Methods for Monitoring and Managing Safety of On-site Working Environment in Power Infrastructure Construction

Abstract

This invention discloses a method for monitoring and managing the operational environment at power infrastructure construction sites, belonging to the fields of intelligent decision-making technology and power safety management. It includes: constructing a semantic network framework based on the construction plan; collecting and calibrating multi-source environmental data to generate a reliable dataset; generating a real-time risk network diagram based on the semantic framework and the reliable dataset; calculating robust risk indicators and performing sensitivity deconstruction; generating closed-loop intervention instructions when risk indicators exceed safety thresholds; and finally, iteratively optimizing the entire system by collecting feedback and generating a reusable intelligent template library across projects. This invention employs a comprehensive technical approach combining semantic modeling, causal inference, and closed-loop adaptive optimization, enabling deep analysis of operational scenarios, proactive quantification and attribution of potential risks, and intelligent generation of optimal intervention strategies. This significantly improves the intelligence, precision, and forward-looking nature of on-site safety management in power infrastructure construction.

Description

Technical Field

[0001] This invention relates to the fields of intelligent decision-making technology and power safety management, and in particular to a method for monitoring and managing the on-site working environment data of power infrastructure projects. Background Technology

[0002] Power infrastructure construction is a crucial component of energy system development, and its on-site operating environments are typically characterized by complexity, dynamism, and high risk. Current technologies generally employ a combination of methods for safety management at power infrastructure construction sites. On one hand, various sensors, such as anemometers, thermometers, hygrometers, and gas detectors, are deployed to collect data on the on-site environment, and fixed alarm thresholds are set to provide initial environmental risk warnings. On the other hand, project management relies on traditional management methods such as construction plans, safety procedures, and on-site inspections to plan and supervise operational activities. Some advanced systems may utilize video surveillance and artificial intelligence image recognition technologies to monitor unsafe personnel behavior or equipment status.

[0003] However, existing technical solutions have significant limitations. First, environmental monitoring data is often disconnected from specific construction operations, only reporting that a certain environmental parameter exceeds the standard, but failing to reveal the extent of risk that this exceedance poses to specific processes or work teams. Second, the quality of multi-source heterogeneous data varies greatly, lacking effective reliability assessment and calibration mechanisms, and decisions based on unreliable data may be distorted. Furthermore, risk assessments are mostly static and lagging, making it difficult to address dynamic changes in the on-site environment and complex risks involving multiple coupled factors. Finally, successful experiences and lessons learned in safety management often remain at the individual or project level, lacking a systematic mechanism for accumulation and reuse. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a method for monitoring and managing the operational environment data at power infrastructure construction sites. Employing a comprehensive technical approach combining semantic modeling, causal inference, and closed-loop adaptive optimization, it can deeply analyze operational scenarios, proactively quantify and attribute potential risks, and intelligently generate optimal intervention strategies. This significantly improves the intelligence, precision, and forward-looking nature of on-site safety management at power infrastructure construction sites.

[0005] The above objectives can be achieved through the following approach:

[0006] A method for monitoring and managing the on-site work environment of power infrastructure includes: acquiring power infrastructure construction plan data; decomposing the construction plan data into standardized scenario units; constructing semantic nodes based on the five-tuple of "personnel-equipment-process-environment-time sequence"; outputting a semantic network framework; collecting multi-source environmental data, performing source labeling and hierarchical reliability assessment, and verifying system processing deviations to obtain an environmental calibration dataset; extracting causal paths based on the semantic network framework; mapping the environmental calibration dataset into controllable nodes and dynamically assigning weights to generate a real-time risk network graph; and implementing safety management measures for the environmental calibration dataset. Multi-scale perturbation simulation and predictive missing reconstruction are performed to calculate robust risk indicators, and multi-dimensional sensitivity deconstruction is conducted to output the factor influence network and innovation contribution spectrum. The robust risk indicators are monitored and compared with preset safety thresholds. When the robust risk indicators exceed the safety thresholds, control measures are implemented and preventive guidance is issued based on the real-time risk network diagram, generating closed-loop intervention instructions. The on-site feedback data and continuous monitoring indicator sequences after the execution of the closed-loop intervention instructions are collected to perform quantitative evaluation of the effects and deviation diagnosis. The semantic network framework and the safety threshold are iteratively optimized based on feedback, and finally, a reusable intelligent template library is generated across projects.

[0007] Optionally, the output semantic network framework includes: performing structured parsing on the power infrastructure construction plan data, identifying work activity boundaries and dependencies, and generating a work activity list; recursively segmenting the work activity list, decomposing it into standardized context units according to time granularity and spatial scope, and assigning a unique identifier to each standardized context unit; extracting the "personnel-equipment-process-environment-time sequence" five-tuple elements from each standardized context unit through semantic feature extraction, establishing an element association mapping table, and generating context nodes; connecting the context nodes according to work logic relationships and spatiotemporal constraints to construct a hierarchical semantic node network, calculating the semantic similarity weights between nodes, and outputting the semantic network framework.

[0008] Optionally, obtaining the environmental calibration dataset includes: annotating the multi-source environmental data with source metadata according to data type, device model, and collection time, and calculating the initial credibility score of each data source to generate an original environmental dataset; performing hierarchical credibility reassessment on the original environmental dataset, identifying abnormal data sources and deviation patterns, and outputting a hierarchical credibility matrix; performing deviation correction on the hierarchical credibility matrix, processing hierarchical errors between data sources, and fusing to generate an environmental calibration dataset.

[0009] Optionally, the output hierarchical credibility matrix includes: performing temporal consistency analysis on the original environmental dataset, calculating and identifying data mutation points and periodic anomalies, and generating a temporal anomaly label sequence; performing cross-validation on different data sources in the original environmental dataset, calculating and evaluating the consistency deviation between data sources, and outputting a deviation pattern feature vector; and dynamically adjusting the initial credibility score of each data source by combining the temporal anomaly label sequence and the deviation pattern feature vector to generate a hierarchical credibility matrix.

