Intelligent agent enterprise project progress management system and method

Abstract

The invention discloses an agent enterprise project progress management system and method, and the system achieves the intelligent task scheduling, the precise progress prediction and the optimal resource distribution through a unique dynamic task network and reinforcement learning, a space-time attention mechanism and variational auto-encoder, a multi-agent cooperation and game theory and other algorithms. The system comprises four layers of architectures of data acquisition and transmission, processing and analysis, intelligent decision and control and user interaction. Experiments prove that the system and the method effectively reduce the project progress deviation rate, improve the resource utilization rate and the task on-time completion rate, reduce the project cost, and assist large model technology enterprises to efficiently manage the project progress.

Description

Technical Field

[0001] The present invention relates to the technical field of information management, and particularly relates to an intelligent enterprise project progress management system and method. Background Art

[0002] Large model technology R & D projects are characterized by high technical difficulty, long R & D cycles, diverse resource requirements, and complex team collaboration. From data collection and preprocessing to model architecture design, training optimization, and then to model deployment and application, each link is interrelated and full of uncertainties. For example, fluctuations in data quality may lead to poor model training effects, requiring readjustment of the data processing process; during the model training process, limitations in hardware resources may affect training speed and efficiency. Traditional tools such as Gantt charts and critical path methods, although they can intuitively display project tasks and time arrangements, are inadequate when dealing with the dynamic changes and complex dependencies of large model technology projects. They are difficult to reflect in real time the progress deviations caused by technical problems, resource allocation, and other factors in the project, and cannot provide timely and effective decision-making support for project managers. Some project management systems that introduce machine learning algorithms mainly focus on the analysis and prediction of historical data and lack effective response mechanisms for the sudden situations and uncertain factors frequently occurring in large model technology projects. For example, when encountering new technical bottlenecks or external environmental changes, the prediction results based on fixed models often deviate significantly from the actual situation. Due to the uniqueness of large model technology projects, existing algorithms and technologies cannot meet their complex and changeable management requirements. Developing a brand-new algorithm specifically for the characteristics of large model technology projects can better handle key issues such as dynamic task allocation, resource optimization, and real-time risk warning in the project, filling the market gap and enhancing the project management level and competitiveness of enterprises in the large model technology field. Summary of the Invention

[0003] The purpose of the present invention is to provide a large model technology enterprise project progress management system, including: A data collection and transmission module, which deploys various sensors and data collection interfaces, is used to collect project progress, resource usage, and personnel status data in real time, and transmits the collected data to the data processing and analysis module through a wired or wireless communication network; A data processing and analysis module, which receives data from the data collection and transmission layer, cleans, preprocesses, extracts features, and analyzes it, and uses data mining and machine learning technologies to establish a data model to provide data support for intelligent decision-making; An intelligent decision-making and control module, based on a task scheduling algorithm of a dynamic task network and reinforcement learning, a progress prediction algorithm of a spatio-temporal attention mechanism and a variational autoencoder, and a resource optimization allocation algorithm of multi-agent collaboration and game theory, accurately predicts the project progress, schedules tasks, and optimizes resource allocation; The user interaction module provides a human-computer interaction interface for project managers and team members, which can display the project progress, resource usage, and risk warning information in real time. At the same time, it supports users to input information such as project plans, task assignments, and resource requirements to achieve manual intervention and management.

[0004] Furthermore, the task scheduling algorithm of the dynamic task network and reinforcement learning includes: The dynamic task network construction module abstracts project tasks into nodes and task dependencies into edges according to the work breakdown structure of the project, constructs a dynamic task network, and updates the task network in real time to reflect changes in project tasks; The reinforcement learning policy module defines the state space, action space, and reward function of reinforcement learning. The state space covers the task network state, available resources, and real-time project progress data. The action space is the scheduling operation of tasks, and the reward function is designed by combining factors such as project progress, resource utilization rate, and cost control; The algorithm execution and optimization module adopts the deep Q-network reinforcement learning algorithm, and through the experience replay mechanism and the target network, enables the intelligent agent to explore and learn in the dynamic task network and optimize the task scheduling strategy.

