Project life cycle budget control method and device

By constructing a topological network and a long short-term memory neural network model for the project lifecycle, the problem of mismatch between budget allocation and project size and priority in the project lifecycle is solved, enabling precise allocation and dynamic adjustment of budget resources, and improving the control efficiency and adaptability of project management.

CN121766700APending Publication Date: 2026-03-31CHINA TOWER CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511979249.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-12-25
Publication Date
2026-03-31

AI Technical Summary

Technical Problem

Existing technologies struggle to effectively depict the complex, multi-dimensional relationships between project requirements and projects throughout the project lifecycle. This leads to a mismatch between budget allocation weights and the actual scale and priority of projects. Serial calculations and unified verification standards are prone to budget aggregation delays or blind spots in 'one-to-many' scenarios. Fixed fault tolerance thresholds cannot adapt to the different market environments and complexities of various projects. The lack of optimization strategies based on historical management results results in a lack of precise direction for budget deviation rectification, leading to insufficient overall management efficiency and adaptability.

Method used

Based on a pre-defined topological network, a multi-dimensional mapping relationship between requirements and project instances is constructed. A long short-term memory neural network prediction model is adopted, combined with dynamic fault tolerance thresholds and intelligent adjustment strategies, to achieve differentiated verification and accurate allocation of project budgets, generate budget aggregation verification results and consumption prediction results, locate the responsible parties for overspending and optimize the budget.

Benefits of technology

It enables precise allocation and dynamic adjustment of budget resources throughout the project lifecycle, improves the accuracy and adaptability of budget control, accurately identifies overspending responsibilities, predicts potential risks and generates targeted rectification plans, and enhances project progress efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121766700A_ABST
    Figure CN121766700A_ABST
Patent Text Reader

Abstract

The invention discloses a project life cycle budget control method and device, and relates to the technical field of project management, and the method comprises the steps: determining a target project instance set corresponding to a target demand based on a preset topological association network, obtaining a target topological association sub-network, carrying out the differentiation verification of the target project instance set based on execution state data, and obtaining a target project instance set; and if the budget is excessive, generating a budget aggregation check result based on the target project instance set, determining a project budget consumption prediction result by adopting a budget consumption prediction model based on the budget aggregation check result and the execution state data, and determining a budget consumption prediction result based on the project budget consumption prediction result and the budget aggregation check result. And adjusting the project data of each target project instance in the target topological association sub-network to obtain a target budget for executing each target project instance. According to the method and the device, the technical problem of relatively low accuracy of managing and controlling the budget in the project life cycle in related technologies is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of project management technology, and more specifically, to a method and apparatus for budget control throughout the project lifecycle. Background Technology

[0002] In the field of project management, such as tower construction and large-scale communication infrastructure, the entire project lifecycle needs to cover key aspects such as project initiation requirements, budget allocation, execution monitoring, and acceptance compliance. In actual business, there are often complex scenarios where a single project initiation requirement corresponds to multiple project instances. At the same time, the budget execution process is easily affected by factors such as fluctuations in raw material prices, changes in project technical complexity, and historical patterns of cost overruns. Therefore, the need for full-process control is proposed for the accurate allocation, real-time verification, risk warning, and deviation correction of budget resources to avoid problems such as budget mismatch and uncontrolled cost overruns affecting project progress efficiency.

[0003] In related technologies, project budget control schemes typically use relational databases to store unidirectional data linking project requirements and projects, allocating the total budget to each project according to a preset fixed ratio. They employ a serial calculation method to summarize the design amount for each project and perform budget verification based on a unified threshold, without distinguishing between "one-to-one" and "one-to-many" mapping differences. A fixed tolerance threshold is set based on the average overrun ratio of historical projects for budget compliance judgment during the acceptance phase. However, these schemes struggle to effectively depict the multi-dimensional and complex relationships between project requirements and projects, leading to a mismatch between budget allocation weights and the actual scale and priority of projects. Furthermore, serial calculations and unified verification standards are prone to budget aggregation delays or control blind spots in "one-to-many" scenarios. Fixed tolerance thresholds cannot adapt to the varying market environments and complexities of different projects, making it difficult to predict budget overrun risks in advance. Moreover, the lack of a mechanism for optimizing strategies based on historical control effects results in a lack of precise direction for budget deviation rectification, leading to insufficient overall control efficiency and adaptability.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This invention provides a method and apparatus for budget control throughout the project lifecycle, which at least addresses the technical problem of low accuracy in budget control throughout the project lifecycle in related technologies.

[0006] According to one aspect of the present invention, a budget control method for a project lifecycle is provided, comprising: determining a set of target project instances corresponding to target requirements based on a preset topology association network to obtain a target topology association subnetwork, wherein the preset topology association network records the relationship between each requirement and at least one project instance, each project instance corresponding to project data, the project data including at least: budget weight coefficients and execution status data, the execution status data including at least: budget amount, design amount, and actual amount of the project instance at each stage; performing differential verification on the set of target project instances based on the execution status data to obtain an initial budget verification result, and generating a budget aggregation verification result based on the set of target project instances if the initial budget verification result indicates budget overrun; determining a project budget consumption prediction result using a budget consumption prediction model based on the budget aggregation verification result and the execution status data; and adjusting the project data of each target project instance in the target topology association subnetwork based on the project budget consumption prediction result and the budget aggregation verification result to obtain a target budget for executing each target project instance.

[0007] Furthermore, before determining the set of target project instances corresponding to the target requirements based on the preset topology association network and obtaining the target topology association subnetwork, the process includes: collecting requirement data for each requirement and instance data for each project instance. The requirement data includes at least: requirement number, requirement type, total budget amount, and requirement priority; the instance data includes at least: project number, project size, project implementation period, and project-related requirement number. The requirement data and instance data are preprocessed to obtain target requirement data and target instance data. Each requirement is represented as a first-type node, and each project instance is represented as a second-type node. Based on the requirement number in the target requirement data corresponding to the first-type node and the project in the target instance data corresponding to the second-type node... Associating requirement numbers, constructing association edges between first-type nodes and second-type nodes, where each association edge has an attribute, including at least: mapping relationship type and association confidence; for each second-type node, determining the budget weight coefficient of the second-type node based on the target requirement data and the target instance data of the second-type node; for each project instance, determining the execution status data of the project instance at each stage based on the budget weight coefficient and the total budget amount, where the stage is a stage divided based on the life cycle of the project instance, and the stages are: project initiation stage, execution stage, and acceptance stage; based on all first-type nodes, all second-type nodes, all association edges, the budget weight coefficient of each second-type node, and the execution status data, a preset topology association network is constructed.

[0008] Further, the step of determining the budget weight coefficient of the second type of node based on the target demand data and the target instance data of the second type of node includes: determining the non-linear transformation value of project scale based on the project scale contained in the target instance data, determining the non-linear transformation value of demand priority based on the demand priority contained in the target demand data, and determining the non-linear transformation value of project cycle based on the project implementation cycle contained in the target instance data; determining the feature row matrix based on the non-linear transformation values ​​of project scale, demand priority, and project cycle; determining the probability of risk associated with project scale based on project scale, demand priority based on demand priority, and project cycle based on project implementation cycle; determining the risk probability column matrix based on the probability of risk associated with project scale, demand priority, and project cycle; determining the non-linear amplification value based on demand priority; and determining the budget weight coefficient of the second type of node based on the feature row matrix, the multi-feature collaborative weight matrix, the risk probability column matrix, the historical feedback dynamic adjustment value, and the non-linear amplification value, wherein the element values ​​in the multi-feature collaborative weight matrix are determined based on historical project instance data, and the historical feedback dynamic adjustment value is determined based on the execution status data of historical project instances.

[0009] Furthermore, the step of performing differentiated verification on the target project instance set based on execution status data to obtain the initial budget verification result includes: for each target project instance in the target project instance set, if the mapping relationship between the target project instance and the target requirement is a one-to-one mapping relationship, extracting the design amount and budget amount from the execution status data and comparing the design amount with the budget amount; if the design amount is less than or equal to the budget amount, determining the initial budget verification result as compliant, or if the design amount is greater than the budget amount, determining the initial budget verification result as budget overrun; if the mapping relationship between the target project instance and the target requirement is a one-to-many mapping relationship, determining the total design amount and the total budget amount of the target requirement based on the execution status data of all target project instances associated with the target requirement; if the total design amount is less than or equal to the total budget amount, determining the initial budget verification result as compliant, or if the total design amount is greater than the total budget amount, determining the initial budget verification result as budget overrun.

[0010] Furthermore, the step of generating a budget aggregation verification result based on the target project instance set includes: for each target project instance in the target project instance set, determining the contribution of the target project instance to the budget overrun based on the execution status data of the target project instance; identifying the target project instance with the largest contribution as the preset project instance, and generating an adjustment strategy for the preset project instance; and integrating the preset project instance and the adjustment strategy with the initial budget verification result to obtain the budget aggregation verification result.

[0011] Further, the steps for determining the contribution of target project instances to budget overruns based on the execution status data of target project instances include: determining the deviation percentage based on the design amount, budget amount, and the sum of positive deviations, where the sum of positive deviations is determined based on the overrun amount of all overrun project instances in the target requirements, and overrun project instances refer to target project instances whose design amount exceeds the budget amount; determining the project feature vector based on project complexity and the association confidence of the association edges between target project instances and target requirements, where project complexity is determined based on the project scale contained in the target instance data of target project instances; determining the strategic weight adjustment value based on the requirement priority contained in the target requirement data of target requirements; and determining the contribution of target project instances to budget overruns based on the deviation percentage, multi-dimensional weight matrix, project feature vector, historical risk correction value, and strategic weight adjustment value, where the elements in the multi-dimensional weight matrix include: project complexity weight, association strength weight, and interaction influence weight; and the historical risk correction value is determined based on the execution status data of historical project instances.

[0012] Furthermore, before determining the project budget consumption prediction result using the budget consumption prediction model based on the budget aggregation verification results and execution status data, the process includes: constructing an initial budget consumption prediction model based on a long short-term memory neural network; collecting historical instance data from multiple historical project instances and processing all historical instance data to obtain historical budget time series data; training the initial budget consumption prediction model using the historical budget time series data until the budget prediction value output by the initial budget consumption prediction model and the actual amount contained in the historical instance data are less than a preset threshold, thus obtaining the budget consumption prediction model.

[0013] Furthermore, the budget consumption prediction model includes at least the following steps: a long short-term memory neural network structure, and determining the project budget consumption prediction result based on the budget aggregation verification results and execution status data. These steps include: processing the budget aggregation verification results and execution status data to obtain budget time series data; inputting the budget time series data into the budget consumption prediction model to capture the dependency features in the budget time series data using the long short-term memory neural network structure to obtain the budget prediction value; for each target project instance, determining the dynamic fault tolerance threshold of the target project instance based on the project complexity, market volatility coefficient, and historical overspending probability using a dynamic fault tolerance threshold calculation strategy; and determining the project budget consumption prediction result based on the budget prediction value and dynamic fault tolerance threshold of each target project instance.

[0014] Furthermore, based on the project budget consumption prediction results and budget aggregation verification results, the steps for adjusting the project data of each target project instance in the target topology associated sub-network to obtain the target budget for executing each target project instance include: determining the target deviation type and target deviation level based on the budget aggregation verification results; determining the target template from a preset template library based on the target deviation type, wherein the preset template library includes: multiple suggested templates associated with the deviation type; adjusting the model parameters in the target template based on the target deviation level and project budget consumption prediction results to obtain a budget deviation rectification plan; generating a management and control effectiveness evaluation report based on the budget deviation rectification plan, budget aggregation verification results, and project budget consumption prediction results; and adjusting the project data of each target project instance based on the budget deviation rectification plan and management and control effectiveness evaluation report to obtain the target budget for executing each target project instance.

[0015] According to another aspect of the present invention, a budget control device for project lifecycle is also provided, comprising: a first determining unit, configured to determine a set of target project instances corresponding to target requirements based on a preset topology association network, thereby obtaining a target topology association subnetwork, wherein the preset topology association network records the relationship between each requirement and at least one project instance, each project instance corresponding to project data, the project data including at least: budget weight coefficients and execution status data, the execution status data including at least: budget amount, design amount, and actual amount of the project instance at each stage; a verification unit, configured to perform differential verification on the set of target project instances based on the execution status data, thereby obtaining an initial budget verification result, and, if the initial budget verification result indicates budget overrun, generate a budget aggregation verification result based on the set of target project instances; a second determining unit, configured to determine a project budget consumption prediction result using a budget consumption prediction model based on the budget aggregation verification result and the execution status data; and an adjustment unit, configured to adjust the project data of each target project instance in the target topology association subnetwork based on the project budget consumption prediction result and the budget aggregation verification result, thereby obtaining a target budget for executing each target project instance.

[0016] Furthermore, the budget control device also includes: a first acquisition module, used to acquire demand data for each demand and instance data for each project instance before determining the target project instance set corresponding to the target demand based on a preset topology association network and obtaining the target topology association subnetwork. The demand data includes at least: demand number, demand type, total budget amount, and demand priority; the instance data includes at least: project number, project size, project implementation period, and project-related demand number; a first processing module, used to preprocess the demand data and instance data to obtain target demand data and target instance data; a first representation module, used to represent each demand as a first-type node and each project instance as a second-type node; and a first construction module, used to construct the target demand data based on the demand number in the target demand data corresponding to the first-type node and the target project instance data corresponding to the second-type node. The project-related requirement numbers in the instance data are used to construct the association edges between the first type of nodes and the second type of nodes. Each association edge has an edge attribute, which includes at least: mapping relationship type and association confidence. The first determination module is used to determine the budget weight coefficient of each second type of node based on the target requirement data and the target instance data of the second type of node. The second determination module is used to determine the execution status data of each project instance at each stage based on the budget weight coefficient and the total budget amount. The stage is a stage divided based on the life cycle of the project instance, which is divided into: project initiation stage, execution stage, and acceptance stage. The first composition module is used to construct a preset topology association network based on all first type of nodes, all second type of nodes, all association edges, the budget weight coefficient of each second type of node, and the execution status data.

[0017] Further, the first determining module includes: a first determining submodule, used to determine a non-linear transformation value for project size based on the project size contained in the target instance data, a non-linear transformation value for requirement priority based on the requirement priority contained in the target requirement data, and a non-linear transformation value for project period based on the project implementation period contained in the target instance data; a second determining submodule, used to determine a feature row matrix based on the non-linear transformation values ​​for project size, requirement priority, and project period; and a third determining submodule, used to determine the probability of risk associated with project size based on project size, the probability of risk associated with requirement priority based on requirement priority, and the probability of risk associated with project implementation. The project cycle is determined in three sub-modules: the fourth sub-module determines the probability of risks associated with the project cycle; the fifth sub-module determines the non-linear amplification value based on the priority of the requirements; and the sixth sub-module determines the budget weight coefficient of the second type of node based on the feature row matrix, the multi-feature collaborative weight matrix, the risk probability column matrix, the historical feedback dynamic adjustment value, and the non-linear amplification value. The element values ​​in the multi-feature collaborative weight matrix are determined based on historical project instance data, and the historical feedback dynamic adjustment value is determined based on the execution status data of historical project instances.

[0018] Furthermore, the verification unit includes: a first extraction module, used for extracting the design amount and budget amount from the execution status data for each target project instance in the target project instance set, when the mapping relationship between the associated edge and the target requirement is a one-to-one mapping relationship, and comparing the design amount with the budget amount; a third determination module, used for determining the initial budget verification result as compliant when the design amount is less than or equal to the budget amount, or determining the initial budget verification result as budget overrun when the design amount is greater than the budget amount; a fourth determination module, used for determining the total design amount and the total budget amount of the target requirement based on the execution status data of all target project instances associated with the target requirement, when the mapping relationship between the associated edge and the target requirement is a one-to-many mapping relationship; and a fifth determination module, used for determining the initial budget verification result as compliant when the total design amount is less than or equal to the total budget amount, or determining the initial budget verification result as budget overrun when the total design amount is greater than the total budget amount.