[0010] Optionally, the method further includes: associating context nodes in the semantic network framework with data source credibility in the hierarchical credibility matrix through semantic matching to generate a node credibility association table; based on the node credibility association table, performing credibility propagation calculation on the semantic network framework, and spreading trust values ​​according to the semantic similarity weights between nodes and the data source credibility distribution to output an enhanced semantic credibility graph.

[0011] Optionally, generating the real-time risk network graph includes: performing path traversal analysis on the semantic network framework, extracting multi-layer causal paths, calculating the dependency strength between paths, and generating a set of causal paths; projecting environmental variables from the environmental calibration dataset onto the set of causal paths, identifying the position and attributes of nodes, adjusting the node control priority, and outputting a mapped node graph; dynamically assigning weights to the mapped node graph, calculating edge weight vectors and updating propagation weights, and generating the real-time risk network graph.

[0012] Optionally, the output factor influence network and innovation contribution spectrum includes: performing multi-scale perturbation simulation on the environmental calibration dataset, calculating the perturbation variation sequence, and incorporating the enhanced semantic credibility map for credibility-weighted filtering to output the perturbation simulation dataset; applying predictive missing value reconstruction to the perturbation simulation dataset to fill data gaps and outliers, and calculating robust risk indicators; performing multi-dimensional sensitivity deconstruction on the robust risk indicators, analyzing the interaction effects between factors, constructing the network topology, and outputting the factor influence network and innovation contribution spectrum.

[0013] Optionally, generating closed-loop intervention instructions includes: continuously monitoring the robust risk indicators, dynamically comparing them with the safety threshold, and incorporating the factor influence network to adjust the sensitivity of the safety threshold, generating a threshold trigger signal; when the threshold trigger signal is activated, extracting risk paths and control nodes based on the real-time risk network graph, generating execution control instructions; simulating potential consequences for the risk paths, formulating targeted prevention guidance, distributing it to relevant personnel and equipment, and outputting a prevention guidance dataset; fusing the execution control instructions and the prevention guidance dataset, verifying intervention consistency, and constructing closed-loop intervention instructions.

[0014] Optionally, the final cross-project aggregation to generate a reusable intelligent template library includes: collecting on-site feedback data and continuous monitoring indicator sequences after the execution of the closed-loop intervention instructions, integrating time series features and semantic annotations to generate a comprehensive effect evaluation dataset; detecting abnormal patterns and causal biases in the comprehensive effect evaluation dataset, calculating quantitative evaluation scores, and generating an optimized parameter set; iteratively updating the semantic network framework and the safety threshold based on the optimized parameter set to generate a reusable intelligent template library.

[0015] Optionally, generating the optimized parameter set includes: performing multi-dimensional anomaly scanning on the comprehensive effect evaluation dataset, isolating data outliers, and extracting periodic and trend-based anomaly patterns to generate an anomaly pattern identification report; estimating the pre- and post-intervention bias based on the anomaly pattern identification report, calculating the causal effect strength, and outputting a causal bias vector; and fusing the anomaly pattern identification report and the causal bias vector to perform multi-objective optimization calculation of a weighted quantitative evaluation score to generate the optimized parameter set.

[0016] Compared with the prior art, the present invention has the following advantages:

[0017] 1. This invention constructs a contextualized semantic network framework and maps multi-source environmental data into it after dynamic credibility assessment, achieving a deep integration of operational logic and environmental status. This makes risk assessment no longer based on isolated data threshold judgments, but on inference based on the element correlation and causal logic in specific operational contexts, thereby significantly improving the predictability and accuracy of risk identification. It can capture potential complex risks caused by the coupling of multiple factors at the risk incubation stage, effectively avoiding false alarms and false negatives of traditional methods.

[0018] 2. Through the visualization of real-time risk network diagrams, managers can clearly locate the sources and propagation paths of risks. The closed-loop intervention instructions generated by the system not only include immediate control measures but also incorporate forward-looking preventive guidance. This targeted and guiding intervention approach overcomes the shortcomings of vague instructions and crude measures in traditional safety management, greatly improving the efficiency of safety response and the effectiveness of control measures.

[0019] 3. Through quantitative evaluation of intervention effects and deviation diagnosis, the system can automatically generate optimized parameter sets and iteratively update risk models and safety thresholds, enabling the entire safety management system to have self-learning and continuous optimization capabilities. Simultaneously, validated successful experiences are aggregated into a reusable intelligent template library, achieving systematic accumulation and cross-project transfer of safety management knowledge, thus solving the problem of safety management levels being highly dependent on personal experience and difficult to continuously improve.

[0020] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating the method for monitoring and managing on-site work environment data in power infrastructure according to an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram of the hierarchical structure decomposition of the semantic network framework according to an embodiment of the present invention.

[0024] Figure 3 This is a schematic diagram of the multi-scale perturbation simulation and robust risk indicator generation process according to an embodiment of the present invention.

[0025] Figure 4 This is a schematic diagram of the closed-loop intervention and model adaptive optimization process in an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Reference Figure 1 One embodiment of the present invention proposes a method for monitoring and managing the on-site work environment of power infrastructure. It adopts a comprehensive technical approach of semantic modeling, causal inference and closed-loop adaptive optimization, which can deeply analyze the work situation, proactively quantify and attribute potential risks, and intelligently generate the optimal intervention strategy, significantly improving the intelligence, accuracy and foresight of on-site safety management of power infrastructure.

[0028] The method described in this embodiment specifically includes:

[0029] Acquire power infrastructure construction plan data, decompose the power infrastructure construction plan data into standardized context units, and construct semantic nodes based on the five-tuple of "personnel-equipment-process-environment-time sequence" to output a semantic network framework;

[0030] Collect multi-source environmental data, perform source labeling and hierarchical reliability assessment, and verify system processing deviations to obtain an environmental calibration dataset;

[0031] Based on the semantic network framework, causal paths are extracted, the environmental calibration dataset is mapped into adjustable nodes, and weights are dynamically assigned to generate a real-time risk network graph.

[0032] Multi-scale perturbation simulation and predictive missing reconstruction are performed on the environmental calibration dataset to calculate robust risk indicators, and multi-dimensional sensitivity deconstruction is performed to output the factor influence network and innovation contribution spectrum.

