Intelligent risk assessment and early warning system for project management

Abstract

The invention relates to the technical field of data processing, in particular to an intelligent risk assessment and early warning system for project management, which comprises an acquisition module, a judgment module, a determination module, an actual generation module, an ideal generation module, an adjustment module and an alarm module. Real-time data of five dimensions of demand document change, communication text, development activeness and quality feedback are deeply associated, code submission frequency and progress deviation are taken as bottom layer triggering conditions, the health condition of a main path for project execution can be accurately drawn, potential hazards in communication are quantified through risk keyword density, and the risk of project execution is effectively improved. And depending on the consistency of the actual risk index and the ideal model output in the adjustment module, the problems of slow response to the demand change quality type and lagging risk early warning caused by low risk identification accuracy due to excessive dependence on a single risk assessment model and static parameters are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an intelligent risk assessment and early warning system for project management. Background Technology

[0002] In current software project management practices, frequent requirement changes have become the norm. Especially in the context of agile development, multi-party collaboration, and rapid iteration, requirements are not only volatile but also vary greatly in quality, often triggering a chain reaction of development delays, feature rollbacks, and wasted testing resources. Most existing project management platforms only focus on the frequency or quantity of requirement changes, lacking in-depth assessment and classification of the "quality of changes" themselves. This makes it difficult to promptly identify hidden problems that, while not frequent changes, pose serious risks to project stability. Therefore, there is an urgent need for a mechanism capable of identifying "requirement changes," dynamically monitoring and intelligently assessing the depth of content behind these changes, their impact, and the probability of them triggering risks, thereby achieving accurate early warnings and optimized resource scheduling.

[0003] Patent document CN115271672A discloses a project management platform, which includes a capability layer, a process layer, and a presentation layer. The presentation layer serves as the entry point for all users, displaying data for each project task and transmitting user interaction data to the capability layer for processing and feedback to the presentation and process layers. The process layer manages the entire project lifecycle and controls the flow of project tasks at each stage. The capability layer provides the basic services required by the presentation and process layers, and receives and processes all user requests. These basic services include at least access control, online editing, task assignment, task tracking, and risk warning.

[0004] Therefore, the project management platform has the following problems: The platform strictly separates the interface, process, and capability layers, with user operations only initiated at the presentation layer. It lacks the ability to collect and process multi-source data (such as document changes, communication texts, and code submissions) in real time and with fine granularity, making it unable to conduct in-depth dynamic assessments of project risks. The literature only considers permission management and online editing as basic services, without providing a dedicated dynamic adjustment mechanism for "risk warnings." It lacks adaptive optimization functions for trigger thresholds, monitoring windows, model parameters, etc., making it difficult for the warning system to accurately adapt to different project characteristics. The platform treats "risk warnings" as a general service but fails to specifically model and identify different risk types, such as quality fluctuations caused by requirement changes and declines in development activity, making it difficult to support targeted intervention and refined management. Summary of the Invention

[0005] To address this, the present invention provides an intelligent risk assessment and early warning system for project management. This system overcomes the problems in existing technologies, such as low accuracy in risk identification due to over-reliance on a single risk assessment model and static parameters, which leads to slow response to changes in the quality of demand and delayed risk warnings, by integrating multi-source data and introducing a dynamic adjustment mechanism.

[0006] To achieve the above objectives, the present invention provides an intelligent risk assessment and early warning system for project management, comprising: The acquisition module is used to acquire in real time the document change frequency, risk keyword density, code submission frequency, task progress deviation, and defect re-opening rate of software projects containing project data; The determination module, which is connected to the acquisition module, is used to determine the occurrence of abnormal events based on the code submission frequency, the task progress deviation and the preset abnormal occurrence threshold, and form a determination result. A determination module, which is connected to the acquisition module and the judgment module respectively, is used to determine the anomaly type of the abnormal event as the requirement change quality type based on the judgment result, the defect re-opening rate and the document change frequency. The actual generation module is connected to the acquisition module and the determination module respectively, and is used to generate an actual risk index based on the demand change quality type and the risk keyword density. An ideal generation module, connected to the acquisition module, is used to classify and cluster the project data of a preset observation data volume according to a preset machine learning model in order to generate an ideal risk index. An adjustment module, which is connected to the actual generation module and the ideal generation module respectively, is used to adjust the preset anomaly occurrence threshold or the preset observation data volume according to the actual risk index and the ideal risk index; An alarm module, connected to the actual generation module, is used to issue an alarm based on the actual risk index and the preset index range regenerated based on the adjusted preset anomaly occurrence threshold or the adjusted preset observation data volume.

[0007] Furthermore, the determination module includes: The fitting unit is used to perform linear regression on all the code submission frequencies within a preset first determination time period to obtain the slope of the fitted line. A determination unit, connected to the fitting unit, is used to determine an abnormal event based on the task progress deviation and the preset abnormality occurrence threshold when the slope of the fitted line is less than zero and the absolute value of the slope of the fitted line is greater than a preset standard slope, thereby forming the determination result.

[0008] Furthermore, the determination unit includes: The incremental calculation subunit is used to calculate the difference between the task progress deviation at each moment within the preset second determination time and the task progress deviation at the previous moment, and obtain several deviation increments. An accumulation subunit, which is connected to the incremental calculation subunit, is used to calculate the sum of all the positive deviation increments to obtain the incremental accumulation value; A determination subunit, which is connected to the accumulation subunit, is used to determine that an abnormal event has occurred when the incremental accumulation value is greater than zero, and to form the determination result.

[0009] Furthermore, the determining module includes: The reopening fluctuation calculation unit is used to calculate the standard deviation of the defect reopening rate at each time from the initial time to a preset time period when the judgment result is formed, and obtain several reopening fluctuation values. The change fluctuation calculation unit is used to calculate the standard deviation of the document change frequency within the preset time period from the initial time to each time within the preset time period when the judgment result is formed, so as to obtain the change fluctuation value; A determination unit, which is connected to the reopening fluctuation calculation unit and the change fluctuation calculation unit respectively, is used to determine the anomaly type of the abnormal event as the demand change quality type based on all the reopening fluctuation values ​​and the change fluctuation values.

