CN116611592B - Prediction method for geothermal abnormal region along railway corridor based on deep learning - Google Patents
Prediction method for geothermal abnormal region along railway corridor based on deep learning Download PDFInfo
- Publication number
- CN116611592B CN116611592B CN202310901926.7A CN202310901926A CN116611592B CN 116611592 B CN116611592 B CN 116611592B CN 202310901926 A CN202310901926 A CN 202310901926A CN 116611592 B CN116611592 B CN 116611592B
- Authority
- CN
- China
- Prior art keywords
- geothermal
- sample
- abnormal region
- processing module
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000002159 abnormal effect Effects 0.000 title claims abstract description 58
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000013135 deep learning Methods 0.000 title claims abstract description 14
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 30
- 230000006870 function Effects 0.000 claims abstract description 14
- 230000001133 acceleration Effects 0.000 claims abstract description 7
- 238000004364 calculation method Methods 0.000 claims description 42
- 238000012545 processing Methods 0.000 claims description 24
- 238000012549 training Methods 0.000 claims description 13
- 230000010365 information processing Effects 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 8
- 238000012360 testing method Methods 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 230000004927 fusion Effects 0.000 abstract description 5
- 238000000605 extraction Methods 0.000 abstract description 4
- 238000013528 artificial neural network Methods 0.000 abstract 2
- 238000011160 research Methods 0.000 description 11
- 238000010276 construction Methods 0.000 description 9
- 230000006378 damage Effects 0.000 description 7
- 238000011156 evaluation Methods 0.000 description 5
- 230000008447 perception Effects 0.000 description 5
- 208000027418 Wounds and injury Diseases 0.000 description 4
- 230000005856 abnormality Effects 0.000 description 4
- 238000003066 decision tree Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 4
- 208000014674 injury Diseases 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 238000009933 burial Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000011176 pooling Methods 0.000 description 3
- 230000035945 sensitivity Effects 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000011835 investigation Methods 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 238000007477 logistic regression Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- HPTJABJPZMULFH-UHFFFAOYSA-N 12-[(Cyclohexylcarbamoyl)amino]dodecanoic acid Chemical compound OC(=O)CCCCCCCCCCCNC(=O)NC1CCCCC1 HPTJABJPZMULFH-UHFFFAOYSA-N 0.000 description 1
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000002238 attenuated effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000004566 building material Substances 0.000 description 1
- 230000002542 deteriorative effect Effects 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000012502 risk assessment Methods 0.000 description 1
- 238000013058 risk prediction model Methods 0.000 description 1
- 239000011435 rock Substances 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 230000003685 thermal hair damage Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E10/00—Energy generation through renewable energy sources
- Y02E10/10—Geothermal energy
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Business, Economics & Management (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Operations Research (AREA)
- Entrepreneurship & Innovation (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application discloses a prediction method for geothermal abnormal areas along a railway corridor based on deep learning, which comprises the following steps: constructing a geothermal sample data set by taking hot spring points of an area to be predicted as geothermal sample points; taking the surface temperature, fault density, stratum combination entropy, earthquake motion peak acceleration, aeromagnetic anomaly, moholo surface depth, geothermal flow value and river density as prediction influencing factors; constructing a geothermal abnormal region prediction model, wherein the geothermal abnormal region prediction model is realized by adopting a convolutional neural network fused with position information; and inputting geothermal sample points of the geothermal sample data set into a geothermal abnormal region prediction model to obtain a geothermal abnormal region prediction result. The application provides a geothermal abnormal region prediction model based on a convolution neural network of fusion position information, which introduces a new space coordinate attention on the basis of the original sample information extraction function, improves the space position information sensing capability of the convolution neural network of fusion position information and can improve the prediction accuracy.
Description
Technical Field
The application relates to the technical field of engineering geological disaster risk analysis, in particular to a prediction method for geothermal abnormal areas along a railway corridor based on deep learning.
Background
Along with the extension of China railway traffic to mountain areas with difficult terrain and complicated geology, tunnel engineering is developed towards the deep-buried special lengthening direction, and more complicated geological units need to be penetrated by the tunnel body. In particular, in recent years, mountain tunnel engineering on Qinghai-Tibet plateau in China needs to pass through the problem of difficult-to-bypass high-temperature heat injury. The tunnel high temperature heat damage mainly refers to the damage caused by Gao Yanwen and high water temperature to tunnel construction. Constructing a deep buried extra-long tunnel in a region where high-temperature heat and damp heat are liable to occur, and deteriorating the tunnel construction environment, so that personnel cannot normally operate in the tunnel; moreover, the tunnel lining structure, the building material and the construction process under the normal temperature condition are possibly not suitable for tunnel engineering in high-ground-temperature areas, so that the safety and the durability of the tunnel structure are greatly threatened, and the normal operation environment of the tunnel can be deteriorated by serious people to influence the normal use function of the tunnel. The serious high-temperature heat damage problem is encountered in the construction process of the Sang Zhu-ridge tunnel in the south valley of China, the highest ground temperature is 89.6 ℃ when the maximum burial depth is 1500m, and the tunnel is the highest rock temperature tunnel encountered in the construction process of the railway tunnel in China at present. The same problem occurs in other countries, such as the maximum ground temperature of 63.8 ℃ for the maximum burial depth of 2000m for the sub-parallel railway tunnel of italy; the maximum ground temperature at the maximum burial depth of the japanese house highway tunnel is 75.5 ℃.
