Underground powerhouse construction risk identification and disposal method, system, equipment and medium

By building a multi-dimensional analysis model and a multi-modal machine learning model, combined with a fuzzy comprehensive evaluation algorithm, efficient identification and accurate emergency decision-making of underground factory construction risks are achieved, the problem of insufficient multi-scale risk analysis in the existing technology is solved, and the real-time response and risk feature extraction capabilities of construction safety management are improved.

CN120450411APending Publication Date: 2025-08-08ECONOMIC & TECH RES INST OF HUBEI ELECTRIC POWER COMPANY SGCC
View PDF 0 Cites 8 Cited by

Patent Information

Application Number
CN202510358566.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-03-25
Publication Date
2025-08-08

AI Technical Summary

Technical Problem

The existing technology lacks multi-scale risk analysis, cannot fully grasp the construction risks of underground factory buildings, has low computing efficiency, is difficult to process large-scale underground engineering monitoring data, and cannot dynamically adjust computing resources according to risk changes.

Method used

Collect multi-source heterogeneous data in real time, build a multi-dimensional analysis model, conduct multi-scale risk analysis, conduct in-depth feature recognition and quantitative analysis through multi-modal machine learning model, divide risk levels with fuzzy comprehensive evaluation algorithm, and generate the optimal emergency response plan.

Benefits of technology

It has achieved efficient identification, quantitative assessment and accurate emergency decision-making of underground factory construction risks, improved the real-time response capabilities of construction safety management, and enhanced the perception ability of abnormal equipment status in complex construction scenarios and comprehensiveness of risk characteristics extraction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120450411A_ABST
    Figure CN120450411A_ABST
Patent Text Reader

Abstract

The invention relates to the field of underground powerhouse construction risk identification, and provides an underground powerhouse construction risk identification and disposal method, system, device and medium, and the method comprises the steps: collecting multi-source heterogeneous data in real time, obtaining historical risk case data, and carrying out the preprocessing to obtain structured time-space correlation data; constructing a multi-dimensional analysis model based on a parallel computing algorithm, and performing multi-scale risk analysis on the structured time-space associated data to obtain multi-level risk feature data; performing risk feature recognition through a multi-modal machine learning model to obtain risk quantitative indexes, and performing recognition based on a fuzzy comprehensive evaluation algorithm to obtain construction risk levels; and matching emergency strategies of construction risk levels, carrying out parameter expansion through a combinatorial optimization algorithm, generating a plurality of candidate disposal schemes, carrying out weight calculation and sorting on the candidate disposal schemes by adopting a multi-criterion evaluation model, and outputting an optimal disposal scheme. According to the invention, efficient identification and accurate emergency decision-making of the construction risk of the underground powerhouse are realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of underground powerhouse construction risk identification, and in particular to a method, system, equipment and medium for identifying and handling underground powerhouse construction risks. Background Art

[0002] Underground powerhouse construction refers to the construction of infrastructure such as large-scale industrial facilities, hydropower plant buildings, and underground transportation hubs within underground spaces. This involves complex excavation, support, and structural construction within a limited and enclosed underground space. The underground powerhouse construction process involves multi-source heterogeneous information, including surrounding rock deformation data, microseismic information, temperature and humidity, hazardous gas concentrations, excavation equipment vibration parameters, support equipment oil pressure data, process completion status, and resource consumption. This multi-source information forms the data foundation for identifying and addressing risks in underground powerhouse construction.

[0003] Existing multi-source information analysis and risk identification methods use multiple sensors to collect data, analyze the sensitivity of monitoring indicators using multivariate statistical methods, perform feature-level fusion based on long-short-term memory networks, and use the DS evidence theory modified with the Chebyshev distance for decision fusion. However, this multi-source information analysis and risk identification method lacks multi-scale risk analysis, cannot fully grasp project risks, has low computational efficiency, cannot dynamically adjust computing resources according to risk changes, and is difficult to process large-scale underground project monitoring data. Summary of the Invention

[0004] In view of this, the present invention provides a method, system, equipment and medium for identifying and handling risks in underground powerhouse construction, which solves the problems of the existing technology lacking multi-scale risk analysis, being unable to fully grasp engineering risks, having low computational efficiency, being unable to dynamically adjust computing resources according to risk changes, and having difficulty processing large-scale underground engineering monitoring data.

[0005] In a first aspect, the present invention provides a method for identifying and handling risks in underground powerhouse construction, comprising:

[0006] Real-time collection of multi-source heterogeneous data during the construction of underground powerhouses, acquisition of historical risk case data for underground powerhouse construction, and pre-processing of the multi-source heterogeneous data and the historical risk case data to obtain structured spatiotemporal correlation data;

[0007] Constructing a multi-dimensional analysis model based on a parallel computing algorithm, and performing multi-scale risk analysis on structured spatiotemporal correlation data according to the multi-dimensional analysis model to obtain multi-level risk feature data;

[0008] Identifying risk characteristics of the multi-level risk feature data using a multimodal machine learning model to obtain risk quantification indicators, and identifying the risk quantification indicators based on a fuzzy comprehensive evaluation algorithm to obtain a construction risk level;

[0009] The emergency strategies corresponding to the construction risk levels are matched from the preset disposal knowledge base, and the matched emergency strategies are parameter-expanded through a combinatorial optimization algorithm to generate multiple candidate disposal solutions. The candidate disposal solutions are weighted and ranked using a multi-criteria evaluation model to output the optimal disposal solution.

[0010] Based on the above technical solution, preferably, the real-time collection of multi-source heterogeneous data during the underground powerhouse construction process, obtaining historical risk case data of the underground powerhouse construction, and pre-processing the multi-source heterogeneous data and the historical risk case data to obtain structured spatiotemporal correlation data include:

[0011] The multi-source heterogeneous data includes geological monitoring data, construction environment data, equipment operation data and project progress data;

[0012] Based on the three-dimensional spatial characteristics of the underground powerhouse construction scene, geological monitoring sensors, construction environment sensors, and equipment operation sensors are deployed in layers according to spatial topological relationships. The project progress data is obtained in real time through the construction progress management system, and the historical risk case data is obtained through the construction safety database.

[0013] Kalman filtering is used to reduce noise on geological monitoring data; the sliding interquartile range method is used to detect outliers in construction environment data, and missing periods are filled through cubic spline interpolation; the missing parameters of equipment operation data under abnormal conditions are reconstructed through a long short-term memory network prediction network to obtain corrected equipment operation data; the project progress data and the historical case data are dimensionally aligned based on a spatiotemporal alignment matrix.

