Engineering risk prediction method based on artificial intelligence

Abstract

The invention provides an artificial intelligence-based engineering risk prediction method. The artificial intelligence-based engineering risk prediction method comprises the steps of constructing an engineering risk knowledge graph, designing a quantitative risk index system, developing a risk prediction intelligent algorithm, realizing risk prediction visualization, dynamically optimizing and updating and the like. According to the method, automatic association fusion of concept nodes is realized by adopting a knowledge fusion algorithm based on semantic similarity in knowledge graph construction; determining an index weight by using an analytic hierarchy process and a fuzzy comprehensive evaluation method in risk index quantification; in prediction algorithm development, risk event time sequence features are extracted, and a plurality of prediction models are fused by adopting a Stacking and Blending ensemble learning strategy; and developing a risk knowledge graph visual modeling tool in visual presentation, and introducing quantitative evaluation indexes such as graph density and the like. According to the method, intelligence, digitization and visualization of engineering risk prediction are realized, the problems of high risk assessment subjectivity, low prediction precision and the like in a traditional method are solved, and the method has important application value in engineering project risk management.

Description

Technical Field

[0001] The present invention relates to the technical field of project management, and particularly to an engineering risk prediction method based on artificial intelligence. Background Art

[0002] Project risk management is an important topic in the field of project management. During the implementation of a project, various risk factors such as technology, quality, safety, schedule, and cost are faced. If these risks cannot be identified in a timely manner, accurately evaluated, and effectively controlled, it may lead to serious consequences such as project delays, cost overruns, and quality accidents, causing huge losses to the project. Therefore, implementing effective engineering risk prediction and early warning is of great significance for the realization of the safety, schedule, cost, and quality objectives of the project.

[0003] Traditional engineering risk prediction mainly relies on expert experience judgment and historical data statistical analysis. Experts subjectively evaluate project risks based on their accumulated engineering experience and give qualitative prediction results such as the likelihood of risk occurrence and the degree of impact. On the other hand, by collecting risk event data of implemented projects and using statistical models for trend extrapolation, quantitative prediction of future risks is carried out. However, these traditional methods often have some common problems in practical applications:

[0004] Firstly, the risk assessment process is highly subjective and lacks a scientific quantitative index system. Expert experience judgment is limited by personal knowledge and vision, and the understanding of risks in a complex engineering environment is often not comprehensive enough, and the prediction results are prone to be one-sided. Due to background differences among different experts, the judgment criteria for the same risk are inconsistent, and the prediction results lack consistency.

[0005] Secondly, the accuracy of the risk prediction model is not high and the generalization ability is poor. Existing risk prediction models are mostly built on a small scale of historical data, with insufficient sample coverage, and it is difficult to accurately depict the risk evolution law in a complex engineering scenario. The models generally have problems of underfitting or overfitting, and it is difficult to guarantee the prediction accuracy.

[0006] Thirdly, the existing technology is difficult to fully utilize the value of risk management data. A large amount of unstructured risk knowledge is difficult to systematically organize and mine, and the feature construction method during model training is relatively simple, ignoring the evolution characteristics of time-series data, and not well converting risk knowledge into effective inputs for risk prediction.

[0007] In addition, the interpretability and visualization degree of risk prediction results are not enough. Existing prediction models are mostly black-box models, and the prediction results are difficult to explain. The presentation method of risk early warning information is not intuitive enough, affecting the acceptance and response efficiency of risk managers.

[0008] In recent years, artificial intelligence technologies represented by knowledge graphs and machine learning have made great progress, bringing new ideas for solving the above problems. Scholars at home and abroad have begun to explore introducing artificial intelligence methods into engineering risk prediction and have made some beneficial attempts. For example, some scholars have proposed constructing ontology models to structurally represent risk management knowledge; some scholars have used machine learning models such as support vector machines and decision trees to construct risk prediction models; and some scholars have introduced visualization technologies to establish a risk early warning visualization platform.

