Plateau mountainous area road slope geological disaster automatic identification method fusing multiple data

Abstract

The invention provides a plateau mountain area road slope geological disaster automatic identification method fusing multiple data, relates to a plateau mountain area road slope geological disaster automatic identification method fusing multiple data, and provides a plateau mountain area road slope geological disaster automatic identification method fusing multiple data. High-precision automatic recognition of geological disasters is realized through technologies of multi-modal data fusion, heterogeneous atlas construction, graph attention network and the like, multi-source data of SAR satellites, earthquake monitoring stations, meteorological monitoring stations and the like can be effectively integrated, geological disaster field knowledge is combined, a heterogeneous atlas is constructed, feature aggregation is performed by using the graph attention network, and the method is suitable for high-precision automatic recognition of geological disasters. The accuracy and comprehensiveness of feature expression are remarkably improved, the occurred geological disasters can be accurately recognized, and powerful technical support can be provided for early warning and risk prevention and control of road slope geological disasters.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster identification, and in particular to a method for automatically identifying geological disasters on highway slopes in plateau mountainous areas by fusing multiple data. Background Art

[0002] The geological disasters on highway slopes in plateau mountainous areas are characterized by suddenness and destructiveness, which seriously threaten highway traffic safety and regional economic development. Traditional slope geological disaster monitoring methods mainly rely on single data sources such as satellite images or ground sensors. Although these data sources can reflect the stability changes of slopes to a certain extent, they still have limitations and cannot fully reflect the changes in complex geological environments. 不同数据 The information between sources is isolated and there is a lack of effective fusion mechanism, which leads to the one-sidedness and incompleteness of the monitoring results. In addition, the existing methods mostly use manual analysis or simple statistical models, lack the ability to efficiently integrate and automatically process multi-source data, are inefficient and easily affected by subjective factors. Manual analysis is difficult to cope with large-scale, high-frequency monitoring data, and cannot meet the needs of real-time monitoring and early warning. Traditional statistical models are also difficult to capture the complex nonlinear relationship between multi-source data, resulting in insufficient prediction accuracy. With the development of remote sensing technology, the Internet of Things and artificial intelligence, multimodal data fusion and intelligent analysis have become important directions for geological disaster monitoring, but there is still a lack of a systematic method that can effectively integrate multi-source data, combine domain knowledge and realize automatic identification.

[0003] Therefore, it is necessary to provide an automatic identification method for geological hazards on highway slopes in plateau mountainous areas by integrating multiple data to solve the above technical problems. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides an automatic identification method for geological disasters on the slopes of plateau mountainous roads by integrating multiple data, so as to achieve the beneficial effect of accurately and quickly identifying geological disasters on the slopes of plateau mountainous roads.

[0005] The present invention provides a method for automatically identifying geological hazards on highway slopes in plateau mountainous areas by integrating multiple data. The method comprises:

[0006] S1: Obtain multimodal monitoring data of the target area of ​​highway slopes in plateau mountainous areas through various data sources;

[0007] S2: Use the pre-trained multimodal feature extraction model to extract features from multimodal monitoring data and obtain a unified high-dimensional feature vector representation of each data source;

[0008] S3: Construct heterogeneous graphs based on various data sources and geological disaster domain knowledge;

[0009] S4: The unified high-dimensional feature vector representation and heterogeneous graph of each data source are input into the pre-trained graph attention network model for feature aggregation to generate a comprehensive feature vector representation of each data source;

[0010] S5: Input the comprehensive feature vector representation of each data source into the pre-built Bayesian network model, and obtain the probability distribution of geological disaster conditions in the target area of ​​the highway slope in the plateau mountainous area through the Bayesian network model reasoning;

[0011] S6: Based on the preset decision rules, the probability distribution of geological hazard conditions in the target area of ​​the highway slope in the plateau mountainous area is converted into the geological hazard identification result in the target area of ​​the highway slope in the plateau mountainous area.

[0012] Preferably, the multiple data sources include SAR satellites, seismic monitoring stations, meteorological monitoring stations, soil moisture sensors, rock and soil displacement monitoring equipment and slope angle monitoring equipment, and the multimodal geological monitoring data include SAR image data, seismograph records, meteorological data, soil moisture change data, rock and soil displacement data and slope angle change data.

[0013] Preferably, before extracting features from the multimodal monitoring data in step S2, the method further includes performing data enhancement on the multimodal monitoring data, generating positive sample pairs, and adjusting the pre-trained multimodal feature extraction model using a contrast loss function.

