Project cost expenditure prediction analysis method and system

Abstract

The invention discloses a project cost expenditure prediction analysis method and system, and relates to the technical field of project analysis methods. According to the project cost and expenditure prediction analysis method, accurate prediction and resource scheduling of project cost and progress are realized through real-time data acquisition, dynamic model prediction, monitoring alarm and reinforcement learning optimization, and comprehensiveness and timeliness of information are ensured through multi-source data real-time acquisition; the dynamic prediction model provides accurate cost and progress trend prediction; early recognition of problems is realized through threshold comparison and progress judgment; the generation of the resource scheduling scheme based on reinforcement learning avoids the human subjectivity of decision making; the method solves the problems of data isolation, inaccurate prediction and response delay in traditional project management.

Description

Technical Field

[0001] This invention relates to the field of project analysis methods, specifically to a project cost expenditure prediction and analysis method and system. Background Technology

[0002] Project cost forecasting and analysis is used in modern project management. In various engineering projects, software development, and product research and development, accurate forecasting of costs and control of project progress directly affect the project's return on investment and success rate. Traditional project management mainly relies on the project manager's experience and judgment and static analysis based on historical data. By regularly summarizing various expenditure data and comparing them with the preset budget, corrective measures are taken after deviations are found.

[0003] However, existing technologies have the following problems: First, data collection is scattered and lagging, and project data is stored in different management tools, which are independent of each other and make it difficult to achieve real-time data synchronization. Second, the forecasting methods mostly use static models or simple trend extrapolation, which cannot adapt to the dynamic changes during project execution, resulting in insufficient forecast accuracy. Third, there is a delay in problem identification and response, which is usually only discovered after cost overruns or schedule delays have already occurred. Resource scheduling schemes rely on human experience to formulate, which is highly subjective and makes it difficult to find the optimal solution under multiple constraints. In order to address the shortcomings of the existing technology, this invention provides a project cost expenditure prediction and analysis method and system to solve the above problems. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for predicting and analyzing project costs. Through real-time data acquisition, dynamic prediction models, monitoring and alarm systems, and reinforcement learning optimization, it achieves accurate prediction of project costs and schedules, as well as resource scheduling. Real-time acquisition of multi-source data ensures the comprehensiveness and timeliness of information; the dynamic prediction model provides accurate predictions of cost and schedule trends; threshold comparison and schedule judgment enable early identification of problems; and the reinforcement learning-based resource scheduling scheme generation avoids human subjectivity in decision-making. This method solves the problems of isolated data, inaccurate predictions, and delayed response in traditional project management.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting and analyzing project cost expenditures, the method comprising the following steps: Step S1: Collect the current project's runtime data stream in real time from multiple project management tools through a preset application programming interface; Step S2: Based on the running data stream, use a pre-trained dynamic prediction model to calculate the predicted cost and predicted progress of the current project at future time points; Step S3: Compare the predicted cost with a preset cost threshold and determine whether the predicted progress meets the preset progress requirements. If any judgment result is negative, a deviation alarm is generated. Step S4: In response to the deviation alarm, generate a resource scheduling optimization plan for the current project based on the reinforcement learning model, and output the optimization plan to the user interface.

[0006] Preferably, in step S1, the real-time acquisition of the current project's runtime data stream specifically includes: Step S11: Configure data interfaces corresponding to different project management tools to obtain raw data in an event-driven manner; Step S12: Clean and format the raw data, wherein the formatting process includes converting unstructured log text into structured data entries; Step S13: Publish the processed data to the message middleware as a data stream.

[0007] Preferably, in step S2, the pre-trained dynamic prediction model is a long short-term memory network model. The input of the long short-term memory network model is the running data stream after feature engineering, and the output of the long short-term memory network model is the predicted cost sequence and predicted progress sequence for the next N time periods. The dynamic prediction model is trained using historical project data and continuously receives new data streams during the online phase to incrementally update the model parameters.

[0008] Preferably, the cell state update of the Long Short-Term Memory network model at time step t is described by the following formula:

[0009] in: This represents the cell state at the current time t; This represents the cell state at the previous time t-1; This represents the forget gate, which controls the degree to which the previous state is forgotten; This represents the input gate, which controls the degree to which new candidate values ​​are input; This indicates the current state of the candidate unit.

