CN-121981529-A - Multi-mode risk signal fusion technology project intelligent prediction method, system, equipment and storage medium

CN121981529ACN 121981529 ACN121981529 ACN 121981529ACN-121981529-A

Abstract

The invention discloses a technological project intelligent prediction method, a system, equipment and a storage medium for multi-mode risk signal fusion, which relate to the technical field of project risk management and artificial intelligence, and the method comprises the following steps: the multi-mode data are fused to generate a unified risk characterization vector, inference is carried out by utilizing a project knowledge graph to identify a risk link, meanwhile, the time sequence evolution trend of the risk index is analyzed to carry out dynamic early warning, experience references are provided by combining with historical similar cases, a prediction report containing risk types, grades, probabilities and countermeasures is generated by combining the multi-dimensional analysis results, and early warning notification is automatically triggered; the invention solves the technical bottleneck of the traditional project risk management system in the aspects of data fusion, association analysis and dynamic prediction, provides comprehensive, intelligent and prospective risk prevention and control capability for scientific projects, and has obvious technical progress and practical value.

Inventors

DONG BIN
YIN YUAN
HUANG YITING
Ai Xuhua
YI CHUNFANG
WANG WEI
CHEN QI
MENG ZHIPENG
MENG QI
LIU KAIJIE
LUO SHENGSI
SHI EN

Assignees

广西电网有限责任公司

Dates

Publication Date: 20260505
Application Date: 20251231

Claims (10)