[0014] Preferably, in step S2, feature extraction of multimodal monitoring data includes using a convolutional neural network to extract spatial features from SAR image data, using a long short-term memory network to extract temporal features from seismograph records and meteorological data, and using a fully connected neural network to extract features from soil moisture change data, rock and soil displacement data, and slope angle change data.

[0015] Preferably, step S3 comprises the following steps:

[0016] S301: taking each data source as a node of a heterogeneous graph;

[0017] S302: Preliminary analysis of the potential correlation between the data sources, identification of the correlation coefficients and mutual information between different data sources, and definition of the causal relationship between nodes in combination with geological disaster domain knowledge and historical disaster data to form the edges of the heterogeneous graph;

[0018] S303: Combine nodes and edges to construct a heterogeneous graph.

[0019] Preferably, redundant edges in the heterogeneous graph are checked regularly and removed through a pruning algorithm.

[0020] Preferably, in step S302, the preliminary analysis of the potential correlation between the data sources further includes using statistical methods and information theory methods to calculate the correlation and dependency between different data sources.

[0021] Preferably, step S4 comprises the following steps:

[0022] S401: Input the unified high-dimensional feature vector representation and heterogeneous graphs of each data source into the pre-trained graph attention network model;

[0023] S402: Calculate the attention weight between each node and its neighboring nodes using the attention mechanism in the pre-trained graph attention network model;

[0024] S403: Based on the attention weight and the heterogeneous graph, the features of the node and its neighboring nodes are aggregated through a multi-layer message passing mechanism to generate an aggregated feature vector representation of each node;

[0025] S404: Globally pooling the aggregated feature vector representations of all nodes to generate a comprehensive feature vector representation of each data source.

[0026] Preferably, in step S5, the parameters of the Bayesian network model are adjusted based on historical disaster data by using a maximum likelihood estimation method.

[0027] Preferably, before step S6, the method also includes quantifying the uncertainty of the probability distribution of geological hazard conditions in the target area of ​​the highway slope in the plateau mountainous area by calculating the entropy value and the confidence interval, and dynamically adjusting the threshold of the preset decision rule based on the entropy value and the confidence interval.

[0028] Compared with the related art, the method for automatically identifying geological hazards on highway slopes in plateau mountainous areas by integrating multiple data provided by the present invention has the following beneficial effects:

[0029] The present invention provides a method for automatically identifying geological hazards on the slopes of plateau mountainous highways by fusing multiple data. Through multimodal data fusion, heterogeneous graph construction, and graph attention network technologies, high-precision automatic identification of geological hazards is achieved. Based on various data sources and geological disaster field knowledge, a heterogeneous graph is constructed, each data source is used as a node, and the correlation and causal relationship between data sources are used as edges, which scientifically reflects the complex relationship between multi-source data. The present invention can also effectively integrate multi-source data such as SAR satellites, seismic monitoring stations, and meteorological monitoring stations, and combine with geological disaster field knowledge to construct a heterogeneous graph and use the graph attention network for feature aggregation, which significantly improves the accuracy and comprehensiveness of feature expression. Multimodal data fusion technology can mine the potential correlation between data and improve the accuracy of detection results. The graph attention network can automatically extract the features of multi-source data, capture complex nonlinear relationships, and achieve efficient data analysis and pattern recognition. At the same time, it can combine geological disaster field knowledge, make full use of historical disaster data and expert experience, and improve the reliability of the model. The graph attention network can also fuse global information, and through multi-layer aggregation, the feature vector of each node gradually contains more extensive context information, and better represent its position and role in the heterogeneous graph. The correlation and dependency between data sources are calculated through statistical methods and information theory methods such as Pearson correlation coefficient, mutual information and Granger causality test, and the causal relationship between nodes is defined in combination with historical disaster data to ensure the scientificity and rationality of heterogeneous graphs. In addition, through Bayesian network model reasoning and dynamic decision rule adjustment, the present invention can quantify the uncertainty of geological disasters and provide more reliable identification results. It can not only accurately identify geological disasters that have occurred, but also provide strong technical support for early warning and risk prevention and control of highway slope geological disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic flow chart of a method for automatically identifying geological hazards on highway slopes in plateau mountainous areas by integrating multiple data according to the present invention;

[0031] Figure 2 A block diagram of a computing device capable of implementing various embodiments of the present invention is shown. DETAILED DESCRIPTION

[0032] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only the parts related to the present invention, rather than all structures, are shown in the accompanying drawings. In addition, the embodiments of the present invention and the features in the embodiments may be combined with each other without conflict.

