A SPS trolley material distribution method and system based on digital twin

Abstract

The present invention proposes a material distribution method and system for SPS trolleys based on digital twins, and the method includes: installing multiple sensors in the factory environment, preprocessing real-time operation data, and constructing a digital twin model of the physical factory and the SPS trolley system; constructing a scheduling model based on reinforcement learning according to the digital twin model, and optimizing the material distribution path and strategy of the SPS trolley in real time; using a graph neural network to analyze and model historical data and real-time data, and predict the material requirements of each production node; constructing a multi-agent collaborative scheduling model based on the SPS trolley to dynamically adjust the scheduling strategy according to the predicted material requirements of each production node; sharing and collaboratively optimizing data between multiple factories or production lines; constructing a blockchain network to manage material distribution data, and recording each material distribution of the SPS trolley. The present invention realizes efficient management and optimization of the material distribution process, and promotes the intelligent and digital development of the manufacturing industry.

Description

Technical Field

[0001] The present invention belongs to the technical field of digital twins, and in particular relates to a SPS trolley material distribution method and system based on digital twins. Background Art

[0002] In modern manufacturing, especially in large-scale assembly industries such as automobile manufacturing, material distribution is one of the key factors affecting production efficiency and quality. The SPS system proposed by Toyota greatly improves production efficiency and flexibility by pre-arranging and supplying the required materials in sets in specific areas and distributing them synchronously with the production line. However, with the complexity of production lines and the acceleration of production rhythm, the traditional SPS system has also exposed some shortcomings.

[0003] First, the traditional SPS system has limited optimization capabilities for scheduling and distribution routes in the face of complex and ever-changing production environments. Existing systems usually rely on pre-set rules and routes and lack the ability to adjust and optimize in real time. This results in the system being unable to respond and adjust quickly when encountering emergencies during the production process, such as equipment failures, order changes, etc., resulting in production interruptions or insufficient material supply.

[0004] Secondly, the accuracy of material demand forecasting is low. Traditional material demand forecasting methods mostly rely on historical data and simple statistical analysis, which makes it difficult to fully consider the impact of various factors in the production process, such as market demand fluctuations, changes in production rhythm, etc., resulting in low accuracy in material distribution, and often material shortages or surpluses, affecting production efficiency.

[0005] Third, the existing system's collaborative scheduling capabilities are insufficient. In a large manufacturing environment, multiple SPS carts are often required to deliver materials simultaneously. The existing scheduling system lacks an effective collaborative mechanism, which is prone to path conflicts and resource waste between carts, further reducing delivery efficiency.

[0006] In addition, data sharing and privacy protection are also major difficulties of the existing system. In order to achieve better collaborative optimization, material distribution data needs to be shared between different factories or production lines. However, traditional methods have data privacy and security risks in the data sharing process, which limits the collaborative optimization across factories or production lines.

[0007] Finally, the existing system lacks transparency and traceability management. In the process of material distribution, it is often necessary to record each distribution task in detail for easy traceability and management. However, traditional recording methods are prone to problems such as incomplete data and tampering, which affects the transparency and credibility of management. Summary of the invention

[0008] The purpose of the present invention is to design an SPS trolley material distribution method and system based on digital twins, which can effectively overcome the shortcomings of the existing technology and realize efficient, accurate and transparent material distribution management by combining the most advanced artificial intelligence and data processing technology.

[0009] In order to achieve the above-mentioned object, in a first aspect of the present invention, a SPS trolley material distribution method based on digital twin is provided, and the method comprises:

[0010] S1. Install multiple sensors in the factory environment to collect real-time operation data of production equipment, materials and SPS trolleys, pre-process the real-time operation data, and build a digital twin model of the physical factory and SPS trolley system; wherein the sensors include position sensors, speed sensors, weight sensors, environmental sensors and state sensors;

[0011] The digital twin model is constructed as follows:

[0012] Using the collected location information, a three-dimensional spatial model of the factory environment and the SPS trolley system is constructed, as shown below:

[0013] P i =(x i ,y i ,z i )

[0014] M j =(x j ,y j ,z j ,w j )

[0015] Among them, P i represents the three-dimensional coordinates of the i-th sensor, M j represents the three-dimensional coordinates and weight of the jth material, (x i ,y i ,z i ) represents the three-dimensional coordinates of the i-th position sensor;

[0016] Using the data from weight sensors and position sensors, a 3D model of the material is constructed. j , which is expressed as follows:

[0017] M j =(x j ,y j ,z j ,w j )

[0018] Among them, x j ,y j ,z jRespectively represent the x, y, and z coordinates of the material, w j Indicates the weight of the material;

[0019] Build a model of the factory environment based on data from environmental sensors:

[0020] E=(T,H)

[0021] Among them, T represents temperature, H represents humidity, and E represents the model of the factory environment;

[0022] Build a device status model based on the data from the status sensor:

[0023] S=(s1,s2,…,s l )

[0024] Among them, s l represents the state of the lth device, and S represents the device state model;

[0025] Use stream processing technology to process the data transmitted by sensors in real time and update the digital twin model in real time, as shown below:

[0026] D(t)={P i (t),M j (t),E(t),S(t)}

[0027] Where D(t) represents the data stream at time t, which contains the real-time data of all sensors, materials, environment and equipment status; P i (t) represents the three-dimensional coordinates of the i-th sensor at time t, M j (t) represents the three-dimensional model of the material at time t, E(t) represents the model of the factory environment at time t, and S(t) represents the equipment status model at time t;

[0028] Leverage real-time data streams to dynamically update the state of the digital twin model:

[0029]

[0030] E t+1 =E t +Δt·E change

[0031] S t+1 =S t +Δt·S change

[0032] Where Δt represents the time increment, represents the velocity component at the ith position, represents the velocity component at the jth position, E change Indicates the change of environmental parameters, Schange Indicates the change in device status. represents the updated 3D coordinates, The 3D model of the updated material, E t+1 The model representing the updated factory environment, S t+1 Represents the updated device state model;

[0033] S2. Build a scheduling model based on reinforcement learning according to the digital twin model, and optimize the material distribution path and strategy of the SPS trolley in real time according to the scheduling model;

[0034] S3. Use graph neural networks to analyze and model historical and real-time data to predict material demand at each production node.

[0035] S4. According to the predicted material demand of each production node, a multi-agent collaborative scheduling model based on the SPS trolley is constructed to dynamically adjust the scheduling strategy;

[0036] S5. Build a federated learning model to share and collaboratively optimize data among multiple factories;

[0037] S6. Build a blockchain network to manage material distribution data and record each material distribution of the SPS trolley.