1. A multi-mode risk signal fusion intelligent technological project prediction method is characterized by comprising the following steps: accessing and organizing multi-source heterogeneous data of the science and technology item from a plurality of data sources; establishing a cross-modal feature extraction network, performing deep feature learning on multi-source heterogeneous data, extracting intra-modal features, and fusing the intra-modal features into a unified risk feature vector; the method comprises the steps of setting a graph reasoning engine, carrying out reasoning by taking a risk feature vector as a clue through a project knowledge graph, and identifying the activation strength of a risk inference link and a risk result; analyzing the historical project state sequence through a time sequence evolution analysis unit, and analyzing the evolution trend of the risk index, if the change rate of the risk index is detected to exceed a preset threshold value, generating a dynamic early warning signal; setting a project similarity calculation engine, carrying out similarity matching on a current project and a historical project case library, calculating comprehensive similarity, and extracting a risk mode and a response experience of the historical project; The comprehensive graph reasoning engine, the time sequence evolution analysis unit and the project similarity calculation engine determine a final risk prediction conclusion through an integrated learning or voting mechanism; And automatically sending an early warning notice to project management personnel based on the risk prediction conclusion.
2. The intelligent prediction method for technological projects fused by multi-modal risk signals according to claim 1, wherein the establishing of a cross-modal feature extraction network, the deep feature learning of multi-source heterogeneous data, the feature extraction in the modalities and the fusion into a unified risk feature vector comprise: Establishing a cross-modal feature extraction network, wherein the cross-modal feature extraction network comprises a text encoder for processing text data, a numerical encoder for processing structured numerical data, a time sequence encoder for processing time sequence data and a cross-modal attention fusion module; Semantic coding is carried out on the input text modal data through a text encoder, and a text feature vector is generated; Carrying out nonlinear transformation coding on the input numerical mode data through a numerical encoder to generate a numerical characteristic vector; performing time sequence dependency relation coding on the input time sequence modal data through a time sequence coder to generate a time sequence feature vector; inputting the text feature vector, the numerical feature vector and the time sequence feature vector into a cross-modal attention fusion module; In the cross-modal attention fusion module, the correlation weights among the feature vectors of different modalities are dynamically calculated through an attention mechanism, and the feature vectors of a plurality of modalities are subjected to self-adaptive weighted fusion according to the correlation weights, so that unified risk feature vectors are generated.
3. The intelligent prediction method for a scientific and technological project fused by multi-modal risk signals according to claim 2, wherein the set-up graph inference engine infers by project knowledge graph with risk feature vectors as clues, and the identification of the activation strength of a risk inference link and a risk result comprises: Constructing and maintaining a project knowledge graph containing three layers of nodes of risk factors, risk symptoms and risk consequences and causal association relations among the nodes; Using a risk feature vector generated by a cross-modal feature extraction network as a query, carrying out semantic matching with risk factor nodes in a project knowledge graph, and determining one or more currently relevant risk factor nodes as an inference starting point; Taking an reasoning starting point as an initial node, and executing multi-hop reasoning based on the graph neural network on the project knowledge graph, wherein the multi-hop reasoning traverses related risk symptom nodes sequentially from risk factor nodes along a causal association path defined in the knowledge graph through a message transmission and neighborhood aggregation mechanism of the graph neural network, and further reasoning to potential risk result nodes; through a multi-hop reasoning process, the activation intensity of the risk result nodes is calculated and output to represent the occurrence probability, and meanwhile, the complete reasoning path from the risk factors to the risk results through the risk symptoms is recorded as an explanatory basis.
4. The intelligent prediction method of technological project with multi-modal risk signal fusion as claimed in claim 3, wherein said analyzing the historical project state sequence by the time sequence evolution analysis unit analyzes the evolution trend of the risk index, if the change rate of the risk index is detected to exceed the preset threshold, generating the dynamic early warning signal comprises: Continuously collecting project state snapshots containing multi-mode data and corresponding knowledge graph states according to preset time intervals to form a history project state sequence; Encoding each item state in the historical item state sequence by using a time sequence encoding model, and extracting a corresponding time sequence feature vector so as to convert the state sequence into a time sequence feature vector sequence; And calculating the change trend of the key risk index through a trend analysis model based on the time sequence feature vector sequence.
5. The intelligent prediction method for technological projects fused by multimodal risk signals according to claim 4, wherein the step of setting a project similarity calculation engine to match the similarity between the current project and a historical project case library, calculating the comprehensive similarity, and extracting the risk mode and the response experience of the historical project comprises the following steps: Extracting multidimensional feature vectors representing the attributes of the items from the items in the current item and the historical item case library; Calculating the comprehensive similarity between the current item and each historical item based on the multidimensional feature vector, wherein the comprehensive similarity is obtained by respectively calculating the similarity of each dimensional feature and then carrying out weighted fusion according to a preset weight coefficient; selecting a preset number of historical items with highest similarity with the current items as reference cases according to the sequencing result of the comprehensive similarity; And extracting the historical risk mode and the countermeasure information from the reference case, and generating a case comparison report by combining the current project characteristics.
6. The intelligent prediction method for technological projects fused by multi-modal risk signals according to claim 5, wherein the output results of the comprehensive graph inference engine, the time sequence evolution analysis unit and the project similarity calculation engine, the final risk prediction conclusion determined by the ensemble learning or voting mechanism, comprise: and integrating and fusing the output results by adopting a preset multi-source information fusion strategy to generate a comprehensive final risk prediction conclusion, wherein the multi-source information fusion strategy dynamically distributes fusion weights for different analysis results according to the characteristics of the analysis methods corresponding to the analysis results or the confidence level of the output, and performs weighted integration.
7. The intelligent prediction method for a scientific project fused by multi-modal risk signals according to claim 6, wherein the automatically sending the early warning notification to the project manager based on the risk prediction conclusion comprises: Based on a risk prediction conclusion generated by multi-source information fusion, automatically generating a structured risk report containing risk types, risk grades, risk occurrence probabilities and coping suggestions, and assisting in revealing a visual chart of risk association and evolution trend; Comparing the risk level or the risk occurrence probability determined in the risk report with a preset early warning triggering condition, and automatically starting an early warning notification process when the triggering condition is met; And sending an early warning notification message to a designated project manager through at least one preset notification channel according to the configured notification strategy, wherein the early warning notification message at least comprises core risk summary information and access guide of a structured risk report.
8. A multi-modal risk signal fusion intelligent prediction system for scientific projects, applying the method as claimed in any one of claims 1 to 7, comprising: The multi-mode data access module is used for accessing and organizing multi-source heterogeneous data of the technological project from a plurality of data sources; The cross-modal feature extraction module is used for establishing a cross-modal feature extraction network, carrying out deep feature learning on multi-source heterogeneous data, extracting intra-modal features and fusing the intra-modal features into a unified risk feature vector; The diagram inference module is used for setting a diagram inference engine, inferring by taking the risk feature vector as a clue through the project knowledge graph, and identifying the activation strength of a risk inference link and a risk result; The time sequence evolution analysis module is used for analyzing the historical project state sequence and the evolution trend of the risk index through the time sequence evolution analysis unit, and generating a dynamic early warning signal if the change rate of the risk index is detected to exceed a preset threshold value; The project similarity calculation module is used for setting a project similarity calculation engine, carrying out similarity matching on the current project and the historical project case library, calculating comprehensive similarity, and extracting a risk mode and a response experience of the historical project; the comprehensive graph reasoning module is used for determining a final risk prediction conclusion through an integrated learning or voting mechanism according to output results of the comprehensive graph reasoning engine, the time sequence evolution analysis unit and the project similarity calculation engine; And the early warning notification module is used for automatically sending an early warning notification to project management personnel based on the risk prediction conclusion.
9. An electronic device, comprising: A memory and a processor; The memory is for storing computer executable instructions, the processor being for executing the computer executable instructions which when executed by the processor implement the steps of the method of any one of claims 1to 7.
10. A computer-readable storage medium, characterized in that it stores computer-executable instructions which, when executed by a processor, implement the steps of the method of any one of claims 1 to 7.