[0010] Furthermore, the determining unit includes: The change normalization subunit is used to normalize all the reopening fluctuation values ​​to obtain the reopening normalization set, and to normalize all the change fluctuation values ​​to obtain the change normalization set. A consistency calculation subunit, which is connected to the change normalization subunit, is used to calculate the correlation coefficient between the reopened normalization set and the change normalization set to obtain the consistency of fluctuation changes; A subunit is defined, which is connected to the consistency calculation subunit, to determine the anomaly type as the demand change quality type when the fluctuation change consistency is greater than the preset standard consistency.

[0011] Furthermore, the actual generation module includes: The density fluctuation calculation unit is used to calculate the density of all the risk keywords within a preset assessment period when the quality type of the demand change is determined, and to obtain the density fluctuation value. A generation unit, connected to the density fluctuation calculation unit, is used to generate the actual risk index based on the density of all risk keywords within the preset evaluation period when the density fluctuation value is greater than a preset fluctuation density threshold.

[0012] Further, the generation unit includes: The density normalization subunit is used to normalize the density of all the risk keywords to obtain several normalized densities. A sub-unit is assigned, which is connected to the density normalization sub-unit, to assign a linear time weight to each of the normalized densities, thereby obtaining several computational weights. The mean calculation subunit is connected to the density normalization subunit and the assignment subunit respectively, and is used to calculate a weighted average based on all the normalized densities and the calculation weights to obtain a density-weighted mean. A sub-unit is generated, which is connected to the mean calculation sub-unit, to calculate the product of the density-weighted mean and the preset correction coefficient, and truncate it to the interval [0, 1] to generate the actual risk index.

[0013] Furthermore, the adjustment module includes: The index deviation calculation unit is used to calculate the absolute value of the relative deviation between the actual risk index and the ideal risk index to obtain the index deviation. An adjustment unit, connected to the index deviation unit, is used to adjust the preset anomaly occurrence threshold or the preset observation data volume according to the index deviation and the preset index deviation range.

[0014] Furthermore, the adjustment unit includes: The first adjustment subunit is used to reduce the preset anomaly occurrence threshold based on the minimum value of the preset index deviation range, the relative deviation of the index deviation, and the preset adjustment coefficient when the index deviation is less than the minimum value of the preset index deviation range. The second adjustment subunit is used to increase the preset observation data volume based on the relative deviation between the exponential deviation and the maximum value of the preset exponential deviation range, and the preset adjustment coefficient, when the exponential deviation is greater than the maximum value of the preset exponential deviation range.

[0015] Furthermore, the alarm module includes: The first alarm unit is used to issue a risk recovery notification when the actual risk index is less than the minimum value of the preset index range; The second alarm unit is used to issue a severe risk alarm when the actual risk index is greater than the maximum value of the preset index range.

[0016] Compared with existing technologies, the beneficial effects of this invention lie in its deep correlation of real-time data across five dimensions—requirement document changes, communication texts, development activity, and quality feedback—to construct a closed-loop logic encompassing "development rhythm; schedule deviation; quality fluctuation; and risk output." First, using code submission frequency and schedule deviation as underlying triggering conditions, the health status of the project's main execution path can be accurately outlined. Second, the introduction of synchronous fluctuation detection of defect rework rate and document change frequency enables precise identification of quality anomalies caused by requirement changes. Finally, by quantifying potential risks in communication through risk keyword density, and relying on the consistency between the actual risk index and the ideal model output in the adaptive adjustment module, the warning sensitivity and observation window are continuously optimized. This ensures both the interpretability and traceability of risk assessment and, through the underlying causal chain between parameters, achieves efficient and dynamic intelligent early warning. This effectively solves the problem of low accuracy in risk identification due to over-reliance on a single risk assessment model and static parameters, leading to slow response to quality type changes in requirements and delayed risk warnings.

[0017] Furthermore, by organically linking the trend analysis of code submission frequency with the determination of the magnitude of task progress deviation, a logic of "development rhythm; execution deviation; anomaly warning" is formed. When development activity drops sharply (sudden change in slope), the system will not blindly issue an alarm, but will perform secondary verification through the underlying execution indicator of "task progress deviation," thus achieving double confirmation of risk assessment. Through the dual-dimensional collaborative judgment based on time series trends and cumulative deviation, not only is noise caused by occasional submission fluctuations effectively filtered out, but the real risks caused by development delays are also accurately captured, thereby improving the accuracy and interpretability of the warning.

[0018] Furthermore, through incremental analysis of task progress deviations, the system not only detects whether task delays exist but also measures whether the delays are continuously worsening. Task progress deviations reflect the gap between the actual progress of current work and the expected plan, while the deviation increment reveals whether this gap is trending upwards. The cumulative increment value is used to quantify the pressure of accumulated delays. Compared with static deviation judgment, this approach is more dynamic and sensitive, enabling early warning of project execution risks and effectively linking task management and progress monitoring mechanisms.

[0019] Furthermore, by synchronously analyzing the fluctuation intensity of "defect rework rate" and "document change frequency," the system can reveal the causal chain between requirement changes and quality instability. That is, when the fluctuations of the two are consistent, it indicates that frequent requirement adjustments directly lead to repeated rework and defect rework. Utilizing standard deviation, an indicator reflecting the degree of data dispersion, not only can single-point anomalies be filtered out, but the amplitude of oscillations over the entire time window can also be quantified. This allows for precise positioning of the synchronous risk of "change-rework." Through a judgment logic based on the correlation of underlying data fluctuations, the system avoids being misled by a single indicator, effectively improving the accuracy and interpretability of anomaly type identification.

[0020] Furthermore, the fluctuations of "defect rework rate" and "document change frequency" are quantified using standard deviation. After normalization, the correlation coefficient is used to measure the synchronicity between the two, which can directly reflect the underlying causal relationship between requirement changes and quality issues. When frequent adjustments to document requirements lead to a synchronous increase in rework fluctuations, the consistency is high. The system accurately captures the core anomaly type of "requirement change quality". It not only filters out single-point sudden interference, but also ensures the robustness and interpretability of the judgment through statistical correlation between parameters, providing highly reliable risk identification for project management.