[0038] Furthermore, the digital twin model is fused and simulated, and various sensor data are integrated to construct a unified data set, which is expressed as follows:

[0039] F(t)=φ(P(t),M(t),E(t),S(t))

[0040] Where F(t) represents the fused data set at time t, and φ represents the data fusion function;

[0041] Based on the fused data set, simulation is performed to simulate the operation status of the SPS trolley and materials in the factory environment, as shown below:

[0042] S(t)=ψ(F(t))

[0043] Among them, S(t) represents the simulation state at time t, and ψ represents the simulation function.

[0044] Furthermore, the scheduling model based on reinforcement learning is constructed as follows:

[0045] Construct the state space of material distribution to represent the current position, speed, load and factory environment of the SPS trolley. The state vector is expressed as follows:

[0046] s t =(P t ,V t,W t ,E t ,S t )

[0047] Among them, P t Indicates the current position of all trolleys, V t Indicates the current speed of all vehicles, W t Indicates the current load of all trolleys, E t Indicates the current environment state, S t Indicates the current device status;

[0048] Define the action space of the scheduling system, including the movement, acceleration, deceleration, and loading and unloading of the trolley, and the action vector is expressed as follows:

[0049] a t =(a 1,t ,a 2,t ,…,a i,t ,a n,t )

[0050] Among them, a i,t represents the action of the i-th trolley at time t;

[0051] The designed reward function is expressed as follows:

[0052] r t =R(s t ,a t )=α·Efficiency(s t ,a t )-β·Energy(s t ,a t )-γ·Conflict(s t ,a t )

[0053] Among them, α, β, γ are weight coefficients, Efficiency represents the delivery efficiency, Energy represents the energy consumption, and Conflict represents the path conflict;

[0054] Use the Q learning algorithm to construct a Q value function to optimize the reward function, where the Q value function is expressed as follows:

[0055]

[0056] Among them, γ is the discount factor, which indicates the degree of decay of future rewards;

[0057] Using the ε-greedy strategy for action selection, the action selection strategy is expressed as follows:

[0058]

[0059] where ∈ is the exploration rate.

[0060] Furthermore, the optimal path is planned in real time according to the scheduling model based on reinforcement learning, as shown below:

[0061] p i,t+1 =p i,t +Δt·v i,t +a i,t

[0062] Among them, p i,t represents the position of the ith trolley at time t, v i,t represents the speed of the ith trolley at time t, a i,t represents the action of the i-th trolley at time t, and Δt represents the time increment;

[0063] During the path planning process, the path conflicts between the vehicles are detected in real time and adjusted. The conflict detection is expressed as follows:

[0064]

[0065] Among them, ∥p i,t -p j,t ∥ represents the distance between the i-th and j-th trolleys, d min is the minimum safe distance;

[0066] According to the real-time data update, the scheduling strategy is dynamically adjusted, as shown below:

[0067]

[0068] Among them, π t represents the scheduling strategy at time t, α represents the learning rate, Represents the gradient of the Q-value function.

[0069] Furthermore, the loss function of the graph neural network is It is expressed as follows:

[0070]

[0071] in, represents the predicted material demand, d i represents the actual material demand, and N represents the number of nodes;

[0072] Use the gradient descent method to train the graph neural network and update the network parameters to minimize the loss function. The parameter update formula is as follows:

[0073]

[0074] Among them, θ represents the network parameters, η represents the learning rate, Represents the gradient of the loss function with respect to the network parameters.

[0075] Furthermore, real-time demand forecasting is performed based on the trained graph neural network, the attributes of the current nodes and edges are input, and the predicted material demand is output. The forecasting formula is as follows:

[0076]

[0077] in, Represents node v i The predicted material demand, ψ represents the state update function; h i Represents node v i The attribute vector of i , Real-time material requirements i and production status i ;m ij Represents node v in the graph neural network i From Node v j Received messages; Represents node v i The set of neighbor nodes.

[0078] Furthermore, in the multi-agent collaborative scheduling model, each SPS vehicle acts as an agent A. i , A i =(P i ,V i ,W i ,S i ), where P i represents the position of the ith agent, V i represents the speed of the ith agent, W i represents the load of the ith agent, S i represents the state of the i-th agent; the agent state vector is in, represents the state of the i-th agent at time t; the objective function J of the multi-agent collaborative scheduling model is expressed as follows:

[0079]

[0080] Among them, N represents the number of agents, α, β, γ represent weight coefficients, Efficiency represents delivery efficiency, Energy represents energy consumption, and Conflict represents path conflict;

[0081] The multi-agent reinforcement learning algorithm is used to update the strategy. Each agent selects the optimal action based on the current state and the state of the neighboring agents, which is expressed as follows:

[0082]

[0083] Where Q represents the Q-value function, represents the state of the neighboring agent, represents the action of the ith agent at time t; where the state-action value function Indicates that agent i is in state and neighbor status Take action The value is expressed as follows:

[0084]

[0085] in, represents the immediate reward, γ represents the discount factor, represents the updated state of the agent; represents the updated action of the ith agent; represents the updated state of the neighboring agent;

[0086] According to real-time data, the agent dynamically adjusts its strategy. The strategy adjustment formula is as follows:

[0087]

[0088] in, represents the strategy of the ith agent at time t, α represents the learning rate, Represents the gradient of the Q-value function;

[0089] Each agent performs collaborative optimization by sharing information and resources, as shown below:

[0090]

[0091] Among them, O represents the result of collaborative optimization, represents the strategy of the ith agent, Represents the state of the ith agent.

[0092] Furthermore, the global loss function of the federated learning model is It is expressed as follows:

[0093]

[0094] in, represents the local loss function of the ith factory, n i represents the number of data samples of the ith factory, n represents the total number of data samples of all factories, represents the local dataset, and w represents the global model;

[0095] Through the iterative process of federated learning, the global model of the federated learning model is optimized to minimize the global loss function, which is expressed as follows:

[0096]

[0097] Among them, w * Represents the updated global model.

[0098] Furthermore, the S6 specifically includes:

[0099] S601. Create a dedicated blockchain network, including initialization and configuration of each node;

[0100] S602: Define the data that needs to be uploaded to the blockchain, package the data records into transactions, and sign the transactions using public and private key pairs;

[0101] S603. Write a smart contract to manage the business logic of material distribution, define the interface of the smart contract, and deploy the smart contract to the blockchain network. Each node can call the smart contract interface to perform corresponding operations.

