Intelligent monitoring method and system for construction environment of energy storage power station

Abstract

The invention relates to the technical field of data processing, and provides an energy storage power station construction environment intelligent monitoring method and system. The method comprises the following steps: constructing a monitoring network of a current construction environment; performing feature dimension disassembly on the data monitored by the monitoring network, and constructing a feature analysis atlas of the current construction environment; according to the feature vector of each node in the feature analysis map of the current construction environment, granularity sensitivity evaluation is carried out, and recommended granularity configuration of each feature is obtained; performing enhancement processing on the matched multi-modal monitoring data acquisition according to the recommendation granularity configuration corresponding to each feature; and carrying out construction environment risk index assessment, and triggering a safety alarm when the risk index exceeds a threshold value. According to the method, data islands among all stages of construction are broken, cross-stage risk linkage analysis and current construction environment risk prediction are realized, passive response is changed into active intervention, and the safety management capability of the whole construction process of the energy storage power station is improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a method and system for intelligent monitoring of the construction environment of energy storage power stations. Background Technology

[0002] Energy storage power stations play a crucial role in energy storage and grid stability. Their site selection tends to be in remote areas, such as near wind farms or solar power plants, to facilitate on-site energy consumption. These areas often have harsh natural environments and rapid risk transmission, posing numerous technical challenges to the construction of energy storage power stations. The construction of an energy storage power station mainly consists of civil engineering, battery compartment installation, and electrical installation. Traditional construction monitoring data is severely isolated, and effective coordination mechanisms between different construction stages cannot be formed. For example, in the civil engineering stage, if foundation settlement or soil deformation is not monitored in real time, the construction team cannot adjust the installation strategy in time, which may directly lead to deviations in the battery compartment installation angle, potentially resulting in insufficient electrical wiring slack or even connection failure, leaving the construction in a reactive, reactive state. Because most existing monitoring systems are limited to static data monitoring of a single construction stage, such as the installation stage, focusing only on the orientation of the installed battery compartment, they fail to establish a cross-stage risk linkage analysis mechanism, making it difficult to quantify and warn of the transmission path of construction hazards.

[0003] Therefore, there is an urgent need to propose an intelligent monitoring method that can break down data silos between construction stages, achieve cross-stage risk linkage analysis, and have the ability to predict construction environmental risks, so as to improve the safety management of the entire construction process of energy storage power stations and increase the construction speed. Summary of the Invention

[0004] This application provides an intelligent monitoring method and system for the construction environment of energy storage power stations, aiming to solve the technical problems of data silos between construction stages and the difficulty in achieving cross-stage, multi-dimensional dynamic linkage analysis of construction risks. It aims to shift from passive response to proactive intervention and improve the safety management capabilities of the entire construction process of energy storage power stations.

[0005] The first aspect disclosed in this application provides a method for intelligent monitoring of the construction environment of an energy storage power station, the method comprising: The current construction phase of the energy storage power station is obtained from the BIM system. The multimodal monitoring data required for monitoring is matched from the construction environment description of the current construction phase, and the sensors on the relevant monitoring points are activated to form a monitoring network for the current construction environment. Based on the data monitored by the monitoring network, spatial, temporal and physical features are extracted for the monitoring points within the monitoring network. The spatial adjacency and temporal causal relationships of each monitoring point are analyzed to obtain the spatial-temporal correlation between each monitoring point. With the monitoring points as nodes and the correlation relationships as edges, a feature analysis map of the current construction stage is constructed. The spatial-temporal correlation includes the composite correlation strength and causal direction. Multiple sampling granularity levels and their sampling frequencies are preset. The monitoring data of each mode acquired at each monitoring point are downsampled separately. The information loss function at each granularity is calculated. The information loss function at each granularity is compared with the preset feature importance level threshold. According to the priority-progressive decision-making process, the recommended sampling granularity for each monitoring mode is determined. Based on the recommended sampling granularity for each monitoring modality, new multimodal monitoring data will continue to be collected in the monitoring network of the current construction environment. The acquired data will be enhanced, and the composite correlation strength of the three-dimensional features and edges of the feature parsing graph nodes will be updated to realize the dynamic evolution of the graph. Based on the evolved feature parsing map, global topological features, critical path features, and critical node features are extracted, input into a pre-trained risk assessment model, and output a continuous risk index of the current construction environment. When the risk index exceeds a preset safety threshold, a graded early warning response is triggered.

[0006] Another aspect of this application discloses an intelligent monitoring system for the construction environment of an energy storage power station, the system comprising: The monitoring network construction module 11 is used to obtain the current construction stage of the energy storage power station from the BIM system, match the multimodal monitoring data required for monitoring from the construction environment description of the current construction stage, activate the sensors on the relevant monitoring points, and form a monitoring network for the current construction environment. The graph construction module 12 is used to extract spatial, temporal and physical features for monitoring points in the monitoring network based on the data monitored by the monitoring network, analyze the spatial adjacency relationship and temporal causal relationship of each monitoring point, obtain the spatial-temporal correlation relationship between each monitoring point, and construct the feature analysis graph of the current construction stage with monitoring points as nodes and correlation relationships as edges. The spatial-temporal correlation relationship includes the composite correlation strength and causal direction. The granularity configuration module 13 is used to preset multiple sampling granularity levels and their sampling frequencies, downsample each type of monitoring data acquired at each monitoring point, calculate the information loss function at each granularity, compare the information loss function at each granularity with the preset feature importance level threshold, and determine the recommended sampling granularity for each monitoring mode according to the priority-progressive decision-making process. The graph dynamic evolution module 14 is used to continue to collect new multimodal monitoring data in the monitoring network of the current construction environment according to the sampling granularity recommended for each monitoring mode, enhance the acquired data, update the composite correlation strength of the three-dimensional features and edges of the feature parsing graph nodes, and realize the dynamic evolution of the graph. The construction environment risk module 15 is used to extract global topological features, critical path features and critical node features based on the evolved feature analysis map, input the pre-trained risk assessment model, and output the continuous risk index of the current construction environment. When the risk index exceeds the preset safety threshold, a graded early warning response is triggered.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: 1) Combine the stage semantic understanding capability of BIM with the dynamic networking of multimodal sensors to achieve intelligent mapping of "construction environment → monitoring target → data acquisition"; 2) By fusing spatial-temporal features through graph nodes, a computable construction environment risk knowledge network is constructed to achieve dynamic quantification of risk transmission paths and solve the problem of fragmented traditional monitoring data that cannot form a global perspective; 3) Update the feature information of nodes in the map in real time according to the recommended granularity configuration to ensure that the map is synchronized with the actual construction status and adapts to construction changes; 4) Dynamic evolution feature analysis map, which can identify the propagation path of risk areas and support accurate early warning; 5) Real-time assessment of the comprehensive risks of the construction environment, and triggering graded alarms through thresholds to solve the problem of traditional construction being in a "passive patching" state and unable to respond to risks in a timely manner.

[0008] In summary, the technical solution provided in this application integrates the entire lifecycle of BIM data flow and constructs a computable and correlated feature analysis map, completely breaking down data silos between construction stages, enabling risk linkage analysis, supporting dynamic risk prediction, and achieving an upgrade from "post-event rectification" to "pre-event intervention".

