A smart factory raw material inventory warning and replenishment system

Abstract

The present invention discloses a smart factory raw material inventory early warning and replenishment system, which belongs to the field of inventory management technology. The method includes: constructing a three-level digital twin model of factory-workshop-production line by collecting full-chain production data in real time; using a reinforcement learning algorithm to evaluate the supplier's performance capability, generate credit scores and priority lists, drive the replenishment strategy generator to dynamically output the Pareto optimal strategy solution set, and quantify the cost-risk indicators through Monte Carlo simulation; constructing a domino effect propagation model to predict the cascading impact of raw material delays and formulate a dual-track replenishment mechanism; the strategy library optimization model iteratively adjusts replenishment parameters and updates supplier credit scores through real-time data feedback, forming a closed-loop optimization; the present invention realizes the active prediction of supply chain risks, dynamic optimization of replenishment strategies and differentiated management of suppliers, thereby improving production continuity and resource utilization efficiency, and reducing supply chain operating costs and cascading risks.

Description

Technical Field

[0001] The present invention belongs to the technical field of inventory management, and in particular to a smart factory raw material inventory early warning and replenishment system. Background Art

[0002] In the field of smart manufacturing, raw material inventory management directly impacts production efficiency and cost control. Traditional inventory management systems suffer from the following issues: traditional replenishment strategies rely on static safety stock thresholds and cannot dynamically simulate the impact of different replenishment strategies on production; supplier response delays lead to a mismatch between automatic replenishment orders and actual production capacity; the lack of topological modeling for supply chain risk transmission results in poor resilience of replenishment strategies; and the disconnection between IoT data and business systems, resulting in a lack of real-time visualization support for manual decision-making. Summary of the Invention

[0003] In response to the shortcomings of the existing technology, the present invention proposes a smart factory raw material inventory early warning and replenishment system. By collecting real-time production data from the entire chain, a three-level digital twin model of factory-workshop-production line is constructed; a reinforcement learning algorithm is used to evaluate the supplier's performance capability, generate credit scores and priority lists, drive the replenishment strategy generator to dynamically output the Pareto optimal strategy solution set, and quantify the cost-risk indicators through Monte Carlo simulation; a domino effect propagation model is constructed to predict the cascading impact of raw material delays and formulate a dual-track replenishment mechanism; the strategy library optimization model iteratively adjusts replenishment parameters and updates supplier credit scores through real-time data feedback, forming a closed-loop optimization; the present invention realizes the active prediction of supply chain risks, dynamic optimization of replenishment strategies and differentiated management of suppliers, thereby improving production continuity and resource utilization efficiency, and reducing supply chain operating costs and cascading risks.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A smart factory raw material inventory early warning and replenishment system, including: a data acquisition module, a digital twin modeling module, a replenishment strategy module, a risk prediction module, and an evaluation and optimization module;

[0006] The replenishment strategy module inputs the full-chain production data acquired in real time by the data acquisition module and the three-level digital twin model of the digital twin modeling module into a pre-set replenishment strategy generator, and combines the reinforcement learning algorithm and the Monte Carlo simulation method to obtain the Pareto optimal strategy solution set; the full-chain production data includes the raw material consumption rate, the three-dimensional warehouse location distribution, the supplier equipment load rate, and the logistics blockchain timestamp; the three-level digital twin model is based on the Petri net-Agent joint modeling configuration;

[0007] The risk prediction module constructs a domino effect propagation model based on the Pareto optimal strategy solution set and the three-level digital twin model, and formulates a dual-track replenishment mechanism. It combines the full-chain production data and the strategy library optimization model configured in the evaluation and optimization module to dynamically update the strategy parameters and transmit them back to the replenishment strategy module.

[0008] Specifically, the steps for constructing the three-level digital twin model include:

[0009] A1: Define the three elements of the system and establish the dynamic equation of inventory change; the three elements of the system include places, transitions, and directed arcs;

[0010] A2: Based on the production hierarchy reflected by the full-chain production data, define intelligent agents. Simultaneously, analyze the interaction patterns between different layers in the full-chain production data and set interaction rules between intelligent agents. The intelligent agents are factories, workshops, and production lines.

[0011] A3: Combine the discrete events described in the three elements of the system with the autonomous decision-making process of the intelligent agent to form a three-level digital twin model of factory-workshop-production line.

[0012] Specifically, the specific steps of A3 include:

[0013] A3.1: Identify system state variables and combine all system state variables to form a system state space. The system state variables are the state information represented by the library in the system's three elements. Each dimension in the system state space corresponds to a system state variable.

[0014] A3.2: Generate a discrete event list based on the three elements of the system and the system state space;

[0015] A3.3: Based on the agent, generate a list of decision events by combining the agent's decisions under different system states;

[0016] A3.4: For each discrete event in the discrete event list, analyze the change in system state after the discrete event occurs, combined with its triggering conditions. Based on the change in system state, determine the corresponding agent decision for each discrete event from the decision event list, thereby obtaining a mapping relationship between discrete events and agent decisions.

[0017] Specifically, the specific steps of A3 further include:

[0018] A3.5: Build a production-line digital twin model with the production line agent as the core, combining the production line discrete event list and the mapping relationship between discrete events and agent decisions.

[0019] A3.6: Based on the production line-level digital twin model, consider the decision-making of the workshop agent and the collaborative relationship between different production lines within the workshop. Use mapping relationships and workshop-related discrete events to build a workshop-level digital twin model.

[0020] A3.7: Integrate all workshop-level models, combine the macro-decision-making and mapping relationships of the factory agent with discrete events at the factory level, and build a three-level digital twin model of factory-workshop-production line.

[0021] A3.8: Compare the output results of the three-level digital twin model of factory-workshop-production line with the actual full-chain production data, and adjust the parameters of the three-level digital twin model based on the comparison results.

[0022] Specifically, the process of solving the Pareto optimal strategy solution set includes:

[0023] B1: Extract supplier data from the entire production chain; the supplier data includes delivery time, product quality, and order completion rate;

[0024] B2: Extracting features from the supplier data to obtain supplier characteristic data; the supplier characteristic data includes on-time delivery rate and product qualification rate;

[0025] B3: Use statistical methods to construct a supplier performance capability assessment model and use supplier characteristic data to train the supplier performance capability assessment model to obtain a trained supplier performance capability assessment model;

[0026] B4: Use the trained supplier performance assessment model, combined with full-chain production data, to calculate each supplier's credit score. This credit score is obtained by using a linear regression model to perform a weighted summation based on the regression coefficient and input features.

[0027] B5: Sort suppliers according to their credit scores and generate a supplier priority list;

[0028] B6: Define replenishment policy parameters, including order quantity and trigger threshold;

[0029] B7: Develop replenishment strategy generation rules based on full-chain production data, the three-level digital twin model, suppliers’ credit scores, supplier priority lists, and replenishment strategy parameters;

[0030] B8: Obtain full-chain production data, three-level digital twin models, and supplier credit scores in real time, and input them into the replenishment strategy generator. Based on the replenishment strategy generation rules, generate N candidate replenishment strategies.

[0031] Specifically, the process of solving the Pareto optimal strategy solution set further includes:

[0032] B9: Set up Monte Carlo simulation scenarios based on full-chain production data;

[0033] B10: Perform Monte Carlo simulations on each candidate replenishment strategy and record the cost and risk indicators of each simulation;

[0034] B11: Perform statistical analysis on the cost and risk indicators of each candidate replenishment strategy to obtain the average cost and risk value of each candidate replenishment strategy;

[0035] B12: Based on the obtained average cost and risk indicators, determine whether each candidate replenishment strategy is a Pareto optimal strategy;

[0036] B13: Filter out all strategies that are judged to be Pareto optimal to form a Pareto optimal strategy solution set; the Pareto optimal strategy solution set includes order quantity, trigger threshold and supplier priority information.

