Engineering investment project multi-dimensional risk dynamic assessment and early warning system

Abstract

The invention relates to the technical field of computers, particularly discloses an engineering investment project multi-dimensional risk dynamic assessment and early warning system, and aims to solve the problems that existing risk assessment is single in dimension, insufficient in timeliness and lack of dynamic early warning and intelligent decision support. The system comprises a data acquisition and preprocessing module, a multi-dimensional risk feature construction module, a dynamic risk assessment and prediction module, a risk early warning and visualization module and an intelligent decision support and optimization module. Through integration of multivariate data, machine learning and deep learning algorithms, risk dynamic modeling, real-time evaluation and trend prediction are realized, and in combination with intelligent early warning and decision support, risk management is changed from post-remedy to beforehand prevention.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of computer technology, and particularly relates to a multi-dimensional risk dynamic assessment and early warning system for engineering investment projects. BACKGROUND

[0002] In the field of engineering investment project management, dynamic assessment and early warning of multi-dimensional risks of projects are key links to ensure project success and investment return. Traditional project risk management methods usually rely on manual analysis, expert experience and report-type processing of static data, which limits the timeliness and accuracy of risk identification, assessment and early warning. Although existing technologies can obtain basic data such as engineering quantity lists of engineering projects and perform systematic analysis and calculation based on preset analysis models in terms of policies and regulations, economic management, professional technology, finance and other aspects, the output analysis reports are usually stage-based and cannot adapt to the dynamic changes of risk factors in the project implementation process, lacking real-time and comprehensive risk correlation analysis capabilities, so that decision-makers cannot obtain accurate early warning information in a timely manner.

[0003] Among them, the inherent complexity of engineering investment projects and the interweaving of risk factors (such as policy changes, market fluctuations, technical difficulties, and fund chain risks) involved further exacerbate the difficulty of risk assessment. Although existing methods can provide various analysis reports, these reports are usually isolated and static, and it is difficult to achieve dynamic correlation and real-time evolution analysis between multi-dimensional risk factors. For example, changes in a certain policy and regulation may have a linked impact on multiple financial or economic indicators, and traditional report preparation has a long cycle and low update frequency, which cannot capture such chain reactions and minor fluctuations, resulting in delayed risk warning or misjudgment. In addition, existing systems generally lack the ability to continuously monitor and intelligently predict potential risk events, and are limited to post-event analysis, unable to actively identify risk germination and timely issue warnings, thereby causing irreparable losses to the project.

[0004] Therefore, there is an urgent need for a system that can dynamically assess multi-dimensional risks of engineering investment projects and provide timely and accurate early warnings. SUMMARY

[0005] The present application provides a multi-dimensional risk dynamic assessment and early warning system for engineering investment projects, aiming to solve the limitations of existing engineering investment project risk assessment methods, such as single assessment dimension, insufficient timeliness, inability to dynamically track risk evolution, and lack of proactive early warning and intelligent decision support. Existing methods mainly rely on static data before project implementation and manual analysis, making it difficult to cope with complex and changing internal and external risk factors throughout the project life cycle, and unable to achieve early identification, quantitative assessment and real-time warning of potential risks.

[0006] To solve the above technical problems, the present application provides an engineering investment project multi-dimensional risk dynamic assessment and early warning system. The system integrates multiple heterogeneous data sources, builds a cross-dimensional risk index system, uses advanced machine learning and deep learning algorithms for dynamic risk modeling, real-time assessment and trend prediction, and combines intelligent early warning mechanisms and decision support modules to provide continuous, comprehensive and accurate risk management services for the entire life cycle of engineering investment projects.

[0007] According to one aspect of the present application, an engineering investment project multi-dimensional risk dynamic assessment and early warning system is provided, which comprises: a data acquisition and preprocessing module, a multi-dimensional risk feature construction module, a dynamic risk assessment and prediction module, a risk early warning and visualization module, and an intelligent decision support and optimization module. The modules work together to achieve comprehensive, dynamic and intelligent management of engineering investment project risks.

[0008] As an embodiment of the present application, the data acquisition and preprocessing module is used to obtain engineering investment project related data from multiple sources, and to clean, standardize, integrate and process the data in a unified format.

[0009] Further, the data sources include but are not limited to: project management system data, obtaining project schedule plan, actual progress, resource consumption, cost budget, actual cost, contract management information and change records; financial management system data, obtaining project inflow and outflow, financing status, cash flow, return on investment and financial statement data; supply chain management system data, obtaining supplier performance, material procurement information, logistics status and inventory level data; on-site Internet of Things sensor data, obtaining real-time data of environmental parameters, equipment operating status, personnel activities and material consumption at the construction site; external macroeconomic data, obtaining gross domestic product, inflation rate, interest rate, exchange rate, industry sentiment index and related policy and regulation update information; market intelligence data, obtaining target market supply and demand changes, competitor dynamics, technology development trends and consumer behavior data; environmental and social data, obtaining climate change, natural disaster risk, community relations, public opinion and environmental compliance information in the project location.

[0010] As an embodiment of the present application, the data collection and preprocessing module performs uniform timestamp alignment, missing value filling, outlier detection and correction, and data normalization on the collected heterogeneous data. Missing value filling is completed by interpolation algorithm or prediction model based on machine learning. Outlier detection is achieved by statistical method or based on isolation forest algorithm. Data normalization converts data of different dimensions into a unified numerical range to eliminate the influence of dimension difference on subsequent analysis. The data preprocessing and integration module stores the processed data to the distributed data warehouse for efficient access and processing by subsequent modules.

[0011] As an embodiment of the present application, the multi-dimensional risk feature construction module is used to extract and construct multi-dimensional risk features from the preprocessed data.

[0012] Further, the multi-dimensional risk feature construction module constructs risk features by the following ways: Time series feature extraction: For time series data such as project progress, cost, resource consumption, etc., the trend, periodicity, volatility and autocorrelation features are extracted. The mean, standard deviation, maximum, minimum, rate of change and seasonal index are calculated by sliding time window as features.

[0013] Text information feature extraction: For unstructured text data such as project reports, contract documents, policy files, news reports and public comments, natural language processing techniques are used to extract keywords, topic model features, sentiment orientation features and entity relationship features. Word embedding model is used to convert text into high-dimensional vector representation.

[0014] Graph structure feature construction: The entities in the project, including project tasks, resources, personnel, risk events, policies and regulations, and stakeholders, are constructed as nodes, and the relationships between entities, including dependency, influence, membership, causality and interaction, are constructed as edges, forming a project knowledge graph. Node features, edge features and graph topology features are extracted from the knowledge graph.

[0015] Expert knowledge and rule features: Combine the experience of domain experts to convert known risk indicators, risk trigger conditions and risk level evaluation rules into structured features or logical judgment features.

[0016] Further, the multi-dimensional risk feature construction module fuses the above-mentioned various features to form a unified multi-dimensional feature vector space, which fully represents the risk state of the engineering investment project in different dimensions. The feature fusion adopts weighted summation, feature splicing or deep learning-based feature fusion network to preserve the semantic information of different features and enhance their expression ability.

