Deformation monitoring method for mountain collapse disaster body

Abstract

The invention relates to the technical field of geological disaster monitoring, in particular to a mountain collapse disaster body deformation monitoring method, which comprises the following steps: acquiring multi-dimensional monitoring data of a target disaster body, and performing synchronous processing and correlation analysis to obtain a multi-dimensional correlation data set; extracting feature parameters in the multi-dimensional associated data set and performing dimension reduction processing to obtain a compressed feature set; performing feature fusion processing on the compressed feature set to obtain a real-time fusion feature set; if the deformation parameter in the real-time fusion feature set exceeds a preset deformation threshold, inputting the real-time fusion feature set into a pre-trained convolutional neural network to obtain a deformation trend sequence; and based on the deformation trend sequence, in combination with historical deformation data and geological environment parameters of the target disaster body, analyzing a trend change rule by adopting a pre-trained long-short-term memory network, obtaining a health state predicted value and performing early warning, thereby completing deformation monitoring of the target disaster body. According to the invention, real-time monitoring, fusion analysis and early warning of multi-dimensional data of a disaster body are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster monitoring, and particularly to a method for monitoring the deformation of a mountain collapse disaster body. Background Art

[0002] Mountain collapse is a geological disaster with strong suddenness and great destructiveness, often induced by factors such as geological tectonic activities, rainfall, weathering, or human engineering activities. Its deformation process has the characteristics of concealment, non-linearity, and multi-stage evolution. Once it becomes unstable, it may cause heavy casualties and property losses. Therefore, high-precision and real-time monitoring of the deformation of mountain collapse disaster bodies and realizing disaster warning are key technical requirements in the field of disaster prevention and reduction.

[0003] Currently, the deformation monitoring of mountain collapse mainly relies on: (1) Traditional surface displacement monitoring technologies: including total stations, levels, and GNSS (Global Navigation Satellite System), etc., which measure surface displacement data manually at regular intervals. Although such methods have low costs, they rely on manual operations, have poor real-time performance, and are difficult to capture deep deformation information, and their applicability is limited under bad weather or complex terrain conditions. (2) Remote sensing monitoring technologies: such as Interferometric Synthetic Aperture Radar (InSAR) and unmanned aerial vehicle (UAV) aerial photography technology, which can obtain surface deformation data over a large range. However, InSAR is easily affected by clouds and vegetation cover, and the data update frequency is low; UAV monitoring is limited by the endurance capacity and is difficult to achieve all-weather continuous observation. (3) Sensor network technologies: including the deployment of sensors such as inclinometers, crack meters, and microseismic instruments, and judging by collecting local strain, vibration, and other parameters. However, the sensors are easily disturbed by the environment, have insufficient long-term stability, and the deployment and maintenance costs are high on a large-scale mountain. (4) Exploration of new technologies: In recent years, technologies such as fiber optic sensing and three-dimensional laser scanning (LiDAR) have been gradually applied, but they are easily blocked in complex terrains, the data processing algorithms are highly complex, and it is difficult to meet the real-time warning requirements.

[0004] Therefore, aiming at the technical defects such as insufficient multi-source data fusion, inability to have both real-time performance and accuracy simultaneously, poor environmental adaptability, and low intelligence degree mentioned above, a method for monitoring the deformation of a mountain collapse disaster body is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for monitoring the deformation of a mountain collapse disaster body, which integrates multi-dimensional monitoring data, can adapt to complex environments, and improves the accuracy and timeliness of disaster warning.

[0006] To achieve the above purpose, the present invention provides the following solution:

[0007] A method for monitoring the deformation of a mountain collapse disaster body, comprising:

[0008] Collect multi-dimensional monitoring data of the target disaster body, perform synchronous processing and correlation analysis, and obtain a multi-dimensional correlation data set;

[0009] Extract the characteristic parameters from the multi-dimensional correlation data set and perform dimensionality reduction processing to obtain a compressed feature set;

[0010] Perform feature fusion processing on the compressed feature set to obtain a real-time fusion feature set;

[0011] If the deformation parameters in the real-time fusion feature set exceed the preset deformation threshold, input the real-time fusion feature set into a pre-trained convolutional neural network to obtain a deformation trend sequence;

[0012] Based on the deformation trend sequence, combine the historical deformation data and geological environment parameters of the target disaster body, use a pre-trained long short-term memory network to analyze the trend change law, obtain a health state prediction value and issue an early warning, and complete the deformation monitoring of the target disaster body.

[0013] Preferably, collecting the multi-dimensional monitoring data of the target disaster body includes:

[0014] Deploy GNSS displacement monitoring points, fiber optic strain sensors and meteorological sensors on the surface of the target disaster body to collect surface displacement, crack expansion and temperature and humidity data in real time;

[0015] Drill holes in the target disaster body to deploy a multi-parameter sensing chain, including a microseismograph, an inclinometer and a pore water pressure gauge, to obtain deep rock mass fracture, dip angle change and groundwater activity data.

