Enterprise resource planning data efficient processing method based on cloud computing

Abstract

The invention relates to the technical field of cloud computing, and discloses an enterprise resource planning data efficient processing method based on cloud computing, and the method comprises the steps: collecting system data of finance, supply chain, production management and the like through a multi-source interface module, and carrying out the preprocessing through a hierarchical fusion framework, thereby obtaining a standardized data set; the method comprises the following steps of: selecting a multi-target scheduling model, inputting the multi-target scheduling model into a task decomposition model, generating task decomposition parameters based on a block processing mechanism and a service association rule, constructing the multi-target scheduling model aiming at the shortest processing duration and the minimum resource occupation, and outputting an optimal scheduling scheme by adopting a dynamic allocation algorithm. A hierarchical execution control model comprising a strategy layer, a coordination layer and an execution layer is established, the strategy layer performs global task planning, the coordination layer performs local resource adjustment, and the execution layer tracks tasks based on a load balancing algorithm and outputs a processing control instruction to realize efficient processing. The method improves the data processing efficiency and the resource utilization rate, and provides support for enterprise decision making.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cloud computing, and particularly to an enterprise resource planning data efficient processing method based on cloud computing. BACKGROUND

[0002] In today's digital era, enterprise resource planning (ERP) systems serve as the core tool for enterprise management, taking on the important task of integrating data from various aspects such as finance, supply chain, and production. As enterprises continue to expand their business and the market environment becomes increasingly complex, the amount of data faced by enterprises is growing explosively, and the sources of data are becoming more diverse, which poses many challenges to traditional enterprise resource planning data processing methods.

[0003] Traditional data processing methods are inefficient when dealing with multi-source heterogeneous enterprise resource planning data. Different business systems such as financial systems, supply chain systems, and production management systems often use different data formats and interface standards, resulting in a large number of format incompatibility problems during data collection. For example, the financial system may store data in a specific table format, while the production management system may use a completely different data storage format, which requires a large amount of format conversion and cleaning work after data collection, not only consuming a lot of time and effort, but also easily causing data loss or errors.

[0004] In terms of data preprocessing, traditional methods lack an effective hierarchical fusion framework, making it difficult to comprehensively and systematically preprocess multi-source data. For data collected by different systems, traditional methods usually use simple filtering or conversion methods, which cannot deeply mine the association relationships between data and are difficult to accurately identify and handle outliers in the data. This results in low-quality preprocessed data that cannot provide a reliable foundation for subsequent data processing tasks.

[0005] The lack of task decomposition and scheduling mechanism is also a major flaw of traditional methods. Traditional task decomposition is usually based on simple rules or experience, without fully considering business association rules and resource capabilities, resulting in unreasonable task decomposition and unbalanced resource allocation. For example, when dealing with complex enterprise resource planning data tasks, traditional methods may split highly associated task units, requiring frequent cross-task coordination during task execution, increasing processing time and resource consumption. At the same time, traditional scheduling models usually only consider a single optimization goal, such as the shortest processing time or the smallest resource occupation, which cannot balance multiple goals and cannot meet the diverse needs of enterprises for data processing.

[0006] In terms of resource allocation and load balancing, traditional methods lack dynamic adjustment mechanisms and cannot adapt to the dynamic changes of resources in the cloud computing environment. The resources of the cloud computing platform have the characteristics of elastic scaling, but traditional methods often use static allocation strategies when allocating resources. Once the resource allocation is completed, it is difficult to make real-time adjustments according to the actual load and resource usage. This leads to situations where some nodes are heavily loaded while others are idle during data processing, severely affecting the efficiency of data processing and the utilization of resources.

[0007] In addition, traditional enterprise resource planning data processing methods lack effective hierarchical execution control mechanisms when facing large-scale data processing tasks. The data processing process often lacks clear hierarchical division and collaborative management, leading to chaotic task execution process and making it difficult to effectively track and control data processing tasks. This makes it difficult for enterprises to timely understand the progress and status of data processing and to timely handle possible problems, affecting the timeliness and accuracy of enterprise decision-making.

[0008] Traditional enterprise resource planning data processing methods have many shortcomings in data collection, preprocessing, task decomposition, scheduling, resource allocation, and execution control, and cannot meet the needs of enterprises for efficient processing of resource planning data in the cloud computing environment. Therefore, there is an urgent need for an enterprise resource planning data efficient processing method based on cloud computing to solve the above problems and improve the efficiency and quality of enterprise data processing. SUMMARY

[0009] The purpose of the present application is to provide an enterprise resource planning data efficient processing method based on cloud computing to solve the problems raised in the background art.

[0010] To achieve the above purpose, the present application provides the following technical solution: an enterprise resource planning data efficient processing method based on cloud computing, the method comprising:

[0011] Collecting enterprise resource planning data through a multi-source interface module, the multi-source interface module including a financial system interface, a supply chain system interface, and a production management system interface; preprocessing the enterprise resource planning data based on a hierarchical fusion framework to obtain a standardized data set; inputting the standardized data set into a pre-configured task decomposition model, the task decomposition model using a block processing mechanism and based on business association rules to hierarchically divide data processing tasks and generate task decomposition parameters;

[0012] According to the task decomposition parameter, a multi-objective scheduling model is constructed, the multi-objective scheduling model takes the shortest processing time and the smallest resource occupation as the optimization goal, and adopts a dynamic allocation algorithm to perform global scheduling on the data processing task, wherein the dynamic allocation algorithm introduces task priority identification and resource availability evaluation; and an optimal scheduling scheme is output based on the multi-objective scheduling model;

[0013] According to the optimal scheduling scheme, a hierarchical execution control model is established, the hierarchical execution control model includes a strategy layer, a coordination layer and an execution layer, wherein the strategy layer performs global task planning based on the task decomposition parameter, the coordination layer performs local resource adjustment based on the optimal scheduling scheme, and the execution layer realizes data processing task tracking of the cloud computing node based on a load balancing algorithm; and a processing control instruction is output through the hierarchical execution control model, so as to realize efficient processing of enterprise resource planning data.

[0014] Preferably, the standardized data set is input into a pre-configured task decomposition model, the task decomposition model adopts a block processing mechanism, performs hierarchical division on the data processing task based on a business association rule, and generates task decomposition parameters including:

[0015] A standardized data set is obtained, the standardized data set includes financial voucher information, supply chain order information, production process records and cross-system association identifiers; a data feature space is constructed based on the standardized data set, and a resource capability space is constructed based on the amount of computing resources, the storage capacity and the network bandwidth that can be allocated by the cloud computing platform;

[0016] Based on the data feature space and the resource capability space, a multi-dimensional business association rule is constructed, the multi-dimensional business association rule includes a data type matching item, a processing timeliness requirement item, a resource adaptation item and a cross-system collaboration item, wherein the data type matching item is obtained through compatibility evaluation of data format and processing module, the processing timeliness requirement item is obtained through matching degree calculation of data update frequency and task completion time, the resource adaptation item is obtained through load balancing degree evaluation of data volume and computing resources, and the cross-system collaboration item is determined through the correspondence relationship between the association identifier and the collaborative processing flow;

[0017] A block processing mechanism is constructed, the block processing mechanism includes an initial task set, a hierarchical division rule and an association verification module, the initial task set contains a plurality of data processing task units, the hierarchical division rule performs hierarchical sorting on the task units through business process priority, and the association verification module verifies the dependency relationship between the task units based on the multi-dimensional business association rule;

[0018] The initial task set is updated by an iterative optimization method, high-priority task units are retained by priority screening, low-priority task units are merged and complex task units are split, and a hierarchical distribution and resource requirement of the data processing task are calculated based on the updated task set; and task decomposition parameters including the hierarchical distribution of the task, a resource requirement list, a task unit dependency relationship and a cross-system collaboration identifier are output based on the block processing mechanism.

