Production and logistics management method and system based on digital twinning

Abstract

The application discloses a production and logistics management method and system based on digital twinning, obtains real-time data streams from production equipment sensors and logistics nodes, adopts a stream processing technology to clean and format the real-time data streams, and obtains a structured real-time data set; for the structured real-time data set, a time series analysis algorithm is applied to detect equipment abnormalities and order fluctuations; if the equipment abnormalities or the order fluctuations are detected, a pre-established digital twinning model is used to simulate the production and logistics links in real time, and a dynamic change trend of event influence is obtained; according to the dynamic change trend, a reinforcement learning algorithm is used to optimize and adjust the production scheduling and the logistics path, and an optimized production and transportation plan is determined. The application realizes intelligent cooperation of the production and logistics links, can quickly respond to abnormal conditions, dynamically optimizes scheduling decisions, and improves overall operation efficiency.

Description

Technical Field

[0001] This invention relates to the field of digital management technology, and in particular discloses a production and logistics management method and system based on digital twins. Background Technology

[0002] In the field of modern industry and supply chain management, building an efficient production and logistics system has become a key pillar for enhancing corporate competitiveness. The importance of this area is self-evident, as it directly relates to resource utilization efficiency, cost control, and market responsiveness.

[0003] However, current mainstream methods often reveal shortcomings in adapting to changing needs and lacking deep data integration when dealing with complex and dynamic environments. These methods typically rely on the traditional separation of planning and execution, making it difficult to make timely adjustments in the face of unexpected events or multi-stage collaboration, leading to resource waste and inefficiency. A deeper analysis of the challenges in this field reveals that the core issue centers on how to achieve real-time dynamic response and precise optimization across the entire process.

[0004] The primary challenge lies in the high degree of uncertainty inherent in various events within the production and logistics processes, such as equipment malfunctions or order fluctuations. Existing systems struggle to effectively simulate and intervene before or during these events. This uncertainty further exacerbates another critical issue: the fragmentation of data across different stages. The lack of seamless integration from design to execution leads to decisions often based on localized information rather than a holistic perspective, ultimately impacting overall optimization effectiveness.

[0005] Therefore, how to achieve dynamic adjustment and intelligent decision-making of production and logistics processes in complex and ever-changing industrial environments by building a mechanism that integrates real-time simulation and full-chain data has become a key issue that urgently needs to be addressed. Summary of the Invention

[0006] This invention provides a production and logistics management method and system based on digital twins, aiming to solve at least one of the defects existing in the prior art.

[0007] One aspect of the present invention relates to a production and logistics management method based on digital twins, comprising the following steps: Real-time data streams are acquired from production equipment sensors and logistics nodes. Streaming processing technology is used to clean and format the real-time data streams to obtain structured real-time datasets. For structured real-time datasets, time series analysis algorithms are applied to detect equipment anomalies and order fluctuations; If equipment malfunctions or order fluctuations are detected, the production and logistics processes are simulated in real time using a pre-established digital twin model to obtain the dynamic trend of the event's impact. Based on dynamic trends, reinforcement learning algorithms are used to optimize and adjust production scheduling and logistics routes to determine optimized production and transportation plans. Key decision parameters are extracted from the optimized production and transportation plans, and combined with the full-chain data integration platform to generate cross-link collaborative scheduling instructions; For collaborative scheduling instructions, the instructions are decomposed into specific tasks through a distributed computing framework, which are then allocated to production equipment and logistics nodes to obtain status feedback after execution. Based on the status feedback, online learning algorithms are applied to update the parameters of the digital twin model, thereby optimizing the accuracy of subsequent simulation intervention mechanisms.

[0008] Furthermore, the steps of acquiring real-time data streams from production equipment sensors and logistics nodes, and using streaming processing technology to clean and format the real-time data streams to obtain structured real-time datasets include: Real-time data streams are acquired from sensors and logistics nodes. A message queue tool is used to receive time-series data from the sensors and tracking data from the logistics nodes to obtain the raw data stream. For the original data stream, a streaming processing tool is used to clean it. If there are missing or outlier values ​​in the original data stream, a preset threshold is used to determine whether to fill in the mean or delete them, resulting in a cleaned data stream. Based on the cleaned data stream, a data transformation tool is used to perform a formatting operation, unifying the cleaned data stream into a pre-established structured data model to obtain structured data; Structured data is stored and integrated using database tools. If the structured data meets the preset time window requirements, a real-time dataset is generated.

[0009] Furthermore, for structured real-time datasets, the steps for applying time series analysis algorithms to detect equipment anomalies and order fluctuations include: The device status and operating parameter information are obtained from the structured real-time dataset. The operating parameters are compared item by item using pre-established comparison rules. If the operating parameter exceeds the preset threshold range, it is marked as an abnormal signal, and data records with abnormal markings are obtained. For data records marked with anomalies, obtain the anomaly signal and fault type information, compare it with historical trend data, determine whether the anomaly signal conforms to the known fault type pattern, and determine the specific fault classification result; Based on the fault classification results, combined with the order quantity and fluctuation range information, the data with anomaly markers are segmented using a time window division method. If the fluctuation range exceeds the preset threshold range, it is marked as an order anomaly, thus obtaining an abnormal fluctuation identifier. To address the needs for identifying abnormal fluctuations and real-time monitoring, alarm thresholds and business impact information are obtained. Data marked with anomalies is comprehensively evaluated using preset logical rules to determine whether an alarm mechanism is triggered, resulting in the final alarm trigger record.

[0010] Furthermore, if equipment malfunctions or order fluctuations are detected, the production and logistics processes are simulated in real time using a pre-established digital twin model to obtain the dynamic trends of the event's impact. The steps include: The device status and operating parameters are obtained from real-time datasets, and the operating parameters are simulated in real time using a pre-established digital twin model to obtain the dynamic trend of device operation. Based on the dynamic changes in equipment operation, abnormal signals and historical trend data are acquired. By comparing the abnormal signals with historical trends through logical rules, it is determined whether they conform to the preset fault type pattern and the fault classification result is determined. Based on the fault classification results and order quantity data, the order quantity data is segmented using a time window method. If the fluctuation range exceeds the preset threshold range, it is marked as abnormal fluctuation, and an abnormal fluctuation identifier is obtained. For abnormal fluctuation indicators, obtain alarm thresholds and business impact information, comprehensively evaluate the abnormal fluctuation indicators through logical rules, determine whether to trigger the alarm mechanism, and obtain alarm trigger records.

[0011] Furthermore, based on dynamic trends, the steps of using reinforcement learning algorithms to optimize and adjust production scheduling and logistics routes to determine the optimized production and transportation plans include: Based on dynamic trends, real-time updated data is obtained from order demand and inventory status. The resource allocation is compared with the pre-established database. If the resource allocation cannot meet the order demand, the adjustment process is triggered to determine the preliminary production scheduling plan. For the initial production scheduling plan, relevant information on logistics routes and route costs is obtained. Route planning tools are used to evaluate the first delivery efficiency. If the first delivery efficiency is lower than the preset threshold, the logistics routes are replanned to obtain the adjusted transportation plan. By adjusting the transportation plan, combining the first time window and demand forecast data, and using scheduling management tools to allocate time for production and delivery, if the delivery task cannot be completed within the first time window, the resource allocation will be adjusted a second time to determine the final production scheduling result. Based on the final production scheduling results, data on changes in inventory status and order demand are obtained. Data comparison tools are used to continuously monitor the efficiency of the first delivery and the route cost to determine whether the optimized transportation plan meets the expected goals.