[0005] Furthermore, the progress prediction algorithm of the spatio-temporal attention mechanism and variational autoencoder includes: The spatio-temporal data collection and preprocessing module collects the time series data and space-related data of the project, and performs cleaning and normalization preprocessing operations on the data; The spatio-temporal attention mechanism network construction module designs a neural network model based on the spatio-temporal attention mechanism, which includes spatio-temporal attention layers, convolutional layers, and fully connected layers, and captures important features of the data in both time and space dimensions simultaneously; The variational autoencoder fusion and uncertainty estimation module combines the variational autoencoder with the spatio-temporal attention mechanism network, extracts features and reconstructs the project progress data, and estimates the uncertainty of the prediction results.

[0006] Furthermore, the resource optimal allocation algorithm of multi-agent cooperation and game theory includes: The multi-agent system construction module regards the resources and tasks in the project as different agents. Each agent has the ability of autonomous decision-making and learning, and the agents complete the project goals through information interaction and cooperation; The game theory model establishment module constructs a resource allocation game model based on game theory, determines the participants, strategy space, and payoff function of the game, and determines the optimal resource allocation strategy by solving the Nash equilibrium; Collaboration mechanism and dynamic adjustment module, establish a collaboration mechanism among agents. When resource conflicts or task emergencies occur, the agents negotiate and coordinate to reallocate resources and adjust task priorities, and dynamically adjust the resource allocation strategy according to the real-time progress of the project.

[0007] Furthermore, the present invention also provides an enterprise project progress management method based on large model technology, including the following steps: Data collection and transmission step, use sensors and data collection interfaces to collect project progress, resource usage, and personnel status data in real time, and transmit them to the data processing and analysis layer through a communication network; Data processing and analysis step, clean, preprocess, extract features and analyze the collected data, and establish a data model; Task scheduling step, construct a dynamic task network, perform task scheduling based on reinforcement learning, define the state space, action space, and reward function, and the agents explore and learn in the task network to optimize the task scheduling strategy; Progress prediction step, collect and preprocess spatio-temporal data, use the spatio-temporal attention mechanism and variational autoencoder for progress prediction, construct a model and train it, and output the progress prediction result and uncertainty estimation; Resource optimal allocation step, construct a multi-agent system, establish a resource allocation model based on game theory, and achieve the optimal allocation and dynamic adjustment of resources.

[0008] Furthermore, in the task scheduling step, when constructing the dynamic task network, create a work breakdown structure of the project according to project requirements and team resources, determine task dependencies, input the information into the dynamic task network construction module to generate an initial dynamic task network, and obtain the initial state of project resources.

[0009] Furthermore, in the progress prediction step, when cleaning the collected spatio-temporal data, remove outliers and noise, use the Z-score normalization method for normalization processing, and use the cross-entropy loss function and Adam optimizer to adjust the model parameters when training the progress prediction model.

[0010] Furthermore, in the resource optimal allocation step, when constructing the multi-agent system, define the initial attributes and capabilities for resource and task agents. In the resource allocation game model, determine the participants, strategy space, and payoff function, and each agent makes autonomous decisions and learns in the game.

[0011] Furthermore, the user interaction layer uses visual charts to display the dynamic changes of the project progress, and at the same time sets operation buttons for users to perform task allocation and resource adjustment operations.

[0012] Furthermore, the data collection and transmission layer supports docking with the API interfaces of various project management software to achieve automatic data collection and synchronization.

[0013] Beneficial effects: Precise progress monitoring: The progress prediction algorithm based on the spatio-temporal attention mechanism and variational autoencoder can accurately predict the project progress, discover potential progress risks in advance, and provide timely and effective decision-making support for project managers.

[0014] Efficient task scheduling: The task scheduling algorithm of the dynamic task network and reinforcement learning realizes the intelligent scheduling and optimization of tasks, improves the project execution efficiency, and ensures the timely completion of the project.

[0015] Optimized resource allocation: The resource optimization allocation algorithm of multi-agent collaboration and game theory can achieve the optimal allocation of resources according to project requirements and resource status, improve resource utilization rate, and reduce project costs.

[0016] Intelligent risk warning: Through the real-time monitoring and analysis of project progress and resource usage, combined with the uncertainty estimation of the prediction results by the variational autoencoder, it can timely discover the risk factors in the project and issue warnings to help project managers formulate countermeasures in advance. Description of the drawings

[0017] Figure 1 System principle flow chart; Specific implementation manners

[0018] The following further describes the implementation manners of the present invention in detail with reference to the drawings and embodiments. The following embodiments are used to illustrate the present invention, but cannot be used to limit the scope of the present invention.

