Product supply method and system and electronic equipment

Abstract

The invention provides a product supply method and system and electronic equipment, and the method comprises the steps: collecting the supply information of multiple links in real time through Internet of Things equipment and an enterprise system, and the supply information comprises at least one of inventory information, order information, logistics information, a supply chain topological structure and supply environment parameters; performing anomaly detection and demand prediction on the supply information by adopting a streaming processing engine to obtain a response priority and a prediction demand; dynamically generating at least one initial supply strategy according to an improved multi-target genetic algorithm, the response priority and the prediction demand, and determining a target supply strategy based on simulation execution results corresponding to the initial supply strategies; and executing product supply based on the target supply strategy, and taking feedback data generated in the execution process as new supply information. Multi-chain real-time collaboration and dynamic evolution are supported, and the supply chain response speed and the resource configuration efficiency are improved through a closed-loop feedback mechanism.

Description

Technical Field

[0001] This invention relates to the field of supply chain management technology, and in particular to a product supply method, system and electronic device. Background Technology

[0002] Supply chain management refers to optimizing the operation of the supply chain to minimize costs, encompassing all processes from procurement to satisfying the end customer.

[0003] Traditional supply chain management suffers from rigid structures, delayed information feedback, and insufficient algorithm adaptability, severely limiting enterprises' ability to cope with complex market environments. These limitations highlight the urgent need to develop real-time, adaptive supply chain management technologies. Summary of the Invention

[0004] To address the above problems, the present invention provides a product supply method, system, and electronic device.

[0005] This invention provides a product supply method, comprising: Through IoT devices and enterprise systems, supply information across multiple links is collected in real time. This supply information includes at least one of the following: inventory information, order information, logistics information, supply chain topology, and supply environment parameters. A streaming processing engine is used to perform anomaly detection and demand forecasting on the supply information to obtain response priorities and predicted demand. Based on the improved multi-objective genetic algorithm, the response priority, and the predicted demand, at least one initial supply strategy is dynamically generated, and a target supply strategy is determined based on the simulation execution results corresponding to each initial supply strategy. Based on the target supply strategy, product supply is executed, and the feedback data generated during the execution process is used as new supply information.

[0006] According to a product supply method provided by the present invention, the method employs a streaming processing engine to perform anomaly detection and demand forecasting on the supply information to obtain response priorities and predicted demand, including: The streaming processing engine is used to input the supply information into the anomaly detection model for anomaly detection, obtain anomaly event information, and determine the response priority corresponding to the anomaly event information based on the anomaly event information. The anomaly detection model has built-in Drools rules, isolated forest algorithm and BERT model. The supply information is input into the demand forecasting model using the streaming processing engine to predict the demand and obtain the predicted demand. The demand forecasting model incorporates a long short-term memory model, a time series model, and various optimization algorithms.

[0007] According to a product supply method provided by the present invention, the anomaly detection includes threshold detection, latent anomaly detection, and text anomaly detection; The process of using the streaming processing engine to input the supply information into an anomaly detection model for anomaly detection, obtaining anomaly event information, and determining the response priority corresponding to the anomaly event information based on the anomaly event information includes: Using the streaming processing engine, the supply information is input into the anomaly detection model. The Drools rule is used to perform threshold detection on the supply information, the isolated forest algorithm is used to perform latent anomaly detection on the supply information, and the BERT model is used to perform text anomaly detection on the supply information to obtain the anomaly event information. The response priority corresponding to the abnormal event information is obtained by weighting the event urgency, impact scope and processing cost in the abnormal event information.

[0008] According to a product supply method provided by the present invention, the step of dynamically generating at least one initial supply strategy based on an improved multi-objective genetic algorithm, the response priority, and the predicted demand includes: Based on the response priority and the predicted demand, an objective function and constraints are constructed. The objective function includes at least one of a function that minimizes total cost, a function that maximizes on-time delivery rate, and a robustness evaluation function. The constraints include at least one of a supply capacity constraint, a minimum order quantity constraint, and an order constraint. The improved multi-objective genetic algorithm is used to solve the objective function based on the constraints to obtain at least one initial supply strategy.

[0009] According to a product supply method provided by the present invention, the processing procedure of the improved multi-objective genetic algorithm includes the following steps: (1) Determine the population initialization strategy: Load the population initialization strategy of the historical best solution from the database, or obtain the population initialization strategy using heuristic generation rules; (2) Adaptively adjust the crossover probability and the mutation probability, wherein the crossover probability and the mutation probability are adjusted by the following formula: in, For crossover probability, This represents the mutation probability of supplier i. This represents the latency rate of supplier i. This represents the maximum latency rate among all suppliers; (3) Multi-objective assessment and elite retention, wherein the multi-objective assessment is a robustness assessment, and the robustness score corresponding to the robustness assessment is determined by Monte Carlo perturbation test, and the calculation formula of the robustness score is as follows: in, For robustness scoring; (4) Hybrid local search, wherein the hybrid local search employs either simulated annealing or tabu search strategies; (5) Dynamic parameter tuning mechanism: A reinforcement learning controller is adopted to switch local search strategies and dynamically modify the weights of the objective function. The state space includes at least one of population diversity, convergence speed, external event type and resource constraints, and the action space includes crossover probability and / or mutation probability. The state is collected at intervals and actions are selected. The actions are executed based on the state to obtain the execution result, the reward is calculated and the parameters of the reinforcement learning controller are updated until convergence.

[0010] According to a product supply method provided by the present invention, before determining the target supply strategy based on the simulation execution results corresponding to each of the initial supply strategies, the method further includes: For each initial supply strategy, the initial supply strategy and the supply information are input into a digital twin simulator for discrete event simulation, proxy modeling, simulation acceleration and non-critical path simplification to obtain the simulation execution result corresponding to the initial supply strategy.

[0011] According to a product supply method provided by the present invention, the process of executing product supply based on the target supply strategy includes: Real-time monitoring and key performance indicator calculation; The target supply strategy is dynamically adjusted based on the deviations corresponding to the key performance indicators. The product supply is delivered in stages when some nodes fail.

[0012] According to a product supply method provided by the present invention, before employing a streaming processing engine to perform anomaly detection and demand forecasting on the supply information to obtain response priorities and predicted demand, the method further includes: The supply information is subjected to format and logic validation. If the verification passes, an anomaly filtering algorithm is used to remove invalid data from the supply information, and a Kalman filtering algorithm is used to complete the missing spatiotemporal data in the supply information.

