A multi-agent cooperation-based intelligent review method for a procurement plan

Abstract

The application discloses a kind of based on the intelligent review method of procurement plan of multi-agent cooperation, it is related to industrial engineering procurement management and artificial intelligence technical field, including the construction of procurement task dependence adjacency matrix based on the equipment installation sequence between procurement tasks, process connection relationship and engineering node constraint relationship, and the risk propagation range is determined in combination with historical procurement execution event;Risk propagation range is sampled by multi-scale disturbance, generate multiple candidate risk propagation ranges, and map it as multi-agent review control object;The collaborative review of the procurement review intelligent agent corresponding to each candidate risk propagation range generates collaborative evaluation value;Finally, the highest candidate risk propagation range of collaborative evaluation value is screened as target risk propagation range according to the propagation structure measurement value constraint, solve the problem of risk omission, invalid review expansion and low review efficiency caused by fixed risk propagation range in the existing multi-agent collaborative procurement plan intelligent review.

Description

Technical Field

[0001] This invention relates to the fields of industrial engineering procurement management and artificial intelligence technology, and more specifically, to a method for intelligent review of procurement plans based on multi-agent collaboration. Background Technology

[0002] As the scale of equipment procurement for large-scale engineering construction projects (such as power substations, rail transit, and industrial plants) expands, the equipment installation sequence, process connection, and engineering node constraints between procurement tasks become increasingly complex. A single intelligent analysis model or manual rules can hardly meet the dynamic risk identification needs under multi-task, multi-dimensional, and multi-constraint coupled scenarios.

[0003] In recent years, multi-agent systems (MAS) and distributed constraint optimization problems (DCOP) have been introduced into procurement plan review. Existing solutions typically configure separate agents for schedule review, supply capacity review, funding review, and equipment-related review based on different review dimensions of the procurement task. Each agent performs risk analysis for its corresponding dimension in parallel, and then achieves collaborative review through information exchange.

[0004] However, existing solutions have the following shortcomings in controlling the propagation scope of procurement constraints (i.e., the scope of related tasks for each agent participating in the review) during multi-agent collaborative review: Firstly, when the scope of risk propagation is too small, some related procurement tasks are excluded from the scope of collaborative review, resulting in the risk propagation chain not being fully unfolded and risks being missed. Secondly, when the scope of risk propagation is too large, a large number of weakly related or even unrelated procurement tasks are included in the review process, causing redundancy in information exchange between intelligent agents, fluctuations in risk judgment, and frequent adjustments to procurement plans, resulting in a significant decrease in review efficiency. Third, existing solutions generally employ fixed propagation ranges or manually set association levels, failing to adaptively adjust the optimal propagation range based on actual review results. Within the framework of DCOP (Distributed Constraint Optimization), the selection of propagation range essentially corresponds to the subgraph sampling problem of the constraint graph.

[0005] Therefore, there is an urgent need for an intelligent review method that can dynamically generate multiple sets of candidate risk propagation ranges and adaptively select the target risk propagation range based on the actual effect of collaborative review. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, the purpose of this invention is to provide a method and system for intelligent review of procurement plans based on multi-agent collaboration. By constructing a dynamic generation and collaborative evaluation and screening mechanism for risk propagation scope, this invention solves the problems of risk omission, invalid review expansion and low review efficiency caused by the fixed risk propagation scope in the existing intelligent review of procurement plans based on multi-agent collaboration.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for intelligent review of procurement plans based on multi-agent collaboration includes the following steps: constructing a procurement task dependency adjacency matrix based on the equipment installation sequence, process connection relationships, and engineering node constraints among procurement tasks; calculating the transitive closure based on the procurement task dependency adjacency matrix and determining the risk propagation range of the procurement tasks by combining the abnormal synchronization changes corresponding to each propagation relationship in historical procurement execution events; performing multi-scale perturbation sampling on the risk propagation range of the procurement tasks to generate multiple candidate risk propagation ranges; establishing a mapping relationship between each candidate risk propagation range and the multi-agent review control quantity, converting each candidate risk propagation range into a corresponding multi-agent... The system involves reviewing and controlling multiple agents; configuring corresponding procurement review agents for each agent, and having these agents conduct collaborative reviews based on their initial and associated review tasks to obtain multi-agent collaborative procurement plan review process data; generating collaborative evaluation values ​​corresponding to the propagation ranges of each candidate risk based on the multi-agent collaborative procurement plan review process data; ranking and evaluating each candidate risk propagation range according to preset propagation structure metric constraints, and performing joint analysis with the corresponding collaborative evaluation values ​​to select the candidate risk propagation range with the highest collaborative evaluation value as the target risk propagation range for intelligent review of the procurement plan.

[0008] A system for intelligent review of procurement plans based on multi-agent collaboration includes a constraint propagation modeling module, a propagation range perturbation module, a review mapping control module, and a multi-agent review module. The constraint propagation modeling module constructs a procurement task dependency matrix based on the dependencies between procurement tasks. The propagation range perturbation module determines the risk propagation range of procurement tasks based on the procurement task dependency matrix and generates multiple candidate risk propagation ranges through perturbation. The review mapping control module configures procurement review agents for collaborative review of each candidate risk propagation range, generating review process data. The multi-agent review module generates a collaborative evaluation value corresponding to each candidate risk propagation range based on the review process data and selects the candidate risk propagation range with the highest collaborative evaluation value for intelligent review of the procurement plan.

[0009] The technical effects and advantages of the intelligent review method for procurement plans based on multi-agent collaboration in this invention are as follows: 1. By constructing an adjacency matrix based on procurement task dependencies and combining transitive closure analysis and historical procurement anomaly propagation verification, the risk propagation range that reflects the actual risk propagation pattern can be obtained, avoiding the unadjustable problem caused by relying solely on fixed association levels.

[0010] 2. By performing multi-scale perturbation sampling on the propagation range of each candidate risk, it can simultaneously cover direct associations, cross-first-level node associations, and associations with weak historical verification, enabling multi-agent collaborative review to compare at various risk deployment depths.

[0011] 3. By mapping the risk propagation range to the multi-agent review control quantity, different propagation ranges correspond to different agent organization methods and information interaction topologies, providing a unified control basis for subsequent selection of the optimal propagation range based on collaborative evaluation values.

[0012] 4. By constructing a coordinated evaluation model based on risk identification coverage, related task omission rate, invalid review task ratio, and agent interaction load, a quantifiable dynamic balance relationship can be established between risk coverage capability and review complexity. Furthermore, by propagating structural metric constraints, local optima can be avoided, thereby improving the accuracy, stability, and efficiency of intelligent review of multi-agent collaborative procurement plans. Attached Figure Description

[0013] To make the technical solution of the present invention clearer and more complete, the following description is provided in conjunction with the accompanying drawings: Figure 1 This is a flowchart illustrating an intelligent review method for procurement plans based on multi-agent collaboration according to the present invention.

[0014] Figure 2 The system architecture diagram provided in this embodiment of the invention includes a dependency modeling module, a disturbance sampling module, a mapping control module, a multi-agent review module, a coordination evaluation module, and a range optimization module, as well as data flows between the modules.

[0015] Figure 3 The procurement task dependency graph and adjacency matrix structure diagram provided in this embodiment of the invention exemplarily shows four procurement tasks (P1-P4) and their corresponding 4×4 adjacency matrices.

[0016] Figure 4 The multi-agent information interaction sequence diagram provided in this embodiment of the invention exemplifies the event transmission order of the project schedule review agent, supply capacity review agent, funding review agent, and equipment association review agent in a collaborative review.

[0017] Figure 5The schematic diagram of joint analysis of propagation structure metric and collaborative evaluation value provided in this embodiment of the invention shows that the horizontal axis is the propagation structure metric and the vertical axis is the collaborative evaluation value, with the coordinated reward enhancement area and the target risk propagation range marked. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] To better illustrate the purpose, technical solution, and advantages of 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 a part of the embodiments of the present invention, and not all of them. The following uses a specific power substation equipment procurement process as an exemplary embodiment to illustrate the technical solution of the present invention. Specific values ​​such as "delivery on August 10th," "80 million yuan," and "20 deliveries" involved in the embodiments are only illustrative and do not constitute a limitation on the scope of the claims.

[0020] Example 1, Figure 1 This invention presents an intelligent review method for procurement plans based on multi-agent collaboration, comprising the following steps: S1. Construct a directed graph matrix of procurement task dependencies, and obtain the risk propagation range of the first procurement task based on directed dependency closure analysis.

[0021] In this embodiment, the construction of a directed graph matrix of procurement task dependencies and the acquisition of the risk propagation range of the first procurement task based on directed dependency closure analysis are specifically as follows: Obtain procurement plan data and divide the planned equipment procurement items into several procurement task nodes; Based on the equipment installation sequence, process connection relationship and engineering node constraint relationship of the procurement plan data, a directed graph of procurement task association is constructed, in which procurement task nodes are used as nodes, and directed association edges are established when there are pre- and post-dependencies, installation dependencies and debugging dependencies between two procurement task nodes. Based on the direct relationships between nodes in the directed graph associated with procurement tasks, the mapping is transformed into an initial procurement task dependency adjacency matrix. The initial procurement task depends on the procurement task generated by the matrix behavior risk of the adjacency matrix; The initial procurement task depends on the columns of the adjacency matrix, which are procurement tasks that may be affected. The elements of the adjacency matrix for the initial procurement task are represented by binary values, where 1 indicates that the risk needs to be propagated to the procurement task, and 0 indicates that there is no constraint propagation relationship between the two. Based on the initial procurement task dependency adjacency matrix, a recursive Boolean expansion is performed to search for reachable paths between each procurement task node, and a directed dependency closure is generated based on all reachable paths, which serves as the maximum propagation range of the procurement task risk theory. Collect historical procurement execution events, extract historical procurement exception tasks and the status changes of related tasks after the exception occurred, and count the historical exception synchronization changes corresponding to each propagation relationship in the directed dependency closure to generate a historical propagation consistency matrix. The propagation relationship is expressed in binary form. By performing a logical intersection operation between the directed dependency closure matrix and the historical propagation consistency matrix, the procurement task dependency adjacency matrix corresponding to the procurement task propagation relationship with historical propagation consistency is retained, and the risk propagation range of the first procurement task is generated.

[0022] In this embodiment, the procurement plan data is a set of data describing the procurement process of each piece of equipment in the engineering project. This includes equipment name, procurement batch, planned procurement time, planned delivery time, installation start time, installation completion time, commissioning time, supplier information, payment node information, and information on engineering coordination with other equipment. The procurement plan data is not a single procurement list, but rather a data foundation containing the time and dependencies between equipment procurement activities and the engineering implementation process, used to subsequently identify whether there are risk transmission relationships between different procurement tasks.

[0023] In this embodiment, a procurement task node is a basic analytical unit formed by breaking down independent equipment procurement items in the procurement plan according to the actual engineering execution process. Specifically, the acquisition process is divided according to the equipment category, procurement batch, and engineering installation object in the equipment procurement plan. For example, the procurement of main transformers, protection cabinets, secondary commissioning systems, and auxiliary equipment in a substation construction project are each taken as independent procurement task nodes. Each procurement task node corresponds to a specific equipment procurement item and the corresponding time plan, supply status, and engineering-related information.

[0024] In this embodiment, the equipment installation sequence is determined according to the order of equipment installation as specified in the engineering construction organization plan. This sequence describes whether a particular piece of equipment must be installed only after another piece of equipment has been completed. Specifically, the installation sequence is determined by reading the installation procedures in the engineering construction plan. For example, if the protection cabinet can only be installed after the main transformer is installed, then there is an installation sequence constraint between the main transformer procurement task node and the protection cabinet procurement task node.

[0025] In this embodiment, the process connection relationship refers to the dependency relationship between equipment due to the construction process flow. It is used to indicate the situation where a change in the procurement status of one piece of equipment will affect the subsequent construction process execution of another piece of equipment. Specifically, during the acquisition process, based on the engineering design documents, construction process flow documents, and equipment commissioning process records, it is identified whether there is a process relationship between the equipment that must be executed continuously. For example, if secondary protection commissioning can only be carried out after the primary equipment installation is completed, then a process connection relationship is established between the primary equipment procurement task and the secondary commissioning equipment procurement task.

[0026] In this embodiment, the engineering node constraint relationship is the constraint relationship formed between procurement tasks and key project time nodes, used to determine whether a delay in a procurement task will affect the overall project milestones. Specifically, during the acquisition process, the expected delivery time of equipment, installation window time, commissioning start time, and project milestone nodes are analyzed accordingly. When the delivery time of a certain piece of equipment exceeds the corresponding installation window and causes subsequent project nodes to be unable to be executed, an engineering node constraint relationship is established between the equipment procurement task and the affected project tasks.

