CN-121982834-A - Earth surface rock movement disaster early warning method based on multi-mode monitoring data

CN121982834ACN 121982834 ACN121982834 ACN 121982834ACN-121982834-A

Abstract

The invention discloses a ground surface rock movement disaster early warning method based on multi-mode monitoring data, which comprises the steps of obtaining the multi-mode monitoring data, preprocessing, mapping to a rock movement influencing unit, carrying out double-phase crack image registration, improving PSPNet, extracting an abnormal region to form an image mode abnormal evolution field, extracting multi-mode abnormal information, generating an abnormal transfer relation through structural projection, constructing a structural abnormal field, extracting a communication relation and a propagation path based on the structural abnormal field, constructing a rock movement precursor topological skeleton, identifying a critical bridging unit, judging a continuous through path, determining a fission channel triggering state, constructing a phase change state variable with hysteresis memory, determining a rock movement evolution stage, and outputting an early warning result. According to the invention, by improving PSPNet networks, crack abnormal areas are extracted and combined with multi-mode monitoring data to perform structural topology analysis, the advanced identification and stable early warning of the ground mantlerock moving disaster from discrete abnormality to through instability are realized.

Inventors

XU ZELIN
XIAO PENG
WANG SHUO
ZHANG KAIXUAN
YANG MINGLIN

Assignees

山东科技大学

Dates

Publication Date: 20260505
Application Date: 20260409

Claims (8)