[0102] S604. Define the consensus mechanism of the blockchain network. When a new transaction is broadcast to the entire network, the nodes verify the validity of the transaction through the consensus mechanism and package it into a new block. The new block is added to the blockchain and reaches consensus through the consensus mechanism.

[0103] S605. Provide a query interface based on blockchain, and users can query the status and history of materials through smart contracts.

[0104] In a second aspect of the present invention, a SPS trolley material distribution system based on digital twin is provided, the system comprising:

[0105] A data acquisition module is used to install a variety of sensors in the factory environment to collect real-time operation data of production equipment, materials and SPS trolleys, and pre-process the real-time operation data to build a digital twin model of the physical factory and SPS trolley system; wherein the sensors include position sensors, speed sensors, weight sensors, environmental sensors and state sensors;

[0106] The digital twin model is constructed as follows:

[0107] Using the collected location information, a three-dimensional spatial model of the factory environment and the SPS trolley system is constructed, as shown below:

[0108] P i =(x i ,y i ,z i)

[0109] M j =(x j ,y j ,z j ,w j )

[0110] Among them, P i represents the three-dimensional coordinates of the i-th sensor, M j represents the three-dimensional coordinates and weight of the jth material, (x i ,y i ,z i ) represents the three-dimensional coordinates of the i-th position sensor;

[0111] Using the data from weight sensors and position sensors, a 3D model of the material is constructed. j , which is expressed as follows:

[0112] M j =(x j ,y j ,z j ,w j )

[0113] Among them, x j ,y j ,z j Respectively represent the x, y, and z coordinates of the material, w j Indicates the weight of the material;

[0114] Build a model of the factory environment based on data from environmental sensors:

[0115] E=(T,H)

[0116] Among them, T represents temperature, H represents humidity, and E represents the model of the factory environment;

[0117] Build a device status model based on the data from the status sensor:

[0118] S=(s1,s2,…,s l )

[0119] Among them, s l represents the state of the lth device, and S represents the device state model;

[0120] Use stream processing technology to process the data transmitted by sensors in real time and update the digital twin model in real time, as shown below:

[0121] D(t)={P i (t),M j (t),E(t),S(t)}

[0122] Where D(t) represents the data stream at time t, which contains the real-time data of all sensors, materials, environment and equipment status; P i (t) represents the three-dimensional coordinates of the i-th sensor at time t, M j (t) represents the three-dimensional model of the material at time t, E(t) represents the model of the factory environment at time t, and S(t) represents the equipment status model at time t;

[0123] Leverage real-time data streams to dynamically update the state of the digital twin model:

[0124]

[0125] E t+1 =E t +Δt·E change

[0126] S t+1 =S t +Δt·S change

[0127] Where Δt represents the time increment, represents the velocity component at the ith position, represents the velocity component at the jth position, E change Indicates the change of environmental parameters, S change Indicates the change in device status. represents the updated 3D coordinates, The 3D model of the updated material, E t+1 The model representing the updated factory environment, S t+1 Represents the updated device state model;

[0128] The production scheduling module is used to build a scheduling model based on reinforcement learning according to the digital twin model, and optimize the material distribution path and strategy of the SPS trolley in real time according to the scheduling model;

[0129] The material forecasting module is used to analyze and model historical and real-time data using graph neural networks to predict the material demand of each production node;

[0130] The multi-node scheduling module is used to dynamically adjust the scheduling strategy by building a multi-agent collaborative scheduling model based on the SPS trolley according to the predicted material demand of each production node;

[0131] Data sharing module, used to build a federated learning model to share and collaboratively optimize data among multiple factories;

[0132] The data traceability module is used to build a blockchain network to manage material distribution data and record each material distribution of the SPS trolley.

[0133] The beneficial technical effects of the present invention are at least as follows:

[0134] (1) The present invention establishes a digital twin of the physical factory and the SPS trolley to synchronize the virtual and real environments in real time. Data is collected in real time through sensors and IoT devices, and simulation and optimization are performed in the virtual environment, so that the system has stronger adjustment and optimization capabilities when facing complex and changing production environments, quickly responds to emergencies, and ensures the continuous operation of the production line.

[0135] (2) The present invention uses advanced artificial intelligence algorithms, including reinforcement learning and graph neural networks (GNNs), to optimize material distribution paths and scheduling strategies. Reinforcement learning algorithms can gradually optimize scheduling strategies through continuous learning and feedback, and achieve the ability to adapt to complex environments. Graph neural networks can handle complex production network relationships, accurately predict material requirements for each production node, avoid material shortages or surpluses, and ensure smooth production.

[0136] (3) The present invention introduces a multi-agent system (MAS) and a federated learning algorithm to achieve collaborative work of multiple SPS trolleys and data sharing and collaborative optimization across factories or production lines. The multi-agent system enables each SPS trolley to share information and resources, coordinate delivery tasks, avoid path conflicts and waste of resources, and improve delivery efficiency. The federated learning algorithm achieves data sharing and collaborative optimization while ensuring data privacy and security, and promotes collaborative optimization across factories or production lines.

[0137] (4) The present invention uses blockchain technology to record and manage the material distribution process, achieving transparency and traceability of the entire process. Blockchain can ensure the transparency and immutability of data, record each material distribution in detail, and improve the credibility and transparency of management. BRIEF DESCRIPTION OF THE DRAWINGS

[0138] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0139] Figure 1 This is a flow chart of the SPS trolley material distribution method based on digital twins of the present invention.

[0140] Figure 2 This is a framework diagram of the SPS trolley material distribution method based on digital twins in the present invention. DETAILED DESCRIPTION

[0141] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0142] In one or more embodiments, Figure 1 As shown, a SPS trolley material distribution method based on digital twin is disclosed, and the method includes the following steps S1-S6:

[0143] S1. Install multiple sensors in the factory environment to collect real-time operation data of production equipment, materials and SPS trolleys, and pre-process the real-time operation data to build a digital twin model of the physical factory and SPS trolley system; wherein the sensors include position sensors, speed sensors, weight sensors, environmental sensors and state sensors.

[0144] Specifically, a variety of sensors are installed in the factory environment to collect real-time operating data of production equipment, materials, and SPS trolleys. Sensor types include but are not limited to:

[0145] Position sensor: collects the position information of SPS trolley and materials.

[0146] Speed ​​sensor: collects the movement speed of the SPS trolley.