[0010] Preferably, in step S3, determining whether the predicted progress meets the preset progress requirements specifically involves: The schedule deviation index is calculated based on the predicted schedule sequence. When the schedule deviation index exceeds the preset schedule tolerance, it is determined that the schedule requirements are not met. The schedule deviation index D is calculated as follows:

[0011] in: The actual completion time predicted by the model; The specified completion time for the project plan.

[0012] Preferably, in step S3, while generating the deviation alarm, the following is also performed: Based on the principle of PID controller, the initial adjustment required to bring the project back to the budget and schedule is calculated; the input of the PID controller is the cost deviation and schedule deviation, and the output is the proposed adjustment amount for resource input.

[0013] Preferably, in step S4, the generation of a resource scheduling optimization scheme based on a reinforcement learning model specifically includes: Step S41: Define the current project state as the state space of the reinforcement learning model, wherein the state space includes, but is not limited to, the amount of work completed, the remaining budget, and the current manpower allocation; Step S42: Define possible adjustment actions as action space, which includes adding or removing personnel with specific roles, extending or shortening the iteration cycle, and reallocating task priorities; Step S43: The reinforcement learning model explores and selects the action that can obtain the maximum cumulative reward as the optimization scheme in the action space according to the preset reward function. The reward function is positively correlated with cost savings and progress advancement.

[0014] Preferably, the method further includes: The final adopted optimization scheme and its actual effect data after implementation are used as new training samples and stored in the experience replay pool for periodic offline training of the reinforcement learning model to optimize the decision-making strategy of the reinforcement learning model.

[0015] Preferably, the method further includes the following after step S4: Step S5: Visualize and render the predicted costs, prediction progress, deviation alarms, and optimization schemes, and push them to a unified web dashboard for display.

[0016] A project cost forecasting and analysis system is provided for implementing the project cost forecasting and analysis method. The system includes: The data acquisition module is used to collect real-time running data streams of the current project from multiple project management tools through a preset application programming interface; The predictive analysis module is communicatively connected to the data acquisition module and is used to calculate the predicted cost and predicted progress of the current project at future time points based on the running data stream and using a pre-trained dynamic prediction model. The monitoring and alarm module is connected in communication with the predictive analysis module. It is used to compare the predicted cost with a preset cost threshold and determine whether the predicted progress meets the preset progress requirements. If any judgment result is negative, a deviation alarm is generated. The optimization decision module is communicatively connected to the monitoring and alarm module, and is used to generate a resource scheduling optimization scheme for the current project based on a reinforcement learning model in response to the deviation alarm. The human-computer interaction module is communicatively connected to the optimization decision module and is used to output the optimization scheme to the user interface.

[0017] The technical effects and advantages of this invention are as follows: 1. The project's cost forecasting and analysis method achieves accurate prediction of project costs and schedules and resource scheduling through real-time data collection, dynamic prediction models, monitoring and alarms, and reinforcement learning optimization. Real-time collection of multi-source data ensures the comprehensiveness and timeliness of information; the dynamic prediction model provides accurate predictions of cost and schedule trends; threshold comparison and schedule judgment enable early identification of problems; and the resource scheduling scheme generation based on reinforcement learning avoids human subjectivity in decision-making. This method solves the problems of data isolation, inaccurate prediction, and delayed response in traditional project management.

[0018] 2. The project's cost forecasting and analysis method improves data quality and forecast accuracy by establishing a standardized data cleaning process and adopting an advanced LSTM forecasting model. The combination of standardized data cleaning and the specific implementation of the LSTM model eliminates data noise interference and utilizes the spatiotemporal characteristics of time series data, making cost and schedule forecasts closely resemble the actual project evolution, thus providing reliable data support for subsequent decision-making.