1. A surface rock movement disaster early warning method based on multi-mode monitoring data is characterized by comprising the following steps: acquiring multi-mode monitoring data in a monitoring area, preprocessing the multi-mode monitoring data, and mapping various monitoring data to a pre-divided rock movement affecting unit; registering the crack image data at the current moment and the previous moment, inputting the registered crack image data into a modified PSPNet network, extracting each time phase abnormal region, generating an abnormal region time sequence evolution result, mapping the abnormal region time sequence evolution result to a corresponding rock movement influence unit, and forming an image mode abnormal evolution field; Extracting unit abnormal strength, abnormal direction and abnormal continuous state corresponding to each mode data, and carrying out structural projection according to the spatial adjacent relation, geological continuation relation and abnormal propagation direction consistency among rock movement influence units by combining an image mode abnormal evolution field to obtain an abnormal transfer relation and generate a structural abnormal field; based on the structural anomaly field, extracting an anomaly communication relation, an anomaly propagation path, an anomaly bifurcation structure and an un-penetrated section among the rock movement influencing units, and constructing a rock movement precursor topological skeleton; identifying rock movement influencing units which are positioned between different abnormal branches and meet the conditions of consistency of propagation directions, continuity of geological structures and local abnormal enhancement in a rock movement precursor topological skeleton as critical bridging units, and determining whether a continuous through path is formed or not based on the critical bridging units so as to determine the triggering state of a fission channel; based on the structural change condition of the rock movement precursor topological skeleton, the activity degree of the critical bridging unit and the fission channel triggering state, a rock movement change state variable with hysteresis memory is constructed, the current rock movement evolution stage is determined, and a corresponding surface rock movement disaster early warning result is output according to the rock movement evolution stage.
2. The earth surface rock movement disaster early warning method based on multi-modal monitoring data according to claim 1, wherein the multi-modal monitoring data comprises earth surface displacement data, deep displacement data, remote sensing deformation data, remote sensing image data and environment triggering data.
3. The earth surface rock movement disaster warning method based on multi-mode monitoring data according to claim 1, wherein the preprocessing of the multi-mode monitoring data maps various monitoring data to a pre-divided rock movement influencing unit, and the method comprises the following steps: And carrying out time alignment processing and space coordinate unified processing on the multi-mode monitoring data, carrying out denoising and smoothing processing on the abnormal data, dividing the rock movement influencing units according to the space range of the monitoring area, and mapping the multi-mode monitoring data to the corresponding rock movement influencing units according to the space positions corresponding to the monitoring data.
4. The earth surface rock movement disaster early warning method based on multi-mode monitoring data according to claim 1, wherein the generating of the abnormal region time sequence evolution result and mapping to the corresponding rock movement influence unit forms an image mode abnormal evolution field comprises the following steps: Acquiring crack image data at the current moment and crack image data at the previous moment, sequentially carrying out denoising treatment, gray scale normalization treatment and spatial registration treatment on the crack image data at the two moments to obtain a registration image at the current moment and a registration image at the previous moment, and establishing a corresponding relation between an image pixel area and a rock movement influence unit according to the spatial range of the rock movement influence unit; An improved PSPNet network is constructed, the improved PSPNet network comprising a dual-temporal input layer, a high-resolution encoded backbone layer, a disparity-guiding fusion layer, a boundary refinement layer, a pyramid pooling layer, and a segmentation output layer, wherein: the high-resolution coding trunk layer adopts HRNet-W18 architecture, a first input branch receives the registration image at the current moment, a second input branch receives the registration image at the previous moment, and the two input branches respectively extract shallow texture features through a first-stage convolution layer and a second-stage high-resolution parallel convolution layer; Setting a difference guiding fusion layer behind the high-resolution parallel convolution layer in the third stage, performing channel splicing on the current time feature, the previous time feature and the element-by-element difference feature thereof, and outputting fusion features after being compressed by the convolution layer; Setting a boundary refinement layer after the high-resolution parallel convolution layer in the fourth stage, wherein the boundary refinement layer is formed by sequentially connecting a three-by-three convolution layer, a cavity convolution layer and a one-by-one convolution layer; After the pyramid pooling layer is connected with the boundary refinement layer, adopting four different pooling scales to perform context aggregation on the fusion characteristics; The segmentation output layer is connected with the pyramid pooling layer and then respectively outputs an abnormal region segmentation map at the current moment and an abnormal region segmentation map at the previous moment; constructing a training sample set, wherein the training sample set comprises a current time crack image sample, a previous time crack image sample and a corresponding abnormal region label graph, inputting the training sample set into an improved PSPNet network, and training the improved PSPNet network by adopting a joint loss function comprising pixel level segmentation constraint, boundary constraint and double-time phase consistency constraint to obtain a trained improved PSPNet network; Inputting the registration image at the current moment and the registration image at the previous moment into a trained improved PSPNet network, outputting an abnormal region segmentation map at the current moment and an abnormal region segmentation map at the previous moment, extracting the boundary expansion condition, the area increase and decrease condition, the communication state change condition and the adjacent region expansion condition of the abnormal region, and forming an abnormal region time sequence evolution result; Mapping the time sequence evolution result of the abnormal region to a corresponding rock movement influence unit according to the spatial position of the abnormal region, counting the coverage range, the expansion direction, the communication change condition and the continuous activity degree of the abnormal region in each rock movement influence unit, generating an image mode abnormal evolution value corresponding to each rock movement influence unit, and further forming an image mode abnormal evolution field.
5. The earth surface rock movement disaster early warning method based on multi-mode monitoring data according to claim 1, wherein the obtaining of the abnormal transfer relation and the generation of the structural abnormal field comprise the following steps: Performing unit level anomaly extraction on the earth surface displacement data, the deep displacement data, the remote sensing deformation data and the environment triggering data to obtain anomaly strength, anomaly direction and anomaly duration states corresponding to each rock movement influence unit, and obtaining image mode anomaly evolution values corresponding to each rock movement influence unit; Uniformly characterizing the abnormal strength, the abnormal direction, the abnormal continuous state and the abnormal evolution value of the image mode of each rock movement affecting unit to form the current abnormal state information of each rock movement affecting unit; setting up a candidate transfer unit set by taking each rock movement affecting unit as a center according to a space adjacent unit, carrying out direction screening on the candidate transfer unit set according to the abnormal expansion direction in the current abnormal state information, reserving adjacent units positioned in an abnormal expansion forward area as effective transfer units, and identifying target units positioned in an abnormal edge expansion zone and an abnormal non-through gap zone by combining an image mode abnormal evolution value; For each rock movement affecting unit, respectively calculating an abnormal transmission value of the rock movement affecting unit to each effective transmission unit, wherein the abnormal transmission value is determined according to the abnormal strength, the abnormal continuous state, the consistency degree of the abnormal expansion direction and the relative position of the target unit, the edge expansion activity degree corresponding to the image mode abnormal evolution value and whether the target unit is positioned in an abnormal non-through gap zone; And establishing an abnormal transfer relation according to the abnormal transfer value between each rock movement affecting unit and the effective transfer unit, and fusing the current abnormal state information of each rock movement affecting unit with the corresponding abnormal transfer relation to generate a structural abnormal field.
