Project decision optimization control method, device and equipment based on knowledge graph

Abstract

The invention discloses a project decision optimization control method, device and equipment based on a knowledge graph, and relates to the technical field of project decision optimization control, and the method comprises the following steps: constructing a project knowledge graph, and carrying out the consistency verification of the project knowledge graph; performing semantic analysis on the verified knowledge graph based on a multi-target efficiency function to generate a project control strategy initial solution set; performing compression screening on the control strategy initial solution set based on a graph neural network method to obtain a compression control strategy set; converting the compression control strategy set into an executable control instruction, and inputting the executable control instruction into a project implementation system; according to the method, the project knowledge graph based on multi-dimensional semantic association is constructed, deep reasoning and screening of the multi-target efficiency function and the graph neural network are combined, the complex decision problem that optimal configuration and dynamic adjustment are difficult to achieve in project management is effectively solved, and optimal decision control over the whole process of the project is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of project decision optimization control, and more specifically, to a project decision optimization control method, device and equipment based on knowledge graph. Background Art

[0002] With the increasing complexity of project management, especially in engineering projects characterized by multi-resource collaboration, highly coupled tasks, and frequent dynamic risks, traditional project control methods based on work breakdown structures and critical path methods are no longer able to meet the challenges of managing multi-dimensional data interactions, dynamic changes, and uncertainties during project execution. Against this backdrop, achieving intelligent perception, knowledge-based modeling, and optimal decision-making and control across the entire project process has become a critical issue in engineering management that urgently needs to be addressed.

[0003] Existing project management decision-making methods often rely on static process modeling and empirical rules, lacking the ability to systematically model the complex semantic relationships between multidimensional factors such as tasks, resources, constraints, and risks. This "structure-process" approach struggles to effectively capture the semantic logic, causal coupling, and propagation paths between various entities in project execution, making it difficult to optimally configure and dynamically adjust project control strategies. Furthermore, project decision-making processes often face multi-objective optimization challenges, including cost control, schedule assurance, risk avoidance, and efficient resource utilization. Traditional methods are limited in their ability to comprehensively balance trade-offs and achieve global optimization.

[0004] In recent years, knowledge graphs, a complex network modeling technology that integrates structured and semantic information, have been widely used in fields such as intelligent question answering, recommendation systems, and medical diagnosis. They have demonstrated significant advantages in fusion modeling, semantic analysis, and knowledge mining of multi-source heterogeneous data. Introducing knowledge graphs into the field of project management can enable semantic modeling and reasoning support for multiple factors such as tasks, resources, and risks, providing a unified knowledge foundation for multi-objective decision-making and control. This paper proposes a solution to this problem. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a project decision optimization control method, device and equipment based on knowledge graph. By constructing a project knowledge graph based on multi-dimensional semantic association, combining multi-objective performance function with in-depth analysis and screening of graph neural network, it effectively solves the complex decision-making problems in project management that are difficult to achieve optimal configuration and dynamic adjustment, and realizes optimal decision control of the entire project process.

[0006] To achieve the above object, the present invention provides the following technical solutions: In the first aspect, the present application provides a project decision optimization control method based on a knowledge graph, which includes: constructing a project knowledge graph and performing consistency verification on the project knowledge graph; performing semantic analysis on the verified knowledge graph based on a multi-objective performance function to generate an initial solution set of the project control strategy; compressing and screening the initial solution set of the control strategy based on a graph neural network method to obtain a compressed control strategy set; converting the compressed control strategy set into executable control instructions, and inputting them into the project implementation system.

[0007] In one embodiment, a project knowledge graph is constructed, specifically by: collecting heterogeneous data related to project implementation and performing structured processing on the heterogeneous data, wherein the heterogeneous data includes task data, resource data, and constraint data; constructing semantic nodes based on the processed data, wherein the semantic nodes include task nodes, resource nodes, and constraint nodes; constructing semantic relationship edges between the semantic nodes; and structurally fusing the semantic nodes and semantic relationship edges to generate a project knowledge graph.