[0014] On the basis of the above technical solution, preferably, the long short-term memory network prediction network adopts a dual-channel attention mechanism, and the calculation formula for correcting the device operation data is:

[0015]

[0016] in, The predicted value of the equipment operating parameters for correcting the equipment operating data; β j is the spatial attention weight of the jth device running sensor; N2 is the number of device running sensors; α m is the temporal attention weight at time m; τ is the length of the time lookback window; is the original value of the device operating parameter of the jth device operating sensor at time tm; softmax(·) is the normalized exponential function; v a is a trainable weight vector; tanh(·) is the hyperbolic tangent activation function; W a is the time-trainable weight matrix; h t-1is the hidden state vector of the long short-term memory network at time t-1; x t is the input feature vector at time t; exp(·) is the exponential function; s j The spatial encoding vector of the sensor running on the jth device; s k is the spatial encoding vector of the sensor running on the kth device; c is the context vector; W s is the spatial trainable weight matrix.

[0017] On the basis of the above technical solution, preferably, the multi-dimensional analysis model is constructed based on the parallel computing algorithm, and multi-scale risk analysis is performed on the structured spatiotemporal correlation data according to the multi-dimensional analysis model to obtain multi-level risk feature data, including:

[0018] The underground powerhouse construction space is divided into multiple computational blocks according to engineering geological conditions and construction procedures, and independent parallel computing nodes are assigned to each block. A multidimensional analysis model is constructed, which includes a global-local coupling layer, a multi-scale feature extraction layer, and a risk association reasoning layer. The global-local coupling layer uses tensor decomposition to correlate and map global construction parameters with local monitoring data to establish a cross-scale data correlation matrix.

[0019] The multi-scale feature extraction layer extracts multi-frequency features from time series data using wavelet transform and empirical mode decomposition, and extracts topological features from spatial data using nonlinear manifold dimensionality reduction. The risk association inference layer constructs a construction risk impact propagation model based on a graph neural network, which is used to identify the cascading effects between different risk factors. Based on a multi-dimensional analysis model, multi-scale risk features are extracted from structured spatiotemporal correlation data according to the macro-project layer, the meso-regional layer, and the micro-perspective layer.

[0020] At each scale level, a time series anomaly detection model, a spatial clustering analysis model and a spatiotemporal pattern evolution prediction model are constructed respectively. The model outputs of the time series anomaly detection model, the spatial clustering analysis model and the spatiotemporal pattern evolution prediction model are integrated through an adaptive feature fusion algorithm to obtain multi-level risk feature data.

[0021] On the basis of the above technical solution, preferably, the calculation formula of the multi-level risk characteristic data is:

[0022]

[0023] Among them, Risk int It is multi-level risk characteristic data; M, R, and P are macro-scale, meso-scale, and micro-scale respectively; is the scale b in time Dynamic weight function of N bis the number of features at scale b; is the importance coefficient of the cth feature in scale b; is the measured value of the cth feature in scale b; is the historical mean of the cth feature in scale b; is the historical standard deviation of the cth feature in scale b; Φ b (·) is the nonlinear mapping function of scale b.

[0024] Based on the above technical solution, preferably, the multi-level risk feature data is subjected to risk feature identification by a multimodal machine learning model to obtain a risk quantification index, and the risk quantification index is identified based on a fuzzy comprehensive evaluation algorithm to obtain a construction risk level, including:

[0025] Constructing a multimodal machine learning model, the multimodal machine learning model including a feature extraction network, a temporal feature learning network, and a multimodal feature fusion network, and performing deep feature extraction and correlation analysis on the multi-level risk feature data through the multimodal machine learning model to obtain risk quantification indicators;

[0026] A risk factor weight system based on the hierarchical analysis method is established, and a fuzzy comprehensive evaluation matrix is constructed in combination with the fuzzy membership function. The comprehensive evaluation results of the risk quantification indicators are synthesized through fuzzy operations. The comprehensive evaluation results are mapped to construction risk levels according to the preset threshold intervals. The construction risk levels include negligible, low risk, medium risk, high risk and extremely high risk.

[0027] Based on the above technical solution, preferably, the emergency strategies corresponding to the construction risk levels are matched from a preset disposal knowledge base, parameters of the matched emergency strategies are expanded by a combinatorial optimization algorithm to generate multiple candidate disposal solutions, and a multi-criteria evaluation model is used to calculate and rank the weights of the candidate disposal solutions to output the optimal disposal solution, including:

[0028] Matching the corresponding emergency strategy according to the construction risk level from the preset disposal knowledge base to form a preliminary disposal plan;

[0029] Using a combinatorial optimization algorithm to expand and optimize multiple parameters of the preliminary treatment plan to generate multiple candidate treatment plans;

[0030] Based on the multi-criteria evaluation model, the candidate disposal solutions are weighted and ranked, and the optimal disposal solution is output.

[0031] In a second aspect, the present invention further provides a system for identifying and handling risks in underground powerhouse construction, which is applied to the above-mentioned method, and comprises:

[0032] A data acquisition and processing module is used to collect multi-source heterogeneous data during the construction of the underground powerhouse in real time, obtain historical risk case data of the underground powerhouse construction, and pre-process the multi-source heterogeneous data and the historical risk case data to obtain structured spatiotemporal correlation data;

[0033] An analysis model building module is used to build a multi-dimensional analysis model based on a parallel computing algorithm, and perform multi-scale risk analysis on structured spatiotemporal correlation data according to the multi-dimensional analysis model to obtain multi-level risk feature data;

[0034] A risk level identification module is used to identify risk characteristics of the multi-level risk feature data using a multimodal machine learning model to obtain risk quantification indicators, and to identify the risk quantification indicators based on a fuzzy comprehensive evaluation algorithm to obtain a construction risk level;

[0035] The disposal scheme matching module is used to match the emergency strategy corresponding to the construction risk level from the preset disposal knowledge base, expand the parameters of the matched emergency strategy through the combination optimization algorithm, generate multiple candidate disposal schemes, use the multi-criteria evaluation model to calculate the weights and sort the candidate disposal schemes, and output the optimal disposal scheme.

[0036] In a third aspect, the present invention further provides an underground powerhouse construction risk identification and disposal device, comprising a memory and a processor;

[0037] The memory is configured to store computer program code and transmit the computer program code to the processor;

[0038] The processor is configured to execute the method according to the instructions in the computer program code.

[0039] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, and the computer program implements the above-described method when executed by a processor.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. The present invention provides a method, system, equipment and medium for identifying and handling risks in underground powerhouse construction. By real-time acquisition and preprocessing of multi-source heterogeneous data during the construction of underground powerhouses, a multi-dimensional analysis model is constructed. By combining parallel computing and multi-scale risk analysis methods, multi-level risk characteristics in the construction process are comprehensively extracted. A multimodal machine learning model is used for deep feature recognition and quantitative analysis. The fuzzy comprehensive evaluation algorithm is used to accurately divide the construction risk levels. Based on a preset handling knowledge base, a combination optimization algorithm and a multi-criteria evaluation model are used to generate and screen the optimal emergency handling plan, thereby achieving efficient identification, quantitative evaluation and accurate emergency decision-making of underground powerhouse construction risks, and improving the real-time response capability of construction safety management.