[0019] 3. The project's cost forecasting and analysis method adopts a two-layer optimization mechanism combining reinforcement learning and PID control. This enables adaptive optimization of project resource scheduling. The PID controller provides initial adjustment suggestions with a fast response, while the reinforcement learning model conducts in-depth optimization exploration in a more complex decision space. The two complement each other, ensuring both the timeliness of the adjustment plan and the optimality of long-term decisions. At the same time, the decision-making strategy is continuously improved through an experience playback mechanism, enabling the system to have the ability to continuously evolve and update. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a flowchart of the data acquisition sub-steps of the present invention; Figure 3 This diagram illustrates the training and application of the prediction model of this invention. Figure 4 This is a diagram illustrating the monitoring and alarm logic of the present invention. Figure 5 This is a flowchart illustrating the reinforcement learning optimization process of this invention. Detailed Implementation

[0022] 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.

[0023] This embodiment discloses a method for predicting and analyzing project expenses, based on the appendix... Figure 1 To be continued Figure 5 As shown, the method includes the following steps: Step S1: Collect the current project's runtime data stream in real time from multiple project management tools through a preset application programming interface; Step S2: Based on the running data stream, use a pre-trained dynamic prediction model to calculate the predicted cost and predicted schedule for the current project at future time points; Step S3: Compare the predicted cost with the preset cost threshold and determine whether the predicted progress meets the preset progress requirements. If any judgment result is negative, a deviation alarm is generated. Step S4: In response to the deviation alarm, generate a resource scheduling optimization plan for the current project based on the reinforcement learning model, and output the optimization plan to the user interface; This method achieves accurate prediction of project costs and schedules and resource scheduling through real-time data acquisition, dynamic prediction models, monitoring and alarms, and reinforcement learning optimization. Step S1 collects real-time project operation data streams from multiple project management tools through a preset application programming interface (API). This data includes task completion status, resource consumption, and time records. Step S2 processes the data streams using a pre-trained dynamic prediction model and outputs predicted costs and schedules for future time points. Step S3 compares the predicted costs with preset cost thresholds, such as budget limits, and simultaneously determines whether the predicted schedule meets preset schedule requirements, such as deadlines. If any condition is not met, a deviation alarm is generated. Step S4 responds to the alarm by generating a resource scheduling optimization plan based on the reinforcement learning model, such as adjusting human resources or task priorities, and outputs it to the user interface. This method enables real-time monitoring and prediction of project management, can detect risks of cost overruns or schedule delays in advance, and respond automatically through intelligent optimization solutions, thereby improving the initiative and efficiency of project management; the integration of multiple data sources ensures the comprehensiveness of the data, while the introduction of reinforcement learning makes resource scheduling more adaptive and optimized. In step S1, the real-time collection of the current project's runtime data stream specifically includes: Step S11: Configure data interfaces corresponding to different project management tools to obtain raw data in an event-driven manner; Step S12: Clean and format the raw data. The formatting process includes converting unstructured log text into structured data entries. Step S13: Publish the processed data to the message middleware as a data stream; Step S11: Configure data interfaces corresponding to different project management tools, and use an event-driven approach to obtain raw data, such as Webhook or message queue, to ensure timely data updates; Step S12: Clean and format the raw data, including removing duplicate values, handling missing values, standardizing data formats, and converting unstructured log text (such as development logs or meeting minutes) into structured data entries (such as JSON or CSV format); Step S13: Publish the processed data as a data stream to a message middleware, such as Kafka or RabbitMQ, for subsequent modules to consume; Data cleaning and formatting improve data quality and reduce the impact of noise on prediction; the use of message middleware enables asynchronous processing and efficient transmission of data streams, supports high-concurrency scenarios, and enhances the scalability and reliability of the system.

[0024] In step S2, the pre-trained dynamic prediction model is a long short-term memory network model. The input of the long short-term memory network model is the running data stream after feature engineering, and the output of the long short-term memory network model is the predicted cost sequence and the predicted progress sequence for the next N time periods. The dynamic prediction model is trained through historical project data and continuously receives new data streams during the online phase to incrementally update the model parameters. The pre-trained dynamic prediction model used in step S2 is a Long Short-Term Memory (LSTM) network model. LSTM is a recurrent neural network that excels at processing time series data. The input data is a running data stream after feature engineering, including feature selection (such as selecting key features like the number of tasks, working hours, and costs) and feature scaling (such as normalization). The LSTM model outputs a predicted cost sequence and a predicted schedule sequence for the next N time periods (such as N=7 days). The model is trained using historical project data (such as cost and schedule data from similar past projects) and continuously receives new data streams for incremental updates during the online phase (such as using online learning algorithms) to adapt to project changes. LSTM models can capture long-term dependencies in time series, improving prediction accuracy; incremental updates enable the model to adapt to dynamic changes in the project, avoiding model obsolescence and ensuring the real-time nature and accuracy of predictions.