[0027] In this embodiment, the directed graph of procurement tasks is a data structure used to describe the direction of risk propagation between procurement tasks. Procurement task nodes are represented as nodes in the graph, and the dependencies between procurement tasks are represented as directed edges. Specifically, during the construction process, each procurement task node is treated as an element in the graph. When an anomaly occurs in procurement task A, affecting the execution of procurement task B, a directed edge from A to B is established between procurement task A and procurement task B. This allows the graph to reflect the direction of procurement risk propagation from the generating task to the affected task.

[0028] In this embodiment, a directed association edge represents the directional relationship connecting two procurement task nodes, indicating the direction of risk propagation. During the generation process, the relationship is determined based on the equipment installation sequence, process connections, and engineering node constraints. When a prerequisite task is not completed, preventing the execution of subsequent tasks, a directed association edge is generated, with the prerequisite task as the starting node and the affected task as the ending node. For example, if a delay in the main transformer procurement affects the installation of the protection cabinet, a directed association edge is established between the main transformer procurement task and the protection cabinet procurement task.

[0029] In this embodiment, as Figure 3As shown, the initial procurement task adjacency matrix is ​​a two-dimensional data matrix obtained by transforming the directed graph associated with procurement tasks, used to represent the direct risk propagation relationship between procurement tasks. During the specific generation process, all procurement task nodes are arranged in a fixed order to form matrix rows and columns. Matrix rows correspond to risk-generating tasks, and matrix columns correspond to tasks that may be affected. When there is a direct propagation relationship between two procurement tasks, the corresponding position is recorded as 1; otherwise, it is recorded as 0. For example, if the main transformer procurement task directly affects the protection cabinet procurement task, then the positions of the main transformer's corresponding row and the protection cabinet's corresponding column are recorded as 1.

[0030] In this embodiment, the recursive Boolean extension is a process of searching for indirect propagation relationships layer by layer based on the initial procurement task-dependent adjacency matrix. This process is used to discover procurement tasks that are not directly related but have risk transmission paths. Specifically, the first-level direct influence relationship is first obtained based on the initial matrix, for example, A influences B. Then, the search continues to determine whether B influences C. If it does, it is considered that A can indirectly influence C through B, and this indirect influence relationship is recorded in the expanded propagation relationship. The above path search process is continued until no new procurement task influence relationships can be found, thereby obtaining a complete set of risk propagation paths.

[0031] In this embodiment, a reachable path is a risk propagation route formed by connecting procurement task nodes through one or more directed edges. Specifically, the process begins with a procurement task that generates a risk, and searches for subsequent task nodes level by level along the directed edges. For example, if procurement task A is connected to B via an edge, and B is further connected to C, then there is a reachable path from A to C, indicating that when an anomaly occurs at A, the risk can propagate further to C via B.

[0032] In this embodiment, the directed dependency closure is the maximum set of risk propagation relationships generated based on all reachable paths of procurement tasks. It represents the range of all procurement tasks that an anomaly in a theoretical procurement task might affect. Specifically, during the generation process, all direct or indirect path relationships between procurement tasks are summarized. If a procurement task can influence another procurement task through any number of associated nodes, the corresponding propagation relationship is retained in the closure matrix. This result represents the maximum boundary of risk propagation; however, since only engineering dependencies are considered and actual historical probability is not taken into account, the propagation range is usually larger than the actual effective impact range.

[0033] In this embodiment, historical procurement execution events refer to procurement anomaly data recorded during the actual execution of the project, including equipment delay events, supply anomaly events, payment conflict events, delivery anomaly events, and information on changes in the status of related equipment after the anomaly occurred. Specifically, the historical project procurement execution records are read to extract the procurement tasks that experienced anomalies, the time of the anomaly, the duration of the anomaly, and whether other procurement tasks underwent planned changes after the anomaly, in order to analyze the actual risk propagation.

[0034] In this embodiment, historical procurement exception tasks are procurement task nodes where anomalies actually occurred during the historical procurement execution process. Specifically, during the acquisition process, judgments are made based on the difference between the planned time and the actual execution time, changes in supply delivery status, and records of the impact of engineering nodes in the historical procurement execution records. When a procurement task is delayed, cannot be delivered as planned, or leads to adjustments to subsequent tasks, the procurement task is recorded as a historical procurement exception task.

[0035] In this embodiment, the change in the status of associated tasks refers to the actual execution changes of other procurement tasks that are potentially related to the abnormal procurement task after it occurs. Specifically, the process starts with historical abnormal procurement tasks and reads the plan adjustment records of other procurement tasks after the abnormality occurred. When other tasks have adjustments to delivery time, installation plan, supply arrangement, or payment node, these changes are recorded as changes in the status of associated tasks.

[0036] In this embodiment, the historical propagation consistency matrix is ​​a binary matrix used to describe the degree of consistency between theoretical propagation relationships and historical actual propagation results. Specifically, during the generation process, for each risk propagation relationship in the directed dependency closure, it checks whether there has ever been a situation in historical procurement events where the target task synchronously changed state after the initial task became abnormal. If this propagation phenomenon exists in historical data, the corresponding position is recorded as 1, indicating that the propagation relationship has historical verification basis; if this propagation phenomenon has not occurred for a long time, the corresponding position is recorded as 0, indicating that the propagation relationship lacks actual propagation support.

[0037] In this embodiment, the logical intersection operation is a data filtering process that retains data that simultaneously satisfies both theoretical dependencies and historical propagation verification relationships. Specifically, the propagation relationships in the directed dependency closure are compared item by item with the propagation relationships in the historical propagation consistency matrix, retaining only those positions in both matrices that are valid propagation relationships. For example, if an engineering dependency exists between devices, but no risk synchronization impact has ever occurred in historical projects, then the propagation relationship is deleted; if the engineering dependency exists and synchronization impacts have occurred multiple times in history, then the propagation relationship is retained.

[0038] In this embodiment, the risk propagation scope of the first procurement task is a risk propagation scope matrix after being constrained by theoretical propagation scope and corrected by historical execution data. This matrix serves as the basis for generating different review scope schemes. Specifically, the maximum risk propagation boundary is determined using directed dependency closure, and then the historical propagation consistency matrix is ​​used to remove relationships lacking actual propagation basis. This ensures that the final risk propagation scope of the first procurement task covers both critical tasks where risk transmission may occur and avoids including all tasks in the entire project that have no actual impact within the review scope.

[0039] It should be noted that when constructing the directed graph of procurement task relationships, the relationships between procurement tasks can be dynamically updated by incorporating data from different project phases. Since design changes, construction sequence adjustments, and supply scheme changes may occur during the procurement process, the dependencies between procurement tasks are not fixed. By periodically reading newly added procurement execution data and project adjustment data, the propagation relationships between procurement tasks can be reassessed. This allows the risk propagation range of the first procurement task to be updated as the project status changes, providing a more accurate foundation for subsequent multi-scale path disturbances based on the current project status.

[0040] It should be noted that the risk propagation scope of the first procurement task in this step is not used to directly determine the final review scope, but rather serves as the initial propagation structure for subsequent optimization searches. Since directly adopting the theoretical maximum propagation scope can easily lead to a large number of irrelevant procurement tasks entering the agent's review process, increasing the amount of information interaction between agents and reducing review efficiency, it is corrected through historical propagation consistency. This ensures that the initial propagation scope falls between the theoretical maximum scope and the actual effective scope, providing a convergence basis for subsequently finding the optimal propagation scope.

[0041] In a specific embodiment, the equipment procurement plan for a certain 220kV substation includes the procurement tasks P1 for main transformers, P2 for protection cabinets, P3 for secondary commissioning systems, and P4 for auxiliary equipment.

[0042] According to the project installation sequence, the protection cabinet can only be installed after the main transformer is installed, and secondary commissioning can only be carried out after the protection cabinet is installed. Therefore, there is a directed dependency edge between P1, P2, and P3.

[0043] The auxiliary equipment has no strong dependency on the main link mentioned above, and is classified as a weakly related task P4.

[0044] This yields the initial adjacency matrix M: Table 1:

[0045] Note: M[i][j]=1 indicates that an anomaly in task Pi will affect task Pj.

[0046] Based on the adjacency matrix described above, a multi-step path search is performed to obtain the transitive closure M. M The propagation relationships where all elements are 1 constitute the maximum propagation range of risk theory. In this embodiment, M The indirect propagation relationship between P1 and P3 is added relative to M.

[0047] The procurement execution data of 10 similar substation projects in history were collected, and the historical anomaly synchronization occurrences corresponding to each propagation relationship were statistically analyzed. Two propagation relationships with historical verification, P1→P2 and P2→P3, were retained, and the theoretically reachable but historically unverified P1→P3 relationship was deleted, forming the procurement task risk propagation range R1 in this embodiment.

[0048] According to the engineering construction process, the main transformer needs to be installed first before the protection cabinet can be installed. The secondary commissioning system needs to be commissioned after the protection cabinet is installed. Therefore, a directed association relationship is established between P1 and P2, and between P2 and P3.

[0049] Further transformation yields the procurement task dependency adjacency matrix, where P1 affects P2, and P2 affects P3, so the corresponding position in the matrix is ​​recorded as 1.

[0050] Through recursive path search, it was found that although P1 is not directly connected to P3, P1 can influence P3 through P2, thus forming an indirect propagation relationship from P1 to P3, and generating a directed dependency closure containing the propagation relationship of P1, P2, and P3.

[0051] Subsequently, by reading the procurement execution data of multiple substation projects in the past, it was found that when the main transformer was delayed in the past projects, the installation plan of the protection cabinet was adjusted. However, the procurement of auxiliary equipment in some projects was not affected. Therefore, the propagation relationship between the main transformer and the auxiliary equipment that did not have historical verification was deleted, and the risk propagation scope of the first procurement task was finally formed.

[0052] It should be noted that this step, by constructing a directed graph of procurement task associations and combining it with directed dependency closure analysis, achieves a structured description of the scope of procurement risk propagation, enabling the procurement plan review process to transform from a single procurement task inspection to a collaborative review oriented towards the risk propagation chain.

[0053] It should be noted that by using historical procurement execution events to modify the theoretical maximum propagation range, we can avoid the problem in traditional procurement review methods where the risk impact range is set too large, resulting in a large number of irrelevant tasks participating in the review. At the same time, we can avoid the problem of missing the risk propagation chain due to focusing only on directly related tasks. This makes the obtained risk propagation range of the first procurement task more consistent with the risk diffusion law in the actual engineering procurement process, and provides a reliable data foundation for subsequent multi-agent collaborative review scope optimization.

[0054] S2, based on multi-scale path perturbation, extends the risk propagation range of the first procurement task into several risk propagation ranges of the second procurement task.

[0055] In this embodiment, the step of extending the risk propagation range of the first procurement task into several risk propagation ranges of the second procurement task based on multi-scale path perturbation specifically involves: Obtain the procurement task dependency adjacency matrix corresponding to the risk propagation scope of the first procurement task, and obtain the directed dependency closure matrix. The associated positions in the directed dependency closure matrix where the element is 1 and the element in the risk propagation range matrix of the first procurement task is 0 are used as the scalable propagation set. Based on the existing propagation relationships and expandable propagation sets in the risk propagation scope matrix of the first procurement task, different perturbation scales are divided according to the length of the procurement task-related path. Based on different perturbation scales, the propagation range of the associated elements in the risk propagation range matrix of the first procurement task is sampled several times based on the perturbation scale. The expansion operation only adjusts some elements in the expandable propagation set from 0 to 1, while the contraction operation only adjusts the elements in the matrix of the risk propagation scope of the first procurement task from 1 to 0. Several risk propagation range matrices for the second procurement task are generated by combining the expansion and contraction results under different perturbation scales.

[0056] In this embodiment, M The propagation relationships that exist in R1 but not in R1 are considered as an expandable propagation set. In this embodiment, the expandable propagation set only contains P1→P3.

[0057] Disturbance scales are divided according to path length: short-range disturbances (path length = 1) do not adjust any relationships; medium-range disturbances (path length = 2) allow the addition or deletion of P1→P3; long-range disturbances (path length = 3 and above) allow adjustments over a larger range.

[0058] Through multiple rounds of perturbation sampling, five candidate risk propagation ranges R1-R5 were finally generated, each corresponding to a different risk deployment depth.

[0059] In this embodiment, the risk propagation scope of the first procurement task is an initial risk propagation scope matrix obtained after procurement task dependency analysis and correction based on historical procurement execution events. This matrix represents the set of procurement tasks in the current engineering procurement plan for which risk propagation relationships have been verified. Specifically, each element in the matrix corresponds to a propagation relationship between two procurement tasks. An element of 1 indicates that when an anomaly occurs in the corresponding procurement task in a row, the corresponding procurement task in a column may be affected by the risk; an element of 0 indicates that no risk propagation relationship has been established between the two tasks within the current scope. This matrix does not represent all theoretically possible influencing relationships, but rather a basic propagation scope filtered through historical data, used as the initial state for subsequently expanding or narrowing the review scope.