[0033] Monitor and compare the robust risk indicators with the preset safety thresholds. When the robust risk indicators exceed the safety thresholds, execute control measures and issue preventive guidance based on the real-time risk network diagram, generating a closed-loop intervention instruction.

[0034] The system collects on-site feedback data and continuous monitoring indicator sequences after the execution of the closed-loop intervention instructions, performs quantitative evaluation of the effects and diagnosis of deviations, iteratively optimizes the semantic network framework and the safety threshold, and finally aggregates and generates a reusable intelligent template library across projects.

[0035] By adopting a comprehensive technical approach that combines semantic modeling, causal inference, and closed-loop adaptive optimization, the system can deeply analyze the operational context, proactively quantify and attribute potential risks, and intelligently generate optimal intervention strategies, significantly improving the intelligence, precision, and foresight of on-site safety management in power infrastructure projects.

[0036] Optionally, the output semantic network framework includes:

[0037] The power infrastructure construction plan data is structured and parsed to identify the boundaries and dependencies of work activities and generate a list of work activities.

[0038] Specifically, this step uses natural language processing technology or a dedicated parser to identify independent work activities with clearly defined start and end boundaries from unstructured or semi-structured planning data, and analyzes the logical dependencies between these activities, such as the succession relationship between preceding and subsequent tasks. After this step, a structured list of work activities is output, in which each entry clearly defines a macro-level work task and its position in the entire construction process.

[0039] The task activity list is recursively segmented, decomposed into standardized context units according to time granularity and spatial scope, and a unique identifier is assigned to each standardized context unit.

[0040] Specifically, each activity in the task activity list is recursively decomposed. This decomposition process is recursive until each unit reaches the minimum indivisible task situation. After processing, each minimum unit is assigned a unique identifier, thereby generating a series of standardized situation units with unique identifiers.

[0041] For the standardized context unit, the five-tuple elements of "personnel-equipment-process-environment-time sequence" are extracted through semantic feature extraction, and an element association mapping table is established to generate context nodes;

[0042] Specifically, the core of this step is to build a multi-dimensional semantic profile for each minimum work scenario. By analyzing the content of standardized scenario units, relevant information is categorized and extracted into personnel elements, equipment elements, process elements, environmental elements, and temporal elements. Simultaneously, a five-tuple element association mapping table is established, which clarifies the subordinate or collaborative relationships between elements, such as which personnel operates which equipment and performs which process. Finally, the standardized scenario unit, containing a unique identifier, five-tuple elements, and their element association mapping table, is encapsulated into a data structure, forming a scenario node.

[0043] The context nodes are connected according to the task logic relationship and spatiotemporal constraints to construct a hierarchical semantic node network, and the semantic similarity weights between nodes are calculated to output the semantic network framework.

[0044] Specifically, this step weaves discrete context nodes into a cohesive network with inherent logic. The task-related logical relationships are primarily inherited from the task dependencies resolved in the first step, forming directed connections between nodes. Spatiotemporal constraints define the exclusive or collaborative relationships between nodes. After the connections are established, semantic similarity weights between nodes are calculated to quantify the strength of their associations, such as... Figure 2 As shown in the figure, the power infrastructure construction plan is centered on the power infrastructure construction plan and radiates outward to show how it is decomposed into multiple scenario nodes. Each scenario node is further decomposed into a five-element element of personnel, equipment, process, environment and time sequence, revealing the nested and hierarchical data structure of the operation scenario in the method of the present invention.

[0045] After calculating the weights of all node pairs, a semantic network framework is finally formed, consisting of context nodes, logical connection edges, and semantic similarity weights.

[0046] Optionally, the obtained environmental calibration dataset includes:

[0047] The multi-source environmental data is labeled with source metadata according to data type, device model and collection time, and the initial credibility score of each data source is calculated to generate the original environmental dataset.

[0048] Specifically, this step involves identifying and initially assessing the quality of environmental data from multiple sources, including fixed environmental monitoring stations, portable sensors, and drone inspection systems. Each piece of collected data is labeled with its metadata, which includes at least the data type, the model of the source device, and a precise collection timestamp. After labeling, an initial reliability score is assigned based on the official specifications, historical operational stability records, or calibration certificates from authoritative institutions. This score is a quantified inherent reliability indicator. Through this step, all raw data is compiled into a raw environmental dataset with identification and a preliminary quality assessment.

[0049] The original environmental dataset is subjected to hierarchical credibility reassessment to identify anomalous data sources and deviation patterns, and a hierarchical credibility matrix is ​​output.

[0050] Specifically, this process aims to identify anomalous data sources in the data stream that may exhibit sudden failures or persistent deviations, and to identify deviation patterns between different data sources, such as a particular sensor consistently scoring higher or lower than similar sensors. By combining the identification results of anomalous data sources with the analysis of deviation patterns, the initial credibility scores of each data source are dynamically adjusted and updated. Finally, a hierarchical credibility matrix is ​​generated, with data sources as rows and time or contextual units as columns. Its matrix elements represent the dynamic credibility score of each data source at a specific time, thus providing a deep characterization of the dynamic changes in data quality.

[0051] The hierarchical confidence matrix is ​​biased and corrected to handle hierarchical errors between data sources, and then fused to generate an environmental calibration dataset.

[0052] Specifically, this step uses the evaluation results of the hierarchical credibility matrix generated in the previous step to clean and fuse the original environmental dataset. For identified outliers or unreliable data points, interpolation, removal, or correction based on high-credibility data sources can be used. For biased data sources, correction is performed according to their bias patterns. Finally, a weighted fusion algorithm is used to fuse similar data from multiple corrected data sources. The weights during fusion are derived from the dynamic credibility scores of the corresponding data sources in the hierarchical credibility matrix; the higher the score, the greater its contribution to the fusion result, and vice versa. By performing this processing on all environmental parameters and time points, a unified and reliable environmental calibration dataset is ultimately output.