[0019] Preferably, a multi-objective scheduling model is constructed based on the task decomposition parameters, the multi-objective scheduling model takes the shortest processing time and the smallest resource occupation as optimization objectives, a dynamic allocation algorithm is used for global scheduling of the data processing task, and an optimal scheduling scheme is output based on the multi-objective scheduling model, including:

[0020] A multi-objective function of scheduling is constructed, including a processing time optimization objective function and a resource occupation optimization objective function, wherein the processing time optimization objective function is calculated by summing the ratio of task unit processing time to corresponding resource processing speed, and the resource occupation optimization objective function is calculated by weighted sum of the calculation resource occupation rate, the storage resource occupation rate and the network bandwidth occupation rate of each cloud computing node;

[0021] A scheduling constraint condition is constructed based on the multi-objective function, including a time constraint, a resource constraint, a priority constraint and a collaboration constraint, the time constraint is used to limit the completion time range of the data processing task, the resource constraint is used to limit the upper limit of resource use of each cloud computing node, the priority constraint is used to limit the scheduling order of high-priority task units, and the collaboration constraint is used to ensure the processing order consistency of cross-system task units;

[0022] Each node contains a task unit identifier and corresponding cloud computing node information, a task allocation probability is constructed based on a task transfer cost and a heuristic evaluation value, and the next feasible node is determined by the task allocation probability; wherein the heuristic evaluation value is calculated by weighted calculation of the processing time difference and the resource occupation difference between nodes;

[0023] An adaptive task transfer cost mechanism is introduced, the adaptive task transfer cost mechanism uses a time-varying weight coefficient, the time-varying weight coefficient changes linearly with the increase of the number of scheduling iterations, and the task transfer cost is dynamically adjusted by the time-varying weight coefficient;

[0024] A dynamic heuristic evaluation mechanism is introduced, which adopts a time-varying influence factor that linearly increases with the number of scheduling iterations, and adjusts the weight of the heuristic evaluation value in the task allocation probability through the time-varying influence factor; based on the adaptive task transfer cost mechanism and the dynamic heuristic evaluation mechanism, iterative optimization is performed, the scheduling scheme generated in each iteration is evaluated in terms of time cost and resource cost, and non-dominated solutions are added to the Pareto front solution set;

[0025] An optimal solution that meets the time-resource trade-off is selected from the Pareto front solution set as the optimal scheduling scheme, and the optimal scheduling scheme is subjected to process smoothing to generate an optimal node sequence and corresponding cloud computing node resource allocation parameters.

[0026] Preferably, the strategy layer performs global task planning based on the task decomposition parameters, which includes:

[0027] The task processing path is described by using parameterized processes, and the task processing path is represented as a function of process parameters, the value range of the process parameters is 0 to 1, and the task processing path includes a data input stage, an intermediate processing stage, and a result output stage;

[0028] The task processing path is described based on a piecewise linear function, and the time schedule of the task processing path is calculated by summing the product of node coordinates and piecewise weights, the piecewise weights are calculated by the ratio of task unit priority to processing complexity;

[0029] Task processing path constraints are constructed, which include time constraints, sequence constraints, resource constraints, and quality constraints, the time constraints are used to limit the time schedule range of the task processing path, the sequence constraints are used to limit the execution order of the task units, the resource constraints are used to limit the upper limit of resource usage at each stage, and the quality constraints are used to limit the accuracy requirements of data processing results;

[0030] A global multi-objective optimization function is constructed, which includes a total process time item, a resource usage fluctuation item, a sequence deviation item, and a quality compliance item, and each item in the global multi-objective optimization function is combined by weighting coefficients;

[0031] The task processing path interval is discretized into multiple process segments, the global multi-objective optimization function is discretized, a global discretized objective function is constructed, and the global discretized objective function includes process segment time, resource usage, sequence deviation value, and quality compliance rate;

[0032] The gradient descent method is used to iteratively optimize the global discretized objective function, the gradient value of the objective function with respect to the control node is calculated, and the time schedule of the control node is updated along the negative gradient direction;

[0033] The optimized task processing path is linearly interpolated and smoothed, the smooth and continuous task processing path is generated by maintaining the continuity of the time schedule, resource usage and sequential relationship at the interpolation endpoints, and the node sequence and corresponding cloud computing node processing speed are generated based on the smooth and continuous task processing path.

[0034] Preferably, the coordination layer performs local resource adjustment based on the optimal scheduling scheme, including:

[0035] A local adjustment window is constructed based on the current load, current resource occupation and current processing progress of the cloud computing platform, the size of the local adjustment window is adaptively adjusted by the load correlation coefficient, and the size of the local adjustment window is positively correlated with the size of the current load;

[0036] A local resource model is constructed using real-time monitoring data, standardized resource occupation data is obtained by normalizing resource usage data, the availability identifier of the resource state table is updated based on the standardized resource occupation data, and the resource availability probability value of each cloud computing node is calculated by the probability accumulation method;

[0037] An exponential smoothing algorithm is used to predict dynamic resource demand, a future resource demand state vector is predicted based on historical demand data, the predicted state is updated based on real-time monitoring data, and accurate prediction and change trend of resource demand are obtained;

[0038] A resource adjustment model is constructed, the resource allocation rules of the cloud computing platform are taken as a state equation, the state equation includes computing resources, storage resources and network resources, state constraint conditions and allocation constraint conditions are constructed, the state constraint conditions are used to limit the value range of resource occupation and the value range of processing progress, and the allocation constraint conditions are used to limit the value range of resource allocation adjustment amplitude and adjustment frequency;

[0039] A multi-objective cost function is constructed, the multi-objective cost function includes a reference scheduling tracking term, a resource conflict avoidance term, a flow smoothing term and an energy consumption term, and each term in the multi-objective cost function is weighted and combined by a weighting coefficient;

[0040] The multi-objective cost function is optimized and solved by using the Lagrange multiplier method, a Lagrange function is constructed and a constraint condition is introduced, and the optimal adjustment amount is obtained by solving the partial derivative equation set.

[0041] Preferably, the execution layer implements cloud computing node data processing task tracking based on a load balancing algorithm, outputs processing control instructions through the hierarchical execution control model, and realizes efficient processing of enterprise resource planning data, including:

[0042] A cloud computing platform multi-node resource allocation model is established, including a computing resource allocation equation and a storage resource allocation equation, the computing resource allocation equation including processing task volume, node processing capacity, and task priority, and the storage resource allocation equation including data storage volume, node storage capacity, and data access frequency;

[0043] The multi-node resource allocation model is constructed as a state space expression, the state vector of the state space expression including computing resource occupancy, storage resource occupancy, task processing progress, and node load balancing degree, and the control vector of the state space expression including computing resource adjustment amount and storage resource adjustment amount;

[0044] The state space expression is linearized, the partial derivatives of the system state equation with respect to the state vector and the control vector are calculated, and a linearized prediction model is constructed;

[0045] A task tracking prediction cost function is constructed, including a tracking error term, a control amount penalty term, and a control increment penalty term, and each penalty term is combined by weighting through a weight matrix;

[0046] State constraint conditions are constructed, including resource occupancy constraints and task processing progress constraints, control constraint conditions are constructed, including computing resource adjustment amount constraints and storage resource adjustment amount constraints, and control increment constraint conditions are constructed, including computing resource increment constraints and storage resource increment constraints;

[0047] The task tracking prediction cost function is converted into a standard form of a quadratic programming problem, a quadratic matrix and a linear term coefficient are calculated, and an inequality constraint matrix and an equality constraint matrix are constructed;

[0048] The effective set method is used to solve the quadratic programming problem, the optimal solution is gradually approached by identifying effective constraints and solving sub-problems;

[0049] Based on the optimization solution, the control amount is mapped, the total computing resource adjustment amount is distributed to each cloud computing node through a computing resource allocation matrix, and the total storage resource adjustment amount is distributed to each node storage unit through a storage resource allocation matrix;

[0050] The resource allocation of the cloud computing node is limited, the computing resource allocation amount is limited according to the maximum processing capacity of the node, the storage resource allocation amount is limited according to the maximum storage capacity of the node, and finally, a processing control instruction is generated.