[0008] In one of the embodiments, the project knowledge graph is subjected to consistency verification, specifically: the semantic nodes and semantic relationship edges in the project knowledge graph are converted into knowledge graph triples; the triples are encoded into high-order sparse tensors, and tensor decomposition operations are performed to obtain embedding vectors; the high-order sparse tensors are reconstructed through the embedding vectors, and the reconstruction errors of the triples are calculated; the triples whose reconstruction errors exceed a preset threshold are marked as preliminary structural anomaly candidates; based on the preliminary structural anomaly candidates, a regularized tensor decomposition model is constructed, and consistency logic is introduced as a regularization term; a regularized embedding vector is output based on the regularized tensor decomposition model, and secondary decomposition operations and reconstruction are performed to generate a structural anomaly report and perform consistency verification.

[0009] In one embodiment, a semantic analysis is performed on the verified knowledge graph based on a multi-objective effectiveness function to generate an initial solution set of project control strategies, specifically: a multi-objective effectiveness function is constructed, wherein the multi-objective effectiveness function includes the total project cost, project cycle, risk exposure, and resource utilization; based on the knowledge graph, a semantic mapping rule set is constructed and control strategy generation is performed to obtain a candidate control strategy set; for each strategy in the candidate control strategy set, a quantitative evaluation is performed based on the multi-objective effectiveness function to obtain a performance index vector; according to the performance index vector, the candidate control strategy set is sorted by non-inferiority, and the initial solution set of control strategies is obtained according to the sorting result.

[0010] In one embodiment, based on the knowledge graph, a semantic mapping rule set is constructed and control strategy generation is performed to obtain candidate control strategies, specifically: the entities and relationships in the verified knowledge graph are structured, and the relationships in the graph are divided into control relationships according to preset control semantics, and the control relationships include task dependency relationships, resource allocation relationships, risk transfer relationships, and time cost binding relationships; based on the control relationships, a project control strategy construction template is constructed, and the project control strategy construction template includes a task sequence template, a resource allocation template, a risk avoidance template, and a time cost constraint template; a semantic mapping rule set is established, and the semantic mapping rule set maps the semantic relationships in the knowledge graph to the project control strategy construction template; based on the semantic mapping rule set, semantic analysis is performed on the graph through a logic rule engine to generate a candidate control strategy set.

[0011] In one embodiment, the initial solution set of the control strategy is compressed and screened based on the graph neural network method to obtain a compressed control strategy set, specifically: extracting a performance indicator vector from each initial solution set of the control strategy, constructing a control strategy performance graph structure, and inputting it into a preset graph neural network model to obtain a graph embedding feature vector; classifying the graph embedding feature vector through an unsupervised clustering algorithm to obtain a control strategy cluster; calculating the Euclidean distance of all control strategies from the cluster center in the embedded feature space according to the control strategy cluster, selecting the control strategy with the smallest distance as the representative strategy of the cluster, and forming a preliminary representative strategy set; constructing the ε-dominance relationship between the strategies of the preliminary representative strategy set according to the preset performance tolerance threshold ε, and traversing to eliminate the ε-dominated strategies to form the final compressed control strategy set.

[0012] In one embodiment, a control strategy performance graph structure is constructed and input into a preset graph neural network model to obtain a graph embedding feature vector, specifically: the similarity of the performance indicator vectors between the initial solution sets of each control strategy is calculated to form a strategy performance similarity matrix; based on the strategy performance similarity matrix, a control strategy performance graph is constructed; the control strategy performance graph is embedded and learned through a graph neural network model, and an inter-layer information propagation mechanism is executed; after multi-layer graph convolution operations, the final graph embedding feature vector of each control strategy node is obtained.

[0013] In one embodiment, the compressed control policy set is converted into executable control instructions and input into the project implementation system, specifically: a control instruction mapping template is constructed, wherein the control instruction mapping template includes a task scheduling instruction template, a resource allocation instruction template, a time window instruction template, and a risk avoidance instruction template; according to the control instruction mapping template, the compressed control policy set is parsed one by one according to the template format and the corresponding instruction fields are filled to generate a standardized control instruction set; the standardized control instruction set is converted into an interface call format supported by the project scheduling system; after the conversion is completed, it is pushed to the project scheduling execution engine to realize the scheduling execution control of tasks, the dynamic allocation management of resources, the execution monitoring of time windows and the risk response control.

[0014] In a second aspect, the present application provides a project decision optimization control device based on a knowledge graph, comprising: A knowledge graph construction and verification module is used to construct a project knowledge graph and perform consistency verification on the project knowledge graph; The control strategy initial solution set generation module is used to perform semantic analysis on the verified knowledge graph based on the multi-objective effectiveness function to generate the project control strategy initial solution set; The control strategy initial solution set compression and screening module is used to compress and screen the control strategy initial solution set based on the graph neural network method to obtain a compressed control strategy set; The control strategy conversion application module is used to convert the compressed control strategy set into executable control instructions and input them into the project implementation system.