[0009] Generally speaking, the application of artificial intelligence technology in engineering risk prediction is still in the exploration stage, and there is still a large room for innovation and improvement in aspects such as constructing a complete engineering risk knowledge system, developing intelligent prediction models, and realizing the visualization of the whole process of risk early warning. It is urgent to, based on the existing research, follow the general laws and characteristics of engineering risk management, give full play to the advantages of artificial intelligence technology, and systematically develop a set of intelligent solutions specifically for engineering risk prediction. This is exactly the goal of the present invention, "An Engineering Risk Prediction Method Based on Artificial Intelligence". The present invention aims to conduct systematic innovative design from aspects such as methods, models, and systems, break through the limitations of traditional methods, and greatly improve the intelligent level of engineering risk prediction, with a view to achieving more accurate, efficient, and reliable engineering risk prediction and better guiding the practice of engineering risk prevention and control. Summary of the Invention

[0010] The present invention proposes an engineering risk prediction method based on artificial intelligence, aiming to solve problems existing in existing engineering risk prediction, such as strong subjectivity in risk identification and assessment, low prediction accuracy, and difficulty in adapting to complex engineering environments. This method takes the knowledge graph as the core and integrates artificial intelligence technologies such as machine learning, deep learning, and visualization analysis to form an intelligent solution for the whole process of risk identification, assessment, prediction, and early warning.

[0011] The technical solution of the present invention mainly includes the following key links:

[0012] Construct an engineering risk knowledge graph. Through steps such as knowledge acquisition, extraction, fusion, and organization, construct a multi-level engineering risk knowledge graph. Introduce a knowledge fusion algorithm based on semantic similarity to improve the accuracy and conciseness of knowledge representation. Use a graph database to achieve efficient storage and retrieval of knowledge.

[0013] Design a quantitative risk index system. Extract risk assessment indicators from the knowledge graph, use the analytic hierarchy process (AHP) to quantify the indicators, and establish a scientific and comprehensive risk assessment index system. Use the fuzzy comprehensive evaluation method to quantify qualitative indicators and introduce a consistency test to ensure the scientificity of index weight calculation.

[0014] Develop an intelligent algorithm for risk prediction. Integrate machine learning and deep learning technologies to build a risk prediction model. Design a feature engineering method for time series data of risk events to extract effective features such as moving average and exponentially weighted moving average. Adopt Stacking and Blending ensemble learning strategies to further improve the prediction performance. Introduce incremental learning and active learning mechanisms to achieve continuous optimization of the model.

[0015] Realize the visualization of risk prediction. Comprehensively apply information visualization and geographic information visualization technologies to present the risk prediction results from multiple perspectives. Develop visualization modules for risk indicators, risk events, risk trends, and risk knowledge graphs, and introduce interactive analysis functions to provide intuitive and easy-to-use analysis tools for risk managers.

[0016] Apply a dynamic optimization mechanism. Continuously collect risk management data during the engineering implementation process, regularly diagnose the risk indicator system and prediction model, and make dynamic corrections and improvements in combination with expert feedback. At the same time, associate with the project management information system to realize the engineering application of risk prediction.

[0017] The innovative effects of the present invention are mainly reflected in the following aspects:

[0018] For the first time, introduce the knowledge graph into engineering risk prediction. Through graph construction, realize the systematic organization and correlation analysis of risk knowledge, providing a solid knowledge foundation for intelligent prediction.

[0019] Propose a set of scientific risk index quantification methods. Adopt AHP and fuzzy comprehensive evaluation methods to address the problem of strong subjectivity in risk assessment.

[0020] Innovatively develop a time series feature engineering method for engineering risk prediction and introduce a multi-strategy ensemble learning model, greatly improving the accuracy and robustness of risk prediction.

[0021] Realize the visualization of the whole process of risk prediction, covering various visualization functions such as risk indicators, events, trends, and knowledge, greatly enhancing the intuitiveness and interpretability of risk early warning.

[0022] Propose a dynamic optimization mechanism to continuously improve the core components during the engineering implementation, realizing the seamless integration of this solution with engineering management and improving the engineering applicability of the solution.

[0023] In summary, the method of the present invention has carried out systematic innovative design from dimensions such as risk knowledge organization, risk assessment quantification, risk prediction modeling, prediction result visualization, and dynamic application optimization, and constructed a complete intelligent engineering risk prediction system. This system gives full play to the advantages of artificial intelligence technologies such as knowledge graphs, machine learning, and visual analysis, and provides a practical solution for solving the industry problems in engineering risk prediction. The present invention has theoretical innovation and practical application value, and is of great significance for improving the intelligent level of engineering project risk management and ensuring the realization of engineering safety, progress, cost, and quality goals. It is applicable to many engineering fields such as water conservancy and hydropower, transportation, and energy and power. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0025] Figure 1 Schematic diagram of the overall architecture of the engineering risk prediction system according to an embodiment of the present invention;

