Construction environment mutation risk field prediction method fusing geological and meteorological data

Abstract

The application discloses a construction environment mutation risk field prediction method fusing geological meteorological data, and belongs to the technical field of construction engineering risk prediction. The method comprises the following steps: obtaining geological data, meteorological data and construction progress data of a construction area, and generating graph data containing multi-modal features after structurally processing the multi-modal heterogeneous data. The application adopts a spatio-temporal adaptive graph neural network prediction model based on an attention mechanism, can dynamically calculate and distribute the weights of different features according to real-time environment and construction data, and can capture the spatio-temporal propagation relationship of risks between geographical nodes by using the message passing mechanism of a graph. The method can deeply fuse data of different modes, accurately predict the mutation risk of the construction environment, trigger corresponding early warnings according to the risk grade, improve the accuracy and timeliness of risk prediction, and provide scientific support for intelligent decision-making of building engineering.

Description

Technical Field

[0001] This invention relates to the field of construction engineering risk prediction technology, and in particular to a method for predicting the risk field of sudden changes in the construction environment by integrating geological and meteorological data. Background Technology

[0002] In the field of construction engineering, predicting safety risks in the construction environment is a crucial step in ensuring the safety of personnel and property and the smooth progress of projects. With the development of information technology, intelligent risk early warning using data-driven computational models by integrating multi-source data such as geological exploration, meteorological monitoring, and construction progress has become an important development direction. These methods aim to identify potential sudden risks, such as landslides and collapses, in advance by analyzing historical and real-time data.

[0003] In existing technologies, some risk prediction methods attempt to utilize multiple models to process different types of data. For example, one model analyzes geological data to assess the inherent stability of a slope, while another independent model analyzes meteorological data to predict the impact of heavy rainfall events. Finally, the outputs of the two models are simply weighted and fused or logically judged to derive the final risk level. This multi-model cascading processing flow is a common implementation method in current technologies.

[0004] However, the aforementioned existing technical solutions have significant drawbacks. First, processing data from different sources through independent models ignores the complex nonlinear interactions between these data modes, leading to a one-sided understanding of the causes of risk. Second, these methods typically employ fixed model parameters and analytical logic, failing to dynamically adjust the prediction focus based on changes in the construction phase or abrupt changes in environmental conditions. Furthermore, most existing prediction models focus on the risk status of a single monitoring point, making it difficult to effectively simulate the spatial correlation and transmission process of risk throughout the entire construction area. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a method for predicting construction environment abrupt change risk fields by integrating geological and meteorological data. It employs a spatiotemporal adaptive graph neural network model, which can dynamically integrate multi-source heterogeneous data and simulate the spatiotemporal propagation of risks, thereby improving the accuracy and foresight of the prediction.

[0006] The above objectives can be achieved through the following approach:

[0007] A method for predicting the risk field of sudden changes in the construction environment by integrating geological and meteorological data includes: acquiring geological data, meteorological data and construction progress data of the construction area, and performing structured processing on these data;

[0008] Based on the structured data, the construction area is divided into several geographical nodes, and edges between geographical nodes are constructed according to geospatial relationships and geological correlations to generate graph data containing multimodal features.

[0009] The features of different modalities in the graph data are fused and encoded to generate a unified feature vector;

[0010] The unified feature vector is input into the spatiotemporal adaptive graph neural network prediction model. Through the attention mechanism inside the spatiotemporal adaptive graph neural network prediction model, the weights of different modal features are dynamically calculated and allocated based on real-time construction progress data and meteorological data.

[0011] The message passing mechanism of the graph is used to capture the spatiotemporal propagation relationship of risk between geographical nodes and output the mutation risk value of each node.

[0012] A mutation risk field map of the construction environment is generated based on the mutation risk value, and corresponding risk warnings are triggered according to the risk level.

[0013] Optionally, obtaining geological data, meteorological data, and construction progress data of the construction area includes: obtaining geological data through geological sensors deployed at the construction site; obtaining meteorological data through automatic weather stations; and obtaining construction progress data by accessing a construction plan database.

[0014] Optionally, the construction of edges between geographic nodes based on geospatial relationships and geological correlations includes: identifying geographically adjacent geographic nodes and establishing a first type of edge between adjacent geographic nodes; identifying geographic nodes located within the same geological structural unit and establishing a second type of edge between geographic nodes located within the same geological structural unit; and merging the first type of edge and the second type of edge to form a complete edge set of graph data.

[0015] Optionally, fusing and encoding features of different modalities in the graph data to generate a unified feature vector includes: processing the geological data using a convolutional neural network to extract spatial features and generate a geological feature vector; processing the meteorological data and the construction progress data using a recurrent neural network to extract temporal features and generate a temporal feature vector; and performing feature cross-fusion of the geological feature vector and the temporal feature vector to generate a unified feature vector.

[0016] Optionally, the feature cross-fusion of the geological feature vector and the temporal feature vector includes: concatenating the geological feature vector and the temporal feature vector to form a combined vector; performing a linear transformation on the combined vector, and applying a nonlinear activation function to the result of the linear transformation to generate a unified feature vector.