[0009] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating an intelligent monitoring method for the construction environment of an energy storage power station in one embodiment.

[0012] Figure 2 This is an architecture diagram of an intelligent monitoring system for the construction environment of an energy storage power station, as shown in one embodiment.

[0013] Figure labeling: Monitoring network construction module 11, map construction module 12, granularity configuration module 13, map dynamic evolution module 14, construction environment risk module 15. Detailed Implementation

[0014] This application provides an intelligent monitoring method and system for the construction environment of energy storage power stations, which solves the technical problems of low scene-data matching and low recognition efficiency caused by insufficient feature enhancement granularity in the multimodal data processing of rail transit.

[0015] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0016] It should be noted that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0017] Example 1, as Figure 1 As shown, this application provides an intelligent monitoring method for the construction environment of an energy storage power station, including: The current construction phase of the energy storage power station is obtained from the BIM system. Multimodal monitoring data required for monitoring is matched from the construction environment description of the current construction phase, and sensors on relevant monitoring points are activated to form a monitoring network for the current construction environment.

[0018] Specifically, before implementing intelligent monitoring methods, BIM first needs to establish a comprehensive 3D parametric model, independently modeling key components such as battery compartments, foundation piles, and cable trenches, and assigning them unique IDs. Simultaneously, it expands the construction monitoring-specific attribute set in the IFC standard, including key parameters such as component sensitivity coefficients, required sensor types, and monitoring thresholds. This is to transform these key parameters derived from engineering experience into a machine-readable standardized format, thereby achieving a digital expression of engineering semantics. Secondly, the model components are deeply bound to the key process requirements and construction schedule of each construction phase, forming a BIM model with a precise timeline, clearly defining the construction phase time window and corresponding monitoring requirements for each component, as shown in Table 1. Finally, it is also necessary to input environmental baseline data for the construction site (including geological exploration reports, 50-year return period wind speeds, and other climate parameters) and set the normal range and warning thresholds for each monitoring parameter (e.g., the battery compartment tilt angle design threshold <0.5°). Furthermore, the intelligent monitoring system needs to establish a sensor capability matrix and a monitoring point information database. The sensor capability matrix is ​​a structured data table that records the technical specifications and capability parameters of all available sensor devices in the monitoring system. The matrix contains detailed information such as sensor ID, monitoring parameter type (e.g., temperature, humidity, displacement, stress), measurement range, accuracy class, response time, operating environment conditions (temperature range, humidity range, protection level), power consumption parameters, communication protocol, and installation requirements. For example, the matrix record for a temperature sensor is: [ID: T001, Parameter: Temperature, Range: -40~85℃, Accuracy: ±0.1℃, Response Time: 1s, Protection Level: IP67]. The monitoring point information database records static information such as the spatial coordinates, coverage area, and device access capabilities of all potential monitoring points. Simultaneously, a construction environment-monitoring requirements knowledge base is constructed, establishing a mapping relationship between construction environment keywords and monitoring parameter types, providing semantic support for NLP technology to analyze construction environment descriptions.

[0019] Table 1 Furthermore, the multimodal monitoring data required for monitoring is matched from the current construction environment description, including: Obtain a description of the current construction environment of the energy storage power station in text and / or language, analyze the current environment description using NLP technology, and identify the key monitoring needs of the current construction environment; Based on the key monitoring needs of the current construction environment, a pre-constructed sensor capability matrix and monitoring point information database are retrieved. The optimal sensor combination is matched through a greedy algorithm, and the sensors on the relevant monitoring points are activated to form a monitoring network for the current construction environment. The multimodal monitoring data required for monitoring is obtained. The sensor capability matrix is ​​a structured data table used to provide sensor technical specifications and performance parameters, and the monitoring point information database is used to provide the spatial location, coverage area and equipment access information of the monitoring points.

[0020] Specifically, the system acquires scenario description information for the current construction phase through various channels, including text descriptions input by construction management personnel, on-site voice reports, and construction log records. Voice input is converted into text format using speech recognition technology. The acquired text is then preprocessed, including noise removal, punctuation standardization, and synonym unification to ensure text quality. Natural language processing (NLP) technology is then used for deep analysis of the preprocessed text. Named entity recognition (NER) is employed to extract key entities such as construction activities, equipment names, and environmental conditions. Dependency parsing is used to identify semantic relationships between entities, and a word vector model is used to calculate semantic feature vectors. Subsequently, the extracted keywords and semantic features are matched against a pre-built construction environment-monitoring requirements knowledge base. Similarity calculation and fuzzy matching algorithms are used to identify the monitoring parameter types most relevant to the current scenario description. Finally, based on the matching results and confidence scores, a list of key monitoring requirements for the current construction environment is output, including monitoring parameter types, priority levels, and expected monitoring intensity, providing clear guidance for subsequent sensor selection and configuration. The construction of the construction environment-monitoring requirements knowledge base adopts a combination of expert knowledge extraction and historical data mining. First, professional literature, construction specifications, safety standards, and historical construction reports in the field of energy storage power station construction are collected to extract typical scenario description texts for each construction stage. Then, experts in the fields of energy storage power station construction, monitoring, and safety are invited to annotate the collected scenario descriptions, identify key semantic elements such as construction activities, environmental conditions, and equipment status, and label the corresponding monitoring requirement types. Next, natural language processing technology is used to perform word frequency analysis and semantic clustering on the annotated data to extract high-frequency keywords and semantic patterns, such as "battery compartment hoisting" corresponding to "displacement monitoring, stress monitoring," and "cable laying" corresponding to "temperature monitoring, gas monitoring," etc. Finally, a many-to-many mapping relationship is established between the extracted keywords and monitoring parameter types, and the knowledge base is continuously optimized and improved through expert verification and feedback from actual projects to form a scenario-requirement mapping knowledge base covering the entire construction cycle of energy storage power stations.Based on the identified key monitoring needs, the system first filters a candidate set of sensors from the sensor capability matrix that can meet the requirements of various monitoring parameter types. Simultaneously, it queries the monitoring point information database to obtain the location information of available monitoring points that meet spatial coverage requirements. Then, it constructs a sensor-monitoring point matching matrix and calculates the comprehensive score of each sensor at each monitoring point. The score comprehensively considers factors such as sensor accuracy matching, environmental adaptability, power efficiency, spatial coverage capability of the monitoring point, and equipment access capacity. Next, a greedy algorithm is used for optimal combination matching. The algorithm processes monitoring needs one by one according to their priority order. For each monitoring need, it selects the sensor-monitoring point with the highest comprehensive score that has not yet been assigned from the candidate set. The system combines monitoring points and checks whether the combination conflicts with existing combinations in terms of space or resources. If a conflict exists, a suboptimal solution is selected until all critical monitoring needs are assigned to suitable sensor-monitoring point combinations. Finally, the system generates a sensor start command sequence based on the matching results and sends configuration parameters and start commands to the sensors at the selected monitoring points via a communication protocol. After receiving the commands, the sensors begin to collect data according to the specified sampling frequency and accuracy requirements, and transmit the collected multimodal monitoring data back to the intelligent monitoring system in real time, forming a dynamic monitoring network for the current construction environment. Each monitoring point in the network corresponds to one or more activated sensors, and these monitoring points are the monitoring points for the current construction phase.