[0042] 2. In the method, system, equipment and medium for identifying and handling risks in underground plant construction of the present invention, by introducing a long short-term memory network prediction network with a dual-channel attention mechanism, the equipment operation data can be accurately reconstructed under abnormal working conditions. The spatial attention mechanism dynamically evaluates the spatial correlation of different sensor nodes, and the temporal attention mechanism captures the characteristic changes at key time points. By combining sensor topology coding and historical state memory, the spatiotemporal feature information is effectively integrated, the prediction accuracy of missing values of equipment operation parameters is improved, and the perception ability of abnormal equipment status in complex construction scenarios is enhanced.

[0043] 3. In the underground powerhouse construction risk identification and disposal method, system, equipment and medium of the present invention, risk characteristics at the macro, meso and micro scales are quantitatively expressed through multi-level risk feature data, a dynamic weight function is introduced to adaptively adjust the contribution of different scales, the feature importance coefficient is used to distinguish the degree of influence of each feature, and feature standardization is performed in combination with the historical mean and standard deviation. The nonlinear change law of risk characteristics is captured through a nonlinear mapping function, and a numerical representation that comprehensively reflects the multi-scale risk characteristics is obtained, thereby improving the comprehensiveness of risk feature extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 This is a flow chart of a method for identifying and handling risks in underground powerhouse construction provided by the present invention.

[0046] Figure 2 This is a structural block diagram of an underground powerhouse construction risk identification and disposal system provided by the present invention.

[0047] Figure 3 This is a structural block diagram of an underground powerhouse construction risk identification and disposal device provided by the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0049] See also Figure 1 The present invention provides a method for identifying and handling risks in underground powerhouse construction, comprising:

[0050] S1. Real-time collection of multi-source heterogeneous data during the construction of underground powerhouses, acquisition of historical risk case data for underground powerhouse construction, and pre-processing of the multi-source heterogeneous data and the historical risk case data to obtain structured spatiotemporal correlation data.

[0051] S2. Construct a multidimensional analysis model based on a parallel computing algorithm, and perform multiscale risk analysis on the structured spatiotemporal correlation data according to the multidimensional analysis model to obtain multi-level risk characteristic data; the multiscale risk analysis includes multi-level risk identification based on the overall project construction parameters, regional correlation data, and point characteristic parameters.

[0052] S3. Identify the risk characteristics of the multi-level risk characteristic data through a multimodal machine learning model to obtain risk quantification indicators, identify the risk quantification indicators based on a fuzzy comprehensive evaluation algorithm, and obtain the construction risk level.

[0053] S4. Match the emergency strategy corresponding to the construction risk level from the preset disposal knowledge base, expand the parameters of the matched emergency strategy through the combinatorial optimization algorithm, generate multiple candidate disposal plans, use a multi-criteria evaluation model to calculate and rank the weights of the candidate disposal plans, and output the optimal disposal plan; the disposal knowledge base contains the disposal rules and multi-level emergency strategies of historical risk cases.

[0054] Specifically, this embodiment constructs a multi-dimensional analysis model by real-time collection and pre-processing of multi-source heterogeneous data during the construction of underground power plants. It combines parallel computing and multi-scale risk analysis methods to comprehensively extract multi-level risk characteristics in the construction process, and uses multimodal machine learning models for deep feature recognition and quantitative analysis. It also accurately divides construction risk levels through a fuzzy comprehensive evaluation algorithm. Based on a preset disposal knowledge base, it uses a combination optimization algorithm and a multi-criteria evaluation model to generate and screen the optimal emergency disposal plan, thereby achieving efficient identification, quantitative evaluation and accurate emergency decision-making of underground power plant construction risks, and improving the real-time response capability of construction safety management.

[0055] Furthermore, the multi-source heterogeneous data collected in real time during the underground powerhouse construction process is used to obtain historical risk case data of the underground powerhouse construction, and the multi-source heterogeneous data and the historical risk case data are pre-processed to obtain structured spatiotemporal correlation data, including:

[0056] The multi-source heterogeneous data includes geological monitoring data, construction environment data, equipment operation data and project progress data;

[0057] Based on the three-dimensional spatial characteristics of the underground powerhouse construction scene, geological monitoring sensors (including fiber Bragg grating strain sensors and microseismic monitoring arrays), construction environment sensors (including temperature and humidity sensors and harmful gas detectors), and equipment operation sensors (including vibration sensors and oil pressure monitors) are hierarchically deployed according to spatial topological relationships. The project progress data is obtained in real time through the construction progress management system, and the historical risk case data is obtained through the construction safety database.

[0058] In a specific embodiment, the layered deployment includes:

[0059] A dual-redundant sensor ring network was constructed in the area sensitive to surrounding rock deformation. The primary ring network used the ZigBee protocol to transmit geological monitoring data, and the backup ring network used the LoRa protocol to transmit construction environment data. The two networks achieved data synchronization through the following spatiotemporal coupling relationship. The synchronization validity determination formula is:

[0060]

[0061] Where ΔT sync is the time synchronization deviation threshold of the two networks; N1 is the number of geological monitoring sensor groups; The timestamp of the address data transmitted by the i-th group of geological monitoring sensors via the ZigBee protocol; is the timestamp of the environmental data transmitted by the i-th group of geological monitoring sensors through the LoRa protocol; α1 is the signal attenuation coefficient; d i is the physical deployment spacing between the i-th group of geological monitoring sensors;

[0062] When ΔT sync When the value is greater than 50ms, synchronization is determined to be valid.

[0063] Specifically, this embodiment adopts a hierarchical deployment strategy based on three-dimensional spatial features, so that the sensor network layout accurately matches the complex spatial structure of the underground powerhouse, improving the monitoring coverage and representativeness; a dual-redundant sensor ring network is constructed in key rock deformation-sensitive areas. The main ring network is based on the ZigBee protocol and the backup ring network is based on the LoRa protocol. Through the time-space coupling synchronization judgment mechanism, redundant backup and high reliability of data acquisition are achieved, and stable data transmission can be maintained even in underground environments with restricted communication.

[0064] In a specific embodiment, the preprocessing includes data cleaning, standardization, spatiotemporal label alignment and fusion processing, specifically including:

[0065] Perform Kalman filtering noise reduction on geological monitoring data to eliminate pulse interference caused by equipment start-up and shutdown;

[0066] The sliding interquartile range method was used to detect outliers in the construction environment data, and missing periods were filled by cubic spline interpolation.

[0067] Through the long short-term memory network prediction network, the missing parameters of the equipment operation data under abnormal conditions are reconstructed to obtain the corrected equipment operation data;

[0068] The project progress data and the historical case data are dimensionally aligned based on a spatiotemporal alignment matrix.