[0060] In this embodiment, the directed dependency closure matrix is ​​the maximum propagation range matrix obtained by expanding all reachable paths in the directed graph of procurement task associations. It is used to describe all risk propagation relationships that may form between procurement tasks under theoretical dependency conditions. Specifically, in the acquisition process, based on the directed graph of procurement task associations corresponding to the risk propagation range of the first procurement task, starting from each risk-generating task node, subsequent task nodes are searched layer by layer along the existing directed association relationships. When a task can influence another task through one or more intermediate tasks, it is considered that there is a reachable relationship between the two, and the propagation relationship is recorded at the corresponding position in the directed dependency closure matrix. This matrix has a coverage area larger than the risk propagation range of the first procurement task, and is used to provide the maximum boundary when expanding the subsequent propagation range.

[0061] In this embodiment, the scalable propagation set represents the set of risk propagation relationships that may increase beyond the current risk propagation range of the first procurement task, but still within the theoretical propagation range. Specifically, during the acquisition process, the directed dependency closure matrix and the risk propagation range matrix of the first procurement task are compared in terms of their corresponding positions. When a position is recorded as having a propagation relationship in the directed dependency closure matrix but not in the first procurement task risk propagation range matrix, that position is considered a scalable propagation relationship. For example, the main transformer procurement task could theoretically affect the commissioning system through the protection cabinet, but historical data has not yet verified this relationship; therefore, this relationship may be considered as an expansion candidate for subsequent disturbances.

[0062] In this embodiment, the path length of the procurement task association is the number of associated nodes traversed during risk propagation between two procurement task nodes, used to describe the proximity of different propagation relationships to the risk-generating task. Specifically, during the acquisition process, reverse tracing is performed based on the propagation path corresponding to the directed dependency closure matrix, and the number of associated edges traversed between the risk-generating task and the affected task is counted. For example, if the main transformer procurement task directly affects the protection cabinet procurement task, the path length is level one; if the main transformer further affects the commissioning system through the protection cabinet, the path length is level two. The larger the path length, the more the propagation relationship relies on intermediate tasks for transmission.

[0063] In this embodiment, the perturbation scale is the range of propagation adjustment obtained based on the length of the associated path of the procurement task, used to control the extent to which the risk propagation range of the first procurement task expands or contracts. Specifically, during the acquisition process, the propagation path distance in the scalable propagation set is used to create a hierarchical structure. Propagation relationships between directly associated tasks are classified as short-range perturbation ranges, while propagation relationships requiring transmission through multiple task nodes are classified as long-range perturbation ranges. For example, for first-level propagation relationships, only the range between directly associated devices can be adjusted; for second-level and higher propagation relationships, the risk propagation range across device links can be adjusted. By setting different scales, the generated risk propagation range of the second procurement task can cover different levels of risk expansion.

[0064] In this embodiment, multi-scale path perturbation is a process of restrictively changing the propagation relationships within the risk propagation range of the first procurement task based on different propagation path distances, used to generate multiple candidate schemes with different risk propagation ranges. In specific implementation, instead of directly and randomly modifying arbitrary positions in the matrix, within the theoretical range determined by the directed dependency closure, adjustable relationships are selected based on the propagation path distance, and operations to add or reduce propagation relationships are performed respectively, thereby forming candidate matrices with different propagation range sizes but all conforming to engineering dependency constraints.

[0065] In this embodiment, random continuous expansion refers to a processing method that continuously adds multiple related propagation relationships during a single disturbance process. This is used to simulate the increased review scope that may occur when procurement risks further spread. Specifically, during the acquisition process, multiple propagation relationship positions that meet the requirements of the current disturbance scale are selected from the expandable propagation set, and the corresponding matrix elements are adjusted from 0 to 1. For example, the original risk propagation scope of the first procurement task only included the main transformer affecting the protection cabinet. When the expanded disturbance adds the relationship between the protection cabinet and the commissioning system, a new propagation scope is formed, further expanding the scope of intelligent review.

[0066] In this embodiment, random continuous shrinking refers to a process of continuously deleting some existing risk propagation relationships during a single perturbation, used to simulate the adjustment of the review scope when the impact range of procurement risks is reduced. Specifically, during the acquisition process, only the propagation relationships already existing in the risk propagation scope matrix of the first procurement task are adjusted, changing some propagation relationships from a valid state to an invalid state. For example, historical data shows that although a certain auxiliary equipment is engineering-related to the main equipment, it is rarely affected in actual abnormal situations. Therefore, a candidate propagation scope that does not include this propagation relationship can be generated by shrinking the perturbation.

[0067] In this embodiment, the expansion operation is a matrix adjustment process that adds risk propagation relationships between procurement tasks. During execution, only elements in the expandable propagation set are allowed to be modified, adding propagation relationships that were not originally present in the risk propagation scope of the first procurement task to the candidate matrix. No artificial additions of relationships beyond the scope of the directed dependency closure are permitted, thus ensuring that the expanded propagation scope still conforms to the engineering task dependency logic.

[0068] In this embodiment, the contraction operation is a matrix adjustment process that reduces existing risk propagation relationships between procurement tasks. Specifically, only the positions where the element in the risk propagation range matrix of the first procurement task is 1 are processed, deleting some propagation relationships so that the generated candidate propagation range can cover the smaller review scope in cases of insufficient risk propagation.

[0069] In this embodiment, the second procurement task risk propagation range matrix is ​​a plurality of candidate risk propagation range matrices generated based on the first procurement task risk propagation range after perturbation at different scales. Each second procurement task risk propagation range matrix represents a possible procurement task risk propagation structure and corresponds to a different multi-agent review task range. For example, one second procurement task risk propagation range may only include direct equipment impact relationships, while another second procurement task risk propagation range may further include cross-equipment debugging impact relationships, used for subsequent comparison of intelligent review effects under different propagation ranges.

[0070] It should be noted that the perturbation in this step is not an unconstrained random modification of the procurement task's dependency adjacency matrix. Instead, it uses a directed dependency closure matrix to limit the adjustment boundary, taking the risk propagation range of the first procurement task as the base state, and only generates new propagation structures within the theoretically propagable range. This approach avoids generating invalid propagation ranges without engineering dependencies while retaining the ability to explore better review ranges.

[0071] It should be noted that the scope of procurement risk propagation changes dynamically at different project stages. For example, in the early stages of a project, the focus is mainly on equipment procurement and supply fulfillment, while in the installation and commissioning phase, it is necessary to further focus on the inter-equipment impact. Therefore, by setting disturbance scales corresponding to different path lengths, multiple risk propagation scopes for secondary procurement tasks applicable to different risk propagation depths can be generated, providing various control schemes for subsequent multi-agent collaborative review.

[0072] In a specific embodiment, taking the procurement process of large industrial plant equipment as an example, the current procurement tasks include the procurement of production equipment, control cabinets, installation accessories, and system debugging equipment.

[0073] Based on historical procurement execution data, the current risk propagation scope of the primary procurement task only records that delays in the procurement of production equipment will affect the installation of control cabinets, and delays in the procurement of control cabinets will affect the installation of commissioning equipment, thus forming a basic propagation chain: Procurement of production equipment – ​​procurement of control cabinets – procurement of commissioning equipment.

[0074] Further analysis of directed dependency closures revealed that, theoretically, the procurement of production equipment may also affect the procurement of system debugging equipment, i.e., the following exists: Procurement of production equipment – ​​procurement of control cabinets – procurement of commissioning equipment.

[0075] Although this indirect propagation relationship has engineering dependencies, it has not been fully verified in historical data, and therefore serves as a scalable propagation set.

[0076] Subsequently, perturbations are applied based on the propagation path length: Type I disturbances only adjust direct correlations, such as the propagation relationship where increased procurement of production equipment directly affects procurement of installation accessories; The second type of disturbance adjusts the propagation relationship across primary nodes, such as increasing the procurement of production equipment affecting the propagation relationship of the procurement of debugging equipment; The third type of disturbance reduces some of the propagation relationships with weaker historical impacts, such as the impact of removing the procurement of installation accessories on commissioning preparation work.

[0077] The final result is a matrix of multiple different risk propagation ranges for the second procurement task, each representing the scope of intelligent review under different risk levels.

[0078] It should be noted that this step expands the single risk propagation range into multiple candidate ranges with different propagation depths and breadths by perturbing the risk propagation range of the first procurement task at multiple scales. This allows the subsequent multi-agent review process to compare under multiple risk propagation structures, rather than using a fixed review range.

[0079] It should be noted that if the risk propagation range of a procurement task is too small, cross-task risk propagation chains may be missed, preventing the agent from detecting subsequent equipment delays, installation conflicts, and debugging impacts. Conversely, if the risk propagation range of a procurement task is too large, a large number of weakly related tasks will be included in the review, expanding the agent's interaction scope and leading to frequent adjustments to the procurement plan. This step generates multiple candidate propagation structures between the theoretical maximum propagation range and the actual effective propagation range, providing optimization space for subsequent selection of the optimal review range based on collaborative evaluation values.

[0080] S3 establishes a conversion relationship between the scope of risk propagation in procurement tasks and the multi-agent review and control quantity, converting the scope of risk propagation in the second procurement task into the intelligent review and control object of the procurement plan.

[0081] In this embodiment, the establishment of the conversion relationship between the risk propagation scope of the procurement task and the multi-agent review and control quantity, converting the risk propagation scope of the second procurement task into the intelligent review and control object of the procurement plan, specifically involves: Obtain the procurement task dependency adjacency matrix corresponding to the risk propagation range of each second procurement task, and extract the risk impact task set corresponding to each procurement task node based on the row propagation relationship of each procurement task node in the matrix. Based on the risk impact task set, the risk impact task set corresponding to each procurement task node is converted into the task set to be reviewed by the corresponding review agent. Among them, the procurement task that generates the risk is used as the initial review object of the agent, and the procurement task with the corresponding propagation relationship of 1 in the matrix is ​​used as the associated review object. Based on the propagation direction between procurement tasks in the second procurement task risk propagation range matrix, establish the information transmission relationship between each review agent, take the procurement task status information corresponding to the associated review object as the input information of the corresponding review agent, and take the risk judgment result generated by the review agent as the feedback information of the corresponding associated review object. Based on the set of tasks to be reviewed, the range of input information, and the range of output feedback corresponding to each review agent, an intelligent review and control object for the procurement plan corresponding to the risk propagation range of the second procurement task is generated.

[0082] In this embodiment, the risk propagation range of the second procurement task is a matrix of multiple procurement task risk propagation ranges generated through multi-scale path perturbation. This matrix represents the task coverage of multi-agent procurement plan review under different risk diffusion scenarios. Each second procurement task risk propagation range corresponds to a procurement task dependency adjacency matrix. Rows in the matrix represent procurement tasks that may generate risks, and columns represent procurement tasks that may be affected by risks. A matrix element of 1 indicates a risk propagation relationship between two corresponding procurement tasks, while a matrix element of 0 indicates no risk association requiring review within the current propagation range. For example, if the position of the protection cabinet procurement task corresponding to the main transformer procurement task in a certain second procurement task risk propagation range is 1, it means that when the main transformer procurement is abnormal, the protection cabinet procurement task needs to be included in the collaborative review scope.

[0083] In this embodiment, taking R3 as an example, both P1→P2 and P2→P3 in its adjacency matrix are valid. Based on this, three procurement review agents are configured: agent A1 is responsible for P1 and associated with P2; agent A2 is responsible for P2 and associated with both P1 and P3; agent A3 is responsible for P3 and associated with P2.

[0084] The direction of information transmission between agents is determined by the propagation direction of the adjacency matrix, as shown in the appendix. Figure 4 As shown.

[0085] In this embodiment, the procurement task dependency adjacency matrix is ​​a data matrix describing the risk propagation relationship between procurement tasks. It is used to convert the engineering dependencies between procurement tasks into a data structure that can be called by multiple agents. Specifically, during the acquisition process, all procurement tasks in the project are numbered in a fixed order. Each procurement task corresponds to a row position and a column position in the matrix. The matrix elements are filled according to the risk propagation relationship recorded in the risk propagation range of the second procurement task, so that the matrix can reflect the direction of risk impact between different procurement tasks.