[0021] Based on the data monitored by the monitoring network, spatial, temporal and physical features are extracted for the monitoring points within the network. The spatial adjacency and temporal causal relationships of each monitoring point are analyzed to obtain the spatial-temporal correlation between each monitoring point. With the monitoring points as nodes and the correlation relationships as edges, a feature analysis map of the current construction stage is constructed. The spatial-temporal correlation includes the composite correlation strength and causal direction.

[0022] Specifically, in the construction environment of an energy storage power station, there are complex mutual influence relationships between various monitoring points. First, the continuously collected multimodal monitoring data such as temperature, vibration, and displacement are preprocessed, including data cleaning, outlier removal, and missing value imputation. Then, spatiotemporal alignment is performed by synchronizing GPS time with BIM coordinate mapping, unifying the data collected by different sensors to the same time base, and forming a multimodal monitoring data time series with precise timestamps. Based on these time-series data, a three-dimensional feature vector is constructed for each monitoring point: Spatial features are generated by extracting the three-dimensional coordinates (x, y, z) of the monitoring point and using the Delaunay triangulation algorithm to calculate the topological connectivity with adjacent monitoring points, producing a feature vector containing location information, number of adjacent points, average distance, and spatial density. For example, monitoring point A in the battery compartment forms a mechanical conduction triangle with three foundation points. Temporal features are extracted by performing statistical analysis and frequency domain transformation on the time series of monitoring data, extracting temporal evolution characteristics such as mean, variance, trend slope, dominant frequency, or periodic intensity. Physical features are constructed based on the physical properties of the monitoring parameters. For example, for temperature monitoring data, physical features include thermal properties such as current temperature value, rate of temperature rise, or thermal conductivity coefficient; for vibration monitoring data, physical features include dynamic properties such as amplitude, frequency, peak acceleration, or vibration energy density. This three-dimensional feature design achieves a comprehensive characterization of the spatial location, temporal evolution, and physical nature of the monitoring points, providing a rich foundation of feature information for subsequent correlation analysis.

[0023] Furthermore, a feature analysis map of the current construction phase is constructed, including: The monitoring data of the monitoring network is continuously collected, and the multimodal monitoring data of each monitoring point is preprocessed and spatiotemporally aligned to obtain a time series of multimodal monitoring data with timestamps. Based on the time series of multimodal monitoring data and their corresponding monitoring points, spatial, temporal, and physical features of each monitoring point are constructed, including: Spatial features are generated through Delaunay triangulation to establish adjacency relationships and spatial influence weights between monitoring points; Temporal features are used to characterize the patterns of change in monitoring data over time, and at least include any one of the features extracted through a sliding window: mean, variance, and mutation rate. Physical characteristics are used to characterize the physical properties of monitoring data, including at least dimensions, units, and safety thresholds; Granger causality test was performed on the time series of multimodal monitoring data from each monitoring point to analyze the driving and response relationships of the time series of the same physical quantity or physically related physical quantities among different monitoring points. The temporal causal relationship between monitoring points was extracted and fused with the spatial adjacency relationship between monitoring points to generate a spatial-temporal joint correlation relationship. Based on a graph database, with monitoring points as nodes, whose attributes include spatial coordinates, three-dimensional features, and timestamps, and with spatial-temporal joint associations as edges, a feature analysis graph of the current construction environment is constructed.

[0024] Specifically, the Delaunay triangulation method is used to analyze the spatial adjacency relationships between monitoring points. This method can automatically identify the optimal spatial adjacency connections, avoiding the subjectivity of manually setting adjacency thresholds, and ensuring that each monitoring point is connected to its most relevant neighbor in space. The generated triangular mesh can accurately reflect the spatial topology of the energy storage power station construction site. Simultaneously, Granger causality tests are used to analyze the temporal causal relationships between monitoring data. This method can identify whether the historical data of one monitoring point has predictive power for the current data of another monitoring point, thereby discovering the temporal causal transmission paths between monitoring points. For example, Delaunay triangulation is first used to establish a spatial mechanical transmission network of monitoring points. For instance, during the battery compartment hoisting stage, 32 monitoring points distributed at the four corners of the compartment and the foundation are triangularly meshed, automatically identifying the effective transmission paths between adjacent monitoring points and calculating the spatial transmission intensity of each edge. Furthermore, Granger causality tests are used to analyze temporal correlations. For example, it was found that the settlement data of foundation monitoring point A (p < 0.01) can predict the tilt angle change of compartment monitoring point B two hours later, thus establishing directional temporal causal edges. Ultimately, 3D feature fusion is achieved through the node-edge structure of the graph database: each node stores structured data such as "Node ID: VIB_003, spatial coordinates [35.2, 18.7, 2.1], temporal features {trend: 0.12 mm / h, period: 24h}, physical features {vibration energy: 1.2g}", while edge relationships record quantitative correlation indicators such as "spatial conduction intensity 0.83, temporal causal intensity 0.76". When constructing the feature analysis graph based on the graph database, each node stores complete information of a monitoring point, including spatial feature vector, temporal feature vector, physical feature vector, and timestamp. Edges between nodes represent correlation relationships and carry composite correlation strength as weights. This graph structure can intuitively express the complex correlation network between monitoring points in the energy storage power station construction environment, providing a structured knowledge representation for subsequent risk propagation analysis and intelligent decision-making. Compared with traditional table or matrix storage methods, the graph database can efficiently perform complex query operations such as graph traversal, path finding, and subgraph matching, making it more suitable for handling correlation analysis tasks in the monitoring network.

[0025] Furthermore, the fusion generates a joint spatial-temporal relationship, including: The spatial coordinates of monitoring points within the monitoring network are obtained, a set of spatial coordinates is generated, and the Delaunay triangulation method is used to divide the set of spatial coordinates to generate a set of triangular elements. Obtain the technological characteristics of the current construction phase, including the main operations performed in this construction phase and the dominant physical disturbance modes they cause. Based on the technological characteristics, query the corresponding maximum effective conduction distance from the preset technological-spatial attenuation parameter table. Traverse all triangular units in the triangular unit set. Calculate the Euclidean distance between the two monitoring points connected by each edge to obtain the edge length. Combined with the maximum effective transmission distance, calculate the spatial influence weight of each edge. Define edges with a spatial influence weight > 0 as effective transmission paths. The formula for calculating the spatial influence weight is as follows: ; Obtain the key risk patterns based on the main work content of the current construction stage, determine the target modes to be analyzed according to the key risk patterns, and extract the time series of monitoring data of the target modes for a pair of monitoring points connected on each effective transmission path, and perform pairwise Granger causality tests. Based on the F-statistic obtained from the Granger causality test, and under the condition of satisfying the preset significance level, it is normalized to obtain the temporal causal strength of each effective path, and the causal direction is recorded. Calculate the composite correlation strength of each effective transmission path. Composite correlation strength = spatial influence weight × temporal causal strength. The composite correlation strength and causal direction are used together as the spatial-temporal joint correlation of each effective transmission path.

