CN-121614990-B - Geological disaster early warning method and system based on deep learning

CN121614990BCN 121614990 BCN121614990 BCN 121614990BCN-121614990-B

Abstract

The invention discloses a geological disaster early warning method and system based on deep learning, and relates to the technical field of disaster early warning. According to the invention, static geological attribute data are mapped into geological penetration structure feature vectors, and a channel attention weight matrix aiming at dynamic rainfall time sequence data is generated by utilizing the geological penetration structure feature vectors, so that rainfall data are subjected to weighted modulation, effective hydrological response features constrained by geology are extracted, and finally disaster occurrence probability is calculated and early warning is triggered, so that the accurate distinguishing and dynamic early warning of disaster risks caused by the same rainfall condition under different geological environments are achieved, and the problem that risk differences of the same rainfall under different geological environments cannot be distinguished in the prior art is solved.

Inventors

LIU JIA
CHEN GUILI
ZENG XINXIONG
WU SHUTIAN
JIANG JINJIN
LUO YUECHUN
LAI BO
JIANG SHAN
Zhao fengshun
CHEN KAIMING

Assignees

广东省珠海工程勘察院

Dates

Publication Date: 20260512
Application Date: 20260202

Claims (7)

1. The geological disaster early warning method based on deep learning is characterized by comprising the following steps of: collecting static geological attribute data and dynamic rainfall monitoring time sequence data of a target monitoring area, mapping the static geological attribute data into geological penetration structure feature vectors, and mapping the dynamic rainfall monitoring time sequence data into a hydrological load input sequence; generating a channel attention weight matrix aiming at the hydrologic load input sequence by using the geological penetration structure feature vector, performing weighted modulation calculation on the hydrologic load input sequence through the channel attention weight matrix, and extracting to obtain effective hydrologic response features constrained by geological conditions; the extracting to obtain effective hydrographic response characteristics constrained by geological conditions comprises: Inputting the geological penetration structure feature vector into a full-connection layer network for dimension transformation, and outputting to obtain a geological gating adjustment vector matched with the time step dimension of the hydrologic load input sequence; Performing element-by-element point multiplication operation by using a geological gating adjustment vector and a hydrological load input sequence, and outputting to obtain effective hydrological response characteristics constrained by geological conditions by inhibiting rainfall data in a non-sensitive period and enhancing rainfall data in a sensitive period; The geological penetration structure feature vector is input to a full-connection layer network for dimension transformation, and a geological gating adjustment vector matched with the time step dimension of the hydrologic load input sequence is obtained by output, and the geological gating adjustment vector comprises: extracting a first component representing a rock-soil body permeability coefficient and a second component representing a soil layer thickness from a geological permeability structure characteristic vector; Calculating a time attenuation factor capable of reflecting rainfall infiltration hysteresis effect through the numerical value according to the first component and the second component through an activation function; Performing time sequence expansion on the geological penetration structure feature vector based on the time attenuation factor, and constructing a geological gating adjustment vector with a time dimension attribute; The extracting of the first component representing the rock-soil body permeability coefficient and the second component representing the soil layer thickness in the geological penetration structure feature vector comprises the following steps: Carrying out independent thermal coding and normalization processing on the static geological attribute data to construct an original geological data set containing lithology categories, soil porosity and slope structures; The method comprises the steps of aggregating spatial neighborhood features in an original geological data set through a graph convolutional neural network, extracting potential hidden layer vectors representing the hydrographic characteristics of local micro-landforms of monitoring points, and extracting a first component and a second component from the potential hidden layer vectors; Inputting the effective hydrologic response characteristics into a preset disaster evolution analysis network to obtain a disaster occurrence probability value reflecting the slope stability under the current geological condition; and generating a geological disaster early warning signal through a signal device and executing alarm pushing operation in response to the disaster occurrence probability value exceeding a preset dynamic safety threshold value.
2. The method for pre-warning geological disasters based on deep learning as claimed in claim 1, wherein the outputting obtains effective hydrological response characteristics constrained by geological conditions, and the method comprises the following steps: Identifying rainfall intensity values and soil volume water content values of each time step in the hydrologic load input sequence; Taking a weight value corresponding to the time step in the geological gating adjustment vector as a scaling coefficient, and adjusting the amplitude value of the rainfall intensity value and the soil volume water content value; and (3) retaining time sequence data with the amplitude value higher than a preset noise datum line after amplitude adjustment, removing invalid data, and reconstructing to form effective hydrologic response characteristics constrained by geological conditions.
3. The geological disaster early warning method based on deep learning as claimed in claim 1, wherein the inputting the effective hydrologic response characteristic into a preset disaster evolution analysis network to obtain the disaster occurrence probability value reflecting the slope stability under the current geological condition comprises the following steps: inputting the effective hydrologic response characteristics into a long-period memory network, and extracting a hidden state vector at the current moment reflecting the accumulation effect of the rainfall infiltration process; acquiring the historical maximum effective rainfall bearing capacity of a target monitoring area as a reference contrast characteristic; and calculating the Euclidean distance between the hidden state vector at the current moment and the reference contrast characteristic, mapping the Euclidean distance into a numerical value between 0 and 1, and determining the numerical value as a disaster occurrence probability value.