[0026] Figure 2 Schematic diagram of the structure of the engineering risk knowledge graph according to an embodiment of the present invention;

[0027] Figure 3 Schematic diagram of the hierarchical structure of the risk assessment indicators according to an embodiment of the present invention;

[0028] Figure 4 Schematic diagram of the model ensemble learning architecture according to an embodiment of the present invention;

[0029] Figure 5 Schematic diagram of the data processing and feature engineering process according to an embodiment of the present invention;

[0030] Figure 6 Schematic diagram of the interface layout of the risk prediction visualization system according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] The present invention will be described in detail below with reference to the drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; and the drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0032] It should be noted that in the specification, the indication of "an embodiment", "embodiment", "exemplary embodiment", "some embodiments", etc. means that the described embodiments may include specific features, structures or characteristics, but not necessarily every embodiment includes such specific features, structures or characteristics. Additionally, when combining embodiments to describe specific features, structures or characteristics, implementing such features, structures or characteristics in combination with other embodiments (whether explicitly described or not) should be within the knowledge scope of those skilled in the relevant art.

[0033] Generally, terms can be understood at least in part from their use in context. For example, at least in part depending on the context, the term "one or more" used herein can be used to describe any feature, structure or characteristic in a singular sense, or can be used to describe a combination of features, structures or characteristics in a plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey a set of exclusive factors, but rather, at least in part depending on the context, allowing for the existence of other factors that may not be explicitly described.

[0034] As Figure 1 shown, the engineering risk prediction system of the present invention includes a knowledge graph construction module, a risk index quantification module, a risk prediction modeling module, and a risk visual analysis module.

[0035] As Figure 2 shown, the engineering risk knowledge graph constructed by the present invention adopts a three - layer architecture, including a concept layer, a data layer, and an application layer.

[0036] I. Construction of the Engineering Risk Knowledge Graph

[0037] Knowledge Acquisition

[0038] Comprehensively collect literature materials on risk management in the engineering field, including academic papers, industry standards, regulations, patent documents, expert works, and risk case reports. Use web crawler technology to crawl web data related to risk management from authoritative websites. Organize expert interviews to obtain practical experience in engineering risk management.

[0039] Knowledge Extraction

[0040] Adopt an information extraction method based on rules and dictionaries to extract core concept terms of risk management from the original knowledge base. The concept terms include risk events, risk factors, risk consequences, and risk control measures. Identify the semantic relationships between concepts, and the relationship types include causal relationships, hierarchical relationships, and association relationships. Construct a risk ontology model to formally define concept categories and relationship types.

[0041] Knowledge Fusion

[0042] Adopt a knowledge fusion algorithm based on similarity to calculate the semantic similarity between concept terms and set a similarity threshold. For concepts with semantic similarity higher than the threshold, perform knowledge tuple alignment, identify duplicate and redundant knowledge representations, and merge them. Based on the hyponymy relationship, infer and generalize new concepts and relationships;

[0043] In the knowledge fusion stage of constructing the engineering risk knowledge graph, we adopted a fusion algorithm based on semantic similarity. The core of this algorithm is to calculate the semantic similarity between two concept nodes, and the formula is:

[0044]

[0045] where c i and c j represent two concept nodes, and the corresponding feature vectors are v i and v j . The calculation process of this formula is as follows:

[0046] First, it is necessary to extract feature vectors for each concept node in the knowledge graph. Common feature extraction methods include Word2Vec, GloVe, etc. These methods can map concept nodes to a real-valued vector space. Then, for the two concept nodes c i and c j to be fused, their feature vectors v i and v j are obtained respectively. Calculate the dot product v i ·v j of the two feature vectors according to the formula, as well as their respective L2 norms |v i | and |v j |. The L2 norm is also called the Euclidean norm, and its calculation formula is the square root of the sum of the squares of the absolute values of the vector elements. Divide the dot product by the product of the L2 norms of the two vectors to obtain the semantic similarity with a value in the range of [0, 1]. The closer this value is to 1, the more similar the semantics of the two concept nodes are. Set a similarity threshold. When the calculated similarity is greater than this threshold, it can be determined that the two concept nodes can be fused. The selection of the threshold needs to be adjusted according to the specific application scenario and fusion effect.