Geothermal anomaly prediction is an effective method for preventing tunnel thermal damage by evaluating the spatial distribution of the probability of occurrence of geothermal anomalies in a particular region based on local geological environment factors. There are many methods for determining the abnormal state of geothermal heat, and they are mainly classified into qualitative and quantitative models. The qualitative evaluation adopts expert experience to carry out index weight assignment, is suitable for predicting the geothermal abnormal region of the small region, and has stronger subjectivity. As the research goes deep, the research method gradually shifts to semi-quantitative and quantitative models.
Deep learning is a hotspot in the field of machine learning research, has advantages in the inherent law and representation level of learning sample data, and has been widely applied to the fields of natural disaster risk area prediction and the like. The most typical model in deep learning is a convolutional neural network (convolutional neural network, CNN), and the characteristics of local connection, weight sharing, pooling operation and the like of the CNN model can reduce the complexity of the network and training parameters, are easy to train and optimize, and have been widely applied in other fields. Compared with the traditional machine learning method, the CNN can effectively extract deep features of the image through the convolution layer and well solve the nonlinear problem. However, most of the current researches predict based on single pixel values of each factor layer where geothermal sample points are located, and spatial position relations among geothermal samples are often ignored, so that the prediction accuracy cannot be guaranteed.
Disclosure of Invention
The application aims to solve the problem that the accuracy cannot be guaranteed due to neglecting the spatial position relation between geothermal samples when geothermal abnormal areas are predicted in the prior art, and provides a method for predicting geothermal abnormal areas along a railway corridor based on deep learning.
The aim of the application is mainly realized by the following technical scheme:
the prediction method for the geothermal abnormal region along the railway corridor based on deep learning comprises the following steps of:
s1, constructing a geothermal sample data set by taking hot spring points of an area to be predicted as geothermal sample points;
s2, taking the surface temperature, fault density, stratum combination entropy, earthquake motion peak acceleration, aeromagnetic anomaly, moghao depth, geothermal flow value and river density as prediction influence factors;
s3, constructing a geothermal abnormal region prediction model, wherein the geothermal abnormal region prediction model is realized by adopting a convolutional neural network fused with position information; the geothermal abnormal region prediction model is provided with a geothermal abnormal influence factor processing module and a geothermal sample space information processing module, wherein the geothermal abnormal influence factor processing module is used for inputting a predicted influence factor and performing convolution calculation, is introduced into a calculation module of a Token Mixer in a transformation, and adopts the following steps ofSoftmaxFor a pair ofNormalizing according to the second dimension to obtain abstract high-level features; the geothermal sample space information processing module is used for inputting geothermal sample points, reading space coordinates of the geothermal sample points, extracting mutual distances between every two geothermal sample points to obtain geothermal sample space distribution information, and weighting and calculating abstract high-level features obtained by the geothermal sample space distribution information on the geothermal anomaly impact factor processing module;
and S4, inputting geothermal sample points of the geothermal sample data set into a geothermal abnormal region prediction model to obtain a geothermal abnormal region prediction result.
Further, the formula of convolution calculation of the geothermal anomaly impact factor processing module is as follows:
the above expression is fromLayer to->Layer calculation process, wherein ∈>Indicate->Layer characteristics,/->Indicate->Layer characteristics,/->Indicate->Bias term of layer->Indicate->Weight parameters of layer convolution +_>Indicate->Intermediate variables for layer without activation function calculation, +.>To activate the function.
Further, the calculation formula of the Token Mixer is as follows:
in the formula, Q, K and V are taken asThe input is provided with a key-in,,/> , ,wherein->C is the number of channels, ">Representing dot product;
in the geothermal anomaly impact factor processing moduleSoftmaxThe calculation formula of (2) is as follows:
wherein X representsSoftmaxIs used for the input data of the (a),len(X)representation ofXThe number of attributes of the data, i.e. the aboveC。
Further, the specific processing steps of the geothermal sample spatial information processing module in step S3 include:
the space coordinates of geothermal sample points are read, the space distance between every two batch samples is calculated, the space distance is mapped by using Gaussian distribution, and the corresponding calculation formula is as follows:
where n represents the number of geothermal sample points,spatial distribution information representing the geothermal sample;
the space distribution information of the geothermal sample and the abstract high-level features obtained after calculation by the geothermal anomaly impact factor processing module are subjected to matrix multiplication, and mapping is carried out by using a full-connection layer, so that a final sample prediction result is obtained; the corresponding calculation formula is as follows:
where y represents the final sample prediction result,Linearrepresenting a fully connected layer.