Description

Multi-mode risk signal fusion technology project intelligent prediction method, system, equipment and storage medium Technical Field The invention relates to the technical field of project risk management and artificial intelligence, in particular to a technological project intelligent prediction method, system, equipment and storage medium for multi-mode risk signal fusion. Background Along with the increasing of the complexity of the science and technology projects, the data types related to project management also show the characteristics of diversification and isomerization, and the multi-source heterogeneous information such as structured project management system data, semi-structured progress report text, unstructured team communication records, and time-ordered financial indexes is contained. Conventional project risk management methods face a number of challenges. The method has the following defects that the method only depends on risk rule matching and text similarity detection, lacks deep feature learning capability on multi-mode data, cannot capture implicit association relations between different data sources, is limited to static analysis at a single time point, lacks a dynamic tracking mechanism on risk evolution trend, is difficult to realize early warning, lacks a structured item knowledge representation system, cannot establish an inference link between a risk factor and a risk result, causes insufficient comprehensiveness and association of risk analysis, does not have intelligent retrieval and comparison capability of historical item cases, and cannot utilize the historical experience of similar items to assist the risk identification of current items. Disclosure of Invention In view of the above problems, the present invention provides a method, a system, a device and a storage medium for intelligent prediction of a multi-modal risk signal fusion technology project. Therefore, the technical problem solved by the invention is to provide the intelligent prediction method for the scientific and technological project, which can fuse multi-source heterogeneous data, establish a deep risk association relationship and realize dynamic evolution analysis, so as to solve the problems of incomplete early warning and insufficient association analysis in the traditional risk management. In order to solve the technical problems, the invention provides the following technical scheme: In a first aspect, the present invention provides a method for intelligently predicting a technological project by multi-modal risk signal fusion, including: accessing and organizing multi-source heterogeneous data of the science and technology item from a plurality of data sources; establishing a cross-modal feature extraction network, performing deep feature learning on multi-source heterogeneous data, extracting intra-modal features, and fusing the intra-modal features into a unified risk feature vector; the method comprises the steps of setting a graph reasoning engine, carrying out reasoning by taking a risk feature vector as a clue through a project knowledge graph, and identifying the activation strength of a risk inference link and a risk result; analyzing the historical project state sequence through a time sequence evolution analysis unit, and analyzing the evolution trend of the risk index, if the change rate of the risk index is detected to exceed a preset threshold value, generating a dynamic early warning signal; setting a project similarity calculation engine, carrying out similarity matching on a current project and a historical project case library, calculating comprehensive similarity, and extracting a risk mode and a response experience of the historical project; The comprehensive graph reasoning engine, the time sequence evolution analysis unit and the project similarity calculation engine determine a final risk prediction conclusion through an integrated learning or voting mechanism; And automatically sending an early warning notice to project management personnel based on the risk prediction conclusion. As a preferable scheme of the intelligent prediction method of the technological project with multi-mode risk signal fusion, the intelligent prediction method comprises the following steps: Establishing a cross-mode feature extraction network, performing deep feature learning on multi-source heterogeneous data, extracting intra-mode features, and fusing into a unified risk feature vector, wherein the method comprises the following steps: Establishing a cross-modal feature extraction network, wherein the cross-modal feature extraction network comprises a text encoder for processing text data, a numerical encoder for processing structured numerical data, a time sequence encoder for processing time sequence data and a cross-modal attention fusion module; Semantic coding is carried out on the input text modal data through a text encoder, and a text feature vector is generated; Carrying out nonlinear tran