[0013] The present invention also provides a product supply system, comprising: The data acquisition layer is used to collect multi-link supply information in real time through IoT devices and enterprise systems. The supply information includes at least one of inventory information, order information, logistics information, supply chain topology, and supply environment parameters. The feedback analysis layer is used to perform anomaly detection and demand forecasting on the supply information using a streaming processing engine, thereby obtaining response priorities and predicted demand. An evolutionary optimization layer is used to dynamically generate at least one initial supply strategy based on the improved multi-objective genetic algorithm, the response priority, and the predicted demand, and to determine a target supply strategy based on the simulation execution results corresponding to each initial supply strategy. The strategy execution layer is used to execute product supply based on the target supply strategy and to use the feedback data generated during the execution process as new supply information.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the product supply method as described above.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the product supply method as described above.

[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the product supply method as described above.

[0017] This invention provides a product supply method, system, and electronic device. The method includes: real-time collection of multi-chain supply information via IoT devices and enterprise systems, the supply information including at least one of inventory information, order information, logistics information, supply chain topology, and supply environment parameters; using a streaming processing engine to perform anomaly detection and demand forecasting on the supply information to obtain response priorities and predicted demand; dynamically generating at least one initial supply strategy based on an improved multi-objective genetic algorithm, the response priorities, and the predicted demand, and determining a target supply strategy based on the simulation execution results corresponding to each initial supply strategy; executing product supply based on the target supply strategy, and using the feedback data generated during the execution process as new supply information. This invention supports real-time multi-chain collaboration and dynamic evolution, improving supply chain response speed and resource allocation efficiency through a closed-loop feedback mechanism.

[0018] This invention achieves standardized processing and unified management of end-to-end supply chain data through the entire data fusion process, solving pain points such as data fragmentation, inconsistent formats, and varying quality in traditional supply chain systems. By organically integrating real-time data streams and batch-processed data, it dynamically reflects the supply chain's operational status, laying a solid foundation for subsequent intelligent analysis and decision optimization. This not only enhances supply chain transparency but also creates the necessary conditions for enterprises to achieve data-driven, refined management. By monitoring the effectiveness of event handling and cost savings, it dynamically adjusts evaluation strategies, forming a self-improving decision-making mechanism that effectively balances response speed and processing costs, achieving an intelligent upgrade in supply chain anomaly management. Through dynamic genetic algorithms, it achieves minute-level response times, supports multi-chain collaborative optimization, and improves robustness. Combined with reinforcement learning self-adjustment, it significantly improves the efficiency and adaptability of supply chain decision-making. It enables precise implementation of intelligent strategies, supports multi-system collaborative execution, possesses dynamic fault tolerance and real-time feedback capabilities, and significantly improves execution efficiency, ensuring the efficient and reliable implementation of supply chain strategies. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is one of the flowcharts illustrating the product supply method provided by the present invention.

[0021] Figure 2 This is the second flowchart illustrating the product supply method provided by the present invention.

[0022] Figure 3 This is the third flowchart illustrating the product supply method provided by the present invention.

[0023] Figure 4 This is a schematic diagram of the product supply system provided by the present invention.

[0024] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0026] The following is combined with Figures 1-5 The present invention describes a product supply method, system, and electronic device.

[0027] First, a brief description of the relevant content involved in this invention will be given.

[0028] Existing supply chains often employ a single-chain linear structure, such as a fixed "supplier → production center → distributor" model, leading to difficulties in cross-chain resource coordination. For example, when a supplier suddenly experiences a production shortage, the lack of cross-supplier data sharing mechanisms prevents companies from quickly utilizing the idle capacity of other suppliers, resulting in missed order delivery windows. More seriously, regional warehouse inventory data is often not linked to logistics routes in real time, requiring manual coordination for cross-warehouse transfers, with an average delay of over 48 hours. An automotive company once experienced a situation where urgently allocated parts were stuck in a transit warehouse for 12 hours due to the inventory system's failure to synchronize logistics status, causing significant production line losses. Furthermore, data formats in systems such as Enterprise Resource Planning (ERP) and logistics platforms are fragmented, including Extensible Markup Language (XML), Electronic Data Interchange (EDI), and JavaScript Object Notation (JSON), requiring substantial manpower for cleaning and integration, resulting in long daily processing times and a high risk of errors.

[0029] The significant lag in information feedback further exacerbates supply chain risks. Traditional batch-processing data synchronization (such as daily scheduled updates) leads to slow decision-making responses, triggering a chain reaction of economic losses. For example, in a port strike, a six-hour delay in updating logistics status prevented the company from adjusting its plans in time, resulting in the erroneous release of 5,000 undeliverable orders and ultimately huge losses due to customer claims. Statistics show that the median response time for anomalies in traditional supply chains is as high as 4.2 hours, with 60% of cases further amplifying losses due to delays in manual approval processes. This inefficient response mechanism is particularly fatal in dynamic market environments, and companies urgently need a new management method that enables real-time awareness and rapid decision-making.

[0030] Furthermore, traditional static optimization algorithms (such as linear programming and basic genetic algorithms) heavily rely on historical data for training, making them ill-suited to dynamic scenarios involving demand fluctuations and unforeseen events. For example, when a social media trend causes a 300% surge in orders for a certain headphone accessory, static algorithms still allocate inventory based on historical averages, causing the company to miss order opportunities. Similarly, when logistics routes are suddenly restricted (such as temporary traffic restrictions affecting 10% of capacity), static route planning requires several hours of recalculation, during which the on-time delivery rate plummets to 62%. Experimental data shows that static algorithms have an inventory prediction error rate as high as 37% in dynamic environments, far from meeting the agility requirements of modern supply chains. Although existing technologies (such as blockchain traceability patents) have made breakthroughs in data trustworthiness, resource scheduling still relies on manual decision-making; while multi-objective optimization algorithms proposed in academic research, such as NSGA-II, while promising, are limited by computational efficiency (2.3 minutes per iteration) and the complexity of multi-chain coupling constraints.

[0031] To address the above issues, this invention provides a product supply method that supports real-time collaboration and dynamic evolution across multiple supply chains, improving supply chain response speed and resource allocation efficiency through a closed-loop feedback mechanism.

[0032] Figure 1 This is one of the flowcharts illustrating the product supply method provided by the present invention, such as... Figure 1 As shown, the method includes steps 101 to 104.

[0033] Step 101: Collect supply information across multiple links in real time through IoT devices and enterprise systems. The supply information includes at least one of the following: inventory information, order information, logistics information, supply chain topology, and supply environment parameters.