[0019] Furthermore, the verification unit also includes: a sixth determining module, used to determine the contribution of each target project instance to the budget overrun based on the execution status data of the target project instance for each target project instance in the target project instance set; a seventh determining module, used to determine the target project instance with the largest contribution as the preset project instance and generate an adjustment strategy for the preset project instance; and a first integrating module, used to integrate the preset project instance and the adjustment strategy with the initial budget verification result to obtain the budget aggregate verification result.

[0020] Furthermore, the sixth determining module includes: a seventh determining submodule, used to determine the deviation percentage based on the design amount, budget amount, and the sum of positive deviations, wherein the sum of positive deviations is determined based on the overspending amount of all overspending project instances in the target requirements, and overspending project instances refer to target project instances whose design amount exceeds the budget amount; an eighth determining submodule, used to determine the project feature vector based on project complexity and the association confidence of the association edges between the target project instances and the target requirements, wherein project complexity is determined based on the project scale contained in the target instance data of the target project instances; a ninth determining submodule, used to determine the strategic weight adjustment value based on the requirement priority contained in the target requirement data of the target requirements; and a tenth determining submodule, used to determine the contribution of the target project instances to the budget overspending based on the deviation percentage, multi-dimensional weight matrix, project feature vector, historical risk correction value, and strategic weight adjustment value, wherein the elements in the multi-dimensional weight matrix include: project complexity weight, association strength weight, and interaction influence weight; and the historical risk correction value is determined based on the execution status data of historical project instances.

[0021] Furthermore, the budget control device also includes: a second construction module, used to construct an initial budget consumption prediction model based on a long short-term memory neural network before determining the project budget consumption prediction result using a budget consumption prediction model based on budget aggregation verification results and execution status data; a second acquisition module, used to acquire historical instance data of multiple historical project instances and process all historical instance data to obtain historical budget time series data; and a first training module, used to train the initial budget consumption prediction model using historical budget time series data until the budget prediction value output by the initial budget consumption prediction model and the actual amount contained in the historical instance data are less than a preset threshold, thus obtaining the budget consumption prediction model.

[0022] Furthermore, the budget consumption prediction model includes at least a long short-term memory neural network structure, and the second determining unit includes: a second processing module for processing the budget aggregation verification results and execution status data to obtain budget time series data; a first capturing module for inputting the budget time series data into the budget consumption prediction model to capture the dependency features in the budget time series data using the long short-term memory neural network structure to obtain the budget prediction value; an eighth determining module for determining the dynamic fault tolerance threshold of each target project instance based on the project complexity, market volatility coefficient, and historical overspending probability of the target project instance, using a dynamic fault tolerance threshold calculation strategy; and a ninth determining module for determining the project budget consumption prediction result based on the budget prediction value and dynamic fault tolerance threshold of each target project instance.

[0023] Furthermore, the adjustment unit includes: a tenth determination module, used to determine the target deviation type and target deviation level based on the budget aggregation verification results; an eleventh determination module, used to determine the target template from a preset template library based on the target deviation type, wherein the preset template library includes: multiple suggested templates associated with the deviation type; a first adjustment module, used to adjust the model parameters in the target template based on the target deviation level and the project budget consumption prediction results to obtain a budget deviation rectification plan; a first generation module, used to generate a management and control effectiveness evaluation report based on the budget deviation rectification plan, the budget aggregation verification results, and the project budget consumption prediction results; and a second adjustment module, used to adjust the project data of each target project instance based on the budget deviation rectification plan and the management and control effectiveness evaluation report to obtain the target budget for executing each target project instance.

[0024] According to another aspect of the present invention, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements a project lifecycle budget control method as described above.

[0025] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement any of the above-described project lifecycle budget control methods.

[0026] In this invention, based on a preset topology association network, a set of target project instances corresponding to the target requirements is determined, resulting in a target topology association subnetwork. Based on execution status data, differential verification is performed on the set of target project instances to obtain an initial budget verification result. If the initial budget verification result indicates that the budget is over-limited, a budget aggregation verification result is generated based on the set of target project instances. Based on the budget aggregation verification result and execution status data, a budget consumption prediction model is used to determine the project budget consumption prediction result. Based on the project budget consumption prediction result and the budget aggregation verification result, the project data of each target project instance in the target topology association subnetwork is adjusted to obtain the target budget for executing each target project instance. This solves the technical problem of low accuracy in budget control during the project lifecycle in related technologies.

[0027] In this invention, a pre-defined topology network is constructed. By determining the set of target project instances corresponding to the target requirements and forming a target topology network, the complex relationships between requirements and project instances are accurately mapped. This achieves the technical effect of precise allocation of budget resources throughout the project lifecycle, thereby solving the technical problem of mismatch between budget allocation and actual needs such as project size and priority in related technologies. Specifically, the pre-defined topology network records the relationship between each requirement and at least one project instance. Each project instance corresponds to project data, including budget weight coefficients and execution status data. The execution status data covers the budget amount, design amount, and actual amount of the project instance at each stage. Based on the execution status data, differential verification is performed on the set of target project instances to obtain an initial budget verification result. If budget overruns are found, a budget aggregation verification result is further generated to locate the responsible party for the overrun. Then, combining the budget aggregation verification result and the execution status data, a budget consumption prediction model is used to determine the project budget consumption prediction result to predict potential risks. Subsequently, based on the project budget consumption prediction results and budget aggregation verification results, the project data of each target project instance in the target topology associated sub-network is intelligently adjusted to obtain the optimized budget for executing each target project instance, thus realizing the foresight and adaptability of budget control. Attached Figure Description

[0028] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0029] Figure 1 This is a flowchart of an optional project lifecycle budget control method according to an embodiment of the present invention;

[0030] Figure 2This is a schematic diagram of an optional intelligent management and control process for project lifecycle budget based on multidimensional association mapping and dynamic threshold learning according to an embodiment of the present invention;

[0031] Figure 3 This is a schematic diagram of an optional project lifecycle budget control device according to an embodiment of the present invention;

[0032] Figure 4 This is a hardware structure block diagram of an electronic device (or mobile device) for a budget control method for the project lifecycle according to an embodiment of the present invention. Detailed Implementation

[0033] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0034] It should be noted that the terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0035] It should be noted that all related information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, and displayed data) collected and involved in this invention are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of this data comply with the relevant laws, regulations, and standards of the relevant regions, necessary confidentiality measures have been taken, and it does not violate public order and good morals. Corresponding operation entry points are provided for users to choose to authorize or refuse. For example, this system has an interface with relevant users or organizations. Before obtaining relevant information, a request to obtain the information needs to be sent to the aforementioned user or organization through the interface, and the relevant information is obtained only after receiving consent from the aforementioned user or organization.

[0036] This invention proposes an intelligent project budget management method based on multidimensional association mapping and dynamic threshold learning, addressing the following issues in current project budget management technologies: First, how to accurately construct a multidimensional and complex mapping relationship between project requirements and project instances, and dynamically calculate budget allocation weights based on actual project characteristics to solve the mismatch between budget allocation and actual needs such as project scale and priority; Second, how to efficiently identify all related projects under the same requirement, perform differentiated budget verification for different mapping relationships ("one-to-one" and "one-to-many"), and accurately locate the main responsible parties for budget overruns to solve the problems of delayed budget aggregation, poor adaptability to management scenarios, and ambiguous responsibility definition; Third, how to combine historical budget execution data, market volatility coefficients, and project complexity factors to achieve accurate prediction of the final actual amount incurred by the project and intelligent generation of dynamic fault tolerance thresholds to solve the problems of fixed thresholds being difficult to adapt to scenario differences and delayed budget risk prediction; Fourth, how to construct a closed-loop optimization system for budget management strategies, iteratively optimize verification rules and threshold parameters based on full-link management effectiveness feedback, and generate targeted rectification plans to solve the problems of rigid management strategies and insufficient accuracy in deviation rectification.

[0037] The present invention will now be described in detail with reference to various embodiments.

[0038] Example 1

[0039] According to an embodiment of the present invention, an embodiment of a budget control method for project lifecycle is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0040] Figure 1 This is a flowchart of an optional project lifecycle budget control method according to an embodiment of the present invention, such as... Figure 1 As shown, the method includes the following steps:

[0041] Step S101: Based on the preset topology association network, determine the set of target project instances corresponding to the target requirements and obtain the target topology association sub-network. The preset topology association network records the relationship between each requirement and at least one project instance. Each project instance corresponds to project data, which includes at least the budget weight coefficient and execution status data. The execution status data includes at least the budget amount, design amount and actual amount of the project instance at each stage.

[0042] In this embodiment of the invention, the preset topology association network is constructed based on graph database technology, recording the multi-dimensional mapping relationship between each project requirement and project instance. In the network, requirement documents are mapped as first-type nodes, and project instances are mapped as second-type nodes, connected by association edges (including mapping relationship type and association confidence attributes). When a requirement document is determined to be the current processing target (i.e., the target requirement), the association subgraph traversal algorithm can be started. Starting from the target requirement node, the algorithm traverses the project instance nodes directly connected to it to obtain the target project instance set, thereby generating the target topology association subnetwork. Here, each project instance corresponds to project data, which includes: budget weight coefficients (calculated through a dynamic weight allocation algorithm, reflecting characteristic factors such as project scale, requirement priority, and implementation cycle, used to guide the allocation of budget among project instances) and execution status data. The execution status data records the budget amount, design amount, and actual amount of the project instance at different stages, used for real-time monitoring of budget execution and verification of overspending risk.

[0043] Step S102: Based on the execution status data, perform differential verification on the target project instance set to obtain the initial budget verification result. If the initial budget verification result indicates that the budget is over-limited, generate the budget aggregation verification result based on the target project instance set.

[0044] In this embodiment of the invention, execution status data for each target project instance in the target project instance set can be extracted, and the design amounts of each project can be aggregated in parallel using a distributed real-time computing framework. For a one-to-one mapping relationship, the design amount of a single project is directly compared with the budget amount; for a one-to-many mapping, it is verified whether the sum of the design amounts of all projects exceeds the total budget of the requirement order, and an initial budget verification result is generated. When an over-budget situation is found, the contribution decomposition algorithm is further used to quantify the responsibility for the overspending and generate a budget aggregation verification result, clarifying the main responsible projects and adjustment suggestions, laying the foundation for accurate correction.

[0045] Step S103: Based on the budget aggregation verification results and execution status data, the budget consumption prediction model is used to determine the project budget consumption prediction results.

[0046] In this embodiment of the invention, a budget consumption prediction model can be trained using a Long Short-Term Memory (LSTM) neural network based on historical project data from project execution status data. The design amount of the target project instance and the current execution data are input into the model, which outputs the predicted value and confidence interval of the project's final actual expenditure, providing a quantitative basis for risk prediction.

[0047] Step S104: Based on the project budget consumption prediction results and budget aggregation verification results, adjust the project data of each target project instance in the target topology associated sub-network to obtain the target budget for executing each target project instance.

[0048] In this embodiment of the invention, adjustment suggestions are generated by combining the budget aggregation verification results with the project budget consumption prediction results through a multi-path intelligent deviation correction recommendation engine. For example, after identifying projects predicted to be over budget, strategies such as design list optimization and budget adjustment are formulated according to the type and degree of deviation. Then, the verification strategy and threshold parameters are optimized through reinforcement learning algorithms and fed back to the topology association network to achieve dynamic budget adjustment.

[0049] In summary, a pre-defined topology network can be constructed to accurately map the complex relationships between requirements and project instances by identifying the set of target project instances corresponding to the target requirements and forming a target topology network. This achieves the technical effect of precise allocation of budget resources throughout the project lifecycle, thereby solving the technical problem of mismatch between budget allocation and actual needs such as project scale and priority in related technologies. Specifically, the pre-defined topology network records the relationship between each requirement and at least one project instance. Each project instance corresponds to project data, including budget weight coefficients and execution status data. The execution status data covers the budget amount, design amount, and actual amount of the project instance at each stage. Based on the execution status data, differential verification is performed on the set of target project instances to obtain the initial budget verification result. If budget overruns are found, a budget aggregation verification result is further generated to locate the responsible party for the overrun. Then, combining the budget aggregation verification result and the execution status data, a budget consumption prediction model is used to determine the project budget consumption prediction result to predict potential risks. Subsequently, based on the project budget consumption prediction results and budget aggregation verification results, the project data of each target project instance in the target topology associated sub-network is intelligently adjusted to obtain the optimized budget for executing each target project instance, thus realizing the foresight and adaptability of budget control.

[0050] To improve the accuracy of constructing the preset topology network, in the project lifecycle budget control method provided in Embodiment 1 of this application, before determining the target project instance set corresponding to the target requirement based on the preset topology network and obtaining the target topology subnetwork, requirement data for each requirement and instance data for each project instance are collected. The requirement data includes at least: requirement number, requirement type, total budget amount, and requirement priority; the instance data includes at least: project number, project size, project implementation period, and project-related requirement number. The requirement data and instance data are preprocessed to obtain target requirement data and target instance data. Each requirement is represented as a first-type node, and each project instance is represented as a second-type node. Based on the requirement number in the target requirement data corresponding to the first-type node and... The project-related requirement numbers in the target instance data corresponding to the second type of nodes are used to construct the association edges between the first type of nodes and the second type of nodes. Each association edge has an edge attribute, which includes at least: mapping relationship type and association confidence. For each second type of node, the budget weight coefficient of the second type of node is determined based on the target requirement data and the target instance data of the second type of node. For each project instance, the execution status data of the project instance at each stage is determined based on the budget weight coefficient and the total budget amount. The stage is divided into stages based on the life cycle of the project instance, and the stages are: project initiation stage, execution stage, and acceptance stage. Based on all first type of nodes, all second type of nodes, all association edges, the budget weight coefficient of each second type of node, and the execution status data, a preset topology association network is constructed.

[0051] In this embodiment of the invention, the original project requirement form data (requirement data) and original project instance data (instance data) collected in a project management scenario are used as the processing objects. The original project requirement form data includes fields such as requirement number, requirement type, total budget amount, requirement priority, and requirement submission time. The original project instance data includes fields such as project number, project name, project scale (e.g., quantity of construction content, implementation scope), project implementation cycle, project-related requirement number, and project manager. Data preprocessing is performed on the original project requirement form data and original project instance data. Specific processing actions include: removing duplicate data entries, completing missing key fields (e.g., completing the total budget amount field in the project instance data by associating it with the requirement number), and standardizing the data format (e.g., standardizing the budget amount unit to RMB yuan and the date format to year-month-day). After processing, standardized project requirement form data (target requirement data) and standardized project instance data (target instance data) are obtained.

[0052] Using standardized project requirement forms and standardized project instance data as processing objects, a multi-dimensional mapping relationship between the two is constructed using graph database technology: In the graph database, standardized project requirement forms are used as the first type of nodes, with node attributes including requirement number, requirement type, total budget amount, and requirement priority; standardized project instance data are used as the second type of nodes, with node attributes including project number, project name, project scale, project implementation period, and project-related requirement number; based on the "project-related requirement number" in the standardized project instance data and the "requirement number" in the standardized project requirement form data, an association edge is established between the nodes. The attributes of the association edge are set as mapping relationship type (values ​​are "one-to-one" or "one-to-many") and association confidence (calculated by the number of matching fields between the requirement and the project; the more matching fields, the higher the association confidence). After processing, an initial requirement-project association graph (preset topology association network) is obtained.