[0147] Weight sensor: collects material weight information.

[0148] Environmental sensor: collects environmental information such as temperature and humidity.

[0149] Status sensor: collects the working status information of the equipment.

[0150] The sensor data format is as follows:

[0151] P i =(t,x i ,y i ,z i )

[0152]

[0153] W j =(t,w j )

[0154] E k =(t,T,H)

[0155] S l =(t,s l )

[0156] Among them, t represents the timestamp, x i ,y i ,z i represents the three-dimensional coordinates of the i-th position sensor, represents the velocity component of the i-th velocity sensor, w j represents the weight of the jth weight sensor, T represents temperature, H represents humidity, s l Indicates the state of the lth state sensor.

[0157] Furthermore, the data collected by the sensor is preprocessed to ensure the accuracy and integrity of the data. The preprocessing includes:

[0158] Remove noise and outliers using a median filter:

[0159] D clean =MedianFilter(D raw )

[0160] Among them, D raw represents the original data, D clean Represents the cleaned data;

[0161] Scale the data to a uniform scale:

[0162]

[0163] Among them, D norm represents the normalized data;

[0164] Synchronize the data of different sensors to a unified time axis using linear interpolation:

[0165]

[0166] Among them, t1 and t2 are two adjacent time points, D sync Indicates the synchronized data.

[0167] Furthermore, the digital twin model is constructed as follows:

[0168] Using the collected location information, a three-dimensional spatial model of the factory environment and the SPS trolley system is constructed:

[0169] P i =(x i ,y i ,z i )

[0170] M j =(x j ,y j ,z j ,wj )

[0171] Among them, P i represents the three-dimensional coordinates of the i-th sensor, M j Represents the three-dimensional coordinates and weight of the j-th material.

[0172] Build a 3D model of the material based on the data from the weight sensor and position sensor:

[0173] M j =(x j ,y j ,z j ,w j )

[0174] Among them, x j ,y j ,z j Respectively represent the x, y, and z coordinates of the material, w j Indicates the weight of the material.

[0175] Build a model of the factory environment based on data from environmental sensors:

[0176] E=(T,H)

[0177] Among them, T represents temperature and H represents humidity.

[0178] Build a device status model based on the data from the status sensor:

[0179] S=(s1,s2,…,s l )

[0180] Among them, s l Indicates the status of the lth device.

[0181] Furthermore, stream processing technologies (such as Apache Kafka) are used to process the data transmitted in real time by sensors to ensure real-time updates of the digital twin model:

[0182] D(t)={P i (t),M j (t),E(t),S(t)}

[0183] Where D(t) represents the data stream at time t, which contains real-time data of all sensors, materials, environment, and equipment status.

[0184] Furthermore, the status of the digital twin model is dynamically updated using real-time data streams:

[0185]

[0186] E t+1 =Et +Δt·E change

[0187] S t+1 =S t +Δt·S change

[0188] Where Δt represents the time increment, represents the velocity component at the ith position, E change Indicates the change of environmental parameters, S change Indicates the change in device status.

[0189] Furthermore, various sensor data are integrated to build a unified data set:

[0190] F(t)=φ(P(t),M(t),E(t),S(t))

[0191] Among them, F(t) represents the fused data set at time t, and φ represents the data fusion function.

[0192] Simulation based on the fused data set simulates the operation status of the SPS trolley and materials in the factory environment:

[0193] S(t)=ψ(F(t))

[0194] Among them, S(t) represents the simulation state at time t, and ψ represents the simulation function.

[0195] S2. Build a scheduling model based on reinforcement learning according to the digital twin model, and optimize the material distribution path and strategy of the SPS trolley in real time according to the scheduling model.

[0196] Specifically, the state space of material distribution is constructed to represent the current position, speed, load and factory environment of the SPS trolley. The state vector is defined as follows:

[0197] s t =(P t ,V t ,W t ,E t ,S t )

[0198] Among them, P t Indicates the current position of all trolleys, V t Indicates the current speed of all vehicles, W t Indicates the current load of all trolleys, E t Indicates the current environment state, S t Indicates the current device status.

[0199] Define the action space of the scheduling system, including the movement, acceleration, deceleration, and loading and unloading of the trolley. The action vector is defined as follows:

[0200] a t =(a 1,t ,a 2,t ,…,a n,t )

[0201] Among them, a i,t Represents the action of the i-th trolley at time t, including moving to a new position, adjusting speed and load, etc.

[0202] Furthermore, a reward function is designed to evaluate the effect of each dispatch action. The reward function takes into account factors such as delivery efficiency, energy consumption, and path conflicts:

[0203] r t =R(s t ,a t )=α·Efficiency(s t ,a t )-β·Energy(s t ,a t )-γ·Conflict(s t ,a t )

[0204] Among them, α, β, γ are weight coefficients, Efficiency represents the delivery efficiency, Energy represents the energy consumption, and Conflict represents the path conflict.

[0205] Preferably, a Q-learning algorithm is used to construct a Q-value function, which is used to evaluate the long-term benefits of performing an action in a specific state. The Q-value function is defined as follows:

[0206]

[0207] Among them, γ is the discount factor, which indicates the degree of decay of future rewards.

[0208] The ε-greedy strategy is used for action selection, which not only explores new action spaces but also uses the existing optimal strategy. The action selection strategy is defined as follows:

[0209]

[0210] where ∈ is the exploration rate.

[0211] Furthermore, based on the current status, the reinforcement learning algorithm is used to plan the optimal path in real time to ensure that materials can be delivered to the designated location in a timely and accurate manner. Path planning model:

[0212] p i,t+1=p i,t +Δt·v i,t +a i,t

[0213] Among them, p i,t represents the position of the ith trolley at time t, v i,t represents the speed of the ith trolley at time t, a i,t Represents the action of the i-th trolley at time t.

[0214] During the path planning process, the path conflicts between the vehicles are detected in real time and adjusted. Conflict detection model:

[0215]

[0216] Among them, ∥p i,t -p j,t ∥ represents the distance between the i-th and j-th trolleys, d min is the minimum safe distance.

[0217] Furthermore, the scheduling strategy is dynamically adjusted based on real-time data updates to ensure that the system can adapt to changes in the factory environment. Strategy adjustment model:

[0218]

[0219] Among them, π t represents the scheduling strategy at time t, α represents the learning rate, Represents the gradient of the Q-value function.

[0220] Furthermore, the effectiveness of the scheduling strategy is regularly evaluated to ensure that the system achieves optimal performance in actual operation. Strategy evaluation indicators include delivery efficiency, energy consumption, and path conflict rate.