[0021] Furthermore, by combining the analysis of the fluctuation intensity of "risk keyword density" with the overall density level, the system can capture both sudden high-frequency negative signals (large fluctuation values) in communication and reflect the sustained popularity of risk topics (average density), forming a dual quantitative perspective on project risk. The fluctuation density threshold serves as a trigger condition, ensuring that in-depth assessments are only conducted when the frequency of risk discussions fluctuates significantly, avoiding false alarms caused by stable periods or occasional events. Meanwhile, the risk index generation logic of the complete density sequence guarantees that the assessment results are sensitive to both "high-frequency continuous discussion" and "short-term sharp fluctuations," thus accurately and interpretably presenting the overall project risk level in the context of "frequent changes in demand."

[0022] Furthermore, normalization eliminates the dimensional differences in keyword density across different time periods, making density values ​​comparable; linear time weighting enhances the early warning sensitivity of "significantly increased risk in recent times"; weighted averaging organically integrates historical and real-time signals, smoothing out occasional fluctuations; correction coefficients provide type-specific flexible calibration, ensuring that the risk index is not overly conservative or aggressive in the "quality of changing demand" scenario; finally, the [0, 1] truncation ensures the uniform scale and interpretability of the risk index, accurately reflecting the dynamics of potential quality risks in project communication.

[0023] Furthermore, when the system is overly conservative (actually much lower than ideal), more potential risks are captured in a timely manner by reducing the threshold; when the system is overly aggressive (actually much higher than ideal), occasional fluctuations are smoothed by increasing the amount of data. This achieves a balance between sensitivity and stability. The interaction of various parameters (risk index, deviation range, threshold, and data volume) improves the accuracy of early warnings and reduces the false alarm rate.

[0024] Furthermore, the deviation between the actual risk index and the ideal value not only reflects the current model performance but also directly participates in the dynamic adjustment of subsequent parameters. By setting a preset index deviation range, the deviation value is compared with the minimum and maximum values. A preset adjustment coefficient is introduced as a control for the adjustment range, allowing the preset anomaly threshold and preset observation data volume to be continuously optimized based on actual results. This achieves a two-way dynamic balance between risk identification accuracy and system stability, making it suitable for development environments with frequent changes in requirements or large data fluctuations, and continuously maintaining the model's adaptability and judgment efficiency.

[0025] Furthermore, by setting preset index ranges with upper and lower limits, the system dynamically identifies and responds to actual risk states. A logical comparison between the actual risk index and the minimum and maximum values ​​of the preset range establishes a mechanism for the system to perceive risk change trends. Lower limit judgment effectively identifies risk mitigation trends, promptly issuing recovery prompts and avoiding resource waste; upper limit judgment ensures rapid early warning when risks continue to escalate, prompting the system to adjust or intervene in a timely manner. This mechanism based on numerical threshold linkage enhances the system's adaptability and the intelligence level of risk response, achieving closed-loop control from risk identification to risk response. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the intelligent risk assessment and early warning system for project management in this embodiment; Figure 2 This is a logic diagram for determining the occurrence of abnormal events in this embodiment; Figure 3 This is the logic diagram for determining the quality type of demand change in this embodiment; Figure 4 This is a logic diagram for determining the actual risk index in this embodiment. Detailed Implementation

[0027] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0028] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0029] Please see Figure 1 As shown, it is a schematic diagram of the intelligent risk assessment and early warning system for project management in this embodiment; This embodiment provides an intelligent risk assessment and early warning system for project management, including: The acquisition module is used to acquire in real time the document change frequency, risk keyword density, code submission frequency, task progress deviation, and defect re-opening rate of software projects containing project data; The determination module, which is connected to the acquisition module, is used to determine the occurrence of abnormal events based on the code submission frequency, the task progress deviation and the preset abnormal occurrence threshold, and form a determination result. A determination module, which is connected to the acquisition module and the judgment module respectively, is used to determine the anomaly type of the abnormal event as the requirement change quality type based on the judgment result, the defect re-opening rate and the document change frequency. The actual generation module is connected to the acquisition module and the determination module respectively, and is used to generate an actual risk index based on the demand change quality type and the risk keyword density. An ideal generation module, connected to the acquisition module, is used to classify and cluster the project data of a preset observation data volume according to a preset machine learning model in order to generate an ideal risk index. An adjustment module, which is connected to the actual generation module and the ideal generation module respectively, is used to adjust the preset anomaly occurrence threshold or the preset observation data volume according to the actual risk index and the ideal risk index; An alarm module, connected to the actual generation module, is used to issue an alarm based on the actual risk index and the preset index range regenerated based on the adjusted preset anomaly occurrence threshold or the adjusted preset observation data volume.

[0030] The acquisition module seamlessly integrates with multiple systems to retrieve key metrics of the software project in real time. Specifically, it periodically analyzes commit records in the code repository to obtain the "code commit frequency" by calling webhooks or APIs of version control systems (such as GitHub and GitLab); it obtains the planned and actual completion times of each task by integrating the RESTful interface of project management tools (such as Jira and Azure DevOps) to calculate the "task progress deviation"; it subscribes to the "close-reopen" operation logs in defect tracking systems (such as Bugzilla and Jira Issues) to calculate the "defect reopening rate" in real time; it monitors update events in the document version history of document management platforms (such as Confluence and SharePoint) to obtain the "document change frequency"; and it uses NLP engines (such as spaCy and BERT) to perform semantic analysis on the message streams of team chat tools (such as Slack and Teams) and email lists to extract and count the frequency of risk-related words in real time to obtain the "risk keyword density". All data is timestamped and, after being cleaned, denoised, and standardized through an ETL process, is pushed to the subsequent judgment and evaluation modules.

[0031] The preset anomaly occurrence threshold is a boundary value used to determine whether a task progress deviation constitutes an abnormal event. It depends on the project type, project cycle, and historical project progress deviation data, and is usually set between 10% and 30%. In this embodiment, it is set to 20%, which can effectively identify the potential development risks caused by progress deviation.