[0017] Optionally, the unified feature vector is input into the spatiotemporal adaptive graph neural network prediction model. The attention mechanism within the spatiotemporal adaptive graph neural network prediction model dynamically calculates and assigns weights to different modal features based on real-time construction progress data and meteorological data. This includes: encoding the construction progress data and meteorological data into a context query vector; calculating the attention weights of each modal feature based on the context query vector; using the attention weights to weight the corresponding modal features in the unified feature vector to generate a weighted feature vector; and using the weighted feature vector to capture the spatiotemporal propagation relationship of risk between geographical nodes.

[0018] Optionally, calculating the attention weights of each modality feature based on the context query vector includes: inputting the context query vector into a multilayer perceptron to calculate the original weight scores of each modality feature; and applying a softmax function to normalize the original weight scores to generate attention weights.

[0019] Optionally, the step of using the graph message passing mechanism to capture the spatiotemporal propagation relationship of risk among geographic nodes and outputting the mutation risk value of each node includes: for each geographic node in the graph, aggregating the feature information of neighboring nodes to form an aggregated information vector; combining the feature information of the geographic node itself with the aggregated information vector to update the feature representation of the geographic node; and iteratively executing the aggregation and update steps to simulate the multi-step propagation of risk information in the graph structure.

[0020] Optionally, generating a mutation risk field map of the construction environment based on the mutation risk value and triggering corresponding risk warnings according to the risk level includes: comparing the mutation risk value of each node with a multi-level risk threshold system to determine the risk level of each node; generating a structured warning signal based on the risk level; and sending the warning signal to the construction management system to trigger a response action.

[0021] Based on the same inventive concept, this invention also provides a construction environment abrupt change risk field prediction system that integrates geological and meteorological data. The system includes: a data acquisition module for acquiring geological data, meteorological data, and construction progress data of the construction area, and performing structured processing on these data; a graph construction module for dividing the construction area into several geographical nodes based on the structured data, and constructing edges between geographical nodes according to geospatial relationships and geological correlations to generate graph data containing multimodal features; a feature fusion module for fusing and encoding features of different modalities in the graph data to generate a unified feature vector; a prediction module for inputting the unified feature vector into a spatiotemporal adaptive graph neural network prediction model, dynamically calculating and allocating weights of different modal features based on real-time construction progress data and meteorological data through the attention mechanism within the spatiotemporal adaptive graph neural network prediction model, and capturing the spatiotemporal propagation relationship of risk between geographical nodes using the graph message passing mechanism, and outputting the abrupt change risk value of each node; and an output module for generating an abrupt change risk field map of the construction environment based on the abrupt change risk value, and triggering corresponding risk warnings according to the risk level.

[0022] Compared with the prior art, the present invention has the following advantages:

[0023] 1. This invention achieves deep fusion of multi-source heterogeneous data from geology, meteorology, and construction by constructing a unified spatiotemporal adaptive graph neural network model. This method abandons the simple feature splicing or multi-model concatenation of traditional techniques, and can uncover the inherent coupling relationships between different modalities from the data's underlying layers, thereby providing a more comprehensive understanding of the complex causes of sudden risks and improving the accuracy of risk identification.

[0024] 2. The attention mechanism introduced in this invention endows the prediction model with dynamic adaptive capabilities. The model can autonomously adjust its focus on different data sources based on real-time changes in construction stages and environmental conditions, concentrating its computational focus on the most critical risk drivers. This intelligent analysis mode makes risk prediction more timely and targeted, especially enhancing its ability to capture sudden and transient risks.

[0025] 3. This invention leverages the inherent topological advantages of graph neural networks to effectively simulate the spatial propagation path of risks. It not only predicts the risk level at individual locations but also reveals the transmission effects and scope of impact of risks within the construction site. This elevates risk management from discrete, single-point monitoring to continuous, area-wide situational awareness, providing a scientific basis for global disaster prevention and mitigation decision-making and enhancing the practicality and foresight of the prediction results.

[0026] 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

[0027] 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.

[0028] Figure 1 This is a flowchart illustrating a method for predicting the risk field of sudden changes in the construction environment that integrates geological and meteorological data, according to an embodiment of the present invention.

[0029] Figure 2 This is a graph showing the trend of risk values ​​at different nodes over time in an embodiment of the present invention.

[0030] Figure 3 This is a heatmap of different node mutation risk values ​​according to an embodiment of the present invention.

[0031] Figure 4 This is a schematic diagram of the structure of a construction environment sudden change risk field prediction system that integrates geological and meteorological data according to an embodiment of the present invention. Detailed Implementation

[0032] 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.

[0033] Reference Figure 1One embodiment of this invention proposes a method for predicting construction environment abrupt change risk fields by integrating geological and meteorological data. It employs a spatiotemporal adaptive graph neural network model, capable of dynamically fusing multi-source heterogeneous data and simulating the spatial propagation of risks. This method abandons the simple feature splicing or multi-model concatenation methods of traditional techniques. Instead, it constructs a unified graph neural network model to uncover the inherent coupling relationships between different modalities from the data's underlying layers, thereby comprehensively understanding the complex causes of sudden risks. The attention mechanism introduced in this invention endows the prediction model with dynamic adaptive capabilities, enabling it to autonomously adjust the degree of attention to different data sources according to real-time changes in construction stages and environmental conditions. This intelligent analysis mode makes risk prediction more timely and targeted, especially enhancing the ability to capture sudden and instantaneous risks. By utilizing the topological advantages of graph neural networks, this invention can not only predict the risk level of a single location but also reveal the transmission effect and impact range of risks within the construction site, elevating risk management from discrete single-point monitoring to continuous field situational awareness.