[0015] In a third aspect, the present application further provides an electronic device, comprising: memory for storing computer programs; A processor is used to execute the computer program to implement the steps of the aforementioned knowledge graph-based project decision optimization control method.

[0016] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages: 1. By constructing a project knowledge graph driven by multi-source heterogeneous data, combined with high-order sparse tensor decomposition and consistency logic regularization verification, the semantic integrity and logical consistency of project data are systematically guaranteed, effectively avoiding reasoning deviations caused by data conflicts, omissions or semantic errors; based on the verified high-quality knowledge graph, a multi-objective performance function is used for refined semantic analysis, achieving an optimized balance of project control strategies in multiple dimensions such as cost, cycle, risk and resource utilization, with strong decision-making support capabilities and good robustness, significantly improving the scientific nature and execution efficiency of project management.

[0017] 2. By introducing graph neural networks, deep feature extraction is performed on the control strategy performance graph constructed with multi-objective performance indicators. Combined with unsupervised clustering and ε-dominance screening mechanisms, the structural correlation and performance similarity between control strategies are effectively explored, and the precise compression of redundant strategies and efficient screening of representative strategies are achieved, which significantly reduces the scale and computational complexity of the control strategy solution set. Subsequently, the compressed control strategy is converted into executable control instructions through standardized mapping templates to realize the automated control of project task scheduling and resource allocation, improve the scientific nature of strategy selection and the practicality of implementation, and enhance the execution efficiency and robustness of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of the process flow of the project decision optimization control method based on the knowledge graph provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of a project decision optimization control system based on a knowledge graph provided in an embodiment of the present application; Figure 3 Schematic diagram of the project decision optimization control method based on knowledge graph provided in the embodiment of this application Figure 2 . DETAILED DESCRIPTION

[0019] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0020] Reference Figure 1 As shown, the project decision optimization control method based on knowledge graph provided by the present invention includes the following S101 to S104. S101: Build a project knowledge graph and perform consistency check on the project knowledge graph.

[0021] The project knowledge graph is a data model that structures and models the multi-source, heterogeneous data involved in project management (such as tasks, resources, and constraints) into a semantically clear, graph-like representation. Its core concept is to represent entities (nodes) and their relationships (edges) in a graph format, enabling the association, query, reasoning, and decision support of information across the entire project lifecycle.

[0022] While project knowledge graphs offer strong structural expressiveness, logical conflicts, missing data, and semantic errors within them can severely impact subsequent reasoning and control strategy generation. Therefore, consistency verification is a prerequisite for ensuring graph quality and usability, and a fundamental guarantee for ensuring control strategy effectiveness, model robustness, and system enforceability.

[0023] In step S101, a project knowledge graph is constructed, including S201 to S204, wherein: S201 , collecting heterogeneous data related to project implementation and performing structured processing on the heterogeneous data, wherein the heterogeneous data includes task data, resource data, and constraint condition data.

[0024] S202 : Constructing semantic nodes based on the processed data, wherein the semantic nodes include task nodes, resource nodes, and constraint condition nodes.

[0025] S203, constructing semantic relationship edges between the semantic nodes, wherein the semantic relationship edges include edge type, directionality, and optional attribute fields, and are used to represent semantic information such as relationship strength, time slack, and priority.

[0026] S204, structurally fuse the semantic nodes and semantic relationship edges to generate a project knowledge graph. The graph has node type identification and edge type distinction, and can be represented by a graph database structure, graph data objects or triple set.

[0027] Semantic relationship edges include dependency edges between task nodes, which represent the order of tasks; resource requirement edges between task nodes and resource nodes; constraint edges between task or resource nodes and constraint nodes; and conflict edges between resource nodes or task nodes. Task data represents project work content, resource data represents resource allocation capabilities, and constraint data represents project scheduling and execution restrictions. Task nodes represent task units within a project; resource nodes represent schedulable resource objects within a project; and constraint nodes represent task or resource constraints.