[0016] Preferably, obtaining the multi-dimensional correlation data set includes:

[0017] Perform timestamp annotation and spatial coordinate calibration on the multi-dimensional monitoring data to obtain a synchronous data set;

[0018] Extract multi-dimensional parameters based on the synchronous data set, and perform correlation analysis on the multi-dimensional parameters through an algorithm to obtain the multi-dimensional correlation data set.

[0019] Preferably, extracting the characteristic parameters from the multi-dimensional correlation data set and performing dimensionality reduction processing to obtain a compressed feature set includes:

[0020] Extract characteristic parameters from the multi-dimensional correlation data set through an edge computing node, clean the characteristic parameters, and construct a feature set;

[0021] After performing feature decomposition and dimensionality reduction processing on the feature set using the principal component analysis method, then perform compression using the t-SNE algorithm to obtain the compressed feature set.

[0022] Preferably, performing feature fusion processing on the compressed feature set to obtain a real-time fusion feature set includes:

[0023] Adopting a dynamic task scheduling strategy to perform multi-node task allocation on the compressed feature set to obtain an allocated task set;

[0024] Adopting a load balancing technique to adjust the allocated task set to obtain a balanced task set;

[0025] Executing the balanced task set in multi-node task processing to obtain a node processing feature set;

[0026] Integrating and performing feature fusion processing on the node processing feature set through data stream scheduling to obtain the real-time fusion feature set.

[0027] Preferably, inputting the real-time fusion feature set into a pre-trained convolutional neural network to obtain a deformation trend sequence includes:

[0028] Respectively inputting the real-time fusion feature set into a convolutional neural network CNN and a long short-term memory network LSTM to obtain a first deformation trend classification result and a second deformation trend classification result;

[0029] Calculating the difference value between the first deformation trend classification result and the second deformation trend classification result. If the difference value exceeds a preset difference value, performing weighted fusion processing on the first deformation trend classification result and the second deformation trend classification result to obtain a final deformation trend sequence; if not, taking the first deformation trend classification result as the final deformation trend sequence.

[0030] Preferably, based on the deformation trend sequence, combining the historical deformation data and geological environment parameters of the target disaster body, and adopting a pre-trained long short-term memory network to analyze the trend change law to obtain a health status prediction value includes:

[0031] Inputting the deformation trend sequence and historical data into a pre-trained bidirectional long short-term memory network BiLSTM for trend change analysis to obtain a preliminary health status value;

[0032] Based on the preliminary health status value and geological environment parameters, obtaining environmental impact features and analyzing the change law of the environmental impact features to obtain a law sequence;

[0033] Combining the law sequence with historical data, judging the stability of the health status and analyzing the fluctuation of the health status to obtain a final health status value.

[0034] Preferably, the method further includes:

[0035] Perform spatial distribution mapping and time series projection based on the predicted health status value to obtain the health status distribution map of the target disaster body;

[0036] Extract key area parameters based on the health status distribution map. If the key area parameters are abnormal, adjust the monitoring frequency through a dynamic update algorithm to obtain an optimized monitoring plan, and update the distributed sensor network settings according to the optimized monitoring plan, and adjust the sensor data acquisition and processing process.

[0037] The beneficial effects of the present invention are:

[0038] The present invention collects multi-dimensional parameters such as displacement, inclination, strain, and cracks of a mountain collapse disaster body through a distributed sensor network, and performs spatio-temporal synchronization processing to obtain a multi-dimensional correlation data set. Subsequently, feature extraction and dimensionality reduction are performed on this data set to obtain a compressed feature set, and real-time feature fusion is achieved through a parallel computing framework. When the deformation parameters in the fusion result exceed the preset threshold, a convolutional neural network is used to judge the deformation trend of the disaster body, and the historical data and geological environment parameters are analyzed in combination with a long short-term memory network to predict the health status of the disaster body and give an early warning. This method realizes real-time monitoring, fusion analysis and early warning of multi-dimensional data of the disaster body, and provides effective technical support for geological disaster prevention and control. Description of the Drawings

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0040] Figure 1 It is a flow chart of a deformation monitoring method for a mountain collapse disaster body according to an embodiment of the present invention. Detailed Embodiments

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0042] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0043] This embodiment provides a deformation monitoring method for a mountain collapse disaster body, as Figure 1As shown in the figure, it includes:

[0044] Collect multi-dimensional monitoring data of the target disaster body, perform synchronous processing and correlation analysis, and obtain a multi-dimensional correlation data set;

[0045] Extract the characteristic parameters in the multi-dimensional correlation data set and perform dimensionality reduction processing to obtain a compressed feature set;

[0046] Perform feature fusion processing on the compressed feature set to obtain a real-time fusion feature set;

[0047] If the deformation parameter in the real-time fusion feature set exceeds the preset deformation threshold, input the real-time fusion feature set into a pre-trained convolutional neural network to obtain a deformation trend sequence;

[0048] Based on the deformation trend sequence, combine the historical deformation data and geological environment parameters of the target disaster body, and use a pre-trained long short-term memory network to analyze the trend change law, obtain a health status prediction value and issue an early warning, and complete the deformation monitoring of the target disaster body.