[0034] The method described in this embodiment specifically includes:

[0035] Obtain geological data, meteorological data, and construction progress data of the construction area, and then perform structured processing on this data;

[0036] Based on the structured data, the construction area is divided into several geographical nodes, and edges between geographical nodes are constructed according to geospatial relationships and geological correlations to generate graph data containing multimodal features.

[0037] The features of different modalities in the graph data are fused and encoded to generate a unified feature vector;

[0038] The unified feature vector is input into the spatiotemporal adaptive graph neural network prediction model. Through the attention mechanism inside the spatiotemporal adaptive graph neural network prediction model, the weights of different modal features are dynamically calculated and allocated based on real-time construction progress data and meteorological data.

[0039] The message passing mechanism of the graph is used to capture the spatiotemporal propagation relationship of risk between geographical nodes and output the mutation risk value of each node.

[0040] A mutation risk field map of the construction environment is generated based on the mutation risk value, and corresponding risk warnings are triggered according to the risk level.

[0041] Optionally, acquiring geological data, meteorological data, and construction progress data of the construction area includes:

[0042] Geological data is acquired through geological sensors deployed at the construction site;

[0043] Specifically, geological data is acquired from geological sensors. This does not refer to static geological exploration reports, but rather to data acquisition through a series of automated monitoring sensors pre-deployed at key locations in the construction area, such as slopes, foundation pits, and tunnel faces. These geological sensors include, but are not limited to, inclinometers for monitoring slope displacement, pore water pressure gauges for measuring pore water pressure within the soil, and strain gauges for monitoring structural strain. These sensors periodically and automatically upload measurement data via Internet of Things (IoT) technology, forming a dynamic geological data stream reflecting the real-time mechanical response of the geological body under external disturbances.

[0044] Meteorological data is obtained through automatic weather stations;

[0045] Specifically, real-time meteorological data is obtained from weather stations. To ensure the timeliness and relevance of the data, an automatic weather station is typically deployed at or near the construction site. The weather station continuously monitors and records key meteorological parameters, primarily including rainfall, rainfall intensity, wind speed, wind direction, temperature, and humidity. Data is collected at a very high frequency, enabling it to capture sudden meteorological events such as short-duration heavy rainfall, and is transmitted in real-time via a network interface.

[0046] Obtain construction progress data by accessing the construction plan database.

[0047] Specifically, construction progress data is obtained from a construction plan database. The data source is typically a digital management platform for the project, such as a system integrated with Building Information Modeling (BIM) or specialized project management software. This database is queried periodically via an interface to extract information on ongoing construction activities, such as the location and depth of excavation work, the timing and amount of explosives used in blasting operations, and the distribution and movement trajectories of heavy machinery. This data represents dynamic human factors that impact the environment. Figure 2 As shown, the mutation risk values ​​of different geographical nodes change over a specific time period. The line graph clearly depicts the dynamic process of risk value accumulation over time, especially the rapid upward trend of high-risk nodes, intuitively demonstrating the real-time monitoring and prediction capabilities of the method of this invention.

[0048] Optionally, the construction of edges between geographic nodes based on geospatial relationships and geological correlations includes:

[0049] Identify geographically adjacent nodes and establish first-class edges between adjacent geographical nodes;

[0050] Specifically, edges between adjacent nodes are defined based on geographic adjacency. After the construction area is divided into gridded geographic nodes, all node pairs are traversed. If two geographic nodes are spatially directly adjacent—that is, they share a boundary or vertex—or the Euclidean distance between them is less than a preset threshold, an edge is established between them. This edge represents a direct physical connection and is used to simulate the local diffusion effect of risks, such as surface water runoff or small-scale soil slippage.

[0051] Identify geographic nodes located within the same geological structural unit, and establish second-type edges between geographic nodes located within the same geological structural unit;

[0052] Specifically, edges are defined between non-adjacent but geologically continuous nodes based on geological correlation. This step goes beyond simple spatial proximity and aims to capture long-distance risk transmission mediated by subsurface geological structures. First, the input geological exploration data and geological maps are analyzed to identify macroscopic geological units running through the construction area, such as the same fault zone, the same continuous weak interlayer, or the same aquifer. Then, it is determined which geographic nodes reside on the same key geological unit. If two or more geographic nodes, even if they are not geographically adjacent, are located within the same identified geological unit, an edge is established between them. This edge represents a geomechanical or hydrological intrinsic connection.

[0053] Merge the first type of edges and the second type of edges to form the complete edge set of the graph data.