[0086] In this embodiment, the row-wise propagation relationship refers to the process of reading all propagation relationships in the row containing a procurement task, using a specific procurement task as the starting point of risk. Specifically, during the acquisition process, all matrix elements in the row containing a procurement task are read. When a column's corresponding element is 1, it indicates that the procurement task corresponding to that column belongs to the associated tasks that the current risk task can influence. For example, in the row containing the main transformer procurement task, if the corresponding positions for the protection cabinet procurement task and the commissioning system procurement task are both 1, then the protection cabinet and commissioning system are extracted as the risk-affecting tasks corresponding to the main transformer procurement task.

[0087] In this embodiment, the risk impact task set is the set of all related affected procurement tasks extracted from the risk propagation range matrix of the second procurement task, starting from a procurement task that generates a risk. Specifically, for each procurement task node, the propagation relationships in the corresponding matrix row are traversed, and all columns with elements equal to 1 are summarized to form the risk impact task set corresponding to that procurement task. For example, if the positions of P2 and P3 in the matrix row corresponding to procurement task P1 are 1, then the risk impact task set is represented as P2 and P3, indicating that when an anomaly occurs in P1, the execution status of P2 and P3 needs to be monitored simultaneously.

[0088] In this embodiment, the review agent is an intelligent processing unit used to perform risk identification and status analysis for a specific procurement task scope. Each review agent corresponds to a risk-generating task and its associated task scope. In the specific configuration process, review agents are created according to the risk propagation starting point in the second procurement task risk propagation scope matrix, so that each risk source task has independent risk analysis capabilities and receives status information of associated tasks according to the propagation relationship, realizing information interaction between risks of different procurement tasks.

[0089] In this embodiment, the set of tasks to be reviewed is a set of procurement tasks assigned to a single review agent for analysis and processing. Specifically, during the acquisition process, the procurement task that generates the risk is added to the set as the initial task. Then, based on the risk propagation range matrix of the corresponding second procurement task, other procurement tasks that the task can affect are found. Procurement tasks with a propagation relationship of 1 in the matrix are added to the task range of the same review agent, so that the agent can simultaneously analyze the correlation changes between the risk source task and the affected tasks.

[0090] In this embodiment, the initial review object is the procurement task that the review agent first executes risk analysis on, i.e., the procurement task node where the risk arises. Specifically, the starting point of risk propagation is determined based on the row direction of the risk propagation range matrix of the second procurement task, and the corresponding row node is used as the initial review object. For example, if the delay in the main transformer procurement task generates risk, then the main transformer procurement task is the first object analyzed by the corresponding review agent to determine whether the task itself has an abnormal state.

[0091] In this embodiment, the associated review objects are other procurement tasks that need to be analyzed synchronously, besides the risk-generating task, as determined by the propagation relationship in the risk propagation range matrix of the second procurement task. Specifically, during the acquisition process, the associated task is determined based on the position of the element that is 1 in the matrix row where the initial review object is located. For example, if the position of the protection cabinet and the commissioning system corresponding to the main transformer procurement task is 1, then the protection cabinet and the commissioning system are used as associated review objects to analyze the impact status after risk propagation.

[0092] In this embodiment, the propagation direction between procurement tasks is the directional relationship from the risk-generating task to the affected task in the risk propagation range matrix of the second procurement task, used to determine the information interaction path between different review agents. Specifically, the position of an element equal to 1 in the matrix is ​​used as the propagation direction, with the task corresponding to the matrix row acting as the information sender and the task corresponding to the matrix column acting as the information receiver. For example, if the element corresponding to the main transformer procurement task is 1, then the information transmission direction from the main transformer review agent to the protection cabinet is established.

[0093] In this embodiment, the information transmission relationship refers to the connection relationship between different review agents used to exchange risk status information of procurement tasks. Specifically, during the establishment process, the agent connection relationship is established according to the propagation direction in the second procurement task risk propagation range matrix. When there is a risk propagation relationship between two procurement tasks, an information interaction channel is established between the two corresponding review agents, so that the procurement task status that generates risk can be transmitted to the review agent corresponding to the associated task.

[0094] In this embodiment, the procurement task status information is a data set describing the current execution status of the procurement task, including the planned time, actual execution time, supplier delivery status, equipment production progress, payment status, and project milestone completion status. Specifically, during the acquisition process, based on the set of tasks to be reviewed corresponding to each review agent, real-time status data of the corresponding procurement task is read from the procurement management system, project management system, and supply chain execution system, serving as input information for the agent to make risk assessments.

[0095] In this embodiment, the risk assessment result is the risk identification output information obtained by the review agent based on the input procurement task status information. It is used to indicate whether there are any abnormal situations affecting the execution of the project plan in the current procurement task. Specifically, during the generation process, the review agent makes judgments based on the differences between the current status of the procurement task and the planned requirements. For example, it determines whether the actual delivery time exceeds the installation requirement time, whether the supplier's production progress meets the delivery requirements, and whether the funding arrangement meets the payment node requirements, and outputs the corresponding risk assessment result for the procurement task.

[0096] In this embodiment, the feedback information is a risk assessment result generated by one review agent and transmitted to other related review agents, which is used to enable the agents corresponding to the related tasks to synchronously obtain changes in the upstream task status. Specifically, during the transmission process, when a review agent identifies a risk event, it sends the risk assessment result to the related review agents with a propagation relationship according to the propagation direction in the risk propagation range of the second procurement task, so that the related agents can re-analyze the risk impact in conjunction with their own task status.

[0097] In this embodiment, the input information range is the range of procurement task status information that a single review agent can receive during the risk analysis process. Specifically, the input range is determined by considering the propagation relationships within the set of tasks to be reviewed corresponding to the review agent and the risk propagation range matrix of the second procurement task. This includes both the initial review object's own status information and the status information of associated review objects. For example, an agent responsible for the risk analysis of main transformer procurement would input not only the main transformer delivery status but also the protection cabinet installation preparation status and the commissioning equipment procurement status.

[0098] In this embodiment, the output feedback range is the task range within which a single review agent can send risk assessment results to other agents. Specifically, the range is determined based on the column direction propagation relationship corresponding to the current task in the second procurement task risk propagation range matrix. When a procurement task is affected by the current task, the risk assessment result generated by the current agent is sent to the corresponding associated agent.

[0099] In this embodiment, the intelligent review and control object for the procurement plan is the multi-agent collaborative review configuration data formed by converting the risk propagation scope of the second procurement task. This data is used to control how subsequent agents participate in the procurement plan review. Specifically, during the generation process, the risk impact task set corresponding to the risk propagation scope of each second procurement task is converted into an agent task allocation relationship; the procurement task propagation direction is converted into an information interaction relationship between agents; the procurement task status information is converted into an agent input range; and the risk judgment result is converted into an agent feedback range. This ultimately forms a complete control object containing the number of agents, the scope of review tasks, the direction of information interaction, and the feedback relationship.

[0100] It should be noted that the transformation relationship in this step is not a simple conversion of the procurement task risk propagation scope matrix into the number of agents, but rather the generation of task collaboration relationships between agents based on the risk propagation structure in the matrix. Because procurement tasks have varying degrees of engineering dependencies, the agent organization methods corresponding to different second procurement task risk propagation scopes differ. When the propagation scope expands, the number of tasks participating in the review increases, and the scope of information interaction between agents expands; when the propagation scope shrinks, agents only focus on the core risk links, thus forming different intelligent review and control objects for the procurement plan.

[0101] It should be noted that in actual engineering procurement processes, a procurement task may be affected by multiple risk sources simultaneously. Therefore, the information transmission relationship established in this step based on the risk propagation scope of the second procurement task can avoid the problem of isolated risk information in traditional single-task review methods. For example, a delay in the delivery of a piece of equipment not only affects its own delivery plan but may also affect installation, commissioning, and funding arrangements. Through information interaction between related intelligent agents, multiple dimensions of risk status can be jointly involved in the procurement plan judgment.

[0102] In a specific embodiment, taking the procurement process of power substation equipment as an example, the procurement tasks include the procurement of main transformers, protection cabinets, and secondary commissioning systems.

[0103] Based on the risk propagation scope matrix of a certain second procurement task, we obtain: The corresponding protection cabinet procurement task for the main transformer procurement task is located at position 1. The procurement task for the protection cabinet corresponds to the procurement task for the secondary debugging system at position 1.

[0104] Based on the row propagation relationship in the matrix, establish a set of risk impact tasks corresponding to the three procurement tasks: The risk impact set of the main transformer procurement task includes the protection cabinet procurement task; The risk impact set of the protection cabinet procurement task includes the secondary debugging system procurement task.

[0105] Then, the review agents are configured respectively: The main transformer review agent is responsible for analyzing the main transformer supply status and sending the risk assessment results to the protection cabinet review agent. The protection cabinet reviews the intelligent agent receiving risk information from the main transformer and, in conjunction with its own installation status, determines whether it will affect subsequent commissioning. The secondary debugging system reviews the risk feedback from the intelligent agent receiving the protection cabinet and determines whether the debugging plan needs to be adjusted.

[0106] Ultimately, this forms the intelligent review and control object for the procurement plan corresponding to the risk propagation scope of the second procurement task.

[0107] It should be noted that this step establishes a conversion relationship between the scope of risk propagation in procurement tasks and the control quantity of multi-agent review, thereby realizing the automatic generation of the risk propagation structure of procurement tasks into the intelligent review organizational structure, so that different risk propagation scopes can correspond to different agent collaboration methods.

[0108] It should be noted that traditional intelligent review of procurement plans typically has a fixed scope of review tasks, making it impossible to dynamically adjust the tasks involved in the analysis based on different procurement risk propagation scenarios. This can easily lead to problems such as an overly narrow review scope resulting in missed risks, or an overly broad review scope resulting in wasted computational resources. This step utilizes the task relationships, information propagation directions, and feedback relationships within the risk propagation scope of the second procurement task to generate corresponding control objects. This enables subsequent multi-agent reviews to dynamically adjust the collaborative scope based on the risk propagation structure, providing a unified control basis for comparing the review effects under different propagation scopes.

[0109] S4 executes multi-agent collaborative procurement plan review based on the procurement plan intelligent review control object corresponding to the risk propagation scope of each second procurement task.

[0110] In this embodiment, the step of performing multi-agent collaborative procurement plan review based on the procurement plan intelligent review control object corresponding to the risk propagation scope of each second procurement task specifically includes: Obtain the intelligent review and control objects of the procurement plan corresponding to the risk propagation scope of each second procurement task, and configure the corresponding procurement review intelligent agents according to the initial review tasks, related review tasks and information transmission relationships in each control object; Each procurement review agent obtains plan execution data, project node data, supply execution data, and fund execution data from the procurement management system, engineering management system, and supply chain execution system, respectively, to form its own review input data. like Figure 2 As shown, based on the review input data corresponding to each procurement review agent, different types of constraint events during the execution of procurement tasks are analyzed, including: Based on the time differences between the planned delivery time of the procurement task, the equipment installation node time, and the engineering commissioning time, the corresponding time deviation event of the procurement task is determined; like Figure 4 As shown, based on the matching relationship between the supplier's historical delivery records, the current production task occupancy, and the target procurement task delivery plan, the supply performance deviation event corresponding to the procurement task is determined; Based on the matching relationship between the procurement payment plan, the project funding availability plan, and the payment nodes of the procurement tasks, identify the funding conflict events corresponding to the procurement tasks; Based on the propagation relationship of procurement tasks within the risk propagation scope of the second procurement task and the status changes of related review tasks, determine the impact events of related tasks corresponding to the procurement task. The time deviation events, supply performance deviation events, funding conflict events, and related task impact events generated by each procurement review agent are interacted according to the information transmission relationship corresponding to the risk propagation scope of the second procurement task, so that each procurement review agent can obtain the constraint event information corresponding to the related procurement task. Based on the constraint event information after interaction, update the risk assessment results corresponding to each procurement task to form intelligent review process data for the procurement plan.

[0111] In this embodiment, each procurement review agent obtains plan execution data, project node data, supply execution data, and fund execution data from the procurement management system, project management system, and supply chain execution system, respectively, to form its own review input data.

[0112] Based on supply execution data, agent A1 identifies a delay in the production of the main transformer and generates a supply performance deviation event. Agent A1 transmits this event to A2 along the P1→P2 direction. A2, combined with its own engineering node data, identifies that the installation window of P2 has shortened and generates a related task impact event. A2 continues to transmit data along the P2→P3 direction to A3. A3 identifies the P3 debugging startup delay and outputs the risk assessment result.

[0113] The final result is multi-agent collaborative procurement plan review process data, which includes event interaction sequences of each agent, risk assessment result sequences, and task execution sequences.