[0026] Specifically, firstly, the three-dimensional spatial coordinates of all activated monitoring points in the current construction phase are extracted to form a coordinate set. For example, the coordinate set of 32 monitoring points in the battery compartment installation phase is P={(x1, y1, z1), (x2, y2, z2), ..., (x 32 y 32 , z 32 The Delaunay triangulation algorithm is used to perform planar projection triangulation on the coordinate set, projecting the three-dimensional coordinates onto the two-dimensional plane of the construction surface. Triangular meshing is performed by maximizing the minimum interior angle of the triangle, generating a set of triangular elements A = {A1, A2, ..., An}, where each triangular element A... i It consists of three monitoring points, such as A1=(P1, P5, P... 12 ) indicates that the monitoring points are P1, P5, and P 12The resulting triangular mesh is composed of triangles. This triangulation process ensures that the generated triangular mesh has the following characteristics: no four monitoring points are concyclic, that is, the circumcircle of each triangle does not contain other monitoring points, thus guaranteeing the uniqueness and optimality of the triangulation. The generated set of triangular elements covers the convex hull of the entire monitoring area. Each triangle edge represents a potential spatial association path between two monitoring points. For example, the three edges of triangular element A1 are (P1, P5), (P5, P... 12 ), (P 12 These edges (P1) form the basic path network for subsequent calculations of spatial mechanical transmission intensity, ensuring that each monitoring point is connected to its most relevant neighboring monitoring point in space through the shortest path. The maximum effective transmission distance is determined based on the technological characteristics of the current construction stage. These characteristics include parameters such as the type of construction operation, the properties of the components involved, and the scope of the construction impact. Construction operation types include, but are not limited to: foundation excavation, concrete pouring, steel structure hoisting, electromechanical equipment installation, battery module assembly, high-voltage cable laying, and system commissioning. The corresponding dominant physical disturbance types include soil vibration, structural impact, heat conduction, mechanical vibration, structural stress, or electromagnetic interference. The maximum effective transmission distance is determined based on this environmental information because the propagation range of physical impacts varies significantly across different construction stages. For example, the battery compartment installation stage uses precision hoisting technology, and the resulting vibrations and stresses are mainly transmitted through structural connections, resulting in a relatively limited impact range. Based on engineering experience and mechanical analysis, the maximum effective transmission distance is determined to be 15 meters. In contrast, the foundation construction stage uses large-scale mechanical excavation and heavy-duty pouring technology, and the resulting dynamic loads and deformations are propagated through the soil and structural foundation, resulting in a wider impact range. Therefore, the maximum effective transmission distance is set at 25 meters. These conduction distance parameters are pre-set in the system's process-space attenuation parameter table based on engineering practice experience and mechanical propagation theory during energy storage power station construction. The system automatically calls the corresponding conduction distance parameters based on the current construction stage identifier obtained from BIM. Next, the system iterates through all triangular elements generated by Delaunay triangulation, calculating the side length and spatial influence weight for each triangular element's three sides. Taking triangular element A1 as an example, its three sides (P1, P5), (P5, P... 12 ), (P 12 The side lengths of P1 and P2 are calculated as follows: d 512 and d 121Then, a linear decay function is used to calculate the spatial influence weight of each edge. When the calculation result is greater than 0, the edge is defined as an effective transmission path and its weight value is recorded. When the calculation result is equal to 0, it indicates that the edge length exceeds the effective range of physical influence and does not constitute an effective spatial correlation. In this way, the system filters and quantifies the edges of all triangular units, ultimately forming a spatial correlation network composed of effective transmission paths. Each path carries a corresponding spatial influence weight, providing spatial constraints for subsequent time causal analysis. All effective transmission paths are traversed, and a pair of monitoring points connected by each path is extracted. Since each monitoring point may simultaneously collect multiple modal data such as temperature, vibration, and pressure, and the Granger causality test requires input of the same type of time series data, it is necessary to determine the target mode of analysis based on the key risk type of the current construction stage. For example, in the battery installation stage, the main focus is on thermal runaway risk, so the temperature mode is selected for causal analysis; in the structural construction stage, the main focus is on deformation risk, so the displacement mode is selected for causal analysis. After determining the mode of interest, the system extracts data of the same mode from the multimodal time series of each pair of monitoring points, and then implements the specific process of the Granger causality test: Obtain the pair of monitoring points connected to each valid transmission path, P q and P g P q The time series of target modal monitoring data at the monitoring points is y t P g The corresponding time series of the monitoring points is x t , q, g=(1,2,...,N), q≠g; Models are built for different lag orders (typically from 1 to 10) using the Akaike Information Criterion (AIC) or the Bayesian Information Criterion (BIC), and the information criterion values ​​are calculated. The lag order that minimizes AIC or BIC is selected as the optimal k value. Then, two autoregressive models are built. The constrained model is as follows: y t-i i = 1, 2, ..., k Among them, y t For P q The target modal monitoring value at the monitoring point at time t, where c is a constant term. Let y represent the autoregressive coefficient of the i-th lag term. t-i For P q The i-th lag term of the monitoring point, The term represents the residual, and k is the optimal lag order. The unrestricted model is: , where y t For P q The target modal monitoring value at the monitoring point at time t, where c is a constant term. Let y represent the autoregressive coefficient of the i-th lag term. t-i For P q The i-th lag term of the monitoring point, Let be the cross-regression coefficient of the jj-th lag term. For P g The historical target modal monitoring value at time tj at the monitoring point. Here, k represents the residual term, and k is the optimal lag order. The model parameters are then estimated using the least squares method, and the residual sums of squares, RSS1 and RSS2, of the two models are calculated. The F-statistic is then calculated as F = [(RSS1 - RSS2) / RSS2] × [(n - 2k) / k], where n is the number of samples in the target modality monitoring data at a significance level. Find the critical value Fi of the F-distribution with degrees of freedom (k, n-2k) by taking a value of 0.05. 临界 When the calculated F-statistic is greater than F 临界 At that time, the null hypothesis was rejected, and it was concluded that P q For P g If Granger causality exists, then the null hypothesis is accepted; otherwise, the null hypothesis is P. q For P g There is no Granger causality. The temporal causal strength is calculated based on the F-statistic. The formula C=F / (F+n-2k) maps the F-value to the interval [0,1] as a causal strength index, where n-2k is the denominator degrees of freedom of the F-statistic distribution. When the F-statistic is significant, the causal direction is recorded as P. q →P g Simultaneously, reverse test P g For P q By establishing causal relationships, the bidirectional temporal causal strength and directionality of each effective transmission path are obtained, providing a time-dimensional quantitative indicator for subsequent calculations of composite correlation strength.

[0027] Multiple sampling granularity levels and their sampling frequencies are preset. The monitoring data of each mode acquired at each monitoring point are downsampled separately. The information loss function at each granularity is calculated. The information loss function at each granularity is compared with the preset feature importance level threshold. According to the priority-progressive decision-making process, the recommended sampling granularity for each monitoring mode is determined.