[0037] Specifically, the B12 specific steps include:

[0038] B12.1: Obtain the average cost and risk index of all candidate replenishment strategies and perform normalization to obtain the normalized average cost of the candidate replenishment strategies. and risk indicators ,and ;

[0039] B12.2: Define a dominance relationship, where the condition for candidate replenishment strategy j to dominate candidate replenishment strategy i is: and ,in, and The normalized average cost of the i-th and j-th candidate replenishment strategies, respectively, and denote the normalized risk indicators of the i-th and j-th candidate replenishment strategies respectively;

[0040] B12.3: Check for each candidate replenishment strategy whether there is at least one candidate replenishment strategy j that satisfies the dominance condition;

[0041] If there is no candidate replenishment strategy j that satisfies the dominance condition, then the candidate replenishment strategy i is the Pareto optimal strategy.

[0042] Specifically, the risk prediction module constructs a domino effect propagation model based on the Pareto optimal strategy solution set and the three-level digital twin model, and formulates a dual-track replenishment mechanism, including:

[0043] C1: Obtain the Pareto optimal strategy solution set and the three-level digital twin model;

[0044] C2: The supply chain network is abstracted into a directed graph, where different links in the supply chain are defined as nodes in the graph, and the dependencies between nodes are represented by edges. Links include raw material suppliers, production lines, workshops, and factories, and dependencies include material supply relationships and production collaboration relationships.

[0045] C3: Setting fault diffusion rules; the fault diffusion rules include forward diffusion rules and reverse forward diffusion rules;

[0046] The forward diffusion rule states that when there is a delay in the supply of raw materials, it will cause the production line to stop, which will lead to material shortages in the workshop, and the shortage of materials in the workshop will cause a decrease in factory-level production capacity;

[0047] The anti-forward diffusion rule is that when a factory-level emergency response is initiated, the workshop adjusts its priority, and after the workshop priority is adjusted, the production plan of the production line is rearranged.

[0048] Specifically, the risk prediction module constructs a domino effect propagation model based on the Pareto optimal strategy solution set and the three-level digital twin model, and formulates a dual-track replenishment mechanism, which also includes:

[0049] C4: Based on a three-level digital twin model, for each replenishment strategy in the Pareto optimal strategy solution set, we simulate the cascading impact of fault diffusion rules at the production line, workshop, and factory levels. Through simulation, we analyze the extent and scope of the impact of faults starting from the raw material supply side and spreading to the production line, workshop, and factory under different replenishment strategies.

[0050] C5: Determine the safety stock strategy based on the Pareto optimal solution set and replenish raw materials according to the replenishment cycle and replenishment quantity;

[0051] Set a risk threshold. When the simulation results predict that the cascading risk reaches or exceeds the risk threshold, it triggers supplier collaborative replenishment or emergency procurement.

[0052] C6: Generate a risk heat map. Based on the simulation results and cascading impact analysis, mark the potential risk nodes in the supply chain network and display the diffusion path of the failure on the heat map. The potential risk nodes include the supplier nodes with delayed raw materials and the lagging production line nodes.

[0053] Specifically, the combination of full-chain production data and the strategy library optimization model configured in the evaluation and optimization module to dynamically update strategy parameters includes:

[0054] D1: Extract the Pareto optimal strategy solution set and dual-track replenishment execution records, and calculate the statistics of the dual-track replenishment execution records to obtain a structured data set; the dual-track replenishment execution records include supplier collaborative replenishment response time and emergency procurement cost data; the statistics include mean and variance;

[0055] D2: Based on the structured data set, a strategy evaluation indicator system is constructed, including replenishment cost indicator, response time indicator, and system robustness indicator;

[0056] D3: Construct state space equations based on strategy evaluation indicators;

[0057] D4: Use the extended Kalman filter to perform state estimation on the state space equation to obtain parameter estimates, and use the gradient descent method to optimize the parameter estimates to obtain the optimized strategy parameters.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] 1. This paper proposes a smart factory raw material inventory early warning and replenishment system. Through real-time, full-chain data collection and three-level digital twin modeling, it achieves accurate mapping and prediction of supply chain dynamics. It employs Petri net-agent joint modeling technology, breaking through the limitations of traditional discrete event simulation and enabling detailed simulation of the collaborative relationships and material flow patterns across factories, workshops, and production lines. A supplier credit assessment model constructed using a reinforcement learning algorithm dynamically quantifies suppliers' performance capabilities. The integration of Monte Carlo simulation and Pareto optimization ensures the optimal cost-risk balance of replenishment strategies, enhancing the robustness of strategy generation.

[0060] 2. This invention proposes a smart factory raw material inventory early warning and replenishment system. This dual-track replenishment mechanism, based on the domino effect propagation model, effectively identifies and blocks the cascading risk of raw material delays, reducing the probability of production line shutdowns and factory capacity losses. Visual reports, through heat maps and path analysis, enable intuitive location of risk nodes and dynamic tracking of diffusion paths. The strategy library optimization model, through real-time data feedback and credit score updates, forms a closed-loop management system of data collection, strategy generation, and execution optimization, enabling the supply chain system to have adaptive adjustment capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a structural diagram of a smart factory raw material inventory warning and replenishment system;

[0062] Figure 2 Build a flow chart for the three-level digital twin model of factory-workshop-production line;

[0063] Figure 3 The flowchart for solving the Pareto optimal strategy solution set;

[0064] Figure 4 This is the workflow diagram of the risk prediction module. DETAILED DESCRIPTION

[0065] Example 1

[0066] See also Figure 1 The present invention provides an embodiment of a smart factory raw material inventory early warning and replenishment system, comprising the following steps:

[0067] Data acquisition module, digital twin modeling module, replenishment strategy module, risk prediction module, evaluation and optimization module;

[0068] A data acquisition module is used to obtain real-time full-chain production data, including smart factory raw material inventory warnings and replenishment. The full-chain production data includes raw material consumption rate, warehouse three-dimensional location distribution, supplier equipment load rate, and logistics blockchain timestamps.

[0069] Among them, the raw material consumption rate is obtained by installing flow sensors on the raw material delivery pipeline to monitor the use of raw materials in the production process in real time, so as to accurately predict the demand for raw materials; the three-dimensional warehouse location distribution is to use lidar scanning technology to scan the warehouse space in real time, and at the same time install position sensors on the shelves to obtain the three-dimensional warehouse location distribution information, understand the storage location and space usage of raw materials; the supplier equipment load rate is to obtain the load rate of the supplier equipment through data interaction with the supplier's equipment management system, and evaluate the supplier's production capacity and supply stability; the logistics blockchain timestamp is to obtain the timestamp information of the raw material transportation process from the logistics blockchain system, accurately grasp the logistics progress, and predict the arrival time of raw materials.

[0070] The digital twin modeling module builds a three-level digital twin model of the factory, workshop, and production line based on the full-chain production data obtained by the data acquisition module, digitally mapping and simulating the factory's actual production process;

[0071] The replenishment strategy module is used to input the real-time full-chain production data from the data acquisition module and the three-level digital twin model from the digital twin modeling module into the discrete event simulation engine, use the reinforcement learning algorithm to generate candidate replenishment strategies, and evaluate these strategies through Monte Carlo simulation to output the Pareto optimal strategy solution set;

[0072] The risk prediction module builds a domino effect propagation model based on the three-level digital twin model and the Pareto optimal strategy solution set of the replenishment strategy module, predicts the diffusion path of delay events, and formulates a dual-track replenishment mechanism based on the prediction results, while also providing predictive visualization.