[0054] Specifically, all edges defined using these two methods are merged to form a complete edge set for the graph data. The resulting graph structure not only includes geographical neighborhood relationships but also establishes "shortcuts" through geologically related edges, enabling the graph neural network to learn and simulate nonlocal risk propagation patterns. For example, when a node at one end of a fault zone experiences subsidence, the model can directly transmit the impact to a distant node at the other end of the fault zone via this geologically related edge. This dual edge definition mechanism allows the model to simultaneously capture both the gradual diffusion and sudden long-distance transmission of risk, thereby improving its ability to predict systemic and cascading risks under complex geological conditions and making risk field predictions more closely reflect physical reality.

[0055] Optionally, fusing and encoding features of different modalities in the graph data to generate a unified feature vector includes:

[0056] The geological data is processed using a convolutional neural network to extract spatial features and generate geological feature vectors.

[0057] Specifically, geological data, typically existing in the form of geological maps or exploration point distribution maps, possesses strong spatial structure characteristics. Therefore, geological data is processed into two-dimensional or three-dimensional gridded data. Convolutional neural networks (CNNs) are then used to process this gridded data. Through unique convolutional and pooling layers, CNNs can automatically learn and extract local spatial patterns and structural information from geological data, such as the strike of rock fault zones and the location and extent of weak soil layers. Ultimately, the output is a geological feature vector that characterizes the static geological environment stability of the construction area.

[0058] A recurrent neural network is used to process the meteorological data and the construction progress data to extract time-series features and generate a time-series feature vector.

[0059] Specifically, meteorological data and construction progress data are both typical time-series data, valuable for revealing dynamic trends. Therefore, recurrent neural networks or their variants, such as long short-term memory networks, are used to encode this type of time-series data. Through their internal recurrent structure, recurrent neural networks can effectively capture the dependencies of data over time, such as the cumulative effect of continuous rainfall and the temporal impact of specific construction activities, thereby generating a time-series feature vector that summarizes the dynamic changes in the environment and construction.

[0060] The geological feature vector and the temporal feature vector are cross-fused to generate a unified feature vector.

[0061] Specifically, after obtaining the geological feature vector representing static spatial information and the temporal feature vector representing dynamic temporal information, a feature cross-fusion step was designed to achieve deep fusion of the two types of information rather than simple concatenation. The geological feature vector and the temporal feature vector are concatenated to form a combined vector. To enable the model to learn the nonlinear interaction between the two modal features, the combined vector is input into one or more fully connected layers for transformation. This process is expressed by the following formula:

[0062] ;

[0063] in, It is the final unified feature vector; Represents geological feature vectors; Represents the temporal feature vector; The function represents concatenating the beginning and end of two vectors; and These represent the weight matrix and bias vector of the fully connected layer, respectively. They are parameters that the model automatically optimizes by learning from the data during training. It is a non-linear activation function, such as ReLU. This process uses a learnable weight matrix. Mapping features from different sources to a unified feature space effectively avoids the problem of direct computation due to differences in data dimensions and physical meanings. It also enables the model to autonomously discover and quantify complex coupling relationships, such as "what intensity of rainfall" acting on "what type of geological structure" to trigger high-risk events. The resulting unified feature vector provides comprehensive and rich input for subsequent graph neural network models, improving the accuracy of risk prediction and sensitivity to sudden events.

[0064] Optionally, the feature cross-fusion of the geological feature vector and the temporal feature vector includes:

[0065] The geological feature vector and the temporal feature vector are concatenated to form a combined vector;

[0066] Specifically, the geological feature vector extracted by the convolutional neural network and the temporal feature vector extracted by the recurrent neural network are concatenated into a higher-dimensional combined vector through a concatenation operation. This concatenation operation physically places features representing the static geological environment and features representing the dynamic environment and construction activities in the same mathematical space, laying the foundation for subsequent interactive learning.

[0067] The combined vectors are linearly transformed, and a nonlinear activation function is applied to the result of the linear transformation to generate a unified feature vector.

[0068] Specifically, this concatenated combined vector is input into a pre-defined fully connected layer. A fully connected layer is a fundamental structure in neural networks, with a learnable weight matrix at its core. In this layer, each element of the combined vector undergoes matrix multiplication with the weight matrix, meaning that each dimension of the geological feature vector interacts with each dimension of the temporal feature vector. During training, the model automatically adjusts the parameters of the weight matrix through backpropagation, thereby learning meaningful intrinsic correlation patterns between the two modal features. After the linear transformation by the fully connected layer, the output is immediately fed into a nonlinear activation function for processing. Nonlinear activation functions, such as the Corrected Linear Unit (ReLU), are key to enabling deep learning models to learn complex nonlinear relationships. In construction risk prediction scenarios, many risk triggering mechanisms are highly nonlinear. For example, the relationship between increased rainfall and landslide risk is not a simple linear one, but may have a critical threshold. By introducing a nonlinear activation function, the model can effectively fit this complex nonlinear dependency, thus more accurately characterizing the causes of risk.