[0012] Furthermore, the steps for extracting key decision parameters from the optimized production and transportation plans and generating cross-link collaborative scheduling instructions by integrating them with the full-chain data platform include: Real-time data on order demand and inventory status are obtained from the optimized production and transportation plan. The resource allocation is compared with the pre-established database. If the resource allocation cannot meet the order demand, the data integration process is triggered to obtain the initial collaborative scheduling parameters. Based on the preliminary collaborative scheduling parameters, relevant data of the logistics path and the second time window are obtained. The second delivery efficiency is evaluated using a path planning tool. If the second delivery efficiency is lower than the preset threshold, the logistics path is replanned and the adjusted collaborative scheduling instructions are determined. By adjusting the collaborative scheduling instructions, combining demand forecasting and monitoring frequency data, and using scheduling management tools to allocate time for production scheduling and delivery, if the delivery task cannot be completed within the second time window, the resource allocation will be adjusted a second time to determine the final collaborative scheduling scheme. Based on the final collaborative scheduling plan, dynamic data on inventory status and route costs are obtained. The efficiency of the second delivery and the effect of collaborative scheduling are continuously monitored through the data integration platform to obtain optimized cross-link scheduling instructions.

[0013] Furthermore, for collaborative scheduling instructions, the steps of decomposing the instructions into specific tasks using a distributed computing framework, allocating them to production equipment and logistics nodes, and obtaining post-execution status feedback include: Based on the collaborative scheduling instructions, the instructions are decomposed using distributed computing tools to generate multiple specific task units. The tasks are then initially sorted according to their priority and execution sequence to obtain a list of decomposed tasks. Based on the task list, and combined with the equipment load and node capacity data of production equipment and logistics nodes, tasks are allocated using resource allocation tools. If the equipment load exceeds the preset threshold, the task allocation order is adjusted to determine the final task assignment scheme. Based on the task assignment plan, specific tasks are assigned to production equipment and logistics nodes. Status feedback data during the execution process is collected in real time. The execution efficiency is monitored by the feedback frequency to obtain dynamic updates of the execution status. In response to the dynamic updates of the execution status, if the execution efficiency is lower than the preset threshold, the task allocation is optimized through the instruction adjustment tool. The execution sequence is re-planned based on the device load and node capacity data, and the adjusted task execution arrangement is determined.

[0014] Furthermore, based on the status feedback, the steps of updating the parameters of the digital twin model using online learning algorithms and optimizing the accuracy of subsequent simulation intervention mechanisms include: The execution data, which is the status feedback data, is obtained from the production node and the logistics node through data acquisition tools. The execution data is then structured and classified to obtain a list of classified execution data. A real-time update tool is used to dynamically monitor the execution data list. If the execution efficiency is lower than the preset threshold, the parameters of the simulation intervention mechanism are corrected by the parameter adjustment tool to determine the adjusted parameter configuration. By continuously tracking parameter configurations through feedback frequency monitoring tools, performance data under different business loads can be obtained. The data comparison tool is used to compare the performance data with the preset accuracy optimization target. If the preset threshold is not reached, the data backtracking tool is used to sort out the historical status feedback to obtain the updated data basis.

[0015] Furthermore, a real-time update tool is used to dynamically monitor the execution data list. If the execution efficiency is lower than a preset threshold, the parameters of the simulation intervention mechanism are corrected using a parameter adjustment tool. In the step of determining the adjusted parameter configuration, the system execution efficiency is calculated in real time using the following formula and compared with the preset threshold:

[0016] in, Indicates the system during a time period Internal execution efficiency This indicates the total number of sampling points within the monitoring time window. Indicates the first The output quantity of each sampling point Indicates the first Input resource quantity per sampling point Indicates the first The execution time interval for each sampling point; The adjusted parameter configuration is calculated using the following formula:

[0017] in, This indicates the adjusted new parameter configuration. This indicates the current parameter configuration. This represents the learning rate used to adjust the parameters. Describe the objective function For parameters gradient, This represents the preset efficiency threshold. This represents the actual efficiency value currently monitored.

[0018] Another aspect of the present invention relates to a digital twin-based production and logistics management system for executing the aforementioned digital twin-based production and logistics management method. The digital twin-based production and logistics management system includes: The real-time dataset acquisition module is used to acquire real-time data streams from production equipment sensors and logistics nodes, and to clean and format the real-time data streams using streaming processing technology to obtain structured real-time datasets. The anomaly detection module is used to detect equipment anomalies and order fluctuations using time series analysis algorithms on structured real-time datasets. The dynamic trend acquisition module is used to simulate the production and logistics process in real time through a pre-established digital twin model if equipment abnormalities or order fluctuations are detected, so as to obtain the dynamic trend of the event's impact. The planning module is used to optimize and adjust production scheduling and logistics routes based on dynamic trends using reinforcement learning algorithms, and to determine the optimized production and transportation plan. The collaborative scheduling instruction generation module is used to extract key decision parameters from the optimized production and transportation plans, and combine them with the full-chain data integration platform to generate cross-link collaborative scheduling instructions. The status feedback acquisition module is used to decompose the collaborative scheduling instructions into specific tasks through a distributed computing framework, allocate them to production equipment and logistics nodes, and acquire status feedback after execution. The accuracy optimization module is used to update the parameters of the digital twin model based on status feedback and by applying online learning algorithms, thereby optimizing the accuracy of subsequent simulation intervention mechanisms.

[0019] The beneficial effects achieved by this invention are as follows: This invention provides a production and logistics management method and system based on digital twins. It detects anomalies through real-time data stream processing and time series analysis, simulates the impact of events using a digital twin model, optimizes production scheduling and logistics routes using reinforcement learning algorithms, generates cross-stage collaborative scheduling instructions and decomposes them for execution, and finally applies online learning to update model parameters. This invention achieves intelligent collaboration between production and logistics, enabling rapid response to anomalies, dynamic optimization of scheduling decisions, and improved overall operational efficiency. Through data-driven and artificial intelligence technologies, it constructs an adaptive, closed-loop optimized production and logistics collaborative system, providing an effective solution for manufacturing enterprises to achieve intelligent transformation. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating an embodiment of a production and logistics management method based on digital twins according to the present invention. Figure 2 This is a functional block diagram of an embodiment of a production and logistics management system based on digital twins according to the present invention.

[0021] Explanation of icon numbers: 10. Real-time dataset acquisition module; 20. Anomaly detection module; 30. Dynamic change trend acquisition module; 40. Plan determination module; 50. Cooperative scheduling instruction generation module; 60. Status feedback acquisition module; 70. Precision optimization module. Detailed Implementation

[0022] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0023] like Figure 1 As shown, the first embodiment of the present invention proposes a production and logistics management method based on digital twins, including the following steps: Step S100: Obtain real-time data streams from production equipment sensors and logistics nodes, and use streaming processing technology to clean and format the real-time data streams to obtain structured real-time datasets.

[0024] By deploying various sensors (such as temperature sensors, vibration sensors, and pressure sensors) on production equipment (such as machine tools, production lines, and robots) and at key nodes in the logistics process (such as warehouse entrances, GPS on transport vehicles, and sorting equipment), real-time data reflecting equipment operating status and material flow is continuously collected (such as equipment temperature changes, logistics location coordinates, and material throughput). Subsequently, streaming processing technologies (such as Apache Flink and Apache Kafka Streams) are used to clean and format these continuously generated unstructured or semi-structured real-time data streams, ultimately forming a structured real-time dataset.

[0025] Data cleaning: Remove noise (such as abnormal values ​​falsely reported by sensors), duplicate data (such as duplicate reports from the same logistics node), missing values ​​(such as data gaps caused by temporary sensor failures), etc., to ensure the accuracy and integrity of the data.

[0026] Formatting: Converting cleaned data into a uniform structured format (such as JSON or CSV formats that conform to a specific database table structure, or field formats of relational databases, etc.), so that the data has standardized field definitions, data types, and logical relationships.

[0027] A structured real-time dataset is a dynamic dataset that can reflect the operating status of production equipment and the flow of logistics nodes in real time, and has a unified data format and clear data logic.

[0028] Step S200: For the structured real-time dataset, apply time series analysis algorithms to detect equipment anomalies and order fluctuations.

[0029] Based on the constructed structured real-time dataset (containing structured information such as equipment operation indicators and order flow data that are continuously recorded over time), time series analysis algorithms are applied to mine the time correlation patterns and trend characteristics hidden in the data, thereby identifying abnormal equipment operation status and abnormal fluctuations in order volume in real time.

[0030] Step S300: If equipment abnormality or order fluctuation is detected, the production and logistics links are simulated in real time through a pre-established digital twin model to obtain the dynamic trend of the event's impact.

[0031] When abnormal conditions of production equipment (such as sudden temperature rise, abnormal vibration frequency, etc.) or fluctuations in order business (such as sudden increase / decrease in order volume, sudden change in regional order distribution, etc.) are detected in real time through time series analysis algorithms, the key parameters of the abnormal event (such as the time of equipment abnormality, fault type, order fluctuation range, etc.) are input into the model based on the pre-built digital twin model of production and logistics links. The model dynamically simulates and extrapolates the real-time status of production process and logistics scheduling after the event occurs, and finally outputs the scope, degree and dynamic trend of the impact of the event on production efficiency, material flow, order fulfillment and other links.

[0032] Step S400: Based on the dynamic change trend, use reinforcement learning algorithm to optimize and adjust production scheduling and logistics routes to determine the optimized production and transportation plan.

[0033] After acquiring the dynamic impact trends of equipment malfunctions or order fluctuations (such as the quantitative impact of process delays in the production process and changes in transportation capacity gaps in the logistics process over time), reinforcement learning algorithms are deployed to optimize and adjust production scheduling schemes and logistics route planning in real time and dynamically with the goal of minimizing the negative impact of events (such as production downtime and order fulfillment delays). Ultimately, the optimal production plan and transportation scheme adapted to the current abnormal situation are generated.

[0034] Step S500: Extract key decision parameters from the optimized production and transportation plan, and generate cross-link collaborative scheduling instructions by combining them with the full-chain data integration platform.

[0035] Based on the optimized production and transportation plans (including details of production scheduling adjustments and logistics route optimization schemes), core decision parameters affecting the coordinated operation of multiple links such as production and logistics are extracted. Then, relying on a full-chain data integration platform (a unified data hub aggregating data from the entire process of production, warehousing, transportation, and orders), multi-dimensional verification and correlation analysis are performed. Ultimately, standardized scheduling instructions that can coordinate efficient cross-linking of production, logistics, and warehousing are generated, achieving coordinated resource allocation and seamless integration of business processes across all links. In this embodiment, the core decision parameters are collaborative scheduling parameters, which are core decision parameters extracted from the optimized production and transportation plans that directly affect the efficiency of coordinated operation across multiple links such as production, logistics, and warehousing. By clarifying the target requirements, resource constraints, time nodes, and connection logic of each link, collaborative scheduling parameters construct a "dynamic adaptation rule" between production and logistics. This is key data support for eliminating cross-link resource conflicts and achieving efficient full-chain linkage, and is also the core basis for generating cross-link collaborative scheduling instructions.

[0036] Step S600: For the collaborative scheduling instruction, the instruction is decomposed into specific tasks through a distributed computing framework, allocated to production equipment and logistics nodes, and status feedback after execution is obtained.

[0037] Taking cross-process collaborative scheduling instructions (standardized instructions that coordinate the linkage of multiple processes such as production and logistics) as input, and relying on distributed computing frameworks (such as Apache Spark, Hadoop YARN and other computing architectures that support multi-node parallel processing), the instructions are broken down into specific sub-tasks that can be directly executed by production equipment and logistics nodes. These sub-tasks are then allocated to the corresponding execution units (such as machine tools, transport vehicles, warehouse equipment, etc.) through a task scheduling mechanism. The system also collects real-time status feedback such as the execution progress and completion status of each sub-task, forming a closed-loop monitoring of instruction execution.

[0038] Step S700: Based on the status feedback, apply an online learning algorithm to update the parameters of the digital twin model and optimize the accuracy of subsequent simulation intervention mechanisms.

[0039] Based on real-time status feedback data (such as actual equipment operating parameters, actual order fulfillment timeliness, and actual effects of intervention measures) after executing collaborative scheduling instructions from physical entities like production equipment and logistics nodes, the core parameters of the digital twin model are continuously corrected and optimized by applying online learning algorithms (machine learning methods that can receive new data in real time and dynamically adjust the model). This continuously improves the mapping accuracy and simulation prediction capabilities of the digital twin model to the physical system, thereby enhancing the accuracy of subsequent simulations and intervention decisions for abnormal events. The digital twin model is a digital mirror image of the physical entity. Its core parameters are crucial data supporting the "dynamic mapping, precise interaction, and continuous optimization" between the model and the physical entity. They directly determine the model's accuracy in reproducing the physical entity's state, its ability to predict future behavior, and the effectiveness of interventions. Updated digital twin model parameters include those for the simulation intervention mechanism. This mechanism is the core function of the digital twin model, enabling "virtual-real interaction and proactive control." The parameters define the digital twin model's intervention logic, execution strategy, and effect evaluation criteria for the physical entity, ensuring the rationality, executability, and dynamic adaptation of intervention instructions to the entity's state.

[0040] Furthermore, the production and logistics management method based on digital twins provided in this embodiment includes step S100: Step S110: Obtain real-time data streams from sensors and logistics nodes. Use a message queue tool to receive time-series data from sensors and tracking data from logistics nodes to obtain raw data streams.

[0041] In real-time data processing scenarios within the logistics industry, data acquisition from sensors and logistics nodes forms the foundation of the entire process. Sensors may be deployed on transport vehicles to collect environmental data such as temperature and humidity in real time, while logistics nodes record the scanning time and location information of packages. This data is received through message queue tools such as Kafka, forming a raw data stream. Kafka processes large-scale data with high throughput, ensuring the real-time nature and reliability of data transmission, thus laying the foundation for subsequent processing.

[0042] Step S120: For the original data stream, use a streaming processing tool to clean it. If there are missing or outlier values ​​in the original data stream, judge and fill the mean or delete them according to the preset threshold to obtain the cleaned data stream.

[0043] For the cleaning of the raw data stream, streaming processing tools such as Apache Flink are used. If sensor data contains missing or abnormal temperature values—for example, a temperature reading of -50 degrees Celsius, significantly exceeding the preset threshold of -20 to 50 degrees Celsius—the system automatically identifies and marks it as an outlier. It then fills the gap with the average temperature of the past hour, such as 25 degrees Celsius, or simply deletes the record. This cleaning method effectively improves data quality, avoids interference from outliers in subsequent analysis, and ensures the accuracy of the data stream.

[0044] Step S130: Based on the cleaned data stream, use a data conversion tool to perform a formatting operation to unify the cleaned data stream into a pre-established structured data model, thereby obtaining structured data.

[0045] The cleaned data stream needs to be formatted using data transformation tools to unify it into a structured data model. Assuming the original data stream contains sensor data in JSON format and logistics node data in CSV format, the system will use tools like Apache NiFi to convert it into a unified structured format, such as a table model containing fields like timestamps, device IDs, temperature values, and locations. This conversion not only facilitates data storage and retrieval but also improves data consistency, providing convenience for subsequent analysis.

[0046] Step S140: The structured data is stored and integrated using database tools. If the structured data meets the preset time window requirements, the real-time dataset is generated, and the real-time dataset is obtained.

[0047] Structured data storage and integration can be achieved using database tools such as PostgreSQL. Data is stored in timestamp-partitioned partitions, and if a preset time window requirement is met, such as generating a dataset every 5 minutes, a real-time dataset is triggered. This mechanism ensures efficient data management and timely updates, meeting real-time monitoring needs. The generated real-time datasets are used to monitor anomalies in the logistics process, such as excessive temperature or package delays, thereby improving logistics efficiency and customer satisfaction.

[0048] Furthermore, in the production and logistics management method based on digital twins provided in this embodiment, step S200 includes: Step S210: Obtain device status and operating parameter information from the structured real-time dataset, compare the operating parameters item by item using pre-established comparison rules, and mark the operating parameters as abnormal signals if they exceed the preset threshold range, thus obtaining data records with abnormal markings.

[0049] In real-time data processing scenarios within the logistics field, acquiring equipment status and operating parameter information from structured real-time datasets is a core step in anomaly detection. Structured datasets typically contain sensor data from transport vehicles, such as engine speed, oil temperature, and tire pressure, as well as equipment status information, including whether it is operating normally. For example, if a transport vehicle's real-time dataset shows an engine speed of 5000 rpm, exceeding a preset threshold of 4000 rpm, the system will mark this data as an anomaly signal based on pre-established comparison rules.

[0050] Step S220: For data records marked with anomalies, obtain the abnormal signal and fault type information, compare it with historical trend data, determine whether the abnormal signal conforms to the known fault type pattern, and determine the specific fault classification result.

[0051] By comparing data item by item using streaming tools such as Apache Spark Streaming, and combining this with threshold ranges (e.g., 3000-4000 RPM is considered normal), anomalies can be identified. This method can quickly identify potential problems and ensure the accuracy of subsequent analysis.

[0052] Step S230: Based on the fault classification results, combined with the order quantity and fluctuation range information, the data with abnormal markers is segmented using a time window division method. If the fluctuation range exceeds the preset threshold range, it is marked as an order abnormality, and an abnormal fluctuation identifier is obtained.

[0053] For data records marked with anomalies, the system further analyzes the matching between the abnormal signals and historical trend data to determine the fault type. For example, historical trend data indicates that when the engine speed continuously exceeds 4000 rpm and the oil temperature is above 90 degrees Celsius, it corresponds to the "engine overheating" fault mode. Assuming the current abnormal signal shows an engine speed of 4500 rpm and an oil temperature of 95 degrees Celsius, the system compares it with historical data to confirm that it matches the "engine overheating" mode and generates a fault classification result. This comparison method based on historical trends can effectively distinguish between intermittent anomalies and systemic faults, improving the accuracy of fault diagnosis.

[0054] Step S240: For abnormal fluctuation identification and real-time monitoring requirements, obtain alarm thresholds and business impact information, comprehensively evaluate the data with abnormal markings through preset logical rules, determine whether to trigger the alarm mechanism, and obtain the final alarm trigger record.

[0055] By combining order quantity and fluctuation range information, the system segments abnormal data. For example, suppose a logistics node records a surge in the number of orders from the normal 100 to 200 within a 5-minute time window, exceeding a preset threshold by 50%. The system marks this data segment as abnormal by dividing it into time windows. In practice, a database such as ClickHouse is used to store data by time window, and the fluctuation range is calculated using SQL queries. This segmented processing method accurately captures the periods of order anomalies, facilitating subsequent analysis and management.

[0056] To address the needs for identifying abnormal fluctuations and real-time monitoring, the system will comprehensively assess whether to trigger an alarm mechanism. For example, if an order anomaly is combined with information about its impact on business (such as an abnormal order involving high-value goods), the system will determine, based on preset logical rules, that an alarm should be triggered immediately.

[0057] Suppose an abnormal fluctuation indicator shows a surge in orders involving cold chain goods. The system, combined with alarm thresholds (e.g., anomalies in cold chain goods should be reported within 2 minutes), generates an alarm trigger record and notifies dispatchers to adjust transportation plans. This comprehensive assessment mechanism enables rapid response to business risks, improving the timeliness and reliability of logistics management.

[0058] The aforementioned steps, through streaming processing, historical trend comparison, segmented processing, and comprehensive evaluation, form a complete anomaly detection and alarm system. Each step is closely linked, from equipment status monitoring to fault classification, and then to order anomaly detection and alarm triggering, jointly ensuring the stability of the logistics process. This approach not only quickly identifies problems but also optimizes resource scheduling through precise fault classification and alarm mechanisms, significantly improving logistics efficiency.

[0059] Furthermore, in the production and logistics management method based on digital twins provided in this embodiment, step S300 includes: Step S310: Obtain the device status and operating parameters from the real-time dataset, and simulate the operating parameters in real time using a pre-established digital twin model to obtain the dynamic change trend of the device operation.

[0060] In real-time data processing scenarios within the logistics field, monitoring the equipment status and operating parameters of transport vehicles effectively identifies potential problems. For processing real-time datasets, assuming a transport vehicle's sensor data includes information such as speed, fuel consumption, and load, a pre-built digital twin model simulates the vehicle's operating status under different conditions. For example, the model simulation shows that the vehicle's fuel consumption should be 15 liters per 100 kilometers when carrying a 5-ton load and traveling at 80 km / h. If the real-time data deviates significantly from this value, an anomaly may be detected. This simulation method dynamically reflects equipment operating trends, providing a basis for subsequent analysis.

[0061] Step S320: Based on the dynamic change trend of equipment operation, acquire abnormal signals and historical trend data, compare the abnormal signals with historical trends through logical rules, determine whether they conform to the preset fault type pattern, and determine the fault classification result.

[0062] Based on the dynamic changes in equipment operation, the system extracts abnormal signals and compares them with historical trend data. For example, if a vehicle's speed suddenly drops from 60 km / h to 20 km / h within a short period, and historical data indicates that similar situations are often accompanied by braking system problems, the system uses logical rules to determine that it likely fits the "braking anomaly" fault mode, thus classifying the fault. This comparison method can quickly pinpoint the problem type, providing precise guidance for subsequent processing.

[0063] Step S330: Based on the fault classification results and order quantity data, the order quantity data is segmented using a time window division method. If the fluctuation range exceeds the preset threshold range, it is marked as abnormal fluctuation, and an abnormal fluctuation identifier is obtained.

[0064] In processing order quantity data, the system uses a time window approach to segment and analyze the data. For example, if the number of orders at a logistics node surges from 50 to 120 within 10 minutes, exceeding a preset threshold of 30%, the system will mark this period as an abnormal fluctuation. This segmented processing clearly captures key time points of data change, facilitating further analysis of the causes of anomalies. The size of the time window can be adjusted according to business needs; for example, it can be shortened to 5 minutes during peak periods to improve sensitivity.

[0065] Step S340: For the abnormal fluctuation identifier, obtain the alarm threshold and business impact information, comprehensively evaluate the abnormal fluctuation identifier through logical rules, determine whether the alarm mechanism is triggered, and obtain the alarm trigger record.

[0066] For abnormal fluctuation indicators, the system conducts a comprehensive assessment based on alarm thresholds and business impact information. For example, if an abnormal fluctuation involves the transportation of special goods, and the alarm threshold requires a response within 3 minutes, the system uses logical rules to determine if an alarm mechanism needs to be triggered, generates a corresponding alarm trigger record, and notifies relevant personnel to take appropriate measures. This assessment method ensures that important anomalies are handled promptly, reducing business risks. The weight of business impact information can be dynamically adjusted based on the value of the goods or the urgency of transportation to optimize alarm priority.

[0067] The aforementioned stages, through real-time simulation, historical comparison, segmented analysis, and comprehensive evaluation, form a complete monitoring system. Each stage supports the others, from dynamic changes in equipment status to fault classification, and then to the marking of order anomalies and alarm triggering, jointly ensuring the stability of logistics and transportation. This approach enables rapid response to potential problems and improves management efficiency.

[0068] Furthermore, in the production and logistics management method based on digital twins provided in this embodiment, step S400 includes: Step S410: Based on the dynamic trend, obtain real-time updated data from order demand and inventory status, compare the resource allocation with the pre-established database, and if the resource allocation cannot meet the order demand, trigger the adjustment process to determine the preliminary production scheduling plan.

[0069] Monitoring dynamic trends involves analyzing real-time data on order demand and inventory status to determine if resources can meet business needs. For example, if a logistics center processes 2000 orders daily, but current inventory indicates available delivery vehicles can only support 1500 orders, a comparison with a pre-established database reveals a resource allocation gap. The system automatically triggers an adjustment process, generating an initial production scheduling plan to add temporary vehicles. This approach promptly identifies resource shortages, providing a foundation for subsequent optimization.

[0070] Step S420: For the preliminary production scheduling plan, obtain relevant information on logistics routes and route costs, and use route planning tools to evaluate the first delivery efficiency. If the first delivery efficiency is lower than a preset threshold, the logistics route is replanned to obtain an adjusted transportation plan.

[0071] For the initial production scheduling plan, the system will evaluate it by combining logistics route and route cost information. Assuming a delivery route is 100 kilometers long and costs 2 yuan per kilometer, if the route planning tool finds that the delivery efficiency is below a preset threshold (e.g., only 20 kilometers of delivery can be completed per hour on average), the system will replan the route, shortening it to 80 kilometers. The adjusted transportation plan will reduce the cost to 1.8 yuan per kilometer. This replanning effectively improves delivery efficiency and reduces operating costs.

[0072] Step S430: Based on the adjusted transportation plan, combined with the first time window and demand forecast data, use scheduling management tools to allocate time for production and delivery. If the delivery task cannot be completed within the first time window, the resource allocation will be adjusted a second time to determine the final production scheduling result.

[0073] In the adjusted transportation plan, the system combines time windows and demand forecast data to allocate time. For example, if a batch of goods needs to be delivered within two hours, and forecasts indicate that peak-hour traffic congestion may cause delays, the scheduling management tool will advance the delivery time by 30 minutes. If this still cannot meet the time window requirements, resources will be adjusted further, such as adding backup vehicles, to ultimately determine the production scheduling result. This time allocation method ensures the on-time performance of delivery tasks and improves customer satisfaction.

[0074] Step S440: Based on the final production scheduling results, obtain data on changes in inventory status and order demand, continuously monitor the first delivery efficiency and route cost through data comparison tools, and determine whether the optimized transportation plan meets the expected goals.

[0075] Based on the final production scheduling results, the system continuously monitors inventory status and changes in order demand. For example, if the order volume suddenly increases to 2500 orders on a given day, while inventory vehicles are still allocated as originally planned, data comparison tools might detect a 10% decrease in delivery efficiency and a 15% increase in route costs. The system will then promptly adjust the transportation plan to ensure the optimization results meet expected goals. This continuous monitoring mechanism dynamically adapts to business fluctuations, ensuring the stability of the logistics system.

[0076] Furthermore, in the production and logistics management method based on digital twins provided in this embodiment, step S500 includes: Step S510: Obtain real-time data on order demand and inventory status from the optimized production and transportation plan. Compare the resource allocation with the pre-established database. If the resource allocation cannot meet the order demand, trigger the data integration process to obtain preliminary collaborative scheduling parameters.

[0077] In the resource allocation and scheduling optimization scenarios of the logistics field, real-time data acquisition of order demand and inventory status is fundamental to optimizing production and transportation plans. Suppose a logistics center handles 1800 orders daily, but inventory shows available vehicles can only support 1400 delivery tasks. A comparison with a pre-established database reveals a significant resource gap. At this point, the system triggers a data integration process, initially generating collaborative scheduling parameters involving vehicle allocation and personnel arrangements. This allows for rapid identification of resource shortages, providing data support for subsequent adjustments. Collaborative scheduling parameters are core decision-making parameters extracted from optimized production and transportation plans, decisively influencing the collaborative operation across production and logistics stages. These parameters clearly define the target requirements, resource constraints, time nodes, and connection logic of production, transportation, and warehousing, serving as crucial data support for efficient linkage between stages, avoiding resource conflicts, and ensuring smooth operation of the entire chain. They are also the core basis for generating cross-stage collaborative scheduling instructions.

[0078] Step S520: Based on the preliminary collaborative scheduling parameters, obtain relevant data on the logistics path and the second time window, use a path planning tool to evaluate the second delivery efficiency, and if the second delivery efficiency is lower than a preset threshold, re-plan the logistics path and determine the adjusted collaborative scheduling instructions.

[0079] Based on the initial collaborative scheduling parameters, the system combines logistics route and time window data, and uses route planning tools to evaluate delivery efficiency. For example, assuming a route is 120 kilometers long and the preset delivery efficiency threshold is 25 kilometers per hour, but the actual evaluation shows it only reaches 18 kilometers per hour, the system will replan the route, shortening the total length to 90 kilometers, and determine the adjusted collaborative scheduling instructions. This replanning effectively improves delivery speed and reduces wasted time.

[0080] Step S530: Using the adjusted collaborative scheduling instructions, combined with demand forecasting and monitoring frequency data, the scheduling management tool is used to allocate time for production scheduling and delivery. If the delivery task cannot be completed within the second time window, the resource allocation is adjusted a second time to determine the final collaborative scheduling scheme.

[0081] Based on the adjusted collaborative scheduling instructions, the system combines demand forecasting and monitoring frequency data, and utilizes scheduling management tools to allocate time. For example, if a batch of goods needs to be delivered within 3 hours, but forecasts indicate a potential 1-hour delay during peak periods, the system will schedule the delivery task 1 hour in advance. If the time window requirement still cannot be met, resources will be adjusted again, such as temporarily adding two delivery vehicles, ultimately forming a collaborative scheduling solution. This time allocation method ensures timely task completion and improves service reliability.

[0082] Step S540: Based on the final collaborative scheduling scheme, obtain dynamic data on inventory status and route costs, continuously monitor the efficiency of the second delivery and the effect of collaborative scheduling through the data integration platform, and obtain optimized cross-link scheduling instructions.

[0083] For the final collaborative scheduling solution, the system continuously monitors the dynamic changes in inventory status and route costs through a data integration platform. Suppose that on a certain day, order volume suddenly surges to 2200 orders, route costs increase by 12%, and delivery efficiency decreases by 8%. The system will promptly adjust the plan and optimize cross-stage scheduling instructions to ensure maximum resource utilization. This continuous monitoring mechanism can dynamically adapt to business fluctuations and maintain the stability of the logistics system.

[0084] Furthermore, in the production and logistics management method based on digital twins provided in this embodiment, step S600 includes: Step S610: According to the collaborative scheduling instruction, the instruction is decomposed by a distributed computing tool to generate multiple specific task units. The task priority and execution sequence are initially sorted to obtain the decomposed task list.

[0085] In the collaborative scheduling and task allocation scenarios of the logistics field, the application of distributed computing tools can break down complex instructions into multiple executable task units. Suppose a logistics center receives a delivery instruction covering the entire region, involving the delivery of 2000 orders. The system first uses distributed computing tools to decompose the instruction into 10 task units divided by region, with each unit responsible for approximately 200 delivery tasks. During initial sorting, the system generates a task list based on task priority (e.g., prioritizing urgent orders) and execution sequence (e.g., prioritizing deliveries to closer areas). This decomposition method refines large tasks into actionable units, facilitating subsequent allocation and management.

[0086] Step S620: Based on the task list, and combined with the equipment load and node capacity data of production equipment and logistics nodes, tasks are allocated using a resource allocation tool. If the equipment load exceeds the preset threshold, the task allocation order is adjusted to determine the final task assignment scheme.

[0087] For resource allocation of the task list, the system combines the load and capacity data of production equipment and logistics nodes to assign tasks. Suppose a logistics node has 5 delivery vehicles, each with a maximum daily load of 400 orders, while the task list allocates 2200 delivery orders to this node, significantly exceeding the load threshold. The resource allocation tool will adjust the task allocation order, transferring the excess 200 orders to neighboring nodes, while considering that neighboring nodes have a vehicle capacity of 800 orders, ensuring task balance. This adjustment avoids equipment overload and ensures a smooth delivery process.

[0088] Step S630: According to the task assignment plan, specific tasks are issued to production equipment and logistics nodes, and status feedback data during the execution process is collected in real time. The execution efficiency is monitored by the feedback frequency to obtain dynamic updates of the execution status.

[0089] The following formula is used to assign specific tasks to production equipment and logistics nodes: (1) In formula (1), Indicates to the first The time for assigning tasks to each production equipment or logistics node. This indicates the total number of tasks. Indicates the first The workload of each task Indicates the first The priority weight of each task. Indicates the first The processing capacity of each node Indicates the first The availability coefficient of each node.

[0090] Monitoring execution efficiency through feedback frequency is achieved using the following formula: (2) In formula (2), This represents the performance monitoring metrics. Indicates the actual feedback frequency. Indicates the expected feedback frequency. The variance representing the feedback frequency. It represents the square of the mean of the feedback frequency.

[0091] After the task assignment plan is issued, the system collects status feedback data in real time during the execution process and monitors execution efficiency through the feedback frequency. Suppose the feedback frequency for delivery tasks in a certain area is set to update every 30 minutes. If the system detects that a vehicle's delivery efficiency is lower than the preset 20 orders per hour, reaching only 15 orders per hour, the dynamic update indicates that the vehicle may be delayed due to road conditions. The system records this status and provides a basis for subsequent adjustments. This real-time monitoring mechanism helps to identify problems promptly and ensures the controllability of task execution.

[0092] Step S640: Regarding the dynamic update of the execution status, if the execution efficiency is lower than the preset threshold, the task allocation is optimized through the instruction adjustment tool. The execution sequence is re-planned in combination with the device load and node capacity data to determine the adjusted task execution arrangement.

[0093] The adjusted task execution schedule is derived using the following formula: (3) In formula (3), This indicates the optimized task execution sequence. Represents the set of all possible timing schemes. Indicates the total number of tasks. Indicates task Priority weights, Indicates task Execution time, This represents the delay penalty coefficient. Indicates task The actual completion time, Indicates task The deadline.

[0094] When execution efficiency falls below a preset threshold, the instruction adjustment tool intervenes to optimize task allocation. For example, if a node experiences reduced delivery efficiency due to peak-hour traffic congestion, the system, combining equipment load and node capacity data, reschedules the execution timeline, shifting some tasks to nighttime deliveries and adding a backup vehicle. The adjusted schedule shows that the completion rate of nighttime delivery tasks has improved to the preset standard. This optimization method can flexibly respond to unforeseen circumstances, ensuring the smooth progress of the overall delivery mission.

[0095] Furthermore, in the production and logistics management method based on digital twins provided in this embodiment, step S700 includes: Step S710: Obtain execution data of status feedback from production nodes and logistics nodes through data acquisition tools, perform structured classification on the execution data, and obtain a list of classified execution data.

[0096] The application of data acquisition tools obtains real-time execution data from production and logistics nodes, providing a foundation for subsequent analysis. This execution data is then structured and categorized into groups such as delivery progress, vehicle load, and node capacity, forming a clear list of execution data. This classification method facilitates rapid identification of problem areas, laying the foundation for dynamic monitoring.

[0097] Step S720: Use a real-time update tool to dynamically monitor the execution data list. If the execution efficiency is lower than the preset threshold, use a parameter adjustment tool to correct the parameters of the simulation intervention mechanism and determine the adjusted parameter configuration.

[0098] The system execution efficiency is calculated in real time using the following formula and compared with a preset threshold: (4) In formula (4), Indicates the system during a time period Internal execution efficiency This indicates the total number of sampling points within the monitoring time window. Indicates the first The output quantity of each sampling point Indicates the first Input resource quantity per sampling point Indicates the first The execution time interval for each sampling point.

[0099] The adjusted parameter configuration is calculated using the following formula: (5) In formula (5), This indicates the adjusted new parameter configuration. This indicates the current parameter configuration. This represents the learning rate used to adjust the parameters. Describe the objective function For parameters gradient, This represents the preset efficiency threshold. This represents the actual efficiency value currently monitored.

[0100] For dynamic monitoring of the execution data list, the real-time update tool refreshes the data at regular intervals to ensure managers are aware of the latest situation. Suppose a logistics center is responsible for delivering 1000 orders. The system monitors and finds that the delivery efficiency in a certain area is 10 orders per hour, lower than the preset threshold of 15 orders per hour. At this point, the parameter adjustment tool intervenes to correct the parameters of the simulated intervention mechanism. For example, the priority of delivery routes is adjusted to avoid peak hours, or temporary vehicle support is added. The adjusted parameter configuration shows that the efficiency in that area has improved to 14 orders per hour, close to the expected target. The corrected parameters of the simulated intervention mechanism are dynamically adjusted parameters set to address issues such as "intervention effects not meeting expectations" and "model-entity mismatch." Their core function is to give the intervention mechanism "self-calibration" capabilities, continuously correcting the intervention logic, strategies, or thresholds to narrow the gap between simulated intervention and the actual response of the physical entity. The parameters of the modified simulation intervention mechanism need to be specifically designed based on the causes of deviations (such as model prediction errors, actuator lag, and environmental disturbances). These parameters include deviation diagnosis parameters, adjustment magnitude parameters, iterative correction parameters, constraint adaptation parameters, and effect verification parameters. When the intervention effect is unsatisfactory, the problem is first located using the deviation diagnosis parameters, then the adjustment magnitude parameters are used to quantify the correction intensity, the adjustment rhythm is set according to the iterative correction parameters, and the correction is executed under the constraints of the constraint adaptation parameters. Finally, the effectiveness of the correction is determined by the effect verification parameters. This process allows the digital twin model's intervention mechanism to dynamically adapt to the aging of physical entities, environmental changes, or sudden disturbances, continuously improving the accuracy of "virtual-real interaction."

[0101] Step S730: Continuously track parameter configurations using a feedback frequency monitoring tool to obtain performance data under different business loads.

[0102] With the help of specialized monitoring tools (such as real-time data acquisition systems, load pressure sensors, parameter log analysis platforms, etc.), the core parameter configurations of digital twin models, production scheduling systems or other complex collaborative systems are continuously tracked and recorded, and then the actual operating performance data of these parameters under different business load conditions (such as low load, medium load, high load, sudden peak load, etc.) are extracted (such as response speed, error rate, stability, resource utilization, etc.).

[0103] Regarding the continuous tracking of the feedback frequency monitoring tool, the system is set to update the performance data of the parameter configuration every 20 minutes to observe its adaptability under different business loads. During peak business periods, the load of a certain logistics node reached 2,000 orders per day, far exceeding the normal processing capacity of 1,500 orders. The feedback frequency monitoring tool revealed that the adjusted parameter configuration could still maintain delivery efficiency under high load, indicating that the parameter optimization has strong adaptability. This continuous tracking helps to verify the adjustment effect and provides data support for subsequent optimizations.

[0104] Step S740: Use a data comparison tool to compare the performance data with the preset accuracy optimization target. If the preset threshold is not reached, use a data backtracking tool to organize historical status feedback and obtain the updated data basis.

[0105] The data foundation is updated by integrating historical status feedback through data backtracking tools. The updated data foundation is shown below: (6) In formula (6), This indicates the updated basic data state. This represents the weight for maintaining the current state. Indicates the current data status. This represents the total number of historical states. Indicates the first The weight of each historical state Indicates the first Historical status feedback data.

[0106] The data comparison tool analyzes whether the adjusted efficiency has met expectations by comparing the performance data with the preset accuracy optimization targets. If the delivery completion rate target for a certain node is 95%, but the actual rate is only 88%, then the preset threshold has not been reached. In this case, the data backtracking tool will organize historical status feedback from the past 7 days to analyze the reasons for the low efficiency, such as discovering that a certain vehicle is frequently affected by traffic delays. Based on this, the system updates the data foundation to provide a reference for further optimization.

[0107] Please see Figure 2This embodiment provides a digital twin-based production and logistics management system for executing the aforementioned digital twin-based production and logistics management method. The digital twin-based production and logistics management system includes a real-time dataset acquisition module 10, an anomaly detection module 20, a dynamic trend acquisition module 30, a plan determination module 40, a collaborative scheduling instruction generation module 50, a status feedback acquisition module 60, and a precision optimization module 70. The real-time dataset acquisition module 10 acquires real-time data streams from production equipment sensors and logistics nodes, and uses streaming processing technology to clean and format the real-time data streams to obtain a structured real-time dataset. The anomaly detection module 20 applies time series analysis algorithms to the structured real-time dataset to detect equipment anomalies and order fluctuations. The dynamic trend acquisition module 30 is used to detect equipment anomalies... In the event of order fluctuations, a pre-established digital twin model is used to simulate the production and logistics processes in real time, obtaining the dynamic trend of the event's impact. The planning determination module 40 is used to optimize and adjust production scheduling and logistics routes based on the dynamic trend using reinforcement learning algorithms, determining the optimized production and transportation plan. The collaborative scheduling instruction generation module 50 is used to extract key decision parameters from the optimized production and transportation plan, and generate cross-process collaborative scheduling instructions in conjunction with the full-chain data integration platform. The status feedback acquisition module 60 is used to decompose the collaborative scheduling instructions into specific tasks through a distributed computing framework, allocate them to production equipment and logistics nodes, and obtain status feedback after execution. The accuracy optimization module 70 is used to update the parameters of the digital twin model based on the status feedback using online learning algorithms, optimizing the accuracy of subsequent simulation intervention mechanisms.

[0108] The production and logistics management method and system based on digital twins provided in this embodiment, compared with existing technologies, detects anomalies through real-time data stream processing and time series analysis, simulates the impact of events using a digital twin model, optimizes production scheduling and logistics routes using reinforcement learning algorithms, generates cross-stage collaborative scheduling instructions and decomposes them for execution, and finally applies online learning to update model parameters. This embodiment achieves intelligent collaboration between production and logistics, enabling rapid response to anomalies, dynamic optimization of scheduling decisions, and improved overall operational efficiency. Through data-driven and artificial intelligence technologies, it constructs an adaptive, closed-loop optimized production and logistics collaborative system, providing an effective solution for manufacturing enterprises to achieve intelligent transformation.

[0109] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A production and logistics management method based on digital twins, characterized in that, Includes the following steps: Real-time data streams are acquired from production equipment sensors and logistics nodes, and the real-time data streams are cleaned and formatted using streaming processing technology to obtain a structured real-time dataset. For structured real-time datasets, time series analysis algorithms are applied to detect equipment anomalies and order fluctuations; If an abnormality in the equipment or a fluctuation in orders is detected, the production and logistics processes are simulated in real time using a pre-established digital twin model to obtain the dynamic trend of the event's impact. Based on the aforementioned dynamic trends, a reinforcement learning algorithm is used to optimize and adjust production scheduling and logistics routes, determining the optimized production and transportation plan, specifically including: Based on the dynamic trend, real-time updated data is obtained from order demand and inventory status. The resource allocation is compared with the pre-established database. If the resource allocation cannot meet the order demand, the adjustment process is triggered to determine the preliminary production scheduling plan. For the preliminary production scheduling plan, relevant information on logistics routes and route costs is obtained. The first delivery efficiency is evaluated using a route planning tool. If the first delivery efficiency is lower than a preset threshold, the logistics route is replanned to obtain an adjusted transportation plan. By adjusting the transportation plan, combining the first time window and demand forecast data, and using scheduling management tools to allocate time for production and delivery, if the delivery task cannot be completed within the first time window, the resource allocation will be adjusted a second time to determine the final production scheduling result. Based on the final production scheduling results, data on changes in inventory status and order demand are obtained. Data comparison tools are used to continuously monitor the first delivery efficiency and route cost to determine whether the optimized transportation plan meets the expected goals. Key decision parameters are extracted from the optimized production and transportation plans, and combined with the end-to-end data integration platform to generate cross-stage collaborative scheduling instructions, specifically including: Real-time data on order demand and inventory status are obtained from the optimized production and transportation plan. The resource allocation is compared with the pre-established database. If the resource allocation cannot meet the order demand, the data integration process is triggered to obtain the initial collaborative scheduling parameters. Based on the preliminary collaborative scheduling parameters, relevant data of the logistics path and the second time window are obtained. The second delivery efficiency is evaluated using a path planning tool. If the second delivery efficiency is lower than the preset threshold, the logistics path is replanned, and the adjusted collaborative scheduling instructions are determined. By adjusting the collaborative scheduling instructions, combining demand forecasting and monitoring frequency data, and using scheduling management tools to allocate time for production scheduling and delivery, if the delivery task cannot be completed within the second time window, the resource allocation will be adjusted a second time to determine the final collaborative scheduling scheme. Based on the final collaborative scheduling scheme, dynamic data on inventory status and route costs are obtained. The second delivery efficiency and collaborative scheduling effect are continuously monitored through the data integration platform to obtain optimized cross-link collaborative scheduling instructions. For the aforementioned collaborative scheduling instruction, the instruction is decomposed into specific tasks through a distributed computing framework, allocated to production equipment and logistics nodes, and status feedback is obtained after execution. Based on the status feedback, the parameters of the digital twin model are updated using an online learning algorithm to optimize the accuracy of subsequent simulation intervention mechanisms.

2. The production and logistics management method based on digital twins as described in claim 1, characterized in that, The steps of acquiring real-time data streams from production equipment sensors and logistics nodes, and cleaning and formatting the real-time data streams using streaming processing technology to obtain structured real-time datasets include: Real-time data streams are acquired from sensors and logistics nodes. A message queue tool is used to receive the time-series data from the sensors and the tracking data from the logistics nodes to obtain the raw data stream. For the original data stream, a streaming processing tool is used to clean it. If there are missing or outlier values ​​in the original data stream, a preset threshold is used to determine whether to fill in the mean or delete them, so as to obtain a cleaned data stream. Based on the cleaned data stream, a data transformation tool is used to perform a formatting operation, unifying the cleaned data stream into a pre-established structured data model to obtain structured data; The structured data is stored and integrated using database tools. If the structured data meets the preset time window requirements, a real-time dataset is generated, thus obtaining a real-time dataset.

3. The production and logistics management method based on digital twins as described in claim 1, characterized in that, The steps for detecting equipment anomalies and order fluctuations using time series analysis algorithms on structured real-time datasets include: Device status and operating parameter information are obtained from structured real-time datasets. The operating parameters are compared item by item using pre-established comparison rules. If the operating parameter exceeds the preset threshold range, it is marked as an abnormal signal, and data records with abnormal markings are obtained. For data records marked with anomalies, acquire the anomaly signal and fault type information, compare it with historical trend data, determine whether the anomaly signal conforms to a known fault type pattern, and determine the specific fault classification result. Based on the fault classification results, combined with the order quantity and fluctuation range information, the data with anomaly markers are segmented using a time window division method. If the fluctuation range exceeds the preset threshold range, it is marked as an order anomaly, thus obtaining an abnormal fluctuation identifier. In response to the abnormal fluctuation indicators and real-time monitoring requirements, alarm thresholds and business impact information are obtained. Data with abnormal markers is comprehensively evaluated through preset logical rules to determine whether an alarm mechanism is triggered, and the final alarm trigger record is obtained.

4. The production and logistics management method based on digital twins as described in claim 1, characterized in that, For the aforementioned collaborative scheduling instruction, the steps of decomposing the instruction into specific tasks using a distributed computing framework, allocating them to production equipment and logistics nodes, and obtaining post-execution status feedback include: According to the collaborative scheduling instruction, the instruction is decomposed by a distributed computing tool to generate multiple specific task units. The tasks are initially sorted according to their priority and execution order to obtain a decomposed task list. Based on the task list, and combined with the equipment load and node capacity data of production equipment and logistics nodes, tasks are allocated using a resource allocation tool. If the equipment load exceeds a preset threshold, the task allocation order is adjusted to determine the final task assignment scheme. According to the task assignment scheme, specific tasks are assigned to production equipment and logistics nodes, and status feedback data during the execution process is collected in real time. The execution efficiency is monitored by the feedback frequency to obtain dynamic updates of the execution status. In response to the dynamic updates of the execution status, if the execution efficiency is lower than the preset threshold, the task allocation is optimized through the instruction adjustment tool. The execution sequence is re-planned based on the device load and node capacity data, and the adjusted task execution arrangement is determined.

5. The production and logistics management method based on digital twins as described in claim 1, characterized in that, Based on the aforementioned status feedback, the steps for updating the parameters of the digital twin model using an online learning algorithm and optimizing the accuracy of subsequent simulation intervention mechanisms include: The execution data of the status feedback is obtained from the production equipment and logistics nodes through data acquisition tools, and the execution data is structured and classified to obtain a list of classified execution data. The execution data list is dynamically monitored using a real-time update tool. If the execution efficiency is lower than a preset threshold, the parameters of the simulation intervention mechanism are corrected using a parameter adjustment tool to determine the adjusted parameter configuration. The parameter configuration is continuously tracked using a feedback frequency monitoring tool to obtain performance data under different business loads. The performance data is compared with the preset accuracy optimization target using a data comparison tool. If the preset threshold is not reached, the historical status feedback is sorted out by executing a data backtracking tool to obtain the updated data basis.

6. The production and logistics management method based on digital twins as described in claim 5, characterized in that, In the step of using a real-time update tool to dynamically monitor the execution data list, and if the execution efficiency is lower than a preset threshold, adjusting the parameters of the simulation intervention mechanism using a parameter adjustment tool to determine the adjusted parameter configuration, the system execution efficiency is calculated in real time using the following formula and compared with the preset threshold: ； in, Indicates the system during a time period Internal execution efficiency This indicates the total number of sampling points within the monitoring time window. Indicates the first The output quantity of each sampling point Indicates the first Input resource quantity per sampling point Indicates the first The execution time interval for each sampling point.

7. A production and logistics management system based on digital twins, used to execute the production and logistics management method based on digital twins as described in any one of claims 1 to 6, characterized in that, The digital twin-based production and logistics management system includes: The real-time dataset acquisition module (10) is used to acquire real-time data streams from production equipment sensors and logistics nodes, and to clean and format the real-time data streams using streaming processing technology to obtain structured real-time datasets. The anomaly detection module (20) is used to detect equipment anomalies and order fluctuations by applying time series analysis algorithms to structured real-time datasets; The dynamic change trend acquisition module (30) is used to simulate the production and logistics links in real time through a pre-established digital twin model if the equipment abnormality or order fluctuation is detected, so as to obtain the dynamic change trend of the event impact. The planning determination module (40) is used to optimize and adjust the production scheduling and logistics path according to the dynamic change trend, and determine the optimized production and transportation plan; The collaborative scheduling instruction generation module (50) is used to extract key decision parameters from the optimized production and transportation plan, and generate cross-link collaborative scheduling instructions in conjunction with the full-chain data integration platform. The status feedback acquisition module (60) is used to decompose the collaborative scheduling instruction into specific tasks through a distributed computing framework, allocate them to production equipment and logistics nodes, and acquire status feedback after execution. The accuracy optimization module (70) is used to update the parameters of the digital twin model by applying an online learning algorithm based on the state feedback, thereby optimizing the accuracy of the subsequent simulation intervention mechanism.

Images