[0032] The preset observation data volume refers to the length of the historical project data time window required to generate the ideal risk index. It depends on the project size, data sampling frequency and model stability requirements, and is usually set between 5 and 30 days. In this embodiment, it is set to 14 days, which can ensure the representativeness and stability of the data input while maintaining the model response speed.

[0033] The preset index range refers to the reference interval used to determine whether the actual risk index is abnormal. It depends on the output distribution characteristics of the risk scoring model and the statistical boundaries of historical risk data. It is usually set between 0.4 and 0.7. In this embodiment, it is set to 0.6, which can reduce the false alarm rate while ensuring the sensitivity of the warning.

[0034] The preset machine learning model, within the ideal generation module, performs joint classification and clustering analysis on the collected multi-source project data to generate a reliable "ideal risk index" model. This model consists of two parts: 1. Classifier (Supervised Learning) Algorithm selection: XGBoost (Extreme Gradient Boosting Tree) was adopted. Input features: five real-time dynamic indicators, including document change frequency, risk keyword density, code submission frequency, task progress deviation, and defect re-opening rate, as well as their statistical derivative features (such as moving average, standard deviation, maximum / minimum value). Output labels: Project status category (normal / low risk / medium risk / high risk), with historical project risk ratings used as supervisory labels during training; Key hyperparameters: learning rate 0.1, number of trees 100, maximum tree depth 6, subsampling ratio 0.8, column sampling ratio 0.7; Training and validation: Hierarchical k-fold cross-validation (k=5) was used, and early stopping was used in each fold to prevent overfitting; the final model had an average AUC ≥ 0.86 on the validation set.

[0035] 2. Clusterer (Unsupervised Learning) Algorithm selection: K-means clustering; Cluster number: k=4 was selected based on a comprehensive evaluation of the elbow rule and the profile coefficient, corresponding to four risk levels; Distance metric: Standardized Euclidean distance is used, and all input features are first normalized to [0, 1] using Min-Max. Application of results: By combining the probability distribution of the classifier output with the clustering labels, the influence of outliers is smoothed out, forming an ideal risk index for multi-model fusion. Workflow: Offline training: Periodically extract thousands of multi-source feature vectors and their risk labels from the historical project library, and train and optimize the classifier and clusterer respectively; Online inference: The data in the current preset observation window is first normalized, then input into the classifier to generate risk probabilities, and simultaneously sent to the clusterer to determine the distance between cluster centers; Fusion output: The results of the two are fused according to the proportional weight (70% classifier + 30% clusterer) to obtain a continuous ideal risk index (0~1). This preset machine learning model takes into account the accurate classification ability of supervised learning and the global structure capture ability of unsupervised clustering, effectively balancing the accuracy and robustness of the model, and providing a reliable reference for subsequent threshold adjustment and alarm decision-making.

[0036] Document change frequency refers to the number of document version iterations per unit of time, used to measure the stability of requirements or design specifications; risk keyword density refers to the proportion of high-risk words such as "delay," "error," "abandon," and "pending" appearing in communication records, reflecting the team's subjective risk perception; code commit frequency refers to the number of code commits or committers per unit of time, used to assess development activity and collaboration smoothness; task progress deviation refers to the difference or percentage between the actual completion time and the planned completion time, revealing whether the project execution is progressing as expected; defect reopening rate refers to the proportion of closed defects that are reopened, reflecting the stability of code quality and whether the problem fixes are in place.

[0037] First, the acquisition module collects key dynamic indicators of the software project in real time: document change frequency reflects the pace of requirement updates, risk keyword density captures risk signals implied in communication, code commit frequency measures development activity, task progress deviation reveals the magnitude of execution deviation, and defect rework rate indicates quality stability. Then, the judgment module performs linear regression and incremental analysis on the short-term trend and cumulative deviation of code commit frequency and task progress deviation to determine if any abnormal events have occurred. If an anomaly is confirmed, the determination module combines the consistency of fluctuations in defect rework rate and document change frequency to uniquely identify it as a "quality anomaly type caused by requirement changes." Based on this, the actual generation module calculates the fluctuation value of risk keyword density under this type and generates a standardized actual risk index; in parallel, the ideal generation module performs classification and clustering of multi-source project data within the observed data volume based on a preset machine learning model, outputting an ideal risk index. The system then adjusts the deviation between the two by the adjustment module, dynamically adjusts the anomaly threshold or observation window size, and finally, the alarm module compares the regenerated actual risk index based on the adjusted indicators with the preset range, triggering a "recovery notification" or "serious risk alarm."

[0038] By deeply linking real-time data across five dimensions—requirement document changes, communication texts, development activity, and quality feedback—a closed-loop logic of "development rhythm; schedule deviation; quality fluctuation; risk output" is constructed. First, using code submission frequency and schedule deviation as underlying triggering conditions, the health of the project's main execution path is accurately outlined. Then, synchronous fluctuation detection of defect rework rate and document change frequency is introduced to accurately identify quality anomalies caused by requirement changes. Finally, potential risks in communication are quantified through risk keyword density, and the adaptive adjustment module continuously optimizes early warning sensitivity and observation windows based on the consistency between the actual risk index and the ideal model output. This ensures both the interpretability and traceability of risk assessment and, through the underlying causal chain between parameters, achieves efficient and dynamic intelligent early warning. This effectively solves the problem of low accuracy in risk identification due to over-reliance on a single risk assessment model and static parameters, leading to slow response to quality type changes in requirements and delayed risk warnings.

[0039] Please continue reading. Figure 2 As shown, this is the logic diagram for determining the occurrence of an abnormal event in this embodiment; The determination module includes: The fitting unit is used to perform linear regression on all the code submission frequencies within a preset first determination time period to obtain the slope of the fitted line. A determination unit, connected to the fitting unit, is used to determine an abnormal event based on the task progress deviation and the preset abnormality occurrence threshold when the slope of the fitted line is less than zero and the absolute value of the slope of the fitted line is greater than a preset standard slope, thereby forming the determination result.

