Project full-life-cycle intelligent management method, device, equipment and medium

Abstract

The invention discloses a project full-life-cycle intelligent management method and device, equipment and a medium. The method comprises the following steps: acquiring multi-source heterogeneous data in a full life cycle of a project to obtain original multi-source data of the project; preprocessing the original multi-source data of the project, and generating comprehensive project state data through cross-system fusion processing; and based on the comprehensive project state data, performing calculation processing through a reinforcement learning decision framework to obtain a management strategy of each stage, wherein the reinforcement learning decision framework comprises the steps of predicting a time sequence change trend by adopting a GRU model of an embedded model migration sub-module, performing risk assessment by adopting a Bayesian network, and performing calculation by adopting a PPO algorithm after an objective function is adjusted. And actual data and a prediction result can be obtained for comparative analysis, and a management strategy is iteratively optimized. According to the method, the problems of fragmentation and stage splitting of traditional management data are solved, accurate decision support is provided, intelligent management is promoted, and management efficiency and decision scientificity are improved.

Description

Technical Field

[0001] This invention relates to the field of project management technology, and in particular to a method, apparatus, equipment and medium for intelligent management of the entire project lifecycle. Background Technology

[0002] With the continuous development of project lifecycle management, technologies such as the Internet of Things and artificial intelligence are gradually penetrating management scenarios, driving the transformation of project management models from traditional manual-driven to technology-assisted. Although these technologies have basic data collection capabilities, they have not yet formed a mature full-cycle management system, and traditional project lifecycle management methods remain the main application in the industry.

[0003] In traditional technologies, project lifecycle management often adopts a phased, manual control model. At the data level, it relies solely on manually entered structured data such as schedules and costs, while unstructured data such as images and text are not included in the management system. Multi-source data is aggregated through basic project management software, lacking effective integration methods. At the phase management level, data from each phase—planning, design, construction, operation, and maintenance—are stored independently, management strategies are not interconnected, and a data feedback mechanism is not established, making it difficult to form a complete management loop.

[0004] However, current traditional methods have two major problems. First, they lack the ability to collect and integrate data, which cannot cover multi-source heterogeneous data. The data is severely fragmented, making it difficult to generate a comprehensive representation of the project status, resulting in a lack of complete data support for decision-making. Second, the management of the entire life cycle is fragmented, with no data exchange or closed-loop feedback mechanism between stages, making it impossible to achieve continuous improvement in project management. Summary of the Invention

[0005] Based on this, it is necessary to provide a method, device, equipment, and medium for intelligent management of the entire project lifecycle that can integrate multi-source heterogeneous data throughout the project's entire lifecycle to achieve effective cross-system fusion, break down the fragmented management at each stage, and establish a data closed-loop feedback mechanism, thereby generating precise management strategies for each stage and supporting continuous iterative optimization.

[0006] Firstly, this application provides a smart management method for the entire lifecycle of a project, including:

[0007] Acquire multi-source heterogeneous data throughout the entire project lifecycle to obtain the project's original multi-source data;

[0008] The original multi-source data of the project is preprocessed to generate preprocessed multi-source data;

[0009] The preprocessed multi-source data is fused across systems to obtain comprehensive project status data.

[0010] Based on comprehensive project status data, a reinforcement learning decision framework is used to calculate and process the data to obtain management strategies for each stage.

[0011] In one embodiment, management strategies for each stage are obtained by processing comprehensive project status data through a reinforcement learning decision framework, including:

[0012] Based on comprehensive project status data, a GRU model with an embedded model transfer submodule is used to predict the temporal change trend of each stage of the project, and the prediction results are obtained.

[0013] A Bayesian network is used to conduct a risk assessment based on the integrated project status data and prediction results, and the risk analysis results are obtained.

[0014] By adjusting the objective function, the PPO algorithm calculates the comprehensive project status data, prediction results, and risk analysis results to obtain management strategies for each stage.

[0015] In one embodiment, a Bayesian network is used to perform a risk assessment on the integrated project status data and prediction results, yielding risk analysis results, including:

[0016] Data encoding and probability mapping are performed on the comprehensive project status data and prediction results to obtain risk evidence nodes and time-series risk probability data.

[0017] Risk evidence nodes and time-series risk probability data are input into a Bayesian network, and probabilistic inference is performed using the Gibbs sampling algorithm to obtain the probability distribution of each risk factor.

[0018] The probability distribution of each risk factor and the priority data of the risk are calculated by pre-set risk impact weights, and a risk priority assessment matrix is ​​obtained.

[0019] By inputting comprehensive project status data and prediction results into the risk priority assessment matrix, a risk heat map and a set of response strategy suggestions are obtained.

[0020] By integrating the risk heatmap and the set of response strategy recommendations, the risk analysis results are obtained.

[0021] In one embodiment, the original multi-source data of the project is preprocessed to generate preprocessed multi-source data, including:

[0022] The unstructured data in the original multi-source data of the project is denoised to obtain the denoised multi-source data.

[0023] The denoised multi-source data is normalized to obtain standardized multi-source data;

[0024] By using semantic association mapping rules between multi-source data, the standardized multi-source data is semantically integrated to generate an integrated multi-source dataset;

[0025] Feature extraction is performed on the integrated multi-source dataset, and a preset feature selection algorithm is used to select feature indicators that are strongly correlated with the project status, thus obtaining preprocessed multi-source data.

[0026] In one embodiment, the preprocessed multi-source data undergoes cross-system fusion processing to obtain comprehensive project status data, including:

[0027] The preprocessed multi-source data is input into the Transformer multimodal fusion model, and the cross-modal feature association weights are calculated through the multi-head attention mechanism to obtain the preliminary fused feature vector.

[0028] The feature vectors are initially fused and connected to cross-system data sources to obtain a set of external data sources. The set of external data sources is used to make up for the shortcomings of internal data and cover heterogeneous data to improve the external related data set for decision-making data support.

[0029] The initial fused feature vectors are aligned with the feature dimensions of the external data source set to obtain the dataset to be fused.

[0030] The dataset to be fused is non-linearly mapped using a fully connected layer and an activation function to generate comprehensive project status data.