[0051] Preferably, the preprocessing operation of the enterprise resource planning data based on the hierarchical fusion framework includes: performing format conversion processing on the enterprise resource planning data collected by the financial system interface, the supply chain system interface and the production management system interface respectively to obtain uniform data formats; performing outlier cleaning on each format data, calculating the data mean and variance through a sliding window, and eliminating abnormal data beyond the range; extracting the core business information of each data by using a keyword matching algorithm, and establishing the corresponding relationship between the data of different systems through information association; converting the associated business information to a unified data model, fusing the multi-system data information by using a weighted average method, and generating a standardized data set containing a time stamp.

[0052] Preferably, the construction of the block processing mechanism includes: using uniform distribution to initialize a task set, each task unit containing a business type and a time requirement of data processing; calculating the node resource demand corresponding to the task unit through the resource allocation rule of the cloud computing platform, mapping the task unit demand to the resource capacity table based on the interface parameter configuration, and generating a task unit resource demand list; calculating the matching error of the task unit resource demand and the actual resources of the node, and constructing a business association rule based on the error sum of squares; obtaining the priority value of each task unit through normalization processing, retaining the task unit with a priority value greater than a preset threshold, and eliminating the task unit with a priority value less than the preset threshold.

[0053] Preferably, the construction of the task transfer cost includes: calculating the processing time difference between the current node and the next node as the processing time cost; calculating the adjustment change amount of each resource required by the cloud computing platform from the current node to the next node as the resource adjustment cost; calculating the minimum distance of the resource conflict in the scheduling scheme, and increasing the penalty cost if it is less than a safety threshold; and generating the total task transfer cost by linearly combining the processing time cost, the resource adjustment cost and the penalty cost.

[0054] Preferably, the effective set method is used to solve the quadratic programming problem, including: initializing an effective constraint set, adding all constraint conditions to a candidate set; solving an initial sub-problem to obtain an initial feasible solution; calculating the constraint violation amount of the current solution, adding the constraint with the largest violation amount to the effective set; re-solving the sub-problem to update the current solution; and repeating the above steps until the violation amount of all constraints is less than a preset precision to obtain an optimal solution.

[0055] Compared with the prior art, the present application has the following advantages:

[0056] The enterprise resource planning data is collected through a multi-source interface module, which includes a financial system interface, a supply chain system interface, and a production management system interface, enabling comprehensive collection of multi-source heterogeneous data of the enterprise, solving the problem of single data collection interface and incompatible formats in traditional methods, and providing a rich and comprehensive data foundation for subsequent data processing.

[0057] The collected data is preprocessed based on a hierarchical fusion framework, through steps such as format conversion, outlier cleaning, keyword matching, and information fusion, to obtain a standardized data set. This process effectively improves the quality and consistency of the data, reduces errors and biases in the data processing process, and lays a solid foundation for subsequent task decomposition and processing.

[0058] The standardized data set is input into a pre-configured task decomposition model, which uses a block processing mechanism and based on business association rules to hierarchically divide data processing tasks, generating task decomposition parameters. This task decomposition method fully considers the business association and resource capabilities between data, enabling the decomposition of complex tasks into a reasonable hierarchical structure, improving the efficiency and rationality of task processing. By constructing multi-dimensional business association rules, including data type matching items, processing time efficiency requirements, resource adaptation items, and cross-system collaboration items, the scientificity and accuracy of task decomposition are ensured, avoiding the problem of unreasonable task decomposition in traditional methods.

[0059] A multi-objective scheduling model is constructed according to the task decomposition parameters, with the shortest processing time and the smallest resource occupation as the optimization objectives, and a dynamic allocation algorithm is used for global scheduling of data processing tasks. This scheduling model can balance between multiple optimization objectives, and through the introduction of task priority identification and resource availability evaluation, it realizes dynamic allocation and optimal scheduling of resources. At the same time, through the adaptive task transfer cost mechanism and dynamic heuristic evaluation mechanism, the convergence speed and solution quality of the scheduling algorithm are improved, ensuring the optimality of the scheduling scheme and effectively improving the utilization of resources and the efficiency of data processing.

[0060] A hierarchical execution control model is established according to the optimal scheduling scheme, including a strategy layer, a coordination layer, and an execution layer. The strategy layer performs global task planning based on task decomposition parameters, generates smooth and continuous task processing paths through parameterized process description and multi-objective optimization, ensuring the global optimality of task processing. The coordination layer performs local resource adjustment based on the optimal scheduling scheme, realizes dynamic adjustment and optimal configuration of resources through the construction of local adjustment windows, resource models, and adjustment models, improving the adaptability and reliability of the system. The execution layer realizes data processing task tracking of cloud computing nodes based on load balancing algorithms, and through the construction of multi-node resource allocation models and prediction cost functions, it realizes accurate tracking and control of tasks, ensuring efficient execution of data processing tasks.

[0061] Based on the preprocessing operation of the hierarchical fusion framework, the data collected by different system interfaces are processed by format conversion, outlier cleaning and other processing, which can effectively improve the data quality and provide reliable data for subsequent processing. The block processing mechanism makes the task decomposition more in line with the actual business through business association rules and hierarchical division, and improves the task processing efficiency. The multi-objective scheduling model considers the processing time and resource occupation, and the dynamic allocation algorithm combines priority and resource availability to achieve global optimal scheduling and improve resource utilization. The strategy layer, coordination layer and execution layer of the hierarchical execution control model work together to form a complete control system from global planning to local adjustment and task tracking, ensuring efficient and orderly data processing. The adaptive task transfer cost and dynamic heuristic evaluation mechanism improve the performance of the scheduling algorithm, making the scheduling scheme more optimal. The load balancing algorithm realizes node task tracking in the execution layer, combines the resource allocation model and optimization solution to ensure reasonable resource allocation and avoid node load imbalance. This method effectively solves the problems of low data processing efficiency, low resource utilization and unreasonable task scheduling in traditional methods, improves the efficiency and quality of enterprise resource planning data processing, provides more timely and accurate data support for enterprise decision-making, and has important practical application value. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 A working principle diagram of the enterprise resource planning data efficient processing method based on cloud computing is described.

[0063] Figure 2 A flowchart for generating task decomposition parameters is described.

[0064] Figure 3 A flowchart for constructing and optimizing the multi-objective scheduling model is described.

[0065] Figure 4 A flowchart for global task planning in the strategy layer is described.

