a system management module;
[0019] The data collection module is used to collect real-time data on workshop production progress, warehouse inventory, and material distribution, and to perform data preprocessing and storage management;
[0020] The scheduling optimization module establishes a multi-objective scheduling optimization model based on a machine learning algorithm and dynamically generates the optimal scheduling decision;
[0021] The plan execution module converts the scheduling plan into an instruction sequence executable by the logistics equipment to realize the automatic distribution and handling of materials;
[0022] The monitoring and early warning module tracks the logistics execution status and issues early warning and adjustments for abnormal situations;
[0023] The system management module provides task management, parameter configuration, and user management system functions to ensure the smooth operation of the system.
[0024] Specifically, the scheduling optimization module includes an adaptive workshop collaboration algorithm, a robust dynamic scheduling algorithm and a multi-granularity knowledge transfer algorithm.
[0025] Specifically, the adaptive workshop collaboration algorithm includes the following steps:
[0026] Step 1: State representation, define the system state S =<Nm,Ns,Nw,P,T> , where Nm, Ns, and Nw represent the material cumulative demand matrix, current inventory matrix, and in-transit material matrix of each workshop, respectively, P is the production plan matrix, and T is the delivery date matrix;
[0027] Step 2: Policy iteration, initialize the Q(S,A) value function, and select the scheduling action A based on the ε-greedy strategy.<Ax,Ay,Aq> , where Ax and Ay are the scheduling paths and quantities of each material, respectively, Aq is the AGV scheduling sequence, after executing A, the state is transferred to S′, and the immediate collaborative benefit R(S,A)=w1*U+w2*C+w3*D is obtained, where U, C, and D are the cumulative material demand satisfaction, logistics cost, and delivery date default, respectively, and w is the weight coefficient. Update the Q value: Q(S,A)←(1-1)α)Q(S,A)+α[R+γ*maxO(S′,A′)], where α is the learning rate and γ is the discount factor;
[0028] Step 3: Policy improvement, apply the policy gradient theorem to the transfer sample (S, A, R, S′) Where π(A|S) is the random strategy, β is the strategy learning rate, For the gradient operator, repeat Step 2-3 until convergence.
[0029] 8. The intelligent factory logistics automation scheduling system according to claim 2, characterized in that the robust dynamic scheduling algorithm comprises the following steps:
[0030] Step 1: Establish a robust optimization model, the objective function min∑c(x,ξ), where x is the scheduling decision variable, ξ is the uncertain interference variable, c(x,ξ) is the scheduling target cost function, and the constraints are: G(x,ξ)≤0 and H(x,ξ)=0, Where U is the uncertainty set of ξ, G and H are resource constraint and timing constraint functions respectively;
[0031] Step 2: Convert it into a robust counterpart. By introducing an appropriate slack variable δ, the optimization model with uncertain parameters is converted into a linear optimization problem:
[0032] min∑c(x,ξ),stG(x,ξ)+M G δ G ≤0,H(x,ξ)+M H δ H =0,∑(δ G +δ H )≤Γ,δ G ,δ H ≥0, where M G 、M H is a constant matrix, Γ controls the robustness level;
[0033] Step 3: Rolling optimization solution. In each scheduling cycle t, observe the current system state x[t], solve the robust counterpart to obtain the scheduling decision x[t:t+T] for the next T cycles, and implement x * [t], rolling forward time domain, repeating optimization until termination.
[0034] Specifically, the multi-granularity knowledge transfer algorithm includes the following steps:
[0035] Step 1: Scene similarity measurement. Given the source scene S and the target scene T, define the similarity Sim(S,T) = w1*Sim(G S ,G T )+w2*Sim(F S ,F T )+w3*Sim(D S ,D T ), where w1+w2+w3=1, G, F, and D represent the production line layout, process parameters, and demand distribution characteristics, respectively;
[0036] Step 2: Multi-granularity knowledge matching, set the similarity threshold ε h >ε m >ε l, if Sim(S,T)>ε h , then K T ={K H ,φ,φ}; if ε m <Sim(S,T)≤ε h , then K T ={K H ,K M ,φ}; otherwise K T ={K H ,K M ,K L}, where K T is the initial scheduling knowledge of T scenarios, K H , K M , K L They are high, medium, and low granularity knowledge migrated from S, respectively;
[0037] Step 3: Knowledge fine-tuning, is the final scheduling knowledge of T scenario, Where D T The scheduling data for the T scenario is fine-tuned to be based on D T K T The process of fine-tuning parameters.