[0053] Using the attribute data of the second type of nodes (standardized project instance data nodes) in the initial demand-project relationship graph as the processing object, a dynamic weight allocation algorithm is used to calculate the weight coefficient of each project in the total budget. First, feature factors for weight calculation are extracted from the attributes of the second type of nodes, including project size (converted to quantitative values ​​according to preset grading standards, such as 3 for large projects, 2 for medium projects, and 1 for small projects), demand priority (3 for high priority, 2 for medium priority, and 1 for low priority), and project implementation period (converted to quantitative values ​​according to the number of months of implementation, such as 3 for 12 months and above, 2 for 6-11 months, and 1 for less than 6 months). Then, a weight percentage is set for each feature factor (the sum of the weight percentages of each factor is 1, which can be adjusted according to the needs of the project management scenario, such as 0.4 for project size, 0.3 for demand priority, and 0.3 for project implementation period). For example, the weighting coefficient of each project instance can be calculated using the weighted summation formula (Project Budget Weighting Coefficient = Quantified Value of Project Size × 0.4 + Quantified Value of Requirement Priority × 0.3 + Quantified Value of Project Implementation Cycle × 0.3). After processing, the project budget weighting coefficient is obtained, which will be used in subsequent steps to calculate the allocation of the total budget among the projects.

[0054] Using standardized project instance data and the requirement-project initial relationship diagram as the processing objects, a budget flow status awareness engine is employed to track the budget execution status of each project stage. First, based on the project management process, the entire project lifecycle is divided into three fixed stages: initiation, execution, and acceptance. Then, for each stage, the corresponding budget amount (the budget amount for the initiation stage is the total requirement budget multiplied by the project budget weighting coefficient), design amount (estimated costs in the project design scheme generated at the beginning of the execution stage), and actual expenditure amount (actual expenses incurred after each stage) are extracted from the standardized project instance data. The budget amount, design amount, and actual expenditure amount for each stage are then associated and stored with the corresponding project number and stage name. After processing, the budget execution status data for each project stage is obtained.

[0055] Using the initial demand-project relationship diagram and the budget execution status data for each project stage as the processing objects, the "project number" in the budget execution status data for each project stage is associated with the "project number" of the second type of node in the initial demand-project relationship diagram. The budget execution status of each project is added to the initial demand-project relationship diagram as a supplementary attribute of the second type of node. After processing, a demand-project topology relationship network is obtained. Thus, through the above continuous processing, three types of results are finally generated: the demand-project topology relationship network, project budget weight coefficients, and budget execution status data for each project stage. These three types of results will be used as processing objects in subsequent technical processes.

[0056] In some optional embodiments, the selection of feature factors for the dynamic weight allocation algorithm differs from the traditional budget allocation method that relies solely on the scale of a single project. It adds two types of feature factors: demand priority and project implementation cycle. This is because in project management scenarios, projects with higher demand priority usually require priority budget allocation, while projects with longer implementation cycles may face more uncertainties and require more reasonable budget space. This design makes the calculation of project budget weight coefficients more in line with actual management needs. At the same time, the application of graph database technology differs from the traditional linear association model. Through the structured modeling method of node-edge-attribute, it can intuitively and efficiently store two complex mapping relationships: "one-to-one" and "one-to-many," avoiding the budget management blind spots that occur in the "one-to-many" scenario in the traditional model.

[0057] In some alternative embodiments, for the raw data preprocessing stage, if the raw project instance data contains missing "project-related requirement numbers", a semantic matching method based on project name and requirement type is used to complete the missing numbers: First, a dictionary of corresponding keywords for requirement types and project names is constructed (e.g., if the requirement type is "communication base station construction", the corresponding project name keywords are "base station" and "communication tower"). Then, the semantic similarity between the project name with missing requirement numbers and the corresponding keywords for each requirement type is calculated. The requirement number corresponding to the requirement type with the highest semantic similarity is taken as the project's related requirement number. After completion, subsequent graph database modeling is performed. This processing method can reduce the problem of incomplete relationship establishment caused by missing data and improve the completeness of the initial requirement-project relationship graph.

[0058] To improve the accuracy of determining the budget weight coefficients for the second type of nodes, the project lifecycle budget control method provided in Embodiment 1 of this application determines the nonlinear transformation value of the project scale based on the project scale contained in the target instance data, the nonlinear transformation value of the demand priority based on the demand priority contained in the target demand data, and the nonlinear transformation value of the project cycle based on the project implementation cycle contained in the target instance data; a feature row matrix is ​​determined based on the nonlinear transformation values ​​of the project scale, demand priority, and project cycle; the probability of project scale-related risks is determined based on the project scale, the probability of demand priority-related risks is determined based on the demand priority, and the probability of project cycle-related risks is determined based on the project implementation cycle; a risk probability column matrix is ​​determined based on the probability of project scale-related risks, the probability of demand priority-related risks, and the probability of project cycle-related risks; a nonlinear amplification value is determined based on the demand priority; and the budget weight coefficients for the second type of nodes are determined based on the feature row matrix, the multi-feature collaborative weight matrix, the risk probability column matrix, the historical feedback dynamic adjustment value, and the nonlinear amplification value. The element values ​​in the multi-feature collaborative weight matrix are determined based on historical project instance data, and the historical feedback dynamic adjustment value is determined based on the execution status data of historical project instances.

[0059] In this embodiment of the invention, the dynamic weight allocation algorithm needs to break through the limitations of the traditional "linear summation + simple product". In the project budget allocation, the importance of features is not linearly superimposed (such as the budget tilt of high priority projects should be "non-linearly amplified"), the risk is not fixed and quantified (such as the market fluctuation risk of long-term projects needs to be described by a probability model), and historical feedback needs to reflect "deviation volatility" (rather than a single deviation rate). This embodiment introduces non-linear transformation function (Sigmoid), probability distribution function (normal CDF (Cumulative Distribution Function)), matrix operation, root mean square error (RMSE) and other operations to construct a multi-dimensional, probabilistic, and dynamically closed-loop weight calculation model.

[0060] Specifically, the formula for calculating the project budget weighting coefficient (budget weighting coefficient for the second type of node) is as follows:

[0061] ;

[0062] in, This is the feature row matrix (1×3 dimensions) after nonlinear transformation, reflecting the nonlinear importance of the features (feature row matrix). It is a multi-feature collaborative weight matrix (3×3 dimension), which reflects the complex interaction effects between features (multi-feature collaborative weight matrix). The risk probability column matrix (3×1 dimension) quantifies the probability of risk occurrence associated with each feature (risk probability column matrix). For historical feedback dynamic adjustment term (based on the exponential decay function of deviation volatility) (historical feedback dynamic adjustment value); This is a priority non-linear amplification term (Sigmoid function, amplifying the weight tilt of high-priority items) (non-linear amplification value). The budget weighting coefficient for a single project (value range [0,1], after normalization to ensure that the sum of the weights of all projects under the same demand order is 1).

[0063] (1) Characteristic row matrix after nonlinear transformation :

[0064] ,in, (Logarithmic transformation function, transforming the original eigenvalues) (1-3) Normalize to the 0-1 interval to smooth out the effects of extreme values):

[0065] Non-linear transformation value of project size: Original project size (1 = small, 2 = medium, 3 = large) After logarithmic transformation, the weighting increase for large projects slows down, avoiding excessive budget bias caused by a "scale-only" approach (e.g.) hour, , hour, Data source: "Project Scale" extracted from the second type of node attributes in the initial requirements-project relationship graph. Substitute the values ​​into the logarithmic function for calculation. This solves the problem in traditional linear quantification where "large projects have excessively high weightings, squeezing the budgets of small and medium-sized strategic projects," and achieves smooth normalization of eigenvalues.

[0066] Demand priority non-linear transformation value: original priority (1=low, 2=medium, 3=high) After logarithmic transformation, the differences in high-priority items are more significant (e.g., hour, , hour, This lays the foundation for subsequent Sigmoid amplification. Data source: "Requirement Priority" extracted from the first type of node attributes in the initial requirement-project relationship graph. Substitute into the logarithmic function to calculate.

[0067] Project cycle nonlinear transformation value: original cycle (1 = short-term, 2 = medium-term, 3 = long-term) After logarithmic transformation, the risk baseline for long-term projects is more reasonable (e.g., hour , hour This avoids weight distortion caused by extreme values ​​in the cycle. Data source: "Project Cycle" extracted from the second type of node attributes in the initial requirement-project relationship graph. Substitute into the logarithmic function to calculate.

[0068] (2) Multi-feature collaborative weight matrix (3×3 dimension):

[0069] Among them, diagonal elements For the feature's own weight, off-diagonal elements For the interaction weights between features, satisfying Default value: .

[0070] Matrix elements embody the dual importance of features themselves and their interactions, such as (Priority has the highest weight, highlighting strategic orientation). (The cycle itself has the second highest weight, highlighting the risk orientation). (Priority and cycle interaction weights reflect the collaborative needs of high-priority long-term projects). Data source: Obtained through "feature-budget deviation" correlation analysis of historical project data (e.g., using a random forest model to fit the relationship between features and budget rationality, and deriving the weight matrix elements in reverse). Thus, it breaks through the limitations of traditional "single feature weight superposition," using matrix operations to describe the complex synergistic effects of multiple features, such as high priority (…). And long-term ( ) project, through The synergistic effect of these factors amplifies the weights in multiple dimensions, making them more in line with actual management and control needs.

[0071] (3) Risk probability column matrix (3×1 dimension): Where each element represents a risk probability based on the cumulative distribution function (CDF) of a normal distribution:

[0072] The probability of risk associated with project size is represented by the following formula: ,in, For the standard normal CDF, (Average size of historical projects) (Standard deviation of size). For example, large projects ( ) risk probability (High risk), small projects ( ) risk probability (Low risk) The larger the scale of quantification, the higher the risk. Data source: This was obtained by statistically analyzing the historical project size distribution from the budget execution status data at each stage of the project.

[0073] This indicates the probability of risk associated with demand priority, calculated using the following formula: , , For example, high-priority projects ( ) risk probability (Low risk, due to priority resource allocation), low priority projects ( ) risk probability (High risk, as resources may be squeezed out).

[0074] The probability of risk associated with the project cycle is represented by the following formula: , , For example, long-term projects ( The probability of risk for short-term projects is approximately 0.89 (high risk, high probability of market volatility). The probability of risk is approximately 0.11 (low risk). Therefore, risk can be quantified using a probability distribution function, overcoming the limitations of traditional "qualitative description of risk" and allowing for precise calculation of the impact of risk on weights.

[0075] (4) Historical feedback dynamic adjustment items The calculation formula is as follows: ,in, The number of similar projects in history, For the first "Actual execution amount / budgeted amount" (deviation rate) for each historical project. The mean of the deviation rate; This is the deviation sensitivity coefficient (industry-appropriate parameter). If the deviation of similar projects in the past fluctuated greatly ( ), then the adjustment item (Weight reduction to avoid repeating the volatility pattern); if the deviation fluctuation is small ( Then the adjustment item ≈ (Weight fine-tuning, trusting historical stability). Data source: The calculation is performed by extracting the actual amount and budgeted amount of similar historical projects from the budget execution status data at each stage of the project. Thus, RMSE is introduced to quantify the volatility of deviations (rather than a single deviation rate), and an exponential function is used to reflect the logic of "the greater the volatility, the more conservative the weighting," enabling refined adjustments based on historical feedback.

[0076] (5) Priority nonlinear amplification term The calculation formula is as follows: This is the amplification factor (industry-adapted parameter). This is the average priority of historical projects. For example, high-priority projects ( () magnification term (Significantly amplify weighting); low-priority items ( The magnified term of ) ≈ (Appropriately reduce weight); Medium priority items ( The amplification term is set to 0.5 (no amplification). Thus, the Sigmoid function is used to achieve "high-priority nonlinear amplification and low-priority moderate compression," solving the problem of "insignificant priority differences" in traditional linear computation and highlighting strategic orientation.

[0077] In this embodiment, nonlinear transformation (logarithmic function) can smooth extreme eigenvalues ​​and avoid weight distortion; probabilistic risk (normal CDF) can transform risk from "qualitative description" to "quantitative calculation", improving accuracy by more than 60%; matrix interaction (3×3 weight matrix) can use matrix operations to describe the synergy of multiple features, covering the "scale-risk-priority" linkage relationship that traditional methods cannot reflect; dynamic feedback (RMSE + exponential function) not only considers the deviation rate but also the deviation volatility, making weight adjustment more robust; strategic amplification (Sigmoid function) can amplify the weight of high-priority projects by more than 4 times that of low-priority projects, highlighting strategic orientation.

[0078] To improve the accuracy of determining the initial budget verification result, in the project lifecycle budget control method provided in Embodiment 1 of this application, for each target project instance in the target project instance set, if the mapping relationship between the target project instance and the target requirement is a one-to-one mapping, the design amount and budget amount are extracted from the execution status data, and the design amount and budget amount are compared; if the design amount is less than or equal to the budget amount, the initial budget verification result is determined to be compliant, or if the design amount is greater than the budget amount, the initial budget verification result is determined to be budget overrun; if the mapping relationship between the target project instance and the target requirement is a one-to-many mapping, the total design amount is determined based on the execution status data of all target project instances associated with the target requirement, and the total budget amount of the target requirement is determined; if the total design amount is less than or equal to the total budget amount, the initial budget verification result is determined to be compliant, or if the total design amount is greater than the total budget amount, the initial budget verification result is determined to be budget overrun.

[0079] In this embodiment of the invention, the demand-project topology association network and the budget execution status data of each stage of the project are used as the initial processing objects. The demand-project topology association network includes first-type nodes (project request single nodes, with attributes including demand number and total demand budget), second-type nodes (project instance nodes, with attributes including project number), and associated edges (with attributes including mapping relationship type). The budget execution status data of each stage of the project includes project number, budget amount, and design amount. First, the demand-project topology association network is processed by an associated subgraph traversal algorithm. Taking the first-type nodes (project request single nodes) as the traversal starting point, all directly connected second-type nodes (project instance nodes) are traversed according to the associated edges between nodes (edges connecting project request singles and project instances). All second-type nodes associated with the same first-type node are grouped together. After processing, associated project grouping data divided by demand singles is obtained. This associated project grouping data clarifies all project instances corresponding to each demand single.

[0080] Then, taking the associated project grouping data and the budget execution status data of each stage of the project as the processing objects, the design amount of the corresponding project is extracted from the budget execution status data of each stage of the project according to the project number in the associated project grouping data. Then, the extracted design amount is grouped and summarized using a distributed real-time computing framework. Each group in the associated project grouping data (the project group corresponding to the same demand order) is treated as a computing task and distributed to different computing nodes to perform the design amount summation operation in parallel. After processing, the project design amount summary data is obtained. The project design amount summary data includes the total design amount of all projects corresponding to each demand order and the design amount of a single project.

[0081] Next, using the project design amount summary data and the association edge attributes (mapping relationship type) in the demand-project topology association network as the processing object, a differentiated verification strategy is executed: For demand unit-project combinations with a "one-to-one" mapping of association edge attributes, the design amount of a single project corresponding to the combination (from the project design amount summary data) and the budget amount of the project (from the budget execution status data of each stage of the project) are extracted. The two values ​​are directly compared for verification. If the project design amount ≤ the budget amount, it is marked as "verification compliant"; if the project design amount > the budget amount, it is marked as "verification exceeding budget". For demand unit-project combinations with a "one-to-many" mapping of association edge attributes, the total project design amount corresponding to the demand unit (from the project design amount summary data) and the total budget amount of the demand unit (from the first type of node attribute in the demand-project topology association network) are extracted for verification. The total project design amount is verified by checking whether it ≤ the total budget amount of the demand unit. If the total ≤ the total budget amount, it is marked as "verification compliant"; if the total > the total budget amount, it is marked as "verification exceeding budget". After the above processing is completed, the preliminary budget verification result (initial budget verification result) is obtained. The preliminary budget verification result includes the verification status (compliant / over budget) of each requirement order-project combination and the total overspending amount when the budget is over (in the case of overspending, the difference between the design amount and the budget amount or the difference between the total design amount and the total budget amount of the requirement order).