[0031] In one embodiment, after processing the comprehensive project status data using a reinforcement learning decision framework to obtain management strategies for each stage, the method further includes:

[0032] Obtain the actual data generated during the execution of each stage of the project, compare and analyze the actual data with the execution results corresponding to the management strategies predicted based on historical data, and obtain the prediction deviation dataset.

[0033] Using preset strategy effectiveness evaluation indicators, a difference analysis was performed on the prediction deviation dataset to obtain information on strategy modules that failed to achieve the expected goals and their corresponding influencing factors.

[0034] The weights of the objective function of the PPO algorithm in the reinforcement learning decision framework are adjusted according to the influence factors to obtain the adjusted objective function.

[0035] Based on the adjusted objective function, the risk assessment parameters of the Bayesian network are corrected to obtain the adjusted parameter data.

[0036] The adjusted parameter data is input into the reinforcement learning decision framework to obtain the iteratively optimized management strategy.

[0037] In one embodiment, multi-source heterogeneous data throughout the entire project lifecycle is acquired to obtain the project's original multi-source data, including:

[0038] Acquire multi-source heterogeneous data throughout the entire project lifecycle through sensor networks;

[0039] Analyze multi-source heterogeneous data and classify it into structured data and unstructured data;

[0040] The collected structured and unstructured data are time-stamped and identified by data source identification, preserving the original data attributes and collection context information to obtain the original multi-source data of the project.

[0041] Secondly, this application also provides a smart management device for the entire life cycle of a project, comprising:

[0042] The data acquisition module is used to acquire multi-source heterogeneous data throughout the entire project lifecycle to obtain the original multi-source data of the project.

[0043] The preprocessing module is used to preprocess the original multi-source data of the project and generate preprocessed multi-source data.

[0044] The data fusion module is used to perform cross-system fusion processing on preprocessed multi-source data to obtain comprehensive project status data;

[0045] The learning decision module is used to calculate and process comprehensive project status data through a reinforcement learning decision framework to obtain management strategies for each stage.

[0046] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods in the first aspect of this application.

[0047] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods in the first aspect of this application.

[0048] The intelligent project lifecycle management method provided in this application obtains original multi-source project data by acquiring multi-source heterogeneous data throughout the project lifecycle. This breaks away from the limitations of traditional management, which relies solely on manually entering structured data such as schedules and costs, by including unstructured data such as images and text in the collection scope, achieving comprehensive coverage of project lifecycle data. Preprocessing the original multi-source project data effectively removes unstructured data noise and standardizes the data. Semantic association mapping and feature filtering are used to extract key information, significantly improving data quality and laying a solid foundation for subsequent processing. The preprocessed multi-source data is then fused across systems. Generating comprehensive project status data can break down data barriers between systems, solve the problems of data fragmentation and independent storage of data at each stage in traditional management, and form a unified and complete understanding of project status. Based on the comprehensive project status data, management strategies for each stage are calculated through a reinforcement learning decision framework, which can realize the linkage and synergy of strategies at each stage, break the fragmented state of traditional phased manual control, provide precise decision support for management, promote the transformation of project management to a technology-driven intelligent model, improve management efficiency and decision-making scientificity, ensure the smooth progress of each stage of the project, and at the same time make up for the shortcomings of traditional management in lacking complete data support and continuous improvement capabilities. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 A schematic diagram of an implementation environment provided for one embodiment of the present invention;

[0051] Figure 2 A flowchart of a project lifecycle intelligent management method according to one embodiment of the present invention;

[0052] Figure 3 This is a schematic diagram of the intelligent management device for the entire life cycle of a project, as shown in one embodiment of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0054] The intelligent project lifecycle management method provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 100 communicates with sensor 101 via a network. Terminal 100 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. Sensors can be, but are not limited to, temperature sensors, pressure sensors, vibration sensors, displacement sensors, optical vision sensors, and fluid flow sensors.

[0055] In one exemplary embodiment, such as Figure 2 As shown, this application provides a smart management method for the entire project lifecycle, which can be applied to... Figure 1 Taking terminal 100 as an example, the method includes:

[0056] S101: Obtain multi-source heterogeneous data throughout the entire project lifecycle to obtain the original multi-source data of the project.

[0057] Here, "project" can refer to a complex set of tasks with clear objectives and multi-stage execution processes, such as engineering construction, IT system development, large equipment R&D, or public services. "Full lifecycle" refers to the complete time span from project initiation and planning, through design, implementation, construction and development, operation and maintenance, to termination and scrapping, encompassing management activities and data flow at each stage. "Multi-source heterogeneous data" refers to a collection of project-related data from different data sources, such as sensors, business systems, and manual input, including structured and unstructured formats, which can be collected through IoT sensor networks, smart terminals, third-party interfaces, etc. For example, terminal 100 collects equipment operating parameters and environmental monitoring data through manual input and sensor 101, and then integrates this data to obtain the project's original multi-source data. The original multi-source data can be data obtained after parsing, classifying, timestamping, and identifying the multi-source heterogeneous data collected throughout the project's lifecycle, while retaining the original data attributes and collection context information. The difference is that the multi-source heterogeneous data is the initial collected data without the aforementioned processing.

[0058] For example, a multi-dimensional data collection system is deployed based on the core management needs of each stage of the project's entire lifecycle. Throughout the entire process from planning and design, implementation and construction, operation and maintenance to termination and decommissioning, structured data such as environmental parameters, equipment operating status, user interaction behavior, and resource consumption are collected through IoT sensors. Simultaneously, the system utilizes interfaces with business management systems to connect to progress tracking systems to obtain task completion rates, cost control systems to retrieve expense amounts, and quality inspection platforms to extract compliance test values. This is supplemented by unstructured data such as on-site video recordings, communication logs, and change documents collected through manual input terminals, forming a multi-source heterogeneous data collection system covering the entire project lifecycle. The dataset employs a multimodal parsing algorithm to identify the format and representation of the collected data. Data with fixed field rules is classified as structured data, while image and text data without fixed formats are classified as unstructured data. Data standardization is implemented, and a unified time benchmark is established to synchronously calibrate all data collection timestamps to ensure cross-source data time consistency. Each data entry is appended with data source identifiers such as collection device information, project stage, and collection scenario. At the same time, the original data attributes and project context information at the time of collection are fully preserved, outputting a raw multi-source dataset containing multiple dimensions, multiple forms, and spatiotemporal and source tags.