[0033] It should also be noted that, for ease of description, only the part relevant to the present invention but not all content is shown in the accompanying drawings. It should be mentioned before discussing exemplary embodiments in more detail that some exemplary embodiments are described as processing or methods depicted as flow charts. Although the flow chart describes each operation (or step) as sequential processing, many operations therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of each operation can be rearranged. When its operation is completed, the processing can be terminated, but it can also have additional steps not included in the accompanying drawings. The processing can correspond to methods, functions, procedures, subroutines, subprograms, etc.

[0034] Embodiment 1

[0035] A method for automatically identifying geological hazards on highway slopes in plateau mountainous areas by integrating multiple data. In the specific implementation process, Figure 1 As shown, it shows a schematic flow chart of a method for automatically identifying geological hazards on highway slopes in plateau mountainous areas by integrating multiple data, including:

[0036] Step S1: Acquire multimodal monitoring data of the target area of ​​the highway slope in the plateau mountainous area through various data sources.

[0037] Specifically, the multiple data sources include SAR satellites, seismic monitoring stations, meteorological monitoring stations, soil moisture sensors, rock and soil displacement monitoring equipment, and slope angle monitoring equipment; the multimodal geological monitoring data include SAR image data, seismograph records, meteorological data, soil moisture change data, rock and soil displacement data, and slope angle change data.

[0038] In the specific implementation process, remote sensing imaging of the target area of ​​the plateau mountain road slope is carried out regularly through the deployed synthetic aperture radar satellite to obtain SAR image data. For example, according to the needs, the synthetic aperture radar satellite scans the target area of ​​the plateau mountain road slope once a day to generate time-series SAR image data. By setting up earthquake monitoring stations around the target area of ​​the plateau mountain road slope, seismograph data is recorded in real time. For example, the sampling frequency is set to dozens of times per second to ensure that small vibrations are captured. Meteorological monitoring stations are set up around the target area of ​​the plateau mountain road slope to collect meteorological data. The meteorological data includes rainfall, wind speed, temperature, and humidity, which are used to evaluate the impact of meteorological conditions on slope stability. For example, a meteorological monitoring station is set to record data once a minute to monitor meteorological changes in real time. Soil moisture sensors are installed at key locations of the target area of ​​the plateau mountain road slope to obtain Real-time monitoring of soil moisture change data is obtained. The soil moisture change data reflects the degree of soil saturation and is used to assess the risk of landslides. For example, the sensor records data once an hour to monitor soil moisture changes in real time. Geotechnical displacement monitoring equipment is deployed inside and on the surface of the target area of ​​the slope of the plateau mountainous road to obtain real-time rock and soil displacement data. The rock and soil displacement data includes the horizontal and vertical displacements of the rock and soil body, which are used to assess the stability of the slope and to identify geological disasters such as landslides that have already occurred. For example, equipment is set to record displacement data every 10 minutes to monitor the movement of the rock and soil body in real time. An inclination sensor is installed on the surface of the target area of ​​the slope of the plateau mountainous road to obtain real-time monitoring of slope angle change data. The slope angle change data reflects the degree of inclination of the slope and is used to assess the risk of collapse. For example, equipment is set to record angle data every 10 minutes to monitor the slope inclination change in real time.

[0039] Step S2: Use the pre-trained multimodal feature extraction model to extract features from the multimodal monitoring data to obtain a unified high-dimensional feature vector representation of each data source.

[0040] Specifically, before extracting features from the multimodal monitoring data in step S2, the method also includes performing data enhancement on the multimodal monitoring data, generating positive sample pairs, and adjusting the pre-trained multimodal feature extraction model using a contrast loss function.

[0041] Specifically, in step S2, feature extraction of multimodal monitoring data includes using a convolutional neural network to extract spatial features from SAR image data, using a long short-term memory network to extract temporal features from seismograph records and meteorological data, and using a fully connected neural network to extract features from soil moisture change data, rock and soil displacement data, and slope angle change data.