[0069] Optionally, the unified feature vector is input into the spatiotemporal adaptive graph neural network prediction model. Through the attention mechanism within the spatiotemporal adaptive graph neural network prediction model, the weights of different modal features are dynamically calculated and allocated based on real-time construction progress data and meteorological data, including:

[0070] Encode construction progress data and meteorological data into context query vectors;

[0071] Specifically, attention weights for different modal features need to be calculated. The model encodes construction phase data and real-time environmental data, which serve as the current spatiotemporal context, to form a contextual query vector. Construction phase data, such as foundation excavation or main structure construction, can be encoded as a category vector; real-time environmental data, such as instantaneous wind speed or cumulative rainfall, can be encoded as a numerical vector. The contextual query vector is then used to evaluate the importance of each modal component in the uniform feature vector.

[0072] Calculate the attention weights of each modality feature based on the context query vector;

[0073] Specifically, to achieve this assessment, the model calculates the correlation score between each modal feature and the current context using a small feedforward neural network. The correlation score reflects the potential contribution of a data modality, such as geology or meteorology, to the risk within a specific construction and environmental context. The scores of all modalities are normalized using a softmax function to generate a set of attention weights that sum to one.

[0074] The attention weights are used to weight the corresponding modal features in the unified feature vector to generate a weighted feature vector, which is then used to capture the spatiotemporal propagation relationship of risk between geographical nodes.

[0075] Specifically, after obtaining the attention weights, the model applies these weights to a unified feature vector. Each modal feature component corresponding to a geographic node in the unified feature vector is multiplied by its corresponding attention weight. This weighting operation dynamically adjusts the contribution of each modal feature in subsequent calculations. For example, during periods of continuous heavy rainfall, the model adaptively assigns higher attention weights to features in meteorological and geological data, thereby amplifying their impact on the final risk prediction. The attention-weighted feature vector serves as input to the graph neural network's message passing mechanism. During message passing, each geographic node aggregates the weighted feature information from its neighboring nodes and updates it in conjunction with its own information. Because the transmitted information has been attention-weighted, the risk propagation simulation focuses more on the most critical risk factors. For example, geological instability features assigned high weights to upstream nodes due to heavy rainfall are more effectively transmitted to downstream nodes, thus accurately simulating the propagation path of debris flow risk.

[0076] Optionally, calculating the attention weights for each modality feature based on the context query vector includes:

[0077] The context query vector is input into a multilayer perceptron to calculate the original weight scores of each modality feature;

[0078] Specifically, the digital description of the current situation involves encoding the current construction phase information and real-time environmental data separately. Construction phase encoding transforms textual descriptions such as "foundation excavation" and "main structure construction" into numerical vectors that the model can understand. Environmental data encoding integrates real-time monitored parameters such as rainfall and wind speed into numerical vectors. These two encoded vectors are then concatenated to form a unified context query vector, which comprehensively describes the "here and now" working conditions and environment. This context query vector is input into a pre-defined multilayer perceptron. A multilayer perceptron is a feedforward neural network that, through internal multilayer nonlinear transformations, can learn and fit extremely complex relationships between inputs and outputs. In this invention, it is trained to learn expert knowledge about "which data modality is more important in which context." The output of the multilayer perceptron is a set of raw, unnormalized weight scores, each corresponding to a data modality such as geology or meteorology, with the numerical value reflecting the initial importance of that modality in the current context.

[0079] The original weight scores are normalized using the softmax function to generate attention weights.

[0080] Specifically, to transform these raw weight scores into a set of normalized values ​​that can be directly used as weights, the system applies the softmax function. The softmax function maps an arbitrary set of real numbers to a probability distribution whose sum is 1. The calculation process is as follows:

[0081] ;

[0082] in, It is the first The final attention weights obtained from each modality feature; It is the correlation score calculated between the modality feature and the current context; This is the exponential sum of all modal correlation scores, used for normalization. Here... It is calculated by a small neural network that takes the current context query vector and the corresponding modality features as input, and its parameters are learned during model training. Through this function, the original weight scores are converted into relative importance, with higher scores receiving higher attention weights, and the sum of all weights is exactly 1.

[0083] Optionally, the step of using the graph messaging mechanism to capture the spatiotemporal propagation relationship of risk among geographical nodes and outputting the mutation risk value of each node includes:

[0084] For each geographical node in the graph, the feature information of the neighboring nodes is aggregated to form an aggregated information vector;

[0085] Specifically, the message passing mechanism in this invention is the core engine for simulating the spatial propagation of risk in graph neural networks. The execution process consists of two closely linked steps: aggregation and updating. These two steps are iterated multiple times, with each iteration representing a step forward in the graph structure to transmit risk information. In the aggregation step, each geographical node in the graph actively collects feature information from all its direct neighbors. Here, "neighbor" is defined based on geographical adjacency and geological relevance. Each node receives not the original features, but a feature vector weighted by an attention mechanism, meaning the transmitted information itself already contains an assessment of importance within the current spatiotemporal context. Aggregation operations are typically performed using permutation-invariant functions, such as summing, averaging, or maximizing the feature vectors of all neighboring nodes, thereby summarizing the discrete information from multiple neighbors into a single aggregated information vector that summarizes the neighborhood state. This process is represented as:

[0086] ;

[0087] in, This is the target node The generated aggregated information vector; Represents aggregate functions; It is a node The set of neighboring nodes; Neighboring nodes The feature vectors have already been attention-weighted.