[0017] As an embodiment of the present application, the dynamic risk assessment and prediction module is used to conduct real-time risk assessment and future risk trend prediction based on the constructed multi-dimensional risk features.

[0018] Further, the dynamic risk assessment and prediction module comprises: A risk factor identification and quantification unit is used to identify and quantify specific risk factors in each dimension. Through principal component analysis, factor analysis or independent component analysis, the latent and independent risk factors such as technical risk, market risk, financial risk, management risk and policy and regulation risk are identified and reduced from high-dimensional features. Each risk factor is quantified by a support vector machine model or a random forest model, and the risk exposure or probability value is output.

[0019] A risk correlation analysis unit is used to analyze the interaction relationship and transmission mechanism between different risk factors. By constructing a Bayesian network or a graph neural network model, the identified risk factors are taken as nodes, and the causal relationship, conditional dependence relationship or correlation relationship between them are taken as edges, and the risk transmission path and influence strength are modeled. This model can reveal the "butterfly effect" type of risk amplification mechanism, i.e. how the risk change in one dimension triggers the risk in other dimensions.

[0020] A dynamic risk state assessment unit is used to calculate the current comprehensive risk index or risk vector of the project based on the real-time updated multi-dimensional risk features and risk correlation model. This unit adopts a recurrent neural network model such as long short-term memory network or gated recurrent unit network, takes the historical risk state, current risk features and risk correlation analysis results as input, and outputs the real-time risk score of the project in each dimension and the overall risk level.

[0021] A risk trend prediction unit is used to predict the risk development trend of the project in the future. This unit is based on a time series prediction model such as ARIMA model, Prophet model or Transformer model, and predicts the future time step of the project's risk indicators and comprehensive risk index. The prediction results include point estimates and confidence intervals of future risk values, providing a time window for taking preventive measures in advance.

[0022] As an embodiment of the present application, the dynamic risk assessment and prediction module is self-adaptively optimized through reinforcement learning mechanism. The evaluation model and prediction model of the system continuously learn and adjust parameters according to the occurrence of actual risk events, the effect of risk mitigation measures and the final risk performance of the project, in order to improve the accuracy of evaluation and prediction. The reinforcement learning agent learns the optimal risk assessment strategy and prediction model weight in the simulated project environment.

[0023] As an embodiment of the present application, the risk early warning and visualization module is used to generate and issue early warning information according to the dynamic risk assessment results, and to display the risk state of the project in an intuitive and friendly manner.

[0024] Further, the risk early warning and visualization module comprises: An early warning rule configuration unit is configured to define early warning thresholds and triggering conditions for different risk levels. The early warning thresholds can be flexibly configured according to the type, stage, industry characteristics and risk preference of the project. For example, set the integrated risk index exceeding 0.6 as yellow early warning, exceeding 0.8 as orange early warning, and exceeding 0.95 as red early warning. The triggering conditions can include single risk factor over-limit, multiple risk factors simultaneously abnormal or risk trend continuously deteriorating.

[0025] An early warning information generation and issue unit is configured to automatically generate early warning information when the early warning rules are met, and notify the relevant stakeholders through various channels. The early warning information includes risk type, risk level, triggering time, impact range, potential consequences and suggested attention matters. The release channels include system built-in message notification, email, mobile phone message and instant messaging software.

[0026] A risk situation visualization unit is configured to visually display the multi-dimensional risk situation of the project through interactive dashboards, risk maps, Gantt charts and trend charts. The dashboard displays the current integrated risk index, risk score of each dimension, risk distribution and risk hotspot area. The risk map is based on geographic information system to display the risk level of different areas or sub-projects. The Gantt chart combines risk events with project progress to show the impact of risk on the construction period. The trend chart shows the historical risk evolution track and the future risk prediction trend.

[0027] Further, the risk early warning and visualization module supports user-defined early warning views and report generation, facilitating users to perform personalized monitoring and reporting according to their own concerns.

[0028] As an embodiment of the present application, the intelligent decision support and optimization module is used to provide targeted risk mitigation suggestions for users after identifying risks and issuing early warnings, and to assist in optimizing project management decisions.

[0029] Further, the intelligent decision support and optimization module comprises: A risk mitigation scheme library is configured to store historical successful risk mitigation strategies, best practice cases and expert experience knowledge. Each scheme in the scheme library is associated with applicable risk types, implementation conditions, expected effects and required resources.

[0030] The intelligent solution matching and recommendation unit is used to intelligently match and recommend the optimal mitigation solution from the risk mitigation solution library based on the specific risk scenario, risk type, risk level, and project resource constraints of the current project. This unit uses a case-based reasoning algorithm or a collaborative filtering algorithm, combining the similarity between the current risk characteristics and historical cases, as well as the effectiveness evaluation of the solutions, to provide the top-ranked recommended solutions.

[0031] The decision-making effectiveness simulation and evaluation unit is used to simulate recommended risk mitigation solutions and evaluate their potential effects and impact on project objectives. Through Monte Carlo simulations or discrete event simulation models, it simulates the implementation effects of different mitigation solutions in the project environment and predicts their degree of improvement on project schedule, cost, quality, and other risk indicators. This unit also evaluates potential new or secondary risks that may arise from the implementation of the solutions.

[0032] The optimization suggestion generation unit is used to generate optimal decision optimization suggestions based on a comprehensive consideration of risk mitigation effectiveness, implementation costs, resource availability, and overall project objectives. For example, when a project schedule delay risk warning is issued, suggestions may be made to adjust resource allocation, optimize workflows, or introduce new technologies to accelerate progress. When a cost overrun risk warning is issued, suggestions may be made to renegotiate supplier contracts, find alternative materials, or optimize procurement strategies.

[0033] Furthermore, the intelligent decision support and optimization module allows users to revise and provide feedback on the recommended solutions, and incorporates the feedback results into the system's learning cycle to continuously optimize the recommendation algorithm and solution library.

[0034] Compared with the prior art, the advantages and positive effects of the present invention are as follows: This system transforms risk assessment for engineering investment projects from a static, discrete approach to a dynamic, continuous one. Traditional methods only conduct a one-time assessment before project commencement, while this system, through real-time data acquisition and processing, continuously monitors and assesses risks, ensuring the timeliness and accuracy of risk information.

[0035] This system breaks through the limitations of a single dimension in risk assessment. By integrating multi-source heterogeneous data and constructing multi-dimensional risk characteristics, this system can comprehensively consider risks at all levels, including policies and regulations, economic management, professional technology, finance, market environment, supply chain, and social environment, forming a more comprehensive and three-dimensional risk view and avoiding one-sided assessments that are like "seeing only a part of the whole."

[0036] This system possesses intelligent risk factor identification and correlation analysis capabilities. By employing advanced models such as Bayesian networks or graph neural networks, this system can not only identify individual risk factors but also reveal the complex correlations and transmission mechanisms between different risk factors, effectively capturing potential risk amplification effects or chain reactions, thus improving the depth and breadth of risk identification.