[0038] Specifically, the plan execution module is used to:
[0039] Translate the scheduling plan into the AGV transportation task table T AGV <src,dst,mat,qty> and robot task table T R <act,obj,param> , where src, dst, mat, and qty represent the starting point, end point, material, and quantity respectively, and act, obj, and param represent the action, object, and parameter respectively; T AGV 、T R It is sent to the AGV cart and robot logistics equipment to synchronize the execution progress; when an abnormality occurs in the equipment, such as obstacle avoidance or failure, it is reported to the scheduling module and waits for rescheduling.
[0040] Specifically, the monitoring and early warning module is used to:
[0041] Real-time collection of material delivery status S M <on t ime r ate,delay t ime>
[0042] and equipment operating status S E <workload,failure r ate>,
[0043] where t ime r ate、delay t ime, workload, failure r ate represents punctuality rate, delay time, load rate, and failure rate respectively; set the KPI warning threshold. M or S E When the threshold is exceeded, an early warning is automatically triggered to process multi-source heterogeneous data, generate visual logistics operation reports, and assist management decisions.
[0044] Specifically, the system management module is used to:
[0045] Provide task-oriented full life cycle management, including task creation, modification, deletion, query, and statistics;
[0046] Manage the hyperparameters required by machine learning algorithms, and provide parameter configuration, online adjustment, and save and load functions;
[0047] Carry out user management, provide account management, authority allocation, and operation audit functions to ensure system security.
[0048] Specifically, the data acquisition module includes:
[0049] RFID reader, frequency 920MHz, reading distance 0-6m;
[0050] Industrial-grade camera, resolution 1920*1080, frame rate 60fps;
[0051] Magnetic navigation AGV car, type, load 100kg, positioning accuracy ±5mm;
[0052] The data acquisition module collects production order number, material code, inventory, AGV location and power data through the above hardware, and organizes and stores them in time series.
[0053] A dispatching control method based on the intelligent factory logistics automation dispatching system comprises the following steps:
[0054] Step 1: Collect real-time data of each logistics node in the workshop and warehouse through the data acquisition module, and store it in the database after pre-processing;
[0055] Step 2: The scheduling optimization module reads real-time data, combines production plans and orders, calls adaptive workshop collaboration algorithms, robust dynamic scheduling algorithms, and multi-granularity knowledge transfer algorithms to generate optimal scheduling decisions;
[0056] Step 3: The plan execution module converts the scheduling decision into control instructions for the AGV and robot, and sends them to the equipment for execution;
[0057] Step 4: The monitoring and early warning module tracks the real-time status of logistics, triggers early warnings based on preset rules, and generates data reports;
[0058] Step 5: Feedback the scheduling performance to the optimization module, continuously improve the algorithm, and record the audit scheduling process through the system management module.
[0059] The present invention has the following beneficial effects:
[0060] 1. Multiple algorithms collaborate to enhance efficiency. The system integrates multiple optimization algorithms such as adaptive workshop collaboration, robust dynamic scheduling, and multi-granularity transfer learning to enhance scheduling capabilities from the perspectives of cross-workshop collaboration, dynamic interference resistance, and cross-scenario generalization. The algorithms work together to form a complete scheduling optimization solution. For example, the cross-workshop collaboration strategy learned by the adaptive collaborative algorithm can be used as prior knowledge for the robust optimization algorithm to accelerate algorithm convergence; the scheduling features extracted by the multi-granularity transfer algorithm can be used to improve the state representation of the adaptive collaborative algorithm. Through mutual empowerment between algorithms, a synergistic efficiency of "1+1>2" is generated.