[0053] Optionally, the output hierarchical credibility matrix includes:

[0054] Perform time-series consistency analysis on the original environmental dataset, calculate and identify data mutation points and periodic anomalies, and generate a time-series anomaly marker sequence;

[0055] Specifically, this step focuses on the data quality within a single data source. For the time-series data of each data source, a combination of anomaly detection algorithms is applied. This combination may include, for example, moving averages, isolated forest algorithms, or statistical methods, to identify outliers in the data stream. These outliers include data mutations whose values ​​far exceed the normal fluctuation range, as well as changes that deviate from expected patterns. For example, a temperature sensor might experience a sharp jump in reading within a short period, or a noise sensor might continuously report high decibel readings during nighttime when there is no construction activity. Whenever an outlier is identified, the data source is marked at the corresponding time point. Ultimately, a time-series anomaly tag sequence is generated for each data source, which records in detail the time points at which the data source exhibited unstable behavior.

[0056] Cross-validation is performed on different data sources in the original environmental dataset to calculate and evaluate the consistency deviation between data sources, and the deviation pattern feature vector is output.

[0057] Specifically, this step focuses on data consistency across different data sources. Multiple data sources monitoring the same environmental parameter are selected and their data are compared within the same time window. Consistency deviations between data sources are assessed by calculating their correlation, difference distribution, or covariance. For example, the degree of consistency can be quantified by calculating the Pearson correlation coefficient between two wind speed sensor reading sequences, or by analyzing the mean and variance of their difference sequences. Specific data sources that exhibit significant and persistent deviations from most other data sources are identified, and their deviation characteristics are modeled. The final output is a deviation pattern feature vector, which describes the behavior of each data source relative to its group of similar data sources using a set of parameterized metrics.

[0058] By combining the time-series anomaly marker sequence and the deviation pattern feature vector, the initial credibility score of each data source is dynamically adjusted to generate a hierarchical credibility matrix.

[0059] Specifically, this step integrates the analysis results of the previous two steps to perform a final dynamic quantification of the data source's reliability. This dynamic adjustment process is accomplished through a non-linear update formula, which can more precisely reflect the impact of different types of biases on reliability. A data source At the point of time Dynamic credibility score It can be calculated using the following formula:

[0060] ,

[0061] in, It is a data source derived from the original environment dataset. The initial credibility score; It is a measure of the severity of a time-series anomalies derived from a sequence of time-series anomaly markers. If at a given time point... If there is no abnormality, its value is 0; if there is an abnormality, its value is a positive number indicating the degree of abnormality. It is a cross-validation bias metric calculated from the bias pattern feature vector, representing the data source. The degree of inconsistency with the overall level of similar data sources; and These are sensitivity control parameters for the effects of timing anomalies and cross-validation bias, used to adjust the intensity of the penalty; This is a time-dependent adaptive weighting parameter, with a value between 0 and 1, used to balance the impact of transient anomalies on the current credibility score. This parameter is periodically updated by an online learning algorithm based on the performance of historical data. By performing this calculation on all data sources and all time points, a hierarchical credibility matrix is ​​ultimately generated, which accurately reflects the comprehensive evaluation of the reliability level of each data source at different times.

[0062] Optionally, the method further includes:

[0063] The context nodes in the semantic network framework are associated and mapped with the data source credibility in the hierarchical credibility matrix through semantic matching to generate a node credibility association table.

[0064] Specifically, this step aims to establish a direct link between structured operational logic and dynamic data quality assessment. Each context node in the semantic network framework is traversed, and its environmental and temporal elements are parsed, such as the spatial extent and time window in which the node is located. Then, the source metadata associated with the hierarchical credibility matrix is ​​queried to find data sources that match the context node in time and space. For example, a context node representing "high-altitude operations in area A in the morning" will be matched with sensor data sources such as wind speed, temperature, and humidity deployed in area A and collecting data during the corresponding morning time period. This matching process can be automatically completed based on spatial inclusion relationships determined by Geographic Information System (GIS) and time stamp interval overlap determination. After matching all context nodes, a node credibility association table is generated, which clearly records all data sources that each context node depends on and their dynamic credibility scores at the corresponding time.

[0065] Based on the node credibility association table, credibility propagation calculation is performed on the semantic network framework. Trust value diffusion is carried out according to the semantic similarity weight between nodes and the credibility distribution of the data source, and an enhanced semantic credibility graph is output.

[0066] Specifically, the core idea of ​​this step is that the final credibility of a task activity depends not only on the quality of the data source directly monitoring it, but also on the credibility of its upstream and downstream related activities in the task process. First, using a node credibility association table, an initial node credibility is calculated for each context node. This initial node credibility can be calculated by weighted averaging of the dynamic credibility scores of all data sources associated with that node. Then, an iterative propagation algorithm is used to diffuse trust values ​​within the semantic network framework. In each iteration, the updated credibility of a node is determined by its own initial node credibility and the current credibility of all its neighboring nodes. The influence of neighboring nodes depends on the semantic similarity weights between them as defined by the semantic network framework. This iterative process continues until the credibility values ​​of all nodes tend to converge stably. Finally, the converged node credibility values ​​are assigned to the corresponding nodes in the semantic network framework, forming an enhanced semantic credibility graph, where each node and each edge contains credibility information calculated through deep fusion and propagation.

[0067] Optionally, generating a real-time risk network graph includes:

[0068] The semantic network framework is subjected to path traversal analysis to extract multi-layer causal paths, and the dependency strength between paths is calculated to generate a set of causal paths.

[0069] Specifically, this step aims to identify potential risk transmission chains from a static semantic network framework. A graph traversal algorithm is employed, starting from context nodes representing initial procedures or critical operations, and exploring all possible paths along operational logic and spatiotemporal constraints until the final node. During the traversal, based on a pre-defined safety management knowledge base, key nodes with strong causal relationships are identified; for example, the "strong wind" environmental node connects to the "high-altitude hoisting" procedure node, and then to the "personnel fall" risk consequence node. For each extracted path, a comprehensive inter-path dependency strength is calculated based on the semantic similarity weights or causal relationship strengths between nodes along the path. Finally, all identified paths and their inter-path dependency strengths are integrated to generate a causal path set.