[0114] In this embodiment, the intelligent review control object for the procurement plan corresponding to the risk propagation range of the second procurement task is a data structure obtained by transforming the risk propagation range matrix of the second procurement task to control the collaborative review process of multiple agents. Specifically, after obtaining the risk propagation range matrix of the second procurement task, the procurement task node corresponding to each row in the matrix is ​​read, and the procurement tasks with all matrix column elements of 1 corresponding to the procurement task node are determined as the risk propagation associated tasks of the procurement task. For example, if the positions of the protection cabinet procurement task and the debugging system procurement task in the row where the main transformer procurement task is located in the risk propagation range matrix of the second procurement task are both 1, then the main transformer is taken as the initial risk task, and the protection cabinet and the debugging system are taken as associated review tasks, further forming the agent review range, information interaction direction, and task feedback range corresponding to the propagation range. This control object is not data directly used for procurement review, but rather a data structure used to limit the review boundaries between multiple agents.

[0115] In this embodiment, the initial review task is the procurement task node in the risk propagation range matrix of the second procurement task, which serves as the starting point for risk propagation. Specifically, this is achieved by reading the row direction relationships in the procurement task dependency adjacency matrix, and using each procurement task with risky input as an independent review starting point. For example, if a delay in equipment procurement first occurs in the main transformer procurement task, then the main transformer procurement task becomes the initial review task. The corresponding agent first analyzes the execution status of this procurement task itself, including planned time, supply status, and funding status, and uses this as the starting point for subsequent risk propagation judgments.

[0116] In this embodiment, the associated review tasks are procurement task nodes in the risk propagation range matrix of the second procurement task that have a propagation relationship with the initial review task. Specifically, this is achieved by reading the element values ​​in the matrix row containing the initial review task, converting the column positions corresponding to elements with a value of 1 into procurement task numbers, and forming a set of associated tasks corresponding to that initial review task. For example, if the matrix row corresponding to the main transformer is P1, and positions P2 and P3 in the row containing P1 are 1, then the protection cabinet and the debugging system are considered associated review tasks for the main transformer. Associated review tasks are used to represent procurement objects that may be further affected by the risk, enabling the agent to analyze not only the current procurement task but also the impact of changes in this task on subsequent equipment procurement.

[0117] In this embodiment, the information transmission relationship is the data interaction direction between agents established based on the directed propagation direction between procurement tasks in the second procurement task risk propagation range matrix. Specifically, the acquisition method involves retaining the positional relationship corresponding to an element of 1 in the matrix. For example, if row P1 and column P2 is 1, then a relationship is established whereby risk information is transmitted from the review agent corresponding to P1 to the review agent corresponding to P2. This information transmission relationship is used to limit the information flow between different agents, ensuring that risk judgment results propagate along the actual dependencies of the procurement tasks, rather than through unconstrained communication between all agents.

[0118] In this embodiment, the procurement review agent is a data processing unit dynamically configured according to the intelligent review control object of the procurement plan to perform reviews of different procurement tasks. Specifically, the configuration involves creating a corresponding agent based on each risk propagation starting point in the risk propagation range matrix of the second procurement task, and binding this agent to the corresponding initial review task and associated review tasks. For example, a procurement review agent is created for the main transformer procurement task. This agent is responsible for acquiring the main transformer procurement data and receiving feedback information from associated tasks such as protection cabinets and commissioning systems. Different risk propagation ranges of the second procurement task correspond to different task association matrices; therefore, the number of procurement review agents generated and the information connection relationships between agents also differ.

[0119] In this embodiment, the planned execution data is a set of data describing the current planned execution status of the procurement task, including the planned start time, planned completion time, planned delivery time, current execution stage, and actual completion status. Specifically, it is obtained by reading procurement plan records from the procurement management system and associating them according to the procurement task number. For example, for a main transformer procurement task, the planned execution data includes the contract signing time, production cycle, estimated delivery date, and current production progress, used to determine whether the procurement task is progressing according to the original plan.

[0120] In this embodiment, the project node data is a set of data describing the installation, commissioning, and operation milestones of the procured equipment during the project construction process. Specifically, it is obtained by reading the project schedule and associating the equipment corresponding to the procurement task with the project nodes. For example, the main transformer needs to be installed after the substation foundation construction is completed and put into operation before the secondary protection system is commissioned. Therefore, the project node data includes the foundation completion date, equipment installation date, and commissioning start date, used to determine whether delays in equipment procurement affect subsequent project nodes.

[0121] In this embodiment, supply execution data is a set of data describing the actual performance status of the supplier during the execution of the procurement contract. Specifically, it is obtained by reading the supplier's historical delivery records, current production task arrangements, and the production progress of the target equipment. For example, if the supplier has experienced multiple delays in the past 20 equipment deliveries, and multiple pieces of the same type of equipment are currently being produced on the production line, then supply execution data is generated and used to determine whether there is a delivery risk in the current procurement task.

[0122] In this embodiment, the funding execution data is a set of data describing the funding arrangements and actual payment status corresponding to the procurement task. Specifically, it is obtained by reading the project's annual funding plan, procurement contract payment milestones, and current fund occupancy. For example, if an equipment procurement contract stipulates a 30% prepayment of the contract amount, but the project funds are insufficient in the current month, funding execution data is generated to determine whether the funding arrangements affect the execution of the procurement task.

[0123] In this embodiment, the review input data is a set of data received by each procurement review agent when performing risk assessment, consisting of plan execution data, engineering node data, supply execution data, and funding execution data. Specifically, the process involves extracting corresponding data from the procurement management system, engineering management system, and supply management system based on the initial review task and associated review tasks bound to the procurement review agent, and combining them according to the procurement task number, so that each agent obtains data input corresponding to its own review scope.

[0124] In this embodiment, a time deviation event is a time constraint anomaly that occurs when the planned delivery time of a procurement task cannot meet the requirements of the engineering installation or commissioning milestones. Specifically, it is determined by comparing the planned delivery time of the procurement task with the equipment installation milestone time. When the expected delivery time of the equipment is later than the required installation time, the delay status corresponding to this time difference is recorded as a time deviation event. For example, if the main transformer is required to arrive before August 10th, but the planned delivery time is August 20th, then a 10-day time deviation event is formed.

[0125] In this embodiment, a supply fulfillment deviation event is an anomaly in supply information resulting from a discrepancy between the supplier's actual production and delivery capabilities and the delivery requirements of the procurement task. Specifically, it is determined by correlating the supplier's historical delivery completion status, current production task occupancy, and the target procurement task delivery date. A supply fulfillment deviation event is generated when the supplier's expected completion date exceeds the required date of the procurement plan. For example, if the supplier expects to complete equipment production by August 25th, but the project requires delivery by August 10th, a supply fulfillment deviation event is generated.

[0126] In this embodiment, a funding conflict event is an anomaly in funding constraints caused by a mismatch between procurement payment requirements and the project's available funding plan. Specifically, it is determined by reading the payment node corresponding to the procurement task, comparing the required payment amount with the project's available funds within the corresponding time period, and generating a funding conflict event when the required payment exceeds the available funds. For example, if equipment procurement requires a payment of 80 million yuan, but the project's available funds for the current month are only 50 million yuan, a funding conflict event occurs.

[0127] In this embodiment, the associated task impact event is a risk event formed when a change in the status of one procurement task affects the execution status of other associated procurement tasks due to the propagation relationship between procurement tasks. Specifically, it is determined by reading the propagation relationships in the second procurement task risk propagation range matrix. When a procurement task experiences a time deviation, supply anomaly, or funding conflict, associated procurement tasks are searched along the direction where elements are 1 in the matrix, and associated task impact events are formed based on the changes in the associated task plans. For example, if a delay in the main transformer causes a change in the installation window of the protection cabinet, then an associated task impact event of the main transformer on the protection cabinet is formed.

[0128] In this embodiment, the constraint event information is a set of information generated by each procurement review agent describing the abnormal state of the procurement task, including the event type, the procurement task it affects, the abnormal state, and the direction of the impact. Specifically, the acquisition method involves sending time deviation events, supply performance deviation events, funding conflict events, and related task impact events to the associated agents according to the information transmission relationship, so that each agent not only obtains its own task's abnormal information, but also obtains the impact information of upstream tasks on its own task.

[0129] In this embodiment, the risk assessment result is a judgment output generated by the procurement review agent based on the received constraint event information, analyzing whether there is execution risk in the procurement task and the scope of its impact. Specifically, the agent combines its own procurement task status data and related task propagation information to determine whether the current procurement task is affected by delays, insufficient supply, insufficient funds, or changes in related tasks, and generates a corresponding risk status. For example, if the agent determines that a delay in the main transformer will prevent the protection cabinet from being installed as planned, it will output a judgment result indicating that there is a risk of installation delay for the protection cabinet.

[0130] In this embodiment, the intelligent review process data for the procurement plan is a complete record of the data generated during the multi-agent collaborative review process, including the data received by each agent, the identified constraint events, the information interaction process between agents, and the final risk assessment result. Specifically, the data input, event analysis, and result feedback process of each agent are recorded according to the information transmission order corresponding to the risk propagation scope of the second procurement task, forming the data foundation for subsequent collaborative evaluation value training.

[0131] It should be noted that in the aforementioned multi-agent collaborative procurement plan review process, different agent collaborative structures correspond to different risk propagation ranges for different second procurement tasks. The same procurement risk will generate different numbers of associated review tasks under different propagation ranges. For example, when the propagation range only includes the main transformer and protection cabinet, only two associated review objects are configured; when the propagation range is further extended to the commissioning system, information interaction for commissioning tasks is added. Therefore, by changing the risk propagation range of the second procurement task, the agent review boundaries can be dynamically adjusted, enabling subsequent comparisons of risk identification effectiveness under different propagation ranges.

[0132] It should be noted that the information exchange between the various procurement review agents does not involve sharing all procurement task data. Instead, it strictly follows the propagation relationships in the second procurement task risk propagation range matrix. 0 in the matrix corresponds to no information exchange between tasks, while 1 corresponds to the transmission of risk status between tasks. This avoids interference with review results due to the exchange of a large amount of irrelevant procurement task information.

[0133] In a specific embodiment, taking the procurement process of large-scale power engineering equipment as an example, the project plans to procure a 220kV main transformer, protection cabinet, and relay protection commissioning system. A procurement task association relationship is established according to the project's installation sequence. The main transformer procurement task affects the protection cabinet installation, and the protection cabinet procurement task affects the relay protection commissioning, thus forming a propagation relationship from the main transformer to the protection cabinet and from the protection cabinet to the commissioning system. When generating the risk propagation range for a certain second procurement task, if the matrix only includes the relationship from the main transformer to the protection cabinet, then a main transformer review agent and a protection cabinet review agent are configured, and the main transformer delay information is transmitted to the protection cabinet agent. If the matrix further includes the commissioning system, then a commissioning agent is added, enabling the commissioning agent to obtain equipment delay information in advance and adjust the commissioning plan risk assessment.

[0134] It should be noted that this step transforms the risk propagation scope of the second procurement task into a multi-agent collaborative review structure. This allows the procurement risk propagation relationship to directly affect the agent configuration and information interaction scope, avoiding information redundancy caused by the fixed participation of all agents in traditional procurement review methods, as well as risk omissions due to insufficient risk propagation scope. By dynamically generating review agents, input data ranges, and feedback paths based on different propagation scopes, the rationality of the propagation scope can be evaluated using review process data generated under different propagation scopes, providing a data foundation for selecting the optimal procurement task risk propagation scope.

[0135] The significance of this step lies in transforming the previously obtained second procurement task risk propagation range matrix from an abstract procurement task association into a practically executable multi-agent collaborative review and control structure. This enables the procurement task propagation relationships in the matrix to directly control the task allocation and information interaction range between agents, thereby achieving intelligent review of multi-scheme procurement plans based on different risk propagation ranges and providing practical operational data for subsequent evaluation of the advantages and disadvantages of different propagation ranges.

[0136] S5 extracts features from the multi-agent review process and trains to generate collaborative evaluation values ​​corresponding to the risk propagation scope of the second procurement task.