[0061] 2. Modular design, high cohesion and low coupling. Modules such as data collection, scheduling optimization, plan execution, monitoring and early warning perform their respective duties and combine to form a complete intelligent scheduling solution. The boundaries of each module are clear, the responsibilities are clear, and the loose coupling design facilitates the expansion and maintenance of the system. At the same time, there is a high degree of collaboration between modules. For example, the data collection module provides data support for optimization decisions, the execution module puts the optimization results into action, and the monitoring module evaluates and feedbacks the optimization effect. Multiple modules are seamlessly connected to ensure the efficient operation of the closed-loop scheduling process.
[0062] 3. End-to-end connection and real-time closed-loop control. The data collection module uses RFID and sensors to realize digital collection of workshops, warehouses, and logistics. The optimization module relies on machine learning to realize the automatic generation of scheduling plans. The execution module realizes the automatic execution of scheduling plans through AGV and robots. The monitoring module performs real-time tracking and evaluation and feedback optimization. End-to-end connection of data flow, decision flow, and execution flow forms a real-time closed-loop control, enabling the scheduling system to accurately regulate and globally optimize all aspects of logistics.
[0063] 4. Strong robustness and good anti-interference ability. Faced with various uncertain interferences in complex scenarios such as workshop production and logistics distribution, this system adopts a robust optimization algorithm, considers interference factors, generates scheduling decisions with a certain margin, and resists interference. At the same time, the system has a certain online scheduling capability. When an interference event is detected, it can quickly re-schedule based on global information, resolve risks, improve system robustness, and ensure the smooth execution of production plans.
[0064] 5. Good generalization and rapid replication and deployment. Factory logistics scenarios are complex and changeable, and the scheduling of different products, production lines, and processes varies greatly. It is difficult for a general scheduling system to meet personalized needs. To address this, this system uses multi-granularity transfer learning to learn and transfer scheduling knowledge from similar scenarios, greatly reducing the demand for training data in the target scenario and achieving rapid deployment of the scheduling algorithm. Through similarity measurement and selective reuse, scheduling knowledge can be reused and shared between different scenarios, significantly improving the system's generalization and deployment efficiency.
[0065] 6. Data-driven, continuous optimization. This system is based on the data-driven machine learning paradigm. By continuously learning the accumulated scheduling experience data, it realizes the continuous iterative optimization of scheduling strategies and algorithms. On the one hand, the various types of data such as logistics distribution and production operation generated during the operation of the system provide a wealth of training samples for the algorithm. The algorithm gradually converges to the optimal strategy network through continuous trial and error learning. On the other hand, the system compares and analyzes the KPI indicators before and after optimization, finds out the weak links, and optimizes the system in a targeted manner through parameter tuning and other methods. The scheduling capability of the system can be continuously improved with the accumulation of experience data.
[0066] In summary, the present invention uses cutting-edge artificial intelligence technologies such as multi-agent reinforcement learning, operations optimization, and transfer learning to build a data-driven intelligent logistics scheduling system, realize end-to-end automation and intelligent scheduling, and significantly improve the scheduling efficiency, robustness, and generalization capabilities in complex dynamic scenarios, accelerating the digitalization and intelligent transformation of industrial logistics. It has broad application prospects in the field of intelligent manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS
[0067] Figure 1 This is a schematic diagram of the framework of an intelligent factory logistics automation scheduling system of the present invention. DETAILED DESCRIPTION
[0068] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of an intelligent factory logistics automation scheduling system proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.
[0069] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0070] The specific scheme of the intelligent factory logistics automation scheduling system provided by the present invention is described in detail below with reference to the accompanying drawings.