[0070] The environmental variables in the environmental calibration dataset are projected onto the causal path set, the position and attributes of the nodes are identified, the node control priority is adjusted, and the mapping node graph is output.

[0071] Specifically, this step involves injecting real-time, quality-verified environmental state information into the logical framework. Each environmental variable in the environmental calibration dataset is precisely mapped to the corresponding environmental element node in the causal path set based on its spatiotemporal attributes, thus updating the abstract state of the node to a concrete real-time value. The control priority of each node is dynamically adjusted based on its real-time state, its critical position in the causal path, and its credibility score derived from the enhanced semantic credibility graph. For example, if a wind speed node's value is close to a critical threshold and it is located on multiple important causal paths, its control priority will be significantly increased. This step outputs a mapped node graph containing real-time data and dynamic priorities.

[0072] The mapped node graph is dynamically weighted, the edge weight vector is calculated and the propagation weight is updated to generate a real-time risk network graph.

[0073] Specifically, this step is the core of quantifying risk and performing propagation calculations within the network. Dynamic risk weights are assigned to each edge and each node in the mapped node graph. A node's own risk weight is calculated using its contained environmental variable values ​​and a pre-defined risk baseline function. The edge's risk weight, i.e., the propagation weight, is determined by the strength of path dependencies, the risk value of upstream nodes, and the edge's regulatory priority. The weight update process is dynamic, using an iterative calculation through a non-linear aggregation formula to simulate the propagation and accumulation of risk. A node... Risk value at the next moment It can be calculated using the following formula:

[0074] ,

[0075] in, It is a node exist The inherent risk value generated by its own environmental state at any given moment is calculated from the environmental variables it maps to. It is a propagation impact factor with a value between 0 and 1, used to balance the proportion of intrinsic risk and external input risk at a node; It is a non-linear activation function, such as the Sigmoid function, used to simulate the saturation effect in the risk accumulation process; Represents all pointer nodes upstream node A set; It is an upstream node exist Risk value at any given moment; It is the edge The propagation weight, which integrates... and The strength of inter-path dependency and the control priority of the propagation path; It originates from nodes in the enhanced semantic credibility graph. A credibility score is used to weight and correct upstream input risks, ensuring that the risk propagation impact of low-credibility nodes is suppressed. Through this calculation, risk values ​​spread and accumulate outward from high-risk source nodes along causal paths. The calculation is performed on the entire network, ultimately generating a real-time risk network graph where each node and edge is assigned a real-time risk weight.

[0076] Optionally, the output factors influencing the network and innovation contribution spectrum include:

[0077] Multi-scale perturbation simulation is performed on the environmental calibration dataset to calculate the perturbation variation sequence, and the enhanced semantic credibility map is incorporated for credibility-weighted filtering to output the perturbation simulation dataset.

[0078] Specifically, this step proactively tests safety boundaries by simulating potential future environmental changes. Key environmental variables from the environmental calibration dataset, such as wind speed, temperature, and rainfall, are selected, and perturbations of different scales and combinations are applied to their existing values. These perturbations can be gradual trends or abrupt changes, simulating various potential severe weather or abnormal operating conditions. The generated perturbation data sequences, i.e., perturbation variant sequences, are integrated into an enhanced semantic credibility graph. This integration process is a crucial filtering step, using the credibility score of each node in the graph to weight and filter the simulation data. This means that perturbation simulation results based on low-credibility data sources will have their weight reduced, thus ensuring the effectiveness and reliability of the simulation. After this step, the final perturbation simulation dataset is output.

[0079] Predictive missing data reconstruction is applied to the aforementioned perturbation simulation dataset to fill in data gaps and outliers, and a robust risk index is calculated.

[0080] Specifically, since perturbation simulations may generate incomplete or outlier data, this step first applies predictive missing data reconstruction techniques, such as imputation methods based on time series models or machine learning models, to fill in and repair the perturbation simulation dataset, ensuring its integrity and consistency. Subsequently, the reconstructed dataset is used as input into the risk calculation model of the real-time risk network graph to simulate risk responses under various perturbation scenarios. Statistical analysis is performed on the risk results under all perturbation scenarios, such as calculating their mean, variance, or specific quantiles, ultimately yielding one or a set of robust risk indicators. A robust risk indicator is one that is insensitive to small fluctuations in the input data and can more stably reflect the true risk level in the face of uncertainty.

[0081] The robust risk indicators are subjected to multidimensional sensitivity deconstruction to analyze the interaction between factors, and a network topology is constructed to output the network of factors influencing innovation contribution spectrum.

[0082] Specifically, this step aims to identify which factors drive risk changes. A sensitivity analysis method, such as Monte Carlo analysis, analysis of variance, or gradient-based methods, is used to quantitatively assess the impact of changes in each input factor on the final calculated robust risk index. This analysis not only assesses the independent impact of individual factors but, more importantly, analyzes the interactive effects between factors, such as the amplifying effect of the combined occurrence of high temperature and high humidity on equipment failure risk. Based on the results of the sensitivity analysis, a network topology is constructed, where nodes represent various influencing factors, and edge weights represent the strength of interactive effects between factors, thus outputting a visualized factor influence network. Simultaneously, the contribution of each factor to the total risk variance is quantified and ranked, generating an innovation contribution spectrum, such as... Figure 3 As shown in the figure, the evolution paths of a large number of possible risk indicators over a period of time can be seen by applying multiple perturbations to key environmental parameters starting from the current point in time. At the final moment, the endpoint values ​​of all paths converge into a probability density distribution. The statistical characteristics of this distribution are defined as robust risk indicators, which reflect a quantitative assessment of future uncertainties.

[0083] Optionally, the method for generating closed-loop intervention commands includes:

[0084] The robust risk indicator is continuously monitored, dynamically compared with the safety threshold, and the factor influence network is incorporated to adjust the sensitivity of the safety threshold and generate a threshold trigger signal.