[0042] In the specific implementation process, the pre-trained multimodal feature extraction model includes a convolutional neural network, a long short-term memory network and a fully connected neural network. The convolutional neural network is used to extract spatial features from SAR image data, the long short-term memory network is used to extract time series features from seismograph records and meteorological data, and the fully connected neural network is used to extract features from soil moisture change data, rock and soil displacement data and slope angle change data. Finally, a unified high-dimensional feature vector representation of each data source is obtained through fully connected layer mapping. In addition, before feature extraction, the multimodal monitoring data is enhanced to improve the generalization ability of the model. For example, SAR image data is enhanced by means including but not limited to rotation, scaling, and adding noise, time series data such as seismograph records and meteorological data are enhanced by means including but not limited to time series interpolation and adding random noise, and numerical data such as soil moisture change data, rock and soil displacement data and slope angle change data are enhanced by means including but not limited to data smoothing and adding noise. After enhanced processing, diversified samples of multimodal monitoring data are generated. Randomly select samples from the enhanced data. For each sample, randomly select a sample from its enhanced version and combine it with the original sample to generate a positive sample pair. A positive sample pair refers to a sample pair from the same data source and has similar semantic information. For example, for SAR image data, randomly select one sample from the generated multiple samples that have been subjected to different enhancement processing and combine it with the SAR image data to form a positive sample pair. For seismic recorder data, randomly select a sample from the enhanced samples after the original seismic waveform data is shifted on the time axis and combine it with the original seismic waveform data to form a positive sample pair. In addition, negative sample pairs need to be generated. Negative sample pairs refer to sample pairs from different data sources or with large differences in semantic information. For example, randomly select samples from the enhanced data. For each sample, randomly select a sample from other data sources or samples with different semantic information to form a negative sample pair. Select a sample from the generated multiple SAR image data that have been subjected to different enhancement processing and select a sample from the enhanced samples after the original seismic waveform data is shifted on the time axis to form a negative sample pair. Then, the contrast loss function is used to adjust the pre-trained multimodal feature extraction model. The goal of the contrast loss function is to make the positive sample pairs as close as possible in the feature space, and to make the different sample pairs as far away as possible in the feature space. The specific formula of the contrast loss function is:

[0043]

[0044] Where N is the number of sample pairs, f(x i ) is the sample x i The characteristic vector of is the positive sample pair with x i The feature vector of the paired samples, is the negative sample pair with x iThe feature vector of the paired sample, m is a marginal parameter, which is used to control the distance between the negative sample pairs. Exemplarily, the feature vectors of the positive sample pair and the negative sample pair are input, the distance between the feature vectors of the positive sample pair is calculated, the distance between the feature vectors of the negative sample pair is calculated, and the loss value is calculated according to the formula of the contrast loss function. According to the loss value, the gradient of the multimodal feature extraction model parameters is calculated, and the multimodal feature extraction model parameters are updated using the gradient descent method until the multimodal feature extraction model converges.

[0045] Step S3: Construct a heterogeneous graph based on various data sources and geological disaster domain knowledge.

[0046] Specifically, step S3 includes the following steps:

[0047] S301: taking each data source as a node of a heterogeneous graph;

[0048] S302: Preliminary analysis of the potential correlation between the data sources, identification of the correlation coefficients and mutual information between different data sources, and definition of the causal relationship between nodes in combination with geological disaster domain knowledge and historical disaster data to form the edges of the heterogeneous graph;

[0049] S303: Combine nodes and edges to construct a heterogeneous graph.

[0050] In the specific implementation process, each data source, namely SAR satellite, earthquake monitoring station, meteorological monitoring station, soil moisture sensor, geotechnical displacement monitoring equipment, slope angle monitoring equipment is taken as a node in the heterogeneous graph, and the unified high-dimensional feature vector representation of each data source is used to initialize the attributes of each node. In addition, the attributes of the node include the type of data source, spatial location, and time range. Then, a preliminary analysis of the correlation is performed, and statistical methods and information theory methods are used to analyze the correlation and dependency between different data sources. For example, the Pearson correlation coefficient between different data sources is calculated. For example, the Pearson correlation coefficient threshold is set to 0.7, and the node pairs exceeding the Pearson correlation coefficient threshold are considered to be strongly correlated. The mutual information between different data sources is calculated. For example, the mutual information threshold is set to 0.5, and the node pairs exceeding the mutual information threshold are considered to be nonlinearly dependent. Granger causality test is performed on seismic waveform data and slope angle change data, and the significance level is set to 0.05 to determine whether there is a causal relationship. After the preliminary analysis is completed, the causal relationship between nodes is defined based on the knowledge of geological disasters and historical disaster data. For example, the surface deformation reflected by SAR images has a causal relationship with the rock and soil displacement data. Rainfall directly affects soil moisture. Heavy rainfall can cause landslides. Seismic activity causes slope angle changes and rock and soil displacement. According to the analysis results and knowledge of geological disasters, directed edges are added between nodes with causal relationships or strong correlations. Each edge contains two types of information. One is the edge type, which identifies the type of causal relationship. Causal relationships are divided into direct causality and indirect causality. For example, in direct causal relationships, rainfall directly leads to an increase in soil moisture. In indirect causal relationships, rainfall leads to an increase in soil moisture, which in turn indirectly leads to rock and soil displacement. The second is the edge weight, which indicates the strength of the causal relationship, which is represented by the correlation coefficient and the mutual information value. After determining the nodes and edges, the nodes and edges are combined to construct a heterogeneous graph.