[0047] Through semantic similarity calculation, concept nodes with similar semantics in the knowledge graph can be automatically discovered and fused, thereby improving the accuracy and conciseness of knowledge representation.

[0048] Knowledge organization

[0049] Organize the fused knowledge based on the ontology model, and construct a multi-level knowledge graph architecture including a concept layer, a data layer, and an application layer. In the concept layer, define the core concept categories and organize the hierarchical relationships of the concepts. In the data layer, instantiate the concepts into specific data and establish the semantic connections between the data. In the application layer, organize the application-oriented knowledge for the risk prediction task.

[0050] Knowledge storage

[0051] Select the graph database Neo4j as the knowledge graph storage engine. Import the concept nodes and relationship edges in the knowledge graph into Neo4j. Each concept node contains information such as name, alias, definition, and attributes, and the relationship edge contains the start node, end node, and relationship type. Establish indexes for the nodes and edges in Neo4j to improve the retrieval efficiency of the graph. At the same time, establish a knowledge graph version management mechanism to record the graph update log for easy graph maintenance.

[0052] As Figure 3 shown, the risk assessment index system designed by the present invention adopts a hierarchical structure and unfolds step by step from the target layer, criterion layer to the specific index layer.

[0053] II. Design a quantitative risk index system

[0054] Index selection

[0055] Make full use of the engineering risk knowledge graph, and identify the risk attributes involved in the risk events and risk factors from the concept layer of the graph as candidate evaluation indexes. Invite experts in the field of engineering risk management, and use methods such as brainstorming and expert scoring to evaluate the importance and measurability of the candidate indexes, and select the key risk assessment indexes.

[0056] Index quantification

[0057] For qualitative indexes, design a scientific scoring rule. Adopt the fuzzy comprehensive evaluation method to establish a multi-level quantification standard, and divide the qualitative indexes into several quantification levels. Formulate detailed index scoring rules to clarify the judgment conditions for each quantification level. For quantitative indexes, perform data standardization processing to eliminate the influence of dimensions and achieve unified representation of the index values.

[0058] Index organization

[0059] Adopt the analytic hierarchy process (AHP) to construct an index evaluation model. Divide the selected risk assessment indexes into a target layer, a criterion layer, and an index layer to form a multi-level index structure. The target layer represents the overall risk assessment goal; the criterion layer includes the likelihood of risk occurrence, impact degree, etc.; the index layer is the specific measurement indexes included in the criterion layer;

[0060] When designing a quantitative risk index system, we use the Analytic Hierarchy Process (AHP) to calculate the index weights. This method first constructs a judgment matrix for pairwise comparison of indices through expert evaluation, and then calculates the index weights from the judgment matrix. The key steps are as follows:

[0061] Expert scoring. Invite domain experts to make pairwise comparisons of the importance of the indices, give a scoring value on a scale of 1 - 9, and fill it into the judgment matrix. Here, 1 means the two indices are equally important, 9 means one index is extremely more important than the other, and 2 - 8 are intermediate values. Calculate the maximum eigenvalue. According to the formula:

[0062]

[0063] where \(A\) is the judgment matrix, \(n\) is the order of the matrix, and \(w\) is the normalized eigenvector, representing the index weights. Since \(\lambda\) max and \(w\) need to be solved iteratively, the commonly used method is power method iteration: (1) Given an initial vector \(w_0\), such as a vector of all 1s. (2) Perform power method iteration until convergence: (3) The weight values of the corresponding indices can be obtained from the components of the normalized eigenvector \(w\).

[0064] Consistency check. To ensure that the judgment matrix has satisfactory consistency, the following check is required:

[0065] Calculate the consistency index:

[0066]

[0067] Look up the average random consistency index \(RI\) in the table and calculate the consistency ratio:

[0068]

[0069] If \(CR < 0.1\), it is considered that the consistency of the judgment matrix is acceptable, and the obtained weights can be used for decision-making; otherwise, the experts need to re-evaluate the importance of the indices until the consistency check is passed. Through the AHP method, the qualitative expert experience judgment can be transformed into quantitative index weights, and the logical consistency of the expert judgment can be tested by mathematical methods, so as to more scientifically and reasonably determine the importance of the risk assessment indices and provide support for subsequent risk prediction.

