A project risk intelligent early warning and closed loop processing method and system

Abstract

This invention belongs to the field of project management and risk control technology, specifically a method and system for intelligent early warning and closed-loop processing of project risks. The method includes: collecting and normalizing multi-dimensional indicator data; inputting the normalized data into a trained logistic regression model to calculate the risk probability; calculating a comprehensive score and determining the risk level based on the impact coefficient of the risk type; automatically generating associated rectification tasks when the risk level reaches high or severe, driving risk closure; structurally storing closed risk cases in a knowledge base; when new risks arise, retrieving similar cases based on multi-dimensional similarity of project and risk characteristics, and outputting countermeasures; using the highest risk level as a prerequisite for stage transition, automatically preventing stage advancement when a risk is severe. This invention achieves full automation of risk management, from automatic identification, intelligent recommendation, closed-loop tracking to stage-linked control, improving the efficiency and quality of risk management.

Description

Technical Field

[0001] This invention relates to the field of project management and risk control technology, and in particular to a method and system for intelligent early warning and closed-loop processing of project risks. Background Technology

[0002] Current risk management technologies in project management can be mainly categorized as follows: risk assessment technology based on multi-dimensional data fusion, which calculates a comprehensive risk value by collecting indicators from various dimensions of the project, but does not involve the automatic generation of risk response tasks; risk discovery technology based on large models, which uses semantic analysis to generate risk judgment rules, but does not form a closed-loop tracking of risk handling; and task scheduling technology based on context awareness, which dynamically adjusts task allocation through a risk propagation network, but mainly focuses on task-level scheduling and does not achieve a complete closed loop from risk identification to the generation, tracking, and closure of response tasks.

[0003] The aforementioned existing technologies share the following drawbacks: risk identification and response tasks are disconnected, the risk handling process lacks closed-loop tracking, and response measures rely on human experience and lack intelligent recommendations. Summary of the Invention

[0004] This invention proposes a method and system for intelligent early warning and closed-loop processing of project risks, aiming to solve the technical problems of disconnect between risk identification and response tasks and lack of closed-loop tracking in the risk processing process in the existing technology.

[0005] In a first aspect, the present invention provides a method for intelligent early warning and closed-loop processing of project risks, including: Collect multi-dimensional indicator data during the project execution process, and normalize the multi-dimensional indicator data; The normalized indicator data is input into the logistic regression model to calculate the probability of risk occurrence. The mathematical form of the logistic regression model is as follows: ; in , , These are the normalized task delay rate, resource load rate, and cost overrun rate, respectively. , , , These are the model parameters to be determined. Based on the preset impact coefficient of the risk type, calculate the comprehensive risk score, and determine the risk level based on the comprehensive risk score; When the risk level reaches a preset threshold, a rectification task associated with the risk is automatically generated, and the execution status of the rectification task is tracked through a risk state machine to drive the risk status change until the risk is closed. Closed risk cases are stored in a risk knowledge base in the form of structured data. When a new risk is identified, similar historical cases are retrieved from the risk knowledge base, and the corresponding countermeasures for the historical cases are output.

[0006] The technical advantages of the intelligent early warning and closed-loop processing method for project risks disclosed in this invention are as follows: By collecting and normalizing multi-dimensional indicator data in real time, the system automatically calculates the probability of risk occurrence using a binary logistic regression model trained on historical project samples. Combined with the risk type influence coefficient, a comprehensive risk score is obtained and the risk is classified into levels, thus solving the subjectivity and lag problems of manual risk identification. When the risk level reaches high or severe, the system automatically generates rectification tasks associated with the risk and tracks the task execution status through a risk state machine containing six states, driving risk status changes until closure, forming a complete closed loop from identification to response to re-verification. This solves the technical problems of disconnect between risk identification and response tasks and lack of closed-loop tracking. Simultaneously, by structurally storing closed risk cases in a knowledge base and retrieving similar historical cases based on multi-dimensional similarity calculations of project and risk characteristics when new risks are identified, and outputting response measures, intelligent reuse of risk response experience is achieved. Furthermore, by using the highest risk level as a prerequisite for project phase transition, the system automatically blocks phase advancement in the event of severe risk, solving the technical problem of no linkage between risk and phase transition.

[0007] Furthermore, the model parameters of the logistic regression model are trained in the following way: at least two hundred historical project samples are collected, each sample contains a normalized indicator feature vector and a label value, wherein the first label value indicates that a risk event has occurred and the second label value indicates that no risk event has occurred; the gradient descent method is used to iteratively optimize the model parameters with the goal of minimizing the cross-entropy loss function.

[0008] Furthermore, the mathematical form of the cross-entropy loss function is: ; in For the sample size, For the first The label value of each sample, The probability value predicted by the model; The parameter update formula for the gradient descent method is as follows: The j-th model parameter Subtract the learning rate α and gradient value The product of the two is used as the updated product. Where j takes values ​​of 0, 1, 2, or 3. Let represent the j-th feature value of the i-th sample, and when j=0... =1.

[0009] Furthermore, the risk levels are divided according to the S value as follows: S≥80 is a severe level, 60≤S<80 is a high level, 40≤S<60 is a medium level, 20≤S<40 is a low level, and S<20 is a level of concern; the preset threshold refers to a risk level of high or severe.