[0059] S102: Preprocess the original multi-source data of the project to generate preprocessed multi-source data.

[0060] The preprocessing process includes unstructured data denoising, multi-source data normalization, semantic-level integration, and feature extraction and selection. After preprocessing, noise interference in the multi-source data is effectively filtered out, data formats are standardized and unified, and key features are accurately extracted. For example, in the preprocessing stage of the project's original multi-source data, unstructured data needs to be denoised to remove invalid information, and the denoised multi-source data is normalized to achieve dimensional uniformity and format standardization. Based on preset semantic association mapping rules, the standardized data is integrated into an integrated dataset according to semantic logic. Feature extraction and selection algorithms are used to select feature indicators strongly correlated with the project status, generating preprocessed multi-source data that meets the requirements of subsequent processing.

[0061] S103: Perform cross-system fusion processing on the preprocessed multi-source data to obtain comprehensive project status data.

[0062] Cross-system fusion processing is a core technological step that integrates multi-source heterogeneous data, breaks down information barriers between systems, and generates unified comprehensive project status data to support intelligent decision-making throughout the entire lifecycle. For example, pre-processed multi-source data is input into a multimodal fusion model. An attention mechanism is used to analyze the semantic relationships and feature dependencies between heterogeneous data, aggregating diverse information such as text, images, and numerical values ​​to generate initial fusion features. Based on a domain knowledge graph, data dimensional gaps are identified, and external data ecosystems such as policies, regulations, and market dynamics are intelligently connected to supplement decision-making information. For example, deep neural networks are used to abstract and reconstruct the elements of the fused data, uncovering complex nonlinear relationships between multi-source data. Core project elements such as schedule, cost, quality, and risk are deeply integrated to generate comprehensive project status data with spatiotemporal continuity, element completeness, and decision-support value.

[0063] S104: Based on comprehensive project status data, management strategies for each stage are obtained through computational processing using a reinforcement learning decision framework.

[0064] The reinforcement learning decision framework is a technical system that enables intelligent agents to continuously interact and learn with the project state environment, dynamically optimizing decision-making strategies based on reward feedback, and realizing the intelligent generation and iterative upgrading of management strategies at each stage of the project lifecycle. For example, based on comprehensive project state data, computational processing is performed through the reinforcement learning decision framework. This framework can use a time-series prediction model to analyze the temporal change trends of project states at each stage to obtain predictive information. It uses a risk assessment model to perform multi-dimensional risk identification and evaluation of the comprehensive project state data and predictive information to generate risk-related results. Through a strategy optimization algorithm, it integrates the comprehensive project state data, predictive information, and risk-related results, dynamically adapting and adjusting the algorithm parameters and objective function based on the characteristics of each stage of project management scenarios and historical project management experience. Through continuous iterative learning, it optimizes the strategy generation logic, outputting management strategies for each stage that accurately match the management needs of each stage of the project lifecycle and possess executability and scenario adaptability. Furthermore, it can flexibly adjust the application methods and parameter configurations of each module within the framework according to differences in project types to cover more project management scenarios.

[0065] The technical solution provided in this application includes the following technical effects: The intelligent management method for the entire project lifecycle provided in this application obtains the original multi-source data of the project by acquiring multi-source heterogeneous data throughout the entire project lifecycle. This breaks the limitation of traditional management that relies solely on manual input of structured data such as schedule and cost, and includes unstructured data such as images and text in the collection scope, achieving comprehensive coverage of project lifecycle data; preprocessing the original multi-source data of the project can effectively remove unstructured data noise, complete data standardization, and extract key information through semantic association mapping and feature screening, significantly improving data quality and laying a solid foundation for subsequent processing; the preprocessed multi-source data... By integrating and processing data across systems to generate comprehensive project status data, we can break down data barriers between systems, solve the problems of data fragmentation and independent storage of data at each stage in traditional management, and form a unified and complete understanding of project status. Based on the comprehensive project status data, we can calculate management strategies for each stage through a reinforcement learning decision-making framework, which can realize the linkage and synergy of strategies at each stage, break the fragmented state of traditional phased manual control, provide precise decision support for management, promote the transformation of project management to a technology-driven intelligent model, improve management efficiency and decision-making scientificity, ensure the smooth progress of each stage of the project, and at the same time make up for the shortcomings of traditional management in lacking complete data support and continuous improvement capabilities.

[0066] Based on the above embodiments, management strategies for each stage are obtained by processing comprehensive project status data through a reinforcement learning decision framework, including:

[0067] Step 201: Based on the comprehensive project status data, use the GRU model embedded in the model transfer submodule to predict the temporal change trend of each stage of the project and obtain the prediction results.

[0068] For example, based on comprehensive project status data, terminal 100 uses a GRU model embedded in the model transfer submodule to predict the temporal change trend of each stage of the project to obtain prediction results. The model transfer submodule first performs feature extraction and pattern learning on the time series dataset of similar historical projects, and transfers the learned key temporal feature parameters to the initialization parameters of the current GRU model. The GRU model then processes the continuous time series indicators in the comprehensive project status data based on these transfer parameters, dynamically adjusts the weight of each time series feature through the gating unit, captures the dependency relationship and change law of indicators at different stages over time, and outputs the specific change values ​​and trend information of key indicators at each stage in the future time period.

[0069] Step 202: Use a Bayesian network to conduct a risk assessment on the integrated project status data and prediction results to obtain the risk analysis results.

[0070] For example, when terminal 100 uses a Bayesian network to conduct risk assessment on comprehensive project status data and prediction results, it first performs unified information transformation and probabilistic processing on the two types of data, converting the data into input information that the Bayesian network can recognize. Then, the processed data is input into the Bayesian network, and combined with risk association rules in the project management domain and historical risk event mining results, it constructs the dependencies between network nodes and risk transmission paths. Probabilistic inference algorithms are used to calculate the probability of occurrence and the scope of impact of each risk factor. Next, a multi-dimensional evaluation system and corresponding weights are set based on the overall project objectives to generate risk priority ranking results. Simultaneously, it integrates the spatiotemporal characteristics of risks to generate visualized risk distribution information and matches it with a preset risk response strategy library to generate targeted suggestions. Finally, it integrates risk probability data, priority ranking, distribution information, and response suggestions to form a complete risk analysis result.