[0028] In an exemplary embodiment, Figure 3 Schematic diagram of the project decision optimization control method based on knowledge graph provided in the embodiment of this application Figure 2 ,exist Figure 1 On the basis of the above, the steps of the project decision optimization control method based on the knowledge graph are exemplified. In step S101, the consistency check of the project knowledge graph is performed, including S301 to S307, wherein: S301: Convert the semantic nodes and their semantic relationship edges in the project knowledge graph into knowledge graph triples ,in, For the Individual entities, No. relationship type, No. The tail entity, is the total number.

[0029] S302, encoding the triplet into a high-order sparse tensor ; The dimension of the tensor corresponds to the combined structure of entities and relationships.

[0030] S303: Perform a tensor decomposition operation on the tensor based on the CP decomposition method to obtain embedding vectors of entities and relationships in the latent semantic space.

[0031] The specific calculation formula for the decomposition operation is as follows:

[0032] Where, is the dimension of the latent semantic space, Head entity In the latent semantic space dimensional embedding vector, For the tail entity In the latent semantic space dimensional embedding vector, The relationship type in the latent semantic space dimensional embedding vector, is the embedding vector of the head entity, is the tail entity embedding vector, Embedding vector for the relationship.

[0033] S304, reconstruct the high-order sparse tensor through the embedding vector and calculate the reconstruction error of the triple .

[0034] The specific calculation formula of the reconstruction error is as follows:

[0035] S305 , marking triples whose reconstruction errors exceed a preset threshold as preliminary structural anomaly candidates, and not marking those that do not exceed the threshold.

[0036] For example, preliminary structural anomaly candidates can be understood as seemingly unreasonable structures such as "task A depends on task A" (it depends on itself) or "resource B is preempted by two tasks at the same time".

[0037] S306: Build a regularized tensor decomposition model based on preliminary structural anomaly candidates , and introduces consistency logic as a regular term.

[0038] Consistency logic is a set of constraints expressed logically within the knowledge graph, defining which structural combinations are "semantically reasonable" and which violate common sense or business logic. This can include task topology rules, which stipulate that project tasks cannot be self-dependent or form a circular dependency structure; resource scheduling rules, which stipulate that the same resource unit cannot be occupied by multiple tasks simultaneously; and constraint logic rules, such as that the start time must be less than the end time and that task priorities cannot conflict with each other.

[0039] For example, introducing consistency logic as a regularization term can be understood as converting the rule into a loss function: when the embedding vector violates the rule, a higher penalty is generated, and when it satisfies the rule, the penalty tends to zero. This is used to penalize latent representations that violate semantic consistency, thereby guiding the model to learn more accurate and logical embedding vectors.

[0040] S307: Output the regularized embedding vector based on the regularized tensor decomposition model, perform secondary decomposition and reconstruction, generate a structural anomaly report, and perform consistency verification for project map optimization or manual proofreading.

[0041] Regularized embeddings refer to the introduction of predefined consistency rules as regularization terms during the knowledge graph tensor decomposition process. This ensures that the learned entity and relationship embeddings not only reconstruct the triple data structure (head entity-relationship-tail entity) but also adhere to the consistent logic of the project domain as much as possible. Structural anomaly reports include high-confidence anomaly triples, their associated semantic subgraphs (i.e., the upstream and downstream paths of the triples in the graph), and the name or expression of the potentially violated rule.

[0042] For example, operation and maintenance experts and project managers can locate problem areas based on this report and determine whether to modify the knowledge graph or update the project schedule based on the task context.

[0043] The consistency logic is introduced as a regular term, and the specific calculation formula is as follows:

[0044] The specific calculation formula of the regularized tensor decomposition model is as follows:

[0045] Where, To introduce consistency logic as a regular term, is the total amount of consistency logic introduced, is the penalty weight of the j-th rule, The regularization function constructed for the j-th consistency logic imposes a penalty constraint on the embedding set, is the embedding set of all entities, is the embedding set of all relations, is the embedding regularization term, is the weight coefficient of the L2 regularization term, is the L2 norm squared sum of all entity embedding vectors, is the sum of the squared L2 norm of all relation embedding vectors.

[0046] S102, based on the multi-objective effectiveness function, performs semantic analysis on the verified knowledge graph to generate an initial solution set of the project control strategy, including S401 to S404, where: S401, constructing a multi-objective effectiveness function, wherein the multi-objective effectiveness function includes total project cost, project cycle, risk exposure, and resource utilization.