[0073] The evaluation and optimization module is used to establish a supplier performance capability evaluation model, combine the full-chain production data of the data acquisition module, calculate the supplier's credit score, optimize the replenishment strategy library based on the reinforcement learning algorithm, dynamically update the optimal strategy parameters, and return the optimized parameters to the replenishment strategy module.

[0074] The replenishment strategy module includes: discrete event simulation engine unit, replenishment strategy generation unit, and Monte Carlo simulation evaluation unit;

[0075] The discrete event simulation engine unit is used to simulate discrete events in the production process, such as raw material consumption, order completion, and replenishment triggering. It combines real-time full-chain production data and a three-level digital twin model to provide a dynamic simulation environment for the generation and evaluation of replenishment strategies.

[0076] The replenishment strategy generation unit, based on a reinforcement learning algorithm, learns and optimizes in a discrete event simulation engine environment to generate candidate replenishment strategies, taking into account factors such as order quantity, trigger threshold, and supplier priority.

[0077] The Monte Carlo simulation evaluation unit generates random scenarios through Monte Carlo simulation, simulates and evaluates the generated replenishment strategies, calculates the performance indicators of each replenishment strategy under different scenarios, and screens out the Pareto optimal strategy solution set.

[0078] The risk prediction module includes: model building unit, dual-track replenishment mechanism unit, and predictive visualization unit;

[0079] Model building unit, used to analyze the dependencies between various links in the production process, build a domino effect propagation model, and predict the diffusion path and impact range of delay events in the production chain;

[0080] The dual-track replenishment mechanism unit, based on the prediction results of the domino effect propagation model, develops a dual-track replenishment mechanism that combines the main replenishment channel and the emergency channel to improve the ability to respond to delay events;

[0081] The predictive visualization unit is used to visualize the prediction results of delay events and the dual-track replenishment mechanism, helping managers understand risk situations and make decisions.

[0082] The evaluation and optimization module includes: capability evaluation unit, credit score calculation unit, strategy optimization unit, and parameter feedback unit;

[0083] The capability assessment unit is used to define assessment indicators and methods, establish a supplier performance capability assessment model, and calculate the supplier's credit score by comprehensively considering factors such as the supplier's equipment load rate and delivery on-time rate;

[0084] The credit score calculation unit calculates each supplier's credit score based on the supplier's performance capability assessment model and the full-chain production data provided by the data collection module, reflecting the supplier's performance capability and reliability;

[0085] The strategy optimization unit uses a reinforcement learning algorithm to optimize the replenishment strategy library in combination with the supplier's credit score, dynamically adjusting the order quantity, trigger threshold, and supplier priority parameters to improve the effectiveness and adaptability of the replenishment strategy;

[0086] The parameter feedback unit is used to return the optimized optimal strategy parameters to the replenishment strategy module, update the input of the strategy generator, and achieve continuous optimization of the strategy.

[0087] Furthermore, the overall implementation process of a smart factory raw material inventory warning and replenishment system includes:

[0088] Step S1: Acquire full-chain production data in real time; the full-chain production data includes raw material consumption rate, warehouse three-dimensional storage location distribution, supplier equipment load rate, and logistics blockchain timestamp;

[0089] Step S2: Based on the production data of the entire chain, a three-level digital twin model of factory-workshop-production line is constructed using the Petri net-Agent joint modeling method;

[0090] It should be noted that Petri net, as a discrete event modeling tool, characterizes system state transitions and event triggering logic through the flow of libraries, transitions and tokens; Agent has autonomous decision-making capabilities and handles dynamic uncertainties, such as abnormal responses, optimization decisions and human-computer interaction. The Petri net-Agent joint modeling method in the present invention integrates deterministic rules, handles dynamic uncertainties, and enhances system robustness. By integrating discrete event modeling and autonomous decision-making capabilities, it provides an efficient and dynamic system characterization method for the three-level digital twin model of factory-workshop-production line, solving the problems of complex modeling, dynamic response and multi-objective optimization in manufacturing systems.

[0091] Step S3: Establish a supplier performance assessment model, combine full-chain production data with reinforcement learning algorithms, calculate the supplier's credit score, and generate a priority list;

[0092] Step S4: Set up a replenishment strategy generator and input real-time full-chain production data, the three-level digital twin model, and the supplier's credit score to generate candidate replenishment strategies. Perform a cost-risk assessment on the generated candidate replenishment strategies through Monte Carlo simulation and output a Pareto optimal strategy solution set. At the same time, the supplier's credit score is introduced as a priority weight to optimize the candidate replenishment strategy parameters; the Pareto optimal strategy solution set includes order quantity, trigger threshold, and supplier priority.

[0093] Step S5: Build a domino effect propagation model based on the three-level digital twin model to predict the cascading impact of raw material delays on production lines, workshops, and factories. Develop a dual-track replenishment mechanism based on the prediction results and generate a visualization report. The visualization report identifies potential risk nodes and diffusion paths. The dual track includes a main replenishment channel and an emergency channel.

[0094] Step S6: Establish a strategy library optimization model, combine real-time full-chain production data with the dual-track replenishment execution effect, adjust the strategy parameters, feed back the revised strategy parameters to the strategy generator, and update the supplier credit score based on the replenishment execution results.

[0095] Example 2

[0096] See also Figure 2 , the steps of constructing the three-level digital twin model in this embodiment include:

[0097] A1: Define the three elements of the system and establish the dynamic equation of inventory change; the three elements of the system include places, transitions, and directed arcs;

[0098] Among them, places represent various states in the production system, such as raw material inventory, work-in-progress inventory, and finished product inventory; transitions represent events in the system, such as the purchase of raw materials, the start and end of production processing, and the shipment of products; directed arcs represent the relationship between places and transitions, describing the transfer of states when events occur. The dynamic equation of inventory change is based on the definition of the three elements of the system and analyzes the change rules of each state under the influence of different events. Taking the inventory state as an example, Describes the dynamic changes of inventory, where Indicates inventory quantity, Indicates the demand for the product, represents the replenishment quantity, and t represents the current time;

[0099] A2: Define intelligent agents and set interaction rules between them; the intelligent agents are factories, workshops, and production lines;

[0100] Among them, each intelligent agent represents a level in the production system, with different functions and decision-making authority. The factory intelligent agent is responsible for the overall production planning and resource allocation; the workshop intelligent agent arranges the production tasks of the workshop according to the factory plan; the production line intelligent agent focuses on specific production operations and equipment management; and the interaction rules between intelligent agents refer to the information transmission and collaboration methods between intelligent agents. For example, the factory intelligent agent issues production tasks to the workshop intelligent agent, and the workshop intelligent agent assigns work to the production line intelligent agent according to the task requirements; the production line intelligent agent feeds back the production progress and equipment status to the workshop intelligent agent, and the workshop intelligent agent then summarizes the information and reports it to the factory intelligent agent; at the same time, it stipulates the decision-making rules of the intelligent agent in different situations, such as how the intelligent agent should make replenishment decisions when the inventory is below a certain threshold.

[0101] A3: Combine the discrete events described by the three elements of the system and the autonomous decision-making process of the intelligent agent to form a three-level digital twin model of factory-workshop-production line.