[0025] The cell state update of a Long Short-Term Memory (LSTM) network model at time step t is described by the following formula:

[0026] in; It represents the cell state at the current time t, the core of "long-term memory", storing key historical information, such as the trend of human resource costs in the previous 8 weeks; This represents the cell state at the previous time t-1, used to inherit historical information; This represents the forget gate, which controls the degree to which the previous state is forgotten. The calculation formula is: ,in This is the sigmoid function (outputs 0-1). This is the weight matrix. This is the hidden state from the previous moment ("short-term memory"). Input features for the current time step. This is a bias term. The closer the value is to 1, the better. The more historical information is retained (e.g., when project costs are stable in the early stages, the forgetting threshold is close to 1, indicating the retention of historical cost trends); the closer it is to 0, the more it indicates that historical information has been forgotten (e.g., when suppliers are changed midway through a project, material costs change abruptly, the forgetting threshold is close to 0, indicating the abandonment of old cost trends). This represents the input gate, which controls the degree to which new candidate values ​​are input. The calculation formula is... Similarly, it outputs 0-1 to control new information. The input ratio, The closer to 1, the easier it is for new information to enter the unit state (e.g., when adding a senior developer, the input gate is close to 1, and the new labor cost information is included). This represents the current state of the candidate unit; the calculation formula is: The tanh function outputs -1 to 1, generating candidate new information for the current moment (such as the predicted cost trend for this week based on current manpower and material data). By using the forget gate to "filter history" and the input gate to "accept new information," the unit state... It can dynamically retain information useful for prediction (such as stable labor costs) and discard useless information (such as one-time equipment maintenance costs), thereby accurately modeling the long-term variation of project costs and schedule.

[0027] The forget gate and input gate are calculated by the sigmoid function, and the candidate state is calculated by the tanh function. Together, they determine the retention and updating of information. This formula reflects the gating mechanism of LSTM, which can selectively memorize important information and avoid the gradient vanishing problem. It is particularly suitable for processing long sequence data and can effectively capture the long-term trends of cost and schedule in project prediction.

[0028] In step S3, determining whether the predicted progress meets the preset progress requirements specifically involves: The schedule deviation index is calculated based on the predicted schedule sequence. When the schedule deviation index exceeds the preset schedule tolerance, it is determined that the schedule requirements are not met. The schedule deviation index D is calculated as follows:

[0029] in; The actual completion time predicted by the model; The specified completion time for the project plan; The schedule deviation index provides a quantitative standard for schedule assessment, making the judgment more objective and accurate; through tolerance settings, it allows for a certain degree of flexibility, reduces false alarms, and improves the usability of alarms. In step S3, while generating the deviation alarm, the following is also performed: Based on the principle of PID controller, the initial adjustment required to bring the project back to the budget and schedule is calculated; the input of the PID controller is the cost deviation and schedule deviation, and the output is the proposed adjustment amount for resource input. While generating deviation alarms, the system also calculates the initial adjustment amount based on the principle of the PID controller. The PID controller (proportional-integral-derivative controller) takes cost deviation (the difference between predicted cost and cost threshold) and schedule deviation (the difference between predicted schedule and planned schedule) as inputs and outputs the suggested adjustment amount for resource input. The proportional term (P) processes the current deviation, the integral term (I) accumulates the historical deviation, and the derivative term (D) predicts the future deviation changes, which together generate the adjustment amount. PID controllers provide a stable and responsive control strategy that can smoothly adjust resources, avoid overreaction, and ensure that projects gradually return to budget and schedule.