[0047] S402: Based on the knowledge graph, a semantic mapping rule set is constructed and control strategy generation is performed to obtain a candidate control strategy set.

[0048] S403, for each strategy in the candidate control strategy set, quantitative evaluation is performed based on the multi-objective effectiveness function to obtain the performance index vector , quantify its performance in the four control target dimensions of total project cost, execution cycle, risk exposure and resource utilization.

[0049] S404 , performing non-inferiority sorting on the candidate control strategy set according to the performance indicator vector, and obtaining an initial solution set of the control strategy according to the sorting result.

[0050] Among them, the candidate control strategy set is sorted by non-inferiority, specifically: If there is a strategy The following two conditions are met: 1. For all performance indicators G∈{1,2,3,4}, , where for resource utilization indicators use Comparison of the advantages and disadvantages; 2. There is at least one performance indicator that makes the inequality true; It is believed that strategy Dominance Strategy ,Will Mark as "bad solution"; Repeat the above comparison process, and finally mark the set of candidate control strategies that are not dominated by other strategies as the "effective control strategy set", and use this set as the initial solution set of control strategies for the subsequent optimization module.

[0051] In step S401, the total project cost Refers to the direct and indirect costs in the process of task execution; project cycle Measures the overall time span from task initiation to completion; risk exposure Measures the cumulative likelihood and severity of risk events during the execution of a strategy; resource utilization Measures the time density and conflict level of resource entities during project execution.

[0052] The project cycle , the specific calculation formula is as follows:

[0053] Where, is the set of all paths from the starting task to the ending task, is the execution time of the tasks in the path, For the path, Number the path.

[0054] Risk exposure , the specific calculation formula is as follows:

[0055] Where, is the kth risk event, For inclusion flag, if the policy Contains and risks If the task path is relevant, it is 1; otherwise, it is 0. is the risk trigger probability, is the severity of the risk, is a collection of risk events.

[0056] Resource utilization , the specific calculation formula is as follows:

[0057] Where, For resource collection, is the jth resource, For resources The actual occupied time, is the total execution time of the project, is the impulse punishment factor, Penalty for impulsiveness.

[0058] Based on the total project cost , project cycle , risk exposure , resource utilization Get the performance indicator vector .

[0059] In step S402, based on the knowledge graph, a semantic mapping rule set is constructed and control strategy generation is performed to obtain candidate control strategies, including S501 to S504, wherein: S501, structure the entities and relationships in the verified knowledge graph, and divide the relationships in the graph into control relationships according to the preset control semantics. The control relationships include task dependency relationships, which are used to represent the order of project tasks; resource allocation relationships, which are used to represent the demand and priority of tasks for resources; risk transfer relationships, which are used to describe the potential risk transmission paths between tasks; and time-cost binding relationships, which are used to define the time window and budget range of tasks.

[0060] S502, constructing a project control strategy construction template based on the control relationship, the project control strategy construction template includes a task sequence template, which is a set of ordered task pairs obtained by reasoning about task dependency relationships; a resource allocation template, which is a set of task-resource-priority triples generated by resource allocation relationships; a risk avoidance template, which is a set of task avoidance paths generated by risk transfer relationships; and a time cost constraint template, which is a set of task time windows and budget constraint triples generated by time-cost binding relationships.

[0061] S503, establishing a semantic mapping rule set, which maps the semantic relationship in the knowledge graph to the project control strategy construction template.

[0062] For example, the semantic mapping rule set includes: if there is a relationship of "task A depends on task B", then a control constraint task B → task A is added to the task sequence template, indicating that task A depends on task B and task B should be executed first; If there is a relationship of "Task A requires resource Z, priority is y", then generate a resource allocation template (Task A, Resource Z, Priority is y), indicating that Task A requires resource Z and its priority is y; If there is a "path from which the risk of Task B spreads to Task F", a risk avoidance path Task B → Task F is generated to represent the propagation path where the risk caused by Task D may affect Task F; If there is a relationship of "Task E's execution time limit is [t1, t2], and the budget is Cost_C", then a time cost constraint template (Task E, [t1, t2], Cost_C) is generated, indicating that the allowed execution time window of Task E is [t1, t2], and the corresponding budget cost upper limit is Cost_C.

[0063] S504, performing semantic analysis on the graph through a logic rule engine based on the semantic mapping rule set to generate a candidate control strategy set, wherein the candidate control strategy set includes task sequence, resource allocation, risk path avoidance and time cost constraint control templates.