[0102] Among them, when the discrete events described by the three elements of the system are combined with the autonomous decision-making process of the intelligent agent: when a discrete event occurs, the relevant intelligent agent makes a decision based on its own state and preset decision-making rules, thereby affecting the state change of the system. For example, when the discrete event of raw materials arriving occurs, the factory intelligent agent decides whether to adjust the subsequent procurement strategy based on the current inventory level and production plan.

[0103] The specific steps of A3 include:

[0104] A3.1: Identify system state variables and combine all system state variables to form a system state space. The system state variables are the state information represented by the library in the system's three elements. Each dimension in the system state space corresponds to a system state variable.

[0105] Furthermore, the specific steps of A3.1 include:

[0106] (1) Obtaining full-chain production data;

[0107] (2) Perform three-element analysis of the system to obtain the system state variables;

[0108] Place analysis: Places represent the state in the system and directly correspond to system state variables. For example, the raw material place corresponds to the state variable of raw material inventory quantity; the equipment place corresponds to the state variables of equipment operation time, number of failures, etc.

[0109] Transition association analysis: Transitions represent events. The occurrence of events will cause changes in the system state. By analyzing the association between transitions and places, the state variables related to the event are determined. For example, the state variables associated with the raw material entry transition include the entry quantity and entry time.

[0110] Directed arc relationship analysis: Directed arcs describe the relationship between places and transitions, clarifying the mutual influence between state variables. For example, a directed arc from the raw material inventory place to the production task transition indicates that the raw material inventory quantity affects the execution of the production task, and the execution of the production task in turn affects the raw material inventory quantity.

[0111] (3) Use the Pearson correlation coefficient to calculate the correlation between system state variables and remove the first A system state variables with the largest correlation;

[0112] (4) According to the degree of influence of system state variables on system performance and decision-making, principal component analysis is used to evaluate their importance, and the top B system state variables with the highest importance are selected as the final system state variables;

[0113] (5) According to the final system state variables, the dimension of the system state space is obtained, where the dimension of the system state space is equal to the number of the final system state variables;

[0114] (6) Determine the value range of each system state variable based on actual production data and business needs;

[0115] (7) Combine all system state variables to form the system state space.

[0116] A3.2: Based on the three system elements and the system state space, generate a list of discrete events, where each discrete event is associated with certain state variables in the system state space and has its triggering conditions.

[0117] For example, when the inventory quantity of a raw material inventory location falls below a preset threshold, this constitutes a discrete event; the start or end moment of a production process transition is also a discrete event; the moment when material flows from one location to another along a directed arc is also a discrete event. These events are recorded to form a discrete event list.

[0118] Furthermore, the specific steps of A3.2 include:

[0119] (1) Define the three elements of the system, system state variables and system state space;

[0120] (2) Define the basic discrete event set ,in, represents the ath basic discrete event, a represents the number of basic discrete events, for example, assuming a=3, Indicates that the production line starts production (triggering raw material consumption), Indicates that the supplier has shipped the goods (triggering an increase in inventory), Indicates equipment failure (triggering production interruption);

[0121] (3) For each basic discrete event , define its triggering conditions and state transition rules;

[0122] For example, a trigger condition: When the raw material inventory quantity falls below a preset safety threshold, a replenishment event is triggered. A state transition rule: After the event is triggered, the state variables are updated according to the rules. For example, if the change in raw material inventory equals the consumption rate multiplied by the production cycle and then takes the opposite value, a production event is triggered, and an event-state transition comparison table is generated, as shown in Table 1.

[0123] Table 1: Event-state transition comparison table

[0124]

[0125] (4) Discretize the continuous state variables into finite intervals and construct a state transition diagram. It is necessary to ensure that all possible system state transitions are covered by events, and when multiple events meet the triggering conditions at the same time, priority rules must be defined.

[0126] A3.3: Based on the agent, combine the agent's decisions under different system states to generate a list of decision events. For example, a factory agent might make a replenishment decision when raw material inventory is low.

[0127] For example, the decision events of the factory agent may include adjusting the production plan, selecting different suppliers, and determining the replenishment quantity; the decision events of the workshop agent may include reallocating production tasks and adjusting equipment operating parameters; the decision events of the production line agent may include changing the production rhythm and activating backup equipment. These decision events are organized into a list to obtain a decision event list.

[0128] Furthermore, the specific steps of A3.3 include:

[0129] (1) Obtain the defined intelligent agent and, based on the system state space constructed in A3.1, divide the system state into different categories or intervals. For example, according to the raw material inventory level, it can be divided into low inventory state, normal inventory state, and high inventory state; according to the equipment load rate, it can be divided into low load, medium load, and high load state;

[0130] (2) Determine the agent’s decision-making rules:

[0131] Factory agent decision rules: Under different system states, the factory agent makes decisions such as adjusting production plans, selecting suppliers, and arranging resource allocation. For example, when raw material inventory is low and market demand is high, the factory agent may decide to increase procurement and prioritize the production of important orders.

[0132] Decision-making rules for workshop agents: Workshop agents are primarily responsible for coordinating production activities within the workshop. Their decisions may include task allocation, equipment scheduling, and personnel arrangements. For example, when equipment load is high, the workshop agent may decide to assign some tasks to other idle equipment.

[0133] Production line agent decision rules: The production line agent focuses on specific production operations on the production line. Its decisions may include adjusting production speed, changing production processes, and handling equipment failures. For example, when a production line device fails, the production line agent may decide to suspend production and notify maintenance personnel.

[0134] (3) Based on the decision-making rules of each agent in different system states, list all possible decision events, such as:

[0135] Decision events of the factory agent: increase purchase volume, adjust production plan priority, select new suppliers;

[0136] Decision-making events of workshop agents: task reallocation, equipment scheduling adjustments, and personnel overtime arrangements;

[0137] Decision-making events of the production line agent: reducing production speed, changing tools, and starting spare equipment;

[0138] (4) For each decision event, clearly define the system state conditions and corresponding agents that trigger it. For example, the trigger condition for the “increase purchase volume” decision event may be that the raw material inventory is below the safety threshold, and the corresponding agent is the factory agent.

[0139] (5) Check the generated decision event list and remove unreasonable decision events; unreasonable decision events refer to decision events that require resources that do not exist in the system at all;

[0140] (6) Optimize the triggering conditions and decision content of decision events based on actual production experience and expert knowledge.

[0141] A3.4: For each discrete event in the discrete event list, analyze the change in system state after the discrete event occurs, combined with its triggering conditions. Based on the change in system state, determine the corresponding agent decision for each discrete event from the decision event list, thereby obtaining a mapping relationship between discrete events and agent decisions.

[0142] Furthermore, the specific steps of A3.4 include:

[0143] (1) Define the trigger condition function for each discrete event;

[0144] (2) For each discrete event, define the rules for how the system state changes after it occurs;

[0145] (3) Generate the system state after the state change for each discrete event;

[0146] (4) According to the system state after the state change, the corresponding agent decision is matched from the decision event list, and then a mapping table between discrete events and agent decisions is generated.

[0147] A3.5: With the production line agent as the core, build a production line-level digital twin model by combining the production line discrete event list and the mapping relationship between discrete events and agent decisions. This production line-level digital twin model should accurately reflect the production process and performance indicators of the production line.

[0148] Furthermore, the specific steps of A3.5 include:

[0149] (1) Collect basic data of the production line, including equipment information, process parameters and raw material information of the production line. The equipment information of the production line includes equipment type, performance parameters and service life; the process parameters include processing speed and processing accuracy; the raw material information includes raw material specifications and consumption rate;

[0150] (2) Based on the system state space and the actual production line situation, determine the state variables at the production line level, such as equipment operating status, raw material inventory level, and the number of work-in-progress. The equipment operating status includes normal, faulty, and maintenance.