[0019] Embodiment 1 Data collection and transmission module: Deploy various sensors and data collection interfaces to collect data such as project progress, resource usage, and personnel status in real time. Through wired or wireless communication networks, the collected data is transmitted to the data processing and analysis module. For example, use the API interface of project management software to obtain task progress data, and collect the operating status data of hardware devices through Internet of Things sensors.

[0020] Data processing and analysis module: Clean, preprocess, extract features and analyze the collected data. Adopt data mining, machine learning and other technologies to establish data models, mine the potential laws and correlation relationships in the data, and provide data support for intelligent decision-making. For example, use clustering algorithms to classify project tasks for better task management and resource allocation.

[0021] Intelligent Decision-making and Control Module: Based on the newly proposed hybrid intelligent algorithm, it accurately predicts the project progress, schedules tasks, and optimally allocates resources. According to the real-time data of the project and the calculation results of the algorithm, it automatically adjusts the project plan and resource allocation to achieve intelligent management of the project. For example, when it is predicted that a certain task may be delayed, it automatically adjusts the priorities and resource allocations of subsequent tasks to ensure that the overall project progress is not affected.

[0022] User Interaction Module: Provides a friendly human-computer interaction interface for project managers and team members, and real-time displays the project progress, resource usage, risk warning information, etc. At the same time, it supports users to input information such as project plans, task assignments, and resource requirements to achieve manual intervention and management of the project. For example, it shows the dynamic changes of the project progress through visual charts, which is convenient for managers to intuitively understand the project situation.

[0023] Dynamic Task Network Construction: According to the Work Breakdown Structure (WBS) of the large model technology project, abstract project tasks as nodes and the dependencies between tasks as edges to construct a dynamic task network. Each node contains basic information of the task, such as task name, duration, resource requirements, etc., and the edge represents the sequence and constraint relationships between tasks. Using the real-time data acquisition module, the task network is dynamically updated to timely reflect changes in project tasks, such as the addition, deletion, or adjustment of tasks.

[0024] Reinforcement Learning Strategy Formulation: Define the state space, action space, and reward function of reinforcement learning. The state space includes the current state of the task network, the available situation of resources, the real-time data of the project progress, etc.; the action space is the scheduling operations on tasks, such as starting, pausing, and adjusting the priorities of tasks; the reward function is designed by combining factors such as project progress, resource utilization rate, and cost control, aiming to guide the intelligent agent to make the optimal task scheduling decision. For example, positive rewards are given for completing key tasks on time, and negative rewards are given for resource waste or schedule delays.

[0025] Algorithm Execution and Optimization: Adopt reinforcement learning algorithms such as Deep Q-Network (DQN) to let the intelligent agent continuously explore and learn in the dynamic task network. Through interaction with the environment, the intelligent agent selects the optimal action according to the current state and gradually optimizes the task scheduling strategy. At the same time, introduce the experience replay mechanism and the target network to improve the stability and convergence speed of the algorithm and avoid falling into local optimal solutions.

[0026] Progress Prediction Algorithm Based on Spatiotemporal Attention Mechanism and Variational Autoencoder Space-time Data Collection and Preprocessing: Collect time-series data of large model technology projects, including task progress, resource consumption, technical indicators, etc., as well as space-related data such as the distribution of team members and the location of hardware devices. Perform preprocessing operations such as data cleaning and normalization on the data to remove noise and outliers, and uniformly convert different types of data into a format suitable for model analysis.

[0027] Space-time Attention Mechanism Network Construction: Design a neural network model based on the space-time attention mechanism, which can capture important features of data in both time and space dimensions simultaneously. In the time dimension, the attention mechanism automatically focuses on historical time points that have a greater impact on project progress; in the space dimension, it pays attention to the correlation relationships between different tasks, team members, and resources. The model structure includes multiple space-time attention layers, convolutional layers, and fully connected layers, and through end-to-end training, accurate prediction of project progress is achieved.