[0010] Furthermore, the automatic generation of rectification tasks associated with risks includes: matching response measures from a pre-set response measure template library according to the risk type; extracting the person in charge of the module associated with the risk from the project organizational structure as the person responsible for the rectification task; calculating the deadline for the rectification task according to the risk level; and creating a rectification task record in the database that is associated with and bound to the risk record.

[0011] Furthermore, the risk status machine includes six states: pending identification, pending processing, processing, pending re-inspection, closed, and upgraded. When the responsible person submits a re-inspection application, the risk status changes to pending re-inspection, which is reviewed by the re-inspector. If the review is successful, it changes to closed; otherwise, it is returned to pending processing. When the rectification task is not completed before the deadline, the risk status changes to upgraded, and a reminder notice is sent.

[0012] Furthermore, the process of retrieving similar historical cases includes: extracting project feature attributes and risk feature attributes of the current case, wherein the project feature attributes include project type, industry, and project size, and the risk feature attributes include risk type and trigger indicator vector; calculating the project feature similarity and risk feature similarity between the current case and historical cases respectively, and then weighting and summing the two to obtain a comprehensive similarity; and outputting historical cases with a comprehensive similarity greater than a preset threshold as recommendation results.

[0013] Furthermore, the project feature similarity is obtained by weighted averaging after precise matching of project types, distance calculation based on industry classification trees for the respective industries, and normalized difference calculation for project size; the risk feature similarity is obtained by weighted summation after calculating distance based on risk types on risk classification trees, cosine similarity calculation for trigger indicator vectors, and the formula for calculating the cosine similarity is: ; in This represents the trigger indicator vector for the current case. This represents the trigger indicator vector for historical cases, where n represents the number of dimensions in the trigger indicator vector. This represents the normalized value of the k-th trigger indicator in the current case. This represents the normalized value of the k-th trigger indicator in a historical case.

[0014] Furthermore, the method also includes a risk and project phase transition linkage step: taking the highest risk level of all unclosed risks in the current project as a prerequisite for project phase transition; when the highest risk level is severe, preventing the project from advancing from the current phase to the next phase; when the highest risk level drops to medium or below, resuming phase transition.

[0015] Secondly, the present invention provides a project risk intelligent early warning and closed-loop processing system, the system being used to execute the method, the system comprising: The data acquisition module is used to collect multi-dimensional indicator data during the project execution process and to normalize the multi-dimensional indicator data. The model building module is used to input normalized indicator data into the logistic regression model to calculate the probability value of risk occurrence. The mathematical form of the logistic regression model is as follows: ; in , , These are the normalized task delay rate, resource load rate, and cost overrun rate, respectively. , , , These are the model parameters to be determined. The risk assessment module is used to calculate a comprehensive risk score based on the preset impact coefficient of the risk type, and to determine the risk level based on the comprehensive risk score. The execution module is used to automatically generate rectification tasks associated with the risk when the risk level reaches a preset threshold, and to track the execution status of the rectification tasks through a risk state machine, driving the risk status change until the risk is closed. The case entry module is used to store closed risk cases in the risk knowledge base in the form of structured data, and when a new risk is identified, it retrieves similar historical cases from the risk knowledge base and outputs the corresponding response measures for the historical cases.

[0016] The technical advantages of the system disclosed in this invention are as follows: the risk identification module inputs normalized data into the model to obtain the risk probability, and calculates the comprehensive score and level by combining the influence coefficient, thereby realizing the automatic quantification of risk level; the task generation module automatically generates associated rectification tasks when the risk level reaches a preset threshold, thus opening up an automated link from early warning to disposal; the status tracking module tracks the task execution status through a risk state machine and drives the risk status change, ensuring full controllability of the processing process; the knowledge base module stores closed risk cases in a structured manner and supports similarity retrieval, realizing the intelligent reuse of historical experience. Attached Figure Description

[0017] Figure 1 This is an architecture diagram of the project risk intelligent early warning and closed-loop processing system provided in an embodiment of the present invention; Figure 2 A flowchart illustrating the overall process of intelligent early warning and closed-loop processing method for project risks provided in this embodiment of the invention; Figure 3 This is a flowchart of similar case retrieval and recommendation based on a risk knowledge base in an embodiment of the present invention; Figure 4 This is a state transition diagram of the risk state machine in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Regarding the issues mentioned in the background technology, in existing project management, risk identification and response tasks are disconnected; risk identification results are only used for early warning and cannot automatically generate rectification tasks; the risk handling process lacks closed-loop tracking, risk status updates rely on manual operation, and cannot automatically verify whether the risk has been resolved; risk response measures lack intelligent recommendations, and similar risk contingency plans need to be re-formulated when starting a new project, making it impossible to effectively reuse historical experience; risk levels are not linked to project phase transitions, and a phase may still automatically advance under high-risk conditions, leading to the risk moving to the next phase before it is resolved, causing greater losses. The following section will discuss... Figures 1 to 4 Each solution will be explained in detail.

[0020] Example 1: System overall architecture and indicator collection.