[0070] Index weights

[0071] Form a weight judgment group composed of engineering risk management experts. Using the 1 - 9 scale method, invite experts to make pairwise importance comparisons of the same-level indices to form a judgment matrix. Use the AHP method to calculate the maximum eigenvalue and the corresponding eigenvector of the judgment matrix, and obtain the index weight vector after normalization. Use the consistency check method to evaluate the logical consistency of the expert judgment and make feedback adjustments if necessary.

[0072] Dynamic optimization

[0073] Establish a dynamic optimization mechanism for the risk index system. During the implementation of the engineering project, continuously collect risk event data and record the actual values of risk indicators.

[0074] Regularly analyze the index data using methods such as factor analysis and data mining to identify the key indicators that have a significant impact on risk assessment. At the same time, organize expert review meetings to diagnose the deficiencies in the index system and dynamically revise the index system based on the feedback.

[0075] As Figure 4 shown, the model ensemble learning architecture adopted by the present invention includes an input layer, a machine learning model layer, a deep learning model layer, and an integration layer:

[0076] III. Develop an intelligent risk prediction algorithm

[0077] As Figure 5 shown, the data processing and feature engineering process of the present invention includes three main stages: data collection, data preprocessing, and feature engineering:

[0078] Data preparation

[0079] Widely collect engineering project risk management data, and the data types include: project basic information data, risk event data, risk loss data, and risk response data. Build a heterogeneous data collection framework, use a relational database to store structured data, and use a document database to store unstructured data. Perform data cleaning and data integration on the collected raw data to identify and eliminate noise data and outlier data.

[0080] Feature engineering

[0081] Based on the quantitative risk index system, construct a machine learning feature space. Extract the time series features of risk events, including the risk occurrence time, frequency, and trend features. Extract the statistical features of risk factors, including mean, variance, quantile, etc. Extract the numerical features of risk consequences, including risk loss amount, loss duration, and loss resource quantity. Build risk factor association features to describe the correlation between risk factors and risk events;

[0082] In the intelligent risk prediction algorithm, we perform feature engineering on the time series data of risk events. Two typical time series features are: moving average (MA) and exponentially weighted moving average (EWMA), and their calculation formulas are respectively:

[0083]

[0084] EWMAt (α) = αx t+(1-α)EWMAt-1 (α)

[0085] Here, x t represents the number of risk events at the t-th moment. The specific calculation process is as follows:

[0086] Determine the time window size k or the smoothing coefficient α. The larger the value of k, the higher the degree of smoothing; the larger the value of α, the higher the weight of recent data. Appropriate parameters need to be selected according to the actual scenario and requirements. For the MA feature, at each time step t, the mean value of the data of k time steps ending with it is selected as the MA feature value at the t-th moment. For the EWMA feature, at t = 1, take Thereafter, calculate according to the recurrence formula at each time step to obtain the EWMA feature value of the sequence at the t-th moment.

[0087] It can be seen that MA is a simple smoothing method that assigns the same weight to each data point within the sliding window; while EWMA is a weighted smoothing method that exponentially reduces the weight of data points over time and pays more attention to recent data. Extracting these two types of time series features can, on the one hand, filter out the noise in the time series data and mine the overall trend information of the data; on the other hand, it can provide more effective and stable input features for subsequent machine learning models, thereby improving the accuracy of risk event prediction. In practice, multiple time series features are often combined for modeling and prediction.

[0088] Model Training

[0089] Adopt a modeling strategy that combines machine learning and deep learning. Split the dataset according to a ratio of 7:3 into a training set and a test set. Use a five-fold cross-validation method on the training set to optimize and train machine learning models, including logistic regression, support vector machine, random forest, and XGBoost. In terms of deep learning, build a multi-layer perceptron neural network and a long short-term memory neural network to construct an end-to-end risk prediction model. Use the grid search method to optimize the model hyperparameters and use the EarlyStopping strategy to control the training process.

[0090] Model Ensemble

[0091] Adopt the Stacking ensemble learning framework to combine multiple trained machine learning models and deep learning models to form a strong learner. Utilize the Stacking principle to use the output of the primary learner as the input of the secondary learner and train the parameters of the secondary learner. Adopt the Blending method to stack the outputs of multiple homogeneous or heterogeneous learners with certain weights to form an integrated prediction result. Evaluate the performance of the integrated model on the test set and select the integrated strategy with the lowest generalization error;

[0092] At the end of the risk prediction intelligent algorithm, two ensemble learning strategies, Stacking and Blending, are adopted to combine the prediction results of multiple base learners in order to further improve the prediction performance. The prediction output of Stacking is:

[0093]

[0094] where f is the secondary learner, which is a meta-model that takes the prediction output of the primary learners as input. The prediction output of Blending is:

[0095]

[0096] where w i is the ensemble weight of the i-th learner. The main difference between the two methods is that Stacking first trains the primary learners on the training set, then uses the primary learners to predict the held-out validation set, and then uses the prediction results as the training data for the secondary learner to train the secondary learner to obtain the final ensemble model; while Blending directly uses the primary learners to predict on the held-out validation set and performs weighted averaging on the prediction results to obtain the ensemble output. Taking Stacking as an example, its calculation process is as follows:

[0097] Divide the data into a training set, a validation set, and a test set. Train multiple primary learners on the training set, such as decision trees, support vector machines, etc. Use the trained primary learners to predict the validation set, and form a new data set with the prediction results. Use the new data set as the training set for the secondary learner to train the secondary learner, usually using logistic regression, etc. Predict on the test set using the primary learners, and then input their prediction results into the secondary learner to obtain the final ensemble prediction output.

[0098] Compared with a single model, ensemble learning can significantly improve the prediction accuracy and robustness by combining the prediction results of multiple models. At the same time, different learners may have different emphases on characterizing the data, and their prediction errors may also compensate for each other, thus playing a complementary role.

[0099] Model Optimization

[0100] Introduce an incremental learning mechanism to regularly perform incremental training and optimization on the prediction model using newly collected risk data. Adopt an active learning strategy to identify new samples with high contribution to improving the model performance for annotation and use them for model retraining. At the same time, utilize the idea of transfer learning to transfer the feature representations and model parameters learned by the model on existing engineering projects to new engineering projects to improve the knowledge reuse efficiency.

[0101] Such as Figure 6As shown in the figure, the interface of the risk prediction visualization system developed by the present invention includes functional areas such as a risk index panel, a knowledge graph navigation, a risk situation visualization area, risk warning information, and trend analysis charts:

[0102] IV. Realize risk prediction visualization

[0103] Risk index visualization

[0104] Following the information visualization design guidelines, using charts such as histograms, pie charts, radar charts, and scatter plots to intuitively display the data characteristics of risk assessment indicators. Develop an exploratory visual analysis interface for risk indicators to achieve interactive functions such as multi-dimensional linkage, drilling down, and slicing of indicator data.

[0105] Risk warning visualization

[0106] Comprehensively apply geographic information visualization technology to design and produce an electronic map of risk events. Connect to the project progress management system to obtain real-time data at the construction site and dynamically mark the locations where risk events occur on the electronic map. According to the risk prediction results, automatically push risk warning information and intuitively display the warning locations and warning levels on the electronic map.

[0107] Risk trend visualization

[0108] Develop a spatio-temporal visualization engine for risk prediction results to dynamically present the risk evolution trend of the entire project life cycle. Through the mapping of the time axis and spatial coordinates, depict the time occurrence sequence and spatial distribution law of risk events. Use visualization metaphors such as streamline diagrams and heat maps to represent the risk propagation path and influence range.

[0109] Risk knowledge graph visualization

[0110] Develop a visualization modeling tool for engineering risk knowledge graphs. Adopt the node-edge paradigm to realize the network topology layout of the knowledge graph and provide a human-computer interaction interface to support the editing, importing, and exporting of graph nodes and relationships. Build a knowledge inference engine inside to realize the visual inference analysis of graph knowledge. At the same time, embed a knowledge graph evaluation model to provide a visual evaluation function for graph quality:

[0111] When evaluating the quality of the constructed engineering risk knowledge graph, quantitative indicators such as graph density, centrality, and coverage rate are introduced. Taking graph density as an example, its calculation formula is:

[0112]

[0113] Among them, |V| and |E| respectively represent the number of concept nodes and relationship edges of the knowledge graph. The value range of graph density is [0,1], which represents the ratio of the actual number of existing edges in the graph to the maximum possible number of edges, reflecting the tightness of the knowledge graph. The calculation process is as follows:

[0114] Traverse the knowledge graph to count the total number of concept nodes |V| and the total number of relationship edges |E|. Calculate the maximum number of possible edges for |V| nodes according to the formula, that is, |V| × (|V| - 1). Note that it is assumed here that self-loops and multiple edges are allowed in the graph. If the graph is set not to allow self-loops or multiple edges, the denominator needs to be adjusted accordingly. Divide the actual number of edges |E| by the maximum number of edges to obtain the graph density value.