[0040] The preset first judgment duration is the length of the data time window used for linear regression analysis of the trend of code submission frequency changes. It depends on the team's submission rhythm and iteration cycle, and is usually set between 3 and 7 days. In this embodiment, it is set to 5 days, which can reliably capture the trend changes of development activity in the short term.

[0041] The preset standard slope is a threshold parameter used to determine whether the decrease in code submission frequency is abnormal. It depends on the project type, the range of historical submission fluctuations, and the team's work rhythm. It is usually set between 0.05 and 0.3. In this embodiment, it is set to 0.15, which can effectively identify abnormal trends in the project caused by a significant decrease in code submission frequency.

[0042] The fitting unit collects code submission frequency data at each time point (such as daily or hourly) within a preset first judgment period, constructs a time series point set, uses time as the independent variable and submission frequency as the dependent variable, and performs linear regression calculation using the least squares method to obtain the slope of the fitted line, which is used to represent the overall trend of code submission frequency change within that time period.

[0043] In the judgment module, the system first uses a fitting unit to perform linear regression analysis on all code submission frequency data within a preset first judgment period to calculate the slope of the fitted straight line representing the trend of development activity. If the slope is negative and its absolute value exceeds the preset standard slope, it indicates that the development pace has declined rapidly in the short term. At this time, the judgment unit further combines the real-time collected task progress deviation with a preset anomaly occurrence threshold. If the deviation is consistently positive and accumulates to exceed the threshold, the system determines that an "abnormal event" has occurred and passes the judgment result to subsequent modules.

[0044] By organically linking the trend analysis of code submission frequency with the determination of the magnitude of task progress deviation, a logic of "development rhythm; execution deviation; anomaly warning" is formed. When development activity drops sharply (sudden change in slope), the system will not blindly issue an alarm, but will perform secondary verification through the underlying execution indicator of "task progress deviation". This achieves double confirmation of risk assessment. Through the dual-dimensional collaborative judgment based on time series trends and cumulative deviation, not only is noise caused by occasional submission fluctuations effectively filtered out, but the real risks caused by development delays are also accurately captured, thereby improving the accuracy and interpretability of the warning.

[0045] Specifically, the determination unit includes: The incremental calculation subunit is used to calculate the difference between the task progress deviation at each moment within the preset second determination time and the task progress deviation at the previous moment, and obtain several deviation increments. An accumulation subunit, which is connected to the incremental calculation subunit, is used to calculate the sum of all the positive deviation increments to obtain the incremental accumulation value; A determination subunit, which is connected to the accumulation subunit, is used to determine that an abnormal event has occurred when the incremental accumulation value is greater than zero, and to form the determination result.

[0046] The preset second judgment period refers to the length of the time interval used to calculate the incremental task progress deviation. It depends on the accuracy requirements of project management and the granularity of the task, and is usually set between 1 day and 14 days. In this embodiment, it is set to 7 days, which can better balance the response speed of progress deviation and the stability of data fluctuation, and accurately identify the trend of task delay.

[0047] First, within a preset second judgment period, task progress deviation data is extracted hourly, and the deviation difference between adjacent time points is calculated to form a set of deviation increments. Then, the accumulation subunit sums all positive deviation increments to obtain a cumulative increment value used to quantify the project task delay trend. Finally, the judgment subunit determines whether this cumulative value is greater than zero. If it is, it is considered that there is a gradually escalating task delay trend, an abnormal event is determined, and the judgment result is output.

[0048] By analyzing the incremental deviation of task progress, the system not only detects whether task delays exist but also measures whether the delays are escalating. Task progress deviation reflects the gap between the actual progress of the current work and the expected plan, while the deviation increment reveals whether the gap is trending upward. The cumulative increment value is used to quantify the pressure of accumulated delays. Compared with static deviation judgment, it is more dynamic and sensitive, and can provide early warning of project execution risks, effectively linking task management and progress monitoring mechanisms.

[0049] Specifically, the determining module includes: The reopening fluctuation calculation unit is used to calculate the standard deviation of the defect reopening rate at each time from the initial time to a preset time period when the judgment result is formed, and obtain several reopening fluctuation values. The change fluctuation calculation unit is used to calculate the standard deviation of the document change frequency within the preset time period from the initial time to each time within the preset time period when the judgment result is formed, so as to obtain the change fluctuation value; A determination unit, which is connected to the reopening fluctuation calculation unit and the change fluctuation calculation unit respectively, is used to determine the anomaly type of the abnormal event as the demand change quality type based on all the reopening fluctuation values ​​and the change fluctuation values.

[0050] After forming the judgment result, the reopening fluctuation calculation unit first collects the defect reopening rate data of each sampling point within the initial time to the preset time period, and calculates the standard deviation of the sequence according to the time segment to obtain a series of reopening fluctuation values; at the same time, the change fluctuation calculation unit performs the same standard deviation calculation on the document change frequency sequence within the same time window to generate a change fluctuation value sequence; finally, the determination unit compares the overall amplitude and distribution characteristics of the two sets of fluctuation values. If both show significant fluctuations, the "demand change quality type" is identified.

[0051] By synchronously analyzing the fluctuation intensity of "defect rework rate" and "document change frequency," the system can reveal the causal chain between requirement changes and quality instability. Specifically, when the fluctuations of these two metrics are consistent, it indicates that frequent requirement adjustments directly lead to repeated rework and defect rework. Utilizing standard deviation, an indicator reflecting data dispersion, the system can not only filter out single-point anomalies but also quantify the amplitude of fluctuations over the entire time window. This allows for precise identification of the synchronous risk of "change-rework." Through a judgment logic based on the correlation of underlying data fluctuations, the system avoids being misled by a single indicator, effectively improving the accuracy and interpretability of anomaly type identification.

[0052] Please continue reading. Figure 3 As shown, this is the logic diagram for determining the quality type of demand change in this embodiment; The determining unit includes: The change normalization subunit is used to normalize all the reopening fluctuation values ​​to obtain the reopening normalization set, and to normalize all the change fluctuation values ​​to obtain the change normalization set. A consistency calculation subunit, which is connected to the change normalization subunit, is used to calculate the correlation coefficient between the reopened normalization set and the change normalization set to obtain the consistency of fluctuation changes; A subunit is defined, which is connected to the consistency calculation subunit, to determine the anomaly type as the demand change quality type when the fluctuation change consistency is greater than the preset standard consistency.