[0151] (3) Filter out production line discrete events from the discrete event list, such as equipment startup, equipment failure, raw material replenishment, and production task start;

[0152] (4) Define trigger conditions for each discrete event of the production line. These trigger conditions are usually related to the state variables of the production line. For example, the trigger condition of an equipment failure event may be that the equipment operating time exceeds a certain threshold or a certain performance indicator of the equipment exceeds the normal range;

[0153] (5) Based on the mapping relationship between discrete events and agent decisions, clarify the change rules of the production line state variables after each discrete event occurs. For example, when the equipment start event occurs, the equipment operating state changes from "stop" to "run", and a certain amount of raw materials will be consumed at the same time;

[0154] (6) Analyze the impact of production line agent decisions on production line state variables. For example, a decision to suspend production will result in the number of work-in-progress no longer increasing, and may also affect the scheduling of production tasks.

[0155] (7) Using the Petri net method to construct the architecture of the production line-level digital twin model, that is, using the Petri net to describe the occurrence and state transition process of discrete events in the production line. The Petri net method is the existing technology content in this field and is not the inventive solution of this application, so it will not be described in detail here;

[0156] (8) Integrate the production line state variables, discrete events, state transition rules and intelligent agent decisions into the architecture of the production line-level digital twin model to form a production line-level digital twin model.

[0157] A3.6: Based on the production line digital twin model, consider the decision-making of the workshop agent and the collaborative relationship between different production lines within the workshop. Utilize mapping relationships and workshop-related discrete events to construct a workshop-level digital twin model. This workshop-level digital twin model should reflect the overall production efficiency and resource utilization of the workshop.

[0158] Furthermore, the specific steps of A3.6 include:

[0159] (1) Integrate the constructed digital twin models of each production line to form a preliminary set of production lines in the workshop, where each production line model contains its own state variables, discrete events, and decision logic;

[0160] (2) Obtaining the layout information, resource allocation, and workshop-level management rules of the workshop. The workshop layout information includes the location of production lines and material transportation routes, the resource allocation includes manpower and equipment sharing, and the workshop-level management rules include production priorities and scheduling strategies.

[0161] (3) Filter out workshop discrete events from the discrete event list and determine the state variables at the workshop level. The workshop discrete events include workshop power outage, emergency order arrival, and workshop equipment maintenance plan start-up. The state variables at the workshop level include the overall production progress of the workshop, the total inventory of the workshop, and the comprehensive utilization rate of the workshop equipment.

[0162] (4) Based on the management responsibilities and goals of the workshop, determine the decisions that the workshop agent may make under different system states, such as adjusting the production line production task allocation, coordinating equipment sharing, and arranging personnel scheduling;

[0163] (5) Analyze the impact of each decision on the state variables of the workshop and the production line. For example, adjusting the production task allocation decision of the production line will change the production load and the number of work-in-progress of each production line, thereby affecting the overall production progress of the workshop.

[0164] (6) Establish coordination rules between different production lines within the workshop based on the workshop’s production processes and resource constraints. For example, when the output of one production line is the input of another, it is necessary to ensure that the production rhythms of the two production lines match; or in the case of equipment sharing, it is necessary to coordinate the order in which each production line uses the equipment;

[0165] (7) Convert the collaborative rules into mathematical models to describe the mutual influence and constraint relationships between production lines;

[0166] (8) Using the system dynamics approach to build a workshop-level digital twin model architecture to integrate the production line model, workshop discrete events, workshop agent decision-making, and production line collaboration;

[0167] (9) Integrate the workshop state variables, workshop discrete events, state transition rules and agent decisions into the workshop-level digital twin model architecture to form a workshop-level digital twin model; the workshop-level digital twin model can simulate the dynamic changes in the status of each production line and the overall status of the workshop after the workshop agent makes a decision when different discrete events occur.

[0168] A3.7: Integrate all workshop-level models, combine the macro-decision-making and mapping relationships of the factory agent with discrete events at the factory level, and build a three-level digital twin model of factory-workshop-production line. This three-level digital twin model should be able to reflect the production operation status and economic benefits of the entire factory.

[0169] Furthermore, the specific steps of A3.7 include:

[0170] (1) Unify and standardize the data format, state variable definition, and event representation of each workshop-level digital twin model to ensure data compatibility between different workshop models;

[0171] (2) All workshop-level models are combined into a whole to construct a collective model of the workshop layer. In the integration process, the material flow and information interaction relationships that may exist between workshops should be considered;

[0172] (3) Filter out discrete events related to the overall operation of the factory from the discrete event list, such as changes in market demand, equipment failure, and supplier delivery delays, and determine factory-level state variables, such as overall factory production efficiency, order delivery rate, total factory inventory, and cash flow;

[0173] (4) Based on the factory’s strategic goals and management responsibilities, determine the macro decisions that the factory agent may make under different system states, such as adjusting production plans, changing product mix, optimizing resource allocation, and adjusting cooperation strategies with suppliers and customers;

[0174] (5) Analyze the impact of each decision on the factory state variables and workshop state variables. For example, adjusting the production plan decision will change the production task allocation of each workshop, thereby affecting the production progress and equipment utilization of the workshop;

[0175] (6) Using the established mapping relationship between discrete events and agent decisions, associate discrete events at the factory level with the decisions of the factory agent. It is important to ensure that the corresponding factory agent decision can be accurately triggered when a specific discrete event occurs. At the same time, it is also necessary to analyze how the factory agent decision is transmitted to the workshop level and production line level, and the impact of these decisions on the state variables at each level during the transmission process. For example, the factory's resource allocation decision will affect the resource acquisition of the workshop, and thus affect the production capacity of the production line;

[0176] (7) Using system simulation methods to build a three-level digital twin model architecture of factory-workshop-production line to integrate the workshop-level model, factory discrete events, factory intelligent agent decision-making and the relationship between each level;

[0177] (8) Integrate the overall factory operation discrete events, factory-level state variables, state transition rules and intelligent agent decisions into the three-level digital twin model architecture of factory-workshop-production line to form a three-level digital twin model of factory-workshop-production line; the three-level digital twin model can simulate the dynamic changes in the status of each workshop and production line after the factory intelligent agent makes a decision when different discrete events occur in the factory.

[0178] A3.8: Compare the output results of the three-level digital twin model of factory-workshop-production line with the actual full-chain production data. Based on the comparison results, adjust the parameters of the three-level digital twin model, including the probability of occurrence of discrete events and the weight of intelligent agent decisions, to make the model more consistent with actual production conditions.

[0179] Example 3

[0180] See also Figure 3 This embodiment discloses a process for solving the Pareto optimal strategy solution set, including:

[0181] B1: Extract supplier data from the entire production chain; the supplier data includes delivery time, product quality, and order completion rate;

[0182] B2: Extracting features from supplier data to obtain supplier characteristic data; the supplier characteristic data includes on-time delivery rate and product qualification rate. Statistical calculation methods are used to extract features from supplier data. The specific process includes:

[0183] (1) Obtain supplier data and clean it;

[0184] (2) Based on the cleaned supplier data, the on-time delivery rate is calculated by calculating the ratio of the number of orders delivered on time to the total number of orders and then multiplying it by the percentage. The product qualification rate is calculated by calculating the ratio of the number of qualified products to the total number of delivered products and then multiplying it by the percentage. The order completion rate is calculated by calculating the number of actual delivered orders to the number of planned delivered orders and then multiplying it by the percentage.