[0082] To improve the accuracy of the generated budget aggregation verification results, in the budget control method for the project lifecycle provided in Embodiment 1 of this application, for each target project instance in the target project instance set, the contribution of the target project instance to the budget overrun is determined based on the execution status data of the target project instance; the target project instance with the largest contribution is determined as the preset project instance, and an adjustment strategy is generated for the preset project instance; the preset project instance and the adjustment strategy are integrated with the initial budget verification results to obtain the budget aggregation verification results.

[0083] In this embodiment of the invention, the record marked as "over-budget verification" in the preliminary budget verification results is taken as the processing object, and the contribution decomposition algorithm is started for processing: First, the total over-budget amount corresponding to the over-budget record is calculated (from the preliminary budget verification results). Then, the difference between the design amount and the budget amount of each project under the over-budget record is extracted (the project design amount comes from the project design amount summary data, and the budget amount comes from the budget execution status data of each stage of the project). By calculating the proportion of the over-budget amount of a single project to the total over-budget amount (over-budget contribution = (over-budget amount of a single project / total over-budget) × 100%), the projects with the highest over-budget contribution ranking (usually projects whose total contribution reaches a preset proportion are selected, and this proportion can be set according to management requirements) are selected as the main responsible projects for over-budget. At the same time, based on the over-budget amount and project attributes of the main responsible projects (such as project scale, design content, and the second type of node attributes from the demand-project topology association network), targeted adjustment suggestions are generated (such as reducing non-core design content and optimizing design schemes to reduce costs). After processing, the main responsible projects for over-budget and adjustment suggestions are obtained.

[0084] Then, taking the preliminary budget verification results and the main responsible items and adjustment suggestions for exceeding the budget as the processing objects, the main responsible items and adjustment suggestions for exceeding the budget are associated and integrated with the corresponding over-budget records in the preliminary budget verification results, and supplemented into the preliminary budget verification results. After processing, a budget aggregate verification result (including the verification status of all requirement orders-project combinations, the total over-budget amount and the corresponding main responsible items) and the main responsible items and adjustment suggestions for exceeding the budget are generated. Among them, the budget aggregate verification result will be used as the processing object of "project design amount in budget aggregate verification result", and the main responsible items and adjustment suggestions for exceeding the budget will be used as the processing object of "deviation type and degree in main responsible items and adjustment suggestions for exceeding the budget", and participate in the subsequent technical process.

[0085] In some optional embodiments, the multi-level budget aggregation verification mechanism differs from the traditional "single-point verification" approach to budget control. Traditional methods only independently verify the budget execution of a single project, failing to correlate multiple projects under the same demand order, easily leading to control loopholes such as "individual projects being compliant but the overall budget exceeding the limit." This embodiment first groups projects using an associated subgraph traversal algorithm, then uses a distributed real-time computing framework for parallel aggregation, and finally performs differentiated verification based on mapping relationship type, forming a multi-level processing flow of "grouping-aggregation-differential verification." This design can cover overall budget control in "one-to-many" mapping scenarios, avoiding blind spots in traditional control. Simultaneously, the application of the contribution decomposition algorithm differs from the traditional system's approach of only providing "budget overrun" prompts. Traditional methods cannot pinpoint specific overrun responsible projects, resulting in a lack of clear direction and low efficiency during rectification. This embodiment quantifies the overrun impact of each project by calculating the overrun contribution metric, accurately identifying the main responsible projects and generating adjustment suggestions. Its design aims to solve the problem of low efficiency in traditional rectification, providing a clear basis for subsequent intelligent correction. This quantitative analysis method represents a targeted technical improvement in the field of budget control.

[0086] In some alternative embodiments, the processing of the distributed real-time computing framework can be segmented by "requirement order number" to group related project data. Each segment contains information on all project instances corresponding to a single requirement order, and the segmented data is distributed to different computing nodes. Each computing node extracts the design amount of the corresponding project within the segment from the budget execution status data of each stage of the project, and performs a summation operation. For example, if a requirement order corresponds to 3 projects with design amounts of X1, X2, and X3, the computing node outputs the total design amount of the projects in the requirement order as X1 + X2 + X3. After the calculation is completed, each node summarizes the results to the master node to generate summary data of project design amounts. In the specific application of the contribution decomposition algorithm, if the total overspending of a certain requirement order is Y, and the overspending of the corresponding 3 projects is Y1, Y2, and Y3 (Y=Y1+Y2+Y3), then the overspending contribution of the 3 projects is (Y1 / Y)×100%, (Y2 / Y)×100%, and (Y3 / Y)×100%, respectively. If the preset total contribution threshold is 80%, and the sum of the contribution of the first 2 projects reaches 85%, then these 2 projects are identified as the main responsible projects for the overspending, and an adjustment suggestion is generated based on their project size (such as medium-sized projects): "Optimize the design list of medium-sized projects and delete 2 unnecessary construction contents to reduce the design cost."

[0087] To improve the accuracy of determining the contribution of target project instances to budget overruns, the budget control method for the project lifecycle provided in Embodiment 1 of this application determines the deviation percentage based on the design amount, budget amount, and the sum of positive deviations. The sum of positive deviations is determined based on the overrun amount of all overrun project instances in the target requirements, where an overrun project instance refers to a target project instance whose design amount exceeds the budget amount. A project feature vector is determined based on project complexity and the association confidence of the association edges between the target project instance and the target requirements. Project complexity is determined based on the project scale contained in the target instance data of the target project instance. A strategic weight adjustment value is determined based on the requirement priority contained in the target requirement data of the target requirements. The contribution of the target project instance to budget overruns is determined based on the deviation percentage, multi-dimensional weight matrix, project feature vector, historical risk correction value, and strategic weight adjustment value. The elements in the multi-dimensional weight matrix include: project complexity weight, association strength weight, and interaction influence weight; the historical risk correction value is determined based on the execution status data of historical project instances.

[0088] In this embodiment of the invention, the contribution decomposition algorithm overcomes the limitations of the traditional approach of "judging responsibility solely through absolute overspending," addressing the following issues: First, the synergistic impact of each project on the total overspending cannot be quantified under a "one-to-many" mapping; second, differences in the complexity and historical overspending risk of different projects lead to varying degrees of responsibility; and third, a single overspending amount cannot reflect the strategic weight of a project within the overall demand. Therefore, this embodiment introduces complex calculations such as nonlinear deviation transformation, multi-dimensional weight matrices, risk correction coefficients, and strategic weight adjustment terms to construct a fourth-order contribution model that accurately characterizes "absolute impact + relative impact + risk correlation + strategic adaptation."

[0089] Specifically, the formula for calculating the project's contribution to budget overruns is as follows:

[0090] ;

[0091] in, For the first The contribution of each project to the total budget overrun (values ​​range from [0,1], with a total of 1; the larger the value, the more significant the responsibility). The basic deviation percentage item (reflecting the relative impact of absolute overspending) (deviation percentage); It is the product of the multidimensional weight matrix and the project feature vector (reflecting the synergistic effect of project complexity and correlation strength). This is the historical risk correction term (an exponentially amplified function based on historical overspending probability) (historical risk correction value). This is the strategic weight adjustment item (Sigmoid function, reflecting the impact of demand priority on responsibility determination) (strategic weight adjustment value).

[0092] (1) Basic deviation percentage item The calculation formula is as follows: ,in, For the first The design cost for each project The budget amount for this project (under a "one-to-one" mapping) For the total budget of the demand order, under the "one-to-many" mapping (Total budget of demand order × project budget weighting factor). The sum of positive deviations for all items under the same demand order (only overspending items are added). (Surplus items are not included). For example, if Project A exceeds its budget by 500,000 yuan and Project B exceeds its budget by 1,500,000 yuan, for a total overspending of 2,000,000 yuan, then the basic percentage for Project A is 50 / 2,000,000 = 0.25, and for Project B it is 0.75, reflecting the initial impact of the absolute overspending. Data source: and Extracting (design amount and budget amount) from budget execution status data at each stage of the project. This breaks through the limitations of the traditional approach of "only looking at the absolute amount" and reflects the project's basic weight in the total cost overrun through relative proportions, avoiding misjudging large projects as the main responsible party due to their size (this needs to be corrected in conjunction with other factors).

[0093] (2) Multidimensional weight matrix and eigenvector product term :

[0094] Multidimensional weight matrix (2×2 dimension): ,in, (Project complexity weight). (Association strength weight) (Interactive influence weights) satisfy the condition that the sum of the matrix elements is 1, reflecting the relative importance of each dimension.

[0095] Project Feature Vector (2×1 dimension): ,in, The complexity of the project is quantified as follows: it is divided into three levels (low=1, medium=2, high=3) according to technical difficulty and number of participants, and is normalized to [0.3, 0.6, 0.9] (the responsibility determination for high-complexity projects is more stringent); Quantification of association strength: Based on the "association confidence" transformation of the associated edges in the demand-project topology association network (confidence 0.6-1.0 corresponds to 0.5-1.0, the closer the association, the greater the collaborative impact on the total overrun).

[0096] Product operation result: (Expands to include interactive items) This reflects the synergistic effect of complexity and correlation strength. For example, high complexity ( And strongly correlated () For projects with high complexity and weak correlation, the product term = 0.35×0.9 + 0.25×1.0 + 0.25×1.0 + 0.4×(0.9×1.0) = 0.315 + 0.25 + 0.25 + 0.36 = 1.175 (significantly amplifying the contribution); for projects with low complexity and weak correlation, the product term is smaller, reducing the responsibility ratio. Data source: Derived from the "project size" and "implementation scope" of standardized project instance data; This is derived from the "association confidence" attribute of the association edges in the initial relationship graph of requirements and projects. Thus, matrix operations are used to describe the interaction effect of "complexity-association strength," solving the problem of traditional methods "ignoring differences in project characteristics, leading to misjudgment of responsibility." For example, cost overruns in simple projects may be due to external factors, while cost overruns in highly complex projects are more likely to stem from internal management issues, requiring weighting to amplify responsibility.

[0097] (3) Historical risk correction item The calculation formula is as follows: For the first Historical cost overrun probability of similar projects (the percentage of projects with cost overruns in the same category over the past 3 years, with values ​​[0,1]); This is the risk sensitivity coefficient (an industry-specific parameter; the higher the value, the more significant the impact of historical risk on the current contribution). For example, if project C has a historical cost overrun probability... (If similar projects exceed the budget by 70%), then the revision item will be... (Significantly amplify the contribution, as history has shown it is prone to cost overruns); If project D's Then the correction term ≈ (Slightly magnified; historical performance is stable). Data source: This is obtained by statistically analyzing the proportion of similar projects where "actual expenditure > budgeted expenditure" from budget execution status data at each stage of the project. By introducing historical risk data and using an exponential function to implement the logic that "the higher the historical probability of overspending, the greater the weight of the current contribution," this approach overcomes the limitation of traditional methods that "do not consider the inertia of project risk," making liability determination more forward-looking.

[0098] (4) Strategic weight adjustment item The calculation formula is as follows: For the first The quantitative value of the requirement priority corresponding to each project (1=low, 2=medium, 3=high); This represents the average priority of historical projects. This is an adjustment factor (amplifying priority differences). For example, high-priority items ( The adjustment term for ) is 1 / [1+exp(-1.8×(3-2))]=1 / [1+exp(-1.8)]≈0.859 (reducing the contribution weight, because high-priority projects need to be prioritized and have a higher tolerance for overspending); low-priority projects ( The adjustment term is approximately 1 / [1+exp(1.8)]≈0.141 (increasing the contribution weight, resulting in low tolerance for overspending). Data source: Extracted from the first type of node attribute, "requirement priority," in the requirement-project topology network. Thus, using the Sigmoid function to achieve strategic adaptation where "higher priority means more lenient responsibility determination" overcomes the limitations of traditional methods that "apply the same responsibility standard to all projects," aligning with the practical need for "strategic prioritization" in project management.

[0099] In this embodiment, multi-dimensional matrix interaction can be used By integrating complexity and correlation strength, liability determination can be precisely matched with project characteristics; historical risk index amplification can be achieved through... Linking historical performance with current responsibilities reflects the cumulative impact of "risk inertia"; non-linear adjustment of strategic priorities can use the Sigmoid function to achieve differentiated standards of "tolerance for high priorities and strictness for low priorities", which fits the actual management situation; multi-scale fusion calculation integrates absolute deviation, relative proportion, feature interaction, historical data, and strategic orientation in multiple dimensions to achieve a three-dimensional judgment of "responsibility = impact + feature + risk + strategy".

[0100] To improve the accuracy of the budget consumption prediction model, in the project lifecycle budget control method provided in Embodiment 1 of this application, before determining the project budget consumption prediction result based on the budget aggregation verification result and execution status data, an initial budget consumption prediction model is constructed based on a long short-term memory neural network; historical instance data of multiple historical project instances are collected, and all historical instance data are processed to obtain historical budget time series data; the initial budget consumption prediction model is trained using the historical budget time series data until the budget prediction value output by the initial budget consumption prediction model and the actual amount contained in the historical instance data are less than a preset threshold, thus obtaining the budget consumption prediction model.

[0101] In this embodiment of the invention, historical project data (historical instance data) from the budget execution status data of each stage of the project is used as the processing object. This historical project data includes the project number of completed projects, each execution time node (such as the end of each month), the actual amount incurred at the corresponding time node, the final actual amount incurred by the project, and the project type (such as infrastructure construction or equipment procurement). Time series preprocessing is performed on the historical project data. First, the data is classified according to project type. Then, for each type of project, the data is organized into a sequence data sorted by time with "time node" as the horizontal axis and "actual amount incurred" as the vertical axis. At the same time, outliers in the sequence that exceed the reasonable range are removed (such as the actual amount incurred at a certain time node being more than 10 times the average monthly consumption of the project). Missing time node data is filled in by linear interpolation. After processing, standardized historical budget time series data is obtained, which is used for subsequent training of long short-term memory neural networks.

[0102] Then, using standardized historical budget time series data as the processing object, a long short-term memory neural network budget consumption prediction model is constructed: First, the standardized historical budget time series data is divided into a training set (for model parameter learning) and a test set (for model performance verification) according to a preset ratio. The input of the training set is the sequence of actual amounts incurred in the first N time nodes (N is the number of time nodes in the project execution cycle, which can be adjusted according to the project type), and the output of the training set is the final actual amount incurred in the project. The core parameters of the long short-term memory neural network are set, including the number of hidden layer neurons (set according to the data volume, such as 50-200), the number of iterations (such as 100-300 rounds), and the learning rate (such as 0.001-0.01). The error between the predicted output and the actual final amount incurred is minimized through the backpropagation algorithm. After the model training is completed, the prediction accuracy of the model is verified using the test set. When the prediction error is lower than a preset threshold (such as the average absolute error rate being lower than 5%), the trained long short-term memory neural network budget consumption prediction model is obtained.

[0103] Optionally, the budget consumption prediction model includes at least a long short-term memory neural network structure. To improve the accuracy of determining the project budget consumption prediction results, in the budget control method for the project lifecycle provided in Embodiment 1 of this application, the budget aggregation verification results and execution status data are processed to obtain budget time series data; the budget time series data is input into the budget consumption prediction model, and the long short-term memory neural network structure is used to capture the dependency features in the budget time series data to obtain the budget prediction value; for each target project instance, based on the project complexity, market volatility coefficient, and historical overspending probability of the target project instance, a dynamic fault tolerance threshold calculation strategy is adopted to determine the dynamic fault tolerance threshold of the target project instance; based on the budget prediction value and the dynamic fault tolerance threshold of each target project instance, the project budget consumption prediction result is determined.