[0115] The larger the density value, the closer the association between concept nodes in the knowledge graph and the finer the granularity of knowledge organization. However, the density should not be too high, otherwise it will lead to an overly complex graph, which is not conducive to understanding and application. It should be noted that a single evaluation index often cannot comprehensively characterize the quality of the knowledge graph. Therefore, in practice, multiple indicators are usually combined for multi-dimensional evaluation. At the same time, quantitative evaluation indicators also need to be combined with qualitative evaluation methods such as expert review to more comprehensively and objectively judge the effectiveness of the knowledge graph.

[0116] The engineering implementation process of the present invention is as follows:

[0117] In the first step, collect and organize literature materials, network data, and standard specifications in the field of engineering risk management, and organize expert interviews to obtain engineering risk management knowledge; then adopt an information extraction method based on rules and dictionaries to construct a risk ontology model, extract core risk management concepts and relationships; use a knowledge fusion algorithm based on similarity to infer and induce new risk concepts, form a multi-level risk knowledge graph, and store it in a graph database.

[0118] In the second step, make full use of the engineering risk knowledge graph to identify the key risk attributes involved in risk events, factors, and consequences, and invite experts to conduct evaluation and screening to form a risk assessment index set; then use the fuzzy comprehensive evaluation method to quantify qualitative indicators and standardize quantitative indicator data; use the analytic hierarchy process to construct a multi-level index evaluation model, and organize experts to make pairwise judgments and weight calculations on the indicators to form a quantitative risk assessment index system.

[0119] In the third step, collect engineering project risk management data, construct a heterogeneous data collection framework, clean and integrate the original data; then, based on the quantitative risk indicators, extract the time series characteristics of risk events and the statistical characteristics of risk factors to construct a machine learning feature space; on this basis, use the cross-validation method to train machine learning models and deep learning models respectively; furthermore, adopt the Stacking ensemble learning framework to combine multiple models to form an ensemble prediction system, and use active learning and transfer learning methods to optimize the models.

[0120] Fourthly, following the information visualization design guidelines, adopt various chart forms to visually display the data characteristics of risk assessment indicators, and develop an exploratory visual analysis interaction interface; comprehensively utilize geographic information visualization technology to develop an early warning electronic map for risk events to visually display warning information; develop a spatio-temporal visualization engine for risk prediction to depict the spatio-temporal evolution of risk events; develop a visualization modeling tool for engineering risk knowledge graphs to support graph interaction editing and reasoning analysis.

[0121] Fifthly, during the implementation process of the engineering project, continuously and dynamically collect risk event data, regularly use methods such as factor analysis and data mining to evaluate the risk index system, and organize expert diagnosis and review to form optimization and improvement opinions; simultaneously implement model incremental learning, and retrain the prediction model with new data; real-time track the project risk event data, and dynamically update the risk event early warning map and the spatio-temporal visualization picture of risk trends; continuously enrich and improve the engineering risk knowledge graph to improve the quality of the graph.

[0122] This invention covers any substitutions, modifications, equivalent methods, and solutions made within the essence and scope of this invention. For the public to have a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments of this invention. However, those skilled in the art can fully understand this invention even without the description of these details. Additionally, well-known methods, processes, procedures, components, and circuits are not described in detail to avoid unnecessary confusion to the essence of this invention.

[0123] The above are only the preferred embodiments of this invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of this invention, several improvements and refinements can still be made, and these improvements and refinements should also be regarded as the protection scope of this invention.