[0071] Step 203: Calculate the integrated project status data, prediction results, and risk analysis results using the PPO algorithm after adjusting the objective function to obtain the management strategies for each stage.

[0072] For example, when calculating the comprehensive project status data, prediction results, and risk analysis results using the PPO algorithm with an adjusted objective function, the penalty coefficients for different risk factors are determined based on the risk priority in the risk analysis results. The penalty coefficients for the top 3 risks are set to 1.5, those for priorities 4-10 are set to 1.2, and the rest are set to 1.0 to strengthen the ability to avoid key risks. The original objective function of the PPO algorithm is adjusted by introducing a comprehensive project status data adaptation term, a prediction result deviation correction term, and a risk penalty term to construct an innovative objective function:

[0073] J(θ)=E[R(S,P,Rk)+β・Adp(S,θ)-γ・|P-P_M|-δ・Rk_Pen(Rk_P)]

[0074] Where J(θ) is the adjusted objective function, θ is the PPO algorithm strategy parameter, E[・] is the expectation operator, R(S,P,Rk) is the immediate reward for strategy execution (positive reward for achieving the target and negative reward for cost overrun), β=0.3 is the state fit coefficient, Adp(S,θ) is the fit function between the strategy and the comprehensive project state data (values ​​0-1), γ=0.8 is the prediction bias penalty coefficient, |P-PM| is the absolute value of the deviation between the algorithm and the GRU model prediction results, δ=1.2 is the risk penalty coefficient, and Rk_Pen(Rk_P) is the penalty value based on risk priority (higher priority results in a larger penalty value). The three types of data are input into the adjusted PPO algorithm in batches of 128. The policy update range is limited to [0.8, 1.2] using the clip function to ensure training stability. The advantage function A is used. t =Q t -V tEstimate the direction and step size of the policy improvement, where A t Let Q be the dominant function at time t. t V represents the action value at time t. t Let be the state value function at time t. After 500 rounds of iterative training, it outputs management strategies for each stage, with each strategy accompanied by an execution threshold and adjustment trigger conditions. This solves the problems in traditional management where strategy generation is disconnected from project status, risks, and time-series trends, and lacks execution standards and dynamic adjustment basis. The resulting management strategies are accurate, executable, and adaptable to different scenarios, significantly improving the scientific nature of decision-making and management efficiency at each stage of the project.

[0075] Based on the above embodiments, a Bayesian network is used to conduct a risk assessment on the integrated project status data and prediction results, yielding risk analysis results, including:

[0076] Step 301: Perform data encoding and probability mapping on the comprehensive project status data and prediction results to obtain risk evidence nodes and time-series risk probability data.

[0077] For example, terminal 100 performs data encoding and probability mapping processing on the comprehensive project status data and prediction results. The structured data in the comprehensive project status data is normalized to achieve dimensional uniformity. The unstructured data is converted into feature vectors through a pre-trained language model using semantic embedding. The feature vectors are then mapped to numerical features corresponding to qualitative labels using a project management risk terminology database. The time-series trend data in the prediction results is divided according to the stage dimension of the project's entire life cycle. Combined with risk event records of similar projects in the project's historical risk database, a mapping relationship between the encoded data and risk probabilities is established. Variables with clear risk meanings are extracted as risk evidence nodes, such as schedule delay risk, cost overrun risk, resource shortage risk, and equipment failure risk. Time-series risk probability data arranged according to the stage dimension of the project's entire life cycle is generated.

[0078] Step 302: Input the risk evidence nodes and time-series risk probability data into the Bayesian network, and perform probabilistic inference through the Gibbs sampling algorithm to obtain the probability distribution of each risk factor.

[0079] For example, risk evidence nodes and time-series risk probability data are input into a pre-constructed Bayesian network. The Bayesian network topology is constructed based on risk association logic in the project management field, such as the impact of a certain type of risk on the phase execution process, the impact of phase execution risks on the overall project goals, and the causal relationship mining results of historical risk events. Risk evidence nodes are used as leaf nodes, the overall project risk is used as the root node, and phase execution risks, resource guarantee risks, etc. are added as intermediate nodes, and the dependencies between nodes are clarified. Prior probabilities are assigned to each node based on historical project data, and probabilistic inference is carried out through the Gibbs sampling algorithm. When the probabilities of other nodes are fixed, the posterior probability of a single node is updated according to the conditional probability formula. This process is repeated until the change in node probability reaches a preset small threshold in multiple consecutive iterations to determine convergence. Data is re-inputted according to the phase dimension of the project's entire life cycle to update the inference results, and the probability distribution of each risk factor in each phase of the project's entire life cycle is obtained.

[0080] Step 303: Calculate the probability distribution of each risk factor and the priority data of the risk through the preset risk impact weights to obtain the risk priority assessment matrix.

[0081] For example, referring to the core objectives of the project's entire lifecycle—schedule, cost, quality, and safety—and by pre-setting risk impact weights, the formula is used:

[0082] RP=P×∑(W j ×S ij )

[0083] Calculate the priority data for each risk factor, where RP is the overall priority of a risk factor, P is the probability of occurrence of that risk factor, and W... j S represents the risk impact weights for each core objective dimension. ij The risk factor is assigned an impact score in the j-th core objective dimension. Based on the calculated priority data of each risk factor, the risk priority assessment matrix is ​​constructed according to the correspondence between risk type, assessment dimension, and priority. The matrix rows contain various risk factors, and the columns contain the weight scores and comprehensive priorities of each core objective dimension. The corresponding calculation results are filled in the cells.

[0084] Step 304: Input the comprehensive project status data and prediction results into the risk priority assessment matrix to obtain the risk heat map and a set of response strategy suggestions.