[0185] (3) Generate a statistical feature table of the supplier dimension based on on-time delivery rate, product qualification rate, and order completion rate;

[0186] (4) Filter the features in the statistical feature table of the supplier dimension and select the top M features as the supplier feature data.

[0187] B3: Using statistical methods to construct a supplier performance capability assessment model, and using the supplier characteristic data to train the supplier performance capability assessment model to obtain a trained supplier performance capability assessment model. The statistical method uses a linear regression method, which is a prior art in this field and does not constitute an inventive solution of this application, and is not described in detail here.

[0188] B4: Use the trained supplier performance assessment model, combined with full-chain production data, to calculate each supplier's credit score. The credit score is obtained by using a linear regression model to perform a weighted summation based on the regression coefficient and input features. The credit score is a comprehensive indicator that reflects the supplier's overall performance capability. The linear regression model is prior art in this field and does not constitute an inventive solution of this application, so it will not be described in detail here.

[0189] B5: Sort suppliers by their credit scores and generate a supplier priority list. Suppliers with higher credit scores have higher priorities.

[0190] B6: Define replenishment policy parameters, including order quantity and trigger threshold;

[0191] B7: Develop replenishment strategy generation rules based on full-chain production data, the three-level digital twin model, suppliers’ credit scores, supplier priority lists, and replenishment strategy parameters;

[0192] Furthermore, the specific steps of B7 include:

[0193] (1) Obtain full-chain production data, three-level digital twin models, supplier credit scores, supplier priority lists, and replenishment strategy parameters;

[0194] (2) Determine the trigger threshold of inventory level based on the full-chain production data and the three-level digital twin model;

[0195] When the inventory level falls below the threshold, a replenishment action is triggered. For example, when the raw material inventory falls below the safety stock level, the replenishment process is initiated.

[0196] (3) Combining the production plan and the prediction results of the three-level digital twin model, when it is expected When material demand in a time period exceeds current inventory, replenishment is triggered. For example, based on the production order schedule, raw material consumption can be predicted in advance to replenish stocks before peak demand arrives.

[0197] (4) Determine the order quantity based on inventory cost and stock-out cost, and select suppliers based on the supplier priority list;

[0198] (5) Integrate the replenishment trigger conditions, supplier selection, and order quantity determination rules to form replenishment strategy generation rules.

[0199] B8: Obtain full-chain production data, the three-level digital twin model, and the supplier's credit score in real time and input them into the replenishment strategy generator. Based on the replenishment strategy generation rules, N candidate replenishment strategies are generated, where each candidate strategy corresponds to a different set of order quantities, trigger thresholds, and supplier selections.

[0200] The logical execution process of candidate replenishment strategy generation includes:

[0201] (1) Trigger condition judgment: Real-time check of inventory levels or production demand to see if they reach replenishment thresholds;

[0202] (2) Supplier prioritization: Calculate the supplier's comprehensive score based on real-time credit score and preset weights;

[0203] (3) Order quantity calculation: Dynamically adjust the order quantity based on real-time demand forecast and cost model.

[0204] B9: Based on the production data of the entire supply chain, Monte Carlo simulation scenarios are set up, including demand fluctuations, supplier delivery delays, production failures, and transportation risk factors. Monte Carlo simulation is the prior art content in this field and does not constitute the inventive solution of this application, so it will not be described in detail here;

[0205] B10: Perform Monte Carlo simulation on each candidate replenishment strategy and record the cost and risk indicators of each simulation. The number of simulations is set to 1000.

[0206] B11: Perform statistical analysis on the cost and risk indicators of each candidate replenishment strategy, and calculate the average cost and risk value of each candidate replenishment strategy;

[0207] Furthermore, the specific steps of B11 include:

[0208] (1) Obtain a list of candidate replenishment strategies and full-chain production data, and set the number of simulations, time range, and cost parameters;

[0209] (2) For each candidate replenishment strategy, extract the core parameters and generate random variable samples; the random variable samples include demand fluctuation, delivery delay time and production yield rate;

[0210] (3) Based on the core parameters and random variable samples, the state transition equation is used to simulate the single-cycle inventory changes, where the inventory at time t+1 is the maximum value between the first factor value and zero; the first factor value is the current inventory plus the product of the order quantity and the yield rate, minus the difference between the actual demand and production consumption;

[0211] (4) Calculate the cost and risk of a single simulation:

[0212] Holding cost is the product of the unit inventory holding cost per unit time and the second factor value; the second factor value is the cumulative sum of the average of the current inventory and the next inventory, multiplied by the inverse of the replenishment cycle;

[0213] The out-of-stock cost is the product of the out-of-stock unit cost and the third factor value; the third factor value is obtained by summing the maximum value between the fourth factor value and zero during the replenishment cycle; the fourth factor value is the current inventory plus the product of the order quantity and the yield rate, then subtracting it from the actual demand, and taking the negation of the difference;

[0214] The risk indicators include out-of-stock rate and service level; the out-of-stock rate is the ratio of the number of out-of-stock cycles to the replenishment cycle; and the service level is the difference between 1 and the out-of-stock rate.

[0215] B12: Based on the obtained average cost and risk indicators, determine whether each candidate replenishment strategy is a Pareto optimal strategy, that is, whether it is impossible to reduce risk without increasing cost, or whether it is impossible to reduce cost without increasing risk;

[0216] B13: Filter out all strategies that are judged to be Pareto optimal to form a Pareto optimal strategy solution set; the Pareto optimal strategy solution set includes order quantity, trigger threshold and supplier priority information.

[0217] The specific steps for B12 include:

[0218] B12.1: Obtain the average cost and risk index of all candidate replenishment strategies and perform normalization to obtain the normalized average cost of the candidate replenishment strategies. and risk indicators ,and ;

[0219] B12.2: Define a dominance relationship, where the condition for candidate replenishment strategy j to dominate candidate replenishment strategy i is: and ,in, and The normalized average cost of the i-th and j-th candidate replenishment strategies, respectively, and denote the normalized risk indicators of the i-th and j-th candidate replenishment strategies respectively;

[0220] B12.3: For each candidate replenishment policy, check whether there is at least one candidate replenishment policy j that satisfies the dominance condition. If there is no candidate replenishment policy j that satisfies the dominance condition, then candidate replenishment policy i is the Pareto optimal policy.

[0221] Example 4

[0222] See also Figure 4 In this embodiment, the risk prediction module constructs a domino effect propagation model based on the Pareto optimal strategy solution set and the three-level digital twin model, and formulates a dual-track replenishment mechanism, including:

[0223] C1: Obtain the Pareto optimal strategy solution set and the three-level digital twin model;

[0224] C2: The supply chain network is abstracted into a directed graph, where different links in the supply chain are defined as nodes in the graph, and the dependencies between nodes are represented by edges. Links include raw material suppliers, production lines, workshops, and factories, and dependencies include material supply relationships and production collaboration relationships.

[0225] C3: Setting fault diffusion rules; the fault diffusion rules include forward diffusion rules and reverse forward diffusion rules;

[0226] The forward diffusion rule states that when there is a delay in the supply of raw materials, it will cause the production line to stop, which will further lead to material shortages in the workshop, and the material shortages in the workshop will eventually cause a decrease in factory-level production capacity;

[0227] The anti-forward diffusion rule is that when a factory-level emergency response is initiated, it will prompt the workshop to adjust its priority. After the workshop priority is adjusted, the production line production plan will be rearranged;

[0228] C4: Based on a three-level digital twin model, for each replenishment strategy in the Pareto optimal strategy solution set, we simulate the cascading impact of fault diffusion rules at the production line, workshop, and factory levels. Through simulation, we analyze the extent and scope of the impact of faults starting from the raw material supply side and spreading to the production line, workshop, and factory under different replenishment strategies.