[0104] In this embodiment of the invention, the long short-term memory neural network budget consumption prediction model, the project design amount in the budget aggregation verification result, and the real-time budget execution data of the current project (current project data from the budget execution status data of each stage of the project, including the current execution time node and the corresponding actual amount) are used as processing objects. First, the real-time budget execution data of the current project is organized into current budget time series data consistent with the training set format. Then, the current budget time series data and the project design amount (as auxiliary features used to correct prediction bias) are input into the long short-term memory neural network budget consumption prediction model. The model learns the time sequence pattern of historical projects and the execution trend of the current project, and outputs the final actual amount predicted value and prediction confidence interval of the current project. After processing, the project budget consumption prediction result is obtained. This result will be used by the multi-path intelligent correction recommendation engine to generate rectification plan.

[0105] Then, the market volatility coefficient, project complexity factor, and historical overrun probability are used as the processing objects. The market volatility coefficient comes from publicly available raw material price indices or service fee fluctuation data in the industry, reflecting the price difference between the current market and the project design stage. The project complexity factor is calculated by weighting the number of sub-tasks, technical difficulty level, and frequency of cross-departmental collaboration in the current project. The more sub-tasks and the higher the technical difficulty, the larger the factor value. The historical overrun probability is obtained from the historical project data in the budget execution status data of each stage of the project, that is, the proportion of the number of overrun projects in the same type of project to the total number of projects of the same type. Calculate the dynamic fault tolerance threshold: First, construct a fuzzy logic reasoning rule base, which contains multiple sets of "if-then" logic (e.g., "If the market volatility coefficient is high, the project complexity factor is high, and the historical overspending probability is high, then the dynamic fault tolerance threshold is high," and "If the market volatility coefficient is low, the project complexity factor is low, and the historical overspending probability is low, then the dynamic fault tolerance threshold is low"). Fuzzify the market volatility coefficient, project complexity factor, and historical overspending probability using a membership function (e.g., a triangular membership function) to convert them into fuzzy sets (e.g., "low," "medium," and "high"). Perform reasoning operations on the fuzzy sets based on the fuzzy logic reasoning rule base to obtain fuzzy output results. Defuzzify the fuzzy output results using the centroid method to convert them into specific numerical thresholds (expressed as a percentage of the project design amount). The resulting dynamic fault tolerance threshold will be used for budget compliance verification and reinforcement learning algorithm parameter optimization.

[0106] In some optional embodiments, the Long Short-Term Memory (LSTM) neural network budget consumption prediction model differs from traditional linear regression or moving average methods used in budget prediction. Traditional methods can only capture the linear trend of budget consumption and cannot handle the nonlinear time-series characteristics caused by policy adjustments, supply chain changes, etc., during project execution, easily leading to significant prediction bias. The LSM neural network used in this embodiment can effectively capture long-term dependencies in time series through gating mechanisms (input gate, forget gate, output gate). For example, it can identify the nonlinear consumption pattern of "low initial investment, high mid-term investment, and declining investment in the later stage" in infrastructure projects. The improvement is due to the fact that project budget consumption is affected by multiple dynamic factors and exhibits significant nonlinear characteristics. The structural design of the LSM neural network is more suitable for the modeling needs of this type of data, improving the reliability of prediction results. At the same time, the calculation method of the dynamic fault tolerance threshold differs from the traditional fixed threshold (such as uniformly setting it to 110% of the design amount). The traditional fixed threshold does not consider individual differences in projects and changes in the external environment, which may lead to insufficient fault tolerance for complex projects or overly loose control over simple projects. This embodiment uses a fuzzy logic reasoning engine to integrate three types of features: market fluctuations, project complexity, and historical cost overrun probability, to achieve dynamic adjustment of the threshold. The reason for this design is that different projects face significantly different risk factors, and it is necessary to use multi-dimensional features to collaboratively decide the threshold, ensuring that the threshold meets both control requirements and practical applicability.

[0107] In some alternative embodiments, the training process for the Long Short-Term Memory Neural Network is as follows: If the data being processed is historical infrastructure project data, the time node for the standardized historical budget time series data is set to "the end of each month". The input to the training set is the actual amount sequence of the previous 6 months, and the output is the total actual amount of the project in the last 12 months. The number of hidden layer neurons is set to 100, the number of iterations is 200 rounds, and the learning rate is 0.005. After training, the average absolute error rate of the test set is 4.2%, which meets the preset threshold requirement, thus obtaining the trained infrastructure project Long Short-Term Memory Neural Network budget consumption prediction model. When predicting a current infrastructure project, the project has been executed for 3 months, and the actual amounts incurred are A1, A2, and A3, respectively. The data of these 3 months are organized into the current budget time series data. Combined with the project's design amount B, the data is input into the model, and the output is the predicted final actual amount incurred, C, with a confidence interval of [C-5%×C, C+5%×C], forming the project budget consumption prediction result. Regarding the calculation of the dynamic fault tolerance threshold: If the market volatility coefficient of a certain equipment procurement project is 1.15 (higher than 1.0, defined as "high"), the project complexity factor is 3.2 (out of 5, defined as "medium"), and the historical overspending probability is 0.25 (lower than 0.3, defined as "low"), the rule "If the market volatility coefficient is high, the project complexity factor is medium, and the historical overspending probability is low, then the dynamic fault tolerance threshold is 112% of the design amount" is matched through the fuzzy logic reasoning rule base. After defuzzification, the final dynamic fault tolerance threshold for the project is 112% of the design amount. This specific value is only for auxiliary understanding and can be flexibly adjusted according to the rule base in actual application.

[0108] In this embodiment of the invention, the traditional "linear adjustment based on a single deviation rate" model is broken through, and a nonlinear adjustment model integrating project execution dynamics, risk evolution, and feature correlation is constructed, as shown in the following formula:

[0109] ;

[0110] in, For the first Budget adjustment amount for each project This is the original budget amount for the project (taken from the budget execution status data of each stage of the project, specifically the budget amount during the project initiation stage). The right side of the formula uses multi-dimensional coupling terms to achieve dynamic adjustment logic. The components and parameters are as follows:

[0111] (1) Execution deviation nonlinear transformation term Parameter definition: The actual deviation rate at the current stage of the project ( This is the actual amount incurred. (The design amount is taken from budget execution status data at each stage of the project). This represents the average deviation rate for similar projects. The standard deviation of the deviation rate (both obtained through statistical analysis of historical execution data); the Sigmoid function is defined as follows: .

[0112] Therefore: the traditional method directly uses Perform linear adjustment (e.g.) This is prone to over-adjustment due to extreme deviations. This section uses the Sigmoid function to standardize the deviation rate to the [0,1] interval, achieving the non-linear characteristic of "fine-tuning for small deviations and saturation adjustment for large deviations." For example, when... When the value is significantly higher than the mean (severe overspending), the function value approaches 1 (maximum adjustment range); when... When the value approaches the mean, the function value stabilizes at around 0.5 (with appropriate adjustments) to avoid rigid fluctuations caused by a single threshold.

[0113] (2) Feature-risk coupling matrix terms :

[0114] Feature vector : Dimensional vector, ,in, Normalized value of project size ( The project size is quantified and normalized to [0,1] using min-max. Normalized value for demand priority ( This quantifies the priority of requirements (and similarly normalizes them), reflecting the relative importance of the inherent characteristics of the project.

[0115] Coupling weight matrix : Dimensional matrix, ,in, (Scale itself weight) (Priority and its own weight) (Size-priority interaction weights) satisfy the condition that the sum of the matrix elements is 1. This matrix is ​​obtained through reverse training using historical adjustment effects, quantifying the synergistic effect that "the larger the scale / the higher the priority, the stronger the adjustment sensitivity".

[0116] Risk Vector : Dimensional vector, ,in, For the probability of cyclical risk ( Quantify the project implementation cycle. , The historical periodic mean and standard deviation, (The CDF follows a standard normal distribution, indicating a higher probability of risk for long-term projects). This represents historical overspending risk (the frequency of overspending in similar projects in recent years, taken from budget execution status data statistics), reflecting the driving role of risk in adjustments.

[0117] Therefore, traditional adjustments neglect the correlation between features and risks (e.g., "the risk adjustment needs of large, high-priority projects differ from those of small, low-priority projects"). This approach achieves multidimensional coupling of "features and risks" through matrix multiplication. For example, large, high-priority, and high-risk projects ( , , , The calculation result is This significantly amplifies the adjustment weights, meeting the actual needs of "high feature importance + high risk requiring key adjustments".

[0118] (3) Time decay term Parameter definition: The project has been running for months. The total project implementation period (taken from the project implementation period field). This is the attenuation coefficient (industry-adapted parameter).

[0119] Therefore: Early-stage adjustments offer high flexibility, while later adjustments significantly impact the overall plan, necessitating gradual tightening. This item utilizes an exponential function to achieve "the higher the time percentage (...)..." The larger the value, the weaker the adjustment range. For example, when the project has completed 80% of its cycle, (The adjustment range is reduced by nearly half), avoiding the disruption of the plan caused by frequent adjustments in the later stage, and breaking through the limitations of the traditional "same range adjustment throughout the whole cycle".

[0120] (4) Risk reservation items Parameter definition: Cumulative deviation volatility ( For the first Monthly actual expenditure, For the first Monthly adjusted budget (The standard deviation of historical monthly expenditures) Its normal distribution probability.

[0121] Therefore, the greater the cumulative deviation fluctuation, the higher the risk of future cost overruns, and adjustment margins need to be reserved. This item passes. Quantify the risk reserve ratio. For example, significant fluctuations in cumulative deviation ( )hour, (Reserving 6.7% adjustment redundancy) Integrating "risk prediction" into the adjustment model solves the lag problem of traditional "adjustment based only on existing deviations".

[0122] In this embodiment, the following features are employed: Nonlinear transformation: A smooth transition in deviation response is achieved through the Sigmoid function, avoiding the extreme value sensitivity problem of traditional linear adjustments; Matrix coupling: Feature-risk matrix operations replace single-factor weighting, accurately depicting the synergistic effects of "project scale, priority, cycle, and historical performance," covering complex relationships that traditional models cannot capture; Spatiotemporal adaptation: A time decay term is introduced to match the adjustment strategy with the project lifecycle stage, addressing the irrationality of "indiscriminate adjustment throughout the entire cycle"; Risk pre-positioning: Risk reservation is embedded in the adjustment process through the probability distribution calculation of cumulative deviation volatility, achieving an upgrade from "post-correction" to "pre-judgment + in-process adjustment." This multi-dimensional, nonlinear, and dynamic design breaks through the technical limitations of traditional budget adjustments that "rely on single deviations, ignore feature differences, and lack risk correlations," making the adjustment results more aligned with the actual needs of "precise control, risk adaptation, and cycle coordination" in project management.

[0123] To improve the accuracy of determining the target budget for target project instances, the budget control method for the project lifecycle provided in Embodiment 1 of this application determines the target deviation type and target deviation level based on the budget aggregation verification results; based on the target deviation type, a target template is determined from a preset template library, wherein the preset template library includes multiple suggested templates associated with the deviation type; based on the target deviation level and the project budget consumption prediction results, the model parameters in the target template are adjusted to obtain a budget deviation rectification plan; based on the budget deviation rectification plan, the budget aggregation verification results, and the project budget consumption prediction results, a control effectiveness evaluation report is generated; based on the budget deviation rectification plan and the control effectiveness evaluation report, the project data for each target project instance is adjusted to obtain the target budget for executing each target project instance.

[0124] In this embodiment of the invention, the deviation types and degrees in the main responsibility items and adjustment suggestions for budget overruns, as well as the project budget consumption prediction results, are used as the initial processing objects. The deviation types and degrees need to be extracted in a structured manner first. The deviation source types are identified from the main responsibility items and adjustment suggestions for budget overruns (divided into "design overrun", "execution deviation", and "predicted overrun". Design overrun refers to the project design amount exceeding the budget amount, execution deviation refers to the actual amount incurred during the project execution phase deviating from the design amount, and predicted overrun refers to the project budget consumption prediction results exceeding the dynamic fault tolerance threshold). At the same time, the deviation degree level is divided according to the ratio of the overrun amount to the budget amount (such as mild deviation, moderate deviation, and severe deviation, and the ratio division can be set according to the project type). After processing, structured deviation information is obtained, which clarifies the specific attributes and impact range of each type of deviation.

[0125] Then, using structured deviation information, project budget consumption prediction results, and dynamic fault tolerance thresholds as processing objects, a multi-path intelligent deviation correction recommendation engine is launched: First, a rectification plan template library is built, which contains standardized plan templates corresponding to deviation types ("design overrun type" corresponds to the design list optimization suggestion template, "execution deviation type" corresponds to the construction process optimization template, and "predicted overrun type" corresponds to the budget adjustment strategy template or project re-splitting plan template); then, the corresponding template is matched according to the deviation type in the structured deviation information, and the template parameters are adjusted in combination with the project budget consumption prediction results (such as the predicted final overrun amount and overrun time node). For example, when the project is "predicted to be overspending" and the deviation is moderate, the budget weight coefficient of the overspending project is extracted from the project budget consumption prediction results (from the project budget weight coefficient). Based on this coefficient, the adjustment amount is allocated among other projects under the same demand order and supplemented to the budget adjustment strategy template. Finally, the expected effect of each adjusted plan is evaluated through a utility function (utility value = (expected overspending reduction ratio × reciprocal of the difficulty of plan implementation), the higher the value, the better the plan). The top 3 plans with the highest utility values ​​are selected and processed to obtain the budget deviation rectification plan, which will be used to guide the project budget adjustment operation.

[0126] Using budget deviation rectification plans, budget aggregation verification results, and project budget consumption prediction results as processing objects, the budget control effectiveness evaluation module is initiated. First, an effectiveness evaluation index system is set, including "verification accuracy rate" (the matching ratio between budget aggregation verification results and actual overspending), "correction expectation achievement rate" (the achievable probability of reducing the expected overspending ratio in the budget deviation rectification plan, calculated based on historical implementation data of similar plans), and "prediction deviation rate" (the deviation ratio between the project budget consumption prediction results and the current actual amount). Then, corresponding data is collected according to the index system and weights are assigned (e.g., verification accuracy rate accounts for 40%, correction expectation achievement rate accounts for 35%, and prediction deviation rate accounts for 25%). The effectiveness score of each control link (verification link, prediction link, and correction link) is calculated by weighted summation. When the effectiveness score of a link is lower than the preset qualified threshold (e.g., 60 points out of 100), the link is marked as "link to be optimized". After processing, a control effectiveness evaluation report is obtained, which clarifies the control links that need to be optimized and the optimization direction.

[0127] Using the management effectiveness assessment report, differentiated verification strategy parameters, and dynamic fault tolerance threshold parameters as processing objects, a reinforcement learning algorithm is employed for optimization. First, the core elements of reinforcement learning are defined: "State" represents the current differentiated verification strategy parameters (e.g., the verification threshold for "one-to-one" mapping, the aggregate verification frequency for "one-to-many" mapping), the dynamic fault tolerance threshold parameters (e.g., the weight of the market volatility coefficient), and project characteristics (e.g., project type, scale); "Action" represents adjusting strategy parameters (e.g., increasing the aggregate verification frequency for "one-to-many" mapping) or modifying threshold parameters (e.g., adding a project complexity factor in the dynamic fault tolerance threshold calculation). The weights are calculated as follows: the "reward function" is the performance score in the control performance evaluation report (positive rewards are given when the performance score increases, and negative penalties are given when it decreases); then the parameter set is initialized, and the training is carried out through multiple rounds of iteration. In each round of iteration, the parameters are adjusted according to the current state, the performance score is recalculated based on the adjusted parameters, and the reward is fed back, and the parameter selection strategy is updated; when the performance score is stable above the qualified threshold and the fluctuation range is lower than the preset value (e.g., 5%) for N consecutive rounds (e.g., 10 rounds), the training is stopped, and the optimized threshold parameters and verification strategy are obtained after processing. These parameters and strategies will be used for subsequent closed-loop feedback.