[0064] S103, compressing and screening the initial control strategy solution set based on the graph neural network method to obtain a compressed control strategy set, including S601 to S604, where: S601, based on the initial solution set of the control strategy, extract the performance indicator vector from each control strategy, construct the control strategy performance graph structure, and obtain the graph embedding feature vector corresponding to each control strategy based on the preset graph neural network model.

[0065] Among them, the graph embedding feature vector refers to the continuous real number vector of fixed dimension used to represent each control strategy node after the control strategy performance graph structure is propagated and hierarchically aggregated by the graph neural network model. While maintaining the original performance indicators, this vector also encodes the relationship pattern between control strategies in the graph structure.

[0066] S602: Classify the graph embedding feature vectors using an unsupervised clustering algorithm to obtain control strategy clusters.

[0067] The clustering algorithm is a K-means clustering algorithm or a spectral clustering algorithm. In the graph embedding feature space, each class represents a group of control strategy clusters with similar structures and performance characteristics. The division of strategy clusters is used for subsequent representative strategy screening and redundancy elimination.

[0068] S603, calculating the Euclidean distance between all control strategies and the cluster center in the embedded feature space according to the control strategy cluster, selecting the control strategy with the smallest distance as the representative strategy of the cluster, and combining the representative strategies of each cluster to form a preliminary representative strategy set.

[0069] The specific calculation formula of the Euclidean distance is as follows:

[0070] Where, For control strategy With cluster center The Euclidean distance, 、 is the embedded feature vector on the w-th dimension, is the dimension of the graph embedding vector.

[0071] In step S604, an ε-dominance relationship is constructed between the strategies for the preliminary representative strategy set based on a preset performance tolerance threshold ε. If a strategy is not inferior to another strategy in all objectives and is superior to the latter in at least one objective by more than the ε threshold, the latter is considered redundant. The strategy pairs are traversed, and the strategies that are ε-dominated are eliminated. The high-value strategies are retained to form the final compression control strategy set.

[0072] In step S601, based on the initial solution set of the control strategies, a performance indicator vector is extracted from each control strategy, and a control strategy performance graph structure is constructed. Based on the preset graph neural network model, a graph embedding feature vector corresponding to each control strategy is obtained, including S701 to S704, where: S701, calculating the similarity of the performance indicator vectors between the initial solution sets of each control strategy to form a strategy performance similarity matrix.

[0073] The specific calculation formula of the similarity is as follows:

[0074] Where, For strategy and strategies The similarity between and Strategy Performance indicator vectors and strategies A vector of performance indicators.

[0075] S702: Construct a control strategy performance graph H=(V, Q, U) based on the strategy performance similarity matrix.

[0076] Among them, V is the node set corresponding to the initial solution set of the control strategy. Each node carries a performance indicator vector as its attribute. Q is the edge set, indicating that there is a high-similarity connection relationship between the strategy pairs. U is the weighted edge, and the edge weight is equal to the similarity value.

[0077] S703: Embed the control strategy performance graph through a graph neural network model and execute an inter-layer information propagation mechanism.

[0078] The inter-layer information propagation mechanism is specifically as follows:

[0079] Where, is the updated node feature matrix, which represents the embedded vector of the node (i.e., control strategy) after a layer of feature propagation and mapping. is the ReLU nonlinear activation function, is the degree matrix, which reflects the connection strength between each control strategy and other strategies. is an adjacency matrix with self-loops, For the The weight parameter matrix of the layer, For the The node feature representation matrix of the layer.

[0080] S704: After multi-layer graph convolution operations, the final graph embedding feature vector of each control strategy node is obtained.

[0081] Among them, the preset graph neural network model is a graph convolutional network with a feature propagation mechanism; in each layer of the graph convolutional network, the node features are obtained by weighted aggregation of its own features and the features of adjacent nodes, and the aggregation weight is determined by the weighted edges; after multi-layer graph convolution operations, the graph embedding feature vector corresponding to each control strategy is obtained, and the graph embedding feature vector simultaneously characterizes the performance characteristics of the strategy and the relationship between the graph structure.