4. The geological disaster warning method based on deep learning as claimed in claim 3, wherein the obtaining of the historical maximum effective rainfall bearing capacity of the target monitoring area as the reference contrast feature comprises: retrieving disaster-causing rainfall event data in a historical disaster record of a target monitoring area; Invoking geological penetration structure feature vectors to perform weighted backtracking calculation on disaster-causing rainfall event data, and reconstructing critical hydrologic feature vectors at historical disaster-causing time; and (3) carrying out cluster center calculation on the critical hydrologic feature vectors at a plurality of historical disaster-causing moments, and marking the obtained cluster center vectors as reference contrast features.
5. The geological disaster early warning method based on deep learning as claimed in claim 3, wherein the calculating of the euclidean distance between the hidden state vector at the current moment and the reference contrast feature, mapping the euclidean distance to a value between 0 and 1, and determining the value as the occurrence probability value of the disaster, comprises: calculating characteristic residual error module length of the hidden state vector and the reference contrast characteristic at the current moment in a high-dimensional characteristic space; And normalizing the reciprocal of the characteristic residual error module length by using the Sigmoid activation function to obtain a disaster occurrence probability value, wherein the disaster occurrence probability value is higher as the characteristic residual error module length is smaller.
6. The geological disaster early warning method based on deep learning as claimed in claim 1, wherein the generating of the geological disaster early warning signal and the execution of the alarm pushing operation by the signaling means in response to the occurrence probability value of the disaster exceeding a preset dynamic safety threshold value comprises: the method comprises the steps of reading threat object grade data of a current target monitoring area, and matching corresponding dynamic security thresholds from a preset threshold lookup table according to the threat object grade data; triggering an alarm instruction when the disaster occurrence probability value continuously preset time period exceeds a dynamic safety threshold value; and converting the alarm instruction into a digital data packet containing the disaster occurrence probability value and the treatment suggestion, and sending the digital data packet to a preset monitoring terminal.
7. A geological disaster early warning system based on deep learning, which is used for realizing the geological disaster early warning method based on deep learning as set forth in any one of claims 1-5, and is characterized by comprising a data mapping module, a feature extraction module, a probability value analysis module and a disaster early warning module; the data mapping module is used for collecting static geological attribute data and dynamic rainfall monitoring time sequence data of a target monitoring area, mapping the static geological attribute data into geological penetration structure feature vectors and mapping the dynamic rainfall monitoring time sequence data into a hydrological load input sequence; The feature extraction module is used for generating a channel attention weight matrix aiming at the hydrologic load input sequence by utilizing the geological penetration structure feature vector, performing weighted modulation calculation on the hydrologic load input sequence through the channel attention weight matrix, and extracting to obtain effective hydrologic response features constrained by geological conditions; the extracting to obtain effective hydrographic response characteristics constrained by geological conditions comprises: Inputting the geological penetration structure feature vector into a full-connection layer network for dimension transformation, and outputting to obtain a geological gating adjustment vector matched with the time step dimension of the hydrologic load input sequence; Performing element-by-element point multiplication operation by using a geological gating adjustment vector and a hydrological load input sequence, and outputting to obtain effective hydrological response characteristics constrained by geological conditions by inhibiting rainfall data in a non-sensitive period and enhancing rainfall data in a sensitive period; The geological penetration structure feature vector is input to a full-connection layer network for dimension transformation, and a geological gating adjustment vector matched with the time step dimension of the hydrologic load input sequence is obtained by output, and the geological gating adjustment vector comprises: extracting a first component representing a rock-soil body permeability coefficient and a second component representing a soil layer thickness from a geological permeability structure characteristic vector; Calculating a time attenuation factor capable of reflecting rainfall infiltration hysteresis effect through the numerical value according to the first component and the second component through an activation function; Performing time sequence expansion on the geological penetration structure feature vector based on the time attenuation factor, and constructing a geological gating adjustment vector with a time dimension attribute; The extracting of the first component representing the rock-soil body permeability coefficient and the second component representing the soil layer thickness in the geological penetration structure feature vector comprises the following steps: Carrying out independent thermal coding and normalization processing on the static geological attribute data to construct an original geological data set containing lithology categories, soil porosity and slope structures; The method comprises the steps of aggregating spatial neighborhood features in an original geological data set through a graph convolutional neural network, extracting potential hidden layer vectors representing the hydrographic characteristics of local micro-landforms of monitoring points, and extracting a first component and a second component from the potential hidden layer vectors; The probability value analysis module is used for inputting the effective hydrologic response characteristics into a preset disaster evolution analysis network to obtain a disaster occurrence probability value reflecting the slope stability under the current geological condition; the disaster early warning module is used for responding to the disaster occurrence probability value exceeding a preset dynamic safety threshold value, generating a geological disaster early warning signal through the signal device and executing alarm pushing operation.