[0229] The specific steps of C4 include:

[0230] (1) Obtain a three-level digital twin model, extract the Pareto optimal strategy solution set, and clarify the fault diffusion rules;

[0231] (2) Determine the total time range and time step of the simulation based on the actual production situation. In the present invention, the total time range is set to one month and the time step is set to one hour;

[0232] (3) Set the initial status of each component of the production line, workshop and factory based on real-time full-chain production data, including inventory levels, equipment operating status, and order progress;

[0233] (4) For each replenishment strategy in the Pareto optimal strategy solution set, perform the following simulation steps:

[0234] Introducing failure events: At the beginning of the simulation or during the simulation, introduce raw material delay failure events according to the preset probability distribution;

[0235] Fault propagation simulation: Based on fault propagation rules, the cascading impact of faults at the production line, workshop, and factory levels is simulated. Specifically, when a raw material delay occurs, the probability and duration of a production line shutdown due to material shortage are calculated. After the production line shutdown, the impact on the workshop's material supply is evaluated to determine the extent and duration of the workshop's material shortage. Finally, the impact of the workshop material shortage on factory-level production capacity is analyzed.

[0236] Update system status: At each time step, the status of production lines, workshops, and factories is updated based on simulation results, including inventory levels, equipment operating status, and order completion status.

[0237] (5) During the simulation process, record key indicators at each time step, such as production line downtime, workshop material shortage, and factory capacity loss;

[0238] (6) Analyze the simulation results under each replenishment strategy to evaluate the extent and scope of the impact of fault propagation on the production line, workshop and factory.

[0239] C5: Determine the safety stock strategy based on the Pareto optimal solution set and replenish raw materials according to the replenishment cycle and replenishment quantity;

[0240] Set a risk threshold. When the cascading risk predicted by the simulation results reaches or exceeds the risk threshold, supplier collaborative replenishment or emergency procurement is triggered.

[0241] C6: Generate a risk heat map. Based on the simulation results and cascading impact analysis, mark the potential risk nodes in the supply chain network and display the diffusion path of the failure on the heat map. The potential risk nodes include the supplier nodes with delayed raw materials and the lagging production line nodes.

[0242] Furthermore, the specific steps of C6 include:

[0243] (1) Obtain the results of fault diffusion simulation based on the three-level digital twin model, as well as the data obtained from the analysis of cascading impacts, for example, the failure probability, impact level, and diffusion path of each node in the supply chain network, such as raw material suppliers, production lines, workshops, and factories;

[0244] (2) Obtain the cascading risk predicted by the simulation results and the preset risk threshold;

[0245] (3) Use Python's matplotlib library to draw a heat map. Nodes are drawn in the map based on their positions and cascade values, and the corresponding colors are filled in to visually display the risk level of the nodes. In the map, low risk is represented by green, medium risk is represented by yellow, and high risk is represented by red.

[0246] (4) Based on the set risk threshold, screen out potential risk nodes, that is, nodes whose cascade risk value exceeds the risk threshold;

[0247] (5) Add labels to potential risk nodes on the heat map. The labeling content may include node name, risk index value, and risk level;

[0248] (6) Extract the fault diffusion path information from the simulation results and determine the nodes and edges passed through the path;

[0249] (7) Use different colors or line styles to draw the fault diffusion path on the heat map to clearly show the propagation direction and scope of the fault.

[0250] The strategy parameters are dynamically updated by combining the full-chain production data with the strategy library optimization model configured in the evaluation and optimization module, including:

[0251] D1: Extract the Pareto optimal strategy solution set and dual-track replenishment execution records, and calculate statistics of the dual-track replenishment execution records to obtain a structured data set; the dual-track replenishment execution records include supplier collaborative replenishment response time and emergency procurement cost data; the statistics include mean and variance. The calculation process of the mean and variance is the content of the prior art in this field and does not constitute the inventive solution of this application, and is not detailed here;

[0252] D2: Based on the structured data set, a strategy evaluation indicator system is constructed, including replenishment cost indicator, response time indicator, and system robustness indicator;

[0253] Among them, the replenishment cost index is the sum of the collaborative replenishment costs of all suppliers and the emergency purchase premium; the response time index is the sum of the time values ​​from the triggering of all orders to the arrival of materials; the system robustness index is the difference between 1 and the production time ratio, among which the production time ratio refers to the sum of the ratio of the production line downtime time to the planned production time.

[0254] D3: Construct state space equations based on strategy evaluation indicators ,in, Represents the state transition matrix, which describes how the state variables are transferred from one moment to the next in the absence of control input. Represents the control matrix, which describes the influence of the control input on the state variable. represents the state vector at time k, represents the control input vector at time k, including but not limited to replenishment cycle, replenishment quantity, and safety stock level, represents the process noise vector at time k, which is zero-mean Gaussian white noise in the present invention. represents the state vector at time k+1;

[0255] D4: Use the extended Kalman filter to perform state estimation on the state space equation to obtain parameter estimation values, and use the gradient descent method to optimize the parameter estimation values ​​to obtain optimized strategy parameters. Among them, the extended Kalman filter and the gradient descent method are the existing technical contents in this field, and are not the creative solutions of this application, so they will not be elaborated here.

[0256] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific embodiments. The above-mentioned specific embodiments are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also change, modify, replace and modify the above-mentioned embodiments without departing from the purpose and scope of protection of the present invention. These are all protected by the present invention.