[0049] Specifically, in this embodiment, a distributed sensor network is used to collect multi-dimensional parameters such as displacement, inclination, strain, and cracks of the mountain collapse disaster body, and perform spatio-temporal synchronization processing to obtain a multi-dimensional correlation data set. Subsequently, feature extraction and dimensionality reduction are performed on this data set to obtain a compressed feature set, and real-time feature fusion is achieved through a parallel computing framework. When the deformation parameter in the fusion result exceeds the preset threshold, a convolutional neural network is used to judge the deformation trend of the disaster body, and a long short-term memory network is combined to analyze the historical data and geological environment parameters to predict the health status of the disaster body and issue an early warning. This method realizes the real-time monitoring, fusion analysis and early warning of multi-dimensional data of the disaster body, and provides effective technical support for geological disaster prevention and control.

[0050] Furthermore, obtaining the multi-dimensional correlation data set includes:

[0051] Perform timestamp annotation and spatial coordinate calibration on the multi-dimensional monitoring data to obtain a synchronous data set;

[0052] Extract multi-dimensional parameters based on the synchronous data set, and perform correlation analysis on the multi-dimensional parameters through an algorithm to obtain the multi-dimensional correlation data set.

[0053] Among them, collecting the multi-dimensional monitoring data of the target disaster body includes:

[0054] Layout GNSS displacement monitoring points, fiber optic strain sensors and meteorological sensors on the surface of the target disaster body to collect surface displacement, crack expansion and temperature and humidity data in real time;

[0055] Drill holes inside the target disaster body to deploy a multi-parameter sensing chain, including a microseismic instrument, an inclinometer, and a pore water pressure gauge, to obtain data on deep rock mass fractures, dip angle changes, and groundwater activities.

[0056] Specifically, in this embodiment, sensors are deployed through a distributed network to collect multi-dimensional monitoring data of the target disaster body, and an original data set is constructed; the synchronous set is labeled with timestamps to obtain time-consistent data, and the time-consistent data is processed by the mapping method using a coordinate system to obtain spatially calibrated data, and a synchronous data set is constructed; multi-dimensional parameters are extracted from the synchronous data set, and the correlation analysis of the multi-dimensional parameters is carried out through an algorithm to obtain a multi-dimensional correlation data set.

[0057] In the synchronous data processing, first, the time reference unification algorithm is applied. The precise time signal accurate to the nanosecond level is obtained through the Global Positioning System (GPS) to ensure that the timestamps of all sensors are aligned under the unified time reference. For example, the Network Time Protocol (NTP) is used to calibrate the sensor time, and the error is controlled within 1 millisecond. Then, the spatial reference consistency algorithm is used to match the position information of each sensor with a high-precision geographic coordinate system (such as WGS84). The spatial coordinate mapping of the sensor data is carried out through the Kalman filter algorithm to eliminate the deviation caused by installation errors and environmental interference, and the spatial error is controlled within 5 millimeters. Subsequently, the multiple linear regression analysis algorithm is used to carry out the correlation analysis on the data after time alignment and spatial mapping, and multi-dimensional features are extracted. For example, by analyzing the correlation of displacement, tilt, strain, and crack data, a dynamic response model is established, and the least squares method is used to optimize the model parameters to ensure that the goodness of fit reaches more than 95. Finally, the processed multi-dimensional correlation data set is imported into the distributed database, and the spatio-temporal indexing technology is used to achieve efficient query and analysis, supporting subsequent real-time monitoring and prediction tasks.