[0051] Specifically, the redundant edges in the heterogeneous graph are checked regularly and removed through a pruning algorithm.

[0052] Specifically, in step S302, the preliminary analysis of the potential correlation between the data sources also includes using statistical methods and information theory methods to calculate the correlation and dependency between different data sources.

[0053] In the specific implementation process, high-performance computing devices are used to accelerate the graph construction and pruning process, and parallel computing technology is used to accelerate the calculation of statistical methods and information theory methods. For example, a graph computing library is used to accelerate the execution of the pruning algorithm, and the heterogeneous graph is updated regularly. According to the latest monitoring data and historical disaster data, the properties of nodes and edges are dynamically adjusted to ensure the real-time and accuracy of the heterogeneous graph. In addition, redundant edges in the heterogeneous graph are checked regularly. Redundant edges refer to edges that contribute little to the heterogeneous graph or are repeated. If two edges represent the same correlation and have similar weights, one of the edges is considered redundant, and a pruning algorithm is used to remove the redundant edge. The pruning algorithm includes but is not limited to threshold-based pruning and graph-theory-based pruning. For example For example, the minimum spanning tree algorithm is used to detect redundant edges, and the edges with weights lower than the preset threshold are removed. During the pruning process, the geological disaster field knowledge and historical disaster data are combined to ensure that the pruned heterogeneous graph can still accurately reflect the correlation and causal relationship between the data sources. The pruned heterogeneous graph is adjusted to ensure the simplicity and effectiveness of the graph. For example, the heterogeneous graph is displayed through visualization tools to check whether the distribution of nodes and edges is reasonable, and the accuracy of the heterogeneous graph is verified using historical disaster data. For example, according to the changing rules of SAR image data and rock and soil displacement data in historical landslide events, it is verified whether the edges between SAR satellite nodes and rock and soil displacement monitoring equipment nodes in the heterogeneous graph accurately reflect the actual correlation.

[0054] Step S4: The unified high-dimensional feature vector representation and heterogeneous graph of each data source are input into the pre-trained graph attention network model for feature aggregation to generate a comprehensive feature vector representation of each data source.

[0055] Step S4 includes the following steps:

[0056] S401: Input the unified high-dimensional feature vector representation and heterogeneous graphs of each data source into the pre-trained graph attention network model;

[0057] S402: Calculate the attention weight between each node and its neighboring nodes using the attention mechanism in the pre-trained graph attention network model;

[0058] S403: Based on the attention weight and the heterogeneous graph, the features of the node and its neighboring nodes are aggregated through a multi-layer message passing mechanism to generate an aggregated feature vector representation of each node;

[0059] S404: Globally pooling the aggregated feature vector representations of all nodes to generate a comprehensive feature vector representation of each data source.

[0060] During the specific implementation process, the unified high-dimensional feature vector representation of each data source extracted is used as the initial feature of the node, and the constructed heterogeneous graph is used as the graph structure to input the pre-trained graph attention network model. The graph attention network model can capture the relationship between the node and its neighbor nodes through the attention mechanism, and calculate the attention weight between each node and its neighbor nodes. The attention weight reflects the strength of the relationship between the nodes. The higher the weight, the stronger the correlation between the two nodes. For example, the attention weight between the SAR satellite node and the geotechnical displacement monitoring equipment node is high, reflecting the strong correlation between the two. Then, based on the attention weight and the heterogeneous graph, the features of the node and its neighbor nodes are aggregated through a multi-layer message passing mechanism to obtain the aggregated feature vector representation of each node. Specifically, in each layer of message passing, the node will fuse the features of its neighbor nodes to generate a new aggregated feature vector representation. For example, the feature vector of the SAR satellite node will fuse the features of its neighbor nodes. The characteristics of the nodes, such as the geotechnical displacement monitoring equipment nodes, generate new aggregated feature vectors. Through multi-layer aggregation, the feature vector of each node gradually integrates the global information and can better represent its position and role in the heterogeneous graph. The message passing mechanism gradually integrates the global information through multi-layer iteration. The output of each layer is used as the input of the next layer, and finally the aggregated feature vector representation of the node is generated. The aggregated feature vector representations of all nodes are globally pooled to generate the comprehensive feature vector representations of each data source. The pooling methods include but are not limited to average pooling, maximum pooling and weighted pooling. Finally, the comprehensive feature vector representations of each data source are output. Among them, the attention mechanism adopts a multi-head attention mechanism. The multi-head attention mechanism captures the different aspects of the relationship between nodes by parallel calculation of multiple attention heads. For example, four attention heads are set in practice, and each attention head independently calculates the attention weight between nodes. The calculation formula for each attention head to calculate the attention weight between nodes is:

[0061]

[0062] Among them, h i and h j are the feature vectors of node i and node j respectively, W q and W kis a learnable weight matrix, d is the dimension of the feature vector, and the outputs of multiple attention heads are concatenated or averaged to obtain the final attention weight. For example, for SAR satellite nodes and geotechnical displacement monitoring equipment nodes, the four attention heads calculate weights of 0.8, 0.7, 0.9, and 0.6, respectively, and the final weight is an average of 0.75. In addition, in the specific implementation process, high-performance GPU servers are used to accelerate the calculation and feature aggregation process of the graph attention network model, and parallel computing technology is used to distribute the computing tasks of the multi-head attention mechanism and the multi-layer message passing mechanism to multiple GPU cores to improve computing efficiency.

[0063] Step S5: Input the comprehensive feature vector representation of each data source into the pre-constructed Bayesian network model, and obtain the probability distribution of geological disaster conditions in the target area of ​​the highway slope in the plateau mountainous area through the Bayesian network model reasoning.

[0064] Specifically, in step S5, the parameters of the Bayesian network model are adjusted based on the historical disaster data by using the maximum likelihood estimation method.

[0065] In the specific implementation process, the structure of the Bayesian network model is defined according to the heterogeneous graph. Each node represents a data source, and the edge represents the causal relationship between nodes. The comprehensive feature vector representation of each data source is mapped to the observation value of the Bayesian network node, and the mapped observation value is standardized to make its mean 0 and variance 1, so that it conforms to the value range of the Bayesian network node. After the multimodal monitoring data is processed into a comprehensive feature vector, the Bayesian network uses these features as input for reasoning, that is, according to the current observation situation, the prior probability is updated to the posterior probability, thereby calculating the conditional probability distribution of different types of geological disasters occurring at a specific time point. This process not only takes into account the impact of various observation indicators on the risk of geological disasters on the slopes of highways in plateau mountainous areas, but also integrates the interactions between them, providing a comprehensive risk assessment framework. Collect historical disaster data, including observations from various data sources and the occurrence of geological disasters. Based on the observations and the occurrence of geological disasters, use the maximum likelihood estimation method to adjust the parameters of the Bayesian network model to ensure that it can accurately reflect the occurrence pattern of geological disasters under similar conditions in history. For example, adjust the conditional probability table between rainfall and soil moisture to make it more consistent with the distribution of historical disaster data.

[0066] Step S6: Based on the preset decision rules, the probability distribution of geological hazard conditions in the target area of ​​the plateau mountain highway slope is converted into the geological hazard identification result of the target area of ​​the plateau mountain highway slope.

[0067] Specifically, before step S6, the method also includes quantifying the uncertainty of the probability distribution of geological hazard conditions in the target area of ​​the highway slope in the plateau mountainous area by calculating the entropy value and the confidence interval, and dynamically adjusting the threshold of the preset decision rule based on the entropy value and the confidence interval.

[0068] During the specific implementation process, the preset decision rules are usually based on threshold judgments. For example, if the probability of a geological disaster occurring is greater than the preset threshold, it is judged as "there is a geological disaster risk". In order to improve the reliability of the identification results, it is also necessary to quantify the uncertainty in the probability distribution of geological hazard conditions at this stage. The uncertainty of the prediction results is measured by calculating the entropy value, and the confidence interval of the probability estimate of each geological disaster event is determined. First, the entropy value of the probability distribution of geological disaster conditions is calculated to measure its uncertainty. The higher the entropy value, the greater the uncertainty. Then the confidence interval of the probability distribution of geological disaster conditions is calculated to measure the reliability of the estimate. For example, the Bayesian method is used to calculate the 95% confidence interval. According to the entropy value and the confidence interval, the threshold is dynamically adjusted. Finally, the probability distribution of geological disaster conditions in the target area of ​​the plateau mountain highway slope is converted into a specific geological disaster identification result in the target area of ​​the plateau mountain highway slope, and the specific potential geological disaster occurrence information is output. This probability distribution reflects the possibility of different types of geological disasters under given current observation conditions. In the process of parsing the probability distribution of geological disaster conditions in the target area of ​​the plateau mountain highway slope, special attention is paid to events with a higher probability of occurrence, that is, an indication of geological disasters that are about to occur or have occurred. In addition, the differences in different types of geological disasters and their potential impacts, as well as the changes in environmental factors in different time periods, adopt a method of dynamically adjusting the threshold of the decision rule. Specifically, based on the calculation results of the entropy value and the confidence interval, the decision threshold used to determine whether to trigger an alarm or take preventive measures is adjusted in real time. For example, for high entropy values, that is, high uncertainty, the threshold for triggering the alarm is appropriately increased to avoid false alarms, while for low entropy values, that is, low uncertainty, the threshold is lowered so that early warning signals can be issued earlier. Based on real-time monitoring data, through the anomaly detection of real-time monitoring data, it is determined whether a geological disaster has occurred. For example, if the rock and soil displacement data suddenly increases and exceeds the historical threshold, it is determined that "landslide has occurred", and if the slope angle data suddenly changes and exceeds the historical threshold, it is determined that "collapse has occurred". The identification results are combined and output to obtain the final identification results of geological disasters in the target area of ​​the highway slope in the plateau mountainous area.