[0137] In this embodiment, the feature extraction of each multi-agent review process and the training to generate a collaborative evaluation value corresponding to the risk propagation scope of the second procurement task specifically involves: Acquire the intelligent review process data of the procurement plan corresponding to the risk propagation scope of each second procurement task, and extract the constraint event interaction sequence, risk judgment result sequence and intelligent agent task execution sequence corresponding to each procurement review intelligent agent; Based on the risk assessment result sequence, extract the actual abnormal states of each procurement task and the risk assessment states output by each procurement review agent, count the number of procurement tasks that actually had abnormalities and were identified by the risk assessment results, and the number of procurement tasks that actually had abnormalities but were not identified by the risk assessment results, and generate the risk identification coverage rate. Based on the set of propagation tasks corresponding to the events affected by related tasks, obtain the set of related review tasks corresponding to each risk procurement task in the risk propagation scope matrix of the second procurement task, and extract the set of actual impact tasks that generate related impact events after the actual risk occurs. A set difference analysis was performed on the set of related review tasks and the set of actual impact tasks. The number of procurement tasks in the set of actual impact tasks that were not included in the set of related review tasks was counted, and the related task omission rate was generated to characterize the situation where the risk propagation scope of the second procurement task was insufficient, resulting in the risk propagation chain not being fully unfolded. Based on the agent task execution sequence, the number of procurement tasks participating in the procurement plan review under the risk propagation scope of each second procurement task is counted, and the number of review tasks that did not generate constraint events, did not trigger risk judgment result updates, and did not affect the status changes of subsequent procurement tasks is extracted to generate the invalid review task ratio, which is used to characterize the situation where the risk propagation scope of the second procurement task is too large, resulting in invalid tasks entering the review scope. Based on the agent interaction sequence, the number of information transmissions between different procurement review agents, the number of repeated transmissions of the same constraint event, and the number of risk judgment result updates are counted to generate agent interaction load, which is used to characterize the increase in agent interaction and procurement plan adjustment oscillation caused by the expansion of the risk propagation scope of the second procurement task. The risk identification coverage rate, the omission rate of related tasks, the proportion of invalid review tasks, and the intelligent agent interaction load are combined in a fixed order to form a review process feature vector corresponding to the risk propagation scope of each second procurement task. Using the risk propagation range of the second procurement task as the state control object in the feature weight regression environment based on the results of multiple rounds of operation, the feature vector of the review process is used as the input of the environment state, and the feedback sources are the improved risk identification coverage, the reduced omission rate of related tasks, the reduced proportion of invalid review tasks, and the reduced interaction load of the intelligent agent. Thus, a training environment for coordinating and evaluating the risk propagation range of the procurement task is constructed. The weights of each feature are trained based on the multi-agent collaborative procurement plan review process corresponding to the risk propagation scope of different second procurement tasks in multiple rounds. The weights of each feature are then weighted based on the trained weights to generate a collaborative evaluation value.

[0138] In this embodiment, the data from the five review processes corresponding to R1-R5 are used as input to calculate four indicators: risk identification coverage, related task omission rate, invalid review task ratio, and agent interaction load. These indicators are then weighted and summed according to preset weights to obtain five collaborative evaluation values ​​S1-S5.

[0139] In this embodiment, the intelligent review process data for the procurement plan comprises all process records generated by multiple agents during the review of the procurement plan under the influence of the risk propagation range of a certain second procurement task. Specifically, this data is obtained by reading the task allocation records, input data, constraint event identification results, information transmission records between agents, and risk assessment results from each procurement review agent in S4, and archiving them according to the corresponding risk propagation range of the second procurement task. For example, for the same main transformer delay risk, one set of review process data is formed when the propagation range only includes the protection cabinet; another set is formed when the propagation range extends to the commissioning system. The rationality of the propagation range is evaluated through comparative analysis of data from different propagation ranges.

[0140] In this embodiment, the constraint event interaction sequence is a data sequence recording the constraint event information transmitted between various procurement review agents in chronological order. Specifically, it is obtained by reading the information transmission records between agents, sorting them according to the event occurrence time, and treating each constraint event sent by one agent and received by another as an interaction record. For example, after the main transformer agent generates a delay event, it sends the event to the protection cabinet agent, thus forming one constraint event interaction record; if the protection cabinet agent further sends the installation delay impact to the commissioning system agent, then the next interaction record is formed.

[0141] In this embodiment, the risk assessment result sequence is a data sequence that records the risk assessment status output by each procurement review agent according to the order of procurement tasks and the order of review time. Specifically, the acquisition method is to read the risk assessment results generated by each agent in S4 and record the corresponding procurement task, risk type, and assessment status. For example, if an agent judges that the procurement task of the main transformer has a delay risk, it records the corresponding risk assessment status of the main transformer; subsequently, after analysis of related tasks, it judges that the protection cabinet has an installation risk, so it continues to record the corresponding risk assessment status of the protection cabinet.

[0142] In this embodiment, the agent task execution sequence is a data sequence recording the changes in the actual review tasks participated in by each procurement review agent under different risk propagation ranges of the second procurement task. Specifically, the acquisition method involves reading the agent configuration results corresponding to each risk propagation range of the second procurement task, recording the procurement tasks reviewed by each agent, the execution order, and whether a valid review result was generated. For example, when a propagation range includes three tasks—main transformer, protection cabinet, and debugging system—the execution process of the agents corresponding to the three tasks is recorded; when a propagation range only includes the main transformer and protection cabinet, the execution process of the agents corresponding to the two tasks is recorded.

[0143] In this embodiment, the risk identification coverage rate is a feature used to describe the number of actual risk tasks that the risk propagation scope of the second procurement task can cover. Specifically, it is obtained by reading procurement tasks that experienced anomalies during the actual execution of the project and analyzing their correspondence with the risk judgment results output by the intelligent agent. When a procurement task that actually experienced anomalies is simultaneously identified by the risk judgment results, the procurement task is recorded as a covered risk task; when anomalies actually occurred but there is no corresponding risk judgment result, the procurement task is recorded as an omitted risk task. The risk identification coverage rate is ultimately formed based on the relationship between the number of the two types of tasks. For example, if there are 5 procurement tasks that actually experienced delays in the project, and 4 of them are identified in advance by the intelligent agent, a higher risk identification coverage rate is formed; if only 1 is identified, a lower risk identification coverage rate is formed.

[0144] In this embodiment, procurement tasks that actually experience abnormalities are those where there are genuine deviations from the plan, supply issues, funding problems, or related impacts during the engineering procurement process. Specifically, this is achieved by reading the actual completion time, supply delivery records, payment execution records, and engineering milestone change records from the procurement execution system and comparing them with the original procurement plan. When the actual execution status deviates from the planned requirements, the corresponding procurement task is marked as an actual abnormal procurement task.

[0145] In this embodiment, the procurement tasks identified by the risk assessment results are those that have been determined to be at risk after analysis by the procurement review intelligence agent. Specifically, this is achieved by reading the risk assessment results output by each procurement review intelligence agent and summarizing the procurement tasks judged to be at risk. For example, if the intelligence agent outputs risks related to the delay of the main transformer installation and the delay of the protection cabinet installation, then both the main transformer and the protection cabinet fall under the risk assessment result identification tasks.

[0146] In this embodiment, the set of propagation tasks corresponding to the associated task impact events is the set of procurement tasks that actually trigger the associated impact events after an anomaly occurs in a certain risky procurement task. Specifically, this is obtained by reading the associated task impact event records in S4 and extracting all associated procurement tasks after an abnormal procurement task has an impact. For example, if the delay of the main transformer actually leads to the delay of the protection cabinet installation and the delay of the commissioning system startup, then the protection cabinet and the commissioning system together constitute this set of propagation tasks.

[0147] In this embodiment, the set of associated review tasks is the set of procurement tasks obtained by converting the positions of elements with a value of 1 in the corresponding rows of risky procurement tasks in the risk propagation range matrix of the second procurement task. Specifically, it is obtained by reading the row of abnormal procurement tasks in the risk propagation range matrix of the second procurement task and converting the column positions with a value of 1 in the corresponding matrix elements into procurement task numbers. For example, if the position of the protection cabinet and the debugging system in the matrix row of the main transformer is 1, then the protection cabinet and the debugging system constitute the set of associated review tasks corresponding to this risky task.

[0148] In this embodiment, the set of truly impactful tasks refers to the set of procurement tasks affected by an actual anomaly, determined through engineering execution data. Specifically, this is achieved by reading records of changes in engineering nodes, equipment installation, and commissioning plans after the anomaly occurs, to determine which procurement tasks actually experience status changes. For example, if the installation time of the protection cabinet is forced to be adjusted due to the main transformer delay, then the protection cabinet belongs to the set of truly impactful tasks.

[0149] In this embodiment, the associated task omission rate is a characteristic used to describe the fact that the risk propagation scope of the second procurement task does not cover the actual risk propagation chain. Specifically, it is obtained by comparing the procurement tasks in the associated review task set with those in the actual impact task set. When a procurement task exists in the actual impact task set but does not appear in the associated review task set, that procurement task is recorded as an omission task, and the number of omission tasks is counted to form the associated task omission rate. For example, in a real situation, the delay of the main transformer affects the protection cabinet and the commissioning system, but the risk propagation scope of the second procurement task only includes the protection cabinet; therefore, the commissioning system is an omission task.

[0150] In this embodiment, the invalid review task ratio describes the degree of invalid expansion caused by the excessively large risk propagation scope of a second procurement task, which includes procurement tasks with no actual risk relationship within the review scope. Specifically, this is achieved by reading all procurement tasks participating in the review within a certain propagation scope and checking whether these tasks generate constraint events, trigger risk assessment result updates, and affect the status changes of other procurement tasks. If a procurement task participates in the review but does not generate any risk association throughout the process, it is recorded as an invalid expansion task. For example, if a general auxiliary equipment procurement task is included in the risk propagation scope of a main transformer, but its execution status is not affected, then this task is an invalid review expansion task.

[0151] In this embodiment, agent interaction load is a feature used to describe the pressure of information interaction among multiple agents and the degree of collaborative complexity resulting from the expansion of the propagation range. Specifically, it is obtained by reading the constraint event interaction sequence and counting the number of information transmissions between different agents, the number of times the same event is repeatedly transmitted, and the number of times risk judgment results are updated. When the propagation range expands, a large number of agents exchange repeated risk information, thus increasing the corresponding agent interaction load.

[0152] In this embodiment, the review process feature vector is a set of data descriptions formed by combining risk identification coverage, associated task omission rate, invalid review task ratio, and agent interaction load in a fixed order. Specifically, it is formed by first recording the four types of feature values ​​corresponding to the risk propagation range of a certain second procurement task, and then combining them in the order of risk identification coverage, risk omission, invalid extension, and interaction load, so that each propagation range corresponds to a unique set of review process feature descriptions.

[0153] In this embodiment, the state control object in the feature weight regression environment based on multi-round running results is the risk propagation scope of the second procurement task itself. Specifically, different risk propagation scope matrices for the second procurement task are used as different state conditions, enabling the training process to compare the review effects generated under different propagation scope configurations. For example, a propagation matrix containing only directly related tasks can be used as one state, and a propagation matrix containing multi-level related tasks can be used as another state. The quality of the propagation scope is evaluated by comparing the review results under different states.

[0154] In this embodiment, the feedback source is evaluation information used to assess the application effect of the risk propagation scope of the second procurement task. Specifically, the feedback is generated based on changes in risk identification coverage, associated task omission rate, invalid review task ratio, and agent interaction load. When risk identification coverage improves, omission tasks decrease, invalid extensions decrease, and interaction load decreases, it indicates that the propagation scope has a good review effect; conversely, it reduces the evaluation effect.

[0155] In this embodiment, the training environment for coordinating and evaluating the risk propagation scope of procurement tasks is a data interaction process used to train the evaluation capability of different risk propagation scopes for second procurement tasks. Specifically, the construction method involves using the risk propagation scope of the second procurement task as the control object, the corresponding review process feature vector as the state description, and the comprehensive evaluation after running different propagation scopes as feedback. Through multiple rounds of application results for different propagation scopes, the correspondence between the propagation scope and the review effect is learned.

[0156] In this embodiment, the collaborative evaluation value is a comprehensive evaluation value used to assess whether the risk propagation scope of a certain second procurement task is suitable for intelligent review of procurement plans. Specifically, it is generated by comprehensively evaluating the risk identification coverage, associated task omission rate, proportion of invalid review tasks, and agent interaction load based on the feature contribution relationships learned during training. A higher collaborative evaluation value is generated when a certain propagation scope can completely cover the risk propagation chain, while avoiding too many invalid tasks entering the review scope and maintaining a reasonable scale of agent interaction.

[0157] In this embodiment, the weights trained are the different feature influence levels learned during the feature weight regression process based on the review results of a large number of second procurement task risk propagation scope reviews, based on the results of multiple rounds of operation. Specifically, during the multi-round propagation scope test, the influence level of each feature in the comprehensive evaluation is adjusted according to the final evaluation result corresponding to each propagation scope, so that the evaluation result can reflect the actual procurement review effect, rather than manually pre-determining the importance of each factor.

[0158] It should be noted that in the above process, the collaborative evaluation value is not simply determined based on the number of risks identified, but rather considers both issues of excessively small and excessively large propagation scopes. When the propagation scope is too small, although the number of agents is small and the interaction is simple, it is easy to miss subsequent affected procurement tasks. When the propagation scope is too large, more tasks can be covered, but this will lead to a large number of irrelevant procurement tasks entering the review process, increasing agent interactions and the number of plan adjustments. Therefore, by simultaneously introducing four features—risk coverage, omission, ineffective expansion, and interaction load—the collaborative evaluation value can reflect the actual coordination capability of the risk propagation scope of procurement tasks.