[0058] In this embodiment, the synchronization set is labeled with timestamps, which can ensure that the data collected by different sensors are consistent in the time dimension. For example, in the monitoring of disaster bodies, multiple sensors may have slight time deviations due to transmission delays or device differences. Timestamps record the acquisition time of each group of data with millisecond-level accuracy. Through comparison and adjustment, time-consistent data is generated. This method avoids analysis errors caused by time misalignment. In this embodiment, assume that in a certain monitoring, the timestamp recorded by the displacement sensor is 10:00:00.125, while that of the crack sensor is 10:00:00.130. Then, the two groups of data are unified to the same moment, such as 10:00:00.100, through the timestamp alignment algorithm to ensure the reliability of subsequent processing. The mapping method is used to process the time-consistent data with a coordinate system, aiming to correspond the data with spatial positions to form spatially calibrated data. Specifically, the longitude and latitude information provided by the global navigation satellite system can be used to map the position of each sensor into a unified geographic coordinate system. For example, the original position data of a certain sensor is 120.5 degrees east longitude and 30.2 degrees north latitude, and it is adjusted to x = 500 meters and y = 300 meters in the standard grid coordinate system through the coordinate conversion algorithm, with the error controlled at the centimeter level. This spatial calibration provides an accurate spatial reference for subsequent multi-dimensional analysis. Multidimensional parameters are extracted from the spatially calibrated data to obtain parameter set data, which can be understood as separating independent variables such as displacement, tilt, and strain from the original data. Exemplarily, in a monitoring, the displacement parameter shows that a certain point has moved 3 millimeters, the tilt parameter records an angular change of 0.2 degrees, and the strain value has increased by 10 microstrains. These parameters are integrated into a multi-dimensional data set. This extraction process lays the foundation for correlation analysis. Correlation analysis is performed on the parameter set data through an algorithm to generate feature correlation data, aiming to discover the potential relationships between different parameters. In this embodiment, it is found through analysis that the crack width also expands while the displacement increases, and the two show a positive correlation trend, with a correlation coefficient reaching 0.85. This correlation reveals the stress change law inside the disaster body and provides a basis for risk assessment.

[0059] Further, feature parameters in the multi-dimensional correlation data set are extracted and dimensionality reduction processing is performed to obtain a compressed feature set, including:

[0060] Feature parameters are extracted from the multi-dimensional correlation data set by an edge computing node, and the feature parameters are cleaned to construct a feature set;

[0061] After performing feature decomposition and dimensionality reduction processing on the feature set using the principal component analysis method, the t-SNE algorithm is then used for compression to obtain the compressed feature set.

[0062] Specifically, this embodiment extracts feature parameters from multidimensional data through edge computing nodes to obtain an initial feature set; uses a preprocessing method to clean the initial feature set to obtain a standardized feature set; applies a feature decomposition method to the standardized feature set to obtain a decomposed feature set; and processes the decomposed feature set through a dimensionality reduction algorithm to obtain a compressed feature set.

[0063] In disaster monitoring scenarios, multidimensional correlated datasets contain multiple parameters such as displacement, tilt, strain, and cracks. These data are preprocessed by edge computing nodes. Principal component analysis (PCA) is first used to perform feature decomposition on the high-dimensional data. For example, 10-dimensional data can be reduced to 3 dimensions, retaining 95% of the original information and reducing data redundancy. The covariance matrix and eigenvalue decomposition are calculated to extract the main eigenvectors. For example, the first principal component explains 60% of the variance, the second principal component explains 30%, and the third principal component explains 5%. The t-SNE algorithm is then used to further compress the reduced-dimensional data, mapping the 3D data into a 2D space while preserving the local structural relationships between the data. For example, in a 2D plane, the distribution of displacement and strain shows a clear clustering trend.

[0064] Edge computing nodes extract feature parameters from multidimensional data to generate an initial feature set. For example, in disaster monitoring, edge computing nodes can extract parameters such as displacement, strain, and tilt angle from raw sensor data. For example, data acquired by a node from a mountain surface sensor includes a displacement of 2 mm, a strain of 8 microstrain, and a tilt angle of 1.5 degrees. These parameters constitute the initial feature set. This approach shifts computing tasks to the edge, reducing the burden of data transmission. Preprocessing methods are used to cleanse the initial feature set to obtain a standardized feature set. Specifically, data cleaning can be performed by removing noise and outliers. For example, if the displacement value of a sensor suddenly jumps to 50 mm, significantly deviating from the normal range, the preprocessing layer will remove it. In one possible implementation, data normalization maps parameters such as displacement and strain to the range of 0 to 1 to facilitate subsequent analysis. This cleaning ensures data quality and provides a reliable foundation for feature decomposition. Feature decomposition is applied to the standardized feature set to obtain a decomposed feature set. It should be noted that feature decomposition can break complex data into independent components. For example, displacement and strain may be affected by multiple factors, and after decomposition, the main change trend and secondary interference components can be separated. In this embodiment, the displacement of a certain mountain area is decomposed into a 70% trend component and a 30% noise component. This decomposition helps to focus on key information. The decomposed feature set is processed by a dimensionality reduction algorithm to obtain a compressed feature set. Specifically, the principal component analysis method can be used to reduce high-dimensional data to a few main dimensions. For example, multi-dimensional data originally containing displacement, strain, inclination angle, etc., retains two principal components after dimensionality reduction and can still represent 90% of the information. This compression reduces redundancy and improves processing efficiency.

[0065] Furthermore, performing feature fusion processing on the compressed feature set to obtain a real-time fusion feature set includes:

[0066] Using a dynamic task scheduling strategy to perform multi-node task allocation on the compressed feature set to obtain an allocated task set;

[0067] Using a load balancing technique to adjust the allocated task set to obtain a balanced task set;

[0068] Executing the balanced task set in multi-node task processing to obtain a node processing feature set;

[0069] Integrating and performing feature fusion processing on the node processing feature set through data flow scheduling to obtain the real-time fusion feature set.