[0069] The working principle of the automatic identification method of geological hazards on highway slopes in plateau mountainous areas by integrating multiple data provided by the present invention is as follows:

[0070] The present invention provides a method for automatically identifying geological hazards on the slopes of plateau mountainous roads by fusing multiple data, and its technical principle is based on multimodal data fusion, heterogeneous graph construction and graph attention network technology. First, multimodal monitoring data of the target area is obtained through multi-source data such as SAR satellites, seismic monitoring stations, and meteorological monitoring stations, and spatial features, temporal features and numerical features are extracted respectively using pre-trained convolutional neural networks, long short-term memory networks and fully connected neural networks to generate a unified high-dimensional feature vector representation. Next, based on the correlation between data sources and the knowledge in the field of geological hazards, a heterogeneous graph is constructed, and each data source and its causal relationship are represented in a graph structure. Then, the high-dimensional feature vector and the heterogeneous graph are input into the pre-trained graph attention network, and the features of nodes and their neighbor nodes are aggregated through the attention mechanism and multi-layer message passing to generate a comprehensive feature vector representation. Finally, the comprehensive feature vector is inferred using a Bayesian network model to obtain the conditional probability distribution of geological hazards, and it is converted into a specific recognition result through dynamically adjusted decision rules. The present invention realizes high-precision automatic identification and early warning of geological hazards on the slopes of plateau mountainous roads through multi-source data fusion, graph structure modeling and probabilistic reasoning.

[0071] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0072] Figure 2 2 is a schematic block diagram of an example device 200 that can be used to implement an embodiment of the present disclosure. The device 200 can be used to implement Figure 1 One or more operations in the method. As shown in the figure, the device 200 includes a central processing unit (CPU) 201, which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) 202 or computer program instructions loaded from a storage unit 208 to a random access memory (RAM) 203. In RAM 203, various programs and data required for the operation of the device 200 can also be stored. CPU 201, ROM 202 and RAM 203 are connected to each other via a bus 204. An input / output (I / O) interface 205 is also connected to the bus 204.

[0073] A number of components in the device 200 are connected to the I / O interface 205, including: an input unit 206, such as a keyboard, a mouse, etc.; an output unit 207, such as various types of displays, speakers, etc.; a storage unit 208, such as a disk, an optical disk, etc.; and a communication unit 209, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 209 allows the device 200 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0074] The processing unit 201 executes the various methods and processes described above, such as Figure 1 For example, in some embodiments, Figure 1 The various operations in the above can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as storage unit 208. In some embodiments, part or all of the computer program can be loaded and / or installed on device 200 via ROM 202 and / or communication unit 209. When the computer program is loaded into RAM 203 and executed by CPU 201, the above-described Figure 1 Alternatively, in other embodiments, the CPU 201 may be configured to perform the operations in any other appropriate manner (eg, by means of firmware). Figure 1 Each operation in .

[0075] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip systems (SOCs), load programmable logic devices (CPLDs), and the like.

[0076] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0077] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0078] In addition, although each operation is described in a specific order, this should be understood as requiring such operation to be performed in the specific order shown or in a sequential order, or requiring that all illustrated operations should be performed to obtain desired results. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single implementation in combination. On the contrary, the various features described in the context of a single implementation can also be implemented in multiple implementations individually or in any suitable sub-combination mode.

[0079] Although the subject matter has been described in language specific to structural features and / or methodological logical actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims.