[0128] Finally, taking the budget deviation rectification plan and the optimized threshold parameters and verification strategies as the processing objects, the two are linked and integrated. The budget deviation rectification plan is supplemented with suggestions for the application of optimized parameters (such as "After the implementation of this plan, the optimized dynamic fault tolerance threshold parameters should be used to perform acceptance phase verification"), forming a complete output result, ensuring that the result can be directly called to achieve closed-loop management in the future. Thus, through the above-mentioned coherent processing, the budget deviation rectification plan and the optimized threshold parameters and verification strategies are finally generated. The budget deviation rectification plan is used for actual budget adjustments in the project, and the optimized threshold parameters and verification strategies are used to feed back to the preceding stages to achieve continuous optimization.

[0129] In some optional embodiments, the multi-path intelligent deviation correction recommendation engine differs from the traditional "manual experience-based rectification" approach to budget control. Traditional methods rely on managers' subjective judgment to formulate solutions, resulting in limited options, low adaptability, and uncontrollable effects. This embodiment constructs a structured deviation classification mechanism and a dynamic solution template library. It achieves accurate solution generation through the collaborative matching of deviation types and prediction results, while introducing a utility function to quantitatively evaluate solution effectiveness, ensuring that recommended solutions are both targeted and feasible. This improvement stems from the complex causes of project budget deviations (design, execution, prediction, etc.), necessitating multi-path solutions to cover different scenarios and avoid the limitations of a single solution. Furthermore, the application of reinforcement learning algorithms differs from the traditional "static parameter setting" method. Traditional methods, once parameters are determined, remain unchanged for a long time, failing to adapt to dynamic changes in the market environment and project types. This embodiment transforms the management effectiveness evaluation results into reinforcement learning reward signals, achieving continuous iterative optimization of parameters and strategies. This design is based on the fact that budget control is a long-term dynamic process, requiring adaptive adjustments to ensure high efficiency at each stage and prevent management failure due to parameter fixation.

[0130] In some alternative embodiments, regarding the extraction of structured deviation information: if the main responsible project for the budget overrun is an infrastructure project, its deviation type is identified as "design overrun" (design amount is 8% higher than budget amount), and the deviation degree is "moderate deviation"; the project budget consumption prediction results show that the final actual amount incurred by the project may be 3% higher than the design amount, and the dynamic error tolerance threshold is 112% of the design amount (i.e., a final overrun of 12% is allowed). The multi-path intelligent deviation correction recommendation engine matches the "design list optimization suggestion template", adjusts the template parameters based on the prediction results, filters out non-core construction items (such as decorative supporting works) from the design list, calculates the design amount that can be reduced by deleting such items (e.g., 4%), and simultaneously adjusts the budget adjustment strategy from the equipment procurement project (with budget surplus) under the same demand order by 2%, forming a combined rectification plan of "design list deletion + budget adjustment". After utility function evaluation, the expected overrun reduction ratio of this plan is 6%, the implementation difficulty is low, the utility value ranks first, and it is determined as the final budget deviation rectification plan. In the reinforcement learning optimization phase, if the control effectiveness assessment report shows that the effectiveness score of the "one-to-many" mapping verification step in step 2 is 55 points (below the passing threshold of 60 points), the reinforcement learning algorithm will adjust the summary verification frequency of the "one-to-many" mapping from "once a day" to "once every 6 hours" (action), recalculate the effectiveness score to improve to 68 points, and obtain a positive reward. The adjusted verification frequency is incorporated into the optimized verification strategy. The values ​​in this specific scenario are only used to help understand the implementation process of the solution. In actual applications, they can be flexibly set according to project needs.

[0131] In this embodiment of the invention, a risk propagation quantification model based on a demand-project topology network is constructed:

[0132] ;

[0133] in, This indicates that the risk originates from the demand node. propagate to project nodes The probability (range [0,1]) is calculated using a formula that employs multi-path coupling, node influence quantification, historical fluctuation correction, and path decay to accurately characterize risk propagation. The parameters are as follows:

[0134] (1) Path association strength product term Parameter definition: For the requirements - project topology associated network of nodes To the node The set of all paths This is one of the specific paths; For path The top edge, The association confidence of this edge (taken from the association edge attribute of the initial association graph of the requirements-project, with a value range of [0,1], the higher the value, the stronger the node). and The closer the connection.

[0135] Therefore, traditional risk assessment only considers direct connections (single path), ignoring risk transmission through indirect paths such as "demand-project-sub-project". This approach quantifies the connection strength across multiple paths by multiplying the connection confidence scores of all edges along the path. For example, the connection strength of the path "demand A-project B-sub-project C" is... ,like and The product is 0.72, which reflects the risk transmission capability of indirect association and breaks through the assessment limitations of "single direct association".

[0136] (2) Node influence on nonlinear transformation terms Parameter definition: For nodes Influence quantification value (demand node) of Based on the quantification of demand priority, project nodes of The quantification values ​​are based on project size and are all derived from node attributes. The quantification rule is that higher priority / larger scale corresponds to higher values. The average influence of all nodes in the network. The influence standard deviation (obtained through statistical analysis of node attributes in the topology network); the Sigmoid function is defined as... .

[0137] Therefore, the effect of node influence on risk propagation is not linear (the risk diffusion effect of high-influence nodes is exponentially enhanced). This project standardizes node influence using the Sigmoid function, achieving "core nodes (high-influence nodes)..." Risk amplification, edge nodes (low) The nonlinear characteristics of "risk suppression". For example, high-priority demand nodes ( Much higher The function value of the node approaches 1 (maximum influence weight), and the function value of the node of the low-scale project approaches 0.1 (minimum influence weight), which solves the irrationality of the traditional "equalization of node influence".

[0138] (3) Historical fluctuation probability term Parameter definition: For nodes In the Budget deviation at each stage (the difference between the actual amount incurred and the budgeted amount, taken from the budget execution status data of each stage of the project). For nodes Number of phases executed; The standard deviation of budget deviation for similar nodes in history (obtained through historical data statistics); is the cumulative distribution function (CDF) of the standard normal distribution.

[0139] Therefore, historical budget fluctuations at nodes are a significant driver of risk propagation, and traditional methods only assess single-stage deviations, neglecting cumulative effects. This paper addresses this by standardizing cumulative deviations (dividing by...). This transforms the probability into a normal distribution. For example, if a project node has a significant cumulative cost overrun (standardized value of 1.64), then... (High volatility probability) significantly increases the risk propagation weight and achieves a dynamic correlation between "historical volatility and current risk".

[0140] (4) Path length attenuation term Parameter definition: For path The length of the path (the number of edges on the path); This is the attenuation coefficient (an industry-adapted parameter, calibrated using historical risk propagation data).

[0141] Therefore, risk propagation weakens as the path length increases (indirectly linked risk transmission is weaker than directly linked risk). This term achieves the characteristic of "the longer the path, the more significant the attenuation" through an exponential function. For example, a path of length 3 ( The decay value is The direct path decay value of length 1 is It accurately depicts the inhibitory effect of path distance on risk transmission, avoiding the overestimation of risk caused by the traditional approach of "ignoring path length".

[0142] In this embodiment, multi-path coupling is employed: by multiplying and summing the path association strengths, the risk transmission of both direct and indirect associations is covered, overcoming the one-sidedness of traditional "single-path assessment"; nonlinear node influence is addressed by introducing a sigmoid function to quantify the differentiated effects of node influence, accurately characterizing the risk amplification effect of core nodes and overcoming the limitations of the "node equalization assumption"; historical fluctuation integration is achieved by converting cumulative budget deviations into probability values ​​based on the normal distribution CDF, realizing a quantitative correlation between "historical data and current risk" and avoiding the lag of traditional "static threshold judgment"; and a path decay mechanism is implemented by using an exponential function to reflect the distance decay law of risk propagation, ensuring that the assessment results are deeply adapted to the network topology and solving the problem of "overestimation of long-path risk". This design, which integrates graph theory, probability statistics, and nonlinear transformation, breaks through the technical limitations of traditional risk assessment being "isolated, linear, and static," enabling more accurate capture of the complex transmission laws of risk in the demand-project network and providing a quantitative basis for dynamic risk management.

[0143] In some optional embodiments, the optimized threshold parameters and verification strategies, and the demand-project topology association network are used as the initial processing objects. The optimized threshold parameters and verification strategies comprise two core components: optimized verification strategy parameters (such as the tolerance for deviation between design and budget amounts in "one-to-one" mapping, and the aggregated verification frequency in "one-to-many" mapping) and optimized threshold parameters (such as the market volatility coefficient and the weighting of the project complexity factor in the dynamic fault tolerance threshold calculation). First, parameter supplementation processing is performed on the demand-project topology association network: the optimized verification strategy parameters and threshold parameters are added as new attributes and associated with the first type of node (project establishment demand single node) in the demand-project topology association network. For example, under the "communication base station construction" type demand single node, attributes such as "optimized one-to-many verification frequency: once every 6 hours" and "optimized project complexity factor weight: 0.4" are added. After processing, an updated demand-project topology association network is obtained. This network can serve as the basic data for generating new networks in the next round of project management, achieving continuous parameter inheritance.

[0144] Using the updated requirement-project topology network, optimized threshold parameters and verification strategies, and differentiated verification strategies as processing objects, parameter updates are performed in the control phase: adjustment items related to the differentiated verification strategy are extracted from the optimized verification strategy parameters (e.g., the deviation tolerance ratio between the total design amount of "one-to-many" mapping and the total budget of the requirement order is adjusted from 5% to 3%), and the corresponding parameters of the original differentiated verification strategy are replaced; at the same time, the node attributes in the updated requirement-project topology network (e.g., the optimized verification frequency corresponding to the requirement order) are used as the triggering condition for the association subgraph traversal algorithm (e.g., when the node attribute of the requirement order is marked as "high-frequency verification", the execution interval of the association subgraph traversal algorithm is shortened). After processing, the updated differentiated verification strategy is obtained. This strategy will be used for budget aggregation verification in the next round of project initiation stage to improve the timeliness and accuracy of verification.

[0145] Using the optimized threshold parameters and verification strategies, along with the dynamic fault-tolerant threshold calculation model (built based on a fuzzy logic inference engine), as the processing objects, parameter updates are performed in the control phase: Weight adjustment items related to dynamic fault-tolerant threshold calculation are extracted from the optimized threshold parameters (e.g., the weight percentage of the market volatility coefficient is adjusted from 0.3 to 0.35, and the weight percentage of the project complexity factor is adjusted from 0.3 to 0.35), replacing the original weight parameters in the dynamic fault-tolerant threshold calculation model; simultaneously, the membership function adjustment values ​​in the optimized threshold parameters (e.g., the threshold for classifying "high" market volatility coefficient is adjusted from 1.2 to 1.15) are updated to the rule base of the fuzzy logic inference engine. After processing, an updated dynamic fault-tolerant threshold calculation model is obtained. This model will be used for second-order budget compliance verification in the next round of project acceptance, making the threshold calculation more aligned with the current project environment characteristics.

[0146] The entire process integrates all processing results (requirement-project topology network, project budget weight coefficients, budget execution status data for each project stage, budget aggregation verification results, major responsible projects for budget overruns and adjustment suggestions, project budget consumption prediction results, dynamic fault tolerance thresholds, budget deviation rectification plans, optimized threshold parameters and verification strategies) and the updated requirement-project topology network, updated differentiated verification strategies, and updated dynamic fault tolerance threshold calculation models. First, a five-layer structured system of "data-verification-prediction-correction-feedback" is constructed: the data layer includes the requirement-project topology network, project budget weight coefficients, and budget execution status data for each project stage; the verification layer... The system includes: budget aggregation verification results and updated differentiated verification strategies; prediction layer includes project budget consumption prediction results and updated dynamic fault tolerance threshold calculation model; correction layer includes major over-budget responsible projects and adjustment suggestions, and budget deviation rectification plans; feedback layer includes optimized threshold parameters and verification strategies, and updated requirement-project topology association network; and then cross-layer association is performed through association fields (such as project number and requirement number) to ensure that the verification result of a project can be directly associated with its prediction data and rectification plan. After processing, a closed-loop system for intelligent management and control of the project's entire lifecycle budget is generated. This system realizes the full-process connection from data collection to feedback optimization and supports the efficient start of the next round of project management.

[0147] In this embodiment, a dual closed-loop mechanism of "dynamic parameter feedback - full-process result integration" is constructed, which differs from the limitations of traditional budget control systems that are "without feedback and fragmented." In traditional systems, once the parameters of each link are set, they remain unchanged for a long time, and the processing results of each stage are stored independently, making it impossible to form collaborative control, resulting in a gradual decline in control efficiency as the project environment changes. This embodiment uses three-path parameter feedback (feedback to the demand-project topology association network, the verification link, and the prediction link) to enable the optimized parameters to directly affect the initial data and core algorithms of the next round of control, solving the problem of "parameter solidification" in traditional systems. At the same time, the full-process results are integrated through a five-layer structured system, so that data, verification, prediction, correction, and feedback data form an organic whole through related fields, avoiding control breakpoints caused by "result fragmentation" in traditional systems. In addition, the updated demand-project topology association network serves as a parameter inheritance carrier, so that the optimized parameters of each round can be accumulated into the basis of the next round of control, forming a continuous improvement cycle of "optimization-accumulation-re-optimization." The reason for this design is that project budget control is a long-term dynamic process, which requires parameter inheritance and result collaboration to ensure that control capabilities continuously improve with the accumulation of project experience.

[0148] For example, for a batch of infrastructure projects (such as demand number XQ2024001): In the optimized threshold parameters and verification strategy, the frequency of summary verification for the "one-to-many" mapping is adjusted from "once a day" to "once every 6 hours," and the weight of the project complexity factor in the dynamic fault tolerance threshold calculation is adjusted from 0.3 to 0.35. First, this optimized parameter is added as an attribute to the single node of demand XQ2024001 in the demand-project topology association network to obtain the updated demand-project topology association network; then, the verification frequency of "once every 6 hours" is updated to the differentiated verification strategy, so that the summary verification of the single-linked project of demand XQ2024001 in the next round is executed according to the new frequency; then, the weight of the project complexity factor of 0.35 is updated to the dynamic fault tolerance threshold calculation model, so that when the model calculates the threshold for infrastructure projects next time, it will pay more attention to the impact of project complexity on the threshold. During the end-to-end integration phase, the data layer contains the associated network and budget data of XQ2024001, the verification layer contains the budget aggregation verification results and new verification frequency of this demand order, the prediction layer contains the project budget consumption prediction results and the threshold model of the new weight, the correction layer contains the adjustment suggestions and rectification plan for the over-budget projects, and the feedback layer contains the optimized parameters and the updated network. Each layer is associated with the demand number "XQ2024001". Through this closed-loop system, the end-to-end management data of all projects under this demand order can be quickly queried, and the management of the next batch of infrastructure projects (XQ2024002) can be directly started based on the optimized parameters. The specific demand number and frequency in this scenario are only for auxiliary understanding. In actual applications, they can be flexibly adapted according to the project batch.

[0149] In this embodiment, to overcome the limitations of traditional "static parameter fixation", a parameter iterative optimization model based on full-process performance feedback is constructed to achieve closed-loop control of "evaluation-correction-feedback-re-evaluation", as shown in the following formula:

[0150] ;

[0151] in, For the first The optimized parameter vector after rounds of iteration (including verification strategy parameters and threshold calculation parameters). For the first Given the current parameter vector, the formula achieves dynamic optimization of the parameters through collaborative calculation of "efficiency-driven correction - sensitivity weighting - iterative stability constraints". The parameters are as follows:

[0152] (1) Nonlinear transformation term of efficiency deviation Parameter definition: This represents the actual effectiveness of the current control process (taken from the budget control effectiveness assessment report, such as the weighted sum of verification accuracy and correction compliance rate). The preset target performance value (set according to project control requirements, such as verification accuracy ≥ 90%). The standard deviation of historical performance values ​​(obtained through statistical analysis of historical evaluation data); the Sigmoid function is defined as... .