[0070] Specifically, in this embodiment, a parallel computing framework is used to perform task allocation on the compressed feature set to obtain a preliminary allocated task set; a load balancing technique is used to adjust the preliminary allocated task set to obtain a balanced task set; the balanced task set is executed in multi-node processing to obtain a node processing feature set; the node processing feature set is integrated through data flow scheduling to obtain a scheduled feature set; for the scheduled feature set, a fusion algorithm is applied to obtain a real-time fusion feature set.

[0071] Under a distributed computing framework, a dynamic task scheduling strategy is used to allocate the compressed feature set to multiple computing nodes. For example, based on the RDD partitioning mechanism of Spark, the data is divided into 8 parallel blocks, and each block processes approximately 15% of the data volume. The consistent hashing algorithm is used to ensure uniform data distribution, and the node load difference is controlled within 5%. In the feature fusion stage, the MapReduce model is used to perform weighted calculations on parameters such as displacement and inclination angle. The weight coefficients are set according to the sensor accuracy. For example, the displacement weight is 4, the inclination angle weight is 3, the strain weight is 2, and the crack weight is 1. A sliding window mechanism is used to perform dynamic aggregation on time series data. The window length is set to 10 seconds, and the step size is 5 seconds to ensure data continuity. During the fusion process, the Kalman filter algorithm is used to eliminate noise. For example, the displacement measurement error is reduced from ±2 mm to ±3 mm. In terms of real-time guarantee, the task queue is optimized through a delay scheduling strategy, and load migration is automatically triggered when the node processing time exceeds 50 milliseconds. The final fusion result is output in JSON format, including statistical quantities such as mean and variance. For example, the displacement fusion value is 8 mm (variance 9), and the inclination angle fusion value is 2 degrees (variance 4), and it is pushed to the downstream through a message middleware (such as Kafka).

[0072] When allocating tasks to the compressed feature set through a parallel computing framework, the core lies in reasonably splitting the data processing tasks to improve efficiency. For example, in the scenario of disaster body monitoring, the compressed feature set may contain key parameters such as displacement and strain. Assuming the data volume is 100,000 records. The parallel computing framework will evenly divide the data into 5 parts according to the number of nodes, such as 5 edge computing nodes, with 20,000 records in each part, initially forming an allocation task set. This method makes full use of the computing power of multiple nodes to ensure parallel execution of tasks. In this embodiment, when using load balancing technology to adjust the initially allocated task set, the performance differences of each node need to be considered. If a certain node has a slower processing speed, its task volume can be dynamically reduced to 15,000 records, while the high-performance node is increased to 25,000 records, forming a balanced task set. Specifically, the task allocation can be adjusted by monitoring the real-time load of the nodes, such as CPU occupancy or memory usage. This method avoids overloading of a certain node and ensures balanced processing efficiency. When multiple nodes process the balanced task set, the process of obtaining the node processing feature set emphasizes independence. For example, each node extracts the displacement mean value and strain peak value for its allocated 20,000 data respectively. For example, node 1 obtains a displacement mean value of 3 mm and a strain peak value of 10 microstrains, and node 2 obtains 4 mm and 12 microstrains. This distributed processing can quickly generate local feature sets, providing a basis for subsequent integration. It should be noted that when the data stream scheduling integrates the node processing feature sets, the focus is on the time series consistency. In this embodiment, the feature sets of the 5 nodes can be sorted according to the acquisition time through timestamp alignment to generate a scheduled feature set. For example, the displacement data stream is integrated according to the sampling frequency of 1 second per second to ensure that the time series relationship between the data is not lost. This scheduling method is convenient for maintaining the integrity of the data during subsequent analysis. When applying a fusion algorithm to the scheduled feature set, a real-time fusion feature set can be generated through a weighted average method. For example, the displacement mean values of the 5 nodes are weighted and fused, and the weights are allocated according to the node data volume, and finally the global displacement mean value of 3.8 mm is obtained. This fusion can synthesize multi-source data and improve the representativeness of the features.

[0073] Further, inputting the real-time fusion feature set into a pre-trained convolutional neural network to obtain the deformation trend sequence includes:

[0074] Inputting the real-time fusion feature set into the convolutional neural network CNN and the long short-term memory network LSTM respectively to obtain a first deformation trend classification result and a second deformation trend classification result;

[0075] Calculating the difference value between the first deformation trend classification result and the second deformation trend classification result. If the difference value exceeds the preset difference value, performing a weighted fusion process on the first deformation trend classification result and the second deformation trend classification result to obtain the final deformation trend sequence; if not, using the first deformation trend classification result as the final deformation trend sequence.