[0028] Specifically, in the construction environment of energy storage power stations, the importance of different monitoring features for risk identification varies significantly. For example, battery temperature features have a much higher predictive ability for thermal runaway risk than ambient humidity features. Therefore, different monitoring granularities are required for data collection and processing to construct a sensitivity scoring model. This model can quantify the impact of each feature on the accuracy of risk identification at different granularity levels. Finally, the granularity level that maximizes the sensitivity score for each feature is selected as the recommended configuration, ensuring that important features receive high-precision monitoring resources, while secondary features use appropriate monitoring granularity. This maximizes the performance of the subsequent risk assessment model within the constraints of limited computing and storage resources. This granularity optimization strategy avoids the "one-size-fits-all" resource allocation method in traditional approaches, achieving precise allocation of monitoring resources and laying a data quality foundation for accurate risk index assessment.

[0029] Furthermore, the recommended sampling granularity for each monitoring modality is determined, including: The sampling hertz of three granularity levels—fine, standard, and coarse—is preset. For each modal monitoring data acquired at each monitoring point, downsampling is performed according to the three sampling hertz levels to obtain downsampled data at the three granularities. The variance of each downsampled data and the variance of the corresponding acquired monitoring data are calculated. Based on the two variances, the information loss rate at each granularity level is calculated. For each importance level, a first threshold and a second threshold are set. The information loss rate of each modality monitoring data is compared with the corresponding importance level threshold to determine the recommended sampling granularity. If the information loss rate at the coarse granularity level is less than the first threshold, then the coarse granularity level is recommended. Otherwise, if the information loss rate at the standard granularity is less than the second threshold, then the standard granularity is recommended. Otherwise, fine-grained granularity is recommended, where the first threshold is less than the second threshold.

[0030] Specifically, this step is a decision-making process with progressive priorities. Under the premise of meeting accuracy requirements, the lowest possible sampling frequency is used to save resources. The process starts with the most resource-efficient "coarse granularity". If the information loss rate of the coarse granularity is small enough (< the first threshold), then coarse granularity can be used and is selected directly; otherwise, the "standard granularity" is tried. If the information loss rate of the standard granularity is within an acceptable range (< the second threshold), then the standard granularity is selected; otherwise, the "fine granularity" must be used. Three granularity levels are preset with specific sampling frequency parameters: fine granularity corresponds to a 1Hz sampling frequency, standard granularity corresponds to a 0.5Hz sampling frequency, and coarse granularity corresponds to a 0.1Hz sampling frequency. Then, granularity sensitivity tests are performed on the monitoring data of each modality at each monitoring point in the feature analysis map. Specifically, the original monitoring data is downsampled to different degrees according to the preset sampling frequency. For example, standard granularity retains every 2 data points to simulate the effect of 0.5Hz, and coarse granularity retains every 10 data points to simulate the effect of 0.1Hz. Based on the downsampled data, the variance value of each feature at different granularities is recalculated. The information loss is evaluated by comparing the change in the variance of the coarsened feature with that of the original feature. The calculation formula is that the information loss rate is equal to 1 minus the ratio of the variance of the coarsened feature to the variance of the original feature. The system sets judgment thresholds based on feature importance levels: a first threshold of 20% and a second threshold of 50%. For moderately important features, the first threshold is set to 10% and the second to 30%, while for low-importance features, the first threshold is set to 20% and the second to 50%. The calculated information loss rate is compared with the preset importance level thresholds. When the loss rate is less than the first threshold, it indicates that the feature is not sensitive to granularity changes, and a coarse granularity can be selected to save computational resources. When the loss rate is between the two thresholds, a standard granularity is selected to balance effect and efficiency. When the loss rate is greater than the second threshold, it indicates that the feature is sensitive to granularity changes, and a fine granularity must be selected to ensure information integrity. This ultimately forms a personalized granularity configuration scheme for each monitoring modality. Each personalized granularity configuration scheme for each monitoring modality includes feature identifier, monitoring point location, importance level, recommended granularity level (fine / standard / coarse), corresponding sampling frequency parameters, and decision basis information, forming a complete monitoring data-granularity configuration mapping relationship.

[0031] Based on the recommended sampling granularity for each monitoring modality, new multimodal monitoring data are continuously collected in the monitoring network of the current construction environment. The acquired data is enhanced, and the composite correlation strength of the three-dimensional features and edges of the feature parsing graph nodes is updated to achieve dynamic evolution of the graph.

[0032] Specifically, based on the personalized granularity configuration scheme for each feature, the system performs precise enhancement processing on the multimodal monitoring data collected at the energy storage power station construction site. For example, the system reads the feature identifier, recommended granularity level, and corresponding sampling frequency parameters from the granularity configuration scheme, and then sends configuration instructions to the relevant sensors in the monitoring network to adjust their sampling frequency, data precision, and storage strategy. For instance, the recommended fine-grained temperature feature is adjusted to 1Hz high-frequency sampling and 16-bit precision processing, while the recommended coarse-grained illumination feature is adjusted to 0.1Hz low-frequency sampling and 8-bit precision processing, thus achieving a differentiated data acquisition strategy. The system continuously collects multimodal monitoring data that has undergone granularity optimization.

[0033] Furthermore, the feature is characterized by updating the composite association strength between the three-dimensional features and edges of the feature parsing graph nodes to achieve dynamic evolution of the graph, including: For newly collected multimodal monitoring data, differentiated data augmentation strategies are employed to improve its characterization quality. For the recommended fine-grained sampling, a fine-grained local enhancement strategy is adopted. Based on the sampled data, a sliding window is used to extract local dynamic trends, and linear or spline interpolation algorithms are combined to fill in possible short-term gaps, further improving the temporal resolution and signal continuity. For the recommended standard granularity sampling, a hybrid granularity hierarchical enhancement strategy is adopted to perform time-domain filtering on the collected data and combine it with the trend changes of neighboring monitoring points to identify and correct outliers that deviate significantly. For the recommended coarse-grained sampling, a coarse-grained statistical enhancement strategy is adopted to perform mean or variance statistics on low-frequency sampling data, and to apply exponential smoothing or Kalman filtering to suppress noise fluctuations and preserve long-term trend characteristics. The new multimodal monitoring data after data augmentation is processed by spatiotemporal synchronization. Based on the augmented data after spatiotemporal synchronization, the three-dimensional features of each node are re-extracted, and the spatial transmission strength and temporal causality strength between nodes are recalculated. The enhanced monitoring data after spatiotemporal synchronization is fed back into the feature parsing map, the three-dimensional features of each node are re-extracted, and the composite correlation strength between the edges between nodes is recalculated. The updated 3D features of the nodes and the composite correlation strength between edges between nodes are injected into the feature parsing graph to complete the dynamic evolution of the graph and characterize the current state evolution trend of the construction environment.

