Project risk monitoring management method based on general large model

Abstract

The invention relates to a project risk monitoring management method based on a general large model, and the method comprises the specific steps: building a field-enhanced project risk monitoring model based on the general large model of a Transform-Decoder structure, namely, injecting a project management knowledge base and historical risk cases into an Attention module of the general large model; dynamically accessing the field-enhanced project risk monitoring model to a project management system API, and analyzing a logical relationship, resource allocation and progress deviation among project tasks in real time; performing multiple rounds of risk chain reasoning based on the logic relationship among the project tasks, resource allocation and progress deviation and historical risk cases, generating a risk root cause map, and synchronously matching an intervention strategy and a conduction path of each project task; a risk-strategy linkage detector is deployed, monitoring data and environmental parameters of a construction site are collected in real time, a comparison scene in which an intervention strategy is not implemented is simulated, actual benefits of the intervention strategy are quantified, and parameters of a large model are adjusted in real time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence and project management, in particular to a project risk monitoring management method based on a general large model. BACKGROUND

[0002] Current project risk management mainly relies on expert experience library and statistical models, which has the following defects: 1. Knowledge limitation: expert system is difficult to cover the long-tail risks in emerging technical fields; 2. Dynamic response lag: traditional models cannot real-time integrate project progress data; 3. Weak cross-field generalization: special models need to be retrained for different project types.

[0003] In the Chinese invention patent with patent number "CN113255271A", neural networks are used for risk prediction, but a large amount of field annotation data is required; in the American invention patent with patent number "US20220318722A1", a knowledge graph is used to construct a risk network, but it relies on artificial definition rules. Although general large models have made breakthroughs in natural language processing, their direct application in project risk management may produce illusion risks, misjudgments of professional terms, and difficulties in integrating multi-source heterogeneous data. SUMMARY

[0004] In order to solve the problems existing in the prior art, the present application proposes a project risk monitoring management method based on a general large model.

[0005] The technical solution of the present application is as follows: On the one hand, the present application proposes a project risk monitoring management method based on a general large model, the specific steps of which include: A general large model based on Transformer-Decoder structure is used to construct a domain-enhanced project risk monitoring model, that is, project management knowledge base and historical risk cases are injected into the Attention module of the general large model; The domain-enhanced project risk monitoring model is dynamically connected to the project management system API to real-time analyze the logical relationship, resource allocation and progress deviation among various project tasks; Based on the logical relationship, resource allocation and progress deviation among various project tasks, as well as historical risk cases, multi-round risk chain reasoning is performed to generate a risk root cause graph and simultaneously match intervention strategies and the transmission path of various project tasks; A risk-strategy linkage detector is deployed to real-time collect monitoring data and environmental parameters of the construction site, simulate a comparative scenario without implementing intervention strategies, quantify the actual benefits of intervention strategies, and real-time adjust the parameters of the large model; Based on the risk root cause graph and the transmission path of various project tasks, a risk quantization matrix is constructed, the risk values of various project tasks are calculated by combining weight coefficients, and the intervention strategies are dynamically adjusted.

[0006] As a preferred embodiment, the domain enhancement adopts LoRA low-rank adaptive technology, which injects project risk management exclusive parameter matrix into the Attention module of the general large model.

[0007] As a preferred embodiment, the risk chain reasoning adopts RAG architecture to retrieve similar risk cases in real time to correct the output of the general large model.

[0008] As a preferred embodiment, the risk root cause mapping and simultaneous matching of intervention strategies specifically includes: Traverse all end risk nodes in the risk root cause mapping, and match intervention strategies in the strategy library according to the types of each end risk node.

[0009] As a preferred embodiment, the risk-strategy linkage detector includes a risk state tracker, a strategy execution simulator, and an adaptive learner: Risk state tracker: deployed on the project site, continuously collecting physical sensor data, warehouse submission records, and financial system streams; Strategy execution simulator: used to simulate a comparative scenario without implementing intervention strategies; Adaptive learner: automatically adjusts the parameters of the general large model when the monitored risk impact exceeds the preset threshold of the predicted value.

[0010] As a preferred embodiment, the risk quantification matrix includes a three-dimensional structure of impact index, occurrence probability, and response urgency; Wherein, the impact index includes financial loss, delay days, and reputation level, extracted from the risk root cause mapping; The occurrence probability includes the general large model prediction probability, the Monte Carlo simulation calculation prediction value based on historical data, and the similarity case matching degree, and the occurrence probability is obtained by dynamic weighting; The response urgency is classified according to the proximity of the end event burst time to the current time according to the risk root cause mapping, and standardized to a specific numerical value.