[0076] Specifically, in this embodiment, it is set that when the deformation parameters in the real-time fusion feature set exceed the preset threshold (for example, the displacement exceeds 15 mm or the inclination angle exceeds 5°), the deformation trend analysis process is automatically triggered. First, the fused deformation time series data is input into a pre-trained convolutional neural network (CNN) model. The network structure includes 3 convolutional layers (the convolutional kernel size is 3×3, the stride is 1, and the activation function is ReLU) and 2 fully connected layers (the number of nodes is 128 and 64 respectively). The input layer receives time series data segments with a sliding window length of 20 seconds and a stride of 10 seconds. The output layer of the CNN uses the Softmax function for classification, and the deformation trend is divided into three states: slow growth (probability < 3), accelerating growth (3 ≤ probability < 7), and mutation (probability ≥ 7). For example, when the input displacement sequence is [12mm, 15mm, 18mm, 22mm], the model outputs an accelerating growth probability of 68, triggering a warning signal.

[0077] To enhance the reliability of trend judgment, a long short-term memory network (LSTM) is synchronously used to perform time dimension analysis on the same data. The number of hidden layer units is set to 32, and the time step is 8. When the difference between the CNN and LSTM results exceeds 15% (for example, the CNN outputs 68 while the LSTM outputs 52), adaptive weighted fusion is started, and the weight coefficients are CNN 6 and LSTM 4. The finally generated deformation trend sequence is updated at a period of 10 seconds, and the GRU neural network is used to predict the deformation amount change curve within the next 30 seconds. For example, the output prediction value sequence is [25mm, 28mm, 32mm], and its mean square error is controlled within 3 mm.

[0078] Furthermore, based on the deformation trend sequence, combined with the historical deformation data and geological environment parameters of the target disaster body, a pre-trained long short-term memory network is used to analyze the trend change law, and the health state prediction value is obtained, including:

[0079] Input the deformation trend sequence and historical data into a pre-trained bidirectional long short-term memory network BiLSTM for change law analysis to obtain the preliminary health state value;

[0080] Based on the preliminary health state value and geological environment parameters, obtain the environmental impact characteristics and analyze the change law of the environmental impact characteristics to obtain a law sequence;

[0081] Combine the law sequence with historical data to judge the stability of the health state and analyze the fluctuation of the health state to obtain the final health state value.

[0082] Specifically, in this embodiment, the variation law is analyzed by using a long short-term memory network through a deformation trend sequence and historical data to obtain a preliminary health state value. For the preliminary health state value, geological environment parameters are integrated to obtain environmental impact characteristics, and the adjusted state characteristics are obtained. If the adjusted state characteristics exceed the preset threshold, the variation law is determined through trend analysis to obtain a law sequence. According to the law sequence, a data integration method is used to process historical data to obtain an integrated feature set. The stability of the health state is judged through the integrated feature set and the environmental impact characteristics to obtain a stability sequence. For the stability sequence, a state judgment method is used to analyze the sequence fluctuation to obtain a final prediction value. According to the final prediction value and the variation law, the trend evolution characteristics are obtained to determine the evolution direction.

[0083] Based on the deformation trend sequence, the system extracts the historical displacement data of the past 30 days (such as [8mm, 10mm, 12mm, 15mm, 18mm]) and geological environment parameters (including rock mass strength of 35MPa, joint density of 5 joints / meter, and groundwater permeability coefficient of 2×10 -5 m / s) to construct a multi-source feature matrix. A bidirectional long short-term memory network (BiLSTM) is used for modeling. The network structure includes 2 layers of LSTM units (64 neurons in each layer), the time step is set to 5, the input window length is 15 minutes, and the sliding step is 5 minutes. When training the model, the mean square error loss function is used, the learning rate is 0.01, the batch size is 32, and the validation set error stabilizes within 1mm after 100 iterations. For the real-time input deformation sequence [16mm, 18mm, 20mm], the model outputs a health state prediction value of 82 for the next 1 hour (threshold range: 0 - 1, values above 8 are high-risk), and at the same time generates the contribution degrees of various influencing factors (rock mass strength accounts for 28%, joint density accounts for 35%, and groundwater accounts for 22%). The attention mechanism is introduced during the prediction process to automatically weight the key time nodes (the weights reach 18 and 21 at the 3rd minute and the 8th minute respectively). To eliminate data noise, wavelet transform (db4 wavelet basis, decomposition level 3) is used to filter the original signal, and the mean square error of the reconstructed displacement sequence is reduced to 8mm.