[0082] It should be noted that by introducing a graph neural network model to perform graph structure modeling and feature extraction on the initial solution set of control strategies, it is possible to fully exploit the structural similarities and potential correlation characteristics between control strategies while maintaining the original multi-objective performance indicators. The unsupervised clustering and representative strategy screening mechanism based on graph-embedded feature vectors can effectively eliminate control strategies with performance or structural redundancy, retaining only representative non-redundant strategy samples. This significantly compresses the size of the control strategy solution set, reduces the complexity of subsequent optimization calculations, and improves the efficiency of strategy selection. It has excellent dimensionality reduction, expression integrity, and engineering applicability.

[0083] S104, converting the compressed control strategy set into executable control instructions and inputting them into the project implementation system to implement project task scheduling and resource allocation control, including S801 to S804, wherein: S801, construct a control instruction mapping template, which includes a task scheduling instruction template: converting the task sequence relationship into task start trigger logic; a resource allocation instruction template: converting resource requirements and priority information into resource occupancy time slices and exclusive scheduling tables; a time window instruction template: converting time constraints into hard window configuration parameters of the task scheduler; and a risk avoidance instruction template: converting risk path avoidance logic into task selection or jump strategies.

[0084] S802: According to the control instruction mapping template, the compressed control policy set is parsed one by one according to the template format and the corresponding instruction fields are filled to generate a standardized control instruction set.

[0085] Among them, each control strategy generates a set of control instructions, and the format of the control instruction set meets the input interface specification of the project implementation system control module.

[0086] S803, converting the standardized control instruction set into an interface calling format supported by the project scheduling system.

[0087] Among them, the interface call format includes task scheduling table (Gantt chart form), resource allocation table, scheduling script, system call interface configuration, etc., which constitute a deployable system control instruction set.

[0088] S804: After the conversion is completed, it is pushed to the project scheduling execution engine to realize the scheduling execution control of tasks, dynamic allocation management of resources, execution monitoring of time windows and risk response control.

[0089] Reference Figure 2 As shown, the present invention provides Figure 2 This is a schematic diagram of the structure of a project decision optimization control system based on a knowledge graph provided in an embodiment of the present application, including a knowledge graph construction and verification module, a control strategy initial solution set generation module, a control strategy initial solution set compression and screening module, and a control strategy conversion and application module. There are connections between the modules: A knowledge graph construction and verification module is used to construct a project knowledge graph and perform consistency verification on the project knowledge graph; The control strategy initial solution set generation module is used to perform semantic analysis on the verified knowledge graph based on the multi-objective effectiveness function to generate the project control strategy initial solution set; The control strategy initial solution set compression and screening module is used to compress and screen the control strategy initial solution set based on the graph neural network method to obtain a compressed control strategy set; The control strategy conversion application module is used to convert the compressed control strategy set into executable control instructions and input them into the project implementation system.

[0090] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0091] The above embodiments may be implemented in whole or in part through software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0092] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0093] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0094] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0095] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. The project decision optimization control method based on knowledge graph is characterized by: The steps include: Build a project knowledge graph and perform consistency check on the project knowledge graph; Perform semantic analysis on the verified knowledge graph based on the multi-objective effectiveness function to generate the initial solution set of the project control strategy; The initial solution set of the control strategy is compressed and screened based on the graph neural network method to obtain the compressed control strategy set; The compressed control strategy set is converted into executable control instructions and input into the project implementation system.

2. The project decision optimization control method based on knowledge graph according to claim 1 is characterized in that: The construction of the project knowledge graph is specifically as follows: Collecting heterogeneous data related to project implementation and structuring it, the heterogeneous data includes task data, resource data, and constraint data; Constructing semantic nodes based on the processed data, wherein the semantic nodes include task nodes, resource nodes and constraint condition nodes; Constructing semantic relationship edges between the semantic nodes; Structural fusion of semantic nodes and semantic relationship edges is performed to generate a project knowledge graph.

3. The project decision optimization control method based on knowledge graph according to claim 2 is characterized in that: The consistency check of the project knowledge graph is specifically as follows: Convert the semantic nodes and their semantic relationship edges in the project knowledge graph into knowledge graph triples; Encode the triples into a high-order sparse tensor and perform tensor decomposition to obtain the embedding vector; Reconstruct a high-order sparse tensor through the embedding vector and calculate the reconstruction error of the triple; The triplets whose reconstruction error exceeds a preset threshold are marked as preliminary structural anomaly candidates; Based on the preliminary structural anomaly candidates, a regularized tensor decomposition model is constructed, and consistency logic is introduced as a regularization term; Based on the regularized tensor decomposition model, the regularized embedding vector is output, and secondary decomposition operations and reconstruction are performed to generate a structural anomaly report and perform consistency verification.