[0085] For example, comprehensive project status data and prediction results are input into a risk priority assessment matrix and matched with the risk types in the matrix. A risk heat map is drawn with each stage of the project's entire life cycle as the horizontal axis and various risk factors as the vertical axis, according to the rule that different priority intervals correspond to different colors to intuitively present the risk distribution. Based on a preset project risk response strategy library, targeted response measures are selected according to the matching relationship between risk type and priority, such as matching and allocating resources and optimizing resource allocation plans for resource shortage risks, and matching and adjusting the stage task arrangement and supplementing execution manpower for schedule delay risks, to form a set of response strategy suggestions.

[0086] Step 305: Integrate the risk heat map and the set of response strategy suggestions to obtain the risk analysis results.

[0087] For example, risk analysis results are obtained by integrating risk heatmaps and a set of response strategy recommendations. The probability of occurrence and the scope of impact of risk factors in high-priority areas of the risk heatmap are marked next to them. Each measure in the response strategy recommendation set is linked to the corresponding risk area in the heatmap by an identifier, and the expected effect after the implementation of each measure is added, such as the expected reduction of the probability of resource shortage risk to a preset safe range after the allocation of resources. The results are then organized into a structured risk analysis report based on project stage, risk type, risk probability, heatmap location, response strategy, and expected effect. This provides risk dimension support for the subsequent computational management strategy of the reinforcement learning decision framework in this method, and provides precise risk dimension support for the subsequent computational management strategy of the reinforcement learning decision framework.

[0088] In one embodiment of the present invention, the original multi-source data of the project is preprocessed to generate preprocessed multi-source data, including:

[0089] Step 401: Denoise the unstructured data in the original multi-source data of the project to obtain the denoised multi-source data.

[0090] For example, noise reduction processing is performed on unstructured data in the original multi-source data of the project. The unstructured data types in the original multi-source data of the project are identified. For text data, stop word removal technology is used to remove meaningless function words and word sense disambiguation algorithm is used to correct ambiguous expressions. For image data, Gaussian filtering algorithm is used to smooth pixel noise and edge preservation technology is used to avoid blurring of key features. For log data, sliding window outlier detection method is used to remove sudden invalid records. After the above adaptation processing, the noise-reduced multi-source data is obtained.

[0091] Step 402: Normalize the denoised multi-source data to obtain standardized multi-source data.

[0092] For example, the denoised multi-source data is normalized, and the numerical distribution characteristics of the denoised multi-source data are statistically analyzed, covering the completion rate of progress-related tasks, the amount of cost-related expenses, the operating parameter readings of equipment monitoring, and the feature vector values ​​of image conversion, etc. A Min-Max normalization method adapted to the multi-dimensional data of the project is used to construct the formula:

[0093] x norm =

[0094] Where, x norm Here are the normalized data values, and x is the original data value after noise reduction. min x is the minimum value in this type of data. max The maximum value in this type of data is used to map various types of data to a unified numerical range according to a formula to eliminate weight bias caused by differences in dimensions, thus obtaining standardized multi-source data.

[0095] Step 403: Through semantic association mapping rules between multi-source data, the standardized multi-source data are semantically integrated to generate an integrated multi-source dataset.

[0096] For example, when standardizing multi-source data is semantically integrated through semantic association mapping rules between multi-source data, a semantic association mapping rule system is constructed based on project management domain knowledge. The rules include the association between equipment data and fault data, the association between cost data and procurement data, and the association between schedule data and task data. Based on these rules, the standardized multi-source data from the equipment monitoring system, cost management system, and schedule tracking system are classified and associated according to semantic correspondence to form semantic links between data, generating an integrated multi-source dataset.

[0097] Step 404: Perform feature extraction processing on the integrated multi-source dataset, and use a preset feature selection algorithm to filter out feature indicators that are strongly correlated with the project status to obtain preprocessed multi-source data.

[0098] For example, when performing feature extraction processing on an integrated multi-source dataset and using a preset feature selection algorithm to screen feature indicators, multimodal feature extraction technology is used to extract features from the integrated multi-source dataset. Principal component analysis is used to extract core components for numerical data, a bag-of-words model is used to extract keyword features for text data, and a convolutional neural network is used to extract visual features for image data. The preset feature selection algorithm is then called to construct a correlation calculation model.

[0099] R=

[0100] Where R is the correlation between the feature and the project status, Cov(F,S) is the covariance between feature F and project status index S, and σF Let σ be the standard deviation of characteristic F. S The standard deviation of the project status indicator S is used to calculate the correlation between each extracted feature and the project status. Feature indicators with a correlation higher than the preset threshold, such as progress completion rate, cost deviation rate, equipment core component temperature, and material inventory turnover rate, are selected to obtain preprocessed multi-source data. This effectively compresses data dimensions and focuses on key information, improving the efficiency and accuracy of subsequent cross-system data fusion. At the same time, it provides core, high-value feature inputs for the reinforcement learning decision framework, avoiding redundant data from interfering with decision calculations.

[0101] In one embodiment of the present invention, preprocessed multi-source data undergoes cross-system fusion processing to obtain comprehensive project status data, including:

[0102] Step 501: Input the preprocessed multi-source data into the Transformer multimodal fusion model, calculate the cross-modal feature association weights through the multi-head attention mechanism, and obtain the preliminary fused feature vector.

[0103] For example, preprocessed multi-source heterogeneous data is directed to the Transformer multimodal fusion model architecture. Relying on the multi-head attention mechanism, in-depth interactive analysis is carried out on the data features from different business systems, covering both structured and unstructured forms. Each attention head independently captures the long-distance dependency relationship and semantic association pattern between features of a specific dimension. The complementary value and redundancy between cross-modal features are quantified through a dynamic weight allocation strategy. The core information of multi-dimensional features such as text semantics, image vision, and numerical statistics are aggregated to output a preliminary fusion feature vector that integrates the key meaning and association logic of multi-source data.

[0104] Step 502: Connect the preliminary fused feature vectors to cross-system data sources to obtain an external data source set; the external data source set is used to make up for the shortcomings of internal data and cover heterogeneous data to improve the external related data set for decision-making data support.