[0084] When analyzing the variation law through the deformation trend and historical data, the long short-term memory network is good at capturing the long-term dependencies in the time series. In geological deformation monitoring, assuming the displacement data of a certain area for 30 consecutive days is used as the input, the network analyzes the daily displacement changes through 32 hidden units and initially obtains the health status value. For example, if the displacement slowly increases from 5 mm to 10 mm, the network outputs a health status value of 75, indicating that the structure is still stable. When fusing the geological environment parameters with the initial health status value, environmental impact features such as rainfall and temperature can be introduced. Suppose the rainfall reaches 50 mm and the temperature drops suddenly by 10 degrees on a certain day. Combining these parameters with the health status value, the adjusted status feature may drop from 75 to 60, reflecting the impact of the environment on the structure. It should be noted that this adjustment can more realistically reflect the actual health condition. If the adjusted status feature exceeds the preset threshold, for example, the health value is lower than 65, the variation law is determined through trend analysis. Specifically, the displacement data for the past 15 days is analyzed. For example, if it increases from 10 mm to 18 mm, the regular sequence shows that the daily increase rate rises from 0.5 mm to 1 mm, indicating an accelerating change. When integrating historical data according to the regular sequence, for example, combining the displacement records of the past year, the data integration method can be processed through weighted average to obtain the integrated feature set. Exemplarily, the weight of recent data is set to 0.7 and that of far-term data is 0.3. The integrated feature set may show an average displacement rate of 0.8 mm / day. When judging the stability by integrating the integrated feature set and the environmental impact feature, in one embodiment, if the integrated feature set shows a rate of 0.8 mm / day and the rate suddenly increases to 1.5 mm / day after rainfall, the stability sequence may be [stable, slightly fluctuating, unstable], reflecting the dynamic change of the health status. Specifically, this judgment can timely detect potential risks. When analyzing the fluctuations in the stability sequence, it can be understood that the state judgment method can adopt the statistical variance of a sliding window. For example, the window length is 5 days, and the sequence variance increases from 0.2 to 0.8. The final predicted value may be 55, indicating a deterioration of the health status. This method can effectively quantify the degree of fluctuations. When obtaining the trend evolution feature based on the final predicted value and the variation law, for example, the predicted value of 55 combined with the regular sequence [0.5, 0.7, 1.0], the evolution direction can be inferred as accelerating deterioration. This analysis can provide a basis for early warning and avoid sudden collapses.

[0085] Furthermore, the method further includes:

[0086] Performing spatial distribution mapping and time series projection based on the health status prediction value to obtain the health status distribution map of the target disaster body;

[0087] Extracting key area parameters based on the health status distribution map. If the key area parameters are abnormal, the monitoring frequency is adjusted through a dynamic update algorithm to obtain an optimized monitoring plan, and the distributed sensor network settings are updated according to the optimized monitoring plan, and the sensor data acquisition and processing process are adjusted.

[0088] Specifically, in this embodiment, based on the health status and prediction values, the spatial distribution analysis method is used to obtain the initial distribution characteristics; for the initial distribution characteristics, mapping technology is applied to obtain the spatial distribution mapping result; according to the spatial distribution mapping result, time series analysis is performed to obtain the time series characteristics; for the time series characteristics, projection technology is used to obtain the health map projection data; if the health map projection data exceeds the preset threshold, the distribution characteristic changes are judged through the analysis of the disaster body state distribution; according to the distribution characteristic changes, the data integration method is used to obtain the adjusted state distribution; based on the adjusted state distribution and the acquisition method, the health status distribution map of the disaster body is determined.

[0089] The key area parameters are obtained from the health status distribution map, and the preset threshold is used to judge whether the parameters are abnormal to obtain the abnormal judgment result; for the abnormal judgment result, if it is abnormal, the frequency adjustment value is calculated through the dynamic update algorithm to determine the new monitoring frequency; according to the new monitoring frequency, the adjusted monitoring time interval is obtained to get the optimized time series data; through the optimized time series data, the data analysis method is used to extract the health status change trend and judge the change characteristics; for the change characteristics, if they deviate from the normal range, the abnormal detection algorithm is used to identify the abnormal pattern and determine the abnormal type; according to the abnormal type, the monitoring plan is updated by the frequency adjustment strategy to obtain the optimized monitoring plan.

[0090] The initial data stream is obtained through the sensor network, and the acquisition area is divided by the distributed configuration to obtain the partitioned monitoring data; according to the partitioned monitoring data, combined with the preset threshold to judge the data acquisition abnormality, if it is abnormal, the sensor network is adjusted through the feedback mechanism to obtain the updated data stream; for the updated data stream, the processing process is used to separate the noise signal to obtain the filtered monitoring data; through the filtered monitoring data, combined with the optimization scheme to calculate the change trend to obtain the trend analysis result; according to the trend analysis result, if the change trend exceeds the preset range, the data acquisition frequency is adjusted through the control process to obtain the frequency optimized data; for the frequency optimized data, the machine learning algorithm is used to detect the potential abnormal pattern to obtain the abnormal detection result; through the abnormal detection result, combined with the configuration management to update the distributed configuration to obtain the continuously optimized monitoring data.