4. The project decision optimization control method based on knowledge graph according to claim 3 is characterized in that: The semantic analysis of the verified knowledge graph is performed based on the multi-objective effectiveness function to generate an initial solution set of the project control strategy, specifically: Construct a multi-objective effectiveness function, generate control strategies based on the knowledge graph and semantic mapping rule set, and obtain a set of candidate control strategies; Based on the multi-objective effectiveness function, each strategy in the candidate control strategy set is quantitatively evaluated to obtain a performance index vector; According to the performance index vector, the candidate control strategy set is sorted in a non-inferiority manner, and the initial solution set of the control strategy is obtained according to the sorting result.

5. The project decision optimization control method based on knowledge graph according to claim 4 is characterized in that: The control strategy is generated based on the knowledge graph and semantic mapping rule set to obtain a candidate control strategy set, which is specifically: Structural processing is performed on the entities and relationships in the verified knowledge graph, and the relationships in the graph are divided into control relationships according to the preset control semantics; Construct a project control strategy template based on the control relationship, and map the semantic relationship in the knowledge graph to the project control strategy template by establishing a semantic mapping rule set; Based on the semantic mapping rule set, the graph is subjected to semantic analysis by the logic rule engine to generate a set of candidate control strategies.

6. The project decision optimization control method based on knowledge graph according to claim 1 is characterized in that: The initial solution set of the control strategy is compressed and screened based on the graph neural network method to obtain a compressed control strategy set, specifically: Extract performance indicator vectors from each control strategy’s initial solution set, construct a control strategy performance graph structure, and input it into a preset graph neural network model to obtain a graph embedding feature vector; The graph embedding feature vectors are classified by unsupervised clustering algorithm to obtain control strategy clusters; According to the control strategy cluster, the Euclidean distance between all control strategies and the cluster center in the embedded feature space is calculated, and the control strategy with the smallest distance is selected as the representative strategy of the cluster to form a preliminary representative strategy set; For the preliminary representative strategy set, the ε-dominance relationship between the strategies is constructed according to the preset performance tolerance threshold ε, and traversal is performed to eliminate the ε-dominated strategies to form the final compression control strategy set.

7. The project decision optimization control method based on knowledge graph according to claim 6 is characterized in that: The control strategy performance graph structure is constructed and input into the preset graph neural network model to obtain the graph embedding feature vector, specifically: Calculate the similarity of the performance index vectors between the initial solution sets of each control strategy to form a strategy performance similarity matrix; Construct a control strategy performance graph based on the strategy performance similarity matrix; Embedding the control strategy performance graph through the graph neural network model and implementing the inter-layer information propagation mechanism; After multiple layers of graph convolution operations, the final graph embedding feature vector of each control strategy node is obtained.

8. The project decision optimization control method based on knowledge graph according to claim 1 is characterized in that: The compression control strategy set is converted into executable control instructions and input into the project implementation system, specifically: By constructing a control instruction mapping template, the compressed control policy set is parsed one by one according to the template format and the corresponding instruction fields are filled in to generate a standardized control instruction set; Convert the standardized control instruction set into the interface call format supported by the project scheduling system; After the conversion is completed, it is pushed to the project scheduling execution engine to generate control instructions, which include scheduling execution control, dynamic resource allocation management, time window execution monitoring and risk response control.

9. A project decision optimization control device based on knowledge graph, characterized in that: include: The knowledge graph construction and verification module is used to build the project knowledge graph and perform consistency verification on the project knowledge graph; The control strategy initial solution set generation module is used to perform semantic analysis on the verified knowledge graph based on the multi-objective effectiveness function to generate the project control strategy initial solution set; The control strategy initial solution set compression and screening module is used to compress and screen the control strategy initial solution set based on the graph neural network method to obtain a compressed control strategy set; The control strategy conversion application module is used to convert the compressed control strategy set into executable control instructions and input them into the project implementation system.

10. An electronic device, characterized in that: include: memory for storing computer programs; A processor, configured to implement the steps of the project decision optimization control method based on knowledge graph as described in any one of claims 1 to 8 when executing the computer program.

Images