[0071] Example 1
[0072] See also Figure 1 The present invention provides an intelligent factory logistics automation scheduling system, which is mainly composed of the following modules: data acquisition module 1, scheduling optimization module 2, plan execution module 3, monitoring and early warning module 4 and system management module 5. Data acquisition is the basis for realizing intelligent scheduling. Data acquisition module 1 uses the Internet of Things technology to deploy sensors at various logistics nodes such as workshops, warehouses, and transportation equipment to collect logistics data in real time. In terms of hardware selection, the RFID reader has an operating frequency of 920MHz and a reading distance of 0-6m, which can efficiently collect the identification and location information of materials; the industrial-grade camera has a resolution of 1920*1080 and a frame rate of 60fps, which can accurately capture the movement status of materials and equipment; the handling equipment uses a magnetic navigation AGV trolley with a load capacity of 100kg and a positioning accuracy of ±5mm, which can flexibly adapt to complex workshop environments. Through the above hardware, multi-source heterogeneous data such as production order number, material code, inventory, AGV position and power can be collected. The collected real-time data is organized according to time series and stored in a time series database such as InfluxDB to provide data support for subsequent scheduling decisions. At the same time, the original data is preprocessed by cleaning and standardization to improve data quality.
[0073] Scheduling Optimization Module 2 is the core of this system. This module comprehensively considers factors such as production plan, material demand, inventory status, equipment capacity, etc., builds a scheduling optimization model, and designs an efficient machine learning algorithm to solve the optimal scheduling decision. The innovation of this module is that three optimization algorithms are proposed for different scheduling scenarios: adaptive workshop collaboration algorithm, robust dynamic scheduling algorithm, and multi-granularity knowledge transfer algorithm. Through the coordinated cooperation of algorithms, complex and changeable factory logistics needs can be met. The specific algorithm explanation is as follows:
[0074] 1. The MAAC algorithm models each workshop as an intelligent agent, explores the collaborative scheduling strategy between workshops through reinforcement learning, and adaptively modifies the strategy based on feedback, including the following steps:
[0075] Step 1: State representation. Define the system state S =<Nm,Ns,Nw,P,T> , where Nm, Ns, and Nw are the material cumulative demand matrix, inventory matrix, and in-transit material matrix of each workshop, respectively, P is the production plan matrix, and T is the delivery date matrix.
[0076] Step 2: Policy iteration. Initialize the Q(S,A) value function and select the joint scheduling action A based on the ε-greedy strategy.<Ax,Ay,Aq> , where Ax and Ay are the scheduling paths and quantities of each material, and Aq is the AGV scheduling sequence. After executing A, the state is transferred to S′, and the immediate collaborative benefit R(S,A)=w1U+w2C+w3D is obtained, where U, C, and D are material demand satisfaction, logistics cost, and delivery default, respectively, and w is the weight coefficient. Update the Q value based on the Bellman equation: Q(S,A)←(1-1)α)Q(S,A)+α[R+γmaxQ(S′,A′)], α is the learning rate, and γ is the discount factor.
[0077] Step 3: Policy improvement. For the sampled data (S, A, R, S′), use the policy gradient theorem to update the policy network parameters θ: θ←θ+β▽logπ(A|S)Q(S,A), π is the stochastic policy, ▽ is the gradient operator, and β is the learning rate. Repeat Step 2-3 until the policy converges.
[0078] For example, when workshop 1 is short of materials and workshop 2 is in surplus, the collaborative strategy learned by MAAC may be: adjust Ax to transfer materials from workshop 2 to workshop 1; optimize Ay and Aq to make AGV prioritize the delivery of materials to workshop 1. After multiple rounds of iterations, each workshop gradually formed a "mutually beneficial" collaborative mode, and the overall logistics efficiency was improved.
[0079] 2. The robust dynamic scheduling algorithm (RDS) combines rolling time domain optimization with robust optimization to generate rolling dynamic scheduling decisions and consider interference uncertainty to improve the robustness of scheduling. It includes the following steps:
[0080] Step 1: Modeling. Considering the uncertainty interference ξ of the scheduling task, such as equipment failure, material delay, etc., a robust scheduling optimization model is constructed: Objective: min∑c(x,ξ), Constraints: G(x,ξ)≤0,H(x,ξ)=0, Where x is the scheduling decision variable, such as equipment startup time, material delivery batch, etc., c(x,ξ) is the scheduling objective function, such as minimizing production cycle, logistics cost, etc., G and H are resource constraint and timing constraint functions respectively, and U is the uncertainty set of ξ.