[0066] Figure 5 A flowchart for local resource adjustment in the coordination layer is described. DETAILED DESCRIPTION

[0067] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0068] Please refer to Figures 1-5 The enterprise resource planning data efficient processing method based on cloud computing is described, and the specific implementation steps are as follows:

[0069] The enterprise resource planning data is collected through a multi-source interface module, which includes a financial system interface, a supply chain system interface, and a production management system interface. The enterprise resource planning data is preprocessed based on a hierarchical fusion framework to obtain a standardized data set.

[0070] The standardized data set is input into a pre-configured task decomposition model. The task decomposition model uses a block processing mechanism to divide data processing tasks into levels according to business correlation rules, and then generates task decomposition parameters.

[0071] A multi-objective scheduling model is constructed according to the task decomposition parameters. The multi-objective scheduling model takes the shortest processing time and the smallest resource occupation as optimization objectives, and uses a dynamic allocation algorithm to implement global scheduling of data processing tasks. The dynamic allocation algorithm introduces task priority identification and resource availability evaluation. An optimal scheduling scheme is output based on the multi-objective scheduling model.

[0072] A hierarchical execution control model is established according to the optimal scheduling scheme. The hierarchical execution control model includes a strategy layer, a coordination layer, and an execution layer. The strategy layer performs global task planning based on task decomposition parameters, the coordination layer performs local resource adjustment based on the optimal scheduling scheme, and the execution layer implements data processing task tracking of cloud computing nodes based on a load balancing algorithm. Processing control instructions are output through the hierarchical execution control model to achieve efficient processing of enterprise resource planning data.

[0073] Example 1: When processing enterprise resource planning data, first obtain a standardized data set, which contains financial voucher information, supply chain order information, production process records, and cross-system association identifiers. These information comes from different systems and has different formats and characteristics, which need to be uniformly processed and managed.

[0074] A data feature space is constructed based on the standardized data set. The construction of the data feature space needs to consider various properties and characteristics of the data, such as the type, format, length, and value range of the data. Through the analysis and extraction of these characteristics, the essence and characteristics of the data can be better understood, providing a basis for subsequent processing.

[0075] A resource capability space is constructed according to the amount of computing resources, storage capacity, and network bandwidth that can be allocated by the cloud computing platform. The construction of the resource capability space needs to consider the actual resource situation of the cloud computing platform, including computing resources, storage resources, and network resources. Through the analysis and evaluation of these resources, the amount of data that can be processed and the processing speed of the cloud computing platform can be determined, providing a reference for subsequent task decomposition and scheduling.

[0076] The multi-dimensional business association rule is constructed based on a data feature space and a resource capability space. The multi-dimensional business association rule includes a data type matching item, a processing timeliness requirement item, a resource adaptation item, and a cross-system collaboration item. The data type matching item is obtained by evaluating the compatibility of data formats and processing modules. Different data formats require different processing modules to process, and only when the data format is compatible with the processing module can the efficiency and accuracy of data processing be ensured. The processing timeliness requirement item is calculated by matching the data update frequency and the task completion time. Different data update frequencies require different task completion times, and the processing time of the task needs to be reasonably arranged according to the data update frequency to ensure the timeliness and effectiveness of the data. The resource adaptation item is obtained by evaluating the load balancing degree of data volume and computing resources. Different data volumes require different computing resources, and the computing resources need to be reasonably allocated according to the size of the data volume to ensure the load balancing of the computing resources and improve the utilization rate of the computing resources. The cross-system collaboration item is determined according to the correspondence between the association identifier and the collaborative processing flow. The data between different systems needs to be collaboratively processed, and the association identifier can establish the correspondence between the data of different systems, thereby realizing the cross-system data collaborative processing.

[0077] A block processing mechanism is constructed. The block processing mechanism includes an initial task set, a hierarchical division rule, and an association verification module. The initial task set includes multiple data processing task units, each of which has a specific business type and time requirement. The hierarchical division rule sorts the task units in a hierarchical order through business process priority, with the task units of high business process priority being processed first. The association verification module verifies the dependency relationship between the task units based on the multi-dimensional business association rule to ensure the correctness of the dependency relationship between the task units and avoid incorrect task processing order.

[0078] The initial task set is then updated using an iterative optimization method. High-priority task units are retained through priority screening, low-priority task units are merged, and complex task units are split. When merging low-priority task units, the association and similarity between task units need to be considered, and task units with similar business types and time requirements are merged into one task unit to improve the efficiency of task processing. When splitting complex task units, the complex task units need to be split into multiple simple task units to facilitate task allocation and processing. The hierarchical distribution of data processing tasks and resource requirements are calculated based on the updated task set. The hierarchical distribution can help understand the priority and processing order of the tasks, and the resource requirements can help reasonably allocate computing resources and storage resources.

[0079] The task decomposition parameters include task level distribution, resource requirement list, task unit dependency relationship, and cross-system coordination identification. The task level distribution clarifies the priority and processing order of the tasks, the resource requirement list details the required computing resources, storage resources, and network resources for each task unit, the task unit dependency relationship ensures the correctness of task processing, and the cross-system coordination identification enables data collaborative processing between different systems. Through these task decomposition parameters, accurate inputs can be provided for subsequent multi-objective scheduling models, thereby achieving efficient processing of enterprise resource planning data.

[0080] In the construction of the multi-objective scheduling model, a multi-objective function for scheduling is first constructed, which includes a processing time optimization objective function and a resource occupation optimization objective function. The processing time optimization objective function is calculated by summing the ratio of task unit processing time to corresponding resource processing speed, which reflects the processing efficiency of each task unit on different resources and measures the optimization degree of processing time as a whole. The resource occupation optimization objective function is calculated by weighted sum of the computing resource occupation rate, storage resource occupation rate, and network bandwidth occupation rate of each cloud computing node, which comprehensively evaluates the resource occupation of the entire cloud computing platform and provides a basis for optimizing resource occupation.

[0081] Based on the above multi-objective function, corresponding scheduling constraints need to be constructed, including time constraints, resource constraints, priority constraints, and coordination constraints. Time constraints are used to limit the completion time range of data processing tasks, ensuring that tasks can be completed within the specified time and meeting the time requirements of the business. Resource constraints are used to limit the upper limit of resource usage of each cloud computing node to prevent system performance degradation or failure due to excessive resource usage. Priority constraints are used to clarify the scheduling order of high-priority task units to ensure that important tasks are processed first. Coordination constraints ensure the consistency of the processing order of cross-system task units, ensuring that data processing between different systems can be coordinated, avoiding inconsistent data or chaotic processing order.

[0082] When encoding the data processing task, a node sequence is used, and each node contains task unit identification and corresponding cloud computing node information. This encoding method can clearly represent the correspondence between tasks and cloud computing nodes, providing convenience for subsequent task allocation and scheduling. Based on task transfer cost and heuristic evaluation value, task allocation probability is constructed to determine the next feasible node. The heuristic evaluation value is obtained by weighting the processing time difference and resource occupation difference between nodes, which can consider the advantages of different nodes in processing time and resource occupation to provide more reasonable guidance for task allocation.

[0083] To improve the adaptability and optimization effect of the scheduling model, an adaptive task transfer cost mechanism is introduced. This mechanism adopts time-varying weight coefficients, and the time-varying weight coefficients will linearly decay with the increase of the number of scheduling iterations. In this way, the task transfer cost can be dynamically adjusted according to the progress of the scheduling, so that the model can better adapt to the needs of task allocation at different iteration stages. At the same time, a dynamic heuristic evaluation mechanism is introduced, which adopts a time-varying influence factor that linearly increases with the number of scheduling iterations. Through this influence factor, the weight of the heuristic evaluation value in the task allocation probability can be adjusted, so as to gradually strengthen the role of heuristic evaluation in the scheduling process and improve the accuracy and efficiency of task allocation.