[0034] The product supply method provided by this invention can be implemented by a product supply system.

[0035] See Figure 2 , Figure 2This is the second flowchart illustrating the product supply method provided by the present invention: In the product supply system, the data acquisition layer collects data through IoT devices and enterprise systems to obtain supply information. The IoT devices can include vehicle global positioning systems (GPS) and warehouse radio frequency identification (RFID) devices, etc.; the enterprise systems can include ERP, manufacturing execution systems (MES), and transportation management systems (TMS), etc.

[0036] Specifically, the data acquisition layer of the product supply system interfaces with various heterogeneous data sources, including IoT devices using Message Queuing Telemetry Transport (MQTT) or CoAP (Constrained Application Protocol) to collect warehouse environmental data and vehicle location information in real time. It also obtains order and delivery plan data from enterprise systems through Representational State Transfer (API) or Electronic Data Interchange (EDI) interfaces. Furthermore, it can utilize WebSocket to receive transportation status updates from logistics platforms and collect manually entered information on unexpected events via web forms. These data sources vary in their collection frequency, ranging from millisecond-level real-time data to on-demand triggered manual input, all of which the system effectively supports. The data collected from these data sources constitutes the supply information.

[0037] Furthermore, to ensure reliable data transmission, the product supply system's data acquisition layer employs a distributed message queue architecture. (See also...) Figure 2The Apache Kafka cluster, through topic partitioning and replication mechanisms, combined with the Apache Flink stream processing engine, constructs a highly available data processing pipeline. Its multi-level buffering scheme, including in-memory queues, local WAL (Write-Ahead Logging) logs, and cloud cold storage, ensures both real-time processing performance and data persistence. This architecture effectively handles anomalies such as network jitter and node failures, providing stable and reliable data support for upper-layer intelligent decision-making.

[0038] Step 102: Using a streaming processing engine, perform anomaly detection and demand forecasting on the supply information to obtain response priority and predicted demand.

[0039] Specifically, the streaming engine can be the Apache Flink streaming engine.

[0040] In practical applications, see Figure 2 The data acquisition layer transmits supply messages to the streaming processing engine of the feedback analysis layer in the product supply system via the Apache Kafka message queue. Specifically, the Apache Flink stream processing engine receives the Kafka data stream. The Apache Kafka message queue includes a distributed message buffer and high availability design, as well as 3 replicas and ISR (Information Search Ranking) fault tolerance.

[0041] Specifically, the feedback analysis layer adopts a layered architecture design, using an Apache Flink cluster as the stream processing engine layer to provide event temporal semantics and state management capabilities; the model service layer deploys TensorFlow Serving / PyTorch Serve to carry anomaly detection models and demand prediction models; the rule engine layer uses Drools to manage multi-level early warning strategies; and the Redis caching layer is responsible for the temporary storage of frequently accessed data, forming an efficient and collaborative technology stack.

[0042] Furthermore, the Apache Flink stream processing engine inputs the supply data into the anomaly detection model and the demand forecasting model for processing, thereby obtaining response priorities and predicted demand.

[0043] Step 103: Based on the improved multi-objective genetic algorithm, the response priority, and the predicted demand, at least one initial supply strategy is dynamically generated, and a target supply strategy is determined based on the simulation execution results corresponding to each initial supply strategy.

[0044] Specifically, the initial supply strategy includes supply path planning strategy and / or inventory allocation strategy.

[0045] See Figure 2 The feedback analysis layer transmits response priorities and predicted demand to the evolutionary optimization layer in the product supply system. Based on the response priorities and predicted demand, the evolutionary optimization layer uses an improved MultiObjective Genetic Algorithm (MOGA) to dynamically generate multiple initial supply strategies. Then, based on the simulation results corresponding to each initial supply strategy, it selects the optimal initial supply strategy as the target supply strategy. The improved MOGA includes adaptive cross-simulated annealing and a parameter adaptation mechanism based on reinforcement learning.

[0046] Step 104: Based on the target supply strategy, execute product supply and use the feedback data generated during the execution process as new supply information.

[0047] See Figure 2 The evolution and optimization layer distributes the target supply strategy to the strategy execution layer of the product supply system. The execution units, such as the Warehouse Management System (WMS), Transportation Management System (TMS), and supplier collaboration platform, then implement the target supply strategy, supplying products according to it. Furthermore, the feedback data generated during execution is reintegrated into the data acquisition layer, forming a closed-loop mechanism for continuous optimization, ensuring the system can dynamically adapt to changes in the supply chain environment.

[0048] The product supply method provided by this invention achieves standardized processing and unified management of data across the entire supply chain through the data fusion process, solving pain points such as data fragmentation, inconsistent formats, and varying quality in traditional supply chain systems. By organically integrating real-time data streams and batch-processed data, it dynamically reflects the operational status of the supply chain, laying a solid foundation for subsequent intelligent analysis and decision optimization. This not only improves supply chain transparency but also creates the necessary conditions for enterprises to achieve data-driven refined management. By monitoring the effectiveness of event handling and cost savings, it dynamically adjusts evaluation strategies, forming a self-improving decision-making mechanism that effectively balances response speed and processing costs, achieving an intelligent upgrade in supply chain anomaly management. Through dynamic genetic algorithms, it achieves minute-level response, supports multi-chain collaborative optimization, and improves robustness. Combined with reinforcement learning self-adjustment, it significantly improves the efficiency and adaptability of supply chain decision-making. It enables precise implementation of intelligent strategies, supports multi-system collaborative execution, possesses dynamic fault tolerance and real-time feedback capabilities, and significantly improves execution efficiency, ensuring the efficient and reliable implementation of supply chain strategies.

[0049] Optionally, the use of a streaming processing engine to perform anomaly detection and demand forecasting on the supply information to obtain response priorities and predicted demand includes: The streaming processing engine is used to input the supply information into the anomaly detection model for anomaly detection, obtain anomaly event information, and determine the response priority corresponding to the anomaly event information based on the anomaly event information. The anomaly detection model has built-in Drools rules, isolated forest algorithm and BERT model. The supply information is input into the demand forecasting model using the streaming processing engine to predict the demand and obtain the predicted demand. The demand forecasting model incorporates a long short-term memory model, a time series model, and various optimization algorithms.