[0053] The preset standard consistency is a correlation coefficient threshold used to determine the synchronicity of fluctuations in "defect re-opening rate" and "document change frequency". It depends on the statistical correlation strength of the fluctuations of the two in historical projects and is usually set between 0.7 and 0.9. In this embodiment, it is set to 0.8, which can effectively distinguish between occasional fluctuations and risk patterns of true causal synchronization.

[0054] After generating the judgment result, the change normalization subunit first normalizes the previously calculated reopening fluctuation value and change fluctuation value using the Min-Max method to obtain a "reopening normalization set" and a "change normalization set" with the same dimension. Then, the consistency calculation subunit calculates the Pearson correlation coefficient for these two sets of normalized data to obtain the quantified consistency of fluctuation changes. Finally, the determination subunit compares the consistency with the preset standard consistency threshold. If it is greater than the threshold, the abnormal event type is determined to be "quality abnormality type caused by demand change".

[0055] The fluctuations of "defect rework rate" and "document change frequency" are quantified by standard deviation. After normalization, the correlation coefficient is used to measure the synchronicity between the two, which can directly reflect the underlying causal relationship between requirement changes and quality issues. When frequent adjustments to document requirements lead to a synchronous increase in rework fluctuations, the consistency is high. The system accurately captures the core anomaly type of "requirement change quality". It not only filters out single-point sudden interference, but also ensures the robustness and interpretability of the judgment through statistical correlation between parameters, providing highly reliable risk identification for project management.

[0056] Please continue reading. Figure 4 As shown, it is the determination logic diagram for generating the actual risk index in this embodiment; The actual generation module includes: The density fluctuation calculation unit is used to calculate the density of all the risk keywords within a preset assessment period when the quality type of the demand change is determined, and to obtain the density fluctuation value. A generation unit, connected to the density fluctuation calculation unit, is used to generate the actual risk index based on the density of all risk keywords within the preset evaluation period when the density fluctuation value is greater than a preset fluctuation density threshold.

[0057] The preset assessment duration is the length of the time window used to calculate the fluctuation of risk keyword density. It depends on the team's communication frequency and the cycle of risk topics. It is usually set between 3 and 10 days. In this embodiment, it is set to 7 days, which can accurately capture the communication fluctuation trend caused by changes in demand while taking into account the sensitivity of risk signals and data stability.

[0058] In the actual generation module, when the determination module identifies the quality type of demand change, the density fluctuation calculation unit calculates the standard deviation of the risk keyword density sequence at each moment within the preset assessment period to obtain a single density fluctuation value, so as to quantify the overall fluctuation amplitude of the risk signal in the communication. Subsequently, the generation unit determines whether the density fluctuation value exceeds the preset fluctuation density threshold. If so, the complete risk keyword density sequence within the assessment period is input into the risk index calculation logic, and the density values ​​at each moment are combined to finally output a standardized actual risk index.

[0059] By combining the analysis of the fluctuation intensity of "risk keyword density" with the overall density level, the system can capture both sudden high-frequency negative signals (large fluctuation values) in communication and reflect the sustained popularity of risk topics (average density), forming a dual quantitative perspective on project risk. The fluctuation density threshold serves as a trigger condition, ensuring that in-depth assessments are only conducted when the frequency of risk discussions fluctuates significantly, avoiding false alarms caused by stable periods or occasional events. Meanwhile, the risk index generation logic of the complete density sequence ensures that the assessment results are sensitive to both "high-frequency continuous discussion" and "short-term sharp fluctuations," thus accurately and interpretably presenting the overall project risk level in the context of "frequent changes in demand."

[0060] Specifically, the generation unit includes: The density normalization subunit is used to normalize the density of all the risk keywords to obtain several normalized densities. A sub-unit is assigned, which is connected to the density normalization sub-unit, to assign a linear time weight to each of the normalized densities, thereby obtaining several computational weights. The mean calculation subunit is connected to the density normalization subunit and the assignment subunit respectively, and is used to calculate a weighted average based on all the normalized densities and the calculation weights to obtain a density-weighted mean. A sub-unit is generated, which is connected to the mean calculation sub-unit, to calculate the product of the density-weighted mean and the preset correction coefficient, and truncate it to the interval [0, 1] to generate the actual risk index.

[0061] The linear time weights are proportionally allocated based on the time index of the normalized density: Assume there are M sampling points within the preset evaluation time, denoted as i = 1, 2, ..., M. Then the calculated weight corresponding to the i-th normalized density is: , where wi is the calculated weight corresponding to the i-th normalized density, and j is the summation index, representing the sequence number from the 1st to the Mth sampling point.

[0062] The preset correction coefficient is a proportional parameter used to adjust the sensitivity of the actual risk index. It depends on the severity of the risk type and the model calibration requirements, and is usually set between 1.0 and 1.5. In this embodiment, it is set to 1.2, which can amplify key risk signals while avoiding excessive amplification of occasional fluctuations.

[0063] In the generation unit, the density normalization subunit first performs Min-Max normalization on the risk keyword density sequence within the preset assessment period, mapping the original fluctuation amplitude to the [0, 1] interval; then, the subunit assigns a linear time weight to each normalized density according to the time sequence, ensuring that recent risk signals account for a higher proportion in the calculation; subsequently, the mean calculation subunit uses the normalized density and the corresponding time weight to calculate the weighted average, obtaining the density-weighted average that reflects the "time-weighted intensity"; finally, the generation subunit multiplies the mean by a preset correction coefficient and truncates the result to the [0, 1] interval, outputting the final actual risk index.

[0064] Normalization eliminates the dimensional differences in keyword density across different time periods, making density values ​​comparable; linear time weighting enhances the early warning sensitivity of "significantly increased risk in recent times"; weighted averaging organically integrates historical and real-time signals, smoothing out occasional fluctuations; correction coefficients provide type-specific flexible calibration, ensuring that the risk index is not overly conservative or aggressive in the "quality of changing demand" scenario; finally, the [0, 1] truncation ensures the uniform scale and interpretability of the risk index, accurately reflecting the dynamics of potential quality risks in project communication.