Further, the step S1 further includes the following steps:
and randomly selecting the same number of non-geothermal points outside a buffer zone with a radius value set from geothermal sample points to construct a non-geothermal sample data set, and randomly dividing the geothermal sample points and the non-geothermal sample points according to a set proportion to obtain a training data set for training a model and a test data set for verifying model prediction accuracy.
In summary, compared with the prior art, the application has the following beneficial effects: (1) When the method is applied, the frequency ratio of the high anomaly area and the extremely high anomaly area is higher than that of the traditional model, the maximum frequency ratio is provided, and the partitioning result is more in accordance with the spatial distribution of the ground hot spots. Therefore, compared with the traditional model, the prediction accuracy can be further improved, and the prediction method is more suitable for the prediction work of the geothermal abnormal region.
(2) The Area Under Curve (AUC) value, the Precision, the Recall ratio (Recall), the F1 fraction (F1-Socre) and the receiver operation characteristic curve (ROC) are all higher than those of the traditional model, so that the prediction Precision of the geothermal abnormal region by using the model provided by the application is better than that of the traditional model, and meanwhile, the deep learning method is far better than that of the traditional machine learning method, so that the nonlinear relation between geothermal abnormal development and the influence factors thereof can be fitted more accurately. The result shows that the method and the device integrate the spatial information of the samples, integrate the attribute of the geothermal samples into the influence of the spatial position, improve the relation perception of the whole model on the geothermal samples to a certain extent, and finally predict the result more accurately, thereby having a certain guiding significance on the prediction research of geothermal abnormal areas.
(3) And qualitatively analyzing that the geothermal points in the geothermal abnormal region prediction graph obtained by using the model provided by the application are basically distributed in the high abnormal region and the extremely high abnormal region.
(4) When the method is applied, the obtained prediction result can be finally made into the prediction area geothermal anomaly hierarchical graph, the result well reflects the distribution current situation of the prediction area geothermal anomaly development, and scientific basis can be provided for disaster prevention and reduction work in the railway corridor planning construction and future safe operation process.
Drawings
The accompanying drawings, which are included to provide a further understanding of embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings:
FIG. 1 is a flow chart of an embodiment of the present application;
FIG. 2 is a diagram of a prediction model of a geothermal anomaly area according to an embodiment of the present application;
FIG. 3 is a graph showing geographical location and geothermal point distribution of a study area when an embodiment of the present application is applied;
FIG. 4 is a graph of ROC for a model of an embodiment of the application versus a conventional model application;
FIG. 5 is a graph of predictive hierarchy for a model and a conventional model application in accordance with one embodiment of the present application;
fig. 6 is a view of a key region of a railway corridor when the model of one embodiment of the present application is applied to a conventional model.
Detailed Description
For the purpose of making apparent the objects, technical solutions and advantages of the present application, the present application will be further described in detail with reference to the following examples and the accompanying drawings, wherein the exemplary embodiments of the present application and the descriptions thereof are for illustrating the present application only and are not to be construed as limiting the present application.
Examples:
as shown in fig. 1 and 2, the prediction method for the geothermal abnormal region along the railway corridor based on deep learning comprises the following steps in sequence: s1, constructing a geothermal sample data set by taking hot spring points of an area to be predicted as geothermal sample points; s2, combining the surface temperature, fault density, stratum combination entropy and earthquake motionPeak acceleration, aeromagnetic anomalies, mojo surface depth, geothermal flow value and river density are taken as prediction influencing factors; s3, constructing a geothermal abnormal region prediction model, wherein the geothermal abnormal region prediction model is realized by adopting a convolutional neural network fused with position information; the geothermal abnormal region prediction model is provided with a geothermal abnormal influence factor processing module and a geothermal sample space information processing module, wherein the geothermal abnormal influence factor processing module is used for inputting a predicted influence factor and performing convolution calculation, is introduced into a calculation module of a Token Mixer in a transformation, and adopts the following steps ofSoftmaxFor a pair ofNormalizing according to the second dimension to obtain abstract high-level features; the geothermal sample space information processing module is used for inputting geothermal sample points, reading space coordinates of the geothermal sample points, extracting mutual distances between every two geothermal sample points to obtain geothermal sample space distribution information, and weighting and calculating abstract high-level features obtained by the geothermal sample space distribution information on the geothermal anomaly impact factor processing module; and S4, inputting geothermal sample points of the geothermal sample data set into a geothermal abnormal region prediction model to obtain a geothermal abnormal region prediction result. In this embodiment, the convolutional neural network fusing the position information is implemented based on point-by-point fusion in a multi-modality, such as CLiP.