[0028] Variational Autoencoder Fusion and Uncertainty Estimation: Combine the variational autoencoder (VAE) with the space-time attention mechanism network to extract features and reconstruct project progress data. The VAE can not only learn the latent feature representation of the data but also estimate the uncertainty of the prediction results. By analyzing the uncertainty, risk warning information is provided for project managers to help them formulate countermeasures in advance.

[0029] Multi-agent System Construction: Regard different resources (such as human resources, computing resources, data storage, etc.) and tasks in the project as different agents, and each agent has the ability of autonomous decision-making and learning. Agents complete project goals through information interaction and cooperation. For example, the human resource agent autonomously selects the tasks to participate in according to the priority of the tasks and its own skill level; the computing resource agent reasonably allocates computing resources according to the computing requirements of the tasks and the load of the current resources.

[0030] Game Theory Model Establishment: Based on game theory, construct a resource allocation game model, and regard the interaction between resource agents and task agents as a game. Each agent pursues the maximization of its own interests in the game while considering the decisions of other agents. By solving the Nash equilibrium of the game model, the optimal resource allocation strategy is determined to achieve efficient utilization of resources. For example, in the computing resource allocation game, the computing resource agent and the task agent reach a balance of interests for both sides by continuously adjusting the resource allocation plan and task requirements.

[0031] Collaboration Mechanism and Dynamic Adjustment: Establish a collaboration mechanism between agents. When resource conflicts or task emergencies occur, agents reallocate resources and adjust task priorities through negotiation and coordination. At the same time, according to the real-time progress and changes of the project, dynamically adjust the resource allocation strategy to ensure that resources are always in the optimal configuration state.

[0032] In the project startup phase, according to project requirements and team resources, use project management tools to create a Work Breakdown Structure (WBS) for the project, break the project into multiple specific tasks, and determine the dependencies between tasks.

[0033] Input the WBS information into the dynamic task network construction module to generate an initial dynamic task network. At the same time, obtain the initial status of project resources through the data acquisition interface, such as the skill distribution of human resources and the configuration of computing resources.

[0034] Initialize the reinforcement learning agent and set the parameters of the state space, action space, and reward function. For example, the dimension of the state space is determined according to the task network and resource information, the action space is defined as common task scheduling operations, and the weights of the reward function are adjusted according to project priorities.

[0035] The agent starts to explore and learn in the dynamic task network, selects actions according to the current state, interacts with the environment to obtain rewards and new states. After a certain number of interactions, store the experience data in the experience replay pool.

[0036] Regularly sample experience data from the experience replay pool to train and optimize the reinforcement learning model. By continuously adjusting the model parameters, make the agent gradually learn the optimal task scheduling strategy.

[0037] During the project execution process, collect the change information of project tasks and resources in real time, and dynamically update the task network and state space. The agent adjusts the task scheduling strategy in real time according to the updated state to ensure the efficient execution of project tasks.

[0038] During the project execution process, through the data acquisition and transmission layer, continuously collect the time series data and space-related data of the project. For example, collect task progress data, resource consumption data, and the work status data of team members at regular intervals.

[0039] Clean and preprocess the collected data, remove outliers and noise, and normalize the data to make it comparable. For example, use the Z - score normalization method to normalize the data.

[0040] Input the preprocessed data into the progress prediction model based on spatio-temporal attention mechanism and variational autoencoder for training. During the training process, use the cross-entropy loss function and the Adam optimizer to adjust the model parameters so that the model can accurately learn the spatio-temporal features and potential laws in the data.

[0041] After the training is completed, the project data collected in real time is input into the model, and the model outputs the prediction results of the project progress. At the same time, the uncertainty of the prediction results is estimated through a variational autoencoder to obtain the confidence interval of the prediction results.

[0042] Based on the prediction results and uncertainty estimation, progress warning information is provided for project managers. For example, when the deviation between the predicted progress and the actual progress exceeds a certain threshold, or the uncertainty of the prediction results is relatively high, a warning is issued in a timely manner to remind the manager to pay attention to the project progress risk.

[0043] In the project startup phase, the resources and tasks in the project are abstracted as agents in a multi-agent system, and initial attributes and capabilities are defined for each agent. For example, attributes such as skill level and work efficiency are defined for the human resource agent, and attributes such as computing power and bandwidth are defined for the computing resource agent.