[0085] Specifically, robust risk indicators are continuously acquired and updated, and compared in real time with a safety threshold. This safety threshold is not fixed but dynamic. The sensitivity of the safety threshold is intelligently adjusted based on the factor influence network generated in previous steps. Specifically, if the factor influence network reveals that a certain environmental factor is a key risk driver in the current operational context, the safety threshold of the corresponding risk indicator is automatically lowered, making it more sensitive and thus enabling early warning of key risks. When a monitored robust risk indicator exceeds the dynamically adjusted safety threshold, a threshold trigger signal is immediately generated, indicating the need to initiate an intervention procedure.

[0086] When the threshold trigger signal is activated, risk paths and control nodes are extracted based on the real-time risk network diagram, and execution control instructions are generated.

[0087] Specifically, upon receiving a threshold trigger signal, the system immediately locks onto the real-time risk network diagram. By analyzing the node with the highest risk value and the path with the greatest risk propagation weight in the diagram, the source of the risk and key risk paths can be accurately located. Based on this location result, control nodes requiring intervention are automatically extracted. These nodes may represent a specific piece of equipment, a personnel team, or a specific process. Subsequently, specific and clear execution control instructions are generated for each control node.

[0088] Simulate potential consequences for the aforementioned risk paths, develop targeted prevention guidelines, distribute them to relevant personnel and equipment, and output a prevention guidance dataset;

[0089] Specifically, this step aims to go beyond passive, immediate control and provide proactive safety guidance. Based on a real-time risk network graph, simulations are performed along identified high-risk paths to demonstrate how risks might evolve and spread over a future period without intervention, and the potential ultimate consequences. Based on the simulated potential consequences, a series of targeted preventative guidelines are matched or generated from a knowledge base. These guidelines are then structured to form a preventative guidance dataset.

[0090] By integrating the execution control instructions and the prevention guidance dataset, and verifying the consistency of interventions, a closed-loop intervention instruction is constructed.

[0091] Specifically, this step combines immediate control with long-term prevention to form a complete action plan. The execution control instructions and prevention guidance datasets are integrated. During the integration process, intervention consistency verification is performed to ensure that there are no logical conflicts between the execution control instructions and prevention guidance, and that they can work synergistically. For example, it verifies whether an instruction to stop a certain operation conflicts with the personnel evacuation routes in the prevention guidance. After verification, the two are packaged into a unified, structured information package, i.e., a closed-loop intervention instruction. This instruction is distributed through the platform to relevant managers, on-site workers, and smart device terminals to ensure that the instruction is accurately and promptly received and executed.

[0092] Optionally, the final cross-project aggregation to generate a reusable smart template library includes:

[0093] The on-site feedback data and continuous monitoring indicator sequences after the execution of the closed-loop intervention instructions are collected, and time series features and semantic annotations are integrated to generate a comprehensive effect evaluation dataset.

[0094] Specifically, this step involves collecting and preparing data on the intervention's effectiveness after the closed-loop intervention instructions are executed. On-site feedback data includes qualitative information such as records of the intervention instruction's execution and personnel observation reports. The continuous monitoring indicator sequence consists of time-series data from relevant environmental and equipment sensors after the intervention measures are implemented. First, these multi-source heterogeneous data are time-aligned and formatted. Then, semantic annotations are added to these data based on contextual information within the semantic network framework. Finally, these data are integrated to generate a comprehensive effectiveness evaluation dataset.

[0095] For the comprehensive effect evaluation dataset, abnormal patterns and causal biases are detected, and quantitative evaluation scores are calculated to generate an optimization parameter set;

[0096] Specifically, this step involves a quantitative assessment and in-depth diagnosis of the effectiveness of the intervention measures. First, multidimensional anomaly scanning techniques, such as clustering-based methods or autoencoder networks, are used to detect anomalous patterns in the comprehensive effectiveness evaluation dataset. This detection focuses not only on numerical outliers but also on identifying periodic and trending anomalous patterns with specific regularities, generating an anomaly pattern identification report. Next, using the execution time of the intervention instruction as a boundary, the continuous monitoring indicator sequences before and after the intervention are compared. Causal inference methods, such as difference-in-differences or regression breakpoint design, are applied to estimate the true causal effect strength of the intervention measures on risk indicators, eliminating interference from other confounding factors and outputting a structured causal bias vector. Finally, the anomaly pattern identification report and the causal bias vector are fused through a multi-objective optimization calculation process to obtain a set of weighted quantitative evaluation scores. These scores comprehensively evaluate the overall effectiveness of this closed-loop intervention. Based on these evaluation scores, an optimization parameter set is finally generated, which includes specific adjustment methods for the model and rules.

[0097] Based on the optimized parameter set, the semantic network framework and the security threshold are iteratively updated to generate a reusable intelligent template library.

[0098] Specifically, this step is the execution phase for achieving self-learning and continuous evolution of the entire safety management system. Based on the adjustment methods included in the optimization parameter set, the node connection weights in the semantic network framework, the causal path strength in the real-time risk network graph, and the safety threshold are iteratively updated. After completing the model optimization for this project, the validated complete "situation-risk-intervention-effect" model is further abstracted into a standardized, reusable intelligent template and stored in a cross-project intelligent template library. When a new power infrastructure project is initiated, the most suitable template can be directly called from this library for initialization, thereby accumulating and transferring safety management experience and wisdom, ultimately generating and continuously expanding the reusable intelligent template library, such as... Figure 4As shown in the figure, the graph uses time as the horizontal axis to illustrate the fluctuation process of robust risk indicators as the work context changes. When the risk indicator reaches a dynamically adaptively adjusted safety threshold, a closed-loop intervention command is triggered. After each successful intervention and effect evaluation, the knowledge volume of the reusable intelligent template library in the background will increase accordingly, while the safety threshold will be optimized, thus demonstrating stronger predictability and adaptability in subsequent risk management, reflecting self-learning and evolutionary capabilities.

[0099] Optionally, the generation of the optimization parameter set includes:

[0100] A multidimensional anomaly scan is performed on the comprehensive effect evaluation dataset to isolate outliers and extract periodic and trend anomaly patterns, generating an anomaly pattern recognition report.