Since hot springs are natural outages of geothermal anomalies, they are a direct indication of mountain-rising (convection-type) geothermal anomalies on the surface. Thus, the present embodiment constructs a geothermal sample dataset with spa points as geothermal sample points. According to the method, a plurality of geothermal anomaly influence factor data and a plurality of landslide influence factor data are selected according to the cause and distribution rules of geothermal anomalies, the correlation of the selected influence factors is measured by using Spearman coefficients from 4 aspects of geological structure, geophysics, natural climate and hydrologic conditions to judge whether the factors have independence, and according to the specification of the Spearman coefficients, the correlation coefficient meets |R| which is less than or equal to 0.75, the independence is indicated, and the absolute value is closer to 1, so that the correlation among variables is stronger. And finally selecting 8 influencing factors including surface temperature, fault density, stratum combination entropy, earthquake motion peak acceleration, aeromagnetic anomaly, moholothurian depth, geothermal flow value and river density through screening.
Step S1 of the present embodiment further includes the steps of: and randomly selecting the same number of non-geothermal points outside a buffer zone with a radius value set from geothermal sample points to construct a non-geothermal sample data set, and randomly dividing the geothermal sample points and the non-geothermal sample points according to a set proportion to obtain a training data set for training a model and a test data set for verifying model prediction accuracy. In the embodiment, a specific value of a radius value set from a geothermal sample point when a non-geothermal sample data set is constructed is set to be 1km, and a set proportion is set to be 7 when the geothermal sample point and the non-geothermal sample point are randomly divided according to the set proportion: 3. in the embodiment, the same number of non-geothermal points are randomly selected outside a buffer zone with a radius of 1km from the geothermal point to construct a non-geothermal sample data set, and only the non-geothermal sample data set is in a research area, and is out of 1km radius from the geothermal point, and can be 5 km or 10km from the radius of the geothermal point. When the embodiment is applied, instead of selecting one non-geothermal point outside the buffer area with the diameter of 1km from each geothermal point, that is, instead of selecting one non-geothermal point corresponding to each geothermal point, only the total number of the geothermal points is equal to the total number of the non-geothermal points, and 0 or 10 non-geothermal points can be arranged outside the buffer area with the diameter of 1km from each geothermal point. The value of 1km is determined by referring to a study similar to the susceptibility evaluation of geological disasters. The purpose of constructing positive and negative samples in this embodiment is for training of the model so that the model can make decisions on other unknown sample points (geothermal points and non-geothermal points).
According to the embodiment, the spatial distribution information of the geothermal sample and the influence factors are processed by corresponding modules, wherein a convolution calculation formula of a geothermal anomaly influence factor processing module is as follows:
the above expression is fromLayer to->Layer calculation process, wherein ∈>Indicate->Layer characteristics,/->Indicate->Layer characteristics,/->Indicate->Bias term of layer->Indicate->Weight parameters of layer convolution +_>Indicate->Intermediate variables for layer without activation function calculation, +.>For activating the function, the activation function of this embodiment is implemented using a ReLU activation function.
In this embodiment, the calculation formula of the calculation module of Token Mixer is:
wherein Q, K and V are used as inputs,,/> , ,wherein->C is the number of channels, ">Representing dot product.
In the geothermal anomaly impact factor processing module in this embodimentSoftmaxThe calculation formula of (2) is as follows:
wherein X representsSoftmaxIs used for the input data of the (a),len(X)representation ofXThe number of attributes of the data, i.e. the aboveC。
In the present embodimentSoftmaxFor aligningNormalization is performed according to the second dimension, so that the relationship between the perceived geothermal samples is improved, the information extraction capacity of the model on the geothermal samples in the same batch is improved, and the overall information extraction capacity and robustness of the model are further improved.
The specific processing steps of the geothermal sample spatial information processing module in step S3 of this embodiment include: the space coordinates of geothermal sample points are read, the space distance (more than 0) between every two batch samples is calculated, the space distance is mapped by using Gaussian distribution, and the corresponding calculation formula is as follows:
where n represents the number of geothermal sample points,the space distribution information of the geothermal samples is represented, and the larger the distance between the samples is, the more the mapped result is approaching to 1;
the space distribution information of the geothermal sample and the abstract high-level features obtained after calculation by the geothermal anomaly impact factor processing module are subjected to matrix multiplication, and mapping is carried out by using a full-connection layer, so that a final sample prediction result is obtained; the corresponding calculation formula is as follows:
where y represents the final sample prediction result,Linearrepresenting a fully connected layer.