[0084] Based on the adaptive task transfer cost mechanism and the dynamic heuristic evaluation mechanism, iterative optimization is carried out. In each iteration process, the time cost and resource cost of the generated scheduling scheme are evaluated, and non-dominated solutions are added to the Pareto frontier solution set. The Pareto frontier solution set contains all solutions that cannot be further optimized simultaneously in terms of time and resources. These solutions are all effective optimal solution candidates.

[0085] The optimal solution that meets the time-resource trade-off is selected from the Pareto frontier solution set as the optimal scheduling scheme. When selecting the optimal solution, both time and resource factors need to be considered to find the best balance point that meets the actual needs. The optimal scheduling scheme is subjected to flow smoothing processing to generate the optimal node sequence and the corresponding cloud computing node resource allocation parameters. Flow smoothing processing can make the execution of tasks more smooth, avoid uneven resource allocation or task execution interruption, and thus ensure the efficiency and stability of the entire data processing process. Through the above steps, the multi-objective scheduling model is constructed and optimized, the global scheduling of data processing tasks is realized, and the optimal scheduling scheme is output to provide reliable input for the subsequent hierarchical execution control model, so as to ensure the efficiency and accuracy of enterprise resource planning data processing.

[0086] In embodiment 3, when the strategy layer performs global task planning based on task decomposition parameters, a parameterized process is first used to describe the task processing path, which is represented as a function of process parameters, where the value range of the process parameters is set to 0 to 1. The task processing path covers three main stages: data input stage, intermediate processing stage, and result output stage, each of which has its specific processing content and requirements. By parameterizing the task processing path, the path can be more flexibly adjusted and optimized to adapt to different data processing needs.

[0087] The task processing path is described based on a piecewise linear function, and the time schedule of the task processing path is calculated by summing the product of the control node coordinates and the piecewise weight. The piecewise weight is calculated by the ratio of the task unit priority and the processing complexity. The weight of the segment with high task unit priority and low processing complexity is relatively large, and the proportion in the time schedule calculation is higher, which means that when planning the task processing path, high-priority and low-complexity task units will be given priority to ensure that important and easy-to-process tasks can be completed as soon as possible.

[0088] The constraints of the task processing path are constructed, including time constraints, sequence constraints, resource constraints and quality constraints. The time constraints are used to limit the time schedule range of the task processing path, to ensure that the entire task processing process can be completed within the specified time, to meet the time requirements of the business. The sequence constraint is used to clarify the execution order of the task unit, to avoid the situation of chaotic task processing sequence, to ensure the logicality and correctness of the task processing. The resource constraint is used to limit the upper limit of the resource usage of each stage, to prevent the system performance from being reduced or the resource from being wasted due to excessive resource usage. The quality constraint is used to limit the accuracy requirement of the data processing result, to ensure that the quality of the data processing meets the business requirements.

[0089] A global multi-objective optimization function is constructed, which includes a total process duration item, a resource usage fluctuation item, a sequence deviation item and a quality compliance item. The total process duration item is used to measure the length of the entire task processing path, the resource usage fluctuation item is used to evaluate the stability of resource usage, the sequence deviation item is used to reflect the difference between the task execution sequence and the expected sequence, and the quality compliance item is used to ensure the quality of the data processing result. The global multi-objective optimization function is combined by weighting coefficients, and the weights of each item are adjusted according to the actual demand, in order to optimize the task processing path.

[0090] The task processing path interval is discretized into multiple process segments, and the global multi-objective optimization function is discretized, thereby constructing a global discretized objective function. The global discretized objective function includes process segment duration, resource usage, sequence deviation value and quality compliance rate, etc. Discretization can convert continuous task processing path into discrete process segments, which is convenient for mathematical calculation and optimization.

[0091] The gradient descent method is used to iteratively optimize the global discretized objective function. The specific process is to calculate the gradient value of the objective function with respect to the control node, and then update the time schedule of the control node along the negative gradient direction. Through continuous iteration and update, the time schedule of the control node is gradually adjusted, so that the value of the objective function gradually decreases, thereby realizing the optimization of the task processing path.

[0092] The optimized task processing path is linearly interpolated and smoothed. By maintaining the continuity of time progress, resource usage and sequential relationship at the interpolation endpoints, a smooth and continuous task processing path is generated. The smoothing process makes the task processing path more smooth and avoids abrupt changes and discontinuities, improving the stability and efficiency of task processing. Finally, based on the smooth and continuous task processing path, a node sequence and corresponding cloud computing node processing speed are generated. The node sequence clearly defines the order and node allocation of task processing, and the cloud computing node processing speed provides specific parameter guidance for each node's task processing, ensuring that the entire task processing process can be carried out efficiently and orderly.

[0093] In example 4, when the coordination layer adjusts resources based on the optimal scheduling scheme, a local adjustment window is constructed based on the current load, current resource occupation and current processing progress of the cloud computing platform. For example, when the current load of the cloud computing platform is high, it means that the system is processing more tasks and the resource occupation is relatively large. At this time, the size of the local adjustment window needs to be expanded to consider the range of resource adjustment more comprehensively. When the current load is low, the size of the local adjustment window can be correspondingly reduced. Here, the size of the local adjustment window is adaptively adjusted by the load correlation coefficient, establishing a positive correlation between the size of the local adjustment window and the size of the current load. This can dynamically adjust the range of the adjustment window according to the actual load situation, making the resource adjustment more accurate.

[0094] A local resource model is constructed using real-time monitoring data. Real-time monitoring data can reflect the usage of each resource in the cloud computing platform in a timely manner. The resource usage data is normalized to obtain standardized resource occupation data. Normalization can convert resource usage data of different types and magnitudes into a unified standard form, facilitating comparison and analysis. The availability identifier of the resource state table is updated based on the standardized resource occupation data. For example, if the computing resource occupation rate of a certain cloud computing node is high, the availability identifier of the computing resource of that node will be correspondingly reduced; conversely, the availability identifier will be increased. The resource availability probability value of each cloud computing node is calculated by the probability accumulation method, which helps to more accurately understand the resource availability of each node and provides a basis for subsequent resource adjustment.

[0095] An exponential smoothing algorithm is used to predict the dynamic resource demand. Based on historical demand data, the exponential smoothing algorithm is used to predict the state vector of future resource demand. For example, according to the usage of computing resources, storage resources and network resources in the past period of time, the demand trend of these resources in the future period of time is predicted. At the same time, the predicted state is updated based on real-time monitoring data to obtain accurate prediction and change trend of resource demand. For example, when real-time monitoring data shows that a certain application suddenly increases the demand for storage resources, the predicted state needs to be adjusted in time to make the resource demand prediction more accurate, so as to prepare for resource allocation in advance.

[0096] A resource adjustment model is constructed. The resource allocation rules of the cloud computing platform are taken as the state equation, and the state equation includes computing resources, storage resources and network resources. For example, the state equation can describe the relationship between the allocation of computing resources and the task processing capacity and the node processing capacity. State constraint conditions and allocation constraint conditions are constructed. The state constraint conditions are used to limit the value range of resource occupation and the value range of processing progress, such as the occupation rate of computing resources cannot exceed the maximum processing capacity of the node, and the processing progress must complete a certain proportion of tasks within a specified time; the allocation constraint conditions are used to limit the value range of the adjustment amplitude and the adjustment frequency of resource allocation, so as to avoid too large or too frequent adjustment amplitude of resource allocation, which leads to system instability.