[0069] Specifically, this embodiment employs differentiated preprocessing strategies for different data types: Kalman filtering technology is used for geological monitoring data to effectively eliminate pulse interference caused by equipment startup and shutdown; the sliding interquartile range method is used to accurately detect outliers in construction environment data, and cubic spline interpolation technology is used to fill missing data to maintain data continuity; and the long short-term memory network prediction model is used to reconstruct missing parameters under abnormal conditions for equipment operation data, enhancing data integrity and reliability. A spatiotemporal alignment matrix is used to unify the dimensions of multi-source heterogeneous data and historical risk case data, solving the problem of difficulty in integrating and analyzing data from different sources, formats, and sampling frequencies.

[0070] Furthermore, the long short-term memory network prediction network adopts a dual-channel attention mechanism, and the calculation formula for correcting the device operation data is:

[0071]

[0072]

[0073] in, The predicted value of the equipment operating parameters for correcting the equipment operating data; β j is the spatial attention weight of the jth device running sensor; N2 is the number of device running sensors; α m is the temporal attention weight at time m; τ is the length of the time lookback window; is the original value of the device operating parameter of the jth device operating sensor at time tm; softmax(·) is the normalized exponential function; v a is a trainable weight vector; tanh(·) is the hyperbolic tangent activation function; W a is the time-trainable weight matrix; h t-1 is the hidden state vector of the long short-term memory network at time t-1; x t is the input feature vector at time t; exp(·) is the exponential function; s j The spatial encoding vector of the sensor running on the jth device; s k is the spatial encoding vector of the sensor running on the kth device; c is the context vector; W s is the spatial trainable weight matrix.

[0074] Specifically, this embodiment introduces a long short-term memory network prediction network with a dual-channel attention mechanism to achieve accurate parameter reconstruction of equipment operation data under abnormal working conditions. The spatial attention mechanism dynamically evaluates the spatial correlation of different sensor nodes, and the temporal attention mechanism captures the feature changes at key time points. It combines sensor topology encoding and historical state memory to effectively integrate spatiotemporal feature information, improve the prediction accuracy of missing values of equipment operation parameters, and enhance the ability to perceive abnormal equipment states in complex construction scenarios.

[0075] Furthermore, the multi-dimensional analysis model is constructed based on the parallel computing algorithm, and multi-scale risk analysis is performed on the structured spatiotemporal correlation data according to the multi-dimensional analysis model to obtain multi-level risk feature data, including:

[0076] A divide-and-conquer strategy is used to divide the underground powerhouse construction space into multiple computing blocks according to engineering geological conditions and construction procedures, and independent parallel computing nodes are allocated to each block.

[0077] Constructing a multi-dimensional analysis model, the multi-dimensional analysis model including a global-local coupling layer, a multi-scale feature extraction layer, and a risk association reasoning layer;

[0078] The global-local coupling layer uses tensor decomposition to correlate and map global construction parameters with local monitoring data to establish a cross-scale data correlation matrix;

[0079] The multi-scale feature extraction layer extracts multi-frequency features by using wavelet transform and empirical mode decomposition on time series data, and extracts topological features by using nonlinear manifold dimensionality reduction method on spatial data;

[0080] The risk association reasoning layer constructs a construction risk impact propagation model based on a graph neural network, and the construction risk impact propagation model is used to identify the cascading effects between different risk factors.

[0081] In a specific embodiment, parallel computing is achieved by regional decoupling and boundary coordination. The speedup ratio of regional decoupling and parallel computing is calculated as follows:

[0082]

[0083] Among them, S par is the parallel computing speedup ratio; T ser is the total time of serial calculation; T par is the total parallel computing time; N3 is the total number of computing blocks; T n is the serial computing time of the nth computing block; T comm is the communication time between calculation blocks; α2 is the communication overhead coefficient.

[0084] Intra-region computing: Each computing node independently performs feature extraction and risk calculation tasks for its region;

[0085] Boundary collaboration: Adjacent regions exchange information between regions by sharing boundary sensor data;

[0086] Global integration: A hierarchical aggregation strategy is used to integrate the calculation results of each region into a unified engineering risk representation;

[0087] A regional weight update mechanism with adaptive learning rate is applied to dynamically adjust the computing resource allocation ratio according to the degree of risk change, so that high-risk areas obtain more computing resources.

[0088] Based on a multi-dimensional analysis model, multi-scale risk features are extracted from structured spatiotemporal correlation data at the macro-project level (overall project level), the meso-regional level (construction zone level), and the micro-point level (sensor node level).

[0089] At each scale level, a time series anomaly detection model, a spatial clustering analysis model and a spatiotemporal pattern evolution prediction model are constructed respectively. The model outputs of the time series anomaly detection model, the spatial clustering analysis model and the spatiotemporal pattern evolution prediction model are integrated through an adaptive feature fusion algorithm to obtain multi-level risk feature data.

[0090] In a specific embodiment, a risk tracing network is constructed based on the propagation relationship between risk features at different levels, and bidirectional mapping and verification of risk features at different scales are performed;

[0091] Multiscale risk analysis includes:

[0092] For macro-engineering risk analysis (project overall level):

[0093] Build a global construction status indicator system, including key indicators such as project progress deviation rate, resource consumption index, and accumulated safety hazards;

[0094] Build a project-level risk baseline based on a historical project experience database and apply time series decomposition techniques to identify abnormal changes in overall project trends and cycle patterns;

[0095] Identify systematic deviations in key project parameters by combining engineering control charts with statistical process control methods;

[0096] For meso-regional risk analysis (construction zone level):

[0097] Establish a multi-factor correlation model of geology, environment, equipment and progress for each construction zone;

[0098] Apply spatial autocorrelation analysis to identify the spatial clustering patterns of risk factors and locate high-risk construction areas;

[0099] Construct risk propagation path maps for different construction processes based on regional operation characteristics to simulate the cascading effects of construction disturbances;

[0100] For micro-point level risk analysis (sensor node level):

[0101] Apply change point detection algorithm to single-point sensor data to identify data mutation points;

[0102] Construct a phase synchronization analysis model for sensor networks to capture abnormal coordinated changes among multiple sensors;

[0103] Establish a point-level risk hotspot detection mechanism based on the sensor topology;

[0104] The hierarchical risk profile includes:

[0105] Design an adaptive weight allocation mechanism to dynamically adjust the contribution weight of each scale based on the rate of change of risk characteristics at different scales;

[0106] The reliability of risk identification is improved through the mutual verification mechanism of risk characteristics between scales. When micro-anomalies appear simultaneously at multiple points and form regional clusters, regional-level risk confirmation is triggered.

[0107] Apply risk feature tracing algorithms to build multi-level risk feature mapping relationships, and realize traceability analysis from superficial risks to root risks.

[0108] Specifically, this embodiment adopts a divide-and-conquer strategy and regional decoupling mechanism to divide the complex underground plant space into multiple computing blocks according to geological conditions and construction procedures. Each block is assigned an independent computing node, which reduces the complexity of the problem while realizing computational parallelism, significantly improving the processing efficiency of large-scale monitoring data.