[0221] S3. Use graph neural networks to analyze and model historical data and real-time data to predict material requirements for each production node.

[0222] Specifically, define the nodes and edges in the graph. Nodes represent production nodes, and edges represent the material flow relationship between production nodes. Node attributes include historical material demand, real-time material demand, production status, etc., and edge attributes include material flow volume, transportation time, etc.

[0223] Let G = (V, E) be a graph, where V is the set of nodes and E is the set of edges. i ∈V represents a production node, and each edge e ij ∈E represents node v i and v j The material flow relationship between them.

[0224] Furthermore, we define the node attribute vector and edge attribute vector:

[0225] h i =(d i ,r i ,s i )

[0226] e ij =(f ij ,t ij )

[0227] Among them, h i Represents node v i The attribute vector of i , Real-time material requirements i and production status i .e ij Represents edge e ij The attribute vector of the material flow rate f ij and transportation time t ij .

[0228] Furthermore, the properties of nodes and edges are updated using the message passing mechanism. Each node receives messages from adjacent nodes and updates its own state. The message passing formula is as follows:

[0229]

[0230] in, Represents node v in the l-layer graph neural network i From Node v j The received message, φ represents the message passing function.

[0231] Use the received message to update the node status. The node status update formula is as follows:

[0232]

[0233] Among them, Represents node v in the l+1 layer graph neural network i The state of ψ represents the state update function. Represents node v i The set of neighbor nodes.

[0234] Use the node state to update the edge state. The edge state update formula is as follows:

[0235]

[0236] in, Represents the edge e in the l+1 layer graph neural network ij The state of , γ represents the edge state update function.

[0237] Define the loss function for training the graph neural network. The loss function considers the accuracy of material demand forecast and the deviation from actual demand:

[0238]

[0239] in, represents the predicted material demand, d i represents the actual material demand, and N represents the number of nodes.

[0240] Use the gradient descent method to train the graph neural network and update the network parameters to minimize the loss function. The parameter update formula is as follows:

[0241]

[0242] Among them, θ represents the network parameters, η represents the learning rate, Represents the gradient of the loss function with respect to the network parameters.

[0243] Furthermore, the trained graph neural network is used to perform real-time demand forecasting, input the attributes of the current nodes and edges, and output the predicted material demand. The forecasting formula is as follows:

[0244]

[0245] in, Represents node v i Forecast material requirements.

[0246] It is understandable that the accuracy of the forecast model is verified by comparing the forecast results with the actual material demand. Verification indicators include mean absolute error (MAE), mean square error (MSE), etc. MAE and MSE are defined as follows:

[0247]

[0248] The material demand forecasting solution of the graph neural network uses the graph structure and message passing mechanism to accurately predict the material demand of each production node. Based on the real-time data provided by the digital twin model, the system can dynamically adjust the demand forecasting model to adapt to changes in the factory environment and ensure that materials can be delivered to the designated location in a timely and accurate manner.

[0249] S4. According to the predicted material demand of each production node, a multi-agent collaborative scheduling model based on the SPS trolley is constructed to dynamically adjust the scheduling strategy.

[0250] Specifically, each SPS trolley is regarded as an intelligent agent, and its attribute vector is defined as follows:

[0251] A i =(Pi ,V i ,W i ,S i )

[0252] Among them, P i represents the position of the ith agent, V i represents the speed of the ith agent, W i represents the load of the ith agent, S i Represents the state of the ith agent.

[0253] Furthermore, the agent state vector is defined as:

[0254]

[0255] in, represents the state of the ith agent at time t.

[0256] Furthermore, the objective function of collaborative scheduling is defined, including factors such as delivery efficiency, energy consumption, and path conflicts:

[0257]

[0258] Among them, N represents the number of agents, α, β, γ are weight coefficients, Efficiency represents delivery efficiency, Energy represents energy consumption, and Conflict represents path conflict.

[0259] Use multi-agent reinforcement learning algorithms (such as MAS-Q learning) to update strategies. Each agent selects the best action based on its current state and the state of its neighboring agents:

[0260]

[0261] Where Q represents the Q-value function, represents the state of the neighboring agent, represents the action of the ith agent at time t.

[0262] Furthermore, the state-action value function Indicates that agent i is in state and neighbor status Take action The value of:

[0263]

[0264] in, represents the immediate reward, and γ represents the discount factor.

[0265] Furthermore, during the path planning process, the path conflicts between agents are detected in real time. The conflict detection formula is:

[0266]

[0267] in, represents the distance between the i-th and j-th agents, d min is the minimum safe distance.

[0268] When a conflict is detected, the agent adjusts the path to avoid the conflict. Path adjustment formula:

[0269]

[0270] in, represents the position of the ith agent at time t+1, Δt represents the time increment, Indicates the adjusted action.

[0271] Furthermore, based on real-time data, the agent dynamically adjusts its strategy to adapt to changes in the factory environment. Strategy adjustment formula:

[0272]

[0273] in, represents the strategy of the ith agent at time t, α represents the learning rate, Represents the gradient of the Q-value function.

[0274] Furthermore, the agents can share information and resources to perform collaborative optimization and improve overall delivery efficiency.

[0275]

[0276] Among them, O represents the result of collaborative optimization, represents the strategy of the ith agent, Represents the state of the ith agent.

[0277] S5. Build a federated learning model to share and collaboratively optimize data among multiple factories.

[0278] Specifically, the initial global model w0 is defined for initialization in each factory or production line:

[0279] w0=InitializeModel()

[0280] The initial global model is then assigned to each participating factory or production line to form a local model:

[0281]

[0282] Where N represents the number of factories or production lines participating in federated learning, and w i,0 represents the initial local model of the ith factory.

[0283] Furthermore, each factory or production line utilizes local data sets Perform model training. Local dataset Contains local material distribution data, production data, etc.

[0284] Each factory trains a local model on a local dataset and updates the model parameters:

[0285]

[0286] Among them, η represents the learning rate, Indicates the i-th factory in the local data set The gradient of the loss function on .

[0287] Furthermore, the update results of all local models are uploaded to the central server for global model aggregation:

[0288]

[0289] Among them, w t+1 represents the global model after aggregation, n i represents the number of data samples of the ith factory, and n represents the total number of data samples of all factories.