[0050] Specifically, see Figure 2 In terms of anomaly detection, the feedback layer system establishes a multimodal intelligent recognition framework, namely the anomaly detection model. This model incorporates Drools rules, the Isolation Forest algorithm, and the BERT model, outputting information about anomalous events, such as stockouts and delays. The demand forecasting model employs a hybrid model architecture, intelligently selecting the optimal algorithm from multiple optimization algorithms based on the characteristics of the smallest stock keeping unit (SKU). Furthermore, the demand forecasting model incorporates a Long Short-Term Memory (LSTM) model and a time series model (Prophet). The LSTM model maintains data timeliness through a rolling window training mechanism, while the Prophet model automatically identifies trend abrupt changes. The forecast results dynamically incorporate external factors such as weather and establish a feedback correction mechanism. When actual demand deviates from the predicted value by more than 20% three consecutive times, model retraining is automatically triggered to ensure continuous optimization of forecast accuracy.

[0051] Optionally, the anomaly detection includes threshold detection, latent anomaly detection, and text anomaly detection; The process of using the streaming processing engine to input the supply information into an anomaly detection model for anomaly detection, obtaining anomaly event information, and determining the response priority corresponding to the anomaly event information based on the anomaly event information includes: Using the streaming processing engine, the supply information is input into the anomaly detection model. The Drools rule is used to perform threshold detection on the supply information, the isolated forest algorithm is used to perform latent anomaly detection on the supply information, and the BERT model is used to perform text anomaly detection on the supply information to obtain the anomaly event information. The response priority corresponding to the abnormal event information is obtained by weighting the event urgency, impact scope and processing cost in the abnormal event information.

[0052] In practical applications, for scenarios with explicit business rules (such as inventory falling below a safety threshold), the Drools rule engine is used for threshold detection; for latent anomalies (such as sudden surges in demand), the Isolation Forest algorithm is used for unsupervised learning and identification, i.e., latent anomaly detection; and for text-based anomalies (such as keywords like "strike" in logistics notes), the BERT model is used for semantic understanding, i.e., text anomaly detection, thereby obtaining information about the anomaly event. This combined detection scheme ensures comprehensive coverage of various anomalies.

[0053] Response priority is evaluated using a combination of quantitative calculation and dynamic adjustment. Response priority is derived by weighting three dimensions: event urgency, scope of impact, and processing cost, with the weighting coefficients continuously optimized through reinforcement learning. The formula for calculating response priority is as follows: in, For response priority, , and These are the weights corresponding to the urgency of the event, the scope of its impact, and the cost of handling it.

[0054] Optionally, dynamically generating at least one initial supply strategy based on the improved multi-objective genetic algorithm, the response priority, and the predicted demand includes: Based on the response priority and the predicted demand, an objective function and constraints are constructed. The objective function includes at least one of a function that minimizes total cost, a function that maximizes on-time delivery rate, and a robustness evaluation function. The constraints include at least one of a supply capacity constraint, a minimum order quantity constraint, and an order constraint. The improved multi-objective genetic algorithm is used to solve the objective function based on the constraints to obtain at least one initial supply strategy.

[0055] First, a mathematical model for multi-chain collaborative optimization is established.

[0056] Given decision variables ,in Indicates supplier For orders The number of parts allocated Characterizes all suppliers, This represents all orders. The possible objective functions include at least one of the following: minimizing the total cost function, maximizing the on-time delivery rate function, and robustness evaluation function, in that order: in, For total cost, Indicates supplier Fulfilling orders The unit transportation cost; Indicates supplier Unit inventory cost; Indicates order The corresponding delay penalty coefficient is related to the order priority; Indicates the actual delivery date. Indicates the delivery date of the request; A1, A2, and A3 are all weights, with values ​​of 1 or 0; The on-time delivery rate is given by m, where m is the total number of orders. For the delay cost function, for The change in; For robustness scoring, This is the robustness evaluation function.

[0057] Supply capacity constraints are: in, Let i be the overproduction coefficient of supplier i. The ideal production capacity for supplier i.

[0058] The minimum order quantity constraint is: in, It is the minimum order quantity for supplier i.

[0059] Order constraints, i.e., order fulfillment in, This is the requirement for order j.

[0060] Furthermore, the improved algorithm is used to solve the optimization model. The key improvement of the algorithm is to introduce adaptive crossover probability and a local search strategy based on simulated annealing to accelerate convergence and obtain at least one initial supply strategy.

[0061] Optionally, the improved multi-objective genetic algorithm includes the following steps: (1) Determine the population initialization strategy: Load the population initialization strategy of the historical best solution from the database, or obtain the population initialization strategy using heuristic generation rules; (2) Adaptively adjust the crossover probability and the mutation probability, wherein the crossover probability and the mutation probability are adjusted by the following formula: in, For crossover probability, This represents the mutation probability of supplier i. This represents the latency rate of supplier i. This represents the maximum latency rate among all suppliers; (3) Multi-objective assessment and elite retention, wherein the multi-objective assessment is a robustness assessment, and the robustness score corresponding to the robustness assessment is determined by Monte Carlo perturbation test, and the calculation formula of the robustness score is as follows: in, For robustness scoring; (4) Hybrid local search, wherein the hybrid local search employs either simulated annealing or tabu search strategies; (5) Dynamic parameter tuning mechanism: A reinforcement learning controller is adopted to switch local search strategies and dynamically modify the weights of the objective function. The state space includes at least one of population diversity, convergence speed, external event type and resource constraints, and the action space includes crossover probability and / or mutation probability. The state is collected at intervals and actions are selected. The actions are executed based on the state to obtain the execution result, the reward is calculated and the parameters of the reinforcement learning controller are updated until convergence.

[0062] Specifically, cross-probability refers to the probability of cross-supplying products, while variation probability refers to the probability of changing suppliers.

[0063] In practical applications, the implementation process of the improved multi-objective genetic algorithm (MOGA) is as follows: Determine the population initialization strategy. Two strategies are available: First, load gene fragments from historical best solutions from the database, such as supplier selection patterns and logistics route preferences. Second, utilize heuristic generation rules to obtain the initialization strategy. For example, initially select suppliers with an on-time rate greater than 95% in the last three collaborations as primary suppliers to complete orders; if these suppliers do not meet the constraints, further degrade the selection process.

[0064] Adaptive crossover and mutation. Crossover probability, for emergency events such as P0-level shortages, can increase exploratory power, i.e. (Increase). For targeted mutations, such as those from high-latency suppliers, apply a higher mutation probability.