[0080] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

Claims

1. A method for automatically identifying geological hazards on highway slopes in plateau mountainous areas by integrating multiple data, characterized in that: The automatic identification method comprises the following steps: S1: Obtain multimodal monitoring data of the target area of ​​highway slopes in plateau mountainous areas through various data sources; S2: Use the pre-trained multimodal feature extraction model to extract features from multimodal monitoring data and obtain a unified high-dimensional feature vector representation of each data source; S3: Construct heterogeneous graphs based on various data sources and geological disaster domain knowledge; S4: The unified high-dimensional feature vector representation and heterogeneous graph of each data source are input into the pre-trained graph attention network model for feature aggregation to generate a comprehensive feature vector representation of each data source; S5: Input the comprehensive feature vector representation of each data source into the pre-built Bayesian network model, and obtain the probability distribution of geological disaster conditions in the target area of ​​the highway slope in the plateau mountainous area through the Bayesian network model reasoning; S6: Based on the preset decision rules, the probability distribution of geological hazard conditions in the target area of ​​the highway slope in the plateau mountainous area is converted into the geological hazard identification result in the target area of ​​the highway slope in the plateau mountainous area.

2. The method for automatically identifying geological hazards on highway slopes in plateau mountainous areas by integrating multiple data according to claim 1 is characterized in that: The multiple data sources include SAR satellites, seismic monitoring stations, meteorological monitoring stations, soil moisture sensors, rock and soil displacement monitoring equipment and slope angle monitoring equipment, and the multimodal geological monitoring data include SAR image data, seismograph records, meteorological data, soil moisture change data, rock and soil displacement data and slope angle change data.

3. The method for automatically identifying geological hazards on highway slopes in plateau mountainous areas by integrating multiple data according to claim 2 is characterized in that: Before extracting features from the multimodal monitoring data in step S2, the method also includes performing data enhancement on the multimodal monitoring data, generating positive sample pairs, and adjusting the pre-trained multimodal feature extraction model using a contrast loss function.

4. The method for automatically identifying geological hazards on highway slopes in plateau mountainous areas by integrating multiple data according to claim 3 is characterized in that: In step S2, feature extraction of multimodal monitoring data includes using convolutional neural networks to extract spatial features from SAR image data, using long short-term memory networks to extract temporal features from seismograph records and meteorological data, and using fully connected neural networks to extract features from soil moisture change data, rock and soil displacement data, and slope angle change data.

5. The method for automatically identifying geological hazards on highway slopes in plateau mountainous areas by integrating multiple data according to claim 4 is characterized in that: Step S3 includes the following steps: S301: taking each data source as a node of a heterogeneous graph; S302: Preliminary analysis of the potential correlation between the data sources, identification of the correlation coefficients and mutual information between different data sources, and definition of the causal relationship between nodes in combination with geological disaster domain knowledge and historical disaster data to form the edges of the heterogeneous graph; S303: Combine nodes and edges to construct a heterogeneous graph.

6. The method for automatically identifying geological hazards on highway slopes in plateau mountainous areas by integrating multiple data according to claim 5 is characterized in that: Regularly check the redundant edges in the heterogeneous graph and remove them through pruning algorithms.

7. The method for automatically identifying geological hazards on highway slopes in plateau mountainous areas by integrating multiple data according to claim 6 is characterized in that: In step S302, the preliminary analysis of the potential correlation between the data sources also includes using statistical methods and information theory methods to calculate the correlation and dependency between different data sources.

8. The method for automatically identifying geological hazards on highway slopes in plateau mountainous areas by integrating multiple data according to claim 7 is characterized in that: Step S4 includes the following steps: S401: Input the unified high-dimensional feature vector representation and heterogeneous graphs of each data source into the pre-trained graph attention network model; S402: Calculate the attention weight between each node and its neighboring nodes using the attention mechanism in the pre-trained graph attention network model; S403: Based on the attention weight and the heterogeneous graph, the features of the node and its neighboring nodes are aggregated through a multi-layer message passing mechanism to generate an aggregated feature vector representation of each node; S404: Globally pooling the aggregated feature vector representations of all nodes to generate a comprehensive feature vector representation of each data source.

9. The method for automatically identifying geological hazards on highway slopes in plateau mountainous areas by integrating multiple data according to claim 8, characterized in that: In step S5, the parameters of the Bayesian network model are adjusted based on the historical disaster data by using the maximum likelihood estimation method.

10. The method for automatically identifying geological hazards on highway slopes in plateau mountainous areas by integrating multiple data according to claim 9, characterized in that: Before step S6, the method also includes quantifying the uncertainty of the probability distribution of geological hazard conditions in the target area of ​​the highway slope in the plateau mountainous area by calculating the entropy value and the confidence interval, and dynamically adjusting the threshold of the preset decision rule based on the entropy value and the confidence interval.

Images