[0037] The system provides accurate risk trend prediction function. With the help of deep learning technologies such as recurrent neural network and Transformer model, the system can predict the risk development trend and potential risk points of the project in the future period of time according to the historical data and real-time dynamics, and provide valuable early warning time for the project management party, and change the risk management from post-repair to pre-prevention.

[0038] The system has intelligent early warning mechanism and decision support ability. The system can automatically trigger multi-level early warning according to the preset rules and dynamic evaluation results, and notify the relevant parties through various channels. More importantly, the system can combine the actual situation of the project, intelligently match and recommend the optimal solution from the risk mitigation scheme library, and assist decision-making through effect simulation, which significantly improves the efficiency and scientificity of risk response.

[0039] The system introduces self-adaptive learning and optimization mechanism. Through reinforcement learning and continuous feedback, the evaluation model, prediction model and decision recommendation algorithm of the system can be iteratively optimized according to the actual operation effect and project feedback, so that the system performance is continuously improved with time and data accumulation, and the risk management ability is self-evolved.

[0040] In summary, the engineering investment project multi-dimensional risk dynamic evaluation and early warning system provided by the present application provides an efficient, intelligent and comprehensive risk management solution for engineering investment projects, significantly improves the project risk management level, effectively reduces the risk of project failure, and guarantees the investment benefit. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 It is the overall technical scheme architecture schematic diagram of the present application. DETAILED DESCRIPTION

[0042] The embodiment provides an engineering investment project multi-dimensional risk dynamic evaluation and early warning system. The system aims to overcome the limitations of single evaluation dimension, insufficient timeliness, inability to dynamically track risk evolution, and lack of active early warning and intelligent decision support in the existing engineering investment project risk evaluation. By integrating multiple heterogeneous data sources, constructing cross-dimensional risk index system, using advanced machine learning and deep learning algorithms for dynamic risk modeling, real-time evaluation and trend prediction, and combining intelligent early warning mechanism and decision support module, the system provides continuous, comprehensive and accurate risk management services for the whole life cycle of engineering investment projects.

[0043] Please refer to Figure 1The multi-dimensional risk dynamic assessment and early warning system for the engineering investment project includes a data collection and preprocessing module, a multi-dimensional risk feature construction module, a dynamic risk assessment and prediction module, a risk early warning and visualization module, and an intelligent decision support and optimization module. The modules work together to comprehensively, dynamically, and intelligently manage the risks of the engineering investment project.

[0044] Data collection and preprocessing module The data collection and preprocessing module is used to obtain engineering investment project related data from multiple sources and clean, standardize, integrate, and format the data. The data collection process is the basis for system operation, and its accuracy and timeliness directly affect the effectiveness of subsequent risk assessment and early warning.

[0045] The data sources cover various aspects of the engineering investment project, including but not limited to project management system data, financial management system data, supply chain management system data, on-site Internet of Things sensor data, external macroeconomic data, market intelligence data, and environmental and social data.

[0046] The data obtained from the project management system includes the project schedule, actual progress, resource consumption, cost budget, actual cost, contract management information, and change records. These data are obtained from enterprise-level project management platforms through standardized application programming interfaces or database connectors, ensuring real-time and complete data transmission. The system periodically polls the project management system for updates and establishes data verification mechanisms, such as checksum or hash value comparison, to ensure data accuracy and perform preliminary data format verification. Typical progress data includes key path task completion percentage and milestone achievement, and cost data includes detailed spending and budget variance analysis.

[0047] The data obtained from the financial management system includes project cash inflow and outflow, financing status, cash flow, investment return rate, and financial statement data. The system connects with enterprise financial management software through data interfaces to obtain real-time financial transaction records, account balances, and various financial indicators. To ensure data privacy and security, encrypted protocols are used for data transmission, and strict access control strategies are implemented. For example, cash flow data is detailed to daily or weekly net cash flow, and investment return rate is dynamically calculated based on the latest financial performance.

[0048] Data acquired from the supply chain management system includes supplier performance, material procurement information, logistics status, and inventory level data. These data are obtained through integration with the supplier management platform or the logistics information system. The system regularly assesses the on-time delivery rate, quality pass rate, and cost control ability of suppliers, and tracks the status of procurement orders and the logistics track of key materials, such as the comparison of the expected arrival time and the actual arrival time of materials. Inventory level data is used to monitor the reserves of key materials to avoid project delays due to shortages.

[0049] Data acquired from the field Internet of Things sensors includes real-time data on environmental parameters, equipment operating status, personnel activities, and material consumption at the construction site. These data are collected through the Internet of Things platform, and the sensor network covers all key areas of the construction site. Environmental parameters include temperature, humidity, wind speed, and dust concentration; equipment operating status includes the working time, failure rate, and energy consumption of large machinery; personnel activity data is obtained through wearable devices or video analysis systems to monitor personnel safety and work efficiency; material consumption data is recorded in real time through intelligent weighing devices or RFID tags. Data enters the system in a streaming manner and is subjected to real-time anomaly detection.

[0050] Data acquired from external macroeconomic data includes gross domestic product, inflation rate, interest rate, exchange rate, industry sentiment index, and relevant policy and regulation update information. These data are obtained through public data interfaces, third-party data service providers, or web crawler technology. The system configures the frequency of data acquisition, such as quarterly updates for gross domestic product data and real-time updates for exchange rate data. Through the interface with the regulations database, the system can timely obtain and analyze the latest changes in policies and regulations, and assess their impact on the project.

[0051] Data acquired from market intelligence data includes changes in supply and demand in target markets, competitor dynamics, technology development trends, and consumer behavior data. These data are obtained through market research reports, news media monitoring, and industry analysis platforms. The system uses natural language processing technology to extract information and analyze sentiment from unstructured market intelligence to identify potential market risks and opportunities. For example, analyzing the impact of competitors' new product launches or technological breakthroughs on project market share.

[0052] Data acquired from environmental and social data includes climate change, natural disaster risk, community relations, public opinion, and environmental compliance information in the project location. These data are obtained through geographic information systems, weather bureau data interfaces, social media monitoring, and public reports from environmental protection departments. The system analyzes the historical frequency of natural disasters in the project location and conducts sentiment tendency assessment on public attitudes towards the project to identify potential social risks.

[0053] The data collection and preprocessing module aligns the collected heterogeneous data with a unified timestamp, fills in missing values, detects and corrects outliers, and normalizes the data. First, aligning the unified timestamp is to synchronize data from different data sources with different sampling frequencies to a pre-set unified time granularity. For high-frequency data, the system may perform downsampling or aggregation operations; for low-frequency data, the system uses interpolation algorithms such as linear interpolation or cubic spline interpolation to fill in the data to the unified timestamp. This process ensures consistency of all data in the time dimension, laying the foundation for subsequent time series analysis.