[0088] The feature representation of the geographic node is updated by combining its own feature information with the aggregated information vector.

[0089] Specifically, in the update step, each node updates its own feature representation using the newly generated aggregated information vector. This process fuses the node's original feature information with the aggregated information vector from its neighborhood. This fusion is typically achieved through a small neural network layer, which concatenates the node's original feature vector and the aggregated information vector and then performs a non-linear transformation. The formula is expressed as:

[0090] ;

[0091] in, It is a node The updated feature vector; It is a node Feature vectors before update; It is a learnable update function, such as a fully connected layer with a non-linear activation function. For example... Figure 3 As shown, the mutation risk values ​​of different geographical nodes at different times are displayed. The shades of color represent the magnitude of the risk value, enabling managers to easily identify areas of concentrated risk and grasp the spatial distribution of risk.

[0092] The aggregation and update steps are executed iteratively to simulate the multi-step propagation of risk information in the graph structure.

[0093] Specifically, through a single aggregation and update, each node's new feature vector incorporates information from its first-order neighborhood. After multiple iterations, a node's feature vector will contain information from multiple-order neighborhoods, meaning information propagates over long distances on the graph. The message passing mechanism transforms static, discrete node risk assessment into dynamic, continuous risk field evolution simulation. This allows the model to explicitly capture the transmission effects of risk, such as how the rainfall saturation state of upstream nodes gradually affects the stability of downstream nodes through the graph edges, thereby achieving a global and forward-looking prediction of the risk situation across the entire construction area.

[0094] Optionally, a mutation risk field map of the construction environment is generated based on the mutation risk value, and corresponding risk warnings are triggered according to the risk level, including:

[0095] The mutation risk value of each node is compared with a multi-level risk threshold system to determine the risk level of each node;

[0096] Specifically, the step in triggering risk warning in this invention involves converting the quantitative risk prediction output by the model into practically operable safety management instructions. After the spatiotemporal adaptive graph neural network prediction model completes its calculations, it outputs a real-time, continuous sudden risk value for each geographical node within the construction area. This value is a quantitative assessment of the likelihood of a sudden risk occurring at that node within a future period. The core is comparing the sudden risk value with a preset risk threshold. The preset risk threshold is a tiered threshold system jointly developed based on safety management regulations, engineering geological conditions, and historical disaster data, corresponding to different safety levels such as attention, warning, and alert. The sudden risk value of each node is compared step-by-step with this threshold system in real time.

[0097] A structured early warning signal is generated based on the risk level;

[0098] Specifically, when the mutation risk value of a node exceeds a preset risk threshold for a certain level, the corresponding early warning logic will be immediately triggered, automatically generating a structured early warning signal. This early warning signal is a data packet containing key information, such as the geographic node number that triggered the early warning, the current mutation risk value, the triggered early warning level, the timestamp, and the main risk contributing factors that may be identified by the model's attention mechanism, such as whether it is continuous rainfall or a specific construction activity.

[0099] The warning signal is sent to the construction management system to trigger a response action.

[0100] Specifically, the generated warning signal is immediately sent to the construction management system via a standard application programming interface (API) or message queue. Upon receiving the signal, the construction management system executes preset response procedures, such as highlighting the risk area on the electronic sand table or GIS interface, updating the sudden risk field map, and pushing specific and clear warning information to relevant responsible persons such as project managers and safety officers via mobile applications, SMS, or on-site audible and visual alarms.

[0101] Based on the same inventive concept, such as Figure 4 As shown, the present invention also provides a construction environment abrupt change risk field prediction system that integrates geological and meteorological data, the system comprising:

[0102] The data acquisition module is used to acquire geological data, meteorological data, and construction progress data of the construction area, and to perform structured processing on this data;

[0103] The graph construction module is used to divide the construction area into several geographic nodes based on the structured data, and construct the edges between the geographic nodes according to the geospatial relationships and geological correlations, thereby generating graph data containing multimodal features.

[0104] The feature fusion module is used to fuse and encode features of different modalities in the graph data to generate a unified feature vector;

[0105] The prediction module is used to input the unified feature vector into the spatiotemporal adaptive graph neural network prediction model. Through the attention mechanism inside the spatiotemporal adaptive graph neural network prediction model, it dynamically calculates and allocates the weights of different modal features based on real-time construction progress data and meteorological data. It also uses the graph message passing mechanism to capture the spatiotemporal propagation relationship of risk between geographical nodes and outputs the mutation risk value of each node.

[0106] The output module is used to generate a mutation risk field map of the construction environment based on the mutation risk value, and trigger corresponding risk warnings according to the risk level.

[0107] To verify the feasibility of this invention in practice, it was applied to a high slope excavation project for a mountainous highway. The geological conditions in the construction area of ​​this project are complex, including multiple fault zones and weak interlayers. Moreover, it is located in a rainy region, making it highly susceptible to sudden geological disasters such as landslides and collapses due to the coupled effect of heavy rainfall and excavation activities during construction.

[0108] Traditional safety management methods rely on periodic manual inspections and static geological survey reports, making it difficult to predict sudden risks in real time and dynamically. This project aims to use the method of this invention to integrate multi-source heterogeneous data from the site, enabling dynamic prediction and real-time early warning of abrupt changes in the construction environment risk field.