[0081] Step 2: Transformation. Introduce appropriate slack variables δ and transform the above model into the following linear programming problem:
[0082] min∑c(x,ξ)
[0083] stG(x,ξ)+M1δ1≤0,H(x,ξ)+M2δ2=0
[0084] ∑(δ1+δ2)≤Γ,δ1,δ2≥0.
[0085] Among them, M1 and M2 are constant matrices of appropriate dimensions, and Γ controls the robustness.
[0086] Step 3: Rolling optimization. In each scheduling cycle t, observe the current system state x[t], solve the model in step 2, obtain the scheduling decision x[t, t+T] for the next T cycles, implement the decision x[t], roll update t, and repeat the solution until termination.
[0087] For example, when an AGV failure causes material delivery delays, RDS re-optimizes scheduling in the new time domain based on the robustness level set by Γ, and promptly adjusts equipment startup time, material delivery sequence, etc., to minimize the losses caused by the failure and ensure the execution of the production plan.
[0088] 3. The Multi-granularity Knowledge Transfer Algorithm (MGKT), when deployed for new scenarios, extracts scheduling knowledge of different granularities from similar scenarios to assist the scheduling decision of the target scenario, including the following steps:
[0089] Step 1: Scene similarity. For the source scene S and the target scene T, define the similarity:
[0090] Sim(S,T)=w1Sim(Gs,Gt)+w2Sim(Fs,Ft)+w3*Sim(Ds,Dt)
[0091] Among them, Gs, Fs, and Ds represent the production line layout, process parameters, and demand distribution characteristics, respectively, and w1+w2+w3=1.
[0092] Step 2: Multi-granularity matching. Set the threshold ε1>ε2>ε3. When Sim(S,T)>ε1, migrate high-granularity strategy knowledge Kh from S; when ε1≥Sim>ε2, migrate Kh and medium-granularity process knowledge Km; when Sim<ε3, migrate Kh, Km and low-granularity data features Kl.
[0093] Step 3: Fine-tuning. Using a small amount of scheduling data Dt of scenario T, fine-tune the migration knowledge to obtain scheduling knowledge specific to scenario T:
[0094] Kt←finetune(Kh+Km+Kl,Dt)
[0095] For example, when it is necessary to plan logistics scheduling for a new production line, MGKT matches production lines with similar layouts and processes from historical mature production lines, extracts their material distribution strategies (high granularity), picking processes (medium granularity), inventory data distribution (low granularity) and other knowledge as a priori for optimizing the scheduling of the new production line, and then optimizes it in combination with actual operating data, significantly accelerating the deployment and iteration of the new production line logistics system.
[0096] The planning execution module 3 is responsible for converting the scheduling decision output by the scheduling optimization module into an operational logistics control instruction and issuing it for execution. First, the scheduling decision is translated into a task table for different devices. For the AGV car, a task table TAGV is generated.<src,dst,mat,qty> , which means transporting the material mat from the starting point src to the end point dst, with the transport quantity qty; for the robot, generate the task table TR<act,obj,par> , which means to execute the action act on the target obj, and the action parameter is par, such as grabbing, assembling, etc. Then, select industrial communication protocols such as MQTT to distribute the task table to each logistics equipment. At the same time, monitor the execution status feedback of the equipment, such as task completion, failure, etc., and update the task status in real time to facilitate scheduling and tracking. In addition, this module also needs to handle scheduling emergencies. For example, when the AGV avoids obstacles or the battery is insufficient, it is necessary to start the emergency scheduling process, including task reallocation, path replanning, etc., to minimize the impact of the event and restore normal scheduling in time. Emergency scheduling can use rules or learning algorithms, which will not be described in this article.
[0097] Monitoring and early warning module 4 To ensure the smooth operation of the dispatching system, it is necessary to monitor the system operation status in real time and issue early warnings in time when abnormal risks are found. The core indicators of monitoring include two categories: one is the logistics distribution status, such as on-time delivery rate on_time_rate, average delay time delay, etc.; the other is the equipment operation status, such as workload rate, failure rate failure_rate, etc. Set thresholds for indicators, such as on_time_rate <90% or failure_rate>5%, triggering yellow, red and other levels of early warnings, prompting management personnel to handle.