[0030] In step S4, generating a resource scheduling optimization scheme based on the reinforcement learning model specifically includes: Step S41: Define the current project state as the state space of the reinforcement learning model. The state space includes, but is not limited to, the amount of work completed, the remaining budget, and the current manpower allocation. Step S42: Define possible adjustment actions as action space, which includes adding or removing personnel with specific roles, extending or shortening the iteration cycle, and reallocating task priorities; Step S43: The reinforcement learning model explores and selects the action that can obtain the maximum cumulative reward as the optimization scheme in the action space according to the preset reward function. The reward function is positively correlated with cost saving and progress advancement. Step S4 details the process of generating a resource scheduling optimization scheme based on a reinforcement learning model. Step S41 defines the current project state as the state space, including variables such as completed workload, remaining budget, and current manpower allocation. Step S42 defines possible adjustment actions as the action space, including adding or removing personnel for specific roles, extending or shortening the iteration cycle, and reallocating task priorities. In step S43, the reinforcement learning model explores and selects the action that yields the maximum cumulative reward as the optimization scheme based on a preset reward function, such as Q-learning or deep reinforcement learning. The reward function is positively correlated with cost savings and schedule advancement, such as reward = α × cost savings + β × schedule advancement, where α and β are weights. Reinforcement learning can learn the optimal strategy through trial and error, adapting to complex and ever-changing project environments; the setting of the reward function ensures that the optimization scheme takes into account both cost and schedule, achieving multi-objective optimization.

[0031] The method also includes: The final adopted optimization scheme and its actual performance data after implementation are used as new training samples and stored in the experience replay pool for periodic offline training of the reinforcement learning model to optimize the decision-making strategy of the reinforcement learning model. Experience replay and offline training improve the stability and efficiency of the reinforcement learning model and avoid overfitting. Continuous learning enables the model to adapt to new situations and continuously improve the quality of the optimization scheme. The method further includes the following after step S4: Step S5: Visualize the predicted costs, predicted progress, deviation alerts, and optimization solutions, and push them to a unified web dashboard for display. Visualization makes complex data easier to understand and supports decision-makers to respond quickly; the web dashboard provides a centralized monitoring interface, enhancing the user experience.

[0032] A project cost forecasting and analysis system is provided for implementing project cost forecasting and analysis methods. The system includes: The data acquisition module is used to collect real-time running data streams of the current project from multiple project management tools through a preset application programming interface; The predictive analysis module communicates with the data acquisition module and is used to calculate the predicted cost and predicted progress of the current project at future time points based on the running data stream and using a pre-trained dynamic prediction model. The monitoring and alarm module communicates with the predictive analysis module and is used to compare the predicted cost with the preset cost threshold and determine whether the predicted progress meets the preset progress requirements. If any judgment result is negative, a deviation alarm is generated. The optimization decision module communicates with the monitoring and alarm module to respond to deviation alarms and generate resource scheduling optimization schemes for the current project based on a reinforcement learning model. The human-computer interaction module communicates with the optimization decision module and is used to output the optimization solution to the user interface. The system includes a data acquisition module, a predictive analysis module, a monitoring and alarm module, an optimization decision-making module, and a human-computer interaction module. These modules work together through communication connections, such as REST API or message queues. The modular design makes the system easy to expand and maintain. The specialization of each module improves processing efficiency and reliability.