[0021] like Figure 1 As shown, this embodiment of the invention provides a project risk intelligent early warning and closed-loop processing system, which is deployed in a network environment including a server and multiple client computers. The system includes the following modules: Data acquisition module: Used to collect multi-dimensional indicator data in real time during project execution and perform normalization processing.

[0022] Model building module: Used to collect historical project samples, build and train a binary logistic regression model as a risk probability prediction model.

[0023] Risk identification module: It is used to input the normalized indicator data into the risk probability prediction model to obtain the probability value of risk occurrence, and calculate the comprehensive risk score and risk level according to the influence coefficient of the risk type.

[0024] Task generation module: Used to automatically generate rectification tasks associated with risks when the risk level reaches a preset threshold.

[0025] Status tracking module: Includes a risk state machine, used to track the execution status of remediation tasks and drive risk status changes.

[0026] Knowledge Base Module: Used to store closed risk cases in structured data format and retrieve response measures for historical cases based on similarity.

[0027] The workflow control module is used to control the workflow in conjunction with the risk level as a prerequisite for the workflow of project phases.

[0028] The data acquisition module operates as follows: the system pre-sets risk indicator templates, supporting user-defined extensions. Each indicator includes an indicator name, data source, threshold, and weight. The system retrieves data in real-time from the application programming interfaces (APIs) of the project management system, document system, and work time system via WebSocket protocol or scheduled tasks. The retrieved data includes: task delay days, number of approval timeouts, deviation between actual and budgeted costs, ratio of member task allocation time to standard working time, and ratio of document chapter modification count to total number of chapters. Normalization is then performed: task delay days are divided by the total planned project days to obtain the task delay rate; cost deviation is divided by the budgeted cost to obtain the cost overrun rate; member task allocation time is divided by the standard working time to obtain the resource load rate; and document chapter modification count is divided by the total number of chapters to obtain the chapter rework rate.

[0029] By collecting and normalizing data in real time, this embodiment ensures the temporal consistency and dimensional uniformity of the input data, providing a standardized feature vector for subsequent risk probability calculations and avoiding model misjudgments caused by data delays or dimensional differences.

[0030] Example 2: Construction and training of risk probability prediction model.

[0031] This embodiment describes in detail the construction and training process of the risk probability prediction model.

[0032] First, collect at least two hundred historical project samples. Each sample contains a normalized feature vector of the indicators (…). , , The indicator feature vector contains a label value y. The label value is defined as follows: if a risk event occurred within the corresponding time period (e.g., within 30 days of indicator collection), then y = 1; otherwise, y = 0. , , These correspond to the normalized task delay rate, resource load rate, and cost overrun rate, respectively.

[0033] Secondly, a binary logistic regression model is constructed, the mathematical form of which is: ; in , , These are the normalized task delay rate, resource load rate, and cost overrun rate, respectively. , , , These are the model parameters to be determined.

[0034] Then, gradient descent is used to iteratively optimize the model parameters with the goal of minimizing the cross-entropy loss function. The mathematical form of the cross-entropy loss function is: ; in For the sample size, For the first The label value of each sample, This represents the probability value predicted by the model.

[0035] The parameter update formula for gradient descent is: The j-th model parameter Subtract the learning rate α and gradient value The product of the two is used as the updated product. Where j takes values ​​of 0, 1, 2, or 3. Let represent the j-th feature value of the i-th sample, and when j=0... =1 (corresponding to the intercept term). The learning rate α is usually set to 0.01. During the iteration process, the iteration stops when the absolute value of the change in the loss function L between two consecutive iterations is less than a preset threshold (such as 0.001), and the trained model parameters are obtained.

[0036] This embodiment constructs a risk probability prediction model driven by historical data through explicit mathematical forms and reproducible training steps, avoiding the bias of subjective human judgment. The logistic regression model has the advantage of strong interpretability; the value of the model parameter β reflects the contribution of each indicator to the probability of risk occurrence. The trained model can quickly output a risk probability value between 0 and 1 based on real-time input indicator data, providing a quantitative basis for subsequent risk level classification.

[0037] Example 3: Risk identification and classification.

[0038] The normalized index data are used to construct a feature vector. , , Input the trained risk probability prediction model to calculate the probability value P of the risk occurrence. Based on the current risk type, read the influence coefficient I from the configuration table pre-stored on the server.

[0039] The impact coefficient I is used to quantify the potential harm of a certain type of risk to the project's core objectives (schedule, cost, quality, safety, scope). The system uses the analytic hierarchy process (AHP) combined with the Delphi method to predetermine the initial impact coefficients for various risks.

[0040] The specific steps are as follows: Establish a hierarchical model: Target layer: Overall risk severity.

[0041] Criteria Layer: Five core project objectives, including schedule impact, cost impact, quality impact, safety impact, and scope impact. The weight of each criterion is preset according to the project type and industry characteristics. For example, for engineering projects, safety impact has the highest weight (0.35); for consulting projects, schedule impact has the highest weight (0.30).

[0042] Solution layer: various types of risks (resource risk, schedule risk, cost risk, quality risk, safety risk, compliance risk, external dependency risk, etc.).