[0054] Second, missing value filling is completed by interpolation algorithm or machine learning-based prediction model. When there are missing values in the data set, the system will choose the appropriate filling strategy according to the type and characteristics of the data. For time series data, common interpolation methods include linear interpolation, polynomial interpolation or spline interpolation. For non-time series data, the system may use K-Nearest Neighbor algorithm-based filling to predict and fill in missing data points based on the characteristics of the K nearest data points; or use a regression model to predict missing values, such as using a random forest model to predict missing numerical features based on other characteristics. The intrinsic patterns and correlations of the data are considered during the filling process to maintain the original distribution of the data to the greatest extent.

[0055] Third, outlier detection is achieved through statistical methods or based on the Isolation Forest algorithm. Statistical methods include Z-score method or boxplot method to identify data points outside the predetermined statistical range. Z-score method judges the anomaly by calculating the standard deviation multiple of the data point from the mean. The Isolation Forest algorithm is an unsupervised learning method that isolates outliers by randomly selecting features and randomly dividing data points, which is efficient and accurate in handling high-dimensional data. The identified outliers will be corrected according to the preset strategy, such as replacing the mean, median or upper and lower limit values of the feature, or marking them as abnormal data without modification for subsequent model processing.

[0056] Finally, data normalization converts data of different dimensions into a unified numerical range to eliminate the impact of dimension differences on subsequent analysis. Common normalization methods include min-max normalization or Z-score standardization. Min-max normalization scales data to a range of zero to one, suitable for cases where data distribution is not significantly skewed. Z-score standardization converts data to a distribution with a mean of zero and a standard deviation of one, suitable for most machine learning algorithms. The system uses the min-max normalization method, which is represented by the formula: where, represents the original data value, represents the minimum value of the feature in the data set, max_value represents the maximum value of the feature in the dataset, normalized_value represents the normalized data value. This normalization process ensures that all features are consistent in numerical scale, preventing certain features from dominating the model training due to their large value range.

[0057] The data preprocessing and integration module stores the processed data into a distributed data warehouse. The distributed data warehouse adopts columnar storage and distributed computing architecture, such as Apache Parquet or ORC file format, and runs on top of the Hadoop distributed file system. The design of the data warehouse considers the high availability, scalability, and query efficiency of the data, supporting subsequent modules for efficient access and processing through data partitioning, indexing, and caching mechanisms, to meet the storage and analysis needs of large-scale data. Data version control is also implemented during data storage to ensure traceability of the data.

[0058] Multi-dimensional risk feature construction module The multi-dimensional risk feature construction module is used to extract and construct multi-dimensional risk features from pre-processed data. This module is the core link of risk assessment, which transforms raw data into risk features with higher information density, providing high-quality input for subsequent dynamic risk assessment and prediction.

[0059] The multi-dimensional risk feature construction module constructs risk features through the following ways: time series feature extraction, text information feature extraction, graph structure feature construction, and expert knowledge and rule features.

[0060] Time series feature extraction: For time series data such as project progress, cost, resource consumption, etc., extract its trend, periodicity, volatility, and autocorrelation features. Calculate the mean, standard deviation, maximum value, minimum value, rate of change, and seasonal index as features through sliding time window. Specifically, trend features are represented by calculating the linear regression slope or moving average of time series to indicate the long-term upward or downward trend. Periodicity features are identified through Fourier transform or seasonal decomposition to identify repeating patterns in the data, such as quarterly fluctuations in project costs. Volatility features are quantified by calculating the standard deviation, variance, or mean absolute deviation of time series to quantify data instability. Autocorrelation features capture the dependence of data points by calculating the correlation coefficient between the time series and its own lagged values. The parameters of the sliding time window, including window size and sliding step, will be configured according to data characteristics and business needs, such as setting the window size to four to twelve weeks for weekly data to capture monthly or quarterly patterns. Change rate features include year-on-year growth rate, month-on-month growth rate, or simple first-order difference, reflecting the speed and direction of data changes at different time points.

[0061] Text information feature extraction: For unstructured text data such as project reports, contract documents, policy files, news reports, and public comments, natural language processing techniques are used to extract key words, topic model features, sentiment orientation features, and entity relationship features. Text data is first preprocessed through steps such as word segmentation, stop word removal, and morphological restoration. Key word features are calculated using the term frequency-inverse document frequency algorithm (TF-IDF) to highlight important words in the document. Topic model features identify the underlying theme distribution in the document using models such as Latent Dirichlet Allocation (LDA), representing each document as a topic vector. Sentiment orientation features identify the positive, negative, or neutral emotions expressed in the text using sentiment dictionaries and machine learning classifiers, quantifying the sentiment color of public opinion or internal project reports. Entity relationship features identify key entities (such as project names, companies, personnel, locations, and times) and their semantic relationships (such as "Company A invests in Project B") through named entity recognition (NER) and relationship extraction techniques. Word embedding models such as Word2Vec or BERT are used to convert text into high-dimensional vector representations that capture semantic and syntactic information, facilitating subsequent processing by machine learning models.

[0062] Graph structure feature construction: Entities in the project, including project tasks, resources, personnel, risk events, policies and regulations, and stakeholders, are constructed as nodes. Relationships between entities, including dependency, influence, membership, causality, and interaction, are constructed as edges, forming a project knowledge graph. Entity nodes include specific person entities such as project managers and chief engineers, resource entities such as excavators and concrete, task entities such as foundation construction and structural topping, risk event entities such as fluctuations in raw material prices and policy adjustments, policy and regulation entities such as environmental regulations and industry standards, and stakeholder entities such as owners, contractors, and regulatory agencies. These entities are connected through their attributes and relationships, such as "project manager responsible for task A", "excavator used for task B", "raw material price fluctuations affecting cost risk", etc. Node features, edge features, and graph topology features are extracted from the knowledge graph. Node features can be node attributes (such as personnel's professional background, equipment's model parameters) or node vectors generated by graph embedding algorithms (such as Node2Vec, DeepWalk). Edge features describe the type and strength of the relationship. Graph topology features include node degree centrality, betweenness centrality, closeness centrality, and clustering coefficient, reflecting the importance and connection pattern of entities in the knowledge graph. These graph structure features can reveal the complex, nonlinear association between project entities, capturing risk transmission paths that traditional linear models cannot discover.

[0063] Expert knowledge and rule features: Combine domain expert experience to convert known risk indicators, risk trigger conditions, and risk level assessment rules into structured features or logical judgment features. For example, experts set the rule "if project progress lags more than 10% and key material supply is interrupted, trigger high-level progress risk warning" based on historical experience. These rules are coded by the rule engine to generate binary features (e.g. "whether the progress risk trigger condition is met") or numerical features (e.g. "expert assessment of risk weight"). These features compensate for the shortcomings of purely data-driven models in terms of data sparsity or lack of interpretability, introducing human wisdom and domain insight.