[0101] Specifically, in one embodiment of the invention, this step aims to deeply mine unexpected behaviors from massive feedback data after intervention. An unsupervised learning model, such as an autoencoder-based neural network, is used to process the comprehensive effect evaluation dataset. This model first learns a compressed representation of the data under normal operating conditions, and then identifies data outliers through reconstruction errors. Furthermore, a time-series decomposition algorithm is applied to the data sequence after outlier removal to extract statistically significant periodic and trend-based abnormal patterns that newly emerged after the intervention. All identified data outliers and abnormal patterns, along with their statistical characteristics, are summarized to generate an anomaly pattern identification report.

[0102] For the abnormal pattern recognition report, estimate the deviation before and after intervention, calculate the causal effect strength, and output the causal deviation vector;

[0103] Specifically, this step aims to accurately quantify the net effect of the intervention and eliminate the interference of other confounding factors. Using the execution time of the closed-loop intervention instruction as a dividing line, the changes in the statistical distribution of key risk indicators before and after the intervention are compared, i.e., the pre- and post-intervention bias. To accurately calculate the true effect of the intervention, a causal inference method is applied to estimate the strength of the true causal effect of the intervention on the risk indicators. Simultaneously, for each anomalous pattern recorded in the anomaly pattern recognition report, the strength of its causal association with the intervention is also estimated. All calculated causal effect strengths are integrated to output a structured causal bias vector.

[0104] By fusing the abnormal pattern recognition report and the causal deviation vector, a multi-objective optimization calculation is performed to obtain a weighted quantitative evaluation score, thereby generating an optimization parameter set.

[0105] Specifically, this step involves synthesizing all analysis results to generate concrete instructions for guiding the self-evolution of the model and rules. A multi-objective optimization problem is constructed, with the objective function aiming to simultaneously maximize the positive intervention effect represented by the causal bias vector and minimize the unintended negative impacts revealed by the anomalous pattern recognition report, while considering the economic and time costs of the intervention. By solving this multi-objective optimization problem, for example using Pareto front analysis, a set of weighted quantitative evaluation scores is obtained, which comprehensively evaluate the overall effectiveness of this closed-loop intervention. Based on this evaluation score, and combined with the backpropagation or gradient descent results of the optimization algorithm, a set of optimization parameters is finally generated. This parameter set explicitly includes adjustments to specific connection weights in the semantic network framework, corrections to safety thresholds, and update instructions for relevant strategies in the reused intelligent template library.

[0106] To verify the feasibility of the method of this invention, it was applied to a 110kV urban core area cable tunnel project. This project involved a complex environment with high risk of sudden events, making traditional safety management methods inadequate. To achieve intelligent and proactive risk control, the project adopted the method of this invention.

[0107] In this embodiment, the construction plan data is first parsed into a semantic network framework containing a five-tuple of "personnel-equipment-process-environment-time sequence". Simultaneously, multi-source environmental data from various on-site sensors is continuously collected and calibrated to generate a highly reliable environmental calibration dataset, laying the foundation for subsequent risk analysis and decision-making.

[0108] To verify the effectiveness of the method of the present invention, a real-world application scenario during a sudden heavy rainfall event in June 2024 is selected for illustration.

[0109] On the afternoon of June 22, 2024, the critical operation for the day was identified as "cable trench laying." At 15:05, based on external meteorological data and changes in on-site humidity, the predicted risk of rainfall increased sharply, with the robustness risk index rapidly rising from 0.38 to 0.61, approaching the safety threshold. In the real-time risk network diagram, the causal risk path from "heavy rainfall" to "slope stability" and "electrical safety" was highlighted. Multidimensional sensitivity analysis showed that "probability of short-term heavy rainfall" was the primary driving factor. At 15:08, the robustness risk index reached the trigger threshold of 0.67, and a closed-loop intervention command was automatically generated and issued within 3 seconds, requiring "suspension of work in the trench, evacuation of personnel, and initiation of emergency drainage."

[0110] On-site personnel responded immediately, completing evacuation and emergency response before the rainstorm arrived, successfully averting a potential combined collapse and electrocution accident.

[0111] After the event, the feedback optimization module collected on-site feedback and monitoring data, conducted a quantitative evaluation of the effect, and generated an optimization parameter set. The weight of the "rainfall probability" factor was increased by 20%, and an intelligent template for "deep foundation pits responding to severe convective weather" was generated in the reuse intelligent template library.

[0112] Table 1 Real-time Risk Monitoring and Early Warning Data Table

[0113]

[0114] Table 2. Factor Influence Network Analysis and Intervention Instruction Data Table

[0115]

[0116] Table 3. Data on Intervention Effectiveness Evaluation and Model Iteration

[0117]

[0118] As can be seen from the data in Tables 1 to 3, the method of the present invention achieves step-by-step and quantitative early warning of risks, can accurately attribute risks and generate targeted instructions, and continuously optimize the model through closed-loop learning, which fully demonstrates its technical advantages in improving the predictability, accuracy and intelligence of safety management.

[0119] It should be noted that the formulas described above, through the principle of dimensional consistency and mathematical standardization methods (such as normalization, dimensionless parameter conversion, or unit system unification), can translate physical quantities with different properties into unitless standard values ​​or parameters that can be superimposed in the same dimension. This eliminates the interference of different dimensions on the computational logic, allowing the formulas to retain the original data distribution characteristics while possessing mathematical rationality and adaptability to objective laws. These are conventional technical methods and will not be elaborated further. The electrical connections between the various units described above do not necessarily represent direct or indirect connections; any indirect connection method is applicable to the embodiments of this invention as long as it achieves the purpose of this invention. The above descriptions are merely exemplary embodiments of this invention and should not be construed as limiting the scope of this invention.