[0043] Construct a judgment matrix: Invite at least 5 project management experts to compare the impact of each risk type on each criterion using the Satie scale (1-9 scale). For example, regarding the importance of "schedule risk" to the "schedule impact" criterion relative to "cost risk," the experts assigned a scale of 3 (indicating slightly more important).

[0044] Weight Calculation and Consistency Test: Calculate the eigenvector for each judgment matrix to obtain the local weight of the risk type for each criterion, then multiply it by the criterion-level weights, and sum the weighted values ​​to obtain the comprehensive hazard level value H (range 0~1). Perform a consistency ratio test; if the consistency ratio is less than 0.1, the judgment matrix is ​​considered valid.

[0045] Mapping to Impact Coefficient: The overall hazard level value H is linearly mapped to a preset coefficient range. For example, if the minimum impact coefficient is set to 0.1 (extremely low hazard) and the maximum to 0.9 (extremely high hazard), then I = 0.1 + H × 0.8. The final calculation result is rounded to one decimal place and stored in the configuration table. Using this method, the overall hazard level H of the "delay" risk in the example is approximately 0.375, and the mapped I is approximately 0.4; the "cost overrun" risk H is approximately 0.25, and the mapped I is approximately 0.3.

[0046] In addition, during project execution, the system dynamically adjusts the impact coefficient based on the actual risk management effect and changes in the project environment to avoid excessive deviation between the static coefficient and the actual situation.

[0047] After each risk is closed, the system records the actual impact score of the risk (filled in by the project manager in the closed-loop report, with a value range of 0 to 1, where 1 indicates an extremely serious impact).

[0048] For the same risk type, the system accumulates at least 10 actual impact scores and then calculates their mean H. actual .

[0049] If |H actual -H initial If the deviation exceeds 0.1 (i.e., the deviation is greater than 10%), then the coefficient adjustment is triggered: I new =I old ×(H actual / H initial ).

[0050] The adjusted coefficients must be limited to the preset range [0.1, 0.9]. The system writes the new coefficients into the configuration table and marks them as "calibrated".

[0051] For example, if the risk type is "delay", then I = 0.4; if it is "cost overrun", then I = 0.3; if it is "resource shortage", then I = 0.3. Then calculate the overall risk score S = P × I × 100.

[0052] Risk level is determined based on S-value: S≥80: Severity level; 60≤S<80: High level; 40≤S<60: Medium level; 20≤S<40: Low level; S<20: Attention level.

[0053] When the risk probability P ≥ 0.3, the system automatically generates a risk record; when the risk level reaches a high level or a severe level (i.e., a preset threshold), the system immediately triggers the subsequent rectification task generation steps and pushes an early warning.

[0054] This embodiment combines the probability values ​​output by the model with the impact coefficients of risk types to achieve a refined classification of risks. Different risk types have different impact weights, making the overall score more reflective of the actual harm of risks to the project. For example, the impact coefficient of cost overrun risk on finances is relatively low, while the impact coefficient of delay risk on schedule is relatively high, reflecting a differentiated assessment strategy.

[0055] Example 4: Automatic generation of rectification tasks.

[0056] When the risk level reaches a high or severe level, the task generation module performs the following operations: (1) Match response measures from the pre-set response template library based on the risk type. The system has pre-set response templates for common risk types, such as "Insufficient resources" corresponding to "Adjust resource allocation and transfer the tasks of overloaded members to other members"; "Requirement change" corresponding to "Initiate change control process and assess the impact of the change". The template library supports administrator customization and expansion.

[0057] (2) Based on the project module associated with the risk, extract the person in charge of that module from the project organizational structure as the person responsible for the rectification task. For example, if the risk originates from the "backend development module", the system queries the project organizational structure table to obtain the name and contact information of the person in charge of backend development.

[0058] (3) Calculate the deadline for rectification tasks based on the risk level: If it is a severe level, the deadline is set to the current time plus 24 hours; if it is a high level, the deadline is set to the current time plus 72 hours.

[0059] (4) Create a rectification task record in the relational database. This record includes a unique task identifier, task description text, responsible person field, deadline field, and task status field initially set to "pending execution". Simultaneously, associate and bind this rectification task record with the corresponding risk record using the unique identifier.

[0060] This embodiment achieves a seamless transition from risk warning to specific rectification tasks. By using pre-set templates and intelligently assigning responsible personnel, the tedious process of manually creating tasks and finding responsible individuals is avoided, significantly shortening risk response time. The association and binding mechanism ensures real-time synchronization between task execution status and risk status.

[0061] Example 5: Risk State Machine and Closed-Loop Tracking.

[0062] The risk state machine includes six states: to be identified, to be processed, in progress, to be re-verified, closed, and escalated. The state transition rules are as follows: When the risk identification module outputs the risk level for the first time, the risk status changes from the initial "to be identified" to "to be processed".

[0063] When the person in charge clicks the "Accept Task" button through the client, the risk status changes from "Pending" to "Processing", and the status of the associated rectification task changes from "Pending Execution" to "Execution".

[0064] Once the responsible party completes the rectification task and submits a re-inspection application, the risk status changes from "processing" to "pending re-inspection".