[0044] According to the requirements of project tasks and the initial state of resources, a resource allocation game model is established to determine the participants, strategy space, and payoff function of the game. For example, in the computing resource allocation game, the participants are the computing resource agent and the task agent, the strategy space is the resource allocation plan and the task requirement adjustment plan, and the payoff function is determined according to the resource utilization rate and task completion situation.

[0045] Each agent makes autonomous decisions and learns in the game model, and by continuously adjusting its own strategy, it pursues the maximization of its own interests. In each game process, the agent updates its own strategy and knowledge according to the decisions and game results of other agents.

[0046] When resource conflicts or task emergencies occur, the agents negotiate and coordinate through a cooperation mechanism. For example, when multiple tasks request the same computing resource at the same time, the relevant agents determine the allocation priority and allocation amount of the resource through negotiation.

[0047] During the project execution process, the resource allocation strategy is dynamically adjusted according to the real-time progress and changes of the project. For example, when a certain task is completed ahead of schedule or delayed, the allocation of relevant resources is adjusted in a timely manner to ensure the efficient utilization of resources and the smooth progress of the project.

[0048] Select an enterprise engaged in large model technology research and development as the experimental object, and deploy the project progress management system proposed by the present invention in multiple projects of the enterprise.

[0049] Select representative large model technology projects, covering different scales, technical difficulties, and application fields. These projects are divided into an experimental group and a control group. The experimental group adopts the management system and method of the present invention, and the control group adopts traditional project management methods.

[0050] Determine the experimental indicators, including project schedule deviation rate, resource utilization rate, project cost, task on-time completion rate, etc. Evaluate the effectiveness of the present invention by comparing the performance of the experimental group and the control group in these indicators.

[0051] Project schedule deviation rate: The project schedule deviation rate of the experimental group decreased by 30% compared with the control group, and the average deviation was controlled within 5%. This indicates that the progress prediction and task scheduling algorithms of the present invention can accurately grasp the project progress, adjust the plan in a timely manner, and effectively reduce schedule delays.

[0052] Resource utilization rate: The human resource utilization rate of the experimental group increased by 25%, and the computing resource utilization rate increased by 30%. The resource optimization allocation algorithm of multi-agent collaboration and game theory can achieve the optimal allocation of resources according to project requirements and resource status, avoiding resource waste.

[0053] Project cost: Due to the effective control of the project schedule and the improvement of resource utilization rate, the project cost of the experimental group decreased by 20% compared with the control group, mainly reflected in the reduction of labor cost, hardware equipment cost and time cost.

[0054] Task on-time completion rate: The task on-time completion rate of the experimental group reached more than 90%, while that of the control group was only about 70%. The intelligent decision-making and dynamic adjustment mechanism of the present invention can ensure the smooth progress of project tasks according to the plan and improve the overall execution efficiency of the project.

[0055] The experimental results show that the project schedule management system and method for large model technology enterprises based on the new hybrid intelligent algorithm proposed by the present invention can significantly improve the efficiency and accuracy of project schedule management, optimize resource allocation, reduce project costs, and has good application prospects and promotion value.

Claims

1. An intelligent enterprise project progress management system, characterized in that: include: The data collection and transmission module deploys various sensors and data collection interfaces to collect project progress, resource usage, and personnel status data in real time, and transmits the collected data to the data processing and analysis module through wired or wireless communication networks; The data processing and analysis module receives data from the data acquisition and transmission module, cleans, preprocesses, extracts features and analyzes the data, and uses data mining and machine learning techniques to establish data models to provide data support for intelligent decision-making; Intelligent decision-making and control module, based on the task scheduling algorithm of dynamic task network and reinforcement learning, the progress prediction algorithm of spatiotemporal attention mechanism and variational autoencoder, and the resource optimization allocation algorithm of intelligent agent collaboration and game theory, to accurately predict the project progress, schedule tasks and optimize resource allocation; The user interaction module provides a human-computer interaction interface for project managers and team members, showing project progress, resource usage, and risk warning information in real time. It also supports users to input project plans, task assignments, and resource demand information to achieve manual intervention and management. The dynamic task network and reinforcement learning task scheduling algorithm include: Dynamic task network construction module, which abstracts project tasks into nodes and task dependencies into edges according to the project's work breakdown structure, builds a dynamic task network, and updates the task network in real time to reflect changes in project tasks; Reinforcement learning strategy module, which defines the state space, action space and reward function of reinforcement learning. The state space covers the task network status, resource availability, and real-time data of project progress. The action space is the scheduling operation of the task. The reward function is designed based on project progress, resource utilization, and cost control factors. The algorithm execution and optimization module adopts the deep Q network reinforcement learning algorithm, and through the experience replay mechanism and target network, allows the intelligent agent to explore, learn and optimize the task scheduling strategy in the dynamic task network.