[0065] Multi-objective evaluation and elite retention. Robustness assessment was performed using Monte Carlo perturbation tests. [The text abruptly ends here, likely due to an incomplete sentence or missing information.] Inject random perturbations (requirements) Logistics time The proportion of feasible solutions is calculated. For non-dominated sorting (NSGA-III improvement), a reference point adaptation mechanism can be introduced to dynamically adjust the distribution of reference points to improve the balance of the solution set in the high-dimensional objective space.

[0066] Hybrid local search strategy. This strategy includes two specific available strategies. The first is simulated annealing (SA), which performs a domain search on elite individuals, as follows: in, The difference. T represents the current simulated annealing temperature. e is the natural logarithm.

[0067] For example, in a cost minimization optimization problem, if the objective value of the previous iteration was 100 and the objective value of the current iteration is 90, then... Since this iteration reduces the objective function value, we accept this iteration with a probability of 1. initial temperature cooling rate The process is iterated 50 times. The tabu search strategy is as follows: a tabu list records the characteristics of the 10 most recent solutions (such as the selected supplier combinations), and repeated visits are prohibited. Amnesty rule: if a tabu solution is better than the historical best, it is accepted as an exception.

[0068] Dynamic parameter tuning mechanism. 1) This process uses a reinforcement learning controller (DQN), where the state space includes population diversity, convergence speed, external event types, and resource constraints. The action space includes adjusting the crossover probability. Probability of mutation ; Switch local search strategy (SA / TS); dynamically modify the objective function weights. 2) The online learning process is as follows: collect the state every 10 generations. Select Action (such as increasing) Observe the next generation's performance after the action is performed, and calculate the reward. Update the DQN network parameters, prioritizing replay to accelerate convergence. The reward function is as follows: in, As a reward.

[0069] Optionally, before determining the target supply strategy based on the simulation execution results corresponding to each of the initial supply strategies, the method further includes: For each initial supply strategy, the initial supply strategy and the supply information are input into a digital twin simulator for discrete event simulation, proxy modeling, simulation acceleration and non-critical path simplification to obtain the simulation execution result corresponding to the initial supply strategy.

[0070] See Figure 2 The evolution optimization layer also includes a digital twin simulator, which is mainly used to accept optimization strategies and simulate the execution results.

[0071] Specifically, the digital twin simulator includes the following: 1) Simulation model construction: Input data includes supply chain topology (suppliers, warehouses, logistics routes), real-time status (inventory, orders, vehicle locations), and external environmental parameters (weather, policies, market fluctuations). 2) Model building: Discrete event simulation: Using the SimPy framework to simulate the temporal logic of events such as order processing and transportation. 3) Agent modeling: Treating suppliers, warehouses, and logistics companies as independent agents and defining interaction rules. 4) Simulation acceleration: Parallelized simulation, allocating different warehouses / logistics routes to multi-core CPU or GPU threads. 5) Reduced-order model (ROM): Using simplified models (such as linear approximations) for non-critical paths to accelerate computation.

[0072] Optionally, the process of executing product supply based on the target supply strategy includes: Real-time monitoring and key performance indicator calculation; The target supply strategy is dynamically adjusted based on the deviations corresponding to the key performance indicators. The product supply is delivered in stages when some nodes fail.

[0073] In practical applications, strategy execution also requires continuous monitoring of Key Performance Indicators (KPIs) deviations: Real-time monitoring and KPI calculation: Collect execution node data and compare the deviations between expected and actual results; Dynamic strategy corrector: Trigger local optimization or global rescheduling based on deviations; Fault-tolerant executor: Ensure that critical tasks are degraded when some nodes fail.

[0074] Specifically, real-time monitoring and dynamic correction include a KPI monitoring system and dynamic correction strategies. The KPI monitoring system is shown in Table 1.

[0075] Table 1

[0076] Then, the dynamic correction strategy includes local correction, such as allocating resources to nearby suppliers or switching suppliers to address single-node anomalies (e.g., insufficient inventory in a warehouse). Global rescheduling: When local correction cannot meet the requirements, the evolutionary optimization layer is triggered to regenerate the strategy.

[0077] Finally, the fault-tolerant execution mechanism, including the hierarchical execution strategy, is shown in Table 2.

[0078] Table 2

[0079] Optionally, before employing a streaming processing engine to perform anomaly detection and demand forecasting on the supply information to obtain response priorities and predicted demand, the method further includes: The supply information is subjected to format and logic validation. If the verification passes, an anomaly filtering algorithm is used to remove invalid data from the supply information, and a Kalman filtering algorithm is used to complete the missing spatiotemporal data in the supply information.

[0080] See Figure 2 The data acquisition layer is also equipped with a multi-distance adapter to perform tasks such as cleaning and format standardization of purchased multi-distance data.

[0081] Specifically, after obtaining supply information, data processing is carried out.

[0082] During the data preprocessing stage, the product supply system established a comprehensive data quality assurance mechanism. Format validation ensures the integrity of key fields, logical validation maintains business rule constraints, and anomaly filtering algorithms remove invalid data records. Simultaneously, algorithms such as Kalman filtering are used to intelligently complete missing spatiotemporal data, and a unified supplier information system is built through semantic expansion, linking metadata such as geographical location and credit rating, significantly improving data usability and value density.

[0083] The product supply method provided by this invention achieves standardized processing and unified management of data across the entire supply chain at the data acquisition layer, solving pain points such as data fragmentation, inconsistent formats, and varying quality in traditional supply chain systems. Through the organic integration of real-time data streams and batch-processed data, it dynamically reflects the operational status of the supply chain, laying a solid foundation for subsequent intelligent analysis and decision optimization. This not only improves supply chain transparency but also creates the necessary conditions for enterprises to achieve data-driven refined management. At the feedback analysis layer, by monitoring the effectiveness of event handling and cost savings, the evaluation strategy is dynamically adjusted to form a self-improving decision-making mechanism, effectively balancing response speed and processing costs, and achieving an intelligent upgrade of supply chain anomaly management. At the evolutionary optimization layer, a dynamic genetic algorithm achieves minute-level response, supports multi-chain collaborative optimization, and improves robustness. Combined with reinforcement learning self-adjustment, it significantly improves the efficiency and adaptability of supply chain decision-making. At the strategy execution layer, it enables precise implementation of intelligent strategies, supports multi-system collaborative execution, possesses dynamic fault tolerance and real-time feedback capabilities, significantly improves execution efficiency, and ensures the efficient and reliable implementation of supply chain strategies.