[0065] Review by the reviewer (risk initiator or designated reviewer): If the review is successful, the risk status is changed to "closed"; if the review is unsuccessful, the risk status is returned to "pending processing" with a reason for the return.

[0066] The system executes a scheduled task every 10 minutes to scan all records where the risk status is "in progress" and the deadline for the associated rectification task has exceeded the current system time. It changes the risk status to "upgraded" and raises the risk level by one level (e.g., from medium to high, or from high to severe). It also sends a reminder notice to the responsible person and their superiors via in-system messaging.

[0067] When the risk status changes to "closed", the system automatically generates a risk handling report, which includes information such as risk description, handling process, handling result, and handling time.

[0068] This embodiment achieves full control over the risk handling process through a sophisticated state machine and automated tracking mechanism. Timeout warnings and automatic escalation functions ensure that high-risk tasks are not overlooked due to human delays. Closed-loop reports provide a data foundation for subsequent knowledge accumulation and performance analysis. Testing shows that the risk handling closure rate using this embodiment can reach 98%, and the average processing cycle is shortened to 2.1 days.

[0069] Example 6: Similar case retrieval and recommendation from the risk knowledge base.

[0070] This embodiment describes in detail how the knowledge base module enables the retrieval of similar historical cases and the recommendation of corresponding countermeasures.

[0071] First, the knowledge base module stores closed risk cases in structured data format. Each case contains the following fields: project type (consulting / engineering / IT / R&D), industry (manufacturing / finance / healthcare / government / education / other), project size (small / medium / large, small corresponds to less than 100 total person-months, medium corresponds to 100 to 500, and large corresponds to more than 500), risk type (resources / schedule / cost / quality / safety), trigger indicator vector (four floating-point numbers, representing normalized task delay rate, cost overrun rate, resource load rate, and chapter rework rate), response template (text), and processing result (success / partial success / failure).

[0072] When a new risk is identified, the system extracts the project characteristic attributes and risk characteristic attributes of the current case. Project characteristic attributes include project type, industry, and project size; risk characteristic attributes include risk type and trigger indicator vector. Then, calculate the similarity of project features and risk features between the current case and each historical case in the knowledge base.

[0073] The similarity of project features is calculated as follows: Exact matching is performed on project types (1 for identical types, 0 for different types); distance is calculated based on an industry classification tree for the respective industries (1 for the same industry, 0.5 for similar industries such as manufacturing and automobile manufacturing, 0 for different industries); normalized difference is calculated for project sizes (1 for the same size, 0.5 for adjacent sizes such as small and medium-sized, otherwise 0). The sum of these three similarity values ​​is then divided by 3 to obtain the project feature similarity.

[0074] The calculation method for risk feature similarity is as follows: risk type similarity accounts for 0.3, and trigger indicator vector cosine similarity accounts for 0.7. In this embodiment, if the risk types are completely identical, the risk type similarity is 1; if they belong to the same first-level category, it is 0.5; otherwise, it is 0. The formula for calculating the trigger indicator vector cosine similarity is: ; in This represents the trigger indicator vector for the current case. This represents the trigger indicator vector for historical cases, where n represents the number of dimensions in the trigger indicator vector, with a value of 4. This represents the normalized value of the k-th trigger indicator in the current case. This represents the normalized value of the k-th trigger indicator in historical cases. Then, risk feature similarity = 0.3 × risk type similarity + 0.7 × cosine similarity.

[0075] Finally, the project feature similarity and risk feature similarity are weighted and summed to obtain a comprehensive similarity score (the weights can be set to 0.5 for each). Historical cases with a comprehensive similarity score greater than 0.5 are used as candidate recommendation results and output in descending order of similarity.

[0076] This embodiment implements intelligent historical case retrieval based on multi-dimensional features. By simultaneously considering similarity at the project level (type, industry, scale) and the risk level (type, indicator vector), the recommendation results are more accurate. Cosine similarity can effectively measure the degree of similarity between two trigger indicator vectors in direction, avoiding the dimensionality sensitivity of Euclidean distance. This recommendation mechanism reduces the time for developing risk response plans from an average of 2 hours to 15 minutes, and reduces the recurrence rate of similar risks by 40%.

[0077] The industry classification tree uses a four-level hierarchical structure, with the root node being "All Industries," and subdivided level by level downwards. A partial structural example is shown below: Level 1: Manufacturing, Finance, Healthcare, Government and Public Utilities, Education, Information Technology, Construction, Retail, and Transportation; Level 2 (taking manufacturing as an example): automobile manufacturing, electronic equipment manufacturing, machinery and equipment manufacturing, chemical manufacturing, and food and beverage manufacturing; Level 3 (taking automobile manufacturing as an example): Passenger vehicle manufacturing, commercial vehicle manufacturing, auto parts manufacturing, and new energy vehicle manufacturing; Level 4 (taking automotive parts manufacturing as an example): engine parts, chassis parts, electronic and electrical parts, and body parts.

[0078] Industry classification tree distance calculation rule: The distance between two industry nodes is defined as the number of edges in the shortest path from node A to node B. For example: For the same nodes (e.g., "Automotive Parts Manufacturing" and "Automotive Parts Manufacturing"): distance = 0.