Description

Geological disaster early warning method and system based on deep learning Technical Field The invention relates to the technical field of disaster early warning, in particular to a geological disaster early warning method and system based on deep learning. Background Along with the improvement of the control precision requirement of geological disasters, the industry is in urgent need of a technology capable of fusing geological conditions and rainfall processes to realize precise early warning. The critical rainfall criterion method of the current main stream relies on a set fixed rainfall threshold to perform early warning, and has the defect that the critical control effects of static geological properties such as lithology, soil layer thickness and the like on the rainfall infiltration and runoff process are completely ignored, so that the risk difference of the same rainfall in different geological environments cannot be distinguished. Still another method is to simply splice geological parameters and rainfall data by adopting a machine learning model and then input the spliced geological parameters and rainfall data into the model for prediction. However, in the method, rainfall data of continuous time sequences often dominate in the model, so that the effectiveness of static geological features is submerged, the model is difficult to learn how to modulate rainfall disaster effects under geological conditions, and early warning accuracy is insufficient. Disclosure of Invention The embodiment of the application solves the problem that the risk difference of the same rainfall under different geological environments cannot be distinguished in the prior art by providing the geological disaster early warning method and the geological disaster early warning system based on deep learning, and realizes the accurate distinguishing and dynamic early warning of disaster risks caused by the same rainfall under different geological environments. The embodiment of the application provides a geological disaster early warning method based on deep learning, which is applied to a geological disaster early warning system based on deep learning and comprises the following steps: collecting static geological attribute data and dynamic rainfall monitoring time sequence data of a target monitoring area, mapping the static geological attribute data into geological penetration structure feature vectors, and mapping the dynamic rainfall monitoring time sequence data into a hydrological load input sequence; generating a channel attention weight matrix aiming at the hydrologic load input sequence by using the geological penetration structure feature vector, performing weighted modulation calculation on the hydrologic load input sequence through the channel attention weight matrix, and extracting to obtain effective hydrologic response features constrained by geological conditions; Inputting the effective hydrologic response characteristics into a preset disaster evolution analysis network to obtain a disaster occurrence probability value reflecting the slope stability under the current geological condition; and generating a geological disaster early warning signal through a signal device and executing alarm pushing operation in response to the disaster occurrence probability value exceeding a preset dynamic safety threshold value. Further, the extracting obtains effective hydrographic response characteristics constrained by geological conditions, including: Inputting the geological penetration structure feature vector into a full-connection layer network for dimension transformation, and outputting to obtain a geological gating adjustment vector matched with the time step dimension of the hydrologic load input sequence; element-by-element point multiplication operation is performed by using the geological gating adjustment vector and the hydrologic load input sequence, and effective hydrologic response characteristics constrained by geological conditions are obtained through inhibiting rainfall data in a non-sensitive period and enhancing the rainfall data in a sensitive period. Further, the inputting the geological penetration structure feature vector to the full-connection layer network for dimension transformation, outputting to obtain a geological gating adjustment vector matched with the time step dimension of the hydrologic load input sequence, includes: extracting a first component representing a rock-soil body permeability coefficient and a second component representing a soil layer thickness from a geological permeability structure characteristic vector; Calculating a time attenuation factor capable of reflecting rainfall infiltration hysteresis effect through the numerical value according to the first component and the second component through an activation function; And (3) carrying out time sequence expansion on the geological penetration structure feature vector based on the time attenuation factor, and constructin