[0011] As a preferred embodiment, the step of calculating the risk value of each project task by combining the weight coefficients specifically includes: According to the target project industry, call the fuzzy analytic hierarchy library to obtain the three-dimensional weight coefficients of the impact index, occurrence probability, and response urgency of the risk quantification matrix.

[0012] As a preferred embodiment, in the step of generating a risk root cause mapping and simultaneously matching intervention strategies, when simultaneously matching intervention strategies, generate a feasibility analysis of each intervention strategy, including resource consumption estimation and expected return on investment.

[0013] In another aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method for project risk monitoring and management based on a general large model according to any one of the embodiments of the present application when executing the program.

[0014] In another aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the program is executable on a processor to implement the method for project risk monitoring and management based on a general large model according to any one of the embodiments of the present application.

[0015] The present application has the following beneficial effects: 1. The present application improves the accuracy of project management professional terminology understanding through LoRA low-rank adaptive technology, and breaks through the limitations of traditional static models by real-time analysis of task logic relationship and progress deviation.

[0016] 2. The present application generates a visual graph based on multi-round risk chain reasoning, penetrates the core root cause, and reduces fault analysis time.

[0017] 3. The present application automatically matches the strategy library by traversing the end node, synchronously outputs resource consumption and ROI analysis, avoids invalid investment, and makes the strategy matching more scientific.

[0018] 4. The present application deploys a risk-strategy linkage detector and a closed-loop self-optimization mechanism.

[0019] 5. The present application analyzes and calculates the final risk value from three dimensions through a risk quantization matrix, evaluates the superiority of the intervention strategy, and optimizes the intervention strategy in real time to achieve the optimal intervention strategy. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 The present application is a structural diagram. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0022] It should be understood that the step numbers used herein are only for the convenience of description, and are not limited to the execution sequence of the steps.

[0023] It is to be understood that the terms used in the specification are for the purpose of describing particular embodiments only and are not intended to be limiting of the present application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

[0024] The terms "comprise" and "comprising" mean that the described features, integers, steps, operations, elements, and / or components are present, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0025] The term "and / or" means any combination of one or more of the associated listed items and all possible combinations thereof.

[0026] Embodiment One: Referring to Figure 1 A project risk monitoring management method based on a general large model, the specific steps comprising: A project risk monitoring model enhanced in the field based on a general large model with a Transformer-Decoder structure, that is, injecting a project management knowledge base and historical risk cases into the Attention module of the general large model; In this embodiment, the general large model based on the Transformer-Decoder structure preferably has the following two types: Chinese scenario: Tongyi Qianwen Qwen-7B (with code and long text analysis advantages) Multilingual scenario: LLaMA-213B open source and commercially available, supporting 128K context) Selection basis: Project management needs to handle long sequence data such as contract texts and design documents, and requires the model to have strong context understanding ability.

[0027] The injection method of the project management knowledge base is specifically: according to the knowledge type, it is divided into PMBOK / PRINCE2, FIDIC contract terms and industry standards.

[0028] PMBOK / PRINCE2: Extract risk domain definition, coping strategy text, and embed through dynamic vector; FIDIC contract terms: Analyze the claim triggering conditions and responsibility division rules, and inject through knowledge graph triple conversion: Industry standards: Structured hierarchical indicators, embedded using rule templates.

[0029] The project risk monitoring model enhanced in the field is dynamically connected to the project management system API to real-time analyze the logical relationship between various project tasks, resource allocation, and progress deviation; In this embodiment, the logical relationship between each key task directly affects the transmission path and the scope of influence of the risk: the logical relationship includes logical association, tightness coefficient and resource coupling degree; Logical association determines the timing rules of risk propagation, tightness coefficient quantifies the strength of risk chain reaction, and resource coupling degree predicts the resource risk diffusion range.