[0257] If the technical solution disclosed herein involves personal information, the product using the technical solution disclosed herein has clearly informed the individual of the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the technical solution disclosed herein involves sensitive personal information, the product using the technical solution disclosed herein has obtained the individual's separate consent before processing the sensitive personal information and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, a clear and prominent sign is set to inform that the individual has entered the personal information collection scope and that personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that they agree to the collection of their personal information; or on the personal information processing device, when the personal information processing rules are notified by obvious signs / information, the individual's authorization is obtained through pop-up information or by asking the individual to upload their personal information. The personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. A smart factory raw material inventory warning and replenishment system, characterized by: include: Data acquisition module, digital twin modeling module, replenishment strategy module, risk prediction module, evaluation and optimization module; The replenishment strategy module inputs the full-chain production data acquired in real time by the data acquisition module and the three-level digital twin model of the digital twin modeling module into a pre-set replenishment strategy generator, and combines the reinforcement learning algorithm and the Monte Carlo simulation method to obtain the Pareto optimal strategy solution set; the full-chain production data includes the raw material consumption rate, the three-dimensional warehouse location distribution, the supplier equipment load rate, and the logistics blockchain timestamp; the three-level digital twin model is based on the Petri net-Agent joint modeling configuration; The risk prediction module constructs a domino effect propagation model based on the Pareto optimal strategy solution set and the three-level digital twin model, and formulates a dual-track replenishment mechanism. It combines the full-chain production data and the strategy library optimization model configured in the evaluation and optimization module to dynamically update the strategy parameters and transmit them back to the replenishment strategy module. The dual-track replenishment mechanism includes a main replenishment channel and an emergency channel; The steps for building the three-level digital twin model include: A1: Define the three elements of the system and establish the dynamic equation of inventory change; the three elements of the system include places, transitions, and directed arcs; A2: Based on the production hierarchy reflected by the full-chain production data, define intelligent agents. Simultaneously, analyze the interaction patterns between different layers in the full-chain production data and set interaction rules between intelligent agents. The intelligent agents are factories, workshops, and production lines. A3: Combine the discrete events described in the three elements of the system with the autonomous decision-making process of the intelligent agent to form a three-level digital twin model of factory-workshop-production line; The process of solving the Pareto optimal strategy solution set includes: B1: Extract supplier data from the entire production chain; the supplier data includes delivery time, product quality, and order completion rate; B2: Extracting features from the supplier data to obtain supplier characteristic data; the supplier characteristic data includes on-time delivery rate and product qualification rate; B3: Use statistical methods to construct a supplier performance capability assessment model and use supplier characteristic data to train the supplier performance capability assessment model to obtain a trained supplier performance capability assessment model; B4: Use the trained supplier performance assessment model, combined with full-chain production data, to calculate each supplier's credit score. This credit score is obtained by using a linear regression model to perform a weighted summation based on the regression coefficient and input features. B5: Sort suppliers according to their credit scores and generate a supplier priority list; B6: Define replenishment policy parameters, including order quantity and trigger threshold; B7: Develop replenishment strategy generation rules based on full-chain production data, the three-level digital twin model, suppliers’ credit scores, supplier priority lists, and replenishment strategy parameters; B8: Obtain full-chain production data, the three-level digital twin model, and the supplier's credit score in real time, and input them into the replenishment strategy generator. Based on the replenishment strategy generation rules, N candidate replenishment strategies are generated. The process of solving the Pareto optimal strategy solution set also includes: B9: Set up Monte Carlo simulation scenarios based on full-chain production data; B10: Perform Monte Carlo simulations on each candidate replenishment strategy and record the cost and risk indicators of each simulation; B11: Perform statistical analysis on the cost and risk indicators of each candidate replenishment strategy to obtain the average cost and risk value of each candidate replenishment strategy; B12: Based on the obtained average cost and risk indicators, determine whether each candidate replenishment strategy is a Pareto optimal strategy; B13: Filter out all strategies that are judged to be Pareto optimal to form a Pareto optimal strategy solution set; the Pareto optimal strategy solution set includes order quantity, trigger threshold, and supplier priority information; The risk prediction module constructs a domino effect propagation model based on the Pareto optimal strategy solution set and the three-level digital twin model, and formulates a dual-track replenishment mechanism, including: C1: Obtain the Pareto optimal strategy solution set and the three-level digital twin model; C2: The supply chain network is abstracted into a directed graph, where different links in the supply chain are defined as nodes in the graph, and the dependencies between nodes are represented by edges. Links include raw material suppliers, production lines, workshops, and factories, and dependencies include material supply relationships and production collaboration relationships. C3: Setting fault diffusion rules; the fault diffusion rules include forward diffusion rules and reverse forward diffusion rules; The forward diffusion rule states that when there is a delay in the supply of raw materials, it will cause the production line to stop, which will lead to material shortages in the workshop, and the shortage of materials in the workshop will cause a decrease in factory-level production capacity; The anti-forward diffusion rule is that when a factory-level emergency response is initiated, the workshop adjusts its priority, and after the workshop priority is adjusted, the production line production plan is rearranged; The risk prediction module constructs a domino effect propagation model based on the Pareto optimal strategy solution set and the three-level digital twin model, and formulates a dual-track replenishment mechanism, which also includes: C4: Based on a three-level digital twin model, for each replenishment strategy in the Pareto optimal strategy solution set, we simulate the cascading impact of fault diffusion rules at the production line, workshop, and factory levels. Through simulation, we analyze the extent and scope of the impact of faults starting from the raw material supply side and spreading to the production line, workshop, and factory under different replenishment strategies. C5: Determine the safety stock strategy based on the Pareto optimal solution set and replenish raw materials according to the replenishment cycle and replenishment quantity; Set a risk threshold. When the simulation results predict that the cascading risk reaches or exceeds the risk threshold, it triggers supplier collaborative replenishment or emergency procurement. C6: Generate a risk heat map. Based on the simulation results and cascading impact analysis, mark the potential risk nodes in the supply chain network and display the diffusion path of the failure on the heat map. The potential risk nodes include the supplier nodes with delayed raw materials and the lagging production line nodes.

2. The smart factory raw material inventory early warning and replenishment system according to claim 1, characterized in that: The specific steps of A3 include: A3.1: Identify system state variables and combine all system state variables to form a system state space. The system state variables are the state information represented by the library in the system's three elements. Each dimension in the system state space corresponds to a system state variable. A3.2: Generate a discrete event list based on the three elements of the system and the system state space; A3.3: Based on the agent, generate a list of decision events by combining the agent's decisions under different system states; A3.4: For each discrete event in the discrete event list, analyze the change in system state after the discrete event occurs, combined with its triggering conditions. Based on the change in system state, determine the corresponding agent decision for each discrete event from the decision event list, thereby obtaining a mapping relationship between discrete events and agent decisions.

3. The smart factory raw material inventory early warning and replenishment system according to claim 2, characterized in that: The specific steps of A3 also include: A3.5: Build a production-line digital twin model with the production line agent as the core, combining the production line discrete event list and the mapping relationship between discrete events and agent decisions. A3.6: Based on the production line-level digital twin model, consider the decision-making of the workshop agent and the collaborative relationship between different production lines within the workshop. Use mapping relationships and workshop-related discrete events to build a workshop-level digital twin model. A3.7: Integrate all workshop-level models, combine the macro-decision-making and mapping relationships of the factory agent with discrete events at the factory level, and build a three-level digital twin model of the factory-workshop-production line. A3.8: Compare the output results of the three-level digital twin model of factory-workshop-production line with the actual full-chain production data, and adjust the parameters of the three-level digital twin model based on the comparison results.

4. The smart factory raw material inventory early warning and replenishment system according to claim 3, characterized in that: The specific steps of B12 include: B12.1: Obtain the average cost and risk index of all candidate replenishment strategies and perform normalization to obtain the normalized average cost of the candidate replenishment strategies. and risk indicators ,and ; B12.2: Define a dominance relationship, where the condition for candidate replenishment strategy j to dominate candidate replenishment strategy i is: and ,in, and The normalized average cost of the i-th and j-th candidate replenishment strategies, respectively, and denote the normalized risk indicators of the i-th and j-th candidate replenishment strategies respectively; B12.3: Check for each candidate replenishment strategy whether there is at least one candidate replenishment strategy j that satisfies the dominance condition; If there is no candidate replenishment strategy j that satisfies the dominance condition, then the candidate replenishment strategy i is the Pareto optimal strategy.

5. The smart factory raw material inventory early warning and replenishment system according to claim 4, characterized in that: The strategy parameters are dynamically updated by combining the full-chain production data with the strategy library optimization model configured in the evaluation and optimization module, including: D1: Extract the Pareto optimal strategy solution set and dual-track replenishment execution records, and calculate the statistics of the dual-track replenishment execution records to obtain a structured data set; the dual-track replenishment execution records include supplier collaborative replenishment response time and emergency procurement cost data; the statistics include mean and variance; D2: Based on the structured data set, a strategy evaluation indicator system is constructed, including replenishment cost indicator, response time indicator, and system robustness indicator; D3: Construct state space equations based on strategy evaluation indicators; D4: Use the extended Kalman filter to perform state estimation on the state space equation to obtain parameter estimates, and use the gradient descent method to optimize the parameter estimates to obtain the optimized strategy parameters.

Images