[0290] Then the updated global model w t+1 Distribute to each factory and continue the next round of local model training:

[0291]

[0292] Furthermore, the local model training and global model aggregation steps are repeated until the global model converges. Each round of iteration process is as follows:

[0293] Each factory uses local data sets to train local models and update model parameters.

[0294] Each factory uploads the local model update results to the central server.

[0295] The central server aggregates the global model and updates the global model.

[0296] Distribute the updated global model to each factory and start the next round of training.

[0297] Furthermore, the global model convergence condition is defined, and the iteration is stopped when the following conditions are met:

[0298] ∥wt+1 -w t ∥<∈

[0299] Among them, ∈ represents the convergence threshold.

[0300] Furthermore, the goal of federated learning is to minimize the global loss function Defined as:

[0301]

[0302] in, represents the local loss function of the ith factory.

[0303] Through the iterative process of federated learning, the global model is continuously optimized to minimize the global loss function:

[0304]

[0305] Preferably, during the federated learning process, each factory only uploads the model update results and does not share local data to ensure data privacy and security.

[0306] To further protect data privacy, you can add noise when uploading model update results, using differential privacy technology:

[0307]

[0308] in, represents zero mean and variance σ 2 Gaussian noise.

[0309] S6. Build a blockchain network to manage material distribution data and record each material distribution of the SPS trolley. Specifically, it includes the following steps:

[0310] S601. Create a dedicated blockchain network, including initialization and configuration of each node. Assume that Hyperledger Fabric is used to build the blockchain network.

[0311] Define a node set {N1, N2, ..., N k}, where N k Represents the kth node. The initialization of each node includes generating a public and private key pair, joining the blockchain network, and configuring the consensus mechanism. Generate a Genesis Block and start the blockchain network.

[0312] S602, define the data that needs to be uploaded to the blockchain, including material information, transportation process, production status, etc. Each data record is represented by D i , where i represents the index of the record. Pack the data records into transactions Ti , including the transaction timestamp, data content and signature. The definition of a transaction is as follows:

[0313] T i =(timestamp i ,data i ,signature i )

[0314] Transactions are signed using public and private key pairs to ensure data integrity and authenticity of the source.

[0315] signature i =Sign(data i ,private_key i )

[0316] S603. Write a smart contract (SmartContract) to manage the business logic of material distribution. Smart contracts include functions such as adding, updating, and querying material records. Define the interface of the smart contract, such as the interface for adding material records:

[0317] function addMaterialRecord(data){...}

[0318] Smart contracts are deployed to the blockchain network, and each node can call the smart contract interface to perform corresponding operations.

[0319] S604. Define the consensus mechanism of the blockchain network, such as using the Raft consensus mechanism. Each node participates in the consensus process, confirms the validity of the transaction and packages it into a block. New transaction T i It is broadcast to the entire network, and nodes verify the validity of the transaction through a consensus mechanism and package it into a new block.

[0320]

[0321] New Block B n Are added to the blockchain and agreed upon through a consensus mechanism.

[0322] chain t+1 =chain t +B n

[0323] S605. Provide a query interface based on blockchain, and users can query the status and history of materials through smart contracts.

[0324] The query interface is defined as follows:

[0325] function queryMaterialStatus(material_id){...}

[0326] The query results include the current status of the material and all related transaction records to achieve traceability.

[0327] result=Query(material_id).

[0328] In one or more embodiments, Figure 2 As shown, a SPS trolley material distribution system based on digital twin is disclosed, and the system includes:

[0329] The data acquisition module 101 is used to install various sensors in the factory environment to collect real-time operation data of production equipment, materials and SPS trolleys, and pre-process the real-time operation data to build a digital twin model of the physical factory and SPS trolley system; wherein the sensors include position sensors, speed sensors, weight sensors, environmental sensors and state sensors;

[0330] The digital twin model is constructed as follows:

[0331] Using the collected location information, a three-dimensional spatial model of the factory environment and the SPS trolley system is constructed, as shown below:

[0332] P i =(x i ,y i ,z i )

[0333] M j =(x j ,y j ,z j ,w j )

[0334] Among them, P i represents the three-dimensional coordinates of the i-th sensor, M j represents the three-dimensional coordinates and weight of the jth material, (x i ,y i ,z i ) represents the three-dimensional coordinates of the i-th position sensor;

[0335] Using the data from weight sensors and position sensors, a 3D model of the material is constructed. j , which is expressed as follows:

[0336] M j =(x j ,y j ,z j ,wj )

[0337] Among them, x j ,y j ,z j Respectively represent the x, y, and z coordinates of the material, w j Indicates the weight of the material;

[0338] Build a model of the factory environment based on data from environmental sensors:

[0339] E=(T,H)

[0340] Among them, T represents temperature, H represents humidity, and E represents the model of the factory environment;

[0341] Build a device status model based on the data from the status sensor:

[0342] S=(s1,s2,…,s l )

[0343] Among them, s l represents the state of the lth device, and S represents the device state model;

[0344] Use stream processing technology to process the data transmitted by sensors in real time and update the digital twin model in real time, as shown below:

[0345] D(t)={P i (t),M j (t),E(t),S(t)}

[0346] Where D(t) represents the data stream at time t, which contains the real-time data of all sensors, materials, environment and equipment status; P i (t) represents the three-dimensional coordinates of the i-th sensor at time t, M j (t) represents the three-dimensional model of the material at time t, E(t) represents the model of the factory environment at time t, and S(t) represents the equipment status model at time t;

[0347] Leverage real-time data streams to dynamically update the state of the digital twin model:

[0348]

[0349] E t+1 =E t +Δt·E change

[0350] S t+1 =S t +Δt·S change

[0351] Where Δt represents the time increment, represents the velocity component at the ith position, represents the velocity component at the jth position, E change Indicates the change of environmental parameters, S change Indicates the change in device status. represents the updated 3D coordinates, The 3D model of the updated material, E t+1 The model representing the updated factory environment, S t+1 Represents the updated device state model;

[0352] The production scheduling module 102 is used to build a scheduling model based on reinforcement learning according to the digital twin model, and optimize the material distribution path and strategy of the SPS trolley in real time according to the scheduling model;

[0353] The material forecasting module 103 is used to analyze and model historical data and real-time data using graph neural networks to predict the material demand of each production node;

[0354] The multi-node scheduling module 104 is used to construct a multi-agent collaborative scheduling model based on the SPS trolley to dynamically adjust the scheduling strategy according to the predicted material demand of each production node;

[0355] A data sharing module 105 is used to build a federated learning model to share and collaboratively optimize data between multiple factories or production lines;

[0356] The data tracing module 106 is used to build a blockchain network to manage the material distribution data and record each material distribution of the SPS trolley.