[0030] Based on the logical relationship between each project task, the resource allocation and progress deviation, and the historical risk cases, multiple rounds of risk chain reasoning are carried out to generate a risk root cause graph and simultaneously match the intervention strategy and the transmission path of each project task; Deploy the risk-strategy linkage detector to collect monitoring data and environmental parameters in real time, simulate the contrast scene without implementing the intervention strategy, quantify the actual benefits of the intervention strategy, and adjust the parameters of the large model in real time; Based on the risk root cause graph and the transmission path of each project task, a risk quantization matrix is constructed, and the risk value of each project task is calculated by combining the weight coefficient, and the intervention strategy is dynamically adjusted.

[0031] As a preferred embodiment of the present embodiment, the field enhancement adopts LoRA low-rank adaptive technology, and the Attention module of the general large model is injected with a project risk management exclusive parameter matrix.

[0032] In this embodiment, the parameter matrix is As a preferred embodiment of the present embodiment, the risk chain reasoning adopts RAG architecture to retrieve similar risk cases in real time to correct the output of the general large model.

[0033] In this embodiment, the historical risk cases are fused into the general large model using the RAG architecture, and the specific steps are as follows: Step 1, real-time input of project characteristics → FAISS retrieval of Top-3 similar cases; Step 2, splice the case key information to the prompt word; Step 3, the general large model generates risk analysis by referring to the case mode.

[0034] As a preferred embodiment of the present embodiment, the risk root cause graph and the matching of the intervention strategy are as follows: Traverse all end risk nodes in the risk root cause graph, and match the intervention strategy in the strategy library according to the type of each end risk node.

[0035] As a preferred embodiment of the present embodiment, the risk-strategy linkage detector includes a risk state tracker, a strategy execution simulator, and an adaptive learner: Risk state tracker: deployed on the project site, continuously collecting physical sensor data, warehouse submission records, and financial system flow; Strategy execution simulator: used to simulate the contrastive scenario without implementing the intervention strategy; Adaptive learner: automatically adjusts the parameters of the general large model when the monitored risk impact exceeds the preset threshold of the predicted value.

[0036] As a preferred embodiment of the present embodiment, the risk quantification matrix includes an impact index, an occurrence probability, and a response urgency three-dimensional structure; The impact index includes financial loss, delay days, and reputation level, which are extracted from the risk root cause graph; The occurrence probability includes a general large model prediction probability, a prediction value calculated based on historical data through Monte Carlo simulation, and a similarity case matching degree, and the occurrence probability is obtained by dynamically weighting; The response urgency is graded according to the proximity between the estimated outbreak time of the terminal event of the risk root cause graph and the current time, and is standardized to a specific numerical value.

[0037] In the present embodiment, first, the risk root cause graph is preprocessed: Terminal event extraction: traverse all terminal risk nodes (such as delay of construction period, cost overrun, and compliance default) in the graph; Data standardization processing: Financial loss value → converted according to the percentage of total project investment; Time delay value → converted into a multiple of the reference construction period; Quality / reputation loss → quantified by adopting industry standard grading.

[0038] Second, the impact index, occurrence probability, and response urgency three-dimensional independent calculation standard value: Impact index calculation: extract the loss triplets (financial loss value, delay days, and reputation loss value) associated with the terminal event Calculate the impact index standard value according to the triplets:

[0039] In the formula, is the impact index standard value.

[0040] Occurrence probability calculation: dynamically weighted sum of general large model prediction probability, prediction value calculated based on historical data through Monte Carlo simulation, and similarity case matching degree:

[0041] In the formula, is the occurrence probability standard value, is the general large model prediction probability; is the prediction value calculated based on historical data through Monte Carlo simulation; Matching degree for similar cases; 、 、 is a weight coefficient, which is automatically adjusted according to data type completeness.

[0042] Response urgency calculation: first calculate the remaining response window:

[0043] Risk acceleration evaluation:

[0044] wherein, is the risk acceleration; According to the remaining response window and the risk acceleration, formulate grading rules: T1 level: ≤24h and ≥0.1; T2 level: 24h ≤72h or 0.05≤ <0.1; T3 / T4 level: other cases; According to different levels, standardize into specific numerical values according to actual situations.

[0045] Third step, risk value integration: Call the fuzzy analytic hierarchy process library to obtain three-dimensional weight coefficients, and obtain the risk value by weighted calculation:

[0046] wherein, is the risk level, in this embodiment, T1=1.0, T2=0.7, T3=0.4, and T4=0.1.

[0047] As a preferred embodiment of the present embodiment, the step of calculating the risk value of each item task according to the combined weight coefficient is specifically: According to the target project industry, call the fuzzy analytic hierarchy process library to obtain the influence index, occurrence probability and response urgency three-dimensional weight coefficient of the risk quantization matrix.