[0105] For example, the generated preliminary fusion feature vector is pushed to an external data source intelligent docking platform. Based on the data gap identification rules and external data matching strategies preset in the project's life cycle knowledge graph, matching data resources are retrieved in a targeted manner in heterogeneous data ecosystems such as open data cloud platforms, industry-level databases, and IoT sensing terminal networks. High-quality external data that can fill the gaps in internal data dimensions and cover the shortcomings of unstructured data are selected through semantic similarity calculation and pattern matching algorithms. An external related data set containing dimensions such as policy and regulatory dynamics, market supply and demand fluctuations, and real-time status of the supply chain is constructed.

[0106] Step 503: Align the initial fused feature vectors with the feature dimensions of the external data source set to obtain the dataset to be fused.

[0107] For example, the process promotes the initial fusion of feature vectors and external data source sets into a feature dimension collaborative alignment process. By constructing a learnable linear transformation layer and a dynamic embedding mapping mechanism, the feature dimension gap caused by differences in acquisition standards, storage formats, and semantic systems between different data sources is eliminated. The cross-modal feature alignment accuracy is optimized by using contrastive learning algorithms. Semantic-level alignment and tensor-level regularization of multi-source data features are achieved in a unified high-dimensional feature space. The output is a dimensionally consistent and semantically interoperable dataset to be fused, providing a standardized data carrier for deep feature interaction and knowledge distillation, and completely breaking down the semantic barriers of cross-system data fusion.

[0108] Step 504: Apply a fully connected layer and activation function to the dataset to be fused to perform a non-linear mapping to generate comprehensive project status data.

[0109] For example, the dimension-aligned dataset to be fused is input into a deep mapping module consisting of a multi-layer fully connected neural network and a nonlinear activation function. The module leverages the parameterized feature interaction capabilities of the fully connected layers to mine complex nonlinear relationships and potential patterns among multiple data sources. It utilizes the gradient propagation characteristics of activation functions such as ReLU to enhance the nonlinear tension of feature representation. Through the forward propagation and backward correction process of the multi-layer network, it deeply integrates and abstracts project elements such as schedule, cost, quality, and risk, generating comprehensive project status data that covers the status of all project elements, possesses interpretability and decision support value, and provides complete and high-quality decision data support for subsequent calculation of management strategies at each stage based on a reinforcement learning decision framework.

[0110] In one embodiment of the present invention, after processing the comprehensive project status data through a reinforcement learning decision framework to obtain the management strategies for each stage, the method further includes:

[0111] Step 601: Obtain the actual data generated during the execution of each stage of the project, compare and analyze the actual data with the execution results corresponding to the management strategies predicted based on historical data, and obtain the prediction deviation dataset.

[0112] For example, during the implementation of each stage of the project's entire lifecycle, actual operational data generated at the execution end is collected in real time. This includes structured data such as task completion volume, resource consumption, risk occurrence frequency, and quality inspection indicators, as well as unstructured data such as on-site video recordings, communication logs, and change documents. These actual data are compared and analyzed with the management strategy execution results output by the reinforcement learning prediction model trained on historical project data, aligned by timestamps and business dimensions. Quantitative difference indicators such as schedule deviation, cost overrun, and risk response delay duration are identified and integrated to form a prediction deviation dataset that includes the type of difference, the stage of occurrence, and the degree of impact.

[0113] Step 602: Using preset strategy effect evaluation indicators, perform difference analysis on the prediction deviation dataset to obtain information on strategy modules that did not achieve the expected goals and their corresponding influencing factors;

[0114] For example, a pre-defined strategy effectiveness evaluation index system, including deviations in schedule achievement rate, cost saving rate, risk avoidance timeliness index, and quality compliance fluctuation range, is invoked to conduct multi-dimensional difference quantitative analysis on the predicted deviation dataset, using a dynamic weighted correlation analysis model:

[0115] R module =α˙Var(A module -P module )+β˙Corr(A module StagePerf)

[0116] Among them, R module The deviation score represents the management strategy module, where α is the weighting coefficient of the deviation fluctuation term within the module, Var(A) module -P module ) is the variance between the actual and predicted values ​​of the module, β is the weighting coefficient of the module's correlation with stage performance, and Corr(A) is the deviation of the actual and predicted values ​​of the module. module StagePerf is the Pearson correlation coefficient between module deviation and overall stage performance. It yields management strategies for failing to meet expected goals and extracts key influencing factors of corresponding module deviations, such as resource scheduling lag coefficient, risk misjudgment probability value, and task dependency imbalance.

[0117] The management strategy modules that fail to meet expected goals, such as resource allocation strategy module, risk warning strategy module, and schedule control strategy module, are identified, and key influencing factors of deviations in the corresponding modules, such as resource scheduling lag coefficient, risk misjudgment probability value, and task dependency imbalance, are extracted simultaneously.

[0118] Step 603: Adjust the weights of the objective function of the PPO algorithm in the reinforcement learning decision framework according to the impact factors to obtain the adjusted objective function;

[0119] For example, based on the type and degree of the extracted influencing factors, the objective function weight coefficients of the proximal policy optimization algorithm in the reinforcement learning decision-making framework are adjusted accordingly. If the weight of resource-type influencing factors exceeds a preset threshold, an adaptive weight adjustment formula is used:

[0120] ω i ′=ω i ˙(1+γ˙ )

[0121] Where ωi′ is the adjusted target weight coefficient, ω i是 The original weighting coefficients, where γ is the amplification factor of the influence factor on the weights. It is the quantitative value of resource-related impact factors. It is the maximum value of all influencing factors. This formula is used to increase the coefficient of the reward item in the resource dimension. If the risk-related influencing factors have a higher weight, the coefficient of the risk penalty item is strengthened. A new multi-objective dynamic weighted objective function is constructed that integrates the degree of return achievement, the degree of risk loss, the degree of time deviation, and the degree of compliance matching, so as to ensure that key influencing factors are given priority in the allocation of decision weights.

[0122] Step 604: Based on the adjusted objective function, correct the risk assessment parameters of the Bayesian network to obtain the adjusted parameter data;

[0123] For example, based on the adjusted objective function orientation and weight allocation logic, the risk assessment parameter system of the Bayesian network is reverse-corrected. For core parameters such as the prior distribution of risk occurrence probability, the conditional probability of risk impact degree, and the weight of risk transmission path, the parameters are calibrated by combining the actual risk event data and deviation analysis results of the current project stage with the parameters of the Bayesian parameter calibration model. The domain adaptive regularization term is introduced to balance historical experience and current project specificity, resulting in a new set of risk assessment parameters that can accurately match the risk characteristics and historical experience of the current project stage.

