Intelligent optimization scheduling method and system in enterprise resource planning

Abstract

The invention discloses an intelligent optimization scheduling method and system in enterprise resource planning, and relates to the technical field of enterprise resource planning. According to the invention, a multi-source heterogeneous data acquisition technology is used to comprehensively capture real-time data in links of production, logistics and the like, the limitation of traditional data acquisition is broken through, and a comprehensive and timely analysis foundation is laid; through data twinborn body construction, an actual state is accurately mapped to a virtual environment, real-time simulation monitoring of a complex dynamic scene is realized, and the problem that global dynamics is difficult to grasp in a traditional method is solved; by means of deep reinforcement learning and other technologies, a scheduling strategy is optimized in a virtual environment, the defect that a traditional model is poor in adaptability is overcome, the problems that a traditional method is poor in complex dynamic environment coping capacity, data and decision are separated, and continuous optimization capacity is lacked are solved, the scheduling response speed and resource configuration efficiency are improved, and the method is suitable for large-scale popularization and application. And continuous and efficient operation of enterprise resource planning in a dynamic environment is comprehensively guaranteed.

Description

Technical Field

[0001] This invention relates to the field of enterprise resource planning technology, specifically to an intelligent optimization scheduling method and system in enterprise resource planning. Background Technology

[0002] Intelligent optimization and scheduling in enterprise resource planning (ERP) is a key area for improving the operational efficiency and competitiveness of modern enterprises. Its core lies in achieving efficient resource allocation and dynamic adjustment through data-driven decision-making, ensuring the adaptability and high responsiveness of production, logistics, and other processes. Traditional methods often rely on static rules or single optimization models, which are ill-suited to complex and dynamic environments. Especially when facing changing production demands and unforeseen events, the lack of real-time and global perspectives often leads to low scheduling efficiency or resource waste. Furthermore, existing solutions suffer from fragmentation in data integration and decision-making deployment, hindering seamless integration from global planning to local execution and limiting system flexibility and responsiveness. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent optimization scheduling method and system for enterprise resource planning, which solves the problems of traditional enterprise resource planning and scheduling, significantly improves scheduling efficiency, system flexibility and stability, and ensures efficient operation of enterprises in dynamic environments.

[0004] The objective of this invention can be achieved through the following technical solutions: This application provides an intelligent optimization scheduling method for enterprise resource planning, including the following steps: By using a multi-source heterogeneous data acquisition system, real-time data streams of equipment status, inventory information, and external environmental changes in the production and logistics processes of an enterprise are obtained to acquire an initial data set. Based on the initial dataset, data cleaning and fusion techniques are used to denoise and unify the format of the collected multi-source data, generate standardized data sets, and determine the basic input content for constructing the data twin. Based on standardized data sets, a high-precision data twin model is constructed. Through real-time mapping technology, the status of production and logistics links is dynamically reflected in the virtual environment, resulting in real-time updated virtual mapping results. Based on the real-time updated virtual mapping results, a deep reinforcement learning algorithm is applied to simulate and train the resource allocation scenario in the virtual environment to obtain optimized scheduling strategy parameters. Deploy a lightweight decision-making module on edge computing devices, and decompose global policies into locally executable task instructions through a parameter distribution mechanism to determine the basis for local decision execution; Based on locally executable task instructions, real-time decision-making logic runs on edge devices. If a sudden change in the production or logistics process is detected, a preset threshold comparison mechanism is used to determine whether to adjust the local resource configuration scheme. Based on the adjusted local resource configuration scheme, the local execution results are uploaded to the cloud data twin model through the data synchronization mechanism between edge devices and the cloud to obtain updated feedback on the global resource status.

[0005] Furthermore, the initial data set is obtained, specifically including: By integrating multi-source and heterogeneous data, the data acquisition system captures the data flow of the production and logistics links in real time, and then classifies and processes the equipment status and inventory information to determine their respective data feature distribution. Abnormal data features of equipment status are marked as abnormal. Combined with data streams of environmental changes, the support vector machine algorithm is used to analyze the abnormal data and determine potential abnormal patterns. Based on the results of the abnormal patterns, the data flow of inventory information is compared for correlation. Then, combined with real-time data from the production and logistics processes, the key influencing factors in the construction of the data twin are determined, and an initial data set is obtained.

[0006] Furthermore, a standardized data set is generated to determine the basic input content for constructing the data twin, specifically including: By analyzing the initial dataset, data preprocessing tools are used to initially screen the multi-source data, then data cleaning techniques are used to identify noisy data, and smoothing techniques are used to remove noise, resulting in a cleaned dataset. A format conversion tool was used to unify the structure of data from different sources to obtain a dataset with a consistent format. Data fusion methods were then used to integrate the features of multi-source data to determine the fused dataset. A standardized dataset is constructed, and data fields are normalized using pre-established mapping rules to obtain standardized data content. Then, for the data twin construction requirements, logical verification tools are used to check the data integrity. The missing fields are supplemented by interpolation methods to determine the final usable input dataset and generate the basic input content of the data twin. The random forest algorithm is used to rank the data features by importance and determine the key fields in the construction process.