[0048] As a preferred embodiment of the present embodiment, in the step of generating a risk root cause graph and synchronously matching intervention strategies, when synchronously matching intervention strategies, simultaneously generate a feasibility analysis of each intervention strategy, including resource consumption estimation and expected return on investment.

[0049] In this embodiment, the most suitable intervention strategy is selected according to the actual situation, and the final intervention strategy is selected in combination with the simulation results.

[0050] The above merely illustrates the embodiments of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent process transformation, or direct or indirect application in other related technical fields, which are made by using the content of the present application specification and drawings, are also included in the patent protection scope of the present application.

Claims

1. A general large model-based project risk monitoring management method, characterized in that, The specific steps include: Based on the Transformer-Decoder structure, a general large model is constructed to build a domain-enhanced project risk monitoring model, that is, project management knowledge base and historical risk cases are injected into the Attention module of the general large model; The domain-enhanced project risk monitoring model is dynamically connected to the project management system API to real-time analyze the logical relationship, resource allocation and progress deviation among various project tasks; Based on the logical relationship, resource allocation and progress deviation among various project tasks and historical risk cases, multiple rounds of risk chain reasoning are performed to generate a risk root cause graph and simultaneously match intervention strategies and the transmission path of various project tasks; A risk-strategy linkage detector is deployed to real-time collect monitoring data and environmental parameters of the construction site, simulate a comparative scenario without implementing the intervention strategy, quantify the actual benefits of the intervention strategy, and real-time adjust the parameters of the large model; Based on the risk root cause graph and the transmission path of various project tasks, a risk quantification matrix is constructed, and the risk values of various project tasks are calculated by combining weight coefficients to dynamically adjust the intervention strategy.

2. The project risk monitoring management method based on a general large model according to claim 1, characterized in that, The domain enhancement adopts LoRA low-rank adaptive technology to inject a project risk management exclusive parameter matrix into the Attention module of the general large model.

3. The project risk monitoring management method based on a general large model according to claim 1, characterized in that, The risk chain reasoning adopts the RAG architecture to real-time retrieve similar risk cases to correct the output of the general large model.

4. The project risk monitoring management method based on a general large model according to claim 1, characterized in that, The risk root cause graph and the simultaneous matching of intervention strategies are as follows: All end risk nodes in the risk root cause graph are traversed, and intervention strategies are matched in the strategy library according to the types of each end risk node.

5. The project risk monitoring management method based on a general large model according to claim 1, characterized in that, The risk-strategy linkage detector includes a risk state tracker, a strategy execution simulator, and an adaptive learner: Risk state tracker: deployed on the project site, continuously collects physical sensor data, warehouse submission records, and financial system flow; Strategy execution simulator: used to simulate a comparative scenario without implementing the intervention strategy; Adaptive learner: when the monitored risk impact exceeds the preset threshold of the predicted value, the parameters of the general large model are automatically adjusted.

6. The general large model-based project risk monitoring management method according to claim 1, characterized in that, The risk quantification matrix includes an impact index, a probability of occurrence, and a response urgency three-dimensional structure; The impact index includes financial loss, delay days, and reputation level, which are extracted from the risk root cause graph; The probability of occurrence includes the prediction probability of the general large model, the prediction value calculated based on historical data by Monte Carlo simulation, and the similarity case matching degree, and the occurrence probability is dynamically weighted; The response urgency is classified according to the proximity of the end event burst time to the current time in the risk root cause graph and standardized to a specific value.

7. The general large model-based project risk monitoring management method according to claim 6, characterized in that, The risk value of each project task is calculated by combining weight coefficients, which is as follows: According to the target project industry, the fuzzy analytic hierarchy library is called to obtain the weight coefficients of the impact index, the probability of occurrence, and the response urgency of the risk quantification matrix.

8. The general large model-based project risk monitoring management method according to claim 1, characterized in that, In the step of generating a risk root cause graph and simultaneously matching intervention strategies, the feasibility analysis of each intervention strategy is generated simultaneously, including resource consumption estimation and expected return on investment.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the program to realize the general large model-based project risk monitoring management method in any one of claims 1 to 8.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to realize the general large model-based project risk monitoring management method in any one of claims 1 to 8.

Images