[0097] A multi-objective cost function is further constructed, which includes a reference scheduling tracking item, a resource conflict avoidance item, a process smoothing item and an energy consumption item. The reference scheduling tracking item is used to track the deviation of actual resource allocation from the optimal scheduling scheme, to ensure that resource adjustment does not deviate from the general direction of the optimal scheduling scheme; the resource conflict avoidance item is used to avoid the competition for resources between different tasks and reduce resource conflicts; the process smoothing item is used to ensure the continuity and stability of the task processing process and avoid the interruption of task processing due to resource adjustment; the energy consumption item is used to consider the energy consumption problem in the resource adjustment process and achieve energy saving optimization. The multi-objective cost function is weighted and combined by weighting coefficients, and the weights of each item are adjusted according to actual needs to achieve the optimal resource adjustment effect.

[0098] The Lagrange multiplier method is used to optimize and solve the multi-objective cost function. The Lagrange function is constructed and the previously set constraints are introduced. The optimal adjustment amount is obtained by solving the partial derivative equation set. For example, in the solving process, the state constraint condition and the allocation constraint condition need to be considered to limit the resource adjustment amount. The optimal adjustment amount that meets all the constraint conditions and minimizes the multi-objective cost function is found through mathematical calculation. Based on the obtained optimal adjustment amount, the resources of the cloud computing platform are locally adjusted, so that the resource allocation is more reasonable, the utilization rate of resources and the processing efficiency of the system are improved, and on the basis of the optimal scheduling scheme, the processing process of enterprise resource planning data is further optimized through local resource adjustment to achieve the goal of efficient processing.

[0099] In example 5, when the execution layer implements data processing task tracking of the cloud computing node based on the load balancing algorithm, a cloud computing platform multi-node resource allocation model is established, which includes a computing resource allocation equation and a storage resource allocation equation. Taking the computing resource allocation equation as an example, it includes the processing task amount, the node processing capacity, and the task priority. Assuming that a cloud computing platform has three nodes, the processing capacity of node A is 100 task units per second, the currently allocated processing task amount is 80 task units, and the priority of these task units is medium; the processing capacity of node B is 150 task units per second, the allocated processing task amount is 120 task units, and the task priority is high; the processing capacity of node C is 80 task units per second, the allocated processing task amount is 60 task units, and the task priority is low. The storage resource allocation equation includes the data storage amount, the node storage capacity, and the data access frequency. For example, the storage capacity of node A is 10 TB, the current data storage amount is 7 TB, and the data access frequency is 1000 times per hour; the storage capacity of node B is 15 TB, the data storage amount is 10 TB, and the access frequency is 1500 times per hour; the storage capacity of node C is 8 TB, the data storage amount is 5 TB, and the access frequency is 800 times per hour.

[0100] The multi-node resource allocation model is constructed as a state space expression, and the state vector of the state space expression includes the computing resource occupancy, the storage resource occupancy, the task processing progress, and the node load balancing degree. The computing resource occupancy can be calculated by the ratio of the processing task amount to the node processing capacity, such as the computing resource occupancy of node A is 80÷100=80%, the computing resource occupancy of node B is 120÷150=80%, and the computing resource occupancy of node C is 60÷80=75%. The storage resource occupancy is the ratio of the data storage amount to the node storage capacity, the storage resource occupancy of node A is 7÷10=70%, the storage resource occupancy of node B is 10÷15≈66.7%, and the storage resource occupancy of node C is 5÷8=62.5%. The task processing progress is determined according to the ratio of the completed task amount to the total task amount, assuming that the total task amount is 1000 task units, node A has completed 60 task units, and the progress is 6%; node B has completed 80 task units, and the progress is 8%; and node C has completed 40 task units, and the progress is 4%. The node load balancing degree is measured by the difference of the resource occupancy of each node, and if the difference of the resource occupancy of each node is small, the load balancing degree is high. The control vector of the state space expression includes the computing resource adjustment amount and the storage resource adjustment amount, for example, it is planned to increase the computing resource adjustment amount of node A by 10 task units, to decrease the computing resource adjustment amount of node B by 5 task units, to increase the storage resource adjustment amount of node A by 0.5 TB, and to increase the storage resource adjustment amount of node C by 0.3 TB.

[0101] The state space expression is linearized, the partial derivatives of the system state equation with respect to the state vector and the control vector are calculated, and a linearized prediction model is constructed. By analyzing the influence of the changes of the state vector and the control vector on the system state, the trend of the change of the system state under different control vectors is predicted, such as the possible change of the computing resource occupancy when the computing resource adjustment amount is increased, and the influence on the task processing progress and the node load balancing degree.

[0102] A task tracking prediction cost function is constructed, which includes a tracking error term, a control amount penalty term, and a control increment penalty term. The tracking error term is used to measure the deviation between the actual state and the predicted state, such as the difference between the actual computing resource occupancy and the predicted occupancy; the control amount penalty term considers the size of the control vector to avoid excessive adjustment of resources; and the control increment penalty term penalizes the change amount of the control vector to prevent frequent adjustment of resources. The weight matrix is used to weight and combine each penalty term, and the weight of each penalty term is determined according to the actual demand, for example, when more attention is paid to the tracking error, the weight of the tracking error term can be increased.

[0103] The state constraint conditions are constructed, including resource occupancy constraints and task processing progress constraints. The resource occupancy constraints are, for example, that the computing resource occupancy cannot exceed 90%, and the storage resource occupancy cannot exceed 85%. The task processing progress constraints are, for example, that the task processing progress of each node needs to reach more than 10% of the total task quantity within a specified time. The control constraint conditions are constructed, including computing resource adjustment amount constraints and storage resource adjustment amount constraints, for example, the computing resource adjustment amount cannot exceed 10% of the node processing capacity each time, and the storage resource adjustment amount cannot exceed 5% of the node storage capacity each time. The control increment constraint conditions are constructed, including computing resource increment constraints and storage resource increment constraints, for example, the computing resource increment cannot exceed 5 task units each time, and the storage resource increment cannot exceed 0.2 TB each time.

[0104] The task tracking prediction cost function is converted into a standard form of a quadratic programming problem, the quadratic form matrix and the first-order term coefficient are calculated, and the inequality constraint matrix and the equality constraint matrix are constructed. Through mathematical transformation, the complex cost function is converted into a quadratic programming problem, so as to facilitate solving.

[0105] The effective set method is used to solve the quadratic programming problem. First, the effective constraint set is initialized, and all constraint conditions are added to the candidate set. Then, the initial sub-problem is solved to obtain an initial feasible solution. Assuming that the initial feasible solution is that the computing resource adjustment amount of node A is increased by 5 task units, the computing resource adjustment amount of node B is reduced by 3 task units, the storage resource adjustment amount of node A is increased by 0.3 TB, and the storage resource adjustment amount of node C is increased by 0.2 TB. Then, the constraint violation quantity of the current solution is calculated, and whether there is a constraint condition that is not met is checked, for example, whether there is a case that the resource occupancy exceeds the constraint. If there is, the constraint with the largest violation quantity is added to the effective set. The sub-problem is solved again, the current solution is updated, and the above steps are repeated until the violation quantity of all constraints is less than a preset precision, and the optimal solution is obtained.