[0034] Specifically, based on the obtained feature granularity configuration results, differentiated data augmentation strategies are adopted for features of different importance. For high-importance features with fine-grained configuration, the system employs a fine-grained local augmentation strategy. This involves setting a small sliding window (e.g., 5-10 data points) to perform a local scan on the original monitoring data. Cubic spline interpolation or Lagrange interpolation algorithms are used to generate denser interpolation points between adjacent data points, enhancing the original 1Hz sampling data to a high-density time series of 2-5Hz. Simultaneously, bilinear interpolation is used to improve the spatial coordinate positioning accuracy, ensuring that important features receive the most complete information representation. For medium-importance features with standard granularity configuration, the system adopts a hybrid granularity hierarchical augmentation strategy. First, a larger sliding window (e.g., 20-30 data points) is used to calculate local statistical features, including mean, variance, and trend. Then, local interpolation techniques are used for fine-grained processing during critical time periods, and statistical smoothing methods are used for noise reduction during stable time periods. This achieves an organic combination of fine-grained and statistical augmentation, balancing data quality and computational efficiency. For low-importance features with coarse-grained configurations, the system employs a coarse-grained statistical enhancement strategy. This involves segmenting and aggregating the original data using a large window (e.g., 50-100 data points), calculating representative statistical values ​​such as the median and quartiles for each time period, and using moving averages or Gaussian smoothing filters to eliminate high-frequency noise. This compresses the data to a lower temporal resolution while preserving the main trends, saving computational resources while ensuring that secondary features still provide effective monitoring information. The monitoring data after granular enhancement shows significant changes in both quality and accuracy. Important features gain higher temporal and spatial resolution, while secondary features benefit from reduced noise interference through statistical smoothing. These improved data must be fed back into the feature analysis map to be effective; otherwise, the map remains based on the original coarse data and cannot reflect the optimization effect.

[0035] After completing the differential enhancement and spatiotemporal synchronization of multimodal monitoring data, the system initiates the dynamic evolution of the feature analysis map. First, the spatial coordinates of all monitoring points are acquired, and a spatial topology is constructed using Delaunay triangulation to generate a set of triangular elements, ensuring a reasonable geometric distribution. Considering the technological characteristics of the current construction stage (such as concrete pouring, prestressing tensioning, etc.) and the resulting dominant physical disturbance modes, the system queries the corresponding maximum effective transmission distance from the preset "technology-space attenuation parameter table." All edges of the triangular elements are traversed, the Euclidean distance between their connecting nodes is calculated, and the spatial influence weight is evaluated based on the maximum effective transmission distance. Edges with a spatial influence weight greater than zero are selected as effective transmission paths. Based on the preset key risk modes of the current construction stage, the target physical modes to be analyzed (such as strain, temperature, etc.) are determined. For each pair of nodes on an effective path, the monitoring data time series of the target mode within the latest time window is extracted, and Granger causality tests are performed. Under the condition of meeting the preset significance level, the test results are normalized to obtain the temporal causal strength of each path, and the causal direction is recorded. Subsequently, the composite correlation strength of each effective transmission path is calculated. This strength is the product of the spatial influence weight and the temporal causal strength, comprehensively reflecting the joint correlation characteristics of the path in terms of spatial accessibility and temporal driving force. Simultaneously, based on the synchronized enhanced data, the spatial location, temporal state, and physical quantity values ​​of each node are re-extracted to form a three-dimensional feature vector. The updated node features, the composite correlation strength of the effective paths, and the causal direction are injected into the feature parsing graph, replacing the original attributes and completing the graph state update. By periodically executing this process, the graph continuously evolves, dynamically representing the disturbance propagation paths and risk evolution trends in the construction environment, achieving accurate perception and interpretable modeling of the structural state.

[0036] Based on the evolved feature parsing map, global topological features, critical path features, and critical node features are extracted, input into a pre-trained risk assessment model, and output a continuous risk index of the current construction environment. When the risk index exceeds a preset safety threshold, a graded early warning response is triggered.

[0037] Specifically, the construction environment of energy storage power stations is complex and ever-changing, with risk factors interacting with each other. Traditional single-data-source assessment methods are prone to misjudgment or omission. By integrating the accuracy of real-time monitoring data and the correlation of feature maps, the accuracy of risk identification and the reliability of risk index calculation can be significantly improved, enabling precise quantitative assessment of the safety status of the construction environment.

[0038] Furthermore, the continuous risk index of the current construction environment is output, including: The evolved feature analysis graph state is periodically acquired, and global topological features, critical path features, and critical node features are extracted based on graph structure analysis methods. The global topological features include node risk density, mean and variance of associated edge strength, and network clustering coefficient. The critical path features are the top K paths with the highest composite association strength in the graph, and the critical node features are the trend of node feature values ​​on the critical path. The global topological features, critical path features, and critical node features are combined into an input vector, which is then input into a pre-trained gradient boosting tree model to output a continuous risk index. The risk index is compared with the preset safety threshold in real time, and a graded alarm protocol is triggered when the risk index exceeds the threshold.

[0039] Specifically, this step quantifies the risk level of the current construction environment. The specific steps are: dynamic feature extraction of the graph, periodically (at preset time intervals or triggered by events) generating snapshots of the dynamically evolved feature analysis graph, and extracting the following dynamic evolution feature sequences: global topological features, including node risk density (calculating the proportion of nodes with feature values ​​exceeding their physical safety thresholds to the total number of nodes), mean and variance of associated edge strength (calculating the average value and fluctuation of the "spatial-temporal composite association strength" on all edges; a decrease in mean or an increase in variance may indicate system instability), network clustering coefficient (measuring the degree of clustering between monitoring points; an abnormally high clustering coefficient may indicate local risk aggregation), and critical path features (identifying the top K paths (e.g., TOP5) in the graph, ranked from highest to lowest composite association strength, as critical risk transmission paths, and calculating the changing trend of association strength on these paths (e.g., the slope obtained through linear fitting), where K > 0. Negative slopes indicate a weakening or failure of the transmission relationship, while large positive slopes indicate an accelerating accumulation of risk. Node characteristic evolution involves tracking the first-order differences (changes) of characteristic values ​​over time for nodes on the critical path, forming a change sequence. Statistics (such as mean and standard deviation) of this sequence are calculated to determine if the changes are stable. Node characteristic values ​​refer to specific attributes or metrics possessed by each node. These can be physical quantities directly monitored from the construction environment or indicators calculated based on these physical quantities. The specific meaning of node characteristic values ​​may differ in different application scenarios. Physical quantity monitoring values ​​include data directly obtained from sensors, such as temperature, humidity, stress, and strain. Derived indicators are indicators calculated based on the original monitoring data, such as rates of change (e.g., the rate of change of stress) and cumulative damage indicators (calculated based on fatigue damage theory). The changing trends of critical node characteristic values ​​can be obtained by calculating the first-order differences (i.e., the differences between adjacent time points) of these node characteristic values, and further analyzing the statistical characteristics (such as mean and standard deviation) of these change sequences to determine whether they are stable or exhibit abnormal fluctuations. This method helps to identify potential risk concentration areas or impending failure points in a timely manner. The above dynamic features are input into a trained risk rating model (such as Gradient Boosting Tree (GBDT)), which outputs a continuous risk index R (range 0-100). An example of the model's input feature vector is: Input Feature Vector = [Node Risk Density (0.35), Mean Edge Strength (0.72), Variance Edge Strength (0.18), Clustering Coefficient (0.41), Critical Path 1 Strength Trend (-0.05), ..., Standard Deviation of Settlement Change Rate of Critical Node A (0.12)].The risk rating model is a pre-trained gradient boosting tree model. Based on historical construction project data, multiple feature parsing graph snapshots are constructed. The dynamic evolution features corresponding to each snapshot are extracted as input, and a risk index associated with the time distance of risk events is used as a label. Supervised learning is performed using the gradient boosting tree algorithm to obtain a pre-trained model for online assessment of the current construction environment risk level. The specific process is as follows: Historical construction data is acquired by collecting data from multiple completed historical engineering projects. Each data item includes: original time-series observations of each monitoring point throughout the construction process (such as physical quantities like temperature, stress, displacement, and vibration), risk events recorded in the construction log (such as structural cracking, equipment overheating, and support instability) and their occurrence time and severity level, as well as corresponding construction stage information; a historical feature parsing graph containing three-dimensional feature vectors (spatial, temporal, and physical) of nodes and dynamically related edges is constructed; the historical feature vectors are extracted as input to the model, and a corresponding risk index label Rt is generated for each graph snapshot time according to the occurrence time of historical risk events. The value range is [0, 100], and the calculation formula is: [Formula omitted]. Where Δt represents the time interval from the current time t to the occurrence of the next risk event, and λ is the decay coefficient, ranging from (0.1, 1.0), used to control the risk accumulation rate, preferably 0.5. A training sample set is formed by combining the input feature vectors of all historical projects with their corresponding risk index labels, and randomly divided into training and test sets. Supervised learning training is performed using gradient boosting tree algorithms (such as XGBoost or LightGBM), aiming to minimize the mean squared error between the predicted risk index and the true label. Model hyperparameters are optimized through cross-validation, and model performance is evaluated on the test set to ensure the coefficient of determination R0 is accurate. 2 The accuracy is >0.85, and the mean absolute error (MAE) is less than 8. The parameters of the trained and validated GBDT model are serialized and saved (e.g., as a .pkl or .json file) and integrated into the online monitoring system as a "pre-trained risk rating model." The overall risk index of the current construction environment of the energy storage power station is calculated. This index is a continuous value between 0 and 100, with higher values ​​indicating greater construction environment risk. It is compared in real-time with a safety threshold preset based on the energy storage power station construction safety specifications. Multiple threshold levels are typically set, such as a low-risk threshold of 25, a medium-risk threshold of 50, and a high-risk threshold of 75. When the calculated risk index exceeds the corresponding threshold, the system immediately triggers a graded alarm of the corresponding level. Finally, alarm information including specific risk locations and risk levels is output to construction management personnel through various methods such as audible and visual alarms, SMS notifications, and system interface pop-ups, ensuring timely and effective response and handling of construction safety risks.