[0098] Monitoring data comes from production logistics management systems such as WMS, MES, and SCADA, and needs to be processed by data cleaning, mining, and visualization to facilitate intuitive analysis by managers. On the one hand, the data is analyzed in multiple dimensions to generate daily / monthly logistics operation reports to display the completion of various KPI indicators and identify weak links; on the other hand, appropriate machine learning algorithms are used to predict abnormal risks such as equipment failures and delivery delays, and avoidance measures are taken in advance. In addition, monitoring and early warning information can be fed back to the optimization module to assist in parameter adjustment and continuous learning of the optimization algorithm. For example, if a delivery route is frequently delayed, the time tolerance parameter of the route can be appropriately increased, or failure cases of the route can be added to the training data to enhance the robustness of the optimization algorithm.
[0099] System management module 5 provides a series of auxiliary management functions to ensure the efficient and stable operation of the scheduling system. The first is task management, which covers the full life cycle management of scheduling tasks. For example, when a task is issued, the task attributes, priority, deadline, etc. are entered to form a task list; during the task execution, the task status is tracked and the priority is adjusted; after the task is completed, the task completion status is counted and the scheduling performance is comprehensively evaluated.
[0100] The second is parameter management. Intelligent scheduling algorithms usually have many parameters, such as learning rate α, discount factor γ, robustness level Γ, etc. Flexible parameter configuration functions need to be provided to visualize parameters and corresponding scheduling benefits, and assist managers in parameter optimization. In addition, functions such as saving, loading, and reusing parameter groups need to be provided to facilitate rapid deployment in different scenarios.
[0101] Secondly, user and permission management. The system supports the registration and verification of users in different positions, and strictly controls the user's access rights to data and policies to prevent malicious tampering and data leakage. At the same time, it records the user's core operation logs and conducts regular security audits to improve system security.
[0102] The present invention also discloses a dispatching control method based on an intelligent factory logistics automation dispatching system, comprising the following steps:
[0103] Step 1: Through the data acquisition module 1, the real-time operation data of each logistics node such as the workshop and warehouse is collected, and after data preprocessing, it is stored in the database in time series to provide a data basis for scheduling decisions.
[0104] Step 2: Scheduling optimization module 2 reads real-time logistics data, combines production orders, material plans, etc., and builds a multi-objective scheduling optimization model. According to different scenarios, one or more of the adaptive workshop collaboration algorithm, robust dynamic scheduling algorithm, and multi-granularity knowledge transfer algorithm are flexibly selected to dynamically solve the optimal scheduling decision and output the scheduling plan for the future period.
[0105] Step 3: The plan execution module 3 translates the scheduling schedule into an executable task table for AGVs, robots, etc., and sends it to each logistics device through communication protocols such as MQTT, monitors device feedback, and updates the task status. When an emergency occurs, the emergency scheduling process is started to ensure the smooth execution of logistics tasks.
[0106] Step 4: Monitoring and early warning module 4 collects equipment status data in real time, calculates logistics performance indicators, and implements automatic early warning based on preset thresholds. At the same time, it uses visualization methods to generate logistics operation reports, analyze system bottlenecks in multiple dimensions, explore abnormal root causes, and assist management decisions.
[0107] Step 5: Optimization iteration. On the one hand, the actual logistics operation data is fed back to the scheduling optimization module for retraining and optimization of the scheduling algorithm; on the other hand, the scheduling performance is fed back to the system management module 5 to continuously improve the system through parameter adjustment, task management, etc.
[0108] The above steps are iterated in a cycle to form a closed-loop intelligent scheduling process, which enables the system to adaptively optimize according to actual operating conditions and continuously improve scheduling efficiency and robustness.
[0109] In addition, the system management module runs through the entire scheduling process. Through the task management function, it coordinates the work of each business module; through the parameter management function, it dynamically adjusts the parameters of each algorithm model to optimize system performance; through the authority management function, it ensures the security and controllability of the system.