[0106] Based on the optimization solving result, the control quantity mapping is performed. The total computing resource adjustment amount is distributed to each cloud computing node through a computing resource distribution matrix, and the total storage resource adjustment amount is distributed to each node storage unit through a storage resource distribution matrix. For example, the total computing resource adjustment amount is 10 task units, according to the computing resource distribution matrix, node A is allocated 4, node B is allocated 3, and node C is allocated 3. The total storage resource adjustment amount is 1 TB, node A is allocated 0.4 TB, node B is allocated 0.3 TB, and node C is allocated 0.3 TB.

[0107] The resource allocation of the cloud computing node is limited, the computing resource allocation is limited according to the maximum processing capacity of the node, and the storage resource allocation is limited according to the maximum storage capacity of the node. For example, the maximum processing capacity of node A is 100 task units, 80 task units have been allocated, the computing resource allocation is limited to 20 task units, if the computing resource allocation of node A in the optimization result is 25, then the limit is 20; the maximum storage capacity of node B is 15 TB, 10 TB has been stored, the storage resource allocation is limited to 5 TB, if the storage resource allocation of node B in the optimization result is 5.5 TB, then the limit is 5 TB. Finally, a processing control instruction is generated, the tracking of the cloud computing node data processing task and the reasonable allocation of resources are realized, so as to guarantee the efficient enterprise resource planning data processing.

[0108] It should be noted that, in this paper, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0109] Although the embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for efficiently processing enterprise resource planning data based on cloud computing, characterized in that: include: Collecting enterprise resource planning data through a multi-source interface module, wherein the multi-source interface module includes a financial system interface, a supply chain system interface, and a production management system interface; Preprocessing the enterprise resource planning data based on a hierarchical fusion framework to obtain a standardized data set; inputting the standardized data set into a pre-configured task decomposition model, wherein the task decomposition model adopts a block processing mechanism, hierarchically divides the data processing tasks based on business association rules, and generates task decomposition parameters; constructing a multi-objective scheduling model based on the task decomposition parameters, wherein the multi-objective scheduling model takes minimizing processing time and minimizing resource usage as optimization goals, and adopts a dynamic allocation algorithm to globally schedule data processing tasks, wherein the dynamic allocation algorithm introduces task priority identification and resource availability assessment; and outputting an optimal scheduling solution based on the multi-objective scheduling model; Establishing a hierarchical execution control model based on the optimal scheduling solution, the hierarchical execution control model includes a strategy layer, a coordination layer, and an execution layer, wherein the strategy layer performs global task planning based on the task decomposition parameters, the coordination layer performs local resource adjustment based on the optimal scheduling solution, and the execution layer implements data processing task tracking of cloud computing nodes based on a load balancing algorithm; By outputting processing control instructions through the hierarchical execution control model, efficient processing of enterprise resource planning data is achieved.

2. The method for efficiently processing enterprise resource planning data based on cloud computing according to claim 1, characterized in that: The standardized data set is input into a pre-configured task decomposition model. The task decomposition model adopts a block processing mechanism and hierarchically divides the data processing tasks based on business association rules. The generated task decomposition parameters include: Acquire a standardized data set, the standardized data set including financial voucher information, supply chain order information, production process records, and cross-system association identifiers; construct a data feature space based on the standardized data set, and construct a resource capability space based on the amount of computing resources, storage capacity, and network bandwidth that can be allocated by the cloud computing platform; Constructing multi-dimensional business association rules based on the data feature space and the resource capability space, the multi-dimensional business association rules include data type matching items, processing time requirement items, resource adaptation items, and cross-system collaboration items, wherein the data type matching items are obtained by evaluating the compatibility of the data format and the processing module, the processing time requirement items are obtained by calculating the matching degree between the data update frequency and the task completion time, the resource adaptation items are obtained by evaluating the load balance degree between the data volume and the computing resources, and the cross-system collaboration items are determined by the correspondence between the association identifier and the collaborative processing flow; Constructing a block processing mechanism, the block processing mechanism including an initial task set, hierarchical division rules, and an association verification module. The initial task set includes multiple data processing task units. The hierarchical division rules hierarchically sort the task units according to business process priorities. The association verification module verifies the dependency relationship between task units based on the multi-dimensional business association rules. An iterative optimization method is used to update the initial task set, high-priority task units are retained through priority screening, low-priority task units are merged and complex task units are split, and the hierarchical distribution and resource requirements of data processing tasks are calculated based on the updated task set; task decomposition parameters are output based on the block processing mechanism, and the task decomposition parameters include task hierarchical distribution, resource requirement list, task unit dependency and cross-system collaboration identification.

3. The method for efficiently processing enterprise resource planning data based on cloud computing according to claim 1, characterized in that: A multi-objective scheduling model is constructed based on the task decomposition parameters. The multi-objective scheduling model takes minimizing processing time and minimizing resource usage as optimization goals, and uses a dynamic allocation algorithm to globally schedule data processing tasks. Outputting the optimal scheduling solution based on the multi-objective scheduling model includes: Constructing a multi-objective function for scheduling, the multi-objective function including a processing time optimization objective function and a resource occupancy optimization objective function, wherein the processing time optimization objective function is calculated by summing the ratios of the task unit processing time and the corresponding resource processing speed, and the resource occupancy optimization objective function is calculated by weighted summing of the computing resource occupancy rate, storage resource occupancy rate, and network bandwidth occupancy rate of each cloud computing node; Constructing scheduling constraints based on the multi-objective function, the scheduling constraints include time constraints, resource constraints, priority constraints, and coordination constraints. The time constraints are used to limit the completion time range of the data processing task, the resource constraints are used to limit the resource usage upper limit of each cloud computing node, the priority constraints are used to limit the scheduling order of high-priority task units, and the coordination constraints are used to ensure the consistency of the processing order of task units across systems; Data processing tasks are encoded using a node sequence, where each node contains a task unit identifier and corresponding cloud computing node information. A task assignment probability is constructed based on the task transfer cost and a heuristic evaluation value, and the next feasible node is determined based on the task assignment probability. The heuristic evaluation value is obtained by weighted calculation of the processing time difference and resource usage difference between nodes. An adaptive task transfer cost mechanism is introduced. The adaptive task transfer cost mechanism adopts a time-varying weight coefficient. The time-varying weight coefficient changes linearly with the increase of the scheduling iteration number. The task transfer cost is dynamically adjusted by the time-varying weight coefficient. A dynamic heuristic evaluation mechanism is introduced. The dynamic heuristic evaluation mechanism uses a time-varying impact factor that increases linearly with the number of scheduling iterations. The time-varying impact factor is used to adjust the weight of the heuristic evaluation value in the task allocation probability. Iterative optimization is performed based on the adaptive task transfer cost mechanism and the dynamic heuristic evaluation mechanism. The scheduling solution generated by each iteration is evaluated for duration and resource costs, and non-dominated solutions are added to the Pareto front solution set. An optimal solution that satisfies the duration-resource trade-off is selected from the Pareto front solution set as an optimal scheduling solution, and a process smoothing process is performed on the optimal scheduling solution to generate an optimal node sequence and corresponding cloud computing node resource allocation parameters.