[0064] The multi-dimensional risk feature construction module fuses the various types of features extracted above to form a unified multi-dimensional feature vector space, fully representing the risk state of the engineering investment project in different dimensions. Feature fusion uses weighted summation, feature splicing, or deep learning-based feature fusion network to preserve the semantic information of different features and enhance their expression ability. Weighted summation combines different types of features into a comprehensive feature by assigning them preset or learned weights. Feature splicing simply concatenates all feature vectors to form a longer vector. Deep learning-based feature fusion network automatically learns the best combination method and weight distribution between different features through multi-layer perceptron or attention mechanism, generating more representative and discriminative feature representations. For example, through an attention mechanism layer, the system can dynamically assign different attention weights to time series features, text sentiment features, and graph structure features according to the current risk situation, highlighting the most relevant risk dimensions.

[0065] Dynamic risk assessment and prediction module The dynamic risk assessment and prediction module is used to conduct real-time risk assessment and future risk trend prediction based on the constructed multi-dimensional risk features. This module is the core intelligent engine of the system, aiming to provide accurate and forward-looking risk information.

[0066] The dynamic risk assessment and prediction module includes a risk factor identification and quantification unit, a risk correlation analysis unit, a dynamic risk state assessment unit, and a risk trend prediction unit.

[0067] A risk factor identification and quantification unit is configured to identify and quantify specific risk factors in each dimension. Through principal component analysis, factor analysis, or independent component analysis, the high-dimensional features are reduced and potential independent risk factors are identified. For example, "liquidity risk" or "debt serviceability risk" is identified from a large number of financial indicators; "schedule delay risk" or "resource shortage risk" is identified from project schedule and resource consumption data. These risk factors are linear or nonlinear combinations of original features, which can more essentially reflect the risk status of the project. Principal component analysis reduces dimensionality by finding the direction with the largest data variance, and factor analysis assumes that observed variables are driven by a small number of latent factors. Independent component analysis aims to find statistically independent latent components. After dimensionality reduction, each risk factor is quantified by a support vector machine model or a random forest model, and the risk exposure or probability value is output. Support vector machines are used to build classifiers to classify risk factors into different levels, such as low, medium, and high risk; random forest models can be used for regression tasks, directly outputting risk probability values between 0 and 1 or continuous risk exposure levels. The accuracy of the risk factor quantification result is crucial, as it directly affects the reliability of subsequent risk assessment.

[0068] A risk correlation analysis unit is configured to analyze the interaction relationships and transmission mechanisms between different risk factors. By constructing a Bayesian network or a graph neural network model, the identified risk factors are taken as nodes, and their causal relationships, conditional dependencies, or correlation relationships are taken as edges to model risk transmission paths and impact strength. Bayesian networks represent the conditional probability relationships between risk factors through directed acyclic graphs, enabling probabilistic reasoning to reveal how changes in one risk factor affect the probability of other risk factors. For example, "raw material price increase" may increase the probability of "cost overrun risk", which in turn affects "project profit margin decline risk". Graph neural network models can more effectively handle complex nonlinear correlations by using message passing mechanisms between nodes to learn the propagation patterns and impact ranges of risks in a knowledge graph. For example, a "policy and regulation change" node may affect multiple "project task" nodes through edges, leading to the expansion of "schedule delay risk". This model can reveal the "butterfly effect" risk amplification mechanism, i.e., how a change in one dimension of risk triggers risks in other dimensions, such as how a small deviation in technical risk can be amplified into a serious financial risk through supply chain disruption mechanisms.

[0069] The dynamic risk status assessment unit calculates the project's current comprehensive risk index or risk vector based on real-time updated multi-dimensional risk characteristics and risk correlation models. This unit employs a recurrent neural network model, such as a Long Short-Term Memory (LSTM) network or a gated recurrent unit network (GRNN), taking historical risk status, current risk characteristics, and risk correlation analysis results as input, and outputting the project's real-time risk scores across various dimensions and its overall risk level. LTM networks are particularly well-suited for processing time-series data, capturing the temporal dependence and evolution of risk status. LTM networks, through their unique gating mechanisms (input gate, forget gate, output gate), effectively solve the gradient vanishing or exploding problems of traditional LTM networks, enabling them to learn long-term dependencies. The GRNN, as a simplified version, reduces the number of parameters while maintaining similar performance. The model's input includes the quantified risk factor values ​​from the past N time steps, the real-time feature vector of the current time step, and the risk transmission influence factors output by the risk correlation analysis model. The model outputs a multi-dimensional vector, where each dimension represents a specific risk score, such as a technology risk score, market risk score, or financial risk score, as well as a comprehensive risk index, which is calculated using a weighted sum or a more complex aggregation function. The formula for calculating the comprehensive risk index is as follows: in, This represents the overall risk index. Indicates the number of risk dimensions. Indicates the first Scoring of each risk dimension Indicates the first Weights for each risk dimension. It can be pre-set based on expert experience or dynamically learned through machine learning models to reflect the importance of different risk dimensions in a specific project or stage.

[0070] The risk trend prediction unit is used to predict the risk development trend of a project over a future period. This unit is based on time series forecasting models, such as the ARIMA model, Prophet model, or Transformer model, to predict various risk indicators and the overall risk index for future time steps. The ARIMA model models the trend, seasonality, and random fluctuations of data by performing differencing, autoregression, and moving average operations on the time series. The Prophet model is suitable for time series with significant seasonality and holiday effects, using a decomposable model to decompose the time series into trend, seasonality, and holiday effects for prediction. The Transformer model utilizes its self-attention mechanism to capture the dependencies between any two points in the time series, making it particularly suitable for handling long series and complex patterns. The prediction results include point estimates of future risk values ​​and confidence intervals, providing a time window for early intervention. For example, predicting the possible range of the overall risk index within the next quarter, and the probability that technical risks will exceed the warning threshold within the next month. A simplified autoregressive model can be expressed as: in, It is the current moment. The risk index, It is a constant term. It is the autoregressive coefficient. It is a past moment Risk Index It is the order of the model. This is the error term. This model predicts future risk values ​​using a linear combination of historical risk indices.

[0071] The dynamic risk assessment and prediction module employs a reinforcement learning mechanism for adaptive optimization. The system's assessment and prediction models continuously learn and adjust parameters based on the occurrence of actual risk events, the effectiveness of risk mitigation measures, and the project's final risk performance to improve the accuracy of assessment and prediction. The reinforcement learning agent learns the optimal risk assessment strategy and prediction model weights in a simulated project environment. The reinforcement learning framework includes an agent, environment, state, action, and reward. The agent is the risk assessment and prediction module; the environment is the engineering investment project and its dynamic changes; the state is the multi-dimensional risk characteristics of the current project; and the action is adjusting model parameters or selecting different assessment / prediction strategies. When the prediction results closely match the actual situation, and the risk can be identified and effectively mitigated in a timely manner, the agent receives a positive reward; conversely, it receives a negative reward. Through continuous trial and error and learning, the agent can discover the optimal model configuration and strategy, enabling its assessment and prediction capabilities to continuously improve over time and with data accumulation.