[0040] Example 2, based on the same inventive concept as the intelligent monitoring method for the construction environment of energy storage power stations in the foregoing examples, such as... Figure 2As shown, this application provides an intelligent monitoring system for the construction environment of an energy storage power station, including: The monitoring network construction module is used to obtain the current construction stage of the energy storage power station from the BIM system, match the multimodal monitoring data required for monitoring from the construction environment description of the current construction stage, activate the sensors on the relevant monitoring points, and form a monitoring network for the current construction environment. The graph construction module is used to extract spatial, temporal, and physical features from monitoring points within the monitoring network based on the data monitored by the monitoring network. It analyzes the spatial adjacency and temporal causal relationships of each monitoring point to obtain the spatial-temporal correlation between each monitoring point. With monitoring points as nodes and correlation relationships as edges, it constructs a feature analysis graph of the current construction stage. The spatial-temporal correlation includes the composite correlation strength and causal direction. The granularity configuration module is used to preset multiple sampling granularity levels and their sampling frequencies. It performs downsampling on each type of modal monitoring data acquired at each monitoring point, calculates the information loss function at each granularity, compares the information loss function at each granularity with the preset feature importance level threshold, and determines the recommended sampling granularity for each monitoring modality according to a priority-based decision-making process. The graph dynamic evolution module is used to continue collecting new multimodal monitoring data in the current construction environment monitoring network according to the sampling granularity recommended for each monitoring mode, enhance the acquired data, update the composite correlation strength of the three-dimensional features and edges of the feature parsing graph nodes, and realize the dynamic evolution of the graph. The construction environment risk module is used to extract global topological features, critical path features, and critical node features based on the evolved feature analysis map, input them into a pre-trained risk assessment model, and output a continuous risk index of the current construction environment. When the risk index exceeds a preset safety threshold, a graded early warning response is triggered.

[0041] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0042] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0043] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for intelligent monitoring of the construction environment of an energy storage power station, characterized in that, include: The current construction phase of the energy storage power station is obtained from the BIM system. The multimodal monitoring data required for monitoring is matched from the construction environment description of the current construction phase, and the sensors on the relevant monitoring points are activated to form a monitoring network for the current construction environment. Based on the data monitored by the monitoring network, spatial, temporal and physical features are extracted for the monitoring points within the monitoring network. The spatial adjacency and temporal causal relationships of each monitoring point are analyzed to obtain the spatial-temporal correlation between each monitoring point. With the monitoring points as nodes and the correlation relationships as edges, a feature analysis map of the current construction stage is constructed. The spatial-temporal correlation includes the composite correlation strength and causal direction. Multiple sampling granularity levels and their sampling frequencies are preset. The monitoring data of each mode acquired at each monitoring point are downsampled separately. The information loss function at each granularity is calculated. The information loss function at each granularity is compared with the preset feature importance level threshold. According to the priority-progressive decision-making process, the recommended sampling granularity for each monitoring mode is determined. Based on the recommended sampling granularity for each monitoring modality, new multimodal monitoring data will continue to be collected in the monitoring network of the current construction environment. The acquired data will be enhanced, and the composite correlation strength of the three-dimensional features and edges of the feature parsing graph nodes will be updated to realize the dynamic evolution of the graph. Based on the evolved feature parsing map, global topological features, critical path features, and critical node features are extracted, input into a pre-trained risk assessment model, and output a continuous risk index of the current construction environment. When the risk index exceeds a preset safety threshold, a graded early warning response is triggered.

2. The intelligent monitoring method for the construction environment of an energy storage power station according to claim 1, characterized in that, Match the required multimodal monitoring data from the current construction environment description, including: Obtain a description of the current construction environment of the energy storage power station in text and / or language, analyze the current environment description using NLP technology, and identify the key monitoring needs of the current construction environment; Based on the key monitoring needs of the current construction environment, a pre-constructed sensor capability matrix and monitoring point information database are retrieved. The optimal sensor combination is matched through a greedy algorithm, and the sensors on the relevant monitoring points are activated to form a monitoring network for the current construction environment. The multimodal monitoring data required for monitoring is obtained. The sensor capability matrix is ​​a structured data table used to provide sensor technical specifications and performance parameters, and the monitoring point information database is used to provide the spatial location, coverage area and equipment access information of the monitoring points.

3. The intelligent monitoring method for the construction environment of an energy storage power station according to claim 1, characterized in that, Construct a feature analysis map of the current construction phase, including: The monitoring data of the monitoring network is continuously collected, and the multimodal monitoring data of each monitoring point is preprocessed and spatiotemporally aligned to obtain a time series of multimodal monitoring data with timestamps. Based on the time series of multimodal monitoring data and their corresponding monitoring points, spatial, temporal, and physical features of each monitoring point are constructed, including: Spatial features are generated through Delaunay triangulation to establish adjacency relationships and spatial influence weights between monitoring points; Temporal features are used to characterize the patterns of change in monitoring data over time, and at least include any one of the features extracted through a sliding window: mean, variance, and mutation rate. Physical characteristics are used to characterize the physical properties of monitoring data, including at least dimensions, units, and safety thresholds; Granger causality test was performed on the time series of multimodal monitoring data from each monitoring point to analyze the driving and response relationships of the time series of the same physical quantity or physically related physical quantities among different monitoring points. The temporal causal relationship between monitoring points was extracted and fused with the spatial adjacency relationship between monitoring points to generate a spatial-temporal joint correlation relationship. Based on a graph database, with monitoring points as nodes, whose attributes include spatial coordinates, three-dimensional features, and timestamps, and with spatial-temporal joint associations as edges, a feature analysis graph of the current construction environment is constructed.