6. The earth surface rock movement disaster early warning method based on multi-mode monitoring data according to claim 1, wherein the building of the rock movement precursor topology skeleton comprises the following steps: extracting a rock movement influence unit with abnormal activity degree reaching a preset condition in a structural abnormal field as a dominant abnormal unit, and extracting a rock movement influence unit which is positioned in an abnormal edge extension zone or an abnormal non-through clearance zone and has an abnormal continuous change state as a potential extension unit; Based on the abnormal transfer relation between the dominant abnormal units, constructing dominant abnormal communication areas, and extracting main propagation sections in each dominant abnormal communication area according to the abnormal transfer direction and the abnormal transfer strength sequence, wherein the main propagation sections are formed by rock movement affecting units which are continuously adjacent and keep the abnormal transfer direction consistent; Based on an abnormal transfer relation between the potential extension unit and the adjacent dominant abnormal communication area, extracting potential extension paths which are positioned at the edge of the dominant abnormal communication area and distributed along the extension direction of the main propagation section, determining the potential extension paths as sections to be penetrated, and recording the connection positions between each section to be penetrated and the corresponding dominant abnormal communication area; Identifying branch propagation paths in each dominant abnormal communication area, determining the rock movement influencing unit as a bifurcation unit when the same rock movement influencing unit is simultaneously connected with more than two abnormal transmission paths, and establishing a layered connection relationship by combining a main propagation section, a section to be penetrated and the bifurcation unit, wherein the dominant abnormal communication areas and the main propagation section form a dominant skeleton, and the section to be penetrated forms a recessive extension skeleton; and integrating the dominant framework, the recessive extension framework, the bifurcation units and the non-through sections to construct the rock movement precursor topological framework.
7. The method for early warning of a surface rock movement disaster based on multi-modal monitoring data according to claim 1, wherein the determining whether a continuous through path is formed based on the critical bridging unit, thereby determining a fission channel trigger state, comprises: acquiring a rock movement precursor topological skeleton, extracting dominant abnormal communication areas and corresponding main propagation sections, and determining non-through sections and potential extension paths between adjacent dominant abnormal communication areas; in the non-through section and the potential extension path, a rock movement influencing unit which simultaneously meets the conditions of consistent direction of a main propagation section of an adjacent dominant abnormal communication area, abnormal transfer relation with an adjacent abnormal unit and continuous abnormal change state is screened to be used as a candidate bridging unit; For each candidate bridging unit, respectively acquiring a corresponding abnormal intensity value, an abnormal continuous state value and an image modal abnormal evolution value, and determining a bridging potential value of the candidate bridging unit by combining a spatial position relation and an abnormal transfer relation of adjacent dominant abnormal communication areas, wherein the bridging potential value is jointly determined by the abnormal intensity, the abnormal continuous state, the abnormal propagation direction consistency degree and the abnormal area edge expansion activity degree; according to the bridging potential values and the spatial distribution relation of each candidate bridging unit, screening rock movement influencing units, wherein the bridging potential values reach preset conditions and form a continuous connection relation in space, as critical bridging units, and associating the critical bridging units with corresponding dominant abnormal communication areas; Based on the connection relation between the critical bridging unit and the dominant abnormal communication area, judging whether a continuous through path formed by the dominant abnormal communication area, the main propagation section and the critical bridging unit exists, determining that the fission channel is in a triggered state when the continuous through path exists, and determining that the fission channel is in an un-triggered state when the continuous through path does not exist.
8. The earth surface rock movement disaster early warning method based on multi-mode monitoring data according to claim 1, wherein the construction of the rock movement change state variable with hysteresis memory, the determination of the current rock movement evolution stage and the output of the corresponding earth surface rock movement disaster early warning result according to the rock movement evolution stage comprise the following steps: Acquiring a rock movement precursor topological skeleton, a critical bridging unit and a fission channel triggering state, and extracting skeleton communication change, main propagation section extension change, to-be-penetrated section shrinkage change and critical bridging unit active change at the current moment and the previous moment; Based on skeleton communication change, main propagation section extension change, to-be-communicated section shrinkage change, critical bridging unit active change and fission channel trigger state, constructing rock phase shift change state variables comprising immediate evolution values and historical memory values, accumulating and updating the historical memory values according to continuous moments, and maintaining the formed communication expansion state; When skeleton expansion enhancement occurs at the current moment and the bridge unit is continuously active or the fission channel is changed from non-triggering to triggering, the historical memory value is improved, when the current moment abnormally occurs and falls back but the section to be penetrated is not restored or the fission channel is still in a triggering state, the historical memory value is maintained, and when the skeleton is contracted, the bridge unit is deactivated and the fission channel is in a non-triggering state at a plurality of continuous moments, the historical memory value is reduced; determining a current rock movement evolution stage according to the instant evolution value and the historical memory value, wherein the rock movement evolution stage comprises a stabilization stage, an inoculation stage, a penetration stage and a destabilization stage, the rock movement evolution stage is increased when the historical memory value reaches the corresponding stage condition, and the current rock movement evolution stage is kept unchanged when the historical memory value does not drop to the corresponding release condition; And outputting a corresponding earth surface rock movement disaster early warning result according to the current rock movement evolution stage, outputting a normal monitoring result in a stable stage, outputting a low-level early warning result in a inoculation stage, outputting a medium-level early warning result in a through stage, and outputting a high-level early warning result in a destabilization stage.