[0124] Step 605: Input the adjusted parameter data into the reinforcement learning decision framework to obtain the iteratively optimized management strategy.

[0125] For example, the corrected Bayesian risk assessment parameter data and the adjusted PPO algorithm objective function parameters are synchronously input into the reinforcement learning decision framework, driving the agent within the framework to explore the optimal strategy in the project state space with the new parameter rules. Through multiple rounds of state-action interaction learning and policy gradient update iteration, the policy network parameters are optimized, and a new generation of management strategy set that adapts to the actual operating status of the current project stage and integrates the experience of this round of deviation analysis and risk parameter calibration results is output, realizing the continuous evolution and precise optimization of management strategies.

[0126] In one embodiment of the present invention, multi-source heterogeneous data throughout the entire lifecycle of a project is acquired to obtain the original multi-source data of the project, including:

[0127] Step 701: Acquire multi-source heterogeneous data throughout the project's entire lifecycle through a sensor network.

[0128] For example, when acquiring multi-source heterogeneous data throughout the entire lifecycle of a project through a sensor network, the sensor network is first deployed according to the core management needs of each stage of the project's lifecycle. Sensor types adapted to the monitoring targets of each stage are configured for different stages such as planning and design, operation and service, and maintenance and optimization. Each sensor collects data of the corresponding dimension at a preset sampling frequency, covering multiple types of information such as environmental parameters, equipment operating status, user interaction behavior, resource consumption, and safety monitoring, forming a multi-source heterogeneous data set covering the entire lifecycle of the project.

[0129] Step 702: Analyze the multi-source heterogeneous data and divide it into structured data and unstructured data.

[0130] For example, a multimodal data parsing algorithm is used to identify the format and representation of the collected data. Data with a fixed data format that can be directly represented by numerical values ​​or fixed fields is classified as structured data, while data without a fixed format that cannot be directly represented by numerical values ​​is classified as unstructured data. The boundaries and classification of the two types of data are clarified through a unified classification logic.

[0131] Step 703: Perform timestamp synchronization and data source identification processing on the collected structured and unstructured data, retain the original data attributes and collection context information, and obtain the original multi-source data of the project.

[0132] For example, when performing timestamp synchronization and data source identification processing on the collected structured and unstructured data, a unified time benchmark is established to calibrate the collection timestamps of all data to ensure the consistency of data time obtained by different collection entities. A data source identifier containing collection device information, project stage, collection scenario, etc. is added to each data, and the original data attributes and project context information at the time of collection are preserved to obtain complete original multi-source project data.

[0133] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0134] In one exemplary embodiment, such as Figure 3As shown, a project lifecycle intelligent management device 800 is provided, comprising:

[0135] The data acquisition module 801 is used to acquire multi-source heterogeneous data throughout the entire life cycle of the project to obtain the original multi-source data of the project.

[0136] The preprocessing module 802 is used to preprocess the original multi-source data of the project to generate preprocessed multi-source data.

[0137] The data fusion module 803 is used to perform cross-system fusion processing on preprocessed multi-source data to obtain comprehensive project status data.

[0138] The learning decision module 804 is used to calculate and process comprehensive project status data through a reinforcement learning decision framework to obtain management strategies for each stage.

[0139] In one embodiment of the present invention, the learning decision module 804 is further configured to:

[0140] Based on comprehensive project status data, a GRU model with an embedded model transfer submodule is used to predict the temporal change trend of each stage of the project, and the prediction results are obtained.

[0141] A Bayesian network is used to conduct a risk assessment based on the comprehensive project status data and prediction results, and the risk analysis results are obtained.

[0142] By adjusting the objective function, the PPO algorithm calculates the comprehensive project status data, prediction results, and risk analysis results to obtain management strategies for each stage.

[0143] In one embodiment of the present invention, the learning decision module 804 is further configured to:

[0144] Data encoding and probability mapping are performed on the comprehensive project status data and prediction results to obtain risk evidence nodes and time-series risk probability data.

[0145] Risk evidence nodes and time-series risk probability data are input into a Bayesian network, and probabilistic inference is performed using the Gibbs sampling algorithm to obtain the probability distribution of each risk factor.

[0146] By calculating the probability distribution of each risk factor and the priority data of the risk through preset risk impact weights, a risk priority assessment matrix is ​​obtained.

[0147] By inputting comprehensive project status data and forecast results into the risk priority assessment matrix, a risk heatmap and a set of response strategy recommendations are obtained.

[0148] By integrating the risk heatmap and the set of response strategy recommendations, the risk analysis results are obtained.

[0149] In one embodiment of the present invention, the preprocessing module 802 is further configured to:

[0150] The unstructured data in the original multi-source data of the project is denoised to obtain the denoised multi-source data.

[0151] The denoised multi-source data is normalized to obtain standardized multi-source data.

[0152] By using semantic association mapping rules between multi-source data, standardized multi-source data are semantically integrated to generate an integrated multi-source dataset.

[0153] Feature extraction is performed on the integrated multi-source dataset, and a preset feature selection algorithm is used to select feature indicators that are strongly correlated with the project status, thus obtaining preprocessed multi-source data.

[0154] In one embodiment of the present invention, the data fusion module 803 is further configured to:

[0155] The preprocessed multi-source data is input into the Transformer multimodal fusion model, and the cross-modal feature association weights are calculated through the multi-head attention mechanism to obtain the preliminary fused feature vector.

[0156] The feature vectors are initially integrated and connected to cross-system data sources to obtain a set of external data sources. This set of external data sources is used to compensate for internal data deficiencies and cover heterogeneous data to improve the external related data set for decision-making.

[0157] The initial fused feature vectors are aligned with the feature dimensions of the external data source set to obtain the dataset to be fused.

[0158] The dataset to be fused is non-linearly mapped using a fully connected layer and an activation function to generate comprehensive project status data.