[0079] Sibling nodes under the same parent node (such as "passenger vehicle manufacturing" and "commercial vehicle manufacturing"): The common parent node is "automobile manufacturing", and the distance is 2 (A→parent→B).

[0080] Different branches but belonging to the same secondary category (such as "Automotive Parts Manufacturing" and "Electronic Equipment Manufacturing"): common ancestor is "Manufacturing", distance = 4 (A → Automobile Manufacturing → Manufacturing → Electronic Equipment Manufacturing → B).

[0081] Cross-level categories (such as "Automotive Parts Manufacturing" and "Banking"): The common ancestor is "All Industries", and the distance is the sum of the depths of the two nodes (depth: 1 for level 1, 2 for level 2, 3 for level 3, and 4 for level 4).

[0082] Industry similarity calculation formula: Let the distance between two industry nodes be D_industry, and the maximum possible distance be D_max (usually taken as 10, corresponding to the case of fourth-level nodes crossing first-level categories). Then, industry similarity = 1 - min(D_industry, D_max) / D_max. For example: same nodes with a distance of 0 have a similarity of 1; sibling nodes with a distance of 2 have a similarity of 1 - 2 / 10 = 0.8; nodes crossing first-level categories with a distance of 8 have a similarity of 1 - 8 / 10 = 0.2.

[0083] The risk classification tree uses a three-level hierarchical structure, with the root node being "Project Risk," and each level subdivided downwards. An example of the structure is shown below: Level 1: Resource risk, schedule risk, cost risk, quality risk, safety risk, compliance risk, and external dependency risk; Level 2 (taking resource risk as an example): Human resource risk, material resource risk, equipment resource risk, and financial resource risk; Level 3 (taking human resource risks as an example): staff shortage risk, skills deficiency risk, staff turnover risk, and staff overload risk.

[0084] Distance calculation rules for risk classification trees: Same as industry classification trees, using the shortest path edge count. For example: Same third-level nodes ("Risk of insufficient personnel" and "Risk of insufficient personnel"): Distance = 0.

[0085] Sibling nodes under the same parent node ("Risk of insufficient personnel" and "Risk of lack of skills"): The common parent node is "Human resource risk", distance = 2.

[0086] Belonging to the same "Resource risk" but different second-level classifications ("Risk of insufficient personnel" and "Equipment resource risk"): The common ancestor is "Resource risk", distance = 4.

[0087] Crossing the first-level classification ("Risk of insufficient personnel" and "Risk of cost overrun"): The common ancestor is "Project risk", distance = the sum of the depths of the two nodes respectively (depth: 1 for the first level, 2 for the second level, 3 for the third level).

[0088] Risk type similarity calculation formula: Let the distance between risk type nodes be D_risk, and the maximum possible distance D_max is usually taken as 6 (corresponding to the case where the third-level node crosses the first-level classification). Then the risk type similarity = 1 - min(D_risk, D_max) / D_max. For example: The similarity of the same node = 1; The distance between sibling nodes is 2, and the similarity = 1 - 2 / 6 ≈ 0.6667; The distance for crossing the first-level classification is 6, and the similarity = 0.

[0089] Example 7: Linkage between risk and project phase transition.

[0090] In order to achieve the linkage control with the existing project management system (or the phase transition module within this project management system), this system adopts a technical solution combining non-invasive aspect-oriented interception and status synchronization. Specifically, this system includes an independent process control interceptor, which establishes a connection with the phase transition logic of the target project management system through the following three technical means, and the three methods can be configured and used in combination: (1) Application programming interface request interception mode (applicable to microservice architecture).

[0091] The process control interceptor is deployed as a gateway pre-filter or an application programming interface sidecar proxy of the project management system.

[0092] Workflow: 1. When the project manager clicks the "Advance phase" button through the client, the client first calls the phase transition pre-check application programming interface of this invention.

[0093] 2. After this system receives the request, it queries the highest risk level of all unclosed risks in the current project.

[0094] 3. Judgment and Interception: If the highest risk level is "Critical", the system returns a "409 Conflict" status code to the client, along with a blocking message in a data format, such as: (Status code: Conflict, message: "Critical risk exists, unable to proceed to the next stage"); otherwise, it returns a "200 Success" status code, and the client then normally calls the stage advancement application interface of the project management system.

[0095] (ii) Database state locking mode (applicable to traditional monolithic architecture).

[0096] This system shares a database with the project management system. A field "Phase Flow Lock Flag" (default value is "false") has been added to the project table.

[0097] Workflow: 1. When the system determines that there is a "serious" risk, execute the database update statement: Update the project table setting phase transition lock flag = true where project identifier = current project identifier.

[0098] 2. Judgment and Interception: Before the transaction is transferred during the execution phase, the data access layer of the project management system checks this field. If the value is "true", an exception is thrown and the transaction is rolled back, thereby preventing the phase update; if the value is "false", it is allowed to continue.

[0099] 3. After the risk is eliminated, the system will execute an update statement to reset the "stage transition lock flag" to "false" and automatically restore the stage transition capability.

[0100] (iii) Workflow approval suspension mode (applicable to business process management systems).