[0110] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification, replacement, and improvement made within the principles of the present invention should be included in the protection scope of the present invention.
Claims
1. An intelligent factory logistics automation scheduling system, characterized by , including: data acquisition module, scheduling optimization module, plan execution module, monitoring and early warning module and system management module; The data collection module is used to collect real-time data on workshop production progress, warehouse inventory, and material distribution, and to perform data preprocessing and storage management; The scheduling optimization module establishes a multi-objective scheduling optimization model based on a machine learning algorithm and dynamically generates the optimal scheduling decision; The plan execution module converts the scheduling plan into an instruction sequence executable by the logistics equipment to realize the automatic distribution and handling of materials; The monitoring and early warning module tracks the logistics execution status and issues early warning and adjustments for abnormal situations; The system management module provides task management, parameter configuration, and user management system functions to ensure the smooth operation of the system.
2. The intelligent factory logistics automation scheduling system according to claim 1 is characterized in that ,The scheduling optimization module includes an adaptive workshop collaboration algorithm, a ,robust dynamic scheduling algorithm and a multi-granularity knowledge transfer algorithm.
3. The intelligent factory logistics automation scheduling system according to claim 2 is characterized in that ,The adaptive workshop collaboration algorithm includes the following steps: Step 1: State representation, define the system state S =<Nm,Ns,Nw,P,T> , where Nm, Ns, and Nw represent the material cumulative demand matrix, current inventory matrix, and in-transit material matrix of each workshop, respectively, P is the production plan matrix, and T is the delivery date matrix; Step 2: Policy iteration, initialize the Q(S,A) value function, and select the scheduling action A based on the ε-greedy strategy.<Ax,Ay,Aq> , where Ax and Ay are the scheduling paths and quantities of each material, respectively, and Aq is the AGV scheduling sequence. After executing A, the state is transferred to S′, and the immediate collaborative benefit R(S,A)=w1*U+w2*C+w3*D is obtained, where U, C, and D are the cumulative material demand satisfaction, logistics cost, and delivery date default, respectively, and w is the weight coefficient. Update the Q value: Q(S,A)←(1-1)α)A(S,A)+α[R+γ*maxQ(S′,A′)], where α is the learning rate and γ is the discount factor; Step 3: Policy improvement, apply the policy gradient theorem to the transfer sample (S, A, R, S′) Where π(A|S) is the random strategy, β is the strategy learning rate, For the gradient operator, repeat Step 2-3 until convergence.
4. The intelligent factory logistics automation scheduling system according to claim 2 is characterized in that ,The robust dynamic scheduling algorithm includes the following steps: Step 1: Establish a robust optimization model with the objective function min∑c(x,ξ), where X is the scheduling decision variable, ξ is the uncertain interference variable, c(x,ξ) is the scheduling target cost function, and the constraints are: G(x,ξ)≤0 and Where U is the uncertainty set of ξ, G and H are resource constraint and timing constraint functions respectively; Step 2: Convert it into a robust counterpart. By introducing an appropriate slack variable δ, the optimization model with uncertain parameters is converted into a linear optimization problem: min∑c(x,ξ),stG(x,ξ)+M G δ G ≤0,H(x,ξ)+M H δ H =0,∑(δ G +δ H )≤Γ,δ G ,δ H ≥0, where M G 、M H is a constant matrix, Γ controls the robustness level; Step 3: Rolling optimization solution. In each scheduling cycle t, observe the current system state x[t], solve the robust counterpart to obtain the scheduling decision x[t:t+T] for the next T cycles, and implement x * [t], rolling forward time domain, repeating optimization until termination.