Description

Earth surface rock movement disaster early warning method based on multi-mode monitoring data Technical Field The invention relates to the technical field of geological disaster monitoring and early warning, in particular to a method for early warning earth surface rock movement disasters based on multi-mode monitoring data. Background The disaster of the earth mantlerock is widely occurred in mining areas, mountain area engineering construction areas and areas with complex geological conditions, and is mainly represented by the phenomena of earth surface subsidence, crack expansion, local instability and the like. In order to realize effective early warning of such disasters, the prior art generally obtains information such as earth surface displacement, deep displacement, remote sensing deformation, environment triggering data and the like by arranging a multi-source monitoring means, and obtains crack development conditions by combining an image monitoring technology. In the aspect of data processing, a threshold value judgment method, a time sequence analysis method or a simple data fusion method is adopted for analyzing various monitoring data, so that early warning judgment on rock movement disasters is realized. However, the prior art still has shortcomings in multi-modal data utilization. Most methods are used for carrying out independent analysis or simple superposition processing on monitoring data from different sources, lack deep excavation on spatial correlation and structural evolution characteristics among multi-mode data, and are particularly difficult to accurately describe the spatial communication relation between crack expansion, planar deformation and displacement abnormality. Although the image data can be used for identifying the crack area, the image data usually only stays at a single-time phase segmentation level, and the evolution process of the abnormal area is not effectively reflected by combining multi-time-phase information, so that the identification of the disaster precursor still stays at a local abnormal level. The existing early warning method is mostly based on instantaneous monitoring data or short-time trend, lacks continuous description of an abnormal evolution process, is difficult to identify a key stage of gradual evolution from discrete abnormality to integral communication instability, and is easy to cause problems of early warning lag, false alarm or missing report. Therefore, how to provide a method for early warning of earth surface rock movement disasters based on multi-modal monitoring data is a problem to be solved by those skilled in the art. Disclosure of Invention According to the earth surface rock movement disaster early warning method based on the multi-mode monitoring data, the crack image is segmented in an abnormal area by introducing an improved PSPNet network, a structural abnormal field is built by combining the multi-mode data, a rock movement precursor topological skeleton is further reconstructed, a critical bridging unit is identified, a rock movement variable state variable with hysteresis memory is built, the early identification and dynamic early warning of the evolution process of the earth surface rock movement disaster from discrete abnormality to through instability are realized, and the earth surface rock movement disaster early warning method based on the multi-mode monitoring data has the advantages of being high in early warning advance, low in false alarm rate and strong in result stability. According to the embodiment of the invention, the earth surface rock movement disaster early warning method based on the multi-mode monitoring data comprises the following steps: acquiring multi-mode monitoring data in a monitoring area, preprocessing the multi-mode monitoring data, and mapping various monitoring data to a pre-divided rock movement affecting unit; registering the crack image data at the current moment and the previous moment, inputting the registered crack image data into a modified PSPNet network, extracting each time phase abnormal region, generating an abnormal region time sequence evolution result, mapping the abnormal region time sequence evolution result to a corresponding rock movement influence unit, and forming an image mode abnormal evolution field; Extracting unit abnormal strength, abnormal direction and abnormal continuous state corresponding to each mode data, and carrying out structural projection according to the spatial adjacent relation, geological continuation relation and abnormal propagation direction consistency among rock movement influence units by combining an image mode abnormal evolution field to obtain an abnormal transfer relation and generate a structural abnormal field; based on the structural anomaly field, extracting an anomaly communication relation, an anomaly propagation path, an anomaly bifurcation structure and an un-penetrated section among the rock movement influencing units, and const