[0033] Example 1: This example uses a software development project to illustrate the workflow in detail; The workflow is as follows: Data acquisition (step S1): The project uses Jira and GitLab as project management tools; by configuring API interfaces, data streams are collected in real time, including the number of tasks completed, the number of code commits, and work hours (e.g., 8 hours of work per day). Raw data cleaning: Remove invalid records (such as entries with 0 hours) and format log text (such as converting GitLab commit messages into structured data); Data is published to the Kafka message queue, processing 100 data records per second; Predictive analysis (step S2): Prediction is performed using a pre-trained LSTM model; input features include: percentage of tasks completed (currently 70%), cumulative cost (50,000 yuan already spent), and remaining time (planned total duration of 100 days, 70 days already completed). The model outputs a predicted cost sequence and a predicted progress sequence for the next 7 days; for example, the predicted cost is: 5,200 yuan on day 71, 5,300 yuan on day 72, ..., 5,800 yuan on day 77; the predicted progress is: 80% completed on day 77. Incremental model update: The parameters are updated daily using new data, and the learning rate is set to 0.001; Monitoring Alarm (Step S3): The cost threshold is set at a total budget of 100,000 yuan, and the current projected total cost is 105,000 yuan (overspending). Schedule requirements: 100% completion is planned for day 100, and 80% is projected for day 77. Calculate the schedule deviation index D: =100 / 0.8=125 days (estimated actual completion time). =100 days; D = |125-100| / 100 = 0.25 (exceeding the tolerance of 0.1), therefore the schedule requirement is not met; Generate a deviation alarm and use a PID controller to calculate the initial adjustment amount: cost deviation of 5,000 yuan, schedule deviation of 25 days, and the PID output suggests adding 2 more developers. Optimize decision-making (step S4): Reinforcement learning model state space: 70% of the workload has been completed, the remaining budget is 50,000 yuan, and the current workforce is 5 people; Action space: Increase the number of developers, shorten the iteration cycle from 2 weeks to 1 week, and reallocate task priorities; Reward function: Reward = 0.6 × Cost savings + 0.4 × Schedule advancement (unit: yuan / day); Action selected after model exploration: Add 2 developers and prioritize critical tasks; The optimization plan is output to the web interface: It is recommended to add 2 developers within 3 days, which is expected to save 3,000 yuan and advance the schedule by 10 days; Visualization (Step S5): All data is rendered on a web dashboard: displaying forecast cost curves, progress bars, alert icons, and optimization details, with users able to interactively adjust parameters; Example 2: This example uses a construction project to illustrate the workflow in detail; The workflow is as follows: Data acquisition (step S1): The project uses Procore and Excel as management tools; data streams are collected via API, including material consumption, worker hours, and project progress percentage. Raw data cleaning: handling missing values ​​(e.g., filling with the average value), formatting Excel spreadsheets as JSON data; Data is published to the RabbitMQ message middleware, which processes 50 data entries per minute. Predictive analysis (step S2): LSTM model input features: 60% of the project has been completed, the cumulative cost is 800,000 yuan, and the remaining time (the planned total project duration is 200 days, and 120 days have been completed). The model outputs a predicted cost sequence and a predicted progress sequence for the next 30 days; for example, the predicted cost is: 10,000 yuan on day 121, 10,500 yuan on day 122, ..., 12,000 yuan on day 150; the predicted progress is: 70% completed on day 150. Model incremental update: Updated weekly, using data from the past 7 days; Monitoring Alarm (Step S3): The cost threshold is set at a total budget of 1,500,000 yuan, while the current projected total cost is 1,600,000 yuan (overspending). Schedule requirements: 100% completion is planned for day 200, and 70% completion is projected for day 150. Calculate the schedule deviation index D: =200 / 0.7≈285.7 days =200 days; D = |285.7 - 200 | / 200 ≈ 0.4285 (exceeding the tolerance of 0.1), therefore the schedule requirement is not met; A deviation alarm is generated, and the PID controller calculates the initial adjustment amount: cost deviation of 100,000 yuan, schedule deviation of 85.7 days, and outputs a suggestion to increase workers and equipment; Optimize decision-making (step S4): Reinforcement learning model state space: 60% of the work has been completed, the remaining budget is 700,000 yuan, and there are currently 20 workers. Action options: increase workers, extend working hours, adjust material procurement order; Reward function: Reward = 0.5 × Cost Savings + 0.5 × Schedule Advancement; Model selected action: Add 5 workers and extend daily working hours by 2 hours, expected cost savings of 50,000 yuan, schedule advance by 30 days; The optimized solution is output to the user interface: It is recommended to add 5 workers next week, working 10 hours a day, but attention should be paid to fatigue management; Visualization (Step S5): The web dashboard displays Gantt charts, cost bar charts, and optimization plans, which project managers can monitor and approve in real time.

[0034] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A project cost expenditure forecast analysis method characterized by, The method comprises the following steps: Step S1: Real-time collection of running data streams of a current project from multiple project management tools through a preset application program interface; Step S2: Based on the running data streams, a pre-trained dynamic prediction model is used to calculate the predicted cost and predicted progress of the current project at a future time point; Step S3: Comparing the predicted cost with a preset cost threshold, and determining whether the predicted progress meets a preset progress requirement, and if any of the determination results is no, a deviation alarm is generated; Step S4: In response to the deviation alarm, a resource scheduling optimization scheme for the current project is generated based on a reinforcement learning model, and the optimization scheme is output to a user interface.

