CN-121999374-A - Dynamic identification method and system for tunnel construction geological disasters

CN121999374ACN 121999374 ACN121999374 ACN 121999374ACN-121999374-A

Abstract

The invention belongs to the technical field of geological disaster identification, and provides a method and a system for dynamically identifying geological disasters in tunnel construction, wherein geological and structural response data in the tunnel construction process are obtained, data are cleaned and characteristics are enhanced, a dynamic transverse expansion network learning model is utilized, primary image fusion characteristics of the data are extracted, partial characteristic nodes are mapped into enhancement nodes through random weight mapping, an increment learning mechanism is triggered to respond to tunnel construction condition changes, increment is carried out on the characteristic nodes and the enhancement nodes, the obtained data are newly added, the number of the characteristic nodes and the enhancement nodes is dynamically adjusted according to characteristic distribution and complexity of the newly added data, and real-time discrimination and risk grade assessment of disaster types are carried out by utilizing the geological disaster dynamic identification model by taking the characteristic nodes and the enhancement nodes as inputs. The invention can realize real-time sensing and dynamic risk prevention and control of tunnel construction geological disasters.

Inventors

ZHANG YANG
DING MINGWEI
LI YU
LI LI
FANG ZHONGDONG

Assignees

山东大学

Dates

Publication Date: 20260508
Application Date: 20260410

Claims (10)

1. A dynamic identification method for geological disasters in tunnel construction is characterized by comprising the following steps: Geological and structural response data in the tunnel construction process are obtained, and the data are cleaned and characteristics are enhanced; Extracting primary fusion characteristics of the image of the data by using a dynamic transverse expansion network learning model, mapping the primary fusion characteristics into characteristic nodes by random weights, and randomly mapping part of the characteristic nodes into enhancement nodes; responding to the tunnel construction condition change, triggering an increment learning mechanism, increasing the characteristic nodes and the enhancement nodes, and newly adding acquired data, and dynamically adjusting the quantity of the characteristic nodes and the enhancement nodes according to the characteristic distribution and the complexity of the newly added data; and taking the characteristic nodes and the enhancement nodes as input, and carrying out real-time discrimination and risk level assessment on the catastrophe type by using a dynamic geological disaster identification model.
2. The method for dynamically identifying geological disasters in tunnel construction according to claim 1, wherein the process of acquiring geological and structural response data in the tunnel construction process comprises the steps of acquiring video data of each key section of a tunnel, extracting key frame images from the video data, and preprocessing the images.
3. The method for dynamically identifying geological disasters in tunnel construction according to claim 1, wherein the process of cleaning and enhancing the characteristics of the data comprises the steps of carrying out standardization processing on an original image, enhancing the overall and local contrast of the image by limiting contrast self-adaptive histogram equalization, removing image noise by utilizing a method of combining wavelet transformation filtering and median filtering, and retaining edge details.
4. The method for dynamically identifying tunnel construction geological disasters is characterized in that a dynamic transverse expansion network learning model is utilized, the process of extracting primary image fusion features of data comprises a primary feature extraction module and a dynamic transverse expansion learning network, wherein the primary feature extraction module extracts primary image fusion features by adopting a convolution structure with fixed weight, each layer of convolution comprises a plurality of convolution kernels capable of extracting different image features, the obtained features serve as input of the dynamic transverse expansion learning network, the obtained features are mapped into feature nodes through random weights, all feature nodes are further mapped into enhancement nodes at random, and the number of the enhancement nodes is determined through Bayesian algorithm optimization.
5. The method for dynamically identifying geological disasters in tunnel construction according to claim 4, wherein the process of optimizing and determining the number of the enhanced nodes by using a Bayesian algorithm comprises the steps of performing parameter searching by using a Bayesian optimization method, iteratively sampling in a super-parameter space by constructing a probability agent model, and updating posterior distribution according to a historical evaluation result so as to guide the search to converge towards a better area until an optimal parameter configuration is found.
6. The method of claim 1, wherein the step of triggering the incremental learning mechanism in response to a change in tunnel construction conditions includes triggering the incremental learning mechanism when a change in construction phase or geological conditions is detected when the confidence of identification continues a plurality of times when the confidence of identification is below a set threshold, and when a visual pattern is identified that has not occurred in the training set and the number of consecutive occurrences of the visual pattern exceeds the threshold.
7. The method of claim 1, wherein the process of adding feature nodes and enhancement nodes comprises setting n feature nodes and m enhancement nodes, generating newly added n+1th feature nodes by random mapping of current image features, and correspondingly adding newly added feature mapping enhancement nodes or adding m+1th independent enhancement nodes.
8. The method for dynamically identifying geological disasters in tunnel construction according to claim 1, wherein the process of dynamically adjusting the number of feature nodes and enhancement nodes according to the feature distribution and complexity of the newly-acquired data comprises the steps of acquiring monitoring image data of a recent construction site, screening qualified samples, carrying out standardization, denoising and image enhancement processing on the samples to form an incremental training sample set, dynamically adjusting the number of feature nodes and enhancement nodes according to the feature distribution and complexity of the newly-acquired data based on an incremental expansion mechanism of a dynamic transverse expansion network learning model, and updating output weights by using a recursive pseudo-inverse updating algorithm, wherein the updating formula is as follows: ; Wherein, the For the feature mapping matrix corresponding to the newly added sample, For its corresponding tag matrix, Is the weight of the original weight of the steel plate, The matrix is updated for the intermediate obtained by the recursive calculation.
9. The method for dynamically identifying the geological disaster in the tunnel construction according to claim 1, wherein the process of carrying out real-time discrimination and risk level assessment on the disaster type by utilizing the dynamic identification model of the geological disaster comprises the steps of comprehensively weighting and calculating four dimensions of identification confidence, disaster type, spatial position and time evolution trend to obtain a comprehensive risk value, and determining corresponding risk early warning grades according to a risk threshold interval where the comprehensive risk value is located.
10. A tunnel construction geological disaster dynamic identification system is characterized by comprising: the data acquisition module is configured to acquire geological and structural response data in the tunnel construction process, and clean and enhance the characteristics of the data; The dynamic transverse expansion module is configured to extract primary fusion characteristics of the image of the data by utilizing a dynamic transverse expansion network learning model, map the primary fusion characteristics to characteristic nodes through random weights, and randomly map part of the characteristic nodes to enhancement nodes; The incremental learning module is configured to respond to the tunnel construction condition change, trigger an incremental learning mechanism, increment the feature nodes and the enhancement nodes, newly increase acquired data, and dynamically adjust the number of the feature nodes and the enhancement nodes according to the feature distribution and the complexity of the newly increased data; The dynamic identification module of the geological disaster is configured to take the characteristic nodes and the enhancement nodes as input, and the dynamic identification model of the geological disaster is utilized to carry out real-time discrimination and risk level assessment of the disaster type.