[0109] This embodiment constructs a three-layer multi-dimensional analysis model: the global-local coupling layer establishes a cross-scale data association matrix through the tensor decomposition method, solving the mapping problem between global construction parameters and local monitoring data; the multi-scale feature extraction layer combines wavelet transform, empirical mode decomposition and nonlinear manifold dimensionality reduction methods to extract the multi-frequency characteristics of time series data and the topological characteristics of spatial data; the risk association inference layer constructs a construction risk impact propagation model based on graph neural network, which depicts the cascade effect between different risk factors and avoids the limitations of single-dimensional analysis.

[0110] This embodiment achieves precise allocation of computing resources through a regional weight update mechanism with adaptive learning rate, enabling high-risk areas to obtain more computing resources, thereby optimizing resource utilization efficiency while ensuring analysis accuracy in key areas.

[0111] This embodiment establishes a three-level risk analysis framework: macro (overall project level), meso (construction zone level), and micro (sensor node level). The macro level identifies overall project anomaly trends through an engineering status indicator system and time series decomposition. The meso level establishes a multi-factor correlation model between geology, environment, equipment, and progress to identify spatial clustering of risks. The micro level applies change point detection and phase synchronization analysis to accurately capture sensor data fluctuations. This multi-scale risk analysis framework enables a progressive identification of risks from local to global levels, avoiding the coverage gaps of single-scale analysis.

[0112] Through the bidirectional mapping verification and traceability network of risk characteristics, the correlation analysis capability between risk characteristics of different scales is established, which realizes the traceability analysis from superficial risks to root risks and improves the accuracy of risk identification.

[0113] Furthermore, the calculation formula of the multi-level risk characteristic data is:

[0114]

[0115] Among them, Risk int It is multi-level risk characteristic data; M, R, and P are macro-scale, meso-scale, and micro-scale respectively; is the scale b in time Dynamic weight function of N b is the number of features at scale b; is the importance coefficient of the cth feature in scale b; is the measured value of the cth feature in scale b; is the historical mean of the cth feature in scale b; is the historical standard deviation of the cth feature in scale b; Φ b (·) is the nonlinear mapping function of scale b.

[0116] Specifically, this embodiment uses multi-level risk feature data to quantitatively express risk features at the macro, meso, and micro scales, introduces a dynamic weight function to adaptively adjust the contributions of different scales, uses feature importance coefficients to distinguish the degree of influence of each feature, combines historical mean and standard deviation to perform feature standardization, and captures the nonlinear change law of risk features through nonlinear mapping functions, obtaining a numerical representation that comprehensively reflects multi-scale risk features and improving the comprehensiveness of risk feature extraction.

[0117] Furthermore, the multi-level risk feature data is subjected to risk feature identification by a multimodal machine learning model to obtain a risk quantification index, and the risk quantification index is identified based on a fuzzy comprehensive evaluation algorithm to obtain a construction risk level, including:

[0118] A multimodal machine learning model is constructed, which includes a feature extraction network, a time series feature learning network and a multimodal feature fusion network. The multimodal machine learning model is used to perform deep feature extraction and correlation analysis on the multi-level risk feature data to obtain risk quantification indicators; the risk quantification indicators include geological stability index, degree of environmental anomaly, possibility of equipment failure and schedule deviation rate.

[0119] In a specific embodiment, the network components of the multimodal machine learning model include:

[0120] For the feature extraction network part:

[0121] Apply convolutional neural networks to geological monitoring data to extract spatial distribution characteristics and identify surrounding rock deformation trends and rock mass structural anomalies;

[0122] Applying long short-term memory networks to construction environment data to capture the temporal variation characteristics of environmental parameters such as temperature, humidity, and harmful gas concentrations;

[0123] Apply a one-dimensional convolutional network to the equipment operation data to extract vibration spectrum features, and combine it with an autoencoder to detect abnormal equipment operation status;

[0124] Apply gated recurrent unit networks to project progress data to model critical path execution and identify schedule delays and resource allocation anomalies.

[0125] For the temporal feature learning network part:

[0126] A recurrent neural network with a bidirectional attention mechanism is used to model the temporal dependency of each modality feature sequence;

[0127] Capture the evolution of risk patterns at different time scales through temporal convolutional networks;

[0128] Apply the multi-head self-attention mechanism to identify changes in risk characteristics at key time points.

[0129] For the multimodal feature fusion network part:

[0130] Design a cross-modal adaptive weight fusion layer to dynamically adjust the fusion weights based on the reliability of different modal features;

[0131] The residual connection structure is used to retain the original feature information and alleviate the gradient vanishing problem in deep network training;

[0132] Through the multi-task learning framework, multiple risk quantitative indicators are output simultaneously to achieve multi-dimensional representation of risks.

[0133] The calculation steps of risk quantification indicators include:

[0134] Construct a risk assessment function based on the fused feature vector and map it to the standardized risk quantification indicator space;

[0135] Design a dual output mechanism of certainty and uncertainty, and provide risk prediction values and their confidence intervals at the same time;

[0136] Calibrate risk quantification indicators through comparative analysis of historical cases to ensure the actual engineering interpretability of the indicators.

[0137] A risk factor weight system based on the hierarchical analysis method is established, and a fuzzy comprehensive evaluation matrix is constructed in combination with the fuzzy membership function. The comprehensive evaluation results of the risk quantification indicators are synthesized through fuzzy operations. The comprehensive evaluation results are mapped to construction risk levels according to the preset threshold intervals. The construction risk levels include negligible, low risk, medium risk, high risk and extremely high risk.

[0138] In a specific embodiment, a risk assessment factor set and a risk level assessment set are established. The risk assessment factor set includes quantitative indicators of geological stability index, degree of environmental anomaly, possibility of equipment failure and schedule deviation rate; the risk level assessment set includes five levels: negligible, low risk, medium risk, high risk and extremely high risk.

[0139] The analytic hierarchy process is applied to determine the weight of each risk factor by combining expert scoring with historical case verification; a time-varying weight adjustment mechanism is introduced to dynamically adjust the importance of risk factors according to the different characteristics of the project stage; and the rationality of weight distribution is ensured through consistency testing.

[0140] A fuzzy membership function is constructed. For each risk quantification indicator, membership functions of different forms such as semi-trapezoidal, triangular and Gaussian are designed based on engineering experience and statistical analysis results. An adaptive boundary determination method is used to dynamically adjust the membership function parameters according to the distribution characteristics of historical data. Partially overlapping fuzzy intervals are set for different risk levels to reflect the continuity characteristics of risk assessment.

[0141] A fuzzy comprehensive evaluation matrix is established, and each risk quantitative indicator is mapped to the membership value of each risk level through the membership function to form a single factor evaluation vector; all single factor evaluation vectors are combined to construct a complete fuzzy relationship matrix.

[0142] Perform fuzzy synthesis operations, adopt weighted average fuzzy synthesis operators, combine risk factor weight vectors and fuzzy relationship matrices to calculate comprehensive evaluation vectors, determine the final risk level through maximum membership method or weighted average method, design decision rules at the risk level boundary, and improve the stability and reliability of risk level determination.