Claims

1. An engineering risk prediction method based on artificial intelligence, characterized in that, It includes the following steps: Step 1: Construct an engineering risk knowledge graph: Construct a multi-level and structured engineering risk knowledge graph through steps of knowledge acquisition, extraction, fusion, organization, and storage; among them, in the knowledge fusion step, a knowledge fusion algorithm based on semantic similarity is adopted, and by calculating the cosine similarity of the feature vectors of concept nodes, the concept nodes are automatically associated and fused; Step 2: Design a quantitative risk index system: Extract risk assessment indicators from the knowledge graph, use the fuzzy comprehensive evaluation method to quantitatively process qualitative indicators, standardize quantitative indicators, and construct a multi-level index evaluation model; in the calculation of index weights, the analytic hierarchy process (AHP) is used, and by constructing a judgment matrix, calculating the maximum eigenvalue and conducting a consistency test, the weights of each risk indicator are determined; Step 3: Develop an intelligent risk prediction algorithm: Collect engineering risk management data, perform data cleaning and feature engineering, and extract the time series features of risk events; adopt a modeling strategy combining machine learning and deep learning, and train and optimize based on logistic regression, support vector machine, random forest, XGBoost, multi-layer perceptron, and long short-term memory network models to construct a risk prediction model; further adopt the Stacking ensemble learning and Blending ensemble learning frameworks to combine multiple prediction models to improve the prediction accuracy and robustness; Step 4: Realize risk prediction visualization: According to the information visualization design criteria, use any one of the histogram, pie chart, radar chart, and scatter plot to present the characteristics of risk assessment indicators from multiple perspectives respectively; develop a visualization modeling tool for the risk knowledge graph for visual display, editing, reasoning, and analysis of risk knowledge; introduce multi-dimensional evaluation indicators including graph scale, graph density, graph centrality, and graph coverage rate to dynamically evaluate the quality of the risk knowledge graph; Step 5: Dynamic optimization and update: During the entire life cycle of the engineering project, continuously collect risk data and regularly conduct dynamic optimization of the risk index system; identify key indicators through factor analysis and data mining methods, perform incremental learning on the risk prediction model, retrain and optimize the model with new data, update the risk warning information and risk evolution trend in real time, and dynamically present the risk panorama.

2. The engineering risk prediction method based on artificial intelligence according to claim 1, wherein In the knowledge fusion link of step (1), the calculation formula of the knowledge fusion algorithm based on semantic similarity is: Among them, c i and c j represent two concept nodes, v i and v j are the corresponding feature vectors; by calculating the dot product of the two feature vectors divided by the product of their moduli, a normalized similarity value is obtained, and a threshold is set to achieve the automatic fusion of concepts.

3. The engineering risk prediction method based on artificial intelligence according to claim 1, wherein In the calculation of index weights in step (2), the analytic hierarchy process AHP is adopted to calculate the maximum eigenvalue of the judgment matrix and conduct a consistency test, and the calculation formula is: Among them, A is the pairwise comparison judgment matrix, λ max is the maximum eigenvalue, w is the weight vector, n is the number of indicators, CI is the consistency index, RI is the random consistency index, and CR is the consistency ratio; through the consistency test, it is ensured that the judgment matrix has satisfactory consistency.

4. The engineering risk prediction method based on artificial intelligence according to claim 1, wherein In the feature extraction process of step (3), the moving average (MA) and exponentially weighted moving average (EWMA) are used for extracting the time series features of risk events, and the calculation formula is: EWMAt (α) = αx t + (1 - α)EWMAt-1 (α) where x i is the number of risk events at the i-th moment, k is the time window size, and α is the smoothing coefficient; by extracting multi-dimensional time series features, a more robust input is provided for the machine learning model.

5. The engineering risk prediction method based on artificial intelligence according to claim 1, characterized in that In the model training of step (3), the Stacking and Blending ensemble learning strategies are adopted for multi-model fusion prediction output. Among them, the prediction function of the Stacking secondary learner is: The weighted average prediction of the Blending ensemble is: Among them, is the predicted output of the i-th primary learner, f is the secondary learner function, and w i is the integrated weight coefficient and satisfies the normalization condition; the integrated weight coefficient is optimized by the cross-validation method to improve the prediction accuracy and robustness.

6. The engineering risk prediction method based on artificial intelligence according to claim 1, wherein In the visual analysis of step (4), multiple quantitative indicators are used to evaluate the risk knowledge graph. Among them, the calculation formula for the graph density D is: Where |V| and |E| are the number of concept nodes and the number of relationship edges of the graph respectively; quantitative indicators such as density can characterize the features of the knowledge graph from perspectives such as structure and semantics, providing an objective basis to evaluate the quality of the knowledge graph.

7. An engineering risk prediction system according to any one of claims 1-6, characterized in that, Including: A knowledge graph construction module, a risk index quantification module, a risk prediction modeling module, a risk visual analysis module, a dynamic optimization and update module, as well as a knowledge graph database, a risk management database, and a model parameter library.

8. A computer-readable storage medium, on which a computer program is stored, and the program, when executed by a processor, implements the engineering risk prediction method according to any one of claims 1-6.

Images