[0120] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A method for monitoring and managing the on-site work environment data of power infrastructure construction, characterized in that, The method includes: Acquire power infrastructure construction plan data, decompose the power infrastructure construction plan data into standardized context units, and construct semantic nodes based on the "personnel-equipment-process-environment-time sequence" five-tuple to output a semantic network framework; The process involves collecting multi-source environmental data, performing source-end annotation and hierarchical reliability assessment, and verifying system processing deviations to obtain an environmental calibration dataset. This includes: annotating the multi-source environmental data with source-end metadata according to data type, equipment model, and collection time; calculating the initial reliability score for each data source to generate an original environmental dataset; performing hierarchical reliability reassessment on the original environmental dataset to identify abnormal data sources and deviation patterns, and outputting a hierarchical reliability matrix; and correcting deviations in the hierarchical reliability matrix to handle hierarchical errors between data sources and fuse them to generate the environmental calibration dataset. Based on the semantic network framework, causal paths are extracted, the environmental calibration dataset is mapped into adjustable nodes, and weights are dynamically assigned to generate a real-time risk network graph. The context nodes in the semantic network framework are associated with the data source credibility in the hierarchical credibility matrix through semantic matching to generate a node credibility association table; based on the node credibility association table, credibility propagation calculation is performed on the semantic network framework, and trust value is diffused according to the semantic similarity weight between nodes and the data source credibility distribution to output an enhanced semantic credibility graph. Multi-scale perturbation simulation and predictive missing value reconstruction are performed on the environmental calibration dataset to calculate robust risk indicators. Multi-dimensional sensitivity deconstruction is then performed to output a factor influence network and innovation contribution spectrum. Specifically, the output factor influence network and innovation contribution spectrum includes: performing multi-scale perturbation simulation on the environmental calibration dataset, calculating perturbation variation sequences, incorporating the enhanced semantic credibility graph for credibility-weighted filtering, and outputting a perturbation simulation dataset; applying predictive missing value reconstruction to the perturbation simulation dataset to fill data gaps and outliers, and calculating robust risk indicators; performing multi-dimensional sensitivity deconstruction on the robust risk indicators to analyze the interaction effects between factors, constructing a network topology, and outputting a factor influence network and innovation contribution spectrum. Monitor and compare the robust risk indicators with the preset safety thresholds. When the robust risk indicators exceed the safety thresholds, execute control measures and issue preventive guidance based on the real-time risk network diagram, generating a closed-loop intervention instruction. The system collects on-site feedback data and continuous monitoring indicator sequences after the execution of the closed-loop intervention instructions, performs quantitative evaluation of the effect and diagnosis of deviations, iteratively optimizes the semantic network framework and the safety threshold, and finally aggregates and generates a reusable intelligent template library across projects. The output hierarchical credibility matrix includes: performing temporal consistency analysis on the original environmental dataset, calculating and identifying data mutation points and periodic anomalies, and generating a temporal anomaly label sequence; performing cross-validation on different data sources in the original environmental dataset, calculating and evaluating the consistency deviation between data sources, and outputting a deviation pattern feature vector; and dynamically adjusting the initial credibility score of each data source by combining the temporal anomaly label sequence and the deviation pattern feature vector to generate a hierarchical credibility matrix.

2. The method for monitoring and managing on-site work environment data in power infrastructure construction according to claim 1, characterized in that, The output semantic network framework includes: The power infrastructure construction plan data is structured and parsed to identify the boundaries and dependencies of work activities and generate a list of work activities. The task activity list is recursively segmented and decomposed into standardized context units according to time granularity and spatial scope, and a unique identifier is assigned to each standardized context unit. For the standardized context unit, the five-tuple elements of "personnel-equipment-process-environment-time sequence" are extracted through semantic feature extraction, and an element association mapping table is established to generate context nodes; The context nodes are connected according to the task logic relationship and spatiotemporal constraints to construct a hierarchical semantic node network, and the semantic similarity weights between nodes are calculated to output the semantic network framework.

3. The method for monitoring and managing on-site work environment data and ensuring safety in power infrastructure construction according to claim 1, characterized in that, The generation of the real-time risk network diagram includes: The semantic network framework is subjected to path traversal analysis to extract multi-layer causal paths, and the dependency strength between paths is calculated to generate a set of causal paths. The environmental variables in the environmental calibration dataset are projected onto the causal path set, the position and attributes of the nodes are identified, the node control priority is adjusted, and the mapping node graph is output. The mapped node graph is dynamically weighted, the edge weight vector is calculated and the propagation weight is updated to generate a real-time risk network graph.

4. The method for monitoring and managing on-site work environment data and ensuring safety in power infrastructure construction according to claim 1, characterized in that, The closed-loop intervention command includes: The robust risk indicator is continuously monitored, dynamically compared with the safety threshold, and the factor influence network is incorporated to adjust the sensitivity of the safety threshold and generate a threshold trigger signal. When the threshold trigger signal is activated, risk paths and control nodes are extracted based on the real-time risk network diagram, and execution control instructions are generated. Simulate potential consequences for the aforementioned risk paths, develop targeted prevention guidelines, distribute them to relevant personnel and equipment, and output a prevention guidance dataset; By integrating the execution control instructions and the prevention guidance dataset, and verifying the consistency of interventions, a closed-loop intervention instruction is constructed.

5. The method for monitoring and managing on-site work environment data and ensuring safety in power infrastructure construction according to claim 1, characterized in that, The final cross-project aggregation and generation of reusable smart template library includes: The on-site feedback data and continuous monitoring indicator sequences after the execution of the closed-loop intervention instructions are collected, and time series features and semantic annotations are integrated to generate a comprehensive effect evaluation dataset. For the comprehensive effect evaluation dataset, abnormal patterns and causal biases are detected, and quantitative evaluation scores are calculated to generate an optimization parameter set; Based on the optimized parameter set, the semantic network framework and the security threshold are iteratively updated to generate a reusable intelligent template library.

6. The method for monitoring and managing the on-site work environment data of power infrastructure construction according to claim 5, characterized in that, The generated optimization parameter set includes: A multidimensional anomaly scan is performed on the comprehensive effect evaluation dataset to isolate outliers and extract periodic and trend-based anomaly patterns, generating an anomaly pattern recognition report. For the abnormal pattern recognition report, estimate the deviation before and after intervention, calculate the causal effect strength, and output the causal deviation vector; By fusing the abnormal pattern recognition report and the causal deviation vector, a multi-objective optimization calculation is performed to obtain a weighted quantitative evaluation score, thereby generating an optimization parameter set.

Images