[0159] It should be noted that the features in this step are all derived from data generated during the actual execution of S4, and no additional human input of evaluation parameters is required. Among them, the risk identification coverage rate describes the problem of missed detection caused by insufficient propagation scope, the related task omission rate describes the problem of incomplete risk chain development, the invalid review task ratio describes the problem of excessive propagation scope, and the agent interaction load describes the problem of collaborative complexity caused by the increase in the number of agents. The four types of features together form a complete evaluation of the risk propagation scope of the second procurement task.

[0160] In a specific embodiment, taking the procurement process for the construction of a power grid substation as an example, the procurement tasks include the procurement of main transformers, protection cabinets, and relay protection commissioning systems. When the risk propagation scope of the second procurement task only includes the relationship between the main transformer and the protection cabinet, the agent can identify that the delay of the main transformer causes the delay of the protection cabinet, but cannot identify that the commissioning system is affected. At this time, the risk identification coverage is reduced, and the omission rate of related tasks is increased.

[0161] When the risk propagation scope of the second procurement task expands to include the main transformer, protection cabinet, and commissioning system, the agent can completely identify the risk propagation chain, reducing the omission rate of related tasks. However, if it continues to expand to the procurement tasks of auxiliary equipment that are not directly related to the main transformer, the agent for auxiliary equipment will not generate effective risk assessments, while increasing the number of information interactions, thus increasing the proportion of invalid review tasks and the agent's interaction load. Based on the above results of different propagation scopes, collaborative evaluation values ​​corresponding to different propagation scopes can be trained.

[0162] It should be noted that the beneficial effect of this step lies in the fact that, by extracting features from the multi-agent review process corresponding to different risk propagation ranges of the second procurement task, and generating a collaborative evaluation value using a feature weight regression method based on multi-round operation results, the actual application effect of the risk propagation range of the procurement task is automatically evaluated. Compared with the method of setting a fixed risk propagation range, this step can simultaneously identify risk omissions caused by insufficient propagation range and waste of review resources caused by excessive propagation range, enabling subsequent selection of a risk propagation range more suitable for the current project procurement plan based on the collaborative evaluation value.

[0163] The significance of this step lies in transforming the risk propagation scope of different second procurement tasks from a simple task-related structure into an evaluable intelligent review scheme. By using data generated from the actual multi-agent collaborative review process, it quantitatively judges whether a certain propagation scope can achieve a balance between risk coverage capability and review complexity, providing a basis for ultimately selecting the optimal risk propagation scope for procurement tasks.

[0164] S6. Based on the propagation structure metric, the risk propagation range of the second procurement task is sorted, and the optimal risk propagation range of the procurement task is selected by fitting and analyzing the corresponding collaborative evaluation value and conducting intelligent review of the procurement plan.

[0165] In this embodiment, the process of ranking the risk propagation range of the second procurement task based on the propagation structure metric, fitting and analyzing it with the corresponding collaborative evaluation value to select the optimal risk propagation range for procurement task intelligent review of the procurement plan, specifically involves: Obtain the procurement task dependency adjacency matrix corresponding to the risk propagation range of each second procurement task, extract the risk impact task set corresponding to each procurement task node according to the row direction of the matrix, and count the number of tasks in the risk impact task set corresponding to each procurement task node. Based on the number of risk-affected tasks corresponding to each procurement task node, a risk propagation coverage distribution for procurement tasks is constructed, and the dispersion of risk propagation range for different second procurement tasks is analyzed based on the propagation coverage distribution. like Figure 3 As shown, based on the procurement task dependency adjacency matrix corresponding to the risk propagation range of each second procurement task, the longest propagation path between procurement task nodes is searched along the directed propagation relationship in the matrix, and the propagation depth of the corresponding second procurement task risk propagation range is determined according to the length of the longest propagation path. By fusing discreteness and propagation depth, a propagation structure representation value corresponding to the risk propagation range of each second procurement task is generated; The risk propagation range of each second procurement task is sorted according to the size of the propagation structure characterization value. The sorted propagation structure characterization value is used as the horizontal axis and the corresponding collaborative evaluation value is used as the vertical axis to construct a discrete propagation range evaluation sequence. Based on the relationship between the change of the cooperative evaluation value corresponding to the characterization value of adjacent propagation structures in the discrete propagation range evaluation sequence, the change of reward gain in the expansion stage of each propagation structure is calculated, and the coordinated reward enhancement region and the coordinated reward decline region are identified according to the change of reward gain. Based on the range of propagation structure representation values ​​corresponding to the coordinated reward enhancement area, determine the range of continuous expansion constraints of the propagation structure, eliminate the risk propagation range of the second procurement task that exceeds the range of continuous expansion constraints of the propagation structure, and generate a set of constraint candidate propagation ranges. Compare the collaborative evaluation values ​​corresponding to the risk propagation ranges of each second procurement task in the set of candidate propagation ranges, select the second procurement task risk propagation range with the highest collaborative evaluation value as the optimal procurement task risk propagation range, and apply it to the intelligent review of procurement plans.

[0166] In this embodiment, the propagation structure metric (obtained by fusing dispersion and propagation depth) corresponding to each candidate risk propagation range R1-R5 is calculated, and a scatter plot is drawn using this metric as the abscissa and the corresponding collaborative evaluation values ​​S1-S5 as the ordinate (see attached diagram). Figure 5 ).

[0167] Based on the increase or decrease of the collaborative evaluation value between adjacent propagation structure metrics in the scatter plot, the coordination reward enhancement area is identified. R2-R4 in this area are selected as the set of constraint candidate propagation ranges, and R3 with the highest collaborative evaluation value is selected as the target risk propagation range for subsequent intelligent review of procurement plans.

[0168] In this embodiment, the procurement task dependency adjacency matrix corresponding to the risk propagation range of the second procurement task is a binary matrix composed of the risk propagation relationship between procurement task nodes. Each row in the matrix corresponds to a risk-generating task, and each column corresponds to a procurement task that may be affected. When the row and column intersection position is 1, it indicates that the corresponding risk task can propagate the constraint influence to the target procurement task. When the row and column intersection position is 0, it indicates that there is no direct or indirect risk propagation relationship between the two.

[0169] In this embodiment, the risk impact task set is determined by reading the row of each procurement task node in the matrix corresponding to the risk propagation scope of the second procurement task. The procurement task node with an element of 1 in that row is taken as the complete set of tasks that the risk task can affect. For example, when a procurement task corresponds to row P1 in the matrix, and the elements at positions P2 and P3 are both 1, then P2 and P3 together form the risk impact task set corresponding to P1, which describes the scope of tasks that need to be reviewed synchronously when an anomaly occurs in P1.

[0170] In this embodiment, dispersion is a parameter used to describe the balance of the distribution of the impact range corresponding to different risk tasks. The process of obtaining it is as follows: count the number of tasks in the risk impact task set corresponding to each procurement task node to form a risk propagation coverage distribution; when most procurement tasks only affect a small number of tasks, or a small number of procurement tasks affect a large number of tasks, it indicates that the risk propagation range distribution is relatively concentrated and the dispersion is low; when there are significant differences in the number of tasks affected by different procurement tasks, it indicates that the risk propagation range is highly dispersed. This parameter reflects the changes in the risk propagation range of different second procurement tasks in the task coverage structure.

[0171] In this embodiment, the propagation depth is used to describe the maximum level of procurement risk propagation along task dependencies. The process is as follows: Any risk-generating task is selected as the starting node from the second procurement task risk propagation range matrix. Subsequent procurement task nodes are searched level by level along the propagation direction where elements are 1. The number of procurement task connections from the starting node to the farthest affected node is counted, and the maximum number of connections corresponding to all risk tasks is taken as the propagation depth of this propagation range. For example, if procurement task P1 can affect P2, and P2 further affects P3, then P1 to P3 forms a two-level propagation path, and the propagation range's expansion depth is 2.

[0172] In this embodiment, the propagation structure characterization value is a comprehensive evaluation parameter used to convert the risk propagation range of the second procurement task into a sortable value. The acquisition process is as follows: the dispersion and propagation depth corresponding to the risk propagation range of each second procurement task are obtained respectively, and the two are combined according to a unified evaluation direction so that the propagation range that can cover more risk propagation chains but does not produce excessive diffusion has a higher propagation structure characterization value, which is used to represent the structural characteristics of the propagation range between risk coverage capability and propagation complexity.

[0173] In this embodiment, the discrete propagation range evaluation sequence is a set of correspondences between the risk propagation range of the second procurement task and the collaborative evaluation value, arranged in ascending order of propagation structure representation values. The specific acquisition process is as follows: First, the propagation structure representation value corresponding to each risk propagation range of the second procurement task is calculated. Then, the corresponding propagation ranges are rearranged according to this value. The sorted propagation structure representation values ​​are used as the horizontal axis, and the collaborative evaluation value obtained after multi-agent review of the propagation range is used as the vertical axis, forming a discrete data evaluation sequence. Each coordinate point corresponds to an actually generated risk propagation range of the second procurement task; no interpolation or virtual expansion is performed on non-existent propagation ranges.

[0174] In this embodiment, the change in reward gain is the trend of change between the corresponding collaborative evaluation values ​​of two adjacent propagation structure characterization values. The acquisition process is as follows: according to the arrangement order of the discrete propagation range evaluation sequence, the changes in the collaborative evaluation values ​​of two adjacent propagation ranges are compared in turn. When the reward value corresponding to the later propagation range is higher than that of the previous propagation range, it indicates that the expansion of the propagation structure generates positive benefits; when the reward value corresponding to the later propagation range decreases, it indicates that continuing to expand the propagation range leads to a decrease in the review effect.

[0175] In this embodiment, the coordinated reward enhancement region is the range of propagation structure representation values ​​where the reward gain is positive consecutively in the discrete propagation range evaluation sequence. The specific acquisition process is as follows: traverse the evaluation sequence according to the increasing direction of the propagation structure representation values. When the coordinated evaluation values ​​corresponding to multiple consecutive propagation ranges are all higher than the previous propagation range, the corresponding consecutive interval of propagation structure representation values ​​is determined as the coordinated reward enhancement region; when a position of decreasing reward occurs, the current continuous region division ends.

[0176] In this embodiment, the continuous expansion constraint range of the propagation structure serves as a screening boundary to limit the excessive expansion of the risk propagation range of procurement tasks. Its acquisition process involves: reading the minimum and maximum propagation structure representation values ​​corresponding to the coordination reward improvement area, and using the range of propagation structure representation values ​​corresponding to this continuously improving area as the allowed propagation structure range for further expansion. In other words, this constraint range is not manually set but automatically determined based on changes in the actual review effect of different propagation ranges. If the coordination reward decreases after the propagation structure continues to expand, it is considered that continuing to add related tasks would lead to an expansion of the review scope but a decrease in benefits, and therefore these tasks are not included in the candidate range.

[0177] In this embodiment, the set of candidate propagation ranges is the set of second procurement task risk propagation ranges that satisfy the constraint range of continuous expansion of the propagation structure. The acquisition process is as follows: traverse all second procurement task risk propagation ranges, retaining each propagation range whose propagation structure representation value falls within the constraint range of continuous expansion of the propagation structure, and deleting propagation ranges whose propagation structure representation value is outside this range. For example, if the three propagation ranges correspond to propagation structure representation values ​​of 1, 2, 3, 4, and 5, respectively, and the coordination reward only continuously increases within the range of 2 to 4, then only the propagation ranges with representation values ​​of 2, 3, and 4 are retained as the candidate set.

[0178] It should be noted that the propagation structure metric in this step does not directly determine the propagation range based on the matrix size, but rather evaluates the extent of expansion of different propagation ranges through changes in the risk propagation structure of procurement tasks. Since the risk propagation range of procurement tasks is essentially a task association matrix, it cannot be directly numerically ranked. This step first converts the matrix into a propagation structure representation value that reflects the task coverage distribution and propagation chain length, and then uses this representation value to establish a ranking relationship for the propagation range, so that the propagation ranges of different matrix sizes and different numbers of tasks can be compared within a unified evaluation space.

[0179] It should be noted that the purpose of introducing continuous expansion constraints in this step is to avoid the local optimum problem caused by directly selecting the highest collaborative evaluation value. For example, if a certain propagation range only includes a small number of directly related tasks, although the number of agents is small and the interaction load is low, it may obtain a higher reward, but this propagation range may miss subsequent influencing tasks; while a fully expanded propagation range, although with sufficient coverage, will introduce a large number of invalid review tasks. Therefore, this step analyzes the changing trend of rewards as the propagation structure expands, and selects candidate ranges only during the stage of continuous reward improvement, so that the finally selected propagation range simultaneously meets the requirements of risk coverage and review efficiency.