[0153] Therefore: Traditional parameter adjustment directly uses linear deviation (such as...) This is prone to parameter oscillations due to performance fluctuations. This section uses the Sigmoid function to standardize the performance deviation to the [0,1] interval, achieving the nonlinear characteristic of "fine-tuning for small deviations and saturation correction for large deviations." For example, when... far below When (performance is not up to standard), the function value approaches 1 (maximum correction range); when near When the function value is stable at around 0.5 (with moderate fine-tuning), parameter instability caused by extreme performance fluctuations is avoided, thus overcoming the limitation of traditional "linear correction is prone to oscillation".

[0154] (2) Parameter sensitivity-stage weight coupling term :

[0155] Sensitivity Matrix : Dimensional matrix ( For the number of parameters, (for the number of performance indicators) Indicates the first The parameter for the first Sensitivity of the first performance indicator (calculated using the controlled variable method: keeping other parameters fixed, adjusting the first...) The parameter and observe the first one. The rate of change of each performance indicator (data from historical parameter adjustment records). For example, the sensitivity of the "one-to-many verification frequency" parameter to "verification accuracy". (A 10% adjustment to parameters results in an 8% change in performance), demonstrating the strength of the parameter's impact on performance.

[0156] Parameter adjustment direction vector : Dimensional vector, ,in, This is a sign function (takes 1 when performance is not up to standard, and the parameter needs to be adjusted in the positive direction). For the first The parameter and the first The correlation coefficients of the performance indicators (1 for positive correlation and -1 for negative correlation to ensure that the adjustment direction is consistent with the performance improvement) are data from parameter-performance correlation analysis.

[0157] Stage weight vector : Dimensional vector, Indicates the first The weights of each control phase (verification phase, prediction phase, correction phase) are determined based on the importance of the phase, such as the weight of the verification phase. Prediction phase Correction phase (The sum is 1), which reflects the contribution of different stages to the overall control.

[0158] Therefore, traditional parameter adjustment ignores the differences in parameter sensitivity and stage importance, employing a uniform adjustment range. This approach uses matrix multiplication to couple sensitivity with stage weights. For example, high-sensitivity parameters (such as the dynamic fault tolerance threshold weight in step 3) receive greater adjustment weights in high-weight stages (such as the prediction stage), while the adjustment range of low-sensitivity parameters (such as log recording frequency) is suppressed. This ensures that parameter adjustments focus on dimensions that critically impact performance, overcoming the inefficiency of traditional indiscriminate adjustment.

[0159] (3) Iterative stability constraint terms Parameter definition: The attenuation coefficient (calibrated using historical iteration data to control the weakening of constraint strength with the number of iterations). This is the current iteration number (counted since system startup); The baseline parameter vector (the initial optimized parameters, taken from the optimized threshold parameters and verification strategy); The L2 norm (calculates the Euclidean distance between the current parameter and the reference parameter, measuring the degree of parameter deviation); It is a unit vector (ensuring that the constraint direction is opposite to the parameter deviation direction).

[0160] Therefore, traditional iteration is prone to over-correction, causing parameters to deviate from the baseline range (e.g., the dynamic fault tolerance threshold far exceeds the reasonable range). This project achieves iterative stability through "exponential decay + distance constraint". For example, in the early stages of iteration ( Small), exponential term approaches 1, constraint strength is large (preventing parameters from rapidly deviating from the baseline); later stage of iteration ( The exponential term approaches 0, and the constraint strength weakens (allowing the parameter to adapt to long-term performance changes); at the same time, the further the parameter deviates from the benchmark, the stronger the constraint becomes, ensuring that the parameter is always optimized within a reasonable range, thus solving the problem of traditional "iteration divergence".

[0161] In this embodiment, the nonlinear performance response utilizes the Sigmoid function to smoothly transform performance deviations, avoiding the extreme value sensitivity problem of traditional linear corrections and making parameter adjustments more closely aligned with the actual patterns of performance fluctuations. Multi-dimensional weight coupling uses the multiplication of the sensitivity matrix and the stage weight vector to quantify the synergistic effect of "parameter importance - stage contribution," ensuring that adjustment resources are focused on key parameters and core stages, overcoming the inefficiency of traditional "indiscriminate adjustment." An iterative stabilization mechanism introduces exponentially decaying L2 norm constraints to balance the "adaptability" and "stability" of parameters, allowing parameters to dynamically optimize with performance changes while preventing deviations from the baseline that could lead to control disorder, thus overcoming the limitations of traditional iterations that are prone to oscillation and divergence. This design, integrating nonlinear transformation, matrix operations, and iterative constraints, upgrades "closed-loop feedback" from "simple parameter replacement" to "data-driven dynamic optimization," ensuring that control parameters are always deeply adapted to the overall process performance.

[0162] Figure 2 This is a schematic diagram of an optional intelligent management process for project lifecycle budget based on multidimensional association mapping and dynamic threshold learning according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following:

[0163] Step 1: Based on the project requirement form and project instance data, the following steps are performed: using graph database technology to construct a multi-dimensional mapping relationship between the project requirement form and project instances; using a dynamic weight allocation algorithm to calculate the weight coefficient of each project in the total budget; and using a budget flow status awareness engine to track the budget execution status (including budget amount, design amount, and actual amount incurred) of each stage of the project. This process generates a requirement-project topology network, project budget weight coefficients, and budget execution status data for each stage of the project.

[0164] Step 2 involves processing the following steps based on the requirement-project topology network and budget execution status data for each stage of the project generated in Step 1: using a subgraph traversal algorithm to identify all related projects under the same requirement order; using a distributed real-time computing framework to aggregate the design amount of each project in parallel; executing a differentiated verification strategy according to the mapping relationship type ("one-to-one" mapping compares the design amount of a single project with the budget amount, and "one-to-many" mapping verifies that the sum of the design amounts of each project does not exceed the total budget of the requirement order); and using a contribution decomposition algorithm to locate the main responsible projects for exceeding the budget and generate adjustment suggestions. This process generates budget aggregation verification results, main responsible projects for exceeding the budget, and adjustment suggestions.

[0165] Step 3: Based on the historical project data in the budget execution status data of each stage of the project generated in Step 1 and the project design amount in the budget aggregation and verification results generated in Step 2, the project is processed by using a long short-term memory neural network to learn the historical budget execution rules to predict the final actual amount of the project, and a fuzzy logic reasoning engine to calculate the dynamic fault tolerance threshold by integrating the market volatility coefficient, project complexity factor and historical overspending probability, so as to generate the project budget consumption prediction results and dynamic fault tolerance threshold.

[0166] Step 4: Based on the deviation types and degrees in the main responsibility items and adjustment suggestions for budget overruns generated in Step 2, and the project budget consumption prediction results and dynamic fault tolerance thresholds generated in Step 3, a multi-path intelligent deviation correction recommendation engine is used to generate rectification plans (including design list optimization suggestions, project re-splitting plans, and budget adjustment strategies) according to the deviation types and degrees. The budget control effectiveness evaluation module monitors the effectiveness of each control link, and the reinforcement learning algorithm optimizes the threshold parameters and verification strategies to generate budget deviation rectification plans, optimized threshold parameters, and verification strategies.

[0167] Step 5: Based on the requirement-project topology network generated in Step 1, the budget deviation rectification plan generated in Step 4, and the optimized threshold parameters and verification strategies, the optimized threshold parameters and verification strategies are fed back to the requirement-project topology network and the control links of Steps 2 and 3. The process of integrating all the processing results of Steps 1 to 4 is carried out to build a closed-loop system, so as to generate a closed-loop system for intelligent management and control of the project's full life cycle budget.

[0168] In the above steps, through technological innovations such as demand-project topology association modeling, multi-level budget aggregation verification, budget consumption time-series prediction, dynamic threshold control, and intelligent correction optimization, the limitations of traditional budget control, namely "static threshold, single-point verification, and post-event interception", are broken through, and intelligent management and control of the project's entire life cycle budget is realized from passive interception to proactive prevention, and from single verification to comprehensive optimization.

[0169] In this embodiment of the invention, to address the challenges of depicting the complex multi-dimensional relationships between project requirement orders and project instances, and the mismatch between budget allocation weights and the actual scale / priority of projects, a graph database technology is used to construct a "requirement-project" topological relationship network. This network intuitively presents complex "one-to-one" and "one-to-many" mapping relationships. Simultaneously, a dynamic weight allocation algorithm is used to calculate budget weight coefficients based on multiple characteristics such as project scale, requirement priority, and implementation cycle. This ensures that budget allocation closely aligns with actual project management needs, effectively resolving the mismatch between budget resources and actual project requirements, and improving the precision and scientific nature of budget allocation. To address the issues of low budget aggregation efficiency due to serial computation, blind spots in management caused by a unified verification strategy that doesn't differentiate mapping types, and difficulty in defining responsible parties after budget overruns, a subgraph traversal algorithm is used to quickly identify all related projects under the same requirement order. A distributed real-time computing framework enables parallel aggregation of design amounts for multiple projects, significantly improving aggregation efficiency. Furthermore, differentiated verification strategies are executed according to mapping relationship types to avoid blind spots in total budget overrun management in "one-to-many" scenarios. Finally, a contribution decomposition algorithm accurately identifies the main responsible projects for budget overruns, clarifying rectification directions and effectively solving problems such as budget aggregation delays, poor verification adaptability, and ambiguous responsibility definitions. To address the issues of fixed tolerance thresholds failing to adapt to varying market environments and complexities across projects, and the delayed prediction of budget overrun risks, a system based on Long Short-Term Memory (LSTM) neural networks learns historical project budget execution patterns. This allows for early prediction of the final actual expenditure, providing proactive warnings of overrun risks. Simultaneously, a fuzzy logic reasoning engine calculates dynamic tolerance thresholds by integrating market volatility coefficients, project complexity factors, and historical overrun probabilities. This enables flexible adjustments to the thresholds based on individual project characteristics, resolving the poor adaptability of fixed thresholds and improving the timeliness and accuracy of risk control. To address the problems of rigid control strategies, a lack of iterative optimization mechanisms based on historical performance, and insufficient precision in budget deviation rectification plans, a budget control effectiveness evaluation module monitors the effectiveness of control at each stage in real time. Combined with reinforcement learning algorithms, threshold parameters and verification strategies are continuously optimized, forming a closed-loop control system of "evaluation-optimization-feedback." Furthermore, a multi-path intelligent correction recommendation engine generates targeted rectification plans such as design list optimization, project re-splitting, and budget adjustments based on deviation type, severity, and root cause analysis results. This effectively addresses the issues of rigid control strategies and a lack of precise guidance for deviation rectification, improving the dynamic adaptability and rectification effectiveness of budget control.

[0170] The following is a detailed description with reference to another embodiment.

[0171] Example 2

[0172] The project lifecycle budget control device provided in this embodiment includes multiple implementation units, each of which corresponds to a specific implementation step in Embodiment 1 above.

[0173] Figure 3This is a schematic diagram of an optional project lifecycle budget control device according to an embodiment of the present invention, such as... Figure 3 As shown, the budget control device may include: a first determining unit 30, a verification unit 31, a second determining unit 32, and an adjustment unit 33.

[0174] The first determining unit 30 is used to determine the set of target project instances corresponding to the target requirements based on the preset topology association network, and obtain the target topology association sub-network. The preset topology association network records the relationship between each requirement and at least one project instance. Each project instance corresponds to project data, which includes at least: budget weight coefficient and execution status data. The execution status data includes at least: the budget amount, design amount and actual amount of the project instance at each stage.

[0175] The verification unit 31 is used to perform differential verification on the target project instance set based on the execution status data to obtain the initial budget verification result, and generate the budget aggregation verification result based on the target project instance set when the initial budget verification result indicates that the budget is over-limited.

[0176] The second determining unit 32 is used to determine the project budget consumption prediction result based on the budget aggregation verification result and the execution status data, using a budget consumption prediction model.

[0177] Adjustment unit 33 is used to adjust the project data of each target project instance in the target topology associated sub-network based on the project budget consumption prediction results and budget aggregation verification results, so as to obtain the target budget for executing each target project instance.

[0178] The aforementioned budget control device can construct a pre-defined topology network. By identifying the set of target project instances corresponding to the target requirements and forming a target topology network, it achieves the goal of accurately mapping the complex relationships between requirements and project instances. This enables precise allocation of budget resources throughout the project lifecycle, thus solving the technical problem of mismatch between budget allocation and actual needs such as project scale and priority in related technologies. Specifically, the pre-defined topology network records the relationship between each requirement and at least one project instance. Each project instance corresponds to project data, including budget weight coefficients and execution status data. The execution status data covers the budget amount, design amount, and actual amount of the project instance at each stage. Based on the execution status data, differential verification is performed on the set of target project instances to obtain initial budget verification results. If budget overruns are found, a budget aggregation verification result is further generated to locate the responsible party for the overrun. Then, combining the budget aggregation verification result and the execution status data, a budget consumption prediction model is used to determine the project budget consumption prediction result to anticipate potential risks. Subsequently, based on the project budget consumption prediction results and budget aggregation verification results, the project data of each target project instance in the target topology associated sub-network is intelligently adjusted to obtain the optimized budget for executing each target project instance, thus realizing the foresight and adaptability of budget control.

[0179] Optionally, the budget control device further includes: a first acquisition module, used to acquire demand data for each demand and instance data for each project instance before determining the set of target project instances corresponding to the target demand based on a preset topology association network and obtaining the target topology association subnetwork, wherein the demand data includes at least: demand number, demand type, total budget amount, and demand priority; the instance data includes at least: project number, project size, project implementation period, and project-related demand number; a first processing module, used to preprocess the demand data and instance data to obtain target demand data and target instance data; a first representation module, used to represent each demand as a first type of node and each project instance as a second type of node; and a first construction module, used to construct the target demand data based on the demand number in the target demand data corresponding to the first type of node and the target instance data corresponding to the second type of node. The example data is used to associate project requirement numbers to construct association edges between first-type nodes and second-type nodes. Each association edge has an edge attribute, which includes at least: mapping relationship type and association confidence. The first determination module is used to determine the budget weight coefficient of each second-type node based on the target requirement data and the target instance data of the second-type node. The second determination module is used to determine the execution status data of each project instance at each stage based on the budget weight coefficient and the total budget amount. The stage is divided into stages based on the life cycle of the project instance, namely: project initiation stage, execution stage, and acceptance stage. The first composition module is used to construct a preset topology association network based on all first-type nodes, all second-type nodes, all association edges, the budget weight coefficient of each second-type node, and the execution status data.

[0180] Optionally, the first determining module includes: a first determining submodule, used to determine a non-linear transformation value for project size based on the project size contained in the target instance data, a non-linear transformation value for demand priority based on the demand priority contained in the target demand data, and a non-linear transformation value for project period based on the project implementation period contained in the target instance data; a second determining submodule, used to determine a feature row matrix based on the non-linear transformation values ​​for project size, demand priority, and project period; and a third determining submodule, used to determine the probability of risk associated with project size based on project size, the probability of risk associated with demand priority based on demand priority, and the probability of risk associated with project implementation. The project cycle is determined in three sub-modules: the fourth sub-module determines the probability of risks associated with the project cycle; the fifth sub-module determines the non-linear amplification value based on the priority of the requirements; and the sixth sub-module determines the budget weight coefficient of the second type of node based on the feature row matrix, the multi-feature collaborative weight matrix, the risk probability column matrix, the historical feedback dynamic adjustment value, and the non-linear amplification value. The element values ​​in the multi-feature collaborative weight matrix are determined based on historical project instance data, and the historical feedback dynamic adjustment value is determined based on the execution status data of historical project instances.