[0065] Specifically, the adjustment module includes: The index deviation calculation unit is used to calculate the absolute value of the relative deviation between the actual risk index and the ideal risk index to obtain the index deviation. An adjustment unit, connected to the index deviation unit, is used to adjust the preset anomaly occurrence threshold or the preset observation data volume according to the index deviation and the preset index deviation range.

[0066] The preset index deviation range is a positive range used to accommodate the relative deviation between the actual risk index and the ideal risk index. It depends on the model stability and business tolerance, and is usually set between 0.05 and 0.2. In this embodiment, it is set to 0.1, which can balance the warning sensitivity and stability, and can reduce the false alarm rate while ensuring the warning sensitivity.

[0067] In the adjustment module, the index deviation calculation unit first obtains the actual risk index from the actual generation module and the ideal risk index from the ideal generation module, and calculates the absolute value of their relative deviation, i.e., |(Actual Risk Index − Ideal Risk Index) / Ideal Risk Index|. Subsequently, the adjustment unit compares this index deviation with a preset deviation range boundary: when the deviation is below the minimum value, the system automatically relaxes the preset anomaly occurrence threshold to enhance sensitivity; when the deviation is above the maximum value, it dynamically increases the preset amount of observation data to utilize more historical information to smooth the model. The adjusted threshold or data amount takes effect immediately for the next round of risk assessment and generation.

[0068] When the system is overly conservative (actually much lower than ideal), more potential risks are captured promptly by reducing the threshold; when the system is overly aggressive (actually much higher than ideal), occasional fluctuations are smoothed out by increasing the amount of data. This achieves a balance between sensitivity and stability. The interaction of various parameters (risk index, deviation range, threshold, and data volume) improves the accuracy of early warnings and reduces the false alarm rate.

[0069] Specifically, the adjustment unit includes: The first adjustment subunit is used to reduce the preset anomaly occurrence threshold according to the minimum value of the preset index deviation range, the relative deviation of the index deviation, and the preset adjustment coefficient when the index deviation is less than the minimum value of the preset index deviation range. Where, T'=T×[1-h×(Qmin-Q) / Q], T' is the preset anomaly occurrence threshold after reduction, T is the preset anomaly occurrence threshold before reduction, h is the preset adjustment coefficient, Qmin is the minimum value of the preset index deviation range, and Q is the index deviation. The second adjustment subunit is used to increase the preset observation data volume based on the relative deviation between the index deviation and the maximum value of the preset index deviation range and the preset adjustment coefficient when the index deviation is greater than the maximum value of the preset index deviation range. Here, D'=D×[1+h×(Q-Qmax) / Qmax], D' is the preset observation data volume after the increase, D is the preset observation data volume before the increase, and Qmax is the maximum value of the preset index deviation range.

[0070] The preset adjustment coefficient is a proportional parameter used to control the adjustment range of the threshold or data volume. It depends on the system's need to balance risk response speed and stability, and is usually set between 0.1 and 1. In this embodiment, it is set to 0.3, which can ensure the adjustment effect while avoiding excessive fluctuations and improving the stability and sensitivity of anomaly detection.

[0071] When the relative deviation between the actual risk index and the ideal risk index is lower than the minimum value of the preset index deviation range, the first adjustment subunit actively reduces the preset anomaly occurrence threshold based on the degree of its relative deviation from the minimum value and the preset adjustment coefficient, thereby improving the sensitivity of anomaly detection. Conversely, when the deviation is higher than the maximum value, the second adjustment subunit appropriately increases the amount of observation data by using its relative deviation from the maximum value and the preset adjustment coefficient, so that the system can introduce more data for judgment and reduce the risk of misjudgment caused by accidental anomalies.

[0072] The deviation between the actual risk index and the ideal value not only reflects the current model performance but also directly participates in the dynamic adjustment of subsequent parameters. By setting a preset index deviation range, the deviation value is compared with the minimum and maximum values. A preset adjustment coefficient is introduced as a control for the adjustment range, so that the preset anomaly occurrence threshold and the preset amount of observation data are continuously optimized according to the actual effect. This achieves a two-way dynamic balance between risk identification accuracy and system stability, which is suitable for development environments with frequent changes in requirements or large data fluctuations, and can continuously maintain the model's adaptability and judgment efficiency.

[0073] Specifically, the alarm module includes: The first alarm unit is used to issue a risk recovery notification when the actual risk index is less than the minimum value of the preset index range; The second alarm unit is used to issue a severe risk alarm when the actual risk index is greater than the maximum value of the preset index range.

[0074] The preset index range is the range of risk index values ​​used to determine the risk level. It depends on the system's risk tolerance requirements and historical data analysis results, and is usually set between [0.3, 0.7]. In this embodiment, it is set to [0.35, 0.65], which can accurately divide the status changes of the actual risk index and facilitate triggering corresponding alarms or notifications.

[0075] The alarm module monitors the generated actual risk index in real time during continuous system operation. When the index is less than the minimum value of the preset index range, the first alarm unit is activated, issuing a risk recovery notification, indicating that the system status has returned to a safe range; when the actual risk index is greater than the maximum value of the preset index range, the second alarm unit is activated, issuing a severe risk alarm, indicating that the acceptable risk threshold has been exceeded and immediate intervention or adjustment is required.

[0076] By setting preset index ranges with upper and lower limits, the system dynamically identifies and responds to actual risk states. A logical comparison between the actual risk index and the minimum and maximum values ​​of the preset range establishes a mechanism for the system to perceive risk change trends. Lower limit judgment effectively identifies risk mitigation trends, promptly issuing recovery prompts and avoiding resource waste; upper limit judgment ensures rapid early warning when risks continue to escalate, prompting the system to adjust or intervene in a timely manner. This mechanism based on numerical threshold linkage enhances the system's adaptability and the intelligence level of risk response, achieving closed-loop control from risk identification to risk response.