According to the embodiment, dot multiplication calculation is carried out through the geothermal sample space attribute and the sample evidence factor, the dot multiplication calculation is similar to attention weighted calculation, compared with a common CNN module, the dot multiplication calculation method has a specific space information processing module, and space information of a geothermal sample is well integrated into an integral CNN model, so that the perception of the model on the space position is improved.
In the training process of the model, the parameters of the whole model are optimized mainly by adopting a Mean Square Error (MSE) and a second-class cross entropy loss function (BCE), and the calculation formulas are as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,a label representing the ith sample, +.>Representing the predicted result of the ith sample, the computational loss of the model as a whole +.>The following formula, α, takes a value of 0.5 in this embodiment.
The embodiment provides a convolutional neural network (Location Information Fusion Convolutional Neural Network, LI ⁃ CNN) integrating position information, which introduces a new space coordinate attention on the basis of the original sample information extraction function and improves the space position information perception capability of the LI-CNN. Meanwhile, the interrelationship among the geothermal anomaly impact factors is calculated according to the high-dimensional abstract features of the geothermal anomaly impact factors, so that the final prediction effect of the model is improved. The model of the embodiment mainly comprises a convolution layer, an activation function, a pooling layer and a full-connection layer, and comprises a geothermal anomaly influence factor processing module and a geothermal sample space information processing module, wherein the geothermal anomaly influence factor processing module is similar to a conventional CNN structure and is mainly used for extracting information of geothermal properties of samples to obtain abstract high-level features, the geothermal sample space information processing module is mainly used for extracting mutual distances among samples according to sample distribution in the same batch, and further weighting calculation is achieved on the abstract high-level features obtained by the geothermal evidence factor processing module by using sample space distribution information. In this embodiment, the convolution layer is used to extract information of data, and the ReLU activation function is used in combination with the convolution layer to promote overall nonlinearity of the model; the pooling layer is used for downsampling of the data shape, and can improve robustness of the model; the fully connected layer is used for the classification of final geothermal properties. The tokennixer of the embodiment is used for improving information perception of the model on all input factors.
One specific application example of the application of this embodiment is as follows:
the railway on a certain plateau passes through two geothermal hot spring zones from east to west, the regional hydrothermal activity is frequent and strong, and the deep-buried tunnel is extremely at risk of generating high-temperature heat injury. High temperature heat injury is a common geological disaster in underground engineering, and seriously threatens the life health of constructors and the safety and durability of the engineering during construction and operation. Therefore, the method has important practical significance for effectively reducing the influence of high-temperature heat injury on railway construction maintenance and developing the prediction work of geothermal abnormal areas along the railway. The outline of the investigation region of this example railway corridor geothermal-energy abnormal region prediction selection of the present embodiment is shown in fig. 3.
And (3) sharing a geothermal hot spring 95 in the research area, constructing a geothermal sample data set by utilizing the position of the geothermal hot spring 95, and constructing a non-geothermal sample data set by randomly selecting the same number of non-geothermal points outside a buffer area with a diameter of 1km from the geothermal points. And then, randomly dividing the geothermal sample points and the non-geothermal sample points according to the proportion of 7:3 to obtain a training data set for training the model and a test data set for verifying the model prediction accuracy.
According to the cause and distribution rule of geothermal anomalies, 8 influencing factors including surface temperature, fault density, stratum combination entropy, earthquake peak acceleration, aeromagnetic anomalies, mohopanax depth, geothermal flow value and river density are selected from 4 aspects of geological structure, geophysics, natural climate and hydrologic conditions respectively. And (3) judging whether the factors have independence or not by utilizing the Spearman coefficient to measure the correlation of the selected influence factors, and finally determining the surface temperature, fault density, stratum combination entropy, earthquake motion peak acceleration, aeromagnetic anomaly, mohopanax depth, geothermal flow value and river density influence factors for subsequent research through screening.
To further verify the ability of the LI-CNN model proposed in this example, it was compared with 5 traditional models of Logistic Regression (LR), decision Tree (DT), K Nearest Neighbor (KNN), distributed gradient hoist (XGBoost), and Convolutional Neural Network (CNN). This example uses the Ubuntu18.04 operating system using Python language PyTorr 1.71 (Facebook AI Research, new York, NY, USA) deep-learning framework, version of CUDA 11.1. The CPU is 12th Gen Intel (R) Core (TM) i7-12700KF, and the GPU is NVIDIAGeForceRTX 3090. In the training process, an Adam optimizer is selected, 100 epochs are selected, the batch size is 16, the learning rate is 0.001, and the learning rate is attenuated by 10% when 10 epochs are trained.
The evaluation of the model accuracy is an important component of the prediction modeling of the geothermal abnormal region, and in order to verify the prediction capabilities of the LI-CNN model and the 5 comparison models, the evaluation results are verified by selecting a frequency ratio, an accuracy (Precision), a Recall rate (Recall), an F1 fraction (F1-Socre) and an ROC curve. The test samples were 30% geothermal samples that did not participate in training.