[0072] Risk warning and visualization module The risk early warning and visualization module is used to generate and publish early warning information according to the dynamic risk assessment results, and to display the risk state of the project in an intuitive and friendly manner. This module is the core interface for the risk management system to interact with users, ensuring timely communication and effective understanding of risk information.

[0073] The risk early warning and visualization module includes an early warning rule configuration unit, an early warning information generation and publishing unit, and a risk situation visualization unit.

[0074] The early warning rule configuration unit is used to define early warning thresholds and trigger conditions for different risk levels. Early warning thresholds can be flexibly configured according to the type, stage, industry characteristics and risk preference of the project. For example, set the integrated risk index exceeding 0.6 as a yellow early warning, indicating that the risk is at a moderate level and needs close attention; set the integrated risk index exceeding 0.8 as an orange early warning, indicating that the risk level is high and immediate intervention measures need to be taken; set the integrated risk index exceeding 0.95 as a red early warning, indicating that the project is facing extremely high risk and needs urgent response. These thresholds are configured through a user interface and support statistical analysis based on historical data to suggest reasonable threshold ranges. Trigger conditions can include single risk factor exceeding, multiple risk factors simultaneously abnormal or risk trend continuously deteriorating. For example, a single risk factor such as "insufficient financial cash flow rate" exceeding a certain percentage can trigger an early warning. Multiple risk factors simultaneously abnormal may manifest as "schedule delay rate" and "cost overrun rate" simultaneously exceeding the preset threshold. Risk trend continuously deteriorating refers to a certain risk indicator showing an upward trend over consecutive time periods, even if it has not yet reached the absolute threshold, an early warning should be triggered.

[0075] The early warning information generation and release unit is used to automatically generate early warning information when the early warning rules are met, and notify relevant stakeholders through multiple channels. The early warning information includes risk type, risk level, trigger time, impact range, potential consequences and suggested attention matters. The risk type clearly indicates which type of risk, such as technical risk, market risk or financial risk. The risk level uses yellow, orange or red to identify its severity. The trigger time is accurate to the second, which is convenient for tracing. The impact range describes the affected project area, task or department. The potential consequences quantify the potential losses caused by the risk, such as the expected number of days of delay or additional cost. The suggested attention matters provide preliminary action recommendations to guide further analysis by the recipient. The release channels include system built-in message notification, email, mobile phone message and instant messaging software. The system built-in message notification is pushed in real time through the user interface, ensuring that system users can receive it immediately. Email notification is sent through SMTP protocol, suitable for non-urgent but detailed record notifications. Mobile phone message is sent through SMS gateway API, suitable for urgent and fast-acting early warnings. Instant messaging software notification is pushed through the API interface of enterprise WeChat or DingTalk platform, meeting the needs of team collaboration. The notification priority is dynamically adjusted according to the risk level, and high-level risk information will be sent through multiple channels at the same time, and may be accompanied by voice reminders.

[0076] The risk situation visualization unit is used to visually display the multidimensional risk situation of the project through interactive dashboards, risk maps, Gantt charts and trend charts. The dashboard, as the core visualization interface of the system, displays the real-time value of the current comprehensive risk index, the column chart or radar chart of each dimension risk score, the pie chart or ring chart of risk distribution, and the table or bubble chart of risk hotspot area. Users can drill down on the dashboard to view detailed data of specific risk dimensions. The risk map is based on geographic information system, which visually displays the risk levels of different areas or sub-projects on the map through color coding (such as green for low risk and red for high risk), making it easy to grasp the spatial distribution of risks. The Gantt chart combines risk events with project progress, showing the potential impact of risks on the duration by marking the occurrence time, duration and impact on key tasks on the time axis, such as highlighting the color of risk events on the critical path and showing the potential duration delay. The trend chart shows the historical risk evolution trajectory and future risk prediction trend, and the confidence interval of the predicted value is attached to the line chart to show the change of each risk indicator over time, enabling users to understand the dynamic change rule and future direction of risks. Users can adjust the time range of view through the time slider, or select different risk indicators for comparative analysis.

[0077] The risk early warning and visualization module supports user-defined early warning views and report generation, making it easy for users to monitor and report according to their own concerns. Users can customize the risk indicators, visualization chart types, and early warning rules they are interested in according to their roles and responsibilities, such as project manager, financial director, or technical director, and save these settings as personalized views. The system supports the generation of customized risk reports at regular intervals, and the content and format of the reports can be configured, such as generating weekly or monthly reports in PDF or Excel format and automatically sending them to a specified list of recipients.

[0078] Intelligent decision support and optimization module The intelligent decision support and optimization module is used to provide targeted risk mitigation suggestions for users after identifying risks and issuing early warnings, and to assist in optimizing project management decisions. This module is a reflection of the system's intelligence level, aiming to transform risk information into actionable solutions.

[0079] The intelligent decision support and optimization module includes a risk mitigation scheme library, a scheme intelligent matching and recommendation unit, a decision effect simulation and evaluation unit, and an optimization suggestion generation unit.

[0080] The risk mitigation scheme library stores historical successful risk mitigation strategies, best practice cases, and expert experience knowledge. Each scheme in the scheme library is associated with applicable risk types, implementation conditions, expected effects, and required resources. Risk mitigation schemes are stored in structured data form, including but not limited to scheme name, detailed description, applicable risk dimension, trigger scenario description, expected cost, expected mitigation effect (such as percentage reduction in risk index), required human resources, required material resources, implementation period, and past implementation success rate. For example, for the "raw material price fluctuation risk", the scheme library may contain multiple schemes such as "locking in supplier contracts in advance", "implementing hedging", and "finding alternative material suppliers". Each scheme is accompanied by its actual application case and effect evaluation data in historical projects.

[0081] The project intelligence matching and recommendation unit is used to intelligently match and recommend the optimal mitigation solution from the risk mitigation solution library according to the specific risk scenario, risk type, risk level, and project resource constraints of the current project. This unit uses a case-based reasoning algorithm or a collaborative filtering algorithm, combined with the similarity of the current risk characteristics and historical cases, and the effectiveness evaluation of the solution, to give the top-ranked recommended solution. The case-based reasoning algorithm identifies the most similar risk scenario by calculating the similarity between the feature vector of the current risk scenario and the feature vector of the historical case in the solution library, and recommends the successful mitigation solution in that scenario. The similarity calculation can use cosine similarity, Euclidean distance, or other more complex measurement methods. The collaborative filtering algorithm recommends solutions based on user evaluations of historical solutions or similar project-selected solutions. The recommendation process also considers the current resource constraints of the project, such as budget limitations, personnel availability, and time windows, to ensure the feasibility of the recommended solution. The recommended results are presented in a list form, including the solution name, main characteristics, expected effect, and priority ranking.