Description

Dynamic identification method and system for tunnel construction geological disasters Technical Field The invention belongs to the technical field of geological disaster identification, and particularly relates to a tunnel construction geological disaster dynamic identification method and system. Background The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art. The dynamic identification, prevention and control of geological disasters are widely focused in construction, and the dynamic processes such as crack development, seepage mutation, surrounding rock degradation and the like in the rock mass are performed in tunnel tunneling under complex geological conditions, so that the method has basic and key effects on evaluating construction safety and early warning of water and mud bursting and collapse risks. The modern tunnel construction has widely laid intelligent sensing devices such as video monitoring, three-dimensional laser scanning, multi-element sensors and the like, and has accumulated massive heterogeneous real-time monitoring data, but the current intelligent recognition method based on deep learning has high precision, but the bottlenecks such as high model training cost, difficult iterative updating, difficult adaptation to dynamically changing geological conditions in construction and the like generally exist, and the intelligent level and the engineering dynamic adaptability are seriously insufficient. Disclosure of Invention In order to solve the problems, the invention provides a method and a system for dynamically identifying the geological disasters in tunnel construction. According to some embodiments, the present invention employs the following technical solutions: A dynamic identification method for geological disasters in tunnel construction comprises the following steps: Geological and structural response data in the tunnel construction process are obtained, and the data are cleaned and characteristics are enhanced; Extracting primary fusion characteristics of the image of the data by using a dynamic transverse expansion network learning model, mapping the primary fusion characteristics into characteristic nodes by random weights, and randomly mapping part of the characteristic nodes into enhancement nodes; responding to the tunnel construction condition change, triggering an increment learning mechanism, increasing the characteristic nodes and the enhancement nodes, and newly adding acquired data, and dynamically adjusting the quantity of the characteristic nodes and the enhancement nodes according to the characteristic distribution and the complexity of the newly added data; and taking the characteristic nodes and the enhancement nodes as input, and carrying out real-time discrimination and risk level assessment on the catastrophe type by using a dynamic geological disaster identification model. In an alternative embodiment, the process of acquiring geological and structural response data in the tunnel construction process comprises the steps of acquiring video data of each key section of a tunnel, extracting key frame images from the video data, and preprocessing the images. In an alternative embodiment, the process of cleaning and feature enhancement of the data comprises the steps of carrying out standardization treatment on an original image, enhancing the overall and local contrast of the image through limiting contrast self-adaptive histogram equalization, removing image noise by utilizing a method of combining wavelet transformation filtering and median filtering, and retaining edge details. The method comprises the steps of selecting a dynamic transverse expansion network learning model, wherein the dynamic transverse expansion network learning model comprises a primary feature extraction module and a dynamic transverse expansion learning network, the primary feature extraction module adopts a convolution structure with fixed weight to extract primary fusion features of the image, each layer of convolution comprises a plurality of convolution kernels capable of extracting different image features, the obtained features are used as input of the dynamic transverse expansion learning network, the obtained features are mapped into feature nodes through random weights, all feature nodes are further mapped into enhancement nodes at random, and the number of the enhancement nodes is determined through Bayesian algorithm optimization. In a further embodiment, the process of optimizing and determining the number of the enhancement nodes by using a Bayesian algorithm comprises the steps of carrying out parameter searching by adopting a Bayesian optimization method, iteratively sampling in a super-parameter space by constructing a probability agent model, and updating posterior distribution according to a historical evaluation result so as to guide the searching to converge towards a better a