[0357] In the description of this specification, the description of reference terms such as "one embodiment", "some embodiments", "examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms is not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples without contradiction. It should be noted that the flowcharts and block diagrams in the accompanying drawings of the present invention show the possible implementation architecture, functions and operations of the system, method and machine program product according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flow chart, and combinations of blocks in the block diagram and / or flow chart, may be implemented with a dedicated hardware-based system that performs the specified functions or actions, or may be implemented with a combination of dedicated hardware and machine instructions.

[0358] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0359] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0360] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0361] In addition, each functional unit in each embodiment of the present invention can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium readable by an acquisition machine. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the acquisition machine software product is stored in a storage medium, including several instructions for enabling an acquisition machine (which can be a personal acquisition machine, a server, or a network machine, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0362] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A SPS trolley material distribution method based on digital twin, characterized in that: The method comprises: S1. Install multiple sensors in the factory environment to collect real-time operation data of production equipment, materials and SPS trolleys, pre-process the real-time operation data, and build a digital twin model of the physical factory and SPS trolley system; wherein the sensors include position sensors, speed sensors, weight sensors, environmental sensors and state sensors; The digital twin model is constructed as follows: Using the collected location information, a three-dimensional spatial model of the factory environment and the SPS trolley system is constructed, as shown below: P i =(x i ,y i ,z i ) Among them, P i represents the three-dimensional coordinates of the i-th sensor; Using the data from weight sensors and position sensors, a 3D model of the material is constructed. j , which is expressed as follows: M j =(x j ,y j ,z j ,w j ) Among them, x j ,y j ,z j Respectively represent the x, y, and z coordinates of the material, w j Indicates the weight of the material; Build a model of the factory environment based on data from environmental sensors: E=(T,H) Among them, T represents temperature, H represents humidity, and E represents the model of the factory environment; Build a device status model based on the data from the status sensor: S=(s1,s2,…,s l ) Among them, s l represents the state of the lth device, and S represents the device state model; Use stream processing technology to process the data transmitted by sensors in real time and update the digital twin model in real time, as shown below: D(t)={P i (t),M j (t),E(t),S(t)} Where D(t) represents the data stream at time t, which contains the real-time data of all sensors, materials, environment and equipment status; P i (t) represents the three-dimensional coordinates of the i-th sensor at time t, M j (t) represents the three-dimensional model of the material at time t, E(t) represents the model of the factory environment at time t, and S(t) represents the equipment status model at time t; Leverage real-time data streams to dynamically update the state of the digital twin model: AND t+1 =And t +Δt·E change S t+1 =S t +Δt·S change Where Δt represents the time increment, represents the velocity component at the ith position, represents the velocity component at the jth position, E change Indicates the change of environmental parameters, S change Indicates the change in device status. represents the updated 3D coordinates, The 3D model of the updated material, E t+1 The model representing the updated factory environment, S t+1 Represents the updated device state model; S2. Build a scheduling model based on reinforcement learning according to the digital twin model, and optimize the material distribution path and strategy of the SPS trolley in real time according to the scheduling model; S3. Use graph neural networks to analyze and model historical and real-time data to predict material demand at each production node. S4. According to the predicted material demand of each production node, a multi-agent collaborative scheduling model based on the SPS trolley is constructed to dynamically adjust the scheduling strategy; S5. Build a federated learning model to share and collaboratively optimize data among multiple factories; S6. Build a blockchain network to manage material distribution data and record each material distribution of the SPS trolley.

2. According to the SPS trolley material distribution method based on digital twinning according to claim 1, it is characterized in that: The digital twin model is fused and simulated, the sensor data is fused, and a unified data set is constructed, which is expressed as follows: F(t)=φ(P(t),M(t),E(t),S(t)) Where F(t) represents the fused data set at time t, and φ represents the data fusion function; Simulate the operation status of SPS trolleys and materials in the factory environment based on the fused data set. It is expressed as follows: S(t)=ψ(F(t)) Among them, S(t) represents the simulation state at time t, and ψ represents the simulation function.

3. According to the SPS trolley material distribution method based on digital twinning according to claim 1, it is characterized in that: The scheduling model based on reinforcement learning is constructed as follows: Construct the state space of material distribution to represent the current position, speed, load and factory environment of the SPS trolley. The state vector is expressed as follows: s t =(P t ,V t ,W t ,E t ,S t ) Among them, P t Indicates the current position of all trolleys, V t Indicates the current speed of all vehicles, W t Indicates the current load of all trolleys, E t Indicates the current environment state, S t Indicates the current device status; Define the action space of the scheduling system, including the movement, acceleration, deceleration, and loading and unloading of the trolley, and the action vector is expressed as follows: a t =(a 1,t ,a 2,t ,…,a i,t ,a n,t ) Among them, a i,t represents the action of the i-th trolley at time t; The designed reward function is expressed as follows: r t =R(s t ,a t )=α·Efficiency(s t ,a t )-β·Energy(s t ,a t )-γ·Conflict(s t ,a t ) Among them, α, β, γ are weight coefficients, Efficiency represents the delivery efficiency, Energy represents the energy consumption, and Conflict represents the path conflict; Use the Q learning algorithm to construct a Q value function to optimize the reward function, where the Q value function is expressed as follows: Among them, γ is the discount factor, which indicates the degree of decay of future rewards; Using the ε-greedy strategy for action selection, the action selection strategy is expressed as follows: where ∈ is the exploration rate.

4. According to the SPS trolley material distribution method based on digital twinning according to claim 3, it is characterized in that: According to the scheduling model based on reinforcement learning, the optimal path is planned in real time, which is expressed as follows: p i,t+1 =p i,t +Δt·v i,t +a i,t Among them, p i,t represents the position of the ith trolley at time t, v i,t represents the speed of the ith trolley at time t, a i,t represents the action of the i-th trolley at time t, and Δt represents the time increment; During the path planning process, the path conflicts between the vehicles are detected in real time and adjusted. The conflict detection is expressed as follows: Among them, ∥p i,t -p j,t ∥ represents the distance between the i-th and j-th trolleys, d min is the minimum safe distance; According to the real-time data update, the scheduling strategy is dynamically adjusted, as shown below: Among them, π t represents the scheduling strategy at time t, α represents the learning rate, Represents the gradient of the Q-value function.