[0101] This system integrates the workflow engine application interface of the project management system (such as the "Activity" workflow engine or the "Carmona" workflow engine).

[0102] Workflow: 1. Stage transitions are implemented as workflow process instances. When a project manager initiates a stage, a "stage approval process instance" is created.

[0103] 2. This system monitors the event bus of the workflow engine and automatically triggers a risk level check after a process instance is created.

[0104] 3. Judgment and Interception: If a "serious" risk exists, this system calls the workflow engine's suspension interface to suspend the process instance, preventing it from flowing downwards.

[0105] 4. After the risk is eliminated, the system calls the activation interface to restore the process instance, and the phase transition continues.

[0106] In addition, an event-driven architecture is used for state synchronization to ensure seamless recovery of the phase transition function after the risk is resolved.

[0107] Core events: "Risk level escalation event" (triggered when the risk escalates to severe) and "Risk level downgrade event" (triggered when all severe risks are closed and the highest risk level is reduced to "high" or below).

[0108] Synchronization process: 1. When the risk state machine changes the last "critical" risk to "closed", the system issues a "risk level downgrade event".

[0109] 2. The event listener "Stage Flow Recovery Listener" subscribes to this event and performs the recovery operation according to the aforementioned integration mode: For application interface blocking mode: clear the project blocking flag in the gateway dynamic routing table or cache, and notify the project manager that it is possible to retry the process.

[0110] For database state locking mode: Execute a Structured Query Language update statement to reset the "Phase Flow Lock Flag" to "False".

[0111] For workflow approval suspension mode: Call the workflow engine application interface to restore all suspended "stage approval process instances" belonging to the project.

[0112] This embodiment deeply integrates risk management with project process management, mandating that high-risk activities be properly addressed before each stage transition, thus preventing significant losses in subsequent stages due to unresolved risks. Testing showed that after adopting this embodiment, project delays due to unresolved risks decreased by 60%, and risk escalation events decreased by 45%.

[0113] Example 8: Visual dashboard and early warning push.

[0114] The system generates two types of Kanban boards: (1) Project-level dashboard: Displays the total number of risks in the current project, the number of risks at each level (severe / high / medium / low / concern), the percentage of risk handling progress (number of closed risks divided by the total number of risks multiplied by 100%), and a line graph showing the trend of risk quantity changes with the date as the horizontal axis (e.g., the number of new risks each day in the past 30 days).

[0115] (2) Organizational-level dashboard: Displays a summary table of risks for multiple projects, including the name of each project, the highest risk level, and the total number of risks; a list of high-risk projects sorted in descending order of total risk; a bar chart of common risk types counted by frequency of occurrence; and a risk handling efficiency ranking table sorted by average processing time.

[0116] Regarding early warning notifications: When the risk level reaches a high or severe level, or when the risk status changes to "escalated" due to timeout, the system sends real-time notifications through the following channels: internal messages (displayed in the system's internal message center), emails (sent to the responsible person and their superiors' email addresses), and mobile app push notifications (sent to bound device tokens). The notification content includes a risk description, risk level, and a link to the rectification task.

[0117] This embodiment improves the transparency and response speed of risk information through visualization and multi-channel early warning, ensuring that relevant personnel are aware of the risks and take action as soon as possible.

[0118] Example 9: Self-iterative optimization of the knowledge base.

[0119] The risk knowledge base adopts a self-iterative optimization mechanism: (1) Each time the risk status changes to "closed" and the processing result is successful, the system automatically adds the risk case to the knowledge base after structuring it. (2) Record the user's adoption rate of the recommendation result (the number of adoptions divided by the total number of recommendations). For the rejected recommendations, record the reason for rejection. (3) Perform weight optimization once a month. Use the grid search method to adjust the weight coefficients of project feature similarity and risk feature similarity to maximize the historical adoption rate. (4) Clean up cases with failed processing results or those that have not been adopted for a long time every quarter.

[0120] This mechanism enables the knowledge base to continuously improve its recommendation quality as it is used, forming an intelligent management closed loop that becomes smarter the more it is used.

[0121] Overall system technical effectiveness verification: The above embodiments were integrated into a complete system and tested in 50 parallel projects. The results showed that: risk identification was performed an average of 7.2 days earlier, with an accuracy rate of 92%, and the false negative rate was reduced from 35% to 5% compared to manual methods; the risk handling closure rate increased from 65% to 98%, and the average handling cycle was shortened from 5.3 days to 2.1 days; the time for developing risk response plans was reduced from an average of 2 hours to 15 minutes, the reuse rate reached 75%, and the recurrence rate of similar risks was reduced by 40%; the project delay rate due to unhandled risks decreased by 60%, and the number of risk escalation events decreased by 45%.

[0122] Example embodiments have been disclosed herein, and while specific terminology has been used, it is for illustrative purposes only and should be construed as such, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in conjunction with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in conjunction with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of the invention as set forth in the appended claims.