4. The intelligent monitoring method for the construction environment of an energy storage power station according to claim 3, characterized in that, The fusion generates spatial-temporal joint relationships, including: The spatial coordinates of monitoring points within the monitoring network are obtained, a set of spatial coordinates is generated, and the Delaunay triangulation method is used to divide the set of spatial coordinates to generate a set of triangular elements. Obtain the technological characteristics of the current construction phase, including the main operations performed in this construction phase and the dominant physical disturbance modes they cause. Based on the technological characteristics, query the corresponding maximum effective conduction distance from the preset technological-spatial attenuation parameter table. Traverse all triangular units in the triangular unit set. Calculate the Euclidean distance between the two monitoring points connected by each edge to obtain the edge length. Combined with the maximum effective transmission distance, calculate the spatial influence weight of each edge. Define edges with a spatial influence weight > 0 as effective transmission paths. The formula for calculating the spatial influence weight is as follows: ； Obtain the key risk patterns based on the main work content of the current construction stage, determine the target modes to be analyzed according to the key risk patterns, and extract the time series of monitoring data of the target modes for a pair of monitoring points connected on each effective transmission path, and perform pairwise Granger causality tests. Based on the F-statistic obtained from the Granger causality test, and under the condition of satisfying the preset significance level, it is normalized to obtain the temporal causal strength of each effective path, and the causal direction is recorded. Calculate the composite correlation strength of each effective transmission path. Composite correlation strength = spatial influence weight × temporal causal strength. The composite correlation strength and causal direction are used together as the spatial-temporal joint correlation of each effective transmission path.

5. The intelligent monitoring method for the construction environment of an energy storage power station according to claim 1, characterized in that, Determine the recommended sampling granularity for each monitoring modality, including: The sampling hertz of three granularity levels—fine, standard, and coarse—is preset. For each modal monitoring data acquired at each monitoring point, downsampling is performed according to the three sampling hertz levels to obtain downsampled data at the three granularities. The variance of each downsampled data and the variance of the corresponding acquired monitoring data are calculated. Based on the two variances, the information loss rate at each granularity level is calculated. For each importance level, a first threshold and a second threshold are set. The information loss rate of each modality monitoring data is compared with the corresponding importance level threshold to determine the recommended sampling granularity. If the information loss rate at the coarse granularity level is less than the first threshold, then the coarse granularity level is recommended. Otherwise, if the information loss rate at the standard granularity is less than the second threshold, then the standard granularity is recommended. Otherwise, fine-grained granularity is recommended, where the first threshold is less than the second threshold.

6. The intelligent monitoring method for the construction environment of an energy storage power station according to claim 5, characterized in that, Update the composite association strength of the three-dimensional features and edges of the feature parsing graph nodes to achieve dynamic evolution of the graph, including: For newly collected multimodal monitoring data, differentiated data augmentation strategies are employed to improve its characterization quality. For the recommended fine-grained sampling, a fine-grained local enhancement strategy is adopted. Based on the sampled data, a sliding window is used to extract local dynamic trends, and linear or spline interpolation algorithms are combined to fill in possible short-term gaps, further improving the temporal resolution and signal continuity. For the recommended standard granularity sampling, a hybrid granularity hierarchical enhancement strategy is adopted to perform time-domain filtering on the collected data and combine it with the trend changes of neighboring monitoring points to identify and correct outliers that deviate significantly. For the recommended coarse-grained sampling, a coarse-grained statistical enhancement strategy is adopted to perform mean or variance statistics on low-frequency sampling data, and to apply exponential smoothing or Kalman filtering to suppress noise fluctuations and preserve long-term trend characteristics. The new multimodal monitoring data after data augmentation strategy is spatiotemporally synchronized. Based on the augmented data after spatiotemporal synchronization, the three-dimensional features of each node are re-extracted, and the spatial transmission strength and temporal causality strength between nodes are recalculated. The enhanced monitoring data after spatiotemporal synchronization is fed back into the feature parsing map, the three-dimensional features of each node are re-extracted, and the composite correlation strength between the edges between nodes is recalculated. The updated 3D features of the nodes and the composite correlation strength between edges between nodes are injected into the feature parsing graph to complete the dynamic evolution of the graph and characterize the current state evolution trend of the construction environment.

7. The intelligent monitoring method for the construction environment of an energy storage power station according to claim 6, characterized in that, Conduct a construction environment risk index assessment, including: The evolved feature analysis graph state is periodically acquired, and global topological features, critical path features, and critical node features are extracted based on graph structure analysis methods. The global topological features include node risk density, mean and variance of associated edge strength, and network clustering coefficient. The critical path features are the top K paths with the highest composite association strength in the graph, and the critical node features are the trend of node feature values ​​on the critical path. The global topological features, critical path features, and critical node features are combined into an input vector, which is then input into a pre-trained gradient boosting tree model to output a continuous risk index. The risk index is compared with the preset safety threshold in real time, and a graded alarm protocol is triggered when the risk index exceeds the threshold.

8. An intelligent monitoring system for the construction environment of an energy storage power station, characterized in that, include: The monitoring network construction module is used to obtain the current construction stage of the energy storage power station from the BIM system, match the multimodal monitoring data required for monitoring from the construction environment description of the current construction stage, activate the sensors on the relevant monitoring points, and form a monitoring network for the current construction environment. The graph construction module is used to extract spatial, temporal, and physical features from monitoring points within the monitoring network based on the data monitored by the monitoring network. It analyzes the spatial adjacency and temporal causal relationships of each monitoring point to obtain the spatial-temporal correlation between each monitoring point. With monitoring points as nodes and correlation relationships as edges, it constructs a feature analysis graph of the current construction stage. The spatial-temporal correlation includes the composite correlation strength and causal direction. The granularity configuration module is used to preset multiple sampling granularity levels and their sampling frequencies. It performs downsampling on each type of modal monitoring data acquired at each monitoring point, calculates the information loss function at each granularity, compares the information loss function at each granularity with the preset feature importance level threshold, and determines the recommended sampling granularity for each monitoring modality according to a priority-based decision-making process. The graph dynamic evolution module is used to continue collecting new multimodal monitoring data in the current construction environment monitoring network according to the sampling granularity recommended for each monitoring mode, enhance the acquired data, update the composite correlation strength of the three-dimensional features and edges of the feature parsing graph nodes, and realize the dynamic evolution of the graph. The construction environment risk module is used to extract global topological features, critical path features, and critical node features based on the evolved feature analysis map, input them into a pre-trained risk assessment model, and output a continuous risk index of the current construction environment. When the risk index exceeds a preset safety threshold, a graded early warning response is triggered.

Images