The prediction results are measured in terms of a confusion matrix, wherein the number of Positive classes predicted as Positive classes is denoted by Ture Positive (TP), the Negative classes predicted as Negative classes is denoted by Ture Negative (TN), the number of Negative classes predicted as Positive classes is denoted by False Positive (FP), and the number of Positive classes predicted as Negative classes is denoted by False Negative (FN).
Precision is used to indicate how many of the positive cases predicted by the classifier are true positive cases, i.e. how many of the positive cases predicted by the classifier are accurate, and the calculation formula is:
recall indicates the proportion of positive samples for which the samples were correctly determined to be the total positive samples. The calculation formula is as follows:
the two indexes of Precision and Recall are usually equal to each other, and are difficult to combine, and the embodiment is expressed by a weighted and harmonic mean value F1-score of the two indexes. The calculation formula is as follows:
the DT, KNN, LG, XGBoost, CNN model and the LI-CNN model evaluation indexes proposed in the embodiment are shown in Table 1, and it can be seen that Precision, recall and F1-score of LI-CNN in the embodiment are significantly better than those of other models. The LI-CNN is disclosed to overcome the limitations of other models, and the comprehensive prediction performance is obviously improved through the LI-CNN fused with the ground hot spot position information.
ROC can reflect the correlation between specificity and sensitivity, and is widely applied to evaluating the merits of disaster prediction models. Specificity is also called False Positive Rate (FPR), sensitivity is also called TruePositiveRate (TPR), and a calculation formula is as follows:
ROC is on the abscissa of specificity and on the ordinate of sensitivity. ROC is typically above y=x, so AUC typically ranges in the interval 0.5, 1. The closer the ROC is to the upper left corner, i.e. the greater the AUC, the better the performance of the landslide risk prediction model.
As shown in FIG. 4, the AUC values of the ROC curves of the 6 models are respectively 0.72, 0.77 and 0.81, and the AUC values of the LI-CNN models are 0.04 higher than those of the XGBoost model and the CNN model and 0.09 higher than those of the DT model, the KNN model and the LG model, so that the prediction precision of geothermal abnormal areas by using the proposed models is better than that of the traditional models, and meanwhile, the deep learning method is far better than that of the traditional machine learning method, so that the nonlinear relation between geothermal abnormal development and influence factors of the geothermal abnormal development can be fitted more accurately. The result shows that the LI-CNN is integrated with the attribute of the geothermal sample by the fusion sample space information module and is influenced by the space position, so that the relation perception of the whole model to the geothermal sample is improved to a certain extent, the final prediction result is more accurate, and the method has a certain guiding significance on the prediction research of the geothermal abnormal region.
And predicting the grid units by using the trained model to obtain the probability that each grid unit is predicted to have geothermal abnormality, namely, the risk index, wherein the larger the risk index is, the more easily the geothermal abnormality occurs in the region, and otherwise, the more difficult the geothermal abnormality occurs. The generated research area risk map is divided into 5 grades of extremely low abnormal area, medium abnormal area, high abnormal area and extremely high abnormal area by using a natural break method.
The classification of the geothermal abnormal region of the investigation region obtained using the 6 models is shown in fig. 5. It can be seen that the prediction results of the models have a certain similarity in spatial distribution, most areas in the geothermal anomaly area map are predicted to be low anomaly areas and extremely low anomaly areas, and the geothermal points are basically distributed in the high anomaly areas and the extremely high anomaly areas. In addition, there is a few differences in the 6 partition maps, e.g., the extremely high anomaly area of LI-CNN is minimal.
And (3) carrying out important area analysis on the geotherm abnormal area diagram of the certain plateau railway generated by the optimal model, as shown in a figure 6. It can be seen that a certain plateau railway has 3 geothermal anomaly areas in total: a) In the North southwest direction, two rivers run through and the hot springs around are densely distributed along with two fractures. B) Is in the North southwest direction, is spread parallel to one fracture, has three rivers flowing through, and has dense surrounding hot springs. C) In the North southwest direction, the hot springs are distributed in parallel with three fractures, vertically distributed with one fracture, and have two rivers flowing through and are densely distributed around.
The 3 geothermal abnormal key areas are densely distributed in water and have multiple deep river valleys, the fracture activity is strong, the fracture and fault fracture zone is used as a water guide structure, favorable conditions are provided for the circulation and transportation of deep heat sources, the surface temperature is high, and the method is a section which needs to be used for preventing geothermal abnormality in the construction process of a certain plateau railway.
The foregoing description of the embodiments has been provided for the purpose of illustrating the general principles of the application, and is not meant to limit the scope of the application, but to limit the application to the particular embodiments, and any modifications, equivalents, improvements, etc. that fall within the spirit and principles of the application are intended to be included within the scope of the application.