5. According to the SPS trolley material distribution method based on digital twinning according to claim 1, it is characterized in that: The loss function of the graph neural network It is expressed as follows: in, represents the predicted material demand, d i represents the actual material demand, and N represents the number of nodes; Use the gradient descent method to train the graph neural network and update the network parameters to minimize the loss function. The parameter update formula is as follows: Among them, θ represents the network parameters, η represents the learning rate, Represents the gradient of the loss function with respect to the network parameters.

6. According to the SPS trolley material distribution method based on digital twinning according to claim 5, it is characterized in that: According to the trained graph neural network, real-time demand forecasting is performed, the attributes of the current nodes and edges are input, and the predicted material demand is output. The forecasting formula is as follows: in, Represents node v i The predicted material demand, ψ represents the state update function; h i Represents node v i The attribute vector of i , Real-time material requirements i and production status i ;m ij Represents node v in the graph neural network i From Node v j Received messages; Represents node v i The set of neighbor nodes.

7. The SPS trolley material distribution method based on digital twin according to claim 1 is characterized in that: In the multi-agent collaborative scheduling model, each SPS vehicle acts as an agent A. i , A i =(P i ,V i ,W i ,S i ), where P i represents the position of the ith agent, V i represents the speed of the ith agent, W i represents the load of the ith agent, S i represents the state of the i-th agent; the agent state vector is in, represents the state of the i-th agent at time t; the objective function J of the multi-agent collaborative scheduling model is expressed as follows: Among them, N represents the number of agents, α, β, γ represent weight coefficients, Efficiency represents delivery efficiency, Energy represents energy consumption, and Conflict represents path conflict; The multi-agent reinforcement learning algorithm is used to update the strategy. Each agent selects the optimal action based on the current state and the state of the neighboring agents, which is expressed as follows: Where Q represents the Q-value function, represents the state of the neighboring agent, represents the action of the ith agent at time t; where the state-action value function Indicates that agent i is in state and neighbor status Take action The value is expressed as follows: in, represents the immediate reward, γ represents the discount factor, represents the updated state of the agent; represents the updated action of the ith agent; represents the updated state of the neighboring agent; According to real-time data, the agent dynamically adjusts its strategy. The strategy adjustment formula is as follows: in, represents the strategy of the ith agent at time t, α represents the learning rate, Represents the gradient of the Q-value function; Each agent performs collaborative optimization by sharing information and resources, as shown below: Among them, O represents the result of collaborative optimization, represents the strategy of the ith agent, Represents the state of the ith agent.

8. The SPS trolley material distribution method based on digital twin according to claim 1 is characterized in that: The global loss function of the federated learning model It is expressed as follows: in, represents the local loss function of the ith factory, n i represents the number of data samples of the ith factory, n represents the total number of data samples of all factories, represents the local dataset, and w represents the global model; Through the iterative process of federated learning, the global model of the federated learning model is optimized to minimize the global loss function, which is expressed as follows: Among them, w * Represents the updated global model.

9. The SPS trolley material distribution method based on digital twin according to claim 1 is characterized in that: The S6 specifically includes: S601. Create a dedicated blockchain network, including initialization and configuration of each node; S602: Define the data that needs to be uploaded to the blockchain, package the data records into transactions, and sign the transactions using public and private key pairs; S603. Write a smart contract to manage the business logic of material distribution, define the interface of the smart contract, and deploy the smart contract to the blockchain network. Each node can call the smart contract interface to perform corresponding operations. S604. Define the consensus mechanism of the blockchain network. When a new transaction is broadcast to the entire network, the nodes verify the validity of the transaction through the consensus mechanism and package it into a new block. The new block is added to the blockchain and reaches consensus through the consensus mechanism. S605. Provide a query interface based on blockchain, and users can query the status and history of materials through smart contracts.

10. An SPS trolley material distribution system based on digital twins, characterized in that: The system comprises: A data acquisition module is used to install a variety of sensors in the factory environment to collect real-time operation data of production equipment, materials and SPS trolleys, and pre-process the real-time operation data to build a digital twin model of the physical factory and SPS trolley system; wherein the sensors include position sensors, speed sensors, weight sensors, environmental sensors and state sensors; The digital twin model is constructed as follows: Using the collected location information, a three-dimensional spatial model of the factory environment and the SPS trolley system is constructed, as shown below: P i =(x i ,y i ,z i ) Among them, P i represents the three-dimensional coordinates of the i-th sensor; Using the data from weight sensors and position sensors, a 3D model of the material is constructed. j , which is expressed as follows: M j =(x j ,y j ,z j ,w j ) Among them, x j ,y j ,z j Respectively represent the x, y, and z coordinates of the material, w j Indicates the weight of the material; Build a model of the factory environment based on data from environmental sensors: E=(T,H) Among them, T represents temperature, H represents humidity, and E represents the model of the factory environment; Build a device status model based on the data from the status sensor: S=(s1,s2,…,s l ) Among them, s l represents the state of the lth device, and S represents the device state model; Use stream processing technology to process the data transmitted by sensors in real time and update the digital twin model in real time, as shown below: D(t)={P i (t),M j (t),E(t),S(t)} Where D(t) represents the data stream at time t, which contains the real-time data of all sensors, materials, environment and equipment status; P i (t) represents the three-dimensional coordinates of the i-th sensor at time t, M j (t) represents the three-dimensional model of the material at time t, E(t) represents the model of the factory environment at time t, and S(t) represents the equipment status model at time t; Leverage real-time data streams to dynamically update the state of the digital twin model: AND t+1 =And t +Δt·E change S t+1 =S t +Δt·S change Where Δt represents the time increment, represents the velocity component at the ith position, represents the velocity component at the jth position, E change Indicates the change of environmental parameters, S change Indicates the change in device status. represents the updated 3D coordinates, The 3D model of the updated material, E t+1 The model representing the updated factory environment, S t+1 Represents the updated device state model; The production scheduling module is used to build a scheduling model based on reinforcement learning according to the digital twin model, and optimize the material distribution path and strategy of the SPS trolley in real time according to the scheduling model; The material forecasting module is used to analyze and model historical and real-time data using graph neural networks to predict the material demand of each production node; The multi-node scheduling module is used to dynamically adjust the scheduling strategy by building a multi-agent collaborative scheduling model based on the SPS trolley according to the predicted material demand of each production node; Data sharing module, used to build a federated learning model to share and collaboratively optimize data among multiple factories; The data traceability module is used to build a blockchain network to manage material distribution data and record each material distribution of the SPS trolley.

Images