[0180] In a specific embodiment, for example, a large-scale equipment installation project includes main equipment procurement, auxiliary equipment procurement, control system procurement, and commissioning. When the main equipment procurement is delayed, the risk propagation scope of the first type of procurement task only includes the main and auxiliary equipment, resulting in a high coordination reward. However, during actual operation, it is found that the delay in the control system installation is not identified. With the second type of propagation scope further expanding to include the control system task, the risk identification coverage improves, and the coordination reward continues to increase. With the third type of propagation scope further expanding to include all construction material procurement tasks, although the coverage expands, the amount of information that needs to be interacted between agents increases significantly, leading to reduced review efficiency and a decrease in the coordination reward. Therefore, by using a discrete propagation scope evaluation sequence, it is possible to identify the second type of propagation scope as being in a stage of continuously increasing rewards, and use it as a candidate scope. Finally, the propagation scope with the highest corresponding coordination reward is selected for subsequent procurement plan review.

[0181] It should be noted that the significance of this step lies in solving the problem that the risk propagation scope of procurement tasks cannot be directly compared. By converting the binary task association matrix into a propagation structure representation value, the different propagation scopes can be ranked. At the same time, by combining the trend of coordination reward changes, the expansion boundary of the propagation scope is limited, avoiding the risk omission caused by the propagation scope being too small, and the problems of increased number of agents, complex information interaction, and oscillation of procurement plan adjustments caused by the propagation scope being too large. From multiple risk propagation scopes of second procurement tasks, the target risk propagation scope that matches the current project procurement status is selected, thereby improving the accuracy and stability of intelligent review of procurement plans.

[0182] Example 2: This invention also includes a multi-agent collaborative intelligent review system for procurement plans, comprising a constraint propagation modeling module, a propagation range perturbation module, a review mapping control module, a multi-agent review module, a coordination reward generation module, and a propagation range optimization module. The constraint propagation modeling module is used to construct a directed graph matrix of procurement task dependencies and obtain the risk propagation range of the first procurement task based on directed dependency closure analysis. The propagation range perturbation module is used to derive the risk propagation range of the first procurement task into several risk propagation ranges of the second procurement task based on multi-scale path perturbation. The review mapping control module is used to establish the relationship between the risk propagation range of the procurement task and the multi-agent review module. The system establishes a conversion relationship between control quantities, transforming the risk propagation scope of the second procurement task into a smart review control object for the procurement plan. A multi-agent review module is used to perform multi-agent collaborative procurement plan reviews based on the smart review control object corresponding to each risk propagation scope of the second procurement task. A coordination reward generation module extracts features from each multi-agent review process and trains to generate collaborative evaluation values ​​corresponding to the risk propagation scope of the second procurement task. A propagation scope optimization module sorts the risk propagation scope of the second procurement task based on the propagation structure metric, fits it with the corresponding collaborative evaluation value, and analyzes and selects the optimal risk propagation scope for intelligent review of the procurement plan.

[0183] The present invention also includes an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform a procurement plan intelligent review method based on multi-agent collaboration.

[0184] The present invention also includes a computer-readable storage medium storing a computer program that, when executed by a processor, implements a method for intelligent review of procurement plans based on multi-agent collaboration.

[0185] To verify the technical effect of the present invention, a comparative experiment was conducted on a workstation with an Intel Xeon Gold 6248R@3.0GHz, 128GB of memory, and an NVIDIA Tesla V100 GPU, using real engineering project data of three different scales (Project A: 12 procurement tasks, 45 historical records; Project B: 28 procurement tasks, 160 historical records; Project C: 56 procurement tasks, 420 historical records).

[0186] Table 2: Performance Comparison of Filtering Strategies with Different Propagation Ranges

[0187] As shown in Table 2, compared with the fixed propagation range baseline and DCOP subgraph sampling comparison scheme, the method of the present invention improves the risk identification coverage by an average of approximately 21.3 and 8.6 percentage points, respectively, while reducing the proportion of invalid review tasks by an average of approximately 29.4 and 13.0 percentage points.

[0188] Table 3: Comparison of Multi-Agent Interaction Load and Review Time

[0189] As shown in Table 3, the method of the present invention reduces the total number of information interactions between intelligent agents by an average of about 62% while maintaining a high risk identification coverage rate, and reduces the time consumed by a single collaborative review by an average of about 64%, effectively avoiding the collaborative oscillation problem caused by excessive propagation range.

[0190] Table 4: Changes in Collaborative Evaluation Values ​​with Propagation Structure Metrics

[0191] As shown in Table 4, in Project B, the collaborative evaluation value shows a non-monotonic change of first increasing and then decreasing as the propagation structure metric value increases. The collaborative evaluation value corresponding to R3 (R3=0.47) is the highest (0.74), and R2-R4 constitute a continuously improving region, which is consistent with the screening logic of the continuous expansion constraint range of the propagation structure.

[0192] The following sources of error were identified during the experiment: ① Inconsistent labeling criteria for historical procurement execution events may lead to statistical deviations in the synchronous occurrence of historical anomalies; ② An event loss rate of approximately 0.3%–1.2% exists during multi-agent interaction, affecting the accuracy of the collaborative evaluation value; ③ The weights of the four indicators of the collaborative evaluation value are preset by those skilled in the art based on engineering experience, and the target risk propagation range may fluctuate within ±1 candidate range under different weight configurations. To suppress the above errors, in actual deployment, this invention can employ methods such as averaging multiple independent reviews, setting weighted interval searches, and setting safety thresholds in the region of decreasing coordination rewards to enhance robustness.

[0193] In this embodiment, the data flow between the modules is as follows: Figure 2 As shown: the dependency modeling module outputs the risk propagation range to the disturbance sampling module; the disturbance sampling module outputs multiple candidate risk propagation ranges to the mapping control module; the mapping control module outputs the multi-agent review control object to the multi-agent review module; the multi-agent review module outputs the review process data to the coordination evaluation module; the coordination evaluation module outputs the collaborative evaluation value to the range optimization module; and the range optimization module outputs the target risk propagation range to the downstream procurement plan execution unit.

[0194] In the embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0195] Furthermore, the functional modules 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 in the form of hardware plus software functional modules.

[0196] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0197] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0198] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0199] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0200] In the embodiments provided in this disclosure, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0201] It should be noted that, in this disclosure, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element limited by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0202] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for intelligent review of procurement plans based on multi-agent collaboration, characterized in that, Includes the following steps: Construct a procurement task dependency matrix based on the dependencies between procurement tasks; The scope of risk propagation for the procurement task is determined based on the procurement task dependency matrix, and multiple candidate risk propagation scopes are generated through perturbation. Configure procurement review agents for collaborative review of each candidate's risk propagation scope, and generate review process data; Based on the review process data, a collaborative evaluation value is generated for each candidate risk propagation range, and the candidate risk propagation range with the highest collaborative evaluation value is selected for intelligent review of the procurement plan. The collaborative evaluation value is obtained by extracting features from the review process data and training a proportional coefficient after weighting each feature.

2. The intelligent review method for procurement plans based on multi-agent collaboration as described in claim 1, characterized in that, The dependencies between the procurement tasks include: equipment installation sequence, process connection, and engineering node constraints.

3. The intelligent review method for procurement plans based on multi-agent collaboration according to claim 2, characterized in that, The construction of the procurement task dependency matrix includes: Obtain procurement plan data and divide each equipment procurement item into several procurement task nodes; Based on the dependencies, determine the directed dependencies between each procurement task node; Arrange the procurement task nodes in a preset order, set the matrix elements corresponding to nodes with directed dependencies to a first preset value, and set the rest to a second preset value to obtain the procurement task dependency matrix.

4. The intelligent review method for procurement plans based on multi-agent collaboration according to claim 3, characterized in that, Determining the scope of risk propagation for procurement tasks includes: Collect historical procurement execution events, obtain the status changes of each historical abnormal task and its related tasks, and statistically analyze the synchronous occurrence of historical abnormalities in each propagation relationship. Based on the historical anomaly occurrences, propagation relationships with historical verification are selected from the propagation relationships corresponding to the procurement task dependency matrix to generate the risk propagation scope.

5. The intelligent review method for procurement plans based on multi-agent collaboration according to claim 4, characterized in that, The generation of multiple candidate risk propagation ranges through perturbation includes: The disturbance scale is divided according to the length of the path associated with the procurement task, and the risk propagation range is sampled multiple times. Each perturbation sampling removes several propagation relationships within the risk propagation range and adds several propagation relationships not included in the risk propagation range to obtain the perturbation result; The results of each disturbance are combined to generate the multiple candidate risk propagation ranges.

6. The intelligent review method for procurement plans based on multi-agent collaboration according to claim 5, characterized in that, The configuration procurement review intelligent agent includes: For each candidate risk propagation scope, extract the risk impact task set for each procurement task node; Configure a procurement review agent for each risk source task, use the risk source task as the initial review task of the procurement review agent, and use the associated tasks in its risk impact task set as the associated review tasks of the procurement review agent. Based on the propagation direction between each procurement task in the procurement task dependency matrix, establish the information transmission relationship between each procurement review agent; The initial review task, associated review task, and information transmission relationship are integrated to generate a multi-agent review control model.

7. The intelligent review method for procurement plans based on multi-agent collaboration according to claim 6, characterized in that, The collaborative review includes: Obtain procurement execution data corresponding to each procurement review intelligent agent to form review input data. The procurement execution data includes plan execution data, engineering node data, supply execution data, and fund execution data. Based on the review input data, the intelligent agent identifies abnormal events during the execution of procurement tasks, including time deviation events, supply fulfillment deviation events, funding conflict events, and events affecting related tasks. The abnormal events are exchanged between the intelligent agents according to the information transmission relationship; The intelligent agent updates the risk assessment results based on the event information after the interaction, forming review process data.

8. The intelligent review method for procurement plans based on multi-agent collaboration according to claim 7, characterized in that, The generation of collaborative evaluation values ​​corresponding to the propagation range of each candidate risk includes: Based on the review process data, determine the risk identification coverage, the omission rate of related tasks, the proportion of invalid review tasks, and the agent interaction load. The risk identification coverage rate is the ratio of the number of procurement tasks that actually experienced anomalies and were identified by the risk assessment results to the total number of procurement tasks that actually experienced anomalies. The associated task omission rate is the ratio of the number of procurement tasks in the actual impact task set that are not covered by the associated review task set to the total number of actual impact tasks. The percentage of invalid review tasks is the ratio of the number of review tasks that participated in the review but did not generate constraint events, did not trigger risk assessment updates, and did not affect the status changes of subsequent procurement tasks to the total number of procurement tasks that participated in the review. The intelligent agent interaction load is the sum of the number of information transmissions between each procurement review intelligent agent, the number of repeated transmissions of the same constraint event, and the number of risk judgment updates; The collaborative evaluation value is generated by combining the risk identification coverage, the omission rate of related tasks, the proportion of invalid review tasks, and the interaction load of the intelligent agent.

9. The intelligent review method for procurement plans based on multi-agent collaboration as described in claim 8, characterized in that, Determining the scope of risk propagation for procurement tasks includes: For each candidate risk propagation range, the number of risk-affected tasks corresponding to each procurement task node is counted to construct the risk propagation coverage distribution of procurement tasks. Search for the longest propagation path along the directed propagation relationship of the procurement task dependency matrix to determine the propagation depth of each candidate risk propagation range; Based on the dispersion of the risk propagation coverage distribution and the propagation depth, a propagation structure metric value corresponding to each candidate risk propagation range is generated; Determine the continuous expansion constraint range of the propagation structure metric, and select the candidate with the highest collaborative evaluation value from the candidate risk propagation ranges that are within the continuous expansion constraint range as the target risk propagation range.

10. A system using the intelligent review method for procurement plans based on multi-agent collaboration as described in any one of claims 1-9, characterized in that, It includes a constraint propagation modeling module, a propagation range perturbation module, a review mapping control module, and a multi-agent review module; The constraint propagation modeling module is used to construct a procurement task dependency matrix based on the dependencies between procurement tasks; The propagation range perturbation module is used to determine the risk propagation range of the procurement task based on the procurement task dependency matrix, and to generate multiple candidate risk propagation ranges through perturbation. The review mapping control module is used to configure procurement review agents for collaborative review based on the risk propagation scope of each candidate, and generate review process data. The multi-agent review module is used to generate collaborative evaluation values ​​corresponding to each candidate risk propagation range based on the review process data, and select the candidate risk propagation range with the highest collaborative evaluation value for intelligent review of the procurement plan.

Images