[0082] The decision effect simulation and evaluation unit is used to simulate and evaluate the potential effects of the recommended risk mitigation solutions on the project goals. Through Monte Carlo simulation or discrete event simulation models, the implementation effects of different mitigation solutions in the project environment are simulated, and the degree of improvement of project progress, cost, quality, and other risk indicators is predicted. Monte Carlo simulation estimates the effects of solutions under uncertainty conditions through a large number of random samples, such as simulating the impact of the "accelerate progress" solution on the distribution of project completion time, providing the probability distribution of project advance or delay. Discrete event simulation models simulate the execution of project tasks, allocation of resources, and occurrence of events to evaluate the impact of solutions on project processes and resource utilization, such as simulating how the "increase staff" solution shortens the duration of critical tasks. The unit also evaluates the new risks or secondary risks that may be brought about by the implementation of the solution, such as "increasing overtime to speed up progress may lead to staff fatigue and cause safety risks". The simulation results are presented in quantitative indicators and visual charts, such as the amount of expected project completion time reduction, cost savings, risk index decline percentage.

[0083] An optimization suggestion generation unit is used to generate the optimal decision optimization suggestion based on comprehensive consideration of risk mitigation effect, implementation cost, resource availability and overall project goals. This unit comprehensively analyzes the scheme list given by the scheme intelligent matching and recommendation unit and the simulation results of the decision effect simulation and evaluation unit, uses multi-objective optimization algorithms such as particle swarm optimization or genetic algorithm, and under the multiple constraints of maximizing mitigation effect, minimizing implementation cost, rationalizing resource utilization and optimizing overall project goals, finds the optimal decision combination. For example, when the project progress delay risk is warned, the system suggests adjusting resource allocation, such as deploying non-critical path manpower to critical path tasks; optimizing workflow, such as using parallel operation instead of serial operation; or introducing new technology to speed up the process, such as using modular construction. When the cost overrun risk is warned, the system suggests renegotiating the supplier contract to get a more favorable price; find alternative materials to reduce procurement costs; or optimize procurement strategy to achieve bulk discount. The optimization suggestion also includes specific steps to implement these decisions, expected effect evaluation and potential risk prompt.

[0084] The intelligent decision support and optimization module supports users to revise and feedback the recommended scheme, and incorporates the feedback results into the system learning cycle, continuously optimizes the recommendation algorithm and the scheme library content. Users can score, comment or modify the recommended scheme, and record the effect of the actual implementation scheme. These user feedback data are collected and used for retraining of the reinforcement learning model to continuously improve the accuracy and practicality of the scheme intelligent matching and recommendation algorithm. At the same time, successful revised schemes and new best practice cases will be updated to the risk mitigation scheme library, ensuring the timeliness and comprehensiveness of the scheme library.

[0085] To sum up, the engineering investment project multi-dimensional risk dynamic assessment and early warning system proposed in this embodiment can convert operation scheduling data and equipment state data into feature representation in a unified semantic space through structured embedding coding technology. Then, based on the scheduling scheme analysis and processing of the semantic search architecture, the alternative scheduling scheme can be dynamically matched with real-time operation demand and equipment state, and through the scheme adaptation degree evaluator of the decoder architecture, the equipment utilization, transportation cost and medical staff satisfaction are comprehensively weighed, and the global optimal scheduling decision is generated to solve the problems of data heterogeneity and dynamic optimization. In addition, the intelligent path planning reduces the time and labor cost of equipment transportation, and provides an efficient and adaptive solution for the whole life cycle management of the equipment in the operation area. The system realizes the transformation of the engineering investment project risk assessment from static and discrete to dynamic and continuous, breaks the single dimension limit of risk assessment, has intelligent risk factor identification and correlation analysis capability, provides accurate risk trend prediction function, has intelligent early warning mechanism and decision support capability, and introduces adaptive learning and optimization mechanism. Finally, the system provides an efficient, intelligent and comprehensive risk management solution for the engineering investment project, significantly improves the project risk management level, effectively reduces the risk of project failure, and guarantees the investment benefit. The system can be implemented in various terminal devices, integrated into the terminal device as a software module and / or hardware module. For example, the system can be a software module in the operating system of the terminal device, or an application program developed for the terminal device; of course, the system can also be one of the many hardware modules of the terminal device. Alternatively, the system and the terminal device can also be separate devices, and the system can be connected to the terminal device through a wired and / or wireless network, and transmit interactive information according to the agreed data format.

Claims

1. A multi-dimensional dynamic risk assessment and early warning system for engineering investment projects, characterized in that, include: The data acquisition and preprocessing module is used to acquire relevant data of engineering investment projects from multiple sources, and to clean, standardize, integrate and unify the format of the relevant data; The multi-dimensional risk feature construction module is used to extract and construct multi-dimensional risk features from the preprocessed data to comprehensively characterize the risk status of the engineering investment project in different dimensions. The dynamic risk assessment and prediction module is used to perform real-time risk assessment and future risk trend prediction based on the constructed multi-dimensional risk characteristics. The risk warning and visualization module is used to generate and issue warning information based on the dynamic risk assessment results, and to display the risk status of the engineering investment project in an intuitive and user-friendly manner. as well as The intelligent decision support and optimization module is used to provide users with targeted risk mitigation suggestions and assist in optimizing project management decisions after identifying risks and issuing warnings.

2. The multi-dimensional risk dynamic assessment and early warning system for engineering investment projects according to claim 1, characterized in that, The data acquisition and preprocessing module is used to acquire relevant data of the engineering investment project from at least one of the following data sources: Project management system data, which includes project schedule, actual progress, resource consumption, cost budget, actual cost, contract management information, and change records; Financial management system data, which includes the project's cash inflows and outflows, financing status, cash flow, return on investment, and financial statement data; Supply chain management system data, which includes supplier performance, material procurement information, logistics status, and inventory level data; On-site IoT sensor data includes real-time data on environmental parameters, equipment operating status, personnel activities, and material consumption at the construction site. External macroeconomic data, including gross national product, inflation rate, interest rate, exchange rate, industry prosperity index, and updates to relevant policies and regulations; Market intelligence data, including changes in supply and demand in the target market, competitor dynamics, technological development trends, and consumer behavior data; as well as Environmental and social data, including information on climate change, natural disaster risks, community relations, public opinion, and environmental compliance in the project location.

3. The multi-dimensional risk dynamic assessment and early warning system for engineering investment projects according to claim 1, characterized in that, The data acquisition and preprocessing module is also used to perform unified timestamp alignment, missing value filling, outlier detection and correction, and data normalization on the acquired heterogeneous data. The missing value imputation is accomplished by an interpolation algorithm or a prediction model based on machine learning, wherein the interpolation algorithm includes linear interpolation or cubic spline interpolation; The outlier detection is achieved through statistical methods or based on the isolated forest algorithm; the statistical methods include Z-score method or box plot method; and The data normalization transforms data with different dimensions into a unified numerical range to eliminate the impact of dimensional differences on subsequent analysis. The data normalization adopts the min-max normalization or Z-score standardization method.