Claims (2)
1. The prediction method for the geothermal abnormal region along the railway corridor based on deep learning is characterized by comprising the following steps of:
s1, constructing a geothermal sample data set by taking hot spring points of an area to be predicted as geothermal sample points;
s2, taking the surface temperature, fault density, stratum combination entropy, earthquake motion peak acceleration, aeromagnetic anomaly, moghao depth, geothermal flow value and river density as prediction influence factors;
s3, constructing a geothermal abnormal region prediction model, wherein the geothermal abnormal region prediction model is realized by adopting a convolutional neural network fused with position information; the geothermal abnormal region prediction model is provided with a geothermal abnormal influence factor processing module and a geothermal sample space information processing module, wherein the geothermal abnormal influence factor processing module is used for inputting a predicted influence factor and performing convolution calculation, is introduced into a calculation module of a Token Mixer in a transformation, and adopts the following steps ofSoftmaxFor a pair ofNormalizing according to the second dimension to obtain abstract high-level features; the geothermal sample space information processing module is used for inputting geothermal sample points, reading space coordinates of the geothermal sample points, extracting mutual distances between every two geothermal sample points to obtain geothermal sample space distribution information, and weighting and calculating abstract high-level features obtained by the geothermal sample space distribution information on the geothermal anomaly impact factor processing module;
s4, inputting geothermal sample points of the geothermal sample data set into a geothermal abnormal region prediction model to obtain a geothermal abnormal region prediction result;
the convolution calculation formula of the geothermal anomaly impact factor processing module is as follows:
the above expression is fromLayer to->Layer calculation process, wherein ∈>Indicate->Layer characteristics,/->Represent the firstLayer characteristics,/->Indicate->Bias term of layer->Indicate->Weight parameters of layer convolution +_>Indicate->Intermediate variables for layer without activation function calculation, +.>Is an activation function;
the calculation formula of the calculation module of the Token Mixer is as follows:
wherein Q, K and V are used as inputs, ,/> , /> ,wherein the method comprises the steps ofC is the number of channels, ">Representing dot product;
in the geothermal anomaly impact factor processing moduleSoftmaxThe calculation formula of (2) is as follows:
wherein X representsSoftmaxIs used for the input data of the (a),len(X)representation ofXThe number of attributes of the data, i.e. the aboveC;
The specific processing steps of the geothermal sample space information processing module in the step S3 include:
the space coordinates of geothermal sample points are read, the space distance between every two batch samples is calculated, the space distance is mapped by using Gaussian distribution, and the corresponding calculation formula is as follows:
where n represents the number of geothermal sample points,spatial distribution information representing the geothermal sample;
the space distribution information of the geothermal sample and the abstract high-level features obtained after calculation by the geothermal anomaly impact factor processing module are subjected to matrix multiplication, and mapping is carried out by using a full-connection layer, so that a final sample prediction result is obtained; the corresponding calculation formula is as follows:
where y represents the final sample prediction result,Linearrepresenting a fully connected layer.
2. The method for predicting a geothermal anomaly along a railway corridor based on deep learning according to claim 1, wherein the step S1 further comprises the steps of:
and randomly selecting the same number of non-geothermal points outside a buffer zone with a radius value set from geothermal sample points to construct a non-geothermal sample data set, and randomly dividing the geothermal sample points and the non-geothermal sample points according to a set proportion to obtain a training data set for training a model and a test data set for verifying model prediction accuracy.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310901926.7A CN116611592B (en) | 2023-07-21 | 2023-07-21 | Prediction method for geothermal abnormal region along railway corridor based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310901926.7A CN116611592B (en) | 2023-07-21 | 2023-07-21 | Prediction method for geothermal abnormal region along railway corridor based on deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116611592A CN116611592A (en) | 2023-08-18 |
CN116611592B true CN116611592B (en) | 2023-10-20 |
Family
ID=87675078
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310901926.7A Active CN116611592B (en) | 2023-07-21 | 2023-07-21 | Prediction method for geothermal abnormal region along railway corridor based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116611592B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117312875B (en) * | 2023-10-13 | 2024-04-12 | 河北省气象灾害防御和环境气象中心(河北省预警信息发布中心) | KNN algorithm-based regional high-temperature event similarity discrimination method |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109740830A (en) * | 2017-10-27 | 2019-05-10 | 中国石油化工集团公司 | Geothermal heating system hot water type geothermal well quantity Needs method and system |
