A transient electromagnetic quick advance prediction method

Abstract

The application belongs to the technical field of tunnel advanced prediction, and discloses a transient electromagnetic rapid advanced prediction method, which is based on actual exploration needs and sets an observation system; a large number of theoretical tunnel resistivity models are randomly generated; numerical calculation is performed on the tunnel resistivity models based on the observation system and a numerical simulation algorithm, and transient electromagnetic field data corresponding to each model is obtained; a residual neural network model is constructed, a loss function is designed, and the obtained transient electromagnetic field data corresponding to each model is used to train ResNet; the trained ResNet model is used to predict new observation data, and tunnel TEM data inversion results are obtained. Compared with the existing transient electromagnetic tunnel advanced prediction technology, the transient electromagnetic rapid advanced prediction method is more accurate, and the possibility of false anomalies in inversion is reduced; the inversion can be completed within 1 second, which is more efficient and lays a foundation for real-time tunnel advanced prediction.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of tunnel advance prediction, and particularly relates to a transient electromagnetic rapid advance prediction method. BACKGROUND

[0002] At present, the transient electromagnetic method (TEM) is a commonly used method for tunnel advance prediction. It is a time-domain electromagnetic method, which transmits a step current pulse signal to the rock stratum, collects the secondary field decay signal caused by the underground abnormal body after power-off, and detects the distribution of the underground resistivity abnormal body according to the different electromagnetic field decay signals of different rock resistivities. The main advantage of the method is that the transmitting device and the receiving device can be ungrounded, which makes it possible for airborne and semi-airborne transient electromagnetic method, and also lays a foundation for real-time tunnel advance prediction.

[0003] The transient electromagnetic advance detection technology can predict the resistivity information of the rock in front, and is very sensitive to low-resistance abnormal bodies, so it can judge whether there is a water-rich abnormal body in front of the tunnel, so as to better guide the shield excavation scheme and avoid geological disasters such as stratum subsidence and water surge above the tunnel to the greatest extent. In view of the current domestic situation in the relevant aspects, the tunnel construction safety prevention consciousness is generally weak, and it is very important to carry out tunnel advance prediction, develop a reasonable construction scheme according to the prediction results and take corresponding measures to reduce the probability of geological disasters and eliminate safety hazards. Improving the tunnel transient electromagnetic advance prediction technology and using a precise and effective tunnel advance prediction system for geological advance prediction have broad application prospects in tunnel construction, urban underground space construction and mine energy exploitation.

[0004] The tunnel transient electromagnetic advance prediction technology is mainly based on one-dimensional inversion interpretation, and the central loop device is widely used. It is usually assumed that the area in front of the detection is a layered medium, and the conversion formula of the apparent resistivity of the whole area is derived based on the analytical solution of the half-space or full-space layered model. After the measured induced electromotive force is obtained, the apparent resistivity can be converted, or the induced electromotive force can be directly inverted to obtain the inverted resistivity model. Through the apparent resistivity and the inversion result, the medium structure in front of the tunnel and whether there is a water-rich abnormal body can be judged.

[0005] The tunnel transient electromagnetic advanced prediction technology mainly adopts one-dimensional inversion interpretation, and the inversion has great uncertainty. Unlike the ground transient electromagnetic method, the tunnel internal transient electromagnetic field has stronger non-uniqueness, the data collected by the sensor can be affected by the tunnel front and the surrounding rock at the same time, therefore, the inversion result is easily affected by the surrounding rock, and false anomalies are easily generated in the inversion. In addition, the two-dimensional and three-dimensional tunnel full-space transient electromagnetic inversion has great non-uniqueness and low calculation efficiency, and the time length of one inversion can be as long as ten hours or even several days, which can seriously affect the tunnel excavation construction progress. Therefore, how to reduce the non-uniqueness of the transient electromagnetic inversion and improve the efficiency of the two-dimensional and three-dimensional inversion is a problem to be solved.

[0006] Through the above analysis, the problems and defects of the prior art are:

[0007] (1) The existing tunnel transient electromagnetic advanced prediction technology mainly adopts one-dimensional inversion interpretation, and the inversion result is easily affected by the surrounding rock and false anomalies are easily generated in the inversion, so the one-dimensional inversion has great uncertainty.

[0008] (2) The two-dimensional and three-dimensional tunnel full-space transient electromagnetic inversion has great non-uniqueness and low calculation efficiency, and the time length of one inversion can be as long as ten hours or even several days, which can seriously affect the tunnel excavation construction progress. SUMMARY

[0009] In view of the problems existing in the prior art, the present application provides a transient electromagnetic rapid advanced prediction method, in particular to a transient electromagnetic rapid advanced prediction method based on ResNet.

[0010] The present application is realized in this way, a transient electromagnetic rapid advanced prediction method, the transient electromagnetic rapid advanced prediction method comprises: setting an observation system based on actual exploration needs; a large number of theoretical tunnel resistivity models are randomly generated; based on the observation system and the numerical simulation algorithm, the tunnel resistivity model is numerically calculated to obtain the transient electromagnetic field data corresponding to each model; a residual neural network model is constructed, a loss function is designed, and the obtained transient electromagnetic field data corresponding to each model is used to train the ResNet; the trained ResNet model is used to predict new observation data to obtain the tunnel TEM data inversion result.

[0011] Further, the transient electromagnetic rapid advanced prediction method comprises the following steps:

[0012] Step 1, setting the observation system and numerically simulating the three-dimensional transient electromagnetic field in the tunnel space to obtain the training data set;

[0013] Step 2, constructing a residual neural network model and performing performance evaluation; the hyperparameters of the residual network are constantly updated through back propagation to complete the training of the residual neural network model.