[0084] The following is combined with Figure 3 The product supply method provided by the present invention will be further described. Figure 3 This is the third flowchart illustrating the product supply method provided by the present invention.

[0085] S1: Multi-chain data fusion to build a unified spatiotemporal database.

[0086] By collecting real-time inventory status, order information, and logistics data across multiple links through IoT devices and enterprise systems, comprehensive data support is provided for upper-level decision-making. The collected data then enters the feedback analysis layer, which uses streaming computing engines such as Apache Flink to achieve real-time anomaly detection and accurate demand forecasting, providing crucial information for subsequent optimization decisions. The analyzed data is then passed to the evolutionary optimization layer.

[0087] S2: Real-time feedback triggers the start of a dynamic optimization process.

[0088] A complete real-time feedback triggering mechanism is constructed to achieve autonomous optimization of the supply chain through an intelligent anomaly detection and dynamic response system. A layered architecture is adopted, with an Apache Flink cluster as the stream processing engine layer, providing event temporal semantics and state management capabilities; the model service layer deploys TensorFlow Serving / PyTorch Serve to carry prediction and anomaly detection models; the rule engine layer uses Drools to manage multi-level early warning strategies; and the Redis caching layer is responsible for the temporary storage of frequently accessed data, forming a highly efficient and collaborative technology stack.

[0089] S3, dynamic evolution optimization, digital twin simulation.

[0090] Input real-time demand, inventory distribution, supplier capacity, and logistics cost matrix data. Then set the optimization objective: minimize total cost (transportation + inventory + delay penalty) and maximize on-time delivery rate. Finally, use an improved algorithm to solve the optimization model. The algorithm improvement focuses on introducing adaptive crossover probabilities and a simulated annealing-based local search strategy to accelerate convergence.

[0091] S4, Strategy Execution and Verification, Continuously Detects KPI Deviations.

[0092] The overall verification process includes the following steps to establish the verification process architecture. This architecture includes: (1) Digital twin simulation engine: Constructing a supply chain environment based on real data to simulate the effect of strategy execution. (2) Real-time monitoring and KPI calculation: Collecting data from execution nodes and comparing the deviation between expected and actual results. (3) Dynamic strategy corrector: Triggering local optimization or global rescheduling based on the deviation. (4) Fault-tolerant executor: Ensuring that critical tasks are degraded when some nodes fail.

[0093] The product supply method provided by this invention supports real-time collaboration and dynamic evolution across multiple chains, and improves supply chain response speed and resource allocation efficiency through a closed-loop feedback mechanism.

[0094] The product supply method provided by the present invention will be described below with reference to specific embodiments.

[0095] The background is that due to a sudden change in customs regulations, 50,000 headphone accessories originally scheduled for sea freight by an electronics brand are stuck in port, affecting order deliveries in three major regions worldwide. The company now has the following urgent handling requirements: Time requirement: Orders must be reallocated within 6 hours to avoid production line downtime.

[0096] Cost limit: Total logistics costs shall not increase by more than 20%.

[0097] Priority rule: VIP customer orders (accounting for 15%) must be delivered 100% on time.

[0098] To achieve the above objectives, the product supply method provided by the invention is adopted for the following processing: The first step is for the data acquisition layer to respond, collecting the following real-time data input as shown in Table 3.

[0099] Table 3

[0100] The second step involves real-time decision-making at the feedback analysis layer. This transforms raw data into actionable supply chain events and provides input parameters for evolutionary optimization. Multi-level event detection is implemented. The first layer is a rule engine, responding in milliseconds. Detection metrics include inventory thresholds, logistics delays, and sudden demand surges, then outputting structured events (e.g., P0-P3 priorities). The second layer is an AI model, responding in seconds. The Isolation Forest algorithm is used to detect negative anomalies, such as slowly deteriorating supplier delivery times. LSTM is used to predict demand fluctuations in the next 3 hours in real time, triggering pre-allocation.

[0101] The third step is intelligent decision-making at the evolutionary optimization layer. A multi-objective optimization model is established in this part, where the decision variables are... supplier For the region The proportion of orders shipped via transportation.

[0102] The objective function is as follows: The constraints are as follows: in, Supplier production capacity Supplier overproduction coefficient, This indicates that suppliers are allowed. Overproduction.

[0103] The algorithm is executed as described in the invention, and the key algorithms are as follows: Initial population generation: Injecting solutions based on historically similar events; Hybrid local search: Taboo search is used for the VIP order protection strategy to avoid duplicate and invalid paths; Real-time parameter adjustment: High priority was detected, and the on-time rate target weight was increased from 0.4 to 0.7.

[0104] The fourth step is strategy verification and execution. Using digital twin simulation, the candidate solutions generated in the third step are input, and the execution results are shown in Table 4.

[0105] Table 4

[0106] Simulation Results: Option B was selected due to its balance between cost and timeliness, but the simulation showed that there was still an 8% risk of order delays in some areas. Dynamic Correction: First, local adjustments were made: an additional 10% air freight quota was added to areas with order delay risk (cost increase to 14%, on-time performance improved to 95%). Second, fault-tolerant execution was implemented. Main Execution Path: Instructions were issued to the TMS (Transportation Management System) and the supplier platform. Backup Path: When a supplier interface timed out (>5 seconds), the system automatically switched to a backup API or email ticket.

[0107] Finally, the following implementation results were achieved.

[0108] The achievement of key indicators is shown in Table 5.

[0109] Table 5

[0110] The product supply method provided by this invention shows an improvement in performance compared to traditional manual scheduling, as shown in Table 6.

[0111] Table 6

[0112] This invention achieves multimodal data fusion, namely structured (inventory data) + unstructured (data jointly driving decision-making). Dynamic weight adjustment: Real-time balancing between VIP orders and costs is achieved through reinforcement learning, with a weight adjustment response time of <1 second. Simulation-execution closed loop: Digital twins predict risks, trigger local optimizations, and avoid post-event remediation.

[0113] The product supply system provided by the present invention is described below. The product supply system described below can be referred to in correspondence with the product supply method described above.

[0114] Figure 4 This is a schematic diagram of the product supply system provided by the present invention, such as... Figure 4 As shown, the system includes: The data acquisition layer 401 is used to collect multi-link supply information in real time through IoT devices and enterprise systems. The supply information includes at least one of inventory information, order information, logistics information, supply chain topology, and supply environment parameters. The feedback analysis layer 402 is used to perform anomaly detection and demand forecasting on the supply information using a streaming processing engine to obtain response priority and predicted demand. Evolutionary optimization layer 403 is used to dynamically generate at least one initial supply strategy based on the improved multi-objective genetic algorithm, the response priority and the predicted demand, and to determine the target supply strategy based on the simulation execution results corresponding to each initial supply strategy. The strategy execution layer 404 is used to execute product supply based on the target supply strategy and use the feedback data generated during the execution process as new supply information.