[0109] In this embodiment, the project team first deployed geological sensors such as inclinometers and pore water pressure gauges at key locations in the construction area according to the data acquisition steps of this invention to obtain dynamic geological data; an automatic weather station was set up on site to collect meteorological data such as rainfall and wind speed in real time; and it was connected to the project's BIM management platform to obtain construction progress data such as excavation and blasting. Subsequently, the construction area was divided into multiple 10m × 10m geographical nodes, and a graph structure containing two types of edges was constructed based on geographical adjacency relationships and the direction of the identified geological fault zones.

[0110] To verify the beneficial effects of this invention, a typical operating condition during the rainy season was selected for testing and analysis. During this period, a prediction system using the method of this invention was used as the experimental group, and a monitoring system based on traditional threshold alarms was used as the control group, operating in parallel.

[0111] In the feature fusion and encoding stage, a convolutional neural network was used to process the gridded geological exploration map, automatically extracting spatial features such as fault zone location and stratum dip, generating a geological feature vector. Simultaneously, a long short-term memory network was used to process 48 consecutive hours of rainfall data and construction excavation progress data, capturing cumulative effects and temporal influences, generating a temporal feature vector. Subsequently, the formula was used... By performing feature cross-fusion, the two types of feature vectors are input into a fully connected layer and processed by the ReLU nonlinear activation function to generate a unified feature vector that can characterize the complex coupling relationship between geology, meteorology and construction.

[0112] On a certain day, a new round of blasting and excavation work commenced at the site, while a short-duration heavy rainfall was detected by the weather station. Upon receiving this information, the spatiotemporal adaptive graph neural network prediction model of this invention activates its internal attention mechanism. The model encodes the construction phase information of "blasting and excavation" and real-time rainfall data into a context query vector. Based on this vector, the model calculates, using a multilayer perceptron and softmax function, that the attention weights of meteorological and geological data should be significantly increased in the current context. For example, the weight of meteorological features increases from 0.2 to 0.5, while the weight of geological features related to slope stability increases from 0.4 to 0.45. Subsequently, the attention-weighted feature vectors are used for message passing within the graph neural network. Nodes located upstream of the blasting point where pore water pressure rises sharply have their high-risk characteristics effectively passed to downstream adjacent nodes through graph aggregation and update steps. Meanwhile, since a node numbered A-15 near the blasting point is located on a known geological fault zone, the model directly transmits the risk information caused by the blasting disturbance of this node to another node numbered C-08, which is also located on the fault zone but geographically far away, through a preset geological association edge.

[0113] Ultimately, the model output showed that the mutation risk values ​​of downstream areas B-15 to B-20 and the distant node C-08 of A-15 increased sharply, exceeding the preset "alarm" level threshold. A structured early warning signal was immediately generated, and the aforementioned areas were marked in red on the on-site GIS electronic sand table through the construction management system. An early warning message was also pushed to the project safety manager's mobile phone: "Areas B-15 to B-20 and area C-08 have an extremely high risk of landslides. The main risk contributing factors are heavy rainfall and blasting disturbance. Please evacuate personnel and equipment immediately." In contrast, the control group only issued a general "orange rainstorm warning for the entire site," failing to identify specific areas of concentrated risk. Approximately three hours after the incident, a small-scale landslide did occur in area B-18. Due to the timely warning, on-site personnel and equipment had been evacuated in advance, avoiding losses.

[0114] Table 1 Comparison of Key Node Data Monitoring and Risk Prediction on a Certain Day

[0115]

[0116] Table 1 shows that, under the same environmental and construction conditions, this invention can distinguish the risk levels of different nodes, accurately identify areas with high risk due to the coupling effect of multiple factors, and predict risk values ​​higher than other areas. In contrast, the control group can only provide indiscriminate regional early warnings.

[0117] The comparison results in Table 2 further demonstrate the technical effectiveness of this invention. This invention not only achieves "field-based perception" of risks, revealing the transmission path and scope of impact, but also provides highly targeted and operable early warning information, offering forward-looking decision support for on-site managers. This enables a shift from passive response to proactive avoidance, thereby improving the inherent safety level of construction sites.

[0118] Table 2 Performance Comparison of Early Warning Systems

[0119]

[0120] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.

[0121] 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 predicting the risk field of sudden changes in the construction environment by integrating geological and meteorological data, characterized in that, The method includes: Obtain geological data, meteorological data, and construction progress data of the construction area, and then perform structured processing on this data; Based on the structured data, the construction area is divided into several geographical nodes, and edges between geographical nodes are constructed according to geospatial relationships and geological correlations to generate graph data containing multimodal features. The features of different modalities in the graph data are fused and encoded to generate a unified feature vector. The generation of the unified feature vector includes: processing the geological data using a convolutional neural network to extract spatial features and generate a geological feature vector; processing the meteorological data and construction progress data using a recurrent neural network to extract temporal features and generate a temporal feature vector; performing feature cross-fusion of the geological feature vector and the temporal feature vector to generate a unified feature vector; performing feature cross-fusion of the geological feature vector and the temporal feature vector includes: concatenating the geological feature vector and the temporal feature vector to form a combined vector; performing a linear transformation on the combined vector and applying a nonlinear activation function to the result of the linear transformation to generate a unified feature vector; inputting the unified feature vector into a spatiotemporal adaptive graph neural network prediction model, and dynamically calculating and allocating weights for different modal features based on real-time construction progress data and meteorological data through the attention mechanism within the spatiotemporal adaptive graph neural network prediction model. The message passing mechanism of the graph is used to capture the spatiotemporal propagation relationship of risk between geographical nodes and output the mutation risk value of each node. A mutation risk field map of the construction environment is generated based on the mutation risk value, and corresponding risk warnings are triggered according to the risk level.