[0007] Furthermore, the real-time updated virtual mapping results are obtained, specifically including: Based on standardized data sets, a high-precision model is constructed using data twin technology. The data is mapped to a virtual environment, and a preliminary virtual mapping framework is determined. Through real-time mapping technology, dynamic status information of the production and logistics links is obtained, resulting in a real-time collected status dataset. The data update mechanism is triggered based on the dynamic state information of the state dataset to adjust the mapping parameters in the virtual environment, determine the updated mapping state, and use the support vector machine algorithm to classify the dynamic state information and determine the distribution characteristics of abnormal and normal states. By using data fusion technology, the classification results are synchronized with the virtual environment to obtain real-time updated mapping results. If there is a deviation between the mapping results and the actual dynamic information, the model parameters are adjusted through a feedback mechanism to obtain an optimized virtual mapping output.

[0008] Furthermore, the optimized scheduling strategy parameters are obtained, specifically including: By collecting and organizing virtual mapping data, dynamically changing environmental information is obtained from the real-time updated data stream, the resource distribution status in the virtual environment is determined, and a deep reinforcement learning algorithm is used to conduct a preliminary analysis of the resource allocation scenario in the virtual environment to obtain an initial strategy model for resource allocation. By combining the dynamic characteristics of the environment, strategy simulation training is carried out, the response efficiency of resource allocation during the training process is judged, and then simulation training is carried out again to obtain an improved strategy model. Based on the constraints of the scheduling strategy, the resource allocation in the virtual environment is iteratively optimized multiple times to determine the optimized scheduling strategy parameters. The adaptability of the scheduling strategy is verified for different environmental scenarios in the virtual environment to obtain the final scheduling strategy scheme. The system continuously monitors and processes feedback on the real-time updated virtual mapping data to determine whether the resource configuration meets dynamic requirements, obtains new optimization parameters, and tracks and adjusts the resource configuration in the virtual environment over a long period of time by combining the results of continuous monitoring and feedback with empirical data from simulation training.

[0009] Furthermore, the global strategy is decomposed into locally executable task instructions, and the execution basis for local decisions is determined, specifically including: By deploying a lightweight decision-making module on an edge computing device, relevant data of the global policy is obtained and preliminarily analyzed to determine the part of the global policy related to the local task. Based on the analyzed global policy data, a parameter delivery mechanism is used to decompose it into specific task instructions. The decomposed task instructions are formatted to obtain an instruction set suitable for local decision-making, and the local task execution environment information on the edge computing device is obtained to determine the execution basis after adaptation. The system obtains the optimization requirements of the scheduling strategy parameters and adjusts the parameters to obtain the adjusted strategy parameter set. Then, it obtains the running status data of the local decision module. If the running status data exceeds the preset range, it uses a pre-established logistic regression model to perform state correction and judges the running stability after correction. Based on the corrected operational stability data, real-time load information of the deployed equipment is obtained, the load information is dynamically allocated, and the path to achieve the final optimization goal is determined.

[0010] Further, determine whether to adjust the local resource configuration scheme, specifically including: By using sensor modules on edge devices, operational data from the production and logistics processes are continuously collected. The collected data streams are initially filtered, and an anomaly flag is output using a preset threshold comparison mechanism. Obtain the current configuration of local resources, analyze resource usage in conjunction with the running logic, determine whether there is insufficient or uneven resource allocation, and call the real-time decision module to determine whether to trigger an adjustment strategy by analyzing resource usage. If resources are unevenly or insufficiently allocated, an optimized configuration scheme is generated, the allocation ratio of local resources is automatically adjusted, the operating logic of edge devices is updated synchronously, the adjusted operating status is obtained, and data changes in the production and logistics processes are continuously monitored. The comparison mechanism determines whether the adjustment meets the expected results, outputs monitoring results, records the execution of the adjustment strategy, stores it in the local database of the edge device, and determines the final operational stability.

[0011] Furthermore, obtain updates on the global resource status, specifically including: According to the adjusted local resource configuration scheme, a synchronization mechanism is used to establish a connection with the cloud, complete the data transmission process, determine the data content to be transmitted to the cloud, and if data loss or abnormality is detected during the transmission process, a retransmission mechanism is triggered to verify and complete the abnormal data and obtain the complete cloud data record. By recording data in the cloud, the data twin model is updated, and a consistency check is performed on the updated model to determine whether the model reflects the global resource status. The parameters of the data twin model are verified to obtain state information that conforms to reality. Based on the adjusted status information, a global resource update feedback is generated. The feedback content is formatted to determine the final structure of the feedback data. The feedback data is then transmitted back to the edge device through a synchronization mechanism to obtain the basis for optimizing the local resource configuration.

[0012] Furthermore, after obtaining the update feedback of the global resource state, the process also includes: re-optimizing the training parameters of the deep reinforcement learning algorithm, and iterating the policy on the new state data to obtain a scheduling policy output that is more adapted to the current scenario.

[0013] This application provides an intelligent optimization scheduling system for enterprise resource planning (ERP), which implements intelligent optimization scheduling methods in ERP, including: The multi-source heterogeneous data acquisition module collects real-time data on equipment operating status, inventory changes, and external environment through sensors and interfaces deployed in the production and logistics processes. It supports the access and preliminary processing of multi-source heterogeneous data streams, identifies abnormal data, and marks it. The data cleaning and fusion module performs noise reduction, format unification and structure transformation on the collected raw data, handles data conflicts and fuses multi-source features to generate a standardized dataset, and fills in missing values ​​through logical verification. The data twin construction module, based on standardized data, builds a high-precision virtual model of the enterprise's operation process, maps the production and logistics status in real time, dynamically updates virtual environment parameters, identifies abnormal states, and continuously optimizes the model's accuracy through a feedback mechanism. The deep reinforcement learning optimization module simulates resource allocation scenarios in a virtual environment, trains scheduling strategies using deep reinforcement learning algorithms, iterates and optimizes strategies in conjunction with dynamic changes in the environment, and generates scheduling parameters that adapt to different scenarios. The edge computing deployment module decomposes the global optimization strategy into locally executable task instructions through a parameter distribution mechanism, deploys them on the edge computing device, and adapts and corrects them in combination with the local operating environment to ensure the effective execution of the strategy and the reasonable allocation of resources on the edge node. The local real-time decision-making module runs real-time decision-making logic on edge devices, continuously monitors changes in production and logistics status, identifies abnormal events through threshold comparison mechanisms, dynamically adjusts local resource allocation schemes, and feeds back the execution results to the system. The cloud synchronization and feedback module uploads local execution results to the cloud data twin through the edge and cloud data synchronization mechanism, updates the global resource status, detects model consistency and performs parameter correction, generates global feedback information and sends it back to the edge device; The strategy iteration and update module analyzes resource usage trends based on global feedback data, dynamically adjusts the training parameters of deep reinforcement learning, continuously iterates the scheduling strategy to adapt to new scenarios, verifies the applicability of the strategy, and outputs the final scheduling scheme, forming a closed-loop optimized intelligent scheduling system.

[0014] The beneficial effects of this invention are as follows: By combining multi-source heterogeneous data acquisition, data twin construction, and deep reinforcement learning algorithms, the problem that traditional methods relying on static rules or single optimization models are difficult to cope with complex dynamic environments is effectively solved. Through the acquisition of real-time data streams and virtual mapping technology, the system can accurately capture the dynamic changes in production and logistics links. The strategy simulation training of deep reinforcement learning realizes the dynamic optimization of scheduling strategies, enabling enterprises to improve scheduling response speed and resource allocation efficiency when facing changing production demands and emergencies, and avoiding inefficiency and resource waste caused by insufficient real-time and global capabilities. By employing data cleaning and fusion technologies, lightweight decision-making modules on edge computing devices, and parameter distribution mechanisms, the existing solutions have broken down the disconnect between data integration and decision-making deployment. Standardized data processing ensures the accuracy of data twin construction, providing a reliable foundation for global planning. Global strategies are decomposed into locally executable task instructions, achieving seamless integration from global planning to local execution. This enhances the system's flexibility and response speed, enabling resource scheduling instructions to be quickly implemented and executed, thus improving the coherence and efficiency of the entire scheduling process. Through the data synchronization mechanism between edge devices and the cloud, the preset threshold comparison mechanism, and the strategy iteration and update module, a closed-loop optimized intelligent scheduling system is formed, which solves the problem of the lack of continuous optimization capability in traditional systems. The local execution results are fed back to the cloud in real time to update the global status. When abnormal changes occur, the resource configuration scheme can be adjusted in a timely manner. Moreover, the deep reinforcement learning parameters are continuously iterated, so that the scheduling strategy can continuously adapt to new scenarios, significantly improving the system's adaptability and long-term stability, and ensuring the continuous and efficient operation of enterprise resource planning in dynamic environments. Attached Figure Description

[0015] To better understand and implement this application, the technical solution is described in detail below with reference to the accompanying drawings.

[0016] Figure 1 A flowchart illustrating the intelligent optimization scheduling method in enterprise resource planning provided in Embodiment 1 of this application; Figure 2 A flowchart illustrating the process of obtaining real-time updated virtual mapping results using the intelligent optimization scheduling method in enterprise resource planning provided in Embodiment 1 of this application; Figure 3 A flowchart illustrating the process of obtaining optimized scheduling strategy parameters in the intelligent optimization scheduling method of enterprise resource planning provided in Embodiment 1 of this application; Figure 4 This is a schematic diagram of the structure of the intelligent optimization scheduling system in enterprise resource planning provided in Embodiment 2 of this application. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, exemplary embodiments will be described in detail below, examples of which are illustrated in the accompanying drawings. In the following description relating to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of methods and systems consistent with some aspects of this application as detailed in the appended claims.

[0018] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0019] The following detailed description of the specific implementation methods, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided in detail.

[0020] Example 1

[0021] Please see Figures 1-3 This embodiment provides an intelligent optimization scheduling method in enterprise resource planning, including the following steps: S1. Through a multi-source heterogeneous data acquisition system, real-time data streams of equipment status, inventory information and external environment changes in the production and logistics links of the enterprise are obtained to obtain an initial data set for subsequent processing to improve the accuracy of data twin construction. Furthermore, the initial data set is obtained, specifically including: By integrating multi-source and heterogeneous data, the data acquisition system captures the data flow of the production and logistics links in real time, and then classifies and processes the equipment status and inventory information to determine their respective data feature distribution. If the data characteristics of the device status exceed the preset threshold range, the data is marked as abnormal, the marked device status dataset is obtained, and the abnormal data is further analyzed by the support vector machine algorithm in combination with the data stream of environmental changes to determine potential abnormal patterns. Based on the results of the abnormal patterns, the data stream of inventory information is compared for correlation to obtain the correlation characteristics between inventory fluctuations and environmental changes. Then, combined with real-time data from the production and logistics links, the key influencing factors in the construction of the data twin are determined, and an initial data set is obtained.

[0022] Specifically, it addresses the problems of traditional data collection, such as single data sources, difficulty in integration, and insufficient anomaly identification leading to low data quality and inability to support the construction of data twins. Multi-source heterogeneous data collection comprehensively and accurately collects relevant data, improves data integrity and accuracy, eliminates interference from invalid or erroneous data, clarifies key factors for data twin construction, provides a high-quality foundation for its high-precision construction, and assists in subsequent intelligent optimization and scheduling.

[0023] S2. Based on the initial dataset, data cleaning and fusion techniques are used to denoise and unify the format of the collected multi-source data, generating standardized data sets and determining the basic input content for constructing the data twin.

[0024] Furthermore, a standardized data set is generated to determine the basic input content for constructing the data twin, specifically including: By analyzing the initial dataset, data preprocessing tools are used to initially screen the multi-source data, then data cleaning techniques are used to identify noisy data, and smoothing techniques are used to remove noise, resulting in a cleaned dataset. A format conversion tool is used to unify the structure of data from different sources to obtain a dataset with a consistent format. Data fusion methods are then used to integrate the features of multi-source data. If data conflicts are found during the fusion process, field selection is performed based on priority rules to determine the fused dataset. Based on the merged dataset, a standardized dataset is constructed, and the data fields are normalized using pre-established mapping rules to obtain standardized data content. Then, for the data twin construction requirements, logical verification tools are used to check the data integrity. If missing fields are found, they are supplemented using interpolation methods. The final usable input dataset is determined, and the basic input content of the data twin is generated. The random forest algorithm is used to rank the data features by importance and determine the key fields in the construction process.

[0025] Specifically, through data cleaning and fusion technologies, standardized data sets are generated by denoising, format unification, conflict resolution, and missing data completion. Key data fields are clearly defined, which solves the problems of poor data quality and low applicability. This provides a standardized, complete, and reliable basic input for the construction of data twins, ensuring the accuracy of the construction.

[0026] S3. Based on the standardized data set, construct a high-precision data twin model. Through real-time mapping technology, dynamically reflect the status of production and logistics links in the virtual environment to obtain real-time updated virtual mapping results.

[0027] Furthermore, the real-time updated virtual mapping results are obtained, specifically including: S31. Based on the standardized data set, a high-precision model is constructed using data twin technology. The data is mapped to the virtual environment, and a preliminary virtual mapping framework is determined. Through real-time mapping technology, dynamic status information of the production and logistics links is obtained, resulting in a real-time collected status dataset. S32. Trigger the data update mechanism based on the state dynamic information of the state dataset, adjust the mapping parameters in the virtual environment, judge the updated mapping state, and use the support vector machine algorithm to classify the state dynamic information and determine the distribution characteristics of abnormal and normal states. S33. Through data fusion technology, the classification results are synchronized with the virtual environment to obtain real-time updated mapping results. If there is a deviation between the mapping results and the actual dynamic information, the model parameters are adjusted through a feedback mechanism to obtain the optimized virtual mapping output.

[0028] Among them, the real-time mapping technology adopts an event-driven streaming data processing architecture. It triggers events through sensor data and pushes them to the virtual model in real time for status updates, thereby ensuring the synchronization and timeliness of the virtual environment with the actual production and logistics status.

[0029] Specifically, by constructing a high-precision data twin model and combining it with real-time mapping technology, the problems of disconnection between traditional virtual models and actual states and delayed updates are effectively solved, realizing dynamic synchronization and real-time feedback of production and logistics states. By introducing support vector machine algorithms to classify and identify state data, and combining data fusion and feedback mechanisms to continuously optimize model parameters, the accuracy and response speed of virtual mapping are significantly improved, thereby enhancing the system's ability to identify abnormal states and its adaptability to complex environments, providing a highly reliable virtual environment foundation for subsequent intelligent scheduling.

[0030] S4. Based on the real-time updated virtual mapping results, apply deep reinforcement learning algorithms to perform policy simulation training for resource configuration scenarios in the virtual environment, and obtain optimized scheduling policy parameters.

[0031] Furthermore, the optimized scheduling strategy parameters are obtained, specifically including: S41. By collecting and organizing virtual mapping data, dynamically changing environmental information is obtained from the real-time updated data stream, the resource distribution status in the virtual environment is determined, and a deep reinforcement learning algorithm is used to conduct a preliminary analysis of the resource allocation scenario in the virtual environment to obtain an initial strategy model for resource allocation. S42. Combining the dynamic characteristics of the environment, implement strategy simulation training, judge the response efficiency of resource allocation during the training process, and if the response efficiency is lower than the preset threshold, adjust the model parameters, re-perform simulation training, and obtain the improved strategy model. S43. Combining the constraints of the scheduling strategy, the resource allocation in the virtual environment is iteratively optimized multiple times to determine the optimized scheduling strategy parameters. The adaptability of different environmental scenarios in the virtual environment is verified. If the adaptability result does not meet the preset standard, the parameters are fine-tuned based on the data to obtain the final scheduling strategy scheme. S44. Adopt the final scheduling strategy scheme, continuously monitor and process the real-time updated virtual mapping data, determine whether the resource configuration meets the dynamic requirements, if not, trigger the strategy model update process, obtain new optimization parameters, and then, through the results of continuous monitoring and feedback processing, combined with the experience data of simulation training, track and adjust the resource configuration in the virtual environment over a long period of time to determine the stability and reliability of the resource configuration.

[0032] The initial resource allocation strategy model extracts real-time updated resource status data from the virtual environment, including key variables such as equipment load, inventory levels, and order demand, to construct a state space. Subsequently, it defines the action space for resource allocation, including operations such as task allocation, equipment scheduling, and path selection, and sets a multi-objective reward function with indicators such as resource utilization, response time, and cost. Based on this, deep reinforcement learning algorithms (such as DQN, PPO, or DDPG) are used to train the agent, enabling it to continuously learn the optimal strategy during interaction with the virtual environment, ultimately forming an initial resource allocation strategy model adapted to dynamic scenarios. The scheduling strategy is constrained by factors such as resource capacity, time window, task priority, process dependencies, and cost control. It iterative optimization of resource allocation in the virtual environment ensures that resource utilization is maximized and scheduling efficiency is optimized while meeting various business rules and operational constraints.

[0033] Specifically, by using deep reinforcement learning algorithms to simulate and train strategies in a virtual environment, the problem of poor adaptability and low response efficiency of traditional scheduling strategies in dynamic environments is effectively solved. It can dynamically adjust resource allocation strategies based on real-time updated virtual mapping data, and continuously optimize model parameters through continuous monitoring and feedback mechanisms, thereby significantly improving the flexibility, accuracy and stability of scheduling strategies, and thus achieving high efficiency and reliability in resource allocation.

[0034] S5. For the optimized scheduling strategy parameters, deploy a lightweight decision module on the edge computing device. Through the parameter distribution mechanism, decompose the global strategy into locally executable task instructions and determine the execution basis of the local decision.

[0035] Furthermore, the global strategy is decomposed into locally executable task instructions, and the execution basis for local decisions is determined, specifically including: By deploying a lightweight decision-making module on an edge computing device, relevant data of the global policy is obtained. The obtained data is initially analyzed to determine the part of the global policy that is related to the local task. Based on the analyzed global policy data, a parameter delivery mechanism is used to decompose it into specific task instructions. The decomposed task instructions are formatted to obtain an instruction set suitable for local decision-making. The local task execution environment information on the edge computing device is obtained. If the environment information meets the preset threshold conditions, the local decision-making module is triggered to adapt the task instructions and determine the execution basis after adaptation. The system obtains the optimization requirements of the scheduling strategy parameters and adjusts the parameters to obtain the adjusted strategy parameter set. Then, it obtains the running status data of the local decision module. If the running status data exceeds the preset range, it uses a pre-established logistic regression model to perform state correction and judges the running stability after correction. Based on the corrected operational stability data, real-time load information of the deployed equipment is obtained, the load information is dynamically allocated, and the path to achieve the final optimization goal is determined.

[0036] The pre-established logistic regression model for state correction is trained based on historical operating data and used to identify abnormal patterns in the operating status of edge devices. When the operating status data of the local decision module (such as CPU utilization, memory usage, task execution latency, etc.) exceeds a preset threshold range, the system inputs the status data into the logistic regression model. The model calculates the probability that the status belongs to normal or abnormal operation and outputs the status classification result. If it is judged to be an abnormal status, the system locates key influencing factors based on the feature weights output by the model and automatically adjusts the task scheduling strategy or resource allocation parameters to achieve real-time correction of the operating status, thereby ensuring the stable operation and efficient response of edge devices in complex environments.

[0037] Specifically, by deploying a lightweight decision-making module on edge computing devices and using a parameter distribution mechanism to decompose global scheduling strategies into locally executable task instructions, the system effectively solves the problems of difficult implementation of global strategies and lack of basis for local execution in traditional systems. Through the formatting and environment adaptation of task instructions, combined with real-time monitoring of running status and logical regression correction, the system can achieve efficient and stable local decision-making on resource-constrained edge devices. At the same time, dynamic allocation based on real-time load information further optimizes resource utilization and task execution efficiency, significantly improving the response speed and operational stability of the overall scheduling system.

[0038] S6. Based on locally executable task instructions, run real-time decision-making logic on edge devices. If a sudden change in the production or logistics process is detected, determine whether to adjust the local resource configuration scheme through a preset threshold comparison mechanism.

[0039] Further, determine whether to adjust the local resource configuration scheme, specifically including: By using sensor modules on edge devices, operational data from the production and logistics processes are continuously collected. The collected data stream is initially filtered to obtain basic status information. A preset threshold comparison mechanism is used to analyze whether there are any sudden changes. If the data exceeds the preset threshold, it is determined to be an abnormal state and an abnormality flag is output. Obtain the current configuration of local resources, analyze resource usage in conjunction with the running logic, determine whether there is insufficient or uneven resource allocation, and call the real-time decision module to determine whether to trigger an adjustment strategy by analyzing resource usage. If resources are unevenly or insufficiently allocated, an optimized configuration scheme is generated, the allocation ratio of local resources is automatically adjusted, the operating logic of edge devices is updated synchronously, the adjusted operating status is obtained, and data changes in the production and logistics processes are continuously monitored. The comparison mechanism determines whether the adjustment meets the expected results, outputs monitoring results, records the execution of the adjustment strategy, and stores it in the local database of the edge device for subsequent decision-making reference, thus determining the final operational stability.

[0040] Specifically, by running real-time decision-making logic on edge devices, combined with sensor data acquisition and threshold comparison mechanisms, the system effectively solves the problems of delayed identification of sudden changes and untimely resource allocation response in production or logistics processes. It can monitor operational status in real time, quickly identify abnormal events, and automatically trigger resource adjustment strategies to optimize local resource allocation, thereby improving the system's response speed and processing efficiency to emergencies. Simultaneously, through continuous monitoring and comparison of adjustment effects, the effectiveness and stability of resource allocation schemes are ensured, significantly enhancing the edge devices' adaptability and operational reliability in dynamic environments.

[0041] S7. Based on the adjusted local resource configuration scheme, the local execution results are uploaded to the cloud data twin model through the data synchronization mechanism between edge devices and the cloud to obtain the update feedback of the global resource status.

[0042] Furthermore, obtain updates on the global resource status, specifically including: According to the adjusted local resource configuration scheme, a synchronization mechanism is used to establish a connection with the cloud, complete the data transmission process, determine the data content to be transmitted to the cloud, and if data loss or abnormality is detected during the transmission process, a retransmission mechanism is triggered to verify and complete the abnormal data and obtain the complete cloud data record. By recording data in the cloud, the data twin model is updated, and a consistency check is performed on the updated model to determine whether the model reflects the global resource status. The parameters of the data twin model are verified to obtain state information that conforms to reality. Based on the adjusted status information, a global resource update feedback is generated. The feedback content is formatted to determine the final structure of the feedback data. The feedback data is then transmitted back to the edge device through a synchronization mechanism. The received feedback data is parsed to obtain the basis for optimizing local resource configuration.

[0043] Specifically, by using a data synchronization mechanism between edge devices and the cloud, the problem of delayed global state updates and data inconsistency after local resource configuration adjustments is solved. Local execution results can be reliably uploaded to the cloud data twin model, and data integrity is ensured through retransmission and verification mechanisms during data transmission, improving the stability and accuracy of data synchronization. At the same time, feedback information is generated based on the updated global resource state and sent back to the edge device, realizing real-time alignment and collaborative optimization between local and global states, significantly enhancing the synchronization of the overall system scheduling strategy and the effectiveness of global resource management.

[0044] Furthermore, after obtaining the update feedback of the global resource state, the process also includes: re-optimizing the training parameters of the deep reinforcement learning algorithm, and iterating the policy on the new state data to obtain a scheduling policy output that is more adapted to the current scenario.

[0045] Furthermore, a scheduling strategy more adapted to the current scenario is output, specifically including: By obtaining state feedback data from global resources, analyzing the changing trends of resource usage, determining the initial needs for resource allocation, adjusting the training parameters of the deep reinforcement learning algorithm based on the changing trends of resource allocation needs, optimizing the parameters of the current state data, and obtaining the optimized training configuration. With optimized training configuration, the strategy iteration of deep reinforcement learning algorithm is executed to process the current state data and determine the policy direction that is suitable for the scenario. If there is a deviation between the policy direction and the state feedback data, the scheduling strategy is adjusted by comparative analysis to obtain output results that better meet the resource adjustment requirements. Based on the adjusted scheduling strategy, the applicability of the strategy in different scenarios is determined by verifying the status update data. The verified scheduling strategy is then obtained and finally calibrated by combining the status feedback of global resources to obtain the final scheduling scheme applicable to the current scenario. Through the final scheduling scheme, resource adjustment operations are performed, global resource status data is updated, and the entire scheduling process is completed.

[0046] Specifically, by using a data synchronization mechanism between edge devices and the cloud, the problem of delayed global state updates and data inconsistency after local resource configuration adjustments is solved. Local execution results can be reliably uploaded to the cloud data twin model, and data integrity is ensured through retransmission and verification mechanisms during data transmission, improving the stability and accuracy of data synchronization. At the same time, feedback information is generated based on the updated global resource state and sent back to the edge device, realizing real-time alignment and collaborative optimization between local and global states, significantly enhancing the synchronization of the overall system scheduling strategy and the effectiveness of global resource management.

[0047] Example 2

[0048] Please see Figure 4 This embodiment provides an intelligent optimization scheduling system for enterprise resource planning (ERP), used to implement intelligent optimization scheduling methods in ERP, including: The multi-source heterogeneous data acquisition module collects real-time data on equipment operating status, inventory changes, and external environment through sensors and interfaces deployed in the production and logistics processes. It supports the access and preliminary processing of multi-source heterogeneous data streams, identifies and marks abnormal data, and provides the original data foundation for the subsequent construction of data twins. The data cleaning and fusion module removes noise, unifies formats and transforms structures in the collected raw data, handles data conflicts and fuses multi-source features to generate a standardized dataset, and ensures data integrity and consistency through logical verification and missing value imputation, providing high-quality input for the data twin. The data twin construction module, based on standardized data, builds a high-precision virtual model of the enterprise's operation process, maps the production and logistics status in real time, dynamically updates virtual environment parameters, identifies abnormal states, and continuously optimizes the model accuracy through a feedback mechanism to achieve virtual-real synchronization. The deep reinforcement learning optimization module simulates resource allocation scenarios in a virtual environment, trains scheduling strategies using deep reinforcement learning algorithms, iterates and optimizes strategies in conjunction with dynamic changes in the environment, and generates scheduling parameters that adapt to different scenarios, providing intelligent decision support for the system. The edge computing deployment module decomposes the global optimization strategy into locally executable task instructions through a parameter distribution mechanism, deploys them on the edge computing device, and adapts and corrects them in combination with the local operating environment to ensure the effective execution of the strategy and the reasonable allocation of resources on the edge node. The local real-time decision-making module runs real-time decision-making logic on edge devices, continuously monitors changes in production and logistics status, identifies abnormal events through threshold comparison mechanisms, dynamically adjusts local resource allocation schemes, and feeds back the execution results to the system to ensure the stability and responsiveness of local operation. The cloud synchronization and feedback module uploads local execution results to the cloud data twin through the edge and cloud data synchronization mechanism, updates the global resource status, detects model consistency and performs parameter correction, generates global feedback information and sends it back to the edge device, thereby achieving resource optimization from a global perspective. The strategy iteration and update module analyzes resource usage trends based on global feedback data, dynamically adjusts the training parameters of deep reinforcement learning, continuously iterates the scheduling strategy to adapt to new scenarios, verifies the applicability of the strategy, and outputs the final scheduling scheme, forming a closed-loop optimized intelligent scheduling system.

[0049] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An intelligent optimization scheduling method in enterprise resource planning, characterized by: Includes the following steps: By using a multi-source heterogeneous data acquisition system, real-time data streams of equipment status, inventory information, and external environmental changes in the production and logistics processes of an enterprise are obtained to acquire an initial data set. Based on the initial dataset, data cleaning and fusion techniques are used to denoise and unify the format of the collected multi-source data, generate standardized data sets, and determine the basic input content for constructing the data twin. Based on standardized data sets, a high-precision data twin model is constructed. Through real-time mapping technology, the status of production and logistics links is dynamically reflected in the virtual environment, resulting in real-time updated virtual mapping results. Based on the real-time updated virtual mapping results, a deep reinforcement learning algorithm is applied to simulate and train the resource allocation scenario in the virtual environment to obtain optimized scheduling strategy parameters. Deploy a lightweight decision-making module on edge computing devices, and decompose global policies into locally executable task instructions through a parameter distribution mechanism to determine the basis for local decision execution; Based on locally executable task instructions, real-time decision-making logic runs on edge devices. If a sudden change in the production or logistics process is detected, a preset threshold comparison mechanism is used to determine whether to adjust the local resource configuration scheme. Based on the adjusted local resource configuration scheme, the local execution results are uploaded to the cloud data twin model through the data synchronization mechanism between edge devices and the cloud to obtain updated feedback on the global resource status.

2. The intelligent optimization scheduling method in enterprise resource planning according to claim 1, characterized in that: The initial data set is obtained, specifically including: By integrating multi-source and heterogeneous data, the data acquisition system captures the data flow of the production and logistics links in real time, and then classifies and processes the equipment status and inventory information to determine their respective data feature distribution. Abnormal data features of equipment status are marked as abnormal. Combined with data streams of environmental changes, the support vector machine algorithm is used to analyze the abnormal data and determine potential abnormal patterns. Based on the results of the abnormal patterns, the data flow of inventory information is compared for correlation. Then, combined with real-time data from the production and logistics processes, the key influencing factors in the construction of the data twin are determined, and an initial data set is obtained.

3. The intelligent optimization scheduling method in enterprise resource planning according to claim 1, characterized in that: Generate standardized data sets to determine the basic inputs for constructing data twins, specifically including: By analyzing the initial dataset, data preprocessing tools are used to initially screen the multi-source data, then data cleaning techniques are used to identify noisy data, and smoothing techniques are used to remove noise, resulting in a cleaned dataset. A format conversion tool was used to unify the structure of data from different sources to obtain a dataset with a consistent format. Data fusion methods were then used to integrate the features of multi-source data to determine the fused dataset. A standardized dataset is constructed, and data fields are normalized using pre-established mapping rules to obtain standardized data content. Then, for the data twin construction requirements, logical verification tools are used to check the data integrity. The missing fields are supplemented by interpolation methods to determine the final usable input dataset and generate the basic input content of the data twin. The random forest algorithm is used to rank the data features by importance and determine the key fields in the construction process.

4. The intelligent optimization scheduling method in enterprise resource planning according to claim 1, characterized in that: The virtual mapping results are updated in real time, including: Based on standardized data sets, a high-precision model is constructed using data twin technology. The data is mapped to a virtual environment, and a preliminary virtual mapping framework is determined. Through real-time mapping technology, dynamic status information of the production and logistics links is obtained, resulting in a real-time collected status dataset. The data update mechanism is triggered based on the dynamic state information of the state dataset to adjust the mapping parameters in the virtual environment, determine the updated mapping state, and use the support vector machine algorithm to classify the dynamic state information and determine the distribution characteristics of abnormal and normal states. By using data fusion technology, the classification results are synchronized with the virtual environment to obtain real-time updated mapping results. If there is a deviation between the mapping results and the actual dynamic information, the model parameters are adjusted through a feedback mechanism to obtain an optimized virtual mapping output.

5. The intelligent optimization scheduling method in enterprise resource planning according to claim 1, characterized in that: Obtain the optimized scheduling policy parameters, specifically including: By collecting and organizing virtual mapping data, dynamically changing environmental information is obtained from the real-time updated data stream, the resource distribution status in the virtual environment is determined, and a deep reinforcement learning algorithm is used to conduct a preliminary analysis of the resource allocation scenario in the virtual environment to obtain an initial strategy model for resource allocation. By combining the dynamic characteristics of the environment, strategy simulation training is carried out, the response efficiency of resource allocation during the training process is judged, and then simulation training is carried out again to obtain an improved strategy model. Based on the constraints of the scheduling strategy, the resource allocation in the virtual environment is iteratively optimized multiple times to determine the optimized scheduling strategy parameters. The adaptability of the scheduling strategy is verified for different environmental scenarios in the virtual environment to obtain the final scheduling strategy scheme. The system continuously monitors and processes feedback on the real-time updated virtual mapping data to determine whether the resource configuration meets dynamic requirements, obtains new optimization parameters, and tracks and adjusts the resource configuration in the virtual environment over a long period of time by combining the results of continuous monitoring and feedback with empirical data from simulation training.

6. The intelligent optimization scheduling method in enterprise resource planning according to claim 1, characterized in that: The global strategy is decomposed into locally executable task instructions, and the execution basis for local decisions is determined, specifically including: By deploying a lightweight decision-making module on an edge computing device, relevant data of the global policy is obtained and preliminarily analyzed to determine the part of the global policy related to the local task. Based on the analyzed global policy data, a parameter delivery mechanism is used to decompose it into specific task instructions. The decomposed task instructions are formatted to obtain an instruction set suitable for local decision-making, and the local task execution environment information on the edge computing device is obtained to determine the execution basis after adaptation. The system obtains the optimization requirements of the scheduling strategy parameters and adjusts the parameters to obtain the adjusted strategy parameter set. Then, it obtains the running status data of the local decision module. If the running status data exceeds the preset range, it uses a pre-established logistic regression model to perform state correction and judges the running stability after correction. Based on the corrected operational stability data, real-time load information of the deployed equipment is obtained, the load information is dynamically allocated, and the path to achieve the final optimization goal is determined.

7. The intelligent optimization scheduling method in enterprise resource planning according to claim 1, characterized in that: Determining whether to adjust the local resource configuration scheme includes: By using sensor modules on edge devices, operational data from the production and logistics processes are continuously collected. The collected data streams are initially filtered, and an anomaly flag is output using a preset threshold comparison mechanism. Obtain the current configuration of local resources, analyze resource usage in conjunction with the running logic, determine whether there is insufficient or uneven resource allocation, and call the real-time decision module to determine whether to trigger an adjustment strategy by analyzing resource usage. If resources are unevenly or insufficiently allocated, an optimized configuration scheme is generated, the allocation ratio of local resources is automatically adjusted, the operating logic of edge devices is updated synchronously, the adjusted operating status is obtained, and data changes in the production and logistics processes are continuously monitored. The comparison mechanism determines whether the adjustment meets the expected results, outputs monitoring results, records the execution of the adjustment strategy, stores it in the local database of the edge device, and determines the final operational stability.

8. The intelligent optimization scheduling method in enterprise resource planning according to claim 1, characterized in that: Obtain updates to the global resource status, specifically including: According to the adjusted local resource configuration scheme, a synchronization mechanism is used to establish a connection with the cloud, complete the data transmission process, determine the data content to be transmitted to the cloud, and if data loss or abnormality is detected during the transmission process, a retransmission mechanism is triggered to verify and complete the abnormal data and obtain the complete cloud data record. By recording data in the cloud, the data twin model is updated, and a consistency check is performed on the updated model to determine whether the model reflects the global resource status. The parameters of the data twin model are verified to obtain state information that conforms to reality. Based on the adjusted status information, a global resource update feedback is generated. The feedback content is formatted to determine the final structure of the feedback data. The feedback data is then transmitted back to the edge device through a synchronization mechanism to obtain the basis for optimizing the local resource configuration.

9. The intelligent optimization scheduling method in enterprise resource planning according to claim 8, characterized in that: After obtaining the update feedback of the global resource state, the process also includes: re-optimizing the training parameters of the deep reinforcement learning algorithm, and iterating the policy on the new state data to obtain a scheduling policy output that is more adapted to the current scenario.

10. An intelligent optimization scheduling system for enterprise resource planning, applied to the intelligent optimization scheduling method for enterprise resource planning as described in any one of claims 1-9, characterized in that: include: The multi-source heterogeneous data acquisition module collects real-time data on equipment operating status, inventory changes, and external environment through sensors and interfaces deployed in the production and logistics processes. It supports the access and preliminary processing of multi-source heterogeneous data streams, identifies abnormal data, and marks it. The data cleaning and fusion module performs noise reduction, format unification and structure transformation on the collected raw data, handles data conflicts and fuses multi-source features to generate a standardized dataset, and fills in missing values ​​through logical verification. The data twin construction module, based on standardized data, builds a high-precision virtual model of the enterprise's operation process, maps the production and logistics status in real time, dynamically updates virtual environment parameters, identifies abnormal states, and continuously optimizes the model's accuracy through a feedback mechanism. The deep reinforcement learning optimization module simulates resource allocation scenarios in a virtual environment, trains scheduling strategies using deep reinforcement learning algorithms, iterates and optimizes strategies in conjunction with dynamic changes in the environment, and generates scheduling parameters that adapt to different scenarios. The edge computing deployment module decomposes the global optimization strategy into locally executable task instructions through a parameter distribution mechanism, deploys them on the edge computing device, and adapts and corrects them in combination with the local operating environment to ensure the effective execution of the strategy and the reasonable allocation of resources on the edge node. The local real-time decision-making module runs real-time decision-making logic on edge devices, continuously monitors changes in production and logistics status, identifies abnormal events through threshold comparison mechanisms, dynamically adjusts local resource allocation schemes, and feeds back the execution results to the system. The cloud synchronization and feedback module uploads local execution results to the cloud data twin through the edge and cloud data synchronization mechanism, updates the global resource status, detects model consistency and performs parameter correction, generates global feedback information and sends it back to the edge device; The strategy iteration and update module analyzes resource usage trends based on global feedback data, dynamically adjusts the training parameters of deep reinforcement learning, continuously iterates the scheduling strategy to adapt to new scenarios, verifies the applicability of the strategy, and outputs the final scheduling scheme.

Images