2. The project cost expenditure forecast analysis method according to claim 1, characterized in that, In step S1, the real-time collection of the running data streams of the current project specifically comprises: Step S11: Configuring data interfaces corresponding to different project management tools to obtain raw data in an event-driven manner; Step S12: Cleaning and formatting the raw data, wherein the formatting process includes converting unstructured log text into structured data entries; Step S13: Publishing the processed data to a message middleware in the form of a data stream.

3. The project cost expenditure forecast analysis method according to claim 2, wherein, In step S2, the pre-trained dynamic prediction model is a long short-term memory network model, the input of the long short-term memory network model is the running data stream after feature engineering processing, and the output of the long short-term memory network model is a predicted cost sequence and a predicted progress sequence of N future time periods; the dynamic prediction model is trained by historical project data and continuously receives new data streams for incremental updating of model parameters in the online stage.

4. The project cost expenditure forecast analysis method according to claim 3, wherein, The unit state update of the long short-term memory network model at time step t is described by the following formula: Wherein: a unit state representing a current time t; represents the cell state at the previous time t-1; denotes a forget gate, controlling the degree of forgetting the previous state; represents an input gate that controls the degree of input of the new candidate value; a candidate cell state representing a current time instant.

5. The project cost expenditure forecasting analysis method according to claim 1, wherein, In step S3, the determination of whether the predicted progress meets the preset progress requirement specifically comprises: Calculating a progress deviation index according to the predicted progress sequence, and determining that the progress requirement is not met when the progress deviation index exceeds a preset progress tolerance; The calculation method of the progress deviation index D is as follows: Wherein: actual completion time for the model prediction; To specify the completion time for a project plan.

6. The project cost expenditure forecast analysis method according to claim 5, wherein, In step S3, while generating the deviation alarm, the following is also performed: Based on the principle of a PID controller, the initial adjustment amount required to make the project return to the budget and progress plan is calculated; the input of the PID controller is the cost deviation and the progress deviation, and the output is the adjustment suggestion amount of resource input.

7. The project cost expenditure forecasting analysis method of claim 1, wherein, In step S4, the generation of the resource scheduling optimization scheme based on the reinforcement learning model specifically comprises: Step S41: Defining the current project state as a state space of the reinforcement learning model, wherein the state space includes but is not limited to completed work amount, remaining budget, and current manpower allocation; Step S42: Defining possible adjustment actions as an action space, wherein the action space includes increasing and decreasing specific role personnel, extending and shortening iteration period, and reallocating task priority; Step S43: The reinforcement learning model explores and selects an action that can obtain the maximum cumulative reward in the action space as the optimization scheme according to a preset reward function, wherein the reward function is positively correlated with cost saving and progress advancement.

8. The project cost expenditure forecast analysis method according to claim 7, wherein, The method further comprises: The finally adopted optimization scheme and actual effect data after its execution are stored as new training samples into an experience replay pool for periodic offline training of the reinforcement learning model to optimize the decision-making strategy of the reinforcement learning model.

9. The project cost expenditure forecasting analysis method of claim 1, wherein, The method further comprises, after step S4: Step S5: visualizing and rendering the predicted cost, predicted progress, deviation alarm and optimization scheme, and pushing them to a unified Web dashboard for display.

10. A project cost expenditure forecast analysis system for implementing the method of any one of claims 1-9, characterized by, The system comprises: A data acquisition module configured to acquire, in real time, a running data stream of a current project from a plurality of project management tools through a preset application program interface; A prediction analysis module in communication connection with the data acquisition module and configured to calculate, based on the running data stream, a predicted cost and a predicted progress of the current project at a future time point by using a pre-trained dynamic prediction model; A monitoring and alarming module in communication connection with the prediction analysis module and configured to compare the predicted cost with a preset cost threshold, and determine whether the predicted progress meets a preset progress requirement, and generate a deviation alarm if any of the determination results is negative; An optimization decision module in communication connection with the monitoring and alarming module and configured to generate, in response to the deviation alarm, a resource scheduling optimization scheme for the current project based on a reinforcement learning model; A human-computer interaction module in communication connection with the optimization decision module and configured to output the optimization scheme to a user interface.

Images