[0115] The product supply system provided by this invention achieves standardized processing and unified management of data across the entire supply chain at the data acquisition layer, solving pain points such as data fragmentation, inconsistent formats, and varying quality in traditional supply chain systems. Through the organic integration of real-time data streams and batch-processed data, it dynamically reflects the operational status of the supply chain, laying a solid foundation for subsequent intelligent analysis and decision optimization. This not only enhances supply chain transparency but also creates the necessary conditions for enterprises to achieve data-driven refined management. At the feedback analysis layer, by monitoring the effectiveness of event handling and cost savings, the system dynamically adjusts evaluation strategies, forming a self-improving decision-making mechanism that effectively balances response speed and processing costs, achieving an intelligent upgrade in supply chain anomaly management. At the evolutionary optimization layer, a dynamic genetic algorithm achieves minute-level response, supports multi-chain collaborative optimization, and improves robustness. Combined with reinforcement learning self-adjustment, it significantly improves the efficiency and adaptability of supply chain decision-making. At the strategy execution layer, it enables precise implementation of intelligent strategies, supports multi-system collaborative execution, possesses dynamic fault tolerance and real-time feedback capabilities, and significantly improves execution efficiency, ensuring the efficient and reliable implementation of supply chain strategies.

[0116] Optionally, the feedback analysis layer 402 is specifically used for: The streaming processing engine is used to input the supply information into the anomaly detection model for anomaly detection, obtain anomaly event information, and determine the response priority corresponding to the anomaly event information based on the anomaly event information. The anomaly detection model has built-in Drools rules, isolated forest algorithm and BERT model. The supply information is input into the demand forecasting model using the streaming processing engine to predict the demand and obtain the predicted demand. The demand forecasting model incorporates a long short-term memory model, a time series model, and various optimization algorithms.

[0117] Optionally, the anomaly detection includes threshold detection, latent anomaly detection, and text anomaly detection; The feedback analysis layer 402 is specifically used for: Using the streaming processing engine, the supply information is input into the anomaly detection model. The Drools rule is used to perform threshold detection on the supply information, the isolated forest algorithm is used to perform latent anomaly detection on the supply information, and the BERT model is used to perform text anomaly detection on the supply information to obtain the anomaly event information. The response priority corresponding to the abnormal event information is obtained by weighting the event urgency, impact scope and processing cost in the abnormal event information.

[0118] Optionally, the evolution optimization layer 403 is specifically used for: Based on the response priority and the predicted demand, an objective function and constraints are constructed. The objective function includes at least one of a function that minimizes total cost, a function that maximizes on-time delivery rate, and a robustness evaluation function. The constraints include at least one of a supply capacity constraint, a minimum order quantity constraint, and an order constraint. The improved multi-objective genetic algorithm is used to solve the objective function based on the constraints to obtain at least one initial supply strategy.

[0119] Optionally, the improved multi-objective genetic algorithm includes the following steps: (1) Determine the population initialization strategy: Load the population initialization strategy of the historical best solution from the database, or obtain the population initialization strategy using heuristic generation rules; (2) Adaptively adjust the crossover probability and the mutation probability, wherein the crossover probability and the mutation probability are adjusted by the following formula: in, For crossover probability, This represents the mutation probability of supplier i. This represents the latency rate of supplier i. This represents the maximum latency rate among all suppliers; (3) Multi-objective assessment and elite retention, wherein the multi-objective assessment is a robustness assessment, and the robustness score corresponding to the robustness assessment is determined by Monte Carlo perturbation test, and the calculation formula of the robustness score is as follows: in, For robustness scoring; (4) Hybrid local search, wherein the hybrid local search employs either simulated annealing or tabu search strategies; (5) Dynamic parameter tuning mechanism: A reinforcement learning controller is adopted to switch local search strategies and dynamically modify the weights of the objective function. The state space includes at least one of population diversity, convergence speed, external event type and resource constraints, and the action space includes crossover probability and / or mutation probability. The state is collected at intervals and actions are selected. The actions are executed based on the state to obtain the execution result, the reward is calculated and the parameters of the reinforcement learning controller are updated until convergence.

[0120] Optionally, the evolution optimization layer 403 is further used for: For each initial supply strategy, the initial supply strategy and the supply information are input into a digital twin simulator for discrete event simulation, proxy modeling, simulation acceleration and non-critical path simplification to obtain the simulation execution result corresponding to the initial supply strategy.

[0121] Optionally, the policy execution layer 404 is specifically used for: Real-time monitoring and key performance indicator calculation; The target supply strategy is dynamically adjusted based on the deviations corresponding to the key performance indicators. The product supply is delivered in stages when some nodes fail.

[0122] Optionally, the data acquisition layer is further used for: The supply information is subjected to format and logic validation. If the verification passes, an anomaly filtering algorithm is used to remove invalid data from the supply information, and a Kalman filtering algorithm is used to complete the missing spatiotemporal data in the supply information.

[0123] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5As shown, the electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, communications interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a product supply method. This method includes: real-time acquisition of multi-link supply information through IoT devices and enterprise systems, the supply information including at least one of inventory information, order information, logistics information, supply chain topology, and supply environment parameters; using a streaming processing engine to perform anomaly detection and demand forecasting on the supply information to obtain response priorities and predicted demand; dynamically generating at least one initial supply strategy based on an improved multi-objective genetic algorithm, the response priorities, and the predicted demand, and determining a target supply strategy based on the simulation execution results corresponding to each initial supply strategy; executing product supply based on the target supply strategy, and using the feedback data generated during the execution process as new supply information.

[0124] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0125] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the product supply method provided by the above methods. The method includes: collecting multi-link supply information in real time through IoT devices and enterprise systems, wherein the supply information includes at least one of inventory information, order information, logistics information, supply chain topology, and supply environment parameters; using a streaming processing engine to perform anomaly detection and demand forecasting on the supply information to obtain response priorities and predicted demand; dynamically generating at least one initial supply strategy based on an improved multi-objective genetic algorithm, the response priorities, and the predicted demand, and determining a target supply strategy based on the simulation execution results corresponding to each initial supply strategy; and executing product supply based on the target supply strategy, and using the feedback data generated during the execution process as new supply information.