[0014] Further, in step one, in tunnel advanced detection, a central loop device is used to scan around the tunnel and directly in front, and an observation system is arranged. Set up n data acquisition channels (set according to the needs of the instrument and the detection coverage range), the Z-direction induced electromotive force is collected at the collection points above and below the tunnel, the collection points in front are detected in the normal direction of the receiving coil, and the detection direction is rotated within a range of 180°, a range of [-90°, 90°], and an interval of 10°; 30 time sampling points are set in each channel, and the interval is 10 -5 to 10 -3 s within 10

[0015] Further, in step one, the three-dimensional transient electromagnetic field in the tunnel space is numerically simulated, 5% Gaussian noise is added to each simulation data to obtain a training data set.

[0016] Assume that the tunnel environment is an isotropic non-uniform medium in the whole space, there is a tunnel space with a resistivity of 1e8, and a low-resistivity anomaly body is located at any position around the tunnel. Set the background resistivity, the size of the low-resistivity anomaly body, the spatial position, and the low-resistivity anomaly resistivity to vary uniformly within a certain range. The background resistivity varies in the range of [1e2, 1e5] Ω·m, and the low-resistivity anomaly body resistivity varies in the range of [0.1, 10] Ω·m. A square shape is used instead of any possible anomaly shape, and the side length ranges from 10 to 30 m.

[0017] The transient electromagnetic field is essentially the decay signal of the electromagnetic field. The induced electromotive force is collected and log10 is taken, and then the absolute value is taken to control the order of magnitude of the input data to be within 20 positive numbers. The input data is a two-dimensional structure, the horizontal coordinate is the channel number of the distance or scanning angle, and the vertical coordinate is the channel number of the time. The upper, lower, and front three data sets are combined into one picture as the input of ResNet, so that the neural network can learn the spatial position of the three two-dimensional pictures.

[0018] The center position coordinates of the low-resistivity anomaly body in each model are projected to a one-dimensional space with a space range of x∈[0, 90] m and an interval of 0.2 m, so the number of neurons in the output layer is 450. The label value near the coordinates is set to a Gaussian distribution, the model center point is located at the maximum point of the Gaussian function, and the variance of the Gaussian distribution is set to 5 m. The theoretical model and the corresponding simulation data set are generated, of which 90% is used for training and 10% is used for testing.

[0019] Further, in step two, the deep learning network for tunnel TEM data training has 4 layers on the left and right, a total of 10 convolutional layers and 2 pooling layers. The RELU activation function is selected to act on the output of the convolutional layer, and Batch_Normilization is used to standardize the data; ShortCut is used to directly pass the information of the shallow network to the deep layer; the sigmoid function is used on the last convolutional layer to make the output value range in [0, 1].

[0020] The performance of ResNet is evaluated by the cross entropy between the predicted variable q i and the label p i , and the expression is:

[0021]

[0022] Where i=0,1 respectively represents the class of the existence of the anomaly and the classification of the non-existence of the low resistance anomaly, and the sum of the class of the existence of the anomaly and the classification of the non-existence of the low resistance anomaly is always 1; x [0, 90] represents the horizontal coordinate of the output distribution. The hyperparameters of the residual network are continuously updated through back propagation, and the training of the residual neural network model is completed.

[0023] Further, in step two, the prediction area is divided into ten regions, which are divided into 1-9 in order, and each region represents a class, and 0 class has no anomaly; the classification result of the prediction is obtained according to the coordinates corresponding to the maximum value of the prediction label, and the accuracy of the test set and the training set is calculated according to the following formula and the label:

[0024]

[0025] Where K is the total number of partitions 9, and I is the distance between the predicted classification and the label classification.

[0026] Another object of the application is to provide a transient electromagnetic quick advance prediction system applying the transient electromagnetic quick advance prediction method.

[0027] The tunnel resistivity model generation module is used to set an observation system based on actual exploration needs, and randomly generate a large number of theoretical tunnel resistivity models;

[0028] The model numerical calculation module is used to perform numerical calculation on the tunnel resistivity model based on the observation system and the numerical simulation algorithm, to obtain the transient electromagnetic field data corresponding to each model;

[0029] The model training module is used to construct a residual neural network model, design a loss function, and train the ResNet using the transient electromagnetic field data corresponding to each model;

[0030] An observation data prediction module is configured to utilize the trained ResNet model to predict new observation data, and obtain an inversion result of the tunnel TEM data.

[0031] Another object of the present application is to provide a computer device comprising a memory and a processor, the memory storing a computer program, the computer program being executed by the processor to cause the processor to perform the steps of the transient electromagnetic quick advance prediction method.

[0032] Another object of the present application is to provide a computer readable storage medium storing a computer program, the computer program being executed by the processor to cause the processor to perform the steps of the transient electromagnetic quick advance prediction method.

[0033] Another object of the present application is to provide an information data processing terminal for implementing the transient electromagnetic quick advance prediction system.

[0034] In combination with the above technical solutions and the technical problems solved, the technical solutions of the present application have the following advantages and positive effects:

[0035] First, in view of the technical problems existing in the prior art and the difficulty in solving the problems, the technical solutions of the present application are closely combined with the results and data obtained during the development process, and the technical problems solved by the technical solutions are analyzed in detail and deeply, and some creative technical effects brought about after the problems are solved are described as follows:

[0036] The present application provides a transient electromagnetic two-dimensional quick inversion technology based on residual neural network (ResNet), which is based on three-dimensional data and reduces non-uniqueness compared with conventional one-dimensional inversion; the inversion algorithm obtains a prediction result within 1 second, realizes real-time monitoring of whether there is a dangerous anomaly in front of the tunnel and the approximate direction of the existing anomaly, and greatly improves the data processing efficiency compared with conventional two-dimensional and three-dimensional inversion.

[0037] The present