[0143] Specifically, this embodiment adopts a hierarchical multimodal machine learning architecture to achieve accurate feature extraction of different types of monitoring data: the spatial distribution characteristics and rock structure anomalies of geological monitoring data are captured through convolutional neural networks; the temporal change patterns of environmental parameters are extracted using long short-term memory networks; one-dimensional convolutional networks and autoencoders are combined to identify abnormal characteristics of equipment vibration spectra; and a gated recurrent unit network is used to analyze the execution status of critical paths of project progress.

[0144] This embodiment realizes the dynamic capture of multi-scale risk patterns through a temporal feature learning network, strengthens the dependency modeling of the time dimension by using a recurrent neural network with a bidirectional attention mechanism, effectively identifies the evolution patterns of risks at different time scales using a temporal convolutional network, and accurately locates the key time points of risk changes based on a multi-head self-attention mechanism. This multi-angle temporal learning greatly enhances the perception of the temporal evolution characteristics of risks.

[0145] The adaptive weight fusion layer of this embodiment dynamically adjusts the fusion weights according to the reliability of data of different modalities, avoiding the negative impact of low-quality data; the residual connection structure is used to effectively retain the original feature information, solving the problem of vanishing gradients in deep networks; and a multi-dimensional quantitative representation of risk is achieved through a multi-task learning framework.

[0146] The calculation process of risk quantification indicators introduces a dual output mechanism of certainty and uncertainty, which not only provides risk prediction values, but also gives confidence intervals, thereby improving the reliability of risk assessment; through comparative analysis of historical cases, the risk indicators are calibrated, thereby enhancing the engineering practicality and interpretability of the risk quantification indicators.

[0147] An improved hierarchical analysis method is used to make the risk factor weights more objective; a time-varying weight adjustment mechanism is used to adapt to the risk characteristics of different construction stages; various forms of fuzzy membership functions, including semi-trapezoidal, triangular, and Gaussian types, are adopted to more accurately describe the fuzzy relationship between risk quantification indicators and risk levels based on the adaptive boundary method; and the stability of risk level determination is improved through a weighted average fuzzy synthesis operator and specific decision rules.

[0148] Furthermore, the emergency strategies corresponding to the construction risk levels are matched from the preset disposal knowledge base, parameters of the matched emergency strategies are expanded by a combinatorial optimization algorithm to generate multiple candidate disposal solutions, and the candidate disposal solutions are weighted and ranked by a multi-criteria evaluation model to output the optimal disposal solution, including:

[0149] The corresponding emergency strategy is matched according to the construction risk level from the preset disposal knowledge base to form a preliminary disposal plan.

[0150] In a specific embodiment, an emergency strategy matching rule indexed by construction risk level is constructed, and the multi-level emergency strategies in the preset disposal knowledge base are hierarchically managed according to the construction risk level, with each risk level corresponding to a group of emergency strategies; the applicable conditions of the emergency strategies are semantically annotated, including construction environment parameters, project stages, and resource constraints.

[0151] Match emergency strategies based on semantic retrieval algorithms, and retrieve corresponding emergency strategy sets based on risk level classification results. Use construction environment parameters, risk indicator characteristic data, and resource constraints as search keywords to match applicable strategy subsets. If no fully matching emergency strategy exists, use a similarity matching algorithm to retrieve emergency strategies from historical similar risk cases and adjust and optimize them based on current construction conditions.

[0152] Filter a subset of strategies that cover the risk indicator characteristics from multiple matching emergency strategies; sort them by priority and form a preliminary disposal plan.

[0153] A combinatorial optimization algorithm is used to expand and optimize multiple parameters of the preliminary treatment plan to generate multiple candidate treatment plans.

[0154] In a specific embodiment, the parameter expansion range of the emergency strategy plan is defined, and the key parameters therein, such as the treatment steps, resource allocation, and execution time limit, are extracted based on the design objectives and applicable scope of the preliminary treatment plan; based on the uncertainty analysis of construction conditions, upper and lower limits and change constraints are defined for each key parameter.

[0155] A combinatorial optimization algorithm is used to generate candidate solutions and construct a parameter combination optimization problem model. The objective functions include minimizing processing time, maximizing resource utilization efficiency, and improving construction safety. A multi-objective genetic algorithm or particle swarm optimization algorithm is used to evolve and solve the parameter combination. During the calculation process, the fitness of the solution is dynamically evaluated, and parameter combinations that do not meet the constraints are corrected.

[0156] Based on historical risk scenario data, the generated candidate solutions are simulated and verified to screen out solutions that meet the expected goals; the candidate solutions are classified according to strategy coverage, feasibility and profitability.

[0157] Based on the multi-criteria evaluation model, the candidate disposal solutions are weighted and ranked, and the optimal disposal solution is output.

[0158] In a specific embodiment, a multi-criteria evaluation model is constructed, and the evaluation criteria include the cost, execution efficiency, resource consumption and risk mitigation effect of the disposal plan; specific evaluation indicators and weight values are set for each evaluation criterion, and the weight values are determined by statistical learning methods.

[0159] Perform weight calculation on candidate solutions and calculate the contribution score of each candidate solution to each evaluation criterion based on the hierarchical analysis method or set pair analysis method; take a weighted average of the scores of each criterion to calculate the comprehensive score of the candidate solution; introduce uncertainty analysis in the score calculation to evaluate the scoring stability of the solution under different weight settings.

[0160] Sort and output the optimal solution, sort the candidate solutions from high to low according to their comprehensive scores, and give priority to the solution with the highest score as the optimal solution; if multiple solutions have similar scores, further screening will be carried out based on additional criteria such as simplicity of implementation conditions and resource reuse rate, and the optimal disposal solution will be output, and a quantitative prediction report of the risk mitigation effect will be provided to the end user.

[0161] Specifically, this embodiment constructs emergency strategy matching rules indexed by construction risk level, hierarchically manages the multi-level emergency strategies in the preset disposal knowledge base according to risk level, ensuring that different risk levels can be quickly matched to applicable emergency strategy sets. Semantic annotation methods are used to annotate the applicable conditions of emergency strategies, such as construction environment parameters, project stages, and resource constraints, and semantic retrieval algorithms are used for strategy matching, thereby improving the accuracy of strategy matching. In the absence of a fully matching strategy, a similarity matching algorithm is used to retrieve emergency strategies from historical similar risk cases, and adjustments and optimizations are made based on current construction conditions, solving the problem of insufficient emergency strategies in complex construction scenarios.

[0162] This embodiment uses a combinatorial optimization algorithm to expand the parameters of the initially matched emergency strategies and generate multiple candidate solutions. This parameter expansion is based on an uncertainty analysis of construction conditions and defines upper and lower limits and variation constraints for key parameters, such as response steps, resource allocation, and execution time limits, to ensure that the candidate solutions cover the needs of diverse construction scenarios.