Claims

1. A method for intelligent early warning and closed-loop processing of project risks, characterized in that, include: Collect multi-dimensional indicator data during the project execution process, and normalize the multi-dimensional indicator data; The normalized indicator data is input into the logistic regression model to calculate the probability of risk occurrence. The mathematical form of the logistic regression model is as follows: ； in , , These are the normalized task delay rate, resource load rate, and cost overrun rate, respectively. , , , These are the model parameters to be determined. Based on the preset impact coefficient of the risk type, calculate the comprehensive risk score, and determine the risk level based on the comprehensive risk score; When the risk level reaches a preset threshold, a rectification task associated with the risk is automatically generated, and the execution status of the rectification task is tracked through a risk state machine to drive the risk status change until the risk is closed. Closed risk cases are stored in a risk knowledge base in the form of structured data. When a new risk is identified, similar historical cases are retrieved from the risk knowledge base, and the corresponding countermeasures for the historical cases are output.

2. The method according to claim 1, characterized in that, The model parameters of the logistic regression model are trained in the following way: at least two hundred historical project samples are collected, each sample contains a normalized indicator feature vector and a label value, wherein the first value indicates that a risk event has occurred and the second value indicates that no risk event has occurred. The model parameters are iteratively optimized using gradient descent to minimize the cross-entropy loss function.

3. The method according to claim 2, characterized in that, The mathematical form of the cross-entropy loss function is: ； in For the sample size, For the first The label value of each sample, The probability value predicted by the model; The parameter update formula for the gradient descent method is as follows: The j-th model parameter Subtract the learning rate α and gradient value The product of the two is used as the updated product. Where j takes values ​​of 0, 1, 2, or 3. Let represent the j-th feature value of the i-th sample, and when j=0... =1.

4. The method according to claim 1, characterized in that, The risk levels are divided according to the S value as follows: S≥80 is severe, 60≤S<80 is high, 40≤S<60 is medium, 20≤S<40 is low, and S<20 is a level of concern; the preset threshold refers to a risk level of high or severe.

5. The method according to claim 1, characterized in that, The automatic generation of risk-associated rectification tasks includes: matching response measures from a pre-set response measure template library based on the risk type; extracting the person in charge of the project module associated with the risk from the project organizational structure as the person responsible for the rectification task; calculating the deadline for the rectification task based on the risk level; and creating a rectification task record in the database that is associated with and bound to the risk record.

6. The method according to claim 1, characterized in that, The risk status machine includes six states: pending identification, pending processing, processing, pending re-inspection, closed, and upgraded. When the responsible person submits a re-inspection application, the risk status changes to pending re-inspection, which is reviewed by the re-inspector. If the review is successful, it changes to closed; otherwise, it is returned to pending processing. When the rectification task is not completed before the deadline, the risk status changes to upgraded, and a reminder notice is sent.

7. The method according to claim 1, characterized in that, The process of retrieving similar historical cases includes: extracting project feature attributes and risk feature attributes of the current case, wherein the project feature attributes include project type, industry, and project size, and the risk feature attributes include risk type and trigger indicator vector; calculating the project feature similarity and risk feature similarity between the current case and historical cases respectively, and then weighting and summing the two to obtain a comprehensive similarity; and outputting historical cases with a comprehensive similarity greater than a preset threshold as recommendation results.

8. The method according to claim 7, characterized in that, The project feature similarity is obtained by weighted averaging the precise matching of project types, the distance calculated based on an industry classification tree for the respective industries, and the normalized difference calculated for project size; the risk feature similarity is obtained by weighted summing the distance calculated based on a risk classification tree for risk types, the cosine similarity calculated for trigger indicator vectors, and the formula for calculating the cosine similarity is: ; in This represents the trigger indicator vector for the current case. This represents the trigger indicator vector for historical cases, where n represents the number of dimensions in the trigger indicator vector. This represents the normalized value of the k-th trigger indicator in the current case. This represents the normalized value of the k-th trigger indicator in a historical case.

9. The method according to claim 1, characterized in that, The method also includes a risk and project phase transition linkage step: taking the highest risk level of all unclosed risks in the current project as a prerequisite for project phase transition; When the highest risk level is severe, the project is prevented from proceeding to the next stage; when the highest risk level drops to medium or below, the stage flow is resumed.

10. A project risk intelligent early warning and closed-loop processing system, characterized in that, The system is used to perform the method according to any one of claims 1-9, and the system comprises: The data acquisition module is used to collect multi-dimensional indicator data during the project execution process and to normalize the multi-dimensional indicator data. The model building module is used to input normalized indicator data into the logistic regression model to calculate the probability value of risk occurrence. The mathematical form of the logistic regression model is as follows: ； in , , These are the normalized task delay rate, resource load rate, and cost overrun rate, respectively. , , , These are the model parameters to be determined. The risk assessment module is used to calculate a comprehensive risk score based on the preset impact coefficient of the risk type, and to determine the risk level based on the comprehensive risk score. The execution module is used to automatically generate rectification tasks associated with the risk when the risk level reaches a preset threshold, and to track the execution status of the rectification tasks through a risk state machine, driving the risk status change until the risk is closed. The case entry module is used to store closed risk cases in the risk knowledge base in the form of structured data, and when a new risk is identified, it retrieves similar historical cases from the risk knowledge base and outputs the corresponding response measures for the historical cases.

Images