2. The method for predicting the risk field of sudden changes in the construction environment by integrating geological and meteorological data according to claim 1, characterized in that, The acquisition of geological data, meteorological data, and construction progress data of the construction area includes: Geological data is acquired through geological sensors deployed at the construction site; Meteorological data is obtained through automatic weather stations; Obtain construction progress data by accessing the construction plan database.

3. The method for predicting the risk field of sudden changes in the construction environment by integrating geological and meteorological data according to claim 1, characterized in that, The construction of edges between geographic nodes based on geospatial relationships and geological correlations includes: Identify geographically adjacent nodes and establish first-class edges between adjacent geographical nodes; Identify geographic nodes located within the same geological structural unit, and establish second-type edges between geographic nodes located within the same geological structural unit; Merge the first type of edges and the second type of edges to form the complete edge set of the graph data.

4. The method for predicting the risk field of sudden changes in the construction environment by integrating geological and meteorological data according to claim 1, characterized in that, The unified feature vector is input into the spatiotemporal adaptive graph neural network prediction model. Through the attention mechanism within the model, weights for different modal features are dynamically calculated and allocated based on real-time construction progress data and meteorological data, including: Encode construction progress data and meteorological data into context query vectors; Calculate the attention weights of each modality feature based on the context query vector; The attention weights are used to weight the corresponding modal features in the unified feature vector to generate a weighted feature vector, which is then used to capture the spatiotemporal propagation relationship of risk between geographical nodes.

5. The method for predicting the risk field of sudden changes in the construction environment by integrating geological and meteorological data according to claim 4, characterized in that, The attention weights for each modality feature are calculated based on the context query vector, including: The context query vector is input into a multilayer perceptron to calculate the original weight scores of each modality feature; The original weight scores are normalized using the softmax function to generate attention weights.

6. The method for predicting the risk field of sudden changes in the construction environment by integrating geological and meteorological data according to claim 1, characterized in that, The method of using graph messaging to capture the spatiotemporal propagation relationship of risk among geographical nodes and outputting the mutation risk value of each node includes: For each geographical node in the graph, the feature information of the neighboring nodes is aggregated to form an aggregated information vector; The feature representation of the geographic node is updated by combining its own feature information with the aggregated information vector. The aggregation and update steps are executed iteratively to simulate the multi-step propagation of risk information in the graph structure.

7. The method for predicting the risk field of sudden changes in the construction environment by integrating geological and meteorological data according to claim 1, characterized in that, Based on the aforementioned mutation risk value, a mutation risk field map of the construction environment is generated, and corresponding risk warnings are triggered according to the risk level, including: The mutation risk value of each node is compared with a multi-level risk threshold system to determine the risk level of each node; A structured early warning signal is generated based on the risk level; The warning signal is sent to the construction management system to trigger a response action.

8. A construction environment abrupt change risk field prediction system integrating geological and meteorological data, characterized in that, The system is used in the construction environment abrupt change risk field prediction method that integrates geological and meteorological data as described in any one of claims 1-7, the system comprising: The data acquisition module is used to acquire geological data, meteorological data, and construction progress data of the construction area, and to perform structured processing on this data; The graph construction module is used to divide the construction area into several geographic nodes based on the structured data, and construct the edges between the geographic nodes according to the geospatial relationships and geological correlations, thereby generating graph data containing multimodal features. The feature fusion module is used to fuse and encode features of different modalities in the graph data to generate a unified feature vector. The generation of the unified feature vector includes: processing the geological data using a convolutional neural network to extract spatial features and generate a geological feature vector; processing the meteorological data and the construction progress data using a recurrent neural network to extract temporal features and generate a temporal feature vector; performing feature cross-fusion on the geological feature vector and the temporal feature vector to generate a unified feature vector; and performing feature cross-fusion on the geological feature vector and the temporal feature vector to generate a unified feature vector. The prediction module is used to input the unified feature vector into the spatiotemporal adaptive graph neural network prediction model. Through the attention mechanism inside the spatiotemporal adaptive graph neural network prediction model, it dynamically calculates and allocates the weights of different modal features based on real-time construction progress data and meteorological data. It also uses the graph message passing mechanism to capture the spatiotemporal propagation relationship of risk between geographical nodes and outputs the mutation risk value of each node. The output module is used to generate a mutation risk field map of the construction environment based on the mutation risk value, and trigger corresponding risk warnings according to the risk level.

Images