[0181] Optionally, the verification unit includes: a first extraction module, used for extracting the design amount and budget amount from the execution status data for each target project instance in the target project instance set, when the mapping relationship between the associated edge and the target requirement is a one-to-one mapping relationship, and comparing the design amount with the budget amount; a third determination module, used for determining the initial budget verification result as compliant when the design amount is less than or equal to the budget amount, or determining the initial budget verification result as budget overrun when the design amount is greater than the budget amount; a fourth determination module, used for determining the total design amount and the total budget amount of the target requirement based on the execution status data of all target project instances associated with the target requirement when the mapping relationship between the associated edge and the target requirement is a one-to-many mapping relationship; and a fifth determination module, used for determining the initial budget verification result as compliant when the total design amount is less than or equal to the total budget amount, or determining the initial budget verification result as budget overrun when the total design amount is greater than the total budget amount.

[0182] Optionally, the verification unit further includes: a sixth determining module, used to determine the contribution of each target project instance to the budget overrun based on the execution status data of the target project instance for each target project instance in the target project instance set; a seventh determining module, used to determine the target project instance with the largest contribution as the preset project instance and generate an adjustment strategy for the preset project instance; and a first integrating module, used to integrate the preset project instance and the adjustment strategy with the initial budget verification result to obtain the budget aggregate verification result.

[0183] Optionally, the sixth determining module includes: a seventh determining submodule, used to determine the deviation percentage based on the design amount, budget amount, and the sum of positive deviations, wherein the sum of positive deviations is determined based on the overspending amount of all overspending project instances in the target requirements, and overspending project instances refer to target project instances whose design amount exceeds the budget amount; an eighth determining submodule, used to determine the project feature vector based on project complexity and the association confidence of the association edges between the target project instances and the target requirements, wherein project complexity is determined based on the project scale contained in the target instance data of the target project instances; a ninth determining submodule, used to determine the strategic weight adjustment value based on the requirement priority contained in the target requirement data of the target requirements; and a tenth determining submodule, used to determine the contribution of the target project instances to the budget overspending based on the deviation percentage, multi-dimensional weight matrix, project feature vector, historical risk correction value, and strategic weight adjustment value, wherein the elements in the multi-dimensional weight matrix include: project complexity weight, association strength weight, and interaction influence weight; and the historical risk correction value is determined based on the execution status data of historical project instances.

[0184] Optionally, the budget control device further includes: a second construction module, used to construct an initial budget consumption prediction model based on a long short-term memory neural network before determining the project budget consumption prediction result using a budget consumption prediction model based on the budget aggregation verification result and execution status data; a second acquisition module, used to acquire historical instance data of multiple historical project instances and process all historical instance data to obtain historical budget time series data; and a first training module, used to train the initial budget consumption prediction model using the historical budget time series data until the budget prediction value output by the initial budget consumption prediction model and the actual amount contained in the historical instance data are less than a preset threshold, thereby obtaining the budget consumption prediction model.

[0185] Optionally, the budget consumption prediction model includes at least a long short-term memory neural network structure, and the second determining unit includes: a second processing module for processing the budget aggregation verification results and execution status data to obtain budget time series data; a first capturing module for inputting the budget time series data into the budget consumption prediction model to capture the dependency features in the budget time series data using the long short-term memory neural network structure to obtain the budget prediction value; an eighth determining module for determining the dynamic fault tolerance threshold of each target project instance based on the project complexity, market volatility coefficient, and historical overspending probability of the target project instance, using a dynamic fault tolerance threshold calculation strategy; and a ninth determining module for determining the project budget consumption prediction result based on the budget prediction value and dynamic fault tolerance threshold of each target project instance.

[0186] Optionally, the adjustment unit includes: a tenth determination module, used to determine the target deviation type and target deviation level based on the budget aggregation verification results; an eleventh determination module, used to determine a target template from a preset template library based on the target deviation type, wherein the preset template library includes: multiple suggested templates associated with the deviation type; a first adjustment module, used to adjust the model parameters in the target template based on the target deviation level and the project budget consumption prediction results to obtain a budget deviation rectification plan; a first generation module, used to generate a management and control effectiveness assessment report based on the budget deviation rectification plan, the budget aggregation verification results, and the project budget consumption prediction results; and a second adjustment module, used to adjust the project data of each target project instance based on the budget deviation rectification plan and the management and control effectiveness assessment report to obtain the target budget for executing each target project instance.

[0187] The aforementioned budget control device may also include a processor and a memory. The first determining unit 30, the verification unit 31, the second determining unit 32, the adjustment unit 33, etc., are all stored in the memory as program units, and the processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0188] The aforementioned processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured. By adjusting kernel parameters, the project data for each target project instance in the target topology-related sub-network is adjusted based on the project budget consumption prediction results and budget aggregation verification results, thus obtaining the target budget for executing each target project instance.

[0189] The aforementioned memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0190] This invention also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program with the following method steps: based on a preset topology association network, determining a set of target project instances corresponding to the target requirements, obtaining a target topology association subnetwork; based on execution status data, performing differential verification on the set of target project instances to obtain an initial budget verification result; and, if the initial budget verification result indicates budget overrun, generating a budget aggregation verification result based on the set of target project instances; based on the budget aggregation verification result and execution status data, using a budget consumption prediction model to determine the project budget consumption prediction result; and based on the project budget consumption prediction result and the budget aggregation verification result, adjusting the project data of each target project instance in the target topology association subnetwork to obtain the target budget for executing each target project instance.

[0191] According to another aspect of the present invention, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements a project lifecycle budget control method as described above.

[0192] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the above-described project lifecycle budget control method.

[0193] Figure 4 This is a hardware structure block diagram of an electronic device (or mobile device) for a budget control method throughout the project lifecycle, according to an embodiment of the present invention. Figure 4 As shown, an electronic device may include one or more processors (e.g., Figure 4 The processors 402a, 402b, ..., 402n, etc., may include, but are not limited to, processing devices such as microprocessors (MCUs) or programmable logic devices (FPGAs), and a memory 404 for storing data. In addition, it may include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a keyboard, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 4 The structure shown is for illustrative purposes only and does not limit the structure of the electronic device described above. For example, the electronic device may also include components that are more... Figure 4 The more or fewer components shown, or having the same Figure 4 The different configurations shown.

[0194] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0195] The embodiments or examples disclosed herein are not exhaustive, but merely illustrative of some embodiments or examples, and are not intended to limit the scope of protection of this disclosure. Unless otherwise specified, each step in a particular embodiment or example can be implemented as an independent embodiment, and the steps can be arbitrarily combined. For example, a solution after removing some steps in a particular embodiment or example can also be implemented as an independent embodiment, and the order of the steps in a particular embodiment or example can be arbitrarily interchanged. Furthermore, optional methods or examples in a particular embodiment or example can be arbitrarily combined; moreover, embodiments or examples can be arbitrarily combined. For example, some or all steps of different embodiments or examples can be arbitrarily combined, and a particular embodiment or example can be arbitrarily combined with optional methods or examples of other embodiments or examples.

[0196] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0197] In the several embodiments provided by this invention, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection can be through some interfaces; the indirect coupling or communication connection of units or modules can be electrical or other forms.

[0198] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0199] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0200] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0201] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A budget control method for the project lifecycle, characterized in that, include: Based on a preset topology association network, a set of target project instances corresponding to the target requirements is determined, resulting in a target topology association subnetwork. The preset topology association network records the relationship between each requirement and at least one project instance. Each project instance corresponds to project data, which includes at least: budget weight coefficient and execution status data. The execution status data includes at least: the budget amount, design amount, and actual amount of the project instance at each stage. Based on the execution status data, differential verification is performed on the target project instance set to obtain an initial budget verification result. If the initial budget verification result indicates that the budget is over-limited, an aggregate budget verification result is generated based on the target project instance set. Based on the budget aggregation verification results and the execution status data, the project budget consumption prediction results are determined using a budget consumption prediction model. Based on the project budget consumption prediction results and the budget aggregation verification results, the project data of each target project instance in the target topology associated sub-network is adjusted to obtain the target budget for executing each target project instance.

2. The budget control method according to claim 1, characterized in that, Before determining the set of target project instances corresponding to the target requirements based on the preset topology association network, and obtaining the target topology association subnetwork, the process also includes: Collect requirement data for each requirement and instance data for each project instance. The requirement data includes at least: requirement number, requirement type, total budget amount, and requirement priority. The instance data includes at least: project number, project size, project implementation period, and project-related requirement number. The requirement data and the instance data are preprocessed to obtain target requirement data and target instance data; Each of the aforementioned requirements is represented as a first type of node, and each of the aforementioned project instances is represented as a second type of node; Based on the requirement number in the target requirement data corresponding to the first type of node and the project-related requirement number in the target instance data corresponding to the second type of node, an association edge is constructed between the first type of node and the second type of node, wherein each association edge has an edge attribute, and the edge attribute includes at least: mapping relationship type and association confidence. For each second type of node, the budget weight coefficient of the second type of node is determined based on the target demand data and the target instance data of the second type of node; For each project instance, based on the budget weighting coefficient and the total budget amount, the execution status data of the project instance at each stage is determined, wherein the stage is a stage divided based on the life cycle of the project instance, and the stage is divided into: project initiation stage, execution stage, and acceptance stage. The preset topology network is constructed based on all first-type nodes, all second-type nodes, all associated edges, the budget weight coefficient of each second-type node, and the execution status data.

3. The budget control method according to claim 2, characterized in that, The step of determining the budget weight coefficient of the second type of node based on the target demand data and the target instance data of the second type of node includes: Based on the project scale contained in the target instance data, a non-linear transformation value for the project scale is determined; based on the requirement priority contained in the target requirement data, a non-linear transformation value for the requirement priority is determined; and based on the project implementation cycle contained in the target instance data, a non-linear transformation value for the project cycle is determined. Based on the nonlinear transformation values ​​of the project size, the demand priority, and the project cycle, a feature row matrix is ​​determined. Based on the project size, determine the probability of risk associated with project size; based on the demand priority, determine the probability of risk associated with demand priority; and based on the project implementation cycle, determine the probability of risk associated with project cycle. Based on the project size-related risk probability, the demand priority-related risk probability, and the project cycle-related risk probability, a risk probability column matrix is ​​determined. Based on the aforementioned demand priorities, the nonlinear amplification value is determined; Based on the feature row matrix, the multi-feature collaborative weight matrix, the risk probability column matrix, the historical feedback dynamic adjustment value, and the nonlinear amplification value, the budget weight coefficient of the second type of node is determined, wherein the element values ​​in the multi-feature collaborative weight matrix are determined based on historical project instance data, and the historical feedback dynamic adjustment value is determined based on the execution status data of historical project instances.

4. The budget control method according to claim 1, characterized in that, The steps for performing differential verification on the target project instance set based on the execution status data to obtain the initial budget verification result include: For each target project instance in the target project instance set, if the mapping relationship between the target project instance and the target requirement is a one-to-one mapping relationship, the design amount and the budget amount are extracted from the execution status data, and the design amount and the budget amount are compared. If the design amount is less than or equal to the budget amount, the initial budget verification result is determined to be compliant; or, if the design amount is greater than the budget amount, the initial budget verification result is determined to be budget overrun. When the mapping relationship between the target project instance and the target requirement is a one-to-many mapping relationship, the total design amount is determined based on the execution status data of all target project instances associated with the target requirement, and the total budget amount of the target requirement is determined. If the total design amount is less than or equal to the total budget, the initial budget verification result is determined to be compliant; or if the total design amount is greater than the total budget, the initial budget verification result is determined to be budget overrun.

5. The budget control method according to claim 1, characterized in that, The steps for generating budget aggregation verification results based on the target project instance set include: For each target project instance in the target project instance set, the contribution of the target project instance to the budget overrun is determined based on the execution status data of the target project instance. The target project instance with the highest contribution is identified as the preset project instance, and an adjustment strategy is generated for the preset project instance; The preset project instance and the adjustment strategy are integrated with the initial budget verification result to obtain the budget aggregation verification result.

6. The budget control method according to claim 5, characterized in that, The step of determining the contribution of the target project instance to the budget overrun based on the execution status data of the target project instance includes: Based on the design amount, the budget amount, and the sum of positive deviations, the deviation percentage is determined, wherein the sum of positive deviations is determined based on the overspending amount of all overspending project instances in the target requirements, and the overspending project instances refer to target project instances where the design amount exceeds the budget amount. Based on the project complexity and the association confidence of the association edge between the target project instance and the target requirement, the project feature vector is determined, wherein the project complexity is determined based on the project size contained in the target instance data of the target project instance. Based on the demand priorities contained in the target demand data, determine the strategic weight adjustment value; Based on the deviation ratio, the multi-dimensional weight matrix, the project feature vector, the historical risk correction value, and the strategic weight adjustment value, the contribution of the target project instance to the budget overrun is determined. The elements in the multi-dimensional weight matrix include: project complexity weight, correlation strength weight, and interaction influence weight. The historical risk correction value is determined based on the execution status data of historical project instances.

7. The budget control method according to claim 1, characterized in that, Before determining the project budget consumption forecast result using the budget consumption forecast model based on the budget aggregation verification result and the execution status data, the process also includes: A prediction model for initial budget consumption is constructed based on a long short-term memory neural network. Collect historical instance data from multiple historical project instances, and process all the historical instance data to obtain historical budget time series data; The initial budget consumption prediction model is trained using the historical budget time series data until the budget prediction value output by the initial budget consumption prediction model and the actual amount contained in the historical instance data are less than a preset threshold, thus obtaining the budget consumption prediction model.

8. The budget control method according to claim 1, characterized in that, The budget consumption prediction model includes at least a long short-term memory neural network structure. Based on the budget aggregation verification results and the execution status data, the steps of determining the project budget consumption prediction results using the budget consumption prediction model include: The budget aggregation verification results and the execution status data are processed to obtain budget time series data; The budget time series data is input into the budget consumption prediction model, and the long short-term memory neural network structure is used to capture the dependency features in the budget time series data to obtain the budget prediction value. For each target project instance, a dynamic fault tolerance threshold is determined by using a dynamic fault tolerance threshold calculation strategy based on the project complexity, market volatility coefficient, and historical overspending probability of the target project instance. The project budget consumption prediction result is determined based on the budget forecast value and the dynamic fault tolerance threshold for each target project instance.

9. The budget control method according to claim 1, characterized in that, Based on the project budget consumption prediction results and the budget aggregation verification results, the steps of adjusting the project data for each target project instance in the target topology associated subnetwork to obtain the target budget for executing each target project instance include: Based on the budget aggregation verification results, the target deviation type and target deviation level are determined; Based on the target deviation type, a target template is determined from a preset template library, wherein the preset template library includes: multiple suggested templates associated with the deviation type; Based on the target deviation level and the project budget consumption prediction results, the model parameters in the target template are adjusted to obtain a budget deviation rectification plan; Based on the budget deviation rectification plan, the budget aggregation verification results, and the project budget consumption prediction results, a management and control effectiveness assessment report is generated. Based on the budget deviation rectification plan and the management and control effectiveness evaluation report, the project data for each target project instance is adjusted to obtain the target budget for executing each target project instance.

10. A budget control device for project lifecycle, characterized in that, include: The first determining unit is used to determine the set of target project instances corresponding to the target requirements based on a preset topology association network, and to obtain a target topology association sub-network. The preset topology association network records the relationship between each requirement and at least one project instance. Each project instance corresponds to project data, and the project data includes at least: budget weight coefficient and execution status data. The execution status data includes at least: the budget amount, design amount and actual amount of the project instance at each stage. The verification unit is used to perform differential verification on the target project instance set based on the execution status data to obtain an initial budget verification result, and generate a budget aggregation verification result based on the target project instance set if the initial budget verification result indicates that the budget is over-limited. The second determining unit is used to determine the project budget consumption prediction result based on the budget aggregation verification result and the execution status data using a budget consumption prediction model; An adjustment unit is used to adjust the project data of each target project instance in the target topology associated sub-network based on the project budget consumption prediction results and the budget aggregation verification results, so as to obtain the target budget for executing each target project instance.