[0126] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a product supply method provided by the above methods. This method includes: real-time collection of multi-link supply information via IoT devices and enterprise systems, the supply information including at least one of inventory information, order information, logistics information, supply chain topology, and supply environment parameters; using a streaming processing engine to perform anomaly detection and demand forecasting on the supply information to obtain response priorities and predicted demand; dynamically generating at least one initial supply strategy based on an improved multi-objective genetic algorithm, the response priorities, and the predicted demand, and determining a target supply strategy based on the simulation execution results corresponding to each initial supply strategy; and executing product supply based on the target supply strategy, using feedback data generated during the execution process as new supply information.

[0127] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0128] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0129] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A product supply method, characterized in that, include: Through IoT devices and enterprise systems, supply information across multiple links is collected in real time. This supply information includes at least one of the following: inventory information, order information, logistics information, supply chain topology, and supply environment parameters. A streaming processing engine is used to perform anomaly detection and demand forecasting on the supply information to obtain response priorities and predicted demand. Based on the improved multi-objective genetic algorithm, the response priority, and the predicted demand, at least one initial supply strategy is dynamically generated, and a target supply strategy is determined based on the simulation execution results corresponding to each initial supply strategy. Based on the target supply strategy, product supply is executed, and the feedback data generated during the execution process is used as new supply information.

2. The product supply method according to claim 1, characterized in that, The process employs a streaming engine to perform anomaly detection and demand forecasting on the supply information, obtaining response priorities and predicted demand, including: The streaming processing engine is used to input the supply information into the anomaly detection model for anomaly detection, obtain anomaly event information, and determine the response priority corresponding to the anomaly event information based on the anomaly event information. The anomaly detection model has built-in Drools rules, isolated forest algorithm and BERT model. The supply information is input into the demand forecasting model using the streaming processing engine to predict the demand and obtain the predicted demand. The demand forecasting model incorporates a long short-term memory model, a time series model, and various optimization algorithms.

3. The product supply method according to claim 2, characterized in that, The anomaly detection includes threshold detection, latent anomaly detection, and text anomaly detection; The process of using the streaming processing engine to input the supply information into an anomaly detection model for anomaly detection, obtaining anomaly event information, and determining the response priority corresponding to the anomaly event information based on the anomaly event information includes: Using the streaming processing engine, the supply information is input into the anomaly detection model. The Drools rule is used to perform threshold detection on the supply information, the isolated forest algorithm is used to perform latent anomaly detection on the supply information, and the BERT model is used to perform text anomaly detection on the supply information to obtain the anomaly event information. The response priority corresponding to the abnormal event information is obtained by weighting the event urgency, impact scope and processing cost in the abnormal event information.

4. The product supply method according to claim 1, characterized in that, The step of dynamically generating at least one initial supply strategy based on the improved multi-objective genetic algorithm, the response priority, and the predicted demand includes: Based on the response priority and the predicted demand, an objective function and constraints are constructed. The objective function includes at least one of a function that minimizes total cost, a function that maximizes on-time delivery rate, and a robustness evaluation function. The constraints include at least one of a supply capacity constraint, a minimum order quantity constraint, and an order constraint. The improved multi-objective genetic algorithm is used to solve the objective function based on the constraints to obtain at least one initial supply strategy.

5. The product supply method according to claim 1, characterized in that, The improved multi-objective genetic algorithm includes the following steps: (1) Determine the population initialization strategy: Load the population initialization strategy of the historical best solution from the database, or obtain the population initialization strategy using heuristic generation rules; (2) Adaptively adjust the crossover probability and the mutation probability, wherein the crossover probability and the mutation probability are adjusted by the following formula: in, For crossover probability, This represents the mutation probability of supplier i. This represents the latency rate of supplier i. This represents the maximum latency rate among all suppliers; (3) Multi-objective assessment and elite retention, wherein the multi-objective assessment is a robustness assessment, and the robustness score corresponding to the robustness assessment is determined by Monte Carlo perturbation test, and the calculation formula of the robustness score is as follows: in, For robustness scoring; (4) Hybrid local search, wherein the hybrid local search employs either simulated annealing or tabu search strategies; (5) Dynamic parameter tuning mechanism: A reinforcement learning controller is adopted to switch local search strategies and dynamically modify the weights of the objective function. The state space includes at least one of population diversity, convergence speed, external event type and resource constraints, and the action space includes crossover probability and / or mutation probability. The state is collected at intervals and actions are selected. The actions are executed based on the state to obtain the execution result, the reward is calculated and the parameters of the reinforcement learning controller are updated until convergence.

6. The product supply method according to claim 1, characterized in that, Before determining the target supply strategy based on the simulation execution results corresponding to each of the initial supply strategies, the process further includes: For each initial supply strategy, the initial supply strategy and the supply information are input into a digital twin simulator for discrete event simulation, proxy modeling, simulation acceleration and non-critical path simplification to obtain the simulation execution result corresponding to the initial supply strategy.

7. The product supply method according to claim 1, characterized in that, The process of executing product supply based on the target supply strategy includes: Real-time monitoring and key performance indicator calculation; The target supply strategy is dynamically adjusted based on the deviations corresponding to the key performance indicators. The product supply is delivered in stages when some nodes fail.

8. The product supply method according to any one of claims 1-7, characterized in that, Before employing a streaming processing engine to perform anomaly detection and demand forecasting on the supply information to obtain response priorities and predicted demand, the process further includes: The supply information is subjected to format and logic validation. If the verification passes, an anomaly filtering algorithm is used to remove invalid data from the supply information, and a Kalman filtering algorithm is used to complete the missing spatiotemporal data in the supply information.

9. A product supply system, characterized in that, include: The data acquisition layer is used to collect multi-link supply information in real time through IoT devices and enterprise systems. The supply information includes at least one of inventory information, order information, logistics information, supply chain topology, and supply environment parameters. The feedback analysis layer is used to perform anomaly detection and demand forecasting on the supply information using a streaming processing engine, thereby obtaining response priorities and predicted demand. An evolutionary optimization layer is used to dynamically generate at least one initial supply strategy based on the improved multi-objective genetic algorithm, the response priority, and the predicted demand, and to determine a target supply strategy based on the simulation execution results corresponding to each initial supply strategy. The strategy execution layer is used to execute product supply based on the target supply strategy and to use the feedback data generated during the execution process as new supply information.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the product supply method as described in any one of claims 1 to 8.

Images