[0077] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent risk assessment and early warning system for project management, characterized in that, include: The acquisition module is used to acquire in real time the document change frequency, risk keyword density, code submission frequency, task progress deviation, and defect re-opening rate of software projects containing project data; The determination module, which is connected to the acquisition module, is used to determine the occurrence of abnormal events based on the code submission frequency, the task progress deviation, and a preset abnormal occurrence threshold, and to form a determination result. A determination module, which is connected to the acquisition module and the judgment module respectively, is used to determine the anomaly type of the abnormal event as the requirement change quality type based on the judgment result, the defect re-opening rate and the document change frequency. The actual generation module is connected to the acquisition module and the determination module respectively, and is used to generate an actual risk index based on the demand change quality type and the risk keyword density. An ideal generation module, connected to the acquisition module, is used to classify and cluster the project data of a preset observation data volume according to a preset machine learning model in order to generate an ideal risk index. An adjustment module, which is connected to the actual generation module and the ideal generation module respectively, is used to adjust the preset anomaly occurrence threshold or the preset observation data volume according to the actual risk index and the ideal risk index; An alarm module, connected to the actual generation module, is used to issue an alarm based on the actual risk index and the preset index range regenerated based on the adjusted preset anomaly occurrence threshold or the adjusted preset observation data volume.

2. The intelligent risk assessment and early warning system for project management according to claim 1, characterized in that, The determination module includes: The fitting unit is used to perform linear regression on all the code submission frequencies within a preset first determination time period to obtain the slope of the fitted line. A determination unit, connected to the fitting unit, is used to determine an abnormal event based on the task progress deviation and the preset abnormal event threshold when the slope of the fitted line is less than zero and the absolute value of the slope of the fitted line is greater than a preset standard slope, thereby forming the determination result.

3. The intelligent risk assessment and early warning system for project management according to claim 2, characterized in that, The determination unit includes: The incremental calculation subunit is used to calculate the difference between the task progress deviation at each moment within the preset second determination time and the task progress deviation at the previous moment, and obtain several deviation increments. An accumulation subunit, which is connected to the incremental calculation subunit, is used to calculate the sum of all the positive deviation increments to obtain the incremental accumulation value; A determination subunit, which is connected to the accumulation subunit, is used to determine that an abnormal event has occurred when the incremental accumulation value is greater than zero, and to form the determination result.

4. The intelligent risk assessment and early warning system for project management according to claim 3, characterized in that, The determining module includes: The reopening fluctuation calculation unit is used to calculate the standard deviation of the defect reopening rate at each time from the initial time to a preset time period when the judgment result is formed, and obtain several reopening fluctuation values. The change fluctuation calculation unit is used to calculate the standard deviation of the document change frequency within the preset time period from the initial time to each time within the preset time period when the judgment result is formed, so as to obtain the change fluctuation value; A determination unit, which is connected to the reopening fluctuation calculation unit and the change fluctuation calculation unit respectively, is used to determine the anomaly type of the abnormal event as the demand change quality type based on all the reopening fluctuation values ​​and the change fluctuation values.

5. The intelligent risk assessment and early warning system for project management according to claim 4, characterized in that, The determining unit includes: The change normalization subunit is used to normalize all the reopening fluctuation values ​​to obtain the reopening normalization set, and to normalize all the change fluctuation values ​​to obtain the change normalization set. A consistency calculation subunit, which is connected to the change normalization subunit, is used to calculate the correlation coefficient between the reopened normalization set and the change normalization set to obtain the consistency of fluctuation changes; A subunit is defined, which is connected to the consistency calculation subunit, to determine the anomaly type as the demand change quality type when the fluctuation change consistency is greater than the preset standard consistency.

6. The intelligent risk assessment and early warning system for project management according to claim 5, characterized in that, The actual generation module includes: The density fluctuation calculation unit is used to calculate the density of all the risk keywords within a preset assessment period when the quality type of the demand change is determined, and to obtain the density fluctuation value. A generation unit, connected to the density fluctuation calculation unit, is used to generate the actual risk index based on the density of all risk keywords within the preset evaluation period when the density fluctuation value is greater than a preset fluctuation density threshold.

7. The intelligent risk assessment and early warning system for project management according to claim 6, characterized in that, The generation unit includes: The density normalization subunit is used to normalize the density of all the risk keywords to obtain several normalized densities. A sub-unit is assigned, which is connected to the density normalization sub-unit, to assign a linear time weight to each of the normalized densities, thereby obtaining several computational weights. The mean calculation subunit is connected to the density normalization subunit and the assignment subunit respectively, and is used to calculate a weighted average based on all the normalized densities and the calculation weights to obtain a density-weighted mean. A sub-unit is generated, which is connected to the mean calculation sub-unit, to calculate the product of the density-weighted mean and the preset correction coefficient, and truncate it to the interval [0, 1] to generate the actual risk index.

8. The intelligent risk assessment and early warning system for project management according to claim 7, characterized in that, The adjustment module includes: The index deviation calculation unit is used to calculate the absolute value of the relative deviation between the actual risk index and the ideal risk index to obtain the index deviation. An adjustment unit, connected to the index deviation unit, is used to adjust the preset anomaly occurrence threshold or the preset observation data volume according to the index deviation and the preset index deviation range.

9. The intelligent risk assessment and early warning system for project management according to claim 8, characterized in that, The adjustment unit includes: The first adjustment subunit is used to reduce the preset anomaly occurrence threshold based on the minimum value of the preset index deviation range, the relative deviation of the index deviation, and the preset adjustment coefficient when the index deviation is less than the minimum value of the preset index deviation range. The second adjustment subunit is used to increase the preset observation data volume based on the relative deviation between the exponential deviation and the maximum value of the preset exponential deviation range, and the preset adjustment coefficient, when the exponential deviation is greater than the maximum value of the preset exponential deviation range.

10. The intelligent risk assessment and early warning system for project management according to claim 9, characterized in that, The alarm module includes: The first alarm unit is used to issue a risk recovery notification when the actual risk index is less than the minimum value of the preset index range; The second alarm unit is used to issue a severe risk alarm when the actual risk index is greater than the maximum value of the preset index range.

Images