[0163] A multi-objective optimization algorithm is used to optimize the candidate solutions, such as a multi-objective genetic algorithm or a particle swarm optimization algorithm. The objective functions include minimizing processing time, maximizing resource utilization efficiency, and improving construction safety. The generated candidate solutions have higher comprehensive benefits under the multi-objective trade-off.

[0164] During the optimization process, the fitness of candidate solutions is dynamically evaluated, and parameter combinations that do not meet the constraints are corrected to ensure that the generated candidate solutions are practically feasible.

[0165] Based on historical risk scenario data, generated candidate solutions are simulated and verified to select those that meet the expected objectives and ensure their applicability in actual construction scenarios. Candidate solutions are categorized based on strategic coverage, feasibility, and profitability, providing a foundation for multi-criteria evaluation.

[0166] This embodiment constructs a multi-criteria evaluation model to conduct a comprehensive evaluation of candidate solutions from multiple dimensions such as cost, execution efficiency, resource consumption and risk mitigation effect. The contribution score of the candidate solution on each evaluation criterion is calculated by the hierarchical analysis method or the set pair analysis method, and the weighted average is performed in combination with the weight value to calculate the comprehensive score to ensure the objectivity of the evaluation process. Uncertainty analysis is introduced in the evaluation to evaluate the stability of the score of the candidate solution under different weight settings, thereby enhancing the robustness of the evaluation results. Sort by comprehensive score from high to low, and give priority to the solution with the highest score as the optimal disposal solution; when multiple solutions have similar scores, further screening is carried out in combination with additional criteria, such as simplicity of implementation conditions and resource reuse rate, to ensure the preference of the final solution.

[0167] See also Figure 2 The present invention also provides a system for identifying and handling risks in underground powerhouse construction, which is applied to the above-mentioned method for identifying and handling risks in underground powerhouse construction. The system includes:

[0168] A data acquisition and processing module is used to collect multi-source heterogeneous data during the construction of the underground powerhouse in real time, obtain historical risk case data of the underground powerhouse construction, and pre-process the multi-source heterogeneous data and the historical risk case data to obtain structured spatiotemporal correlation data;

[0169] An analysis model building module is used to build a multi-dimensional analysis model based on a parallel computing algorithm, and perform multi-scale risk analysis on structured spatiotemporal correlation data according to the multi-dimensional analysis model to obtain multi-level risk feature data;

[0170] A risk level identification module is used to identify risk characteristics of the multi-level risk feature data using a multimodal machine learning model to obtain risk quantification indicators, and to identify the risk quantification indicators based on a fuzzy comprehensive evaluation algorithm to obtain a construction risk level;

[0171] The disposal scheme matching module is used to match the emergency strategy corresponding to the construction risk level from the preset disposal knowledge base, expand the parameters of the matched emergency strategy through the combination optimization algorithm, generate multiple candidate disposal schemes, use the multi-criteria evaluation model to calculate the weights and sort the candidate disposal schemes, and output the optimal disposal scheme.

[0172] Specifically, a risk identification and disposal system for underground plant construction in this embodiment realizes efficient collection and preprocessing of multi-source heterogeneous data through a data acquisition and processing module, solving the problems of data noise, anomalies and missing data; the analysis model construction module realizes accurate risk identification from macro, meso to micro levels through parallel computing and multi-dimensional analysis; the risk level identification module uses multimodal machine learning and fuzzy comprehensive evaluation algorithms to realize accurate quantification and classification of construction risks; the disposal plan matching module outputs the optimal disposal plan through knowledge base matching, combination optimization and multi-criteria evaluation, thereby improving the accuracy of construction risk identification.

[0173] See also Figure 3 ,The present invention also provides an underground powerhouse construction risk identification and disposal device, including a memory and a processor;

[0174] The memory is configured to store computer program code and transmit the computer program code to the processor;

[0175] The processor is used to execute the above-mentioned method for identifying and handling risks in underground powerhouse construction according to the instructions in the computer program code.

[0176] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned method for identifying and handling risks in underground plant construction is implemented.

[0177] Generally speaking, computer instructions for implementing the method of the present invention may be carried by any combination of one or more computer-readable storage media. Non-transitory computer-readable storage media may include any computer-readable media except for signals that are temporarily propagating.

[0178] Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or components, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EKROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or device.

[0179] Computer program code for performing the operations of the present invention can be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages, in particular, Python suitable for neural network computing and platform frameworks based on TensorFlow, PyTorch, etc. can be used. The program code can be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or to an external computer (for example, through the Internet using an Internet service provider).

[0180] The above-mentioned equipment and non-temporary computer-readable storage medium can be referred to the specific description of a method for identifying and handling risks in underground powerhouse construction and its beneficial effects, which will not be repeated here.

[0181] Although the embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for identifying and handling risks in underground powerhouse construction, characterized in that: include: Real-time collection of multi-source heterogeneous data during the construction of underground powerhouses, acquisition of historical risk case data for underground powerhouse construction, and pre-processing of the multi-source heterogeneous data and the historical risk case data to obtain structured spatiotemporal correlation data; Constructing a multi-dimensional analysis model based on a parallel computing algorithm, and performing multi-scale risk analysis on structured spatiotemporal correlation data according to the multi-dimensional analysis model to obtain multi-level risk feature data; Identifying risk characteristics of the multi-level risk feature data using a multimodal machine learning model to obtain risk quantification indicators, and identifying the risk quantification indicators based on a fuzzy comprehensive evaluation algorithm to obtain a construction risk level; The emergency strategies corresponding to the construction risk levels are matched from the preset disposal knowledge base, and the matched emergency strategies are parameter-expanded through a combinatorial optimization algorithm to generate multiple candidate disposal solutions. The candidate disposal solutions are weighted and ranked using a multi-criteria evaluation model to output the optimal disposal solution.

2. A method for identifying and handling risks in underground powerhouse construction according to claim 1, characterized in that: The real-time collection of multi-source heterogeneous data during the underground powerhouse construction process, obtaining historical risk case data of the underground powerhouse construction, and pre-processing the multi-source heterogeneous data and the historical risk case data to obtain structured spatiotemporal correlation data include: The multi-source heterogeneous data includes geological monitoring data, construction environment data, equipment operation data and project progress data; Based on the three-dimensional spatial characteristics of the underground powerhouse construction scene, geological monitoring sensors, construction environment sensors, and equipment operation sensors are deployed in layers according to spatial topological relationships. The project progress data is obtained in real time through the construction progress management system, and the historical risk case data is obtained through the construction safety database. Kalman filtering is used to reduce noise on geological monitoring data; the sliding interquartile range method is used to detect outliers in construction environment data, and missing periods are filled through cubic spline interpolation; the missing parameters of equipment operation data under abnormal conditions are reconstructed through a long short-term memory network prediction network to obtain corrected equipment operation data; the project progress data and the historical case data are dimensionally aligned based on a spatiotemporal alignment matrix.