[0091] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention should fall within the protection scope determined by the claims of the present invention.

Claims

1. A deformation monitoring method for a landslide disaster body, characterized in that, Including: Collect multi-dimensional monitoring data of the target disaster body, perform synchronous processing and correlation analysis, and obtain a multi-dimensional correlation data set; Extract characteristic parameters from the multi-dimensional correlation data set and perform dimensionality reduction processing to obtain a compressed feature set; Perform feature fusion processing on the compressed feature set to obtain a real-time fusion feature set; If the deformation parameters in the real-time fusion feature set exceed the preset deformation threshold, input the real-time fusion feature set into a pre-trained convolutional neural network to obtain a deformation trend sequence; Based on the deformation trend sequence, combine the historical deformation data and geological environment parameters of the target disaster body, and use a pre-trained long short-term memory network to analyze the trend change law, obtain a health status prediction value and issue a warning, and complete the deformation monitoring of the target disaster body.

2. The deformation monitoring method for the landslide disaster body according to claim 1, characterized in that, Collecting the multi-dimensional monitoring data of the target disaster body includes: Deploy GNSS displacement monitoring points, fiber optic strain sensors and meteorological sensors on the surface of the target disaster body to collect surface displacement, crack expansion and temperature and humidity data in real time; Drill holes in the target disaster body to deploy a multi-parameter sensing chain, including a microseismic instrument, an inclinometer and a pore water pressure gauge, to obtain deep rock mass fracture, dip angle change and groundwater activity data.

3. The deformation monitoring method for the landslide disaster body according to claim 1, characterized in that, Obtaining the multi-dimensional correlation data set includes: Perform timestamp annotation and spatial coordinate calibration on the multi-dimensional monitoring data to obtain a synchronous data set; Extract multi-dimensional parameters based on the synchronous data set, and perform correlation analysis on the multi-dimensional parameters through an algorithm to obtain the multi-dimensional correlation data set.

4. The deformation monitoring method for mountain collapse disaster bodies according to claim 1, characterized in that, Extracting the characteristic parameters from the multi-dimensional correlation data set and performing dimensionality reduction processing to obtain a compressed feature set includes: Extract characteristic parameters from the multi-dimensional correlation data set through an edge computing node, clean the characteristic parameters, and construct a feature set; After performing feature decomposition and dimensionality reduction processing on the feature set using the principal component analysis method, then perform compression using the t-SNE algorithm to obtain the compressed feature set.

5. The deformation monitoring method for the landslide disaster body according to claim 1, characterized in that, Performing feature fusion processing on the compressed feature set to obtain a real-time fusion feature set includes: Use a dynamic task scheduling strategy to perform multi-node task allocation on the compressed feature set to obtain an allocated task set; Use load balancing technology to adjust the allocated task set to obtain a balanced task set; Execute the balanced task set in multi-node task processing to obtain a node processing feature set; Integrate and perform feature fusion processing on the node processing feature set through data flow scheduling to obtain the real-time fusion feature set.

6. The deformation monitoring method for the landslide disaster body according to claim 1, characterized in that Inputting the real-time fusion feature set into a pre-trained convolutional neural network to obtain a deformation trend sequence includes: Input the real-time fusion feature set into a convolutional neural network CNN and a long short-term memory network LSTM respectively to obtain a first deformation trend classification result and a second deformation trend classification result; Calculate the difference value between the first deformation trend classification result and the second deformation trend classification result. If the difference value exceeds the preset difference value, perform weighted fusion processing on the first deformation trend classification result and the second deformation trend classification result to obtain a final deformation trend sequence; if not, use the first deformation trend classification result as the final deformation trend sequence.

7. The deformation monitoring method for the landslide disaster body according to claim 1, characterized in that, Based on the deformation trend sequence, combined with the historical deformation data and geological environment parameters of the target disaster body, a pre-trained long short-term memory network is used to analyze the trend change law, and the health status prediction value is obtained, including: Input the deformation trend sequence and historical data into the pre-trained bidirectional long short-term memory network BiLSTM for change law analysis to obtain the preliminary health status value; Based on the preliminary health status value and geological environment parameters, obtain the environmental impact characteristics and analyze the change law of the environmental impact characteristics to obtain the law sequence; Combine the law sequence with historical data to judge the stability of the health status and analyze the fluctuation of the health status to obtain the final health status value.

8. The deformation monitoring method for the landslide disaster body according to any one of claims 1-7, characterized in that The method further includes: Perform spatial distribution mapping and time series projection based on the health status prediction value to obtain the health status distribution map of the target disaster body; Extract the key area parameters based on the health status distribution map. If the key area parameters are abnormal, adjust the monitoring frequency through the dynamic update algorithm to obtain the optimized monitoring plan, and update the distributed sensor network settings according to the optimized monitoring plan, and adjust the sensor data collection and processing process.

Images