2. The enterprise project progress management system according to claim 1, characterized in that: The progress prediction algorithm of the spatiotemporal attention mechanism and variational autoencoder includes: The spatiotemporal data acquisition and preprocessing module collects the project's time series data and space-related data, and performs data cleaning and normalization preprocessing operations; The spatiotemporal attention mechanism network building module designs a neural network model based on the spatiotemporal attention mechanism. The model contains spatiotemporal attention layers, convolutional layers, and fully connected layers, which can capture the important features of data in both time and space dimensions. The variational autoencoder fusion and uncertainty estimation module combines the variational autoencoder with the spatiotemporal attention mechanism network to extract and reconstruct the features of project progress data and estimate the uncertainty of the prediction results.

3. The enterprise project progress management system according to claim 1, characterized in that: The resource optimization allocation algorithm of the agent collaboration and game theory includes: The agent system building module treats the resources and tasks in the project as different agents. Each agent has the ability to make independent decisions and learn. Agents achieve project goals through information exchange and collaboration. Game theory model building module: constructs a resource allocation game model based on game theory, determines the game participants, strategy space and profit function, and determines the optimal resource allocation strategy by solving the Nash equilibrium; The collaboration mechanism and dynamic adjustment module establishes a collaboration mechanism between intelligent agents. When resource conflicts or task emergencies occur, the intelligent agents negotiate and coordinate to reallocate resources and adjust task priorities, and dynamically adjust resource allocation strategies based on the real-time progress of the project.

4. A method for enterprise project progress management based on the system of claim 1, characterized in that: The following steps are involved: Data collection and transmission step, using sensors and data collection interfaces to collect project progress, resource usage, and personnel status data in real time, and transmit them to the data processing and analysis module through the communication network; Data processing and analysis steps: cleaning, preprocessing, feature extraction and analysis of the collected data, and establishing a data model; Task scheduling steps: construct a dynamic task network, perform task scheduling based on reinforcement learning, define the state space, action space and reward function, and the agent explores and learns in the task network to optimize the task scheduling strategy; The progress prediction step collects and preprocesses spatiotemporal data, uses the spatiotemporal attention mechanism and variational autoencoder to predict progress, builds and trains the model, and outputs progress prediction results and uncertainty estimates; The resource optimization allocation steps are to build an intelligent system and establish a resource allocation model based on game theory to achieve optimal allocation and dynamic adjustment of resources.

5. The enterprise project progress management method according to claim 4 is characterized in that: In the task scheduling step, when the dynamic task network is constructed, a work breakdown structure of the project is created according to project requirements and team resources, task dependencies are determined, information is input into a dynamic task network construction module to generate an initial dynamic task network, and the initial state of project resources is obtained.

6. The large model technology enterprise project progress management method according to claim 4 is characterized in that: In the progress prediction step, when cleaning the collected spatiotemporal data, outliers and noise are removed, and normalization is performed using the Z-score standardization method. When training the progress prediction model, the cross entropy loss function and the Adam optimizer are used to adjust the model parameters.

7. The enterprise project progress management method according to claim 4, characterized in that: In the resource optimization allocation step, when constructing the intelligent agent system, the initial attributes and capabilities are defined for the resource and task agents. In the resource allocation game model, the participants, strategy space and profit function are determined, and each agent makes autonomous decisions and learns in the game.

8. The enterprise project progress management system according to claim 1, characterized in that: The user interaction layer uses visual charts to display the dynamic changes of project progress, and sets operation buttons for users to perform task allocation and resource adjustment operations.

9. The enterprise project progress management system according to claim 1, characterized in that: The data collection and transmission layer supports docking with the API interfaces of various project management software to achieve automatic data collection and synchronization.

Images