5. The intelligent factory logistics automation scheduling system according to claim 2 is characterized in that ,The multi-granularity knowledge migration algorithm includes the following steps: Step 1: Scene similarity measurement. Given the source scene S and the target scene T, define the similarity Sim(S,T) = w1*Sim(G S ,G T )+w2*Sim(F S ,F T )+w3*Sim(D S ,D T ), where w1+w2+w3=1, G, F, and D represent the production line layout, process parameters, and demand distribution characteristics, respectively; Step 2: Multi-granularity knowledge matching, set the similarity threshold ε h >ε m >ε l , if Sim(S,T)>ε h , then K T ={K H ,φ,φ}; if ε m <Sim(S,T)≤ε h , then K T ={K H ,K M ,φ}; otherwise K T ={K H ,K M ,K L }, where K T is the initial scheduling knowledge of T scenarios, K H , K M , K L They are high, medium, and low granularity knowledge migrated from S, respectively; Step 3: Knowledge fine-tuning, is the final scheduling knowledge of T scenario, Where D T The scheduling data for the T scenario is fine-tuned to be based on D T To K T The process of fine-tuning parameters.
6. The intelligent factory logistics automation scheduling system according to claim 1 is characterized in that ,The plan execution module is used to: Translate the scheduling plan into the AGV transportation task table T AGV <src,dst,mat,qty> and robot task table T R <act,obj,param> , where src, dst, mat, and qty represent the starting point, end point, material, and quantity respectively, and act, obj, and param represent the action, object, and parameter respectively; T AGV , T R It is sent to the AGV cart and robot logistics equipment to synchronize the execution progress; when an abnormality occurs in the equipment, such as obstacle avoidance or failure, it is reported to the scheduling module and waits for rescheduling.
7. The intelligent factory logistics automation scheduling system according to claim 1 is characterized in that The monitoring and early warning module is used for: Real-time collection of material delivery status S M <on t ime r ate,delay t ime> and equipment operating status S E <workload,failure r ate>, Among them, t ime r ate、delay t ime, workload, failure r ate represents punctuality rate, delay time, load rate, and failure rate respectively; set the KPI warning threshold. M or S E When the threshold is exceeded, an early warning is automatically triggered to process multi-source heterogeneous data, generate visual logistics operation reports, and assist management decisions.
8. The intelligent factory logistics automation scheduling system according to claim 1 is characterized in that , the system management module is used to: Provide task-oriented full life cycle management, including task creation, modification, deletion, query, and statistics; Manage the hyperparameters required by machine learning algorithms, and provide parameter configuration, online adjustment, and save and load functions; Carry out user management, provide account management, authority allocation, and operation audit functions to ensure system security.
9. The intelligent factory logistics automation scheduling system according to claim 1 is characterized in that , the data acquisition module includes: RFID reader, frequency 920MHz, reading distance 0-6m; Industrial-grade camera, resolution 1920*1080, frame rate 60fps; Magnetic navigation AGV car, type, load 100kg, positioning accuracy ±5mm; The data acquisition module collects production order number, material code, inventory, AGV location and power data through the above hardware, and organizes and stores them in time series.
10. A dispatching control method based on the intelligent factory logistics automation dispatching system according to claims 1-9, characterized in that , including the following steps: Step 1: Collect real-time data of each logistics node in the workshop and warehouse through the data acquisition module, and store it in the database after pre-processing; Step 2: The scheduling optimization module reads real-time data, combines production plans and orders, calls adaptive workshop collaboration algorithms, robust dynamic scheduling algorithms, and multi-granularity knowledge transfer algorithms to generate optimal scheduling decisions; Step 3: The plan execution module converts the scheduling decision into control instructions for the AGV and robot, and sends them to the equipment for execution; Step 4: The monitoring and early warning module tracks the real-time status of logistics, triggers early warnings based on preset rules, and generates data reports; Step 5: Feedback the scheduling performance to the optimization module, continuously improve the algorithm, and record the audit scheduling process through the system management module.
Citation Information
Cited By
Logistics and warehousing system optimization method based on multi-agent system
CN119273122A
Goods dispatching and distribution management system suitable for stocker vertical warehouse
CN119349075A
Intelligent flow machine dynamic collaborative operation management platform based on digital storage yard
CN119379226A
Material distribution regulation and control method of robot laser welding work center and related device
CN119476880A
Garbage truck real-time monitoring and remote scheduling method based on Internet of Things
CN119647912A