4. The multi-dimensional risk dynamic assessment and early warning system for engineering investment projects according to claim 1, characterized in that, The multi-dimensional risk feature construction module constructs the risk features through at least one of the following methods: Time series feature extraction is used to extract the trend, periodicity, volatility and autocorrelation features of time series data of project progress, cost and resource consumption. The mean, standard deviation, maximum value, minimum value, rate of change and seasonality index are calculated as features through a sliding time window. The rate of change features include year-on-year growth rate, month-on-month growth rate or simple first difference. Text information feature extraction is used to extract keywords, topic model features, sentiment features, and entity relationship features from unstructured text data such as project reports, contract documents, policy documents, news reports, and public comments using natural language processing techniques. The keyword features are calculated using the word frequency-inverse document frequency algorithm, the topic model features are identified using the latent Dirichlet assignment model, and the text information is converted into a high-dimensional vector representation using a word embedding model. Graph structure feature construction is used to construct entities in the project, including project tasks, resources, personnel, risk events, policies and regulations, and stakeholders, as nodes; and to construct the relationships between entities, including dependencies, influence relationships, membership relationships, causal relationships, and interaction relationships, as edges, forming a project knowledge graph. Node features, edge features, and graph topology features are extracted from this knowledge graph. The graph topology features include the degree centrality, betweenness centrality, proximity centrality, and clustering coefficient of the nodes. Expert knowledge and rule features are used to combine domain expert experience to transform known risk indicators, risk triggering conditions, and risk level assessment rules into structured features or logical judgment features. These rules are encoded into binary features or numerical features by a rule engine.

5. The multi-dimensional risk dynamic assessment and early warning system for engineering investment projects according to claim 4, characterized in that, The multi-dimensional risk feature construction module is also used to fuse the extracted risk features to form a unified multi-dimensional feature vector space, which comprehensively represents the risk status of the engineering investment project in different dimensions. The feature fusion is achieved through weighted summation, feature concatenation, or a deep learning-based feature fusion network. The deep learning-based feature fusion network automatically learns the combination methods and weight allocation between different features through a multi-layer perception mechanism or attention mechanism, so as to retain the semantic information of different features and enhance their expressive power.

6. The multi-dimensional risk dynamic assessment and early warning system for engineering investment projects according to claim 1, characterized in that, The dynamic risk assessment and prediction module includes: The risk factor identification and quantification unit is used to identify and quantify specific risk factors in various dimensions. Through principal component analysis, factor analysis, or independent component analysis, it reduces the dimensionality of high-dimensional features and identifies potential, independent risk factors. These risk factors include technical risk, market risk, financial risk, management risk, and policy and regulatory risk. Each risk factor is quantified using a support vector machine model or a random forest model, and its risk exposure degree or probability value is output. The risk correlation analysis unit is used to analyze the interaction and transmission mechanism between different risk factors. By constructing a Bayesian network or graph neural network model, the identified risk factors are used as nodes, and the causal relationship, conditional dependency relationship or correlation relationship between the risk factors are used as edges to model the risk transmission path and the intensity of the impact, so as to reveal the risk amplification mechanism. The dynamic risk status assessment unit is used to calculate the current comprehensive risk index or risk vector of the engineering investment project based on the real-time updated multi-dimensional risk characteristics and risk correlation model, and output the real-time risk score and overall risk level of the engineering investment project in each dimension; and The risk trend prediction unit is used to predict the risk development trend of the engineering investment project in the future. Based on the time series prediction model, ARIMA model, Prophet model or Transformer model, it predicts the various risk indicators and comprehensive risk index of the engineering investment project in the future time step, and outputs the point estimate and confidence interval of the future risk value.

7. The multi-dimensional risk dynamic assessment and early warning system for engineering investment projects according to claim 6, characterized in that, The dynamic risk assessment and prediction module uses a reinforcement learning mechanism for adaptive optimization to improve the accuracy of the assessment and prediction models. The assessment and prediction models continuously learn and adjust their parameters based on the occurrence of actual risk events, the effectiveness of risk mitigation measures, and the final risk performance of the project. The reinforcement learning agent learns the optimal risk assessment strategy and prediction model weights in a simulated project environment, thereby achieving self-evolution of risk management capabilities.

8. The multi-dimensional risk dynamic assessment and early warning system for engineering investment projects according to claim 1, characterized in that, The risk warning and visualization module includes: The early warning rule configuration unit is used to define early warning thresholds and triggering conditions for different risk levels. The early warning thresholds are flexibly configured according to the type, stage, industry characteristics and risk preferences of the engineering investment project. The triggering conditions include a single risk factor exceeding the limit, multiple risk factors being abnormal at the same time, or a risk trend continuing to deteriorate. The early warning information generation and dissemination unit is used to automatically generate early warning information when the aforementioned early warning rules are met, and to notify relevant stakeholders through multiple channels including built-in system message notifications, emails, SMS messages, and instant messaging software. The early warning information includes the risk type, risk level, trigger time, scope of impact, potential consequences, and recommended precautions. The risk situation visualization unit is used to intuitively display the multi-dimensional risk situation of the engineering investment project through interactive dashboards, risk maps, Gantt charts, and trend charts. The risk situation visualization unit supports user-defined early warning views and report generation.

9. The multi-dimensional risk dynamic assessment and early warning system for engineering investment projects according to claim 1, characterized in that, The intelligent decision support and optimization module includes: The risk mitigation solution library stores historically successful risk mitigation strategies, best practice cases, and expert experience and knowledge. Each solution in the library is associated with the applicable risk type, implementation conditions, expected results, and required resources. The intelligent matching and recommendation unit is used to intelligently match and recommend the optimal mitigation solution from the risk mitigation solution library based on the specific risk scenario, risk type, risk level, and project resource constraints of the current engineering investment project. The intelligent matching and recommendation unit employs a case-based reasoning algorithm or a collaborative filtering algorithm. The decision effect simulation and evaluation unit is used to simulate the recommended risk mitigation schemes, evaluate their potential effects and impact on project objectives. The decision effect simulation and evaluation unit simulates the implementation effects of different mitigation schemes in the project environment through Monte Carlo simulation or discrete event simulation models, predicts the degree of improvement on project schedule, cost, quality and other risk indicators, and evaluates the new or secondary risks that the implementation of the schemes may bring.

10. The multi-dimensional risk dynamic assessment and early warning system for engineering investment projects according to claim 9, characterized in that, The intelligent decision support and optimization module also includes an optimization suggestion generation unit, which generates the optimal decision optimization suggestion based on a comprehensive consideration of risk mitigation effect, implementation cost, resource availability and overall goals of the engineering investment project. The optimization suggestion generation unit uses multi-objective optimization algorithms, particle swarm optimization or genetic algorithms. Furthermore, the intelligent decision support and optimization module allows users to revise and provide feedback on recommended solutions, and incorporates the feedback results into the system's learning loop to continuously optimize the recommendation algorithm and solution library content, thereby constantly improving the accuracy and practicality of intelligent solution matching and recommendation algorithms, and updating the risk mitigation solution library.

Images