[0159] In one embodiment of the present invention, the apparatus further includes:

[0160] The predictive data preprocessing module is used to acquire the actual data generated during the execution of each stage of the project, compare and analyze the actual data with the execution results corresponding to the management strategies predicted based on historical data, and obtain the prediction deviation dataset.

[0161] The strategy effectiveness analysis module is used to perform difference analysis on the prediction deviation dataset using preset strategy effectiveness evaluation indicators, and to obtain information on strategy modules that have not achieved the expected goals and the corresponding influencing factors.

[0162] The weight adjustment module is used to adjust the weights of the objective function of the PPO algorithm in the reinforcement learning decision framework according to the influence factors, so as to obtain the adjusted objective function.

[0163] The Bayesian parameter correction module is used to correct the risk assessment parameters of the Bayesian network based on the adjusted objective function, thereby obtaining the adjusted parameter data.

[0164] The strategy iteration and optimization module is used to input the adjusted parameter data into the reinforcement learning decision framework to obtain the iteratively optimized management strategy.

[0165] In one embodiment, this application provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0166] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0167] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0168] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A smart management method for the entire lifecycle of a project, characterized in that, The method includes: Acquire multi-source heterogeneous data throughout the entire project lifecycle to obtain the project's original multi-source data; The original multi-source data of the project is preprocessed to generate preprocessed multi-source data; The preprocessed multi-source data is then subjected to cross-system fusion processing to obtain comprehensive project status data; Based on the comprehensive project status data, management strategies for each stage are obtained through calculation and processing using a reinforcement learning decision framework.

2. The method according to claim 1, characterized in that, The management strategies for each stage are obtained by processing the comprehensive project status data through a reinforcement learning decision framework, including: Based on the comprehensive project status data, the GRU model with embedded model transfer submodule is used to predict the time-series change trend of each stage of the project, and the prediction results are obtained. A risk assessment is performed on the comprehensive project status data and the prediction results using a Bayesian network to obtain risk analysis results. The PPO algorithm, after adjusting the objective function, calculates the comprehensive project status data, the prediction results, and the risk analysis results to obtain the management strategies for each stage.

3. The method according to claim 2, characterized in that, The risk assessment is performed using a Bayesian network on the integrated project status data and the prediction results to obtain risk analysis results, including: The comprehensive project status data and the prediction results are processed by data encoding and probability mapping to obtain risk evidence nodes and time-series risk probability data. The risk evidence nodes and time-series risk probability data are input into a Bayesian network, and probabilistic inference is performed using the Gibbs sampling algorithm to obtain the probability distribution of each risk factor. The probability distribution risk priority data of each risk factor is calculated by using preset risk impact weights, and a risk priority assessment matrix is ​​obtained. Input the comprehensive project status data and the prediction results into the risk priority assessment matrix to obtain a risk heatmap and a set of response strategy suggestions; The risk analysis results are obtained by integrating the risk heatmap and the set of response strategy suggestions.

4. The method according to claim 1, characterized in that, The preprocessing of the original multi-source data of the project to generate preprocessed multi-source data includes: The unstructured data in the original multi-source data of the project is denoised to obtain denoised multi-source data; The denoised multi-source data is normalized to obtain standardized multi-source data; By using semantic association mapping rules between multi-source data, the standardized multi-source data is semantically integrated to generate an integrated multi-source dataset; Feature extraction processing is performed on the integrated multi-source dataset, and a preset feature selection algorithm is used to filter out feature indicators that are strongly correlated with the project status, thus obtaining the preprocessed multi-source data.

5. The method according to claim 1, characterized in that, The process of performing cross-system fusion processing on the preprocessed multi-source data to obtain comprehensive project status data includes: The preprocessed multi-source data is input into the Transformer multimodal fusion model, and the cross-modal feature association weights are calculated through the multi-head attention mechanism to obtain the preliminary fused feature vector. The preliminary fused feature vectors are then connected to cross-system data sources to obtain an external data source set; wherein, the external data source set is used to compensate for internal data shortcomings and cover heterogeneous data to improve the external related data set for decision-making data support; The preliminary fused feature vectors are aligned with the external data source set in terms of feature dimensions to obtain the dataset to be fused. The dataset to be fused is non-linearly mapped using a fully connected layer and an activation function to generate comprehensive project status data.

6. The method according to claim 1, characterized in that, After processing the comprehensive project status data using a reinforcement learning decision framework to obtain management strategies for each stage, the process further includes: Obtain the actual data generated during the execution of each stage of the project, compare and analyze the actual data with the execution results corresponding to the management strategies predicted based on historical data, and obtain the prediction deviation dataset. Using preset strategy effectiveness evaluation indicators, a difference analysis is performed on the prediction deviation dataset to obtain information on strategy modules that did not achieve the expected goals and their corresponding influencing factors. The weights of the objective function of the PPO algorithm in the reinforcement learning decision framework are adjusted according to the influencing factors to obtain the adjusted objective function. Based on the adjusted objective function, the risk assessment parameters of the Bayesian network are corrected to obtain the adjusted parameter data. The adjusted parameter data is input into the reinforcement learning decision framework to obtain the iteratively optimized management strategy.

7. The method according to claim 1, characterized in that, The acquisition of multi-source heterogeneous data throughout the entire project lifecycle, resulting in the original multi-source project data, includes: The multi-source heterogeneous data is acquired throughout the entire lifecycle of the project through a sensor network. The multi-source heterogeneous data is analyzed and divided into structured data and unstructured data. The collected structured and unstructured data are time-stamped and identified by data source, preserving the original data attributes and collection context information to obtain the original multi-source data of the project.

8. A smart management device for the entire life cycle of a project, characterized in that, The device includes: The data acquisition module is used to acquire multi-source heterogeneous data throughout the entire project lifecycle to obtain the original multi-source data of the project. The preprocessing module is used to preprocess the original multi-source data of the project to generate preprocessed multi-source data; The data fusion module is used to perform cross-system fusion processing on the preprocessed multi-source data to obtain comprehensive project status data; The learning decision module is used to calculate and process the comprehensive project status data through a reinforcement learning decision framework to obtain management strategies for each stage.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images