3. A method for identifying and handling risks in underground powerhouse construction according to claim 2, characterized in that: The long short-term memory network prediction network adopts a dual-channel attention mechanism, and the calculation formula for correcting the device operation data is: in, The predicted value of the equipment operating parameters for correcting the equipment operating data; β j is the spatial attention weight of the jth device running sensor; N2 is the number of device running sensors; α m is the temporal attention weight at time m; τ is the length of the time lookback window; is the original value of the device operating parameter of the jth device operating sensor at time tm; softmax(·) is the normalized exponential function; v a is a trainable weight vector; tanh(·) is the hyperbolic tangent activation function; W a is the time-trainable weight matrix; h t-1 is the hidden state vector of the long short-term memory network at time t-1; x t is the input feature vector at time t; exp(·) is the exponential function; s j The spatial encoding vector of the sensor running on the jth device; s k is the spatial encoding vector of the sensor running on the kth device; c is the context vector; W s is the spatial trainable weight matrix.

4. A method for identifying and handling risks in underground powerhouse construction according to claim 1, characterized in that: The multi-dimensional analysis model is constructed based on the parallel computing algorithm, and multi-scale risk analysis is performed on the structured spatiotemporal correlation data according to the multi-dimensional analysis model to obtain multi-level risk feature data, including: The underground powerhouse construction space is divided into multiple computational blocks according to engineering geological conditions and construction procedures, and independent parallel computing nodes are assigned to each block. A multidimensional analysis model is constructed, which includes a global-local coupling layer, a multi-scale feature extraction layer, and a risk association reasoning layer. The global-local coupling layer uses tensor decomposition to correlate and map global construction parameters with local monitoring data to establish a cross-scale data correlation matrix. The multi-scale feature extraction layer extracts multi-frequency features from time series data using wavelet transform and empirical mode decomposition, and extracts topological features from spatial data using nonlinear manifold dimensionality reduction. The risk association inference layer constructs a construction risk impact propagation model based on a graph neural network, which is used to identify the cascading effects between different risk factors. Based on a multi-dimensional analysis model, multi-scale risk features are extracted from structured spatiotemporal correlation data according to the macro-project layer, the meso-regional layer, and the micro-perspective layer. At each scale level, a time series anomaly detection model, a spatial clustering analysis model and a spatiotemporal pattern evolution prediction model are constructed respectively. The model outputs of the time series anomaly detection model, the spatial clustering analysis model and the spatiotemporal pattern evolution prediction model are integrated through an adaptive feature fusion algorithm to obtain multi-level risk feature data.

5. A method for identifying and handling risks in underground powerhouse construction according to claim 4, characterized in that: The calculation formula for the multi-level risk characteristic data is: Among them, Risk int It is multi-level risk characteristic data; M, R, and P are macro-scale, meso-scale, and micro-scale respectively; is the scale b in time Dynamic weight function of N b is the number of features at scale b; is the importance coefficient of the cth feature in scale b; is the measured value of the cth feature in scale b; is the historical mean of the cth feature in scale b; is the historical standard deviation of the cth feature in scale b; Φ b (·) is the nonlinear mapping function of scale b.

6. The method for identifying and handling risks in underground powerhouse construction according to claim 1 is characterized in that: The multi-modal machine learning model is used to identify the risk characteristics of the multi-level risk characteristic data to obtain a risk quantification index, and the risk quantification index is identified based on a fuzzy comprehensive evaluation algorithm to obtain a construction risk level, including: Constructing a multimodal machine learning model, the multimodal machine learning model including a feature extraction network, a temporal feature learning network, and a multimodal feature fusion network, and performing deep feature extraction and correlation analysis on the multi-level risk feature data through the multimodal machine learning model to obtain risk quantification indicators; A risk factor weight system based on the hierarchical analysis method is established, and a fuzzy comprehensive evaluation matrix is constructed in combination with the fuzzy membership function. The comprehensive evaluation results of the risk quantification indicators are synthesized through fuzzy operations. The comprehensive evaluation results are mapped to construction risk levels according to the preset threshold intervals. The construction risk levels include negligible, low risk, medium risk, high risk and extremely high risk.

7. The method for identifying and handling risks in underground powerhouse construction according to claim 1 is characterized in that: The method includes matching the emergency strategies corresponding to the construction risk levels from the preset disposal knowledge base, performing parameter expansion on the matched emergency strategies through a combinatorial optimization algorithm, generating multiple candidate disposal solutions, and using a multi-criteria evaluation model to calculate and rank the candidate disposal solutions, and outputting the optimal disposal solution, including: Matching the corresponding emergency strategy according to the construction risk level from the preset disposal knowledge base to form a preliminary disposal plan; Using a combinatorial optimization algorithm to expand and optimize multiple parameters of the preliminary treatment plan to generate multiple candidate treatment plans; Based on the multi-criteria evaluation model, the candidate disposal plans are weighted and ranked, and the optimal disposal plan is output.

8. A risk identification and treatment system for underground powerhouse construction, characterized in that: The system is applied to the method according to any one of claims 1 to 7, and the system comprises: A data acquisition and processing module is used to collect multi-source heterogeneous data during the construction of the underground powerhouse in real time, obtain historical risk case data of the underground powerhouse construction, and pre-process the multi-source heterogeneous data and the historical risk case data to obtain structured spatiotemporal correlation data; An analysis model building module is used to build a multi-dimensional analysis model based on a parallel computing algorithm, and perform multi-scale risk analysis on structured spatiotemporal correlation data according to the multi-dimensional analysis model to obtain multi-level risk feature data; A risk level identification module is used to identify risk characteristics of the multi-level risk feature data using a multimodal machine learning model to obtain risk quantification indicators, and to identify the risk quantification indicators based on a fuzzy comprehensive evaluation algorithm to obtain a construction risk level; The disposal scheme matching module is used to match the emergency strategy corresponding to the construction risk level from the preset disposal knowledge base, expand the parameters of the matched emergency strategy through the combination optimization algorithm, generate multiple candidate disposal schemes, use the multi-criteria evaluation model to calculate the weights and sort the candidate disposal schemes, and output the optimal disposal scheme.

9. An underground powerhouse construction risk identification and treatment device, characterized in that: including memory and processor; The memory is configured to store computer program code and transmit the computer program code to the processor; The processor is configured to execute the method according to any one of claims 1 to 7 according to instructions in the computer program code.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Citation Information

Cited By

  • Cascade hydropower remote centralized control plan generation method and system

    CN120634283A

  • AI-based construction project risk prediction and decision optimization method and system

    CN120822836A

  • Power consumer anomaly detection and risk assessment method, system and device based on spatial-temporal feature fusion and storage medium

    CN121119699A

  • Urban water affair situation awareness management method and system

    CN121120304A

  • Underground comprehensive pipe gallery geological risk identification method and system

    CN121744837A