4. The method for efficiently processing enterprise resource planning data based on cloud computing according to claim 1, characterized in that: The strategy layer performs global task planning based on the task decomposition parameters, including: A parameterized process is used to describe the task processing path, and the task processing path is expressed as a function of a process parameter, where the value of the process parameter ranges from 0 to 1. The task processing path includes a data input stage, an intermediate processing stage, and a result output stage; The task processing path is described based on a piecewise linear function, and the time schedule of the task processing path is calculated by summing the product of the control node coordinates and the segment weight, wherein the segment weight is calculated by the ratio of the task unit priority to the processing complexity; Constructing task processing path constraints, which include time constraints, sequence constraints, resource constraints, and quality constraints. The time constraints are used to limit the time schedule of the task processing path, the sequence constraints are used to limit the execution order of task units, the resource constraints are used to limit the resource usage upper limit of each stage, and the quality constraints are used to limit the accuracy requirements of data processing results. Constructing a global multi-objective optimization function, the global multi-objective optimization function includes a total process duration item, a resource usage fluctuation item, a sequence deviation item, and a quality compliance item, and performing a weighted combination of each item in the global multi-objective optimization function using a weighting coefficient; Discretize the task processing path interval into multiple process segments, discretize the global multi-objective optimization function, and construct a global discretized objective function, wherein the global discretized objective function includes the process segment duration, resource usage, sequence deviation value, and quality compliance rate; Iteratively optimizing the global discretized objective function using a gradient descent method, calculating the gradient value of the objective function with respect to the control node, and updating the time progress of the control node along the negative gradient direction; The optimized task processing path is subjected to linear interpolation smoothing processing, and a smooth and continuous task processing path is generated by maintaining the continuity of the time progress, resource usage and sequence relationship at the interpolation endpoints. A node sequence and the corresponding cloud computing node processing speed are generated based on the smooth and continuous task processing path.

5. The method for efficiently processing enterprise resource planning data based on cloud computing according to claim 1, characterized in that: The coordination layer performs local resource adjustment based on the optimal scheduling solution, including: A local adjustment window is constructed based on the current load, current resource usage, and current processing progress of the cloud computing platform, and the size of the local adjustment window is adaptively adjusted using a load correlation coefficient to establish a positive correlation between the size of the local adjustment window and the magnitude of the current load; Using real-time monitoring data to build a local resource model, normalizing resource usage data to obtain standardized resource occupancy data, updating the availability indicator of the resource status table based on the standardized resource occupancy data, and calculating the resource availability probability value of each cloud computing node through a probability accumulation method; Use exponential smoothing algorithms to predict dynamic resource demand, predict the state vector of future resource demand based on historical demand data, and update the predicted state based on real-time monitoring data to obtain accurate predictions and changing trends of resource demand; Construct a resource adjustment model, using the cloud computing platform's resource allocation rules as a state equation. The state equation includes computing resources, storage resources, and network resources. Construct state constraints and allocation constraints. The state constraints are used to limit the value ranges of resource occupancy and processing progress, and the allocation constraints are used to limit the value ranges of resource allocation adjustment amplitude and adjustment frequency. Constructing a multi-objective cost function, the multi-objective cost function including a reference scheduling tracking item, a resource conflict avoidance item, a process smoothing item, and an energy consumption item, and weighting each item in the multi-objective cost function by a weighting coefficient; The Lagrange multiplier method is used to optimize and solve the multi-objective cost function, a Lagrange function is constructed and constraints are introduced, and the optimal adjustment amount is obtained by solving the partial derivative equation group.

6. The method for efficiently processing enterprise resource planning data based on cloud computing according to claim 1, characterized in that: The execution layer implements data processing task tracking of cloud computing nodes based on a load balancing algorithm and outputs processing control instructions through the hierarchical execution control model to achieve efficient processing of enterprise resource planning data, including: Establishing a multi-node resource allocation model for a cloud computing platform, the multi-node resource allocation model includes a computing resource allocation equation and a storage resource allocation equation. The computing resource allocation equation includes the processing task amount, node processing capability, and task priority. The storage resource allocation equation includes the data storage amount, node storage capacity, and data access frequency. The multi-node resource allocation model is constructed as a state space expression, wherein the state vector of the state space expression includes computing resource occupancy, storage resource occupancy, task processing progress, and node load balance, and the control vector of the state space expression includes computing resource adjustment amount and storage resource adjustment amount; Linearizing the state space expression, calculating the partial derivatives of the system state equation with respect to the state vector and the control vector, and constructing a linearized prediction model; Constructing a task tracking prediction cost function, wherein the task tracking prediction cost function includes a tracking error term, a control amount penalty term, and a control increment penalty term, and performing a weighted combination of each penalty term through a weight matrix; Constructing state constraints, including resource occupancy constraints and task processing progress constraints; constructing control constraints, including computing resource adjustment constraints and storage resource adjustment constraints; constructing control increment constraints, including computing resource increment constraints and storage resource increment constraints; Converting the task tracking prediction cost function into a standard form of a quadratic programming problem, calculating the quadratic matrix and linear term coefficients, and constructing an inequality constraint matrix and an equality constraint matrix; The effective set method is used to solve the quadratic programming problem, and the optimal solution is gradually approached by identifying effective constraints and solving subproblems; Based on the optimization solution, control quantity mapping is performed, and the total computing resource adjustment quantity is allocated to each cloud computing node through the computing resource allocation matrix, and the total storage resource adjustment quantity is allocated to each node storage unit through the storage resource allocation matrix; The resource allocation of cloud computing nodes is limited. The computing resource allocation is limited according to the maximum processing capability of the node, and the storage resource allocation is limited according to the maximum storage capacity of the node, and the final processing control instructions are generated.

7. The method for efficiently processing enterprise resource planning data based on cloud computing according to claim 1, characterized in that: The enterprise resource planning data is preprocessed based on a hierarchical fusion framework to obtain a standardized data set, including: performing format conversion processing on the enterprise resource planning data collected from the financial system interface, the supply chain system interface, and the production management system interface to obtain a unified data format; performing outlier cleaning on the data in each format, calculating the data mean and variance through a sliding window, and eliminating abnormal data that exceeds the range; using a keyword matching algorithm to extract the core business information of each data, and establishing a corresponding relationship between data from different systems through information association; converting the associated business information into a unified data model, fusing multi-system data information through a weighted average method, and generating a standardized data set containing a timestamp.

8. The method for efficiently processing enterprise resource planning data based on cloud computing according to claim 2, characterized in that: Building a block processing mechanism includes: using a uniformly distributed initialization task set, where each task unit contains the business type and time requirements for data processing; calculating the node resource requirements corresponding to the task unit through the cloud computing platform resource allocation rules, mapping the task unit requirements to the resource capability table based on the interface parameter configuration, and generating a task unit resource requirement list; calculating the matching error between the task unit resource requirements and the actual resources of the node, and building business association rules based on the sum of squares of the error; obtaining the priority value of each task unit through normalization processing, retaining task units with priority values ​​greater than a preset threshold, and eliminating task units with priority values ​​less than the preset threshold.

9. The method for efficiently processing enterprise resource planning data based on cloud computing according to claim 3, characterized in that: Constructing the task transfer cost includes: calculating the processing time difference between the current node and the next node as the processing time cost; calculating the adjustment change of each resource required from the current node to the next node of the cloud computing platform as the resource adjustment cost; calculating the minimum distance with resource conflict in the scheduling plan, and increasing the penalty cost if it is less than the safety threshold; generating the total cost of task transfer by linearly combining the processing time cost, resource adjustment cost and penalty cost.

10. The method for efficiently processing enterprise resource planning data based on cloud computing according to claim 6, characterized in that: Solving the quadratic programming problem using the active set method includes: initializing the active constraint set and adding all constraints to the candidate set; solving the initial subproblem to obtain an initial feasible solution; calculating the constraint violation amount of the current solution and adding the constraint with the largest violation amount to the active set; resolving the subproblem and updating the current solution; repeating the above steps until the violation amount of all constraints is less than the preset precision, thereby obtaining the optimal solution.

Images