CN111912540A (en) * | 2020-06-23 | 2020-11-10 | 中铁第一勘察设计院集团有限公司 | Method for testing and predicting geothermal anomaly tunnel engineering ground temperature in complex and difficult highland mountain areas |
CN115659796A (en) * | 2022-10-21 | 2023-01-31 | 中科卫星应用德清研究院 | Method, device and equipment for predicting geothermal high-temperature abnormal region and readable storage medium |
CN115936234A (en) * | 2022-12-21 | 2023-04-07 | 成都理工大学 | Thin reservoir space distribution prediction method based on deep learning |
CN116451584A (en) * | 2023-04-23 | 2023-07-18 | 广东云湃科技有限责任公司 | Thermal stress prediction method and system based on neural network |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101815356B1 (en) * | 2015-12-24 | 2018-01-04 | 한국전력공사 | Method of estimating the geothermal gradient and apparatus for performing the method |
-
2023
- 2023-07-21 CN CN202310901926.7A patent/CN116611592B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109740830A (en) * | 2017-10-27 | 2019-05-10 | 中国石油化工集团公司 | Geothermal heating system hot water type geothermal well quantity Needs method and system |
CN111912540A (en) * | 2020-06-23 | 2020-11-10 | 中铁第一勘察设计院集团有限公司 | Method for testing and predicting geothermal anomaly tunnel engineering ground temperature in complex and difficult highland mountain areas |
CN115659796A (en) * | 2022-10-21 | 2023-01-31 | 中科卫星应用德清研究院 | Method, device and equipment for predicting geothermal high-temperature abnormal region and readable storage medium |
CN115936234A (en) * | 2022-12-21 | 2023-04-07 | 成都理工大学 | Thin reservoir space distribution prediction method based on deep learning |
CN116451584A (en) * | 2023-04-23 | 2023-07-18 | 广东云湃科技有限责任公司 | Thermal stress prediction method and system based on neural network |
Non-Patent Citations (3)
Title |
---|
Arya Shahdi,etc.Exploratory analysis of machine learning methods in predicting subsurface temperature and geothermal gradient of Northeastern United States.《Geothermal Energy》.2021,第1-22页. * |
杨新亮.拉日铁路吉沃希嘎隧道地热异常特征与防治措施分析.《铁道标准设计》.2014,第58卷(第7期),第107-112页. * |
郝明 等.证据理论融合多视角遥感信息的地热异常探测.《地球科学》.2022,第1-17页. * |
Also Published As
Publication number | Publication date |
---|---|
CN116611592A (en) | 2023-08-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN116611592B (en) | Prediction method for geothermal abnormal region along railway corridor based on deep learning | |
Deng et al. | A general method of spatio-temporal clustering analysis | |
Chen et al. | Resilience assessment of regional areas against earthquakes using multi-source information fusion | |
Gu et al. | Health and safety situation awareness model and emergency management based on multi-sensor signal fusion | |
Arul et al. | Identification of vortex-induced vibration of tall building pinnacle using cluster analysis for fatigue evaluation: Application to Burj Khalifa | |
Niu et al. | Susceptibility assessment of landslides triggered by the Lushan earthquake, April 20, 2013, China | |
Liang et al. | An sand plug of fracturing intelligent early warning model embedded in remote monitoring system | |
Wu et al. | A multi-source information fusion evaluation method for the tunneling collapse disaster based on the artificial intelligence deformation prediction | |
Li et al. | An accurate leakage localization method for water supply network based on deep learning network | |
Chen et al. | A cellular automata approach of urban sprawl simulation with Bayesian spatially-varying transformation rules | |
Chen et al. | Building vulnerability assessment in seismic areas using ensemble learning: A Nepal case study | |
CN106056577A (en) | Hybrid cascaded SAR image change detection method based on MDS-SRM | |
Jiang et al. | Q-method optimization of tunnel surrounding rock classification by fuzzy reasoning model and support vector machine | |
Ma et al. | Nonlinear uncertainty modeling between bridge frequencies and multiple environmental factors based on monitoring data | |
Zhang et al. | Evaluation of the water yield of coal roof aquifers based on the FDAHP‐entropy method: a case study in the Donghuantuo Coal Mine, China | |
Eslamirad et al. | Data generative machine learning model for the assessment of outdoor thermal and wind comfort in a northern urban environment | |
Zhang | Exploration on coal mining-induced rockburst prediction using Internet of things and deep neural network | |
Zhu et al. | Quantitative evaluation of low-permeability gas reservoirs based on an improved fuzzy-gray method | |
Muin et al. | Human-machine collaboration framework for bridge health monitoring | |
CN114298192A (en) | Deep learning model depolarization method based on bias sample pair generation | |
Yang et al. | Research on intelligent identification technology of lithology wireless remote transmission based on element measurement | |
Sui et al. | Application of machine learning to the V s-based soil liquefaction potential assessment | |
Meysami et al. | Estimating the optimal population upper bound for scan methods in retrospective disease surveillance | |
He et al. | Gaussian process model of surrounding rock classification based on digital characterization of rock mass structure and its application | |
Coolbaugh | The prediction and detection of geothermal systems at regional and local scales in Nevada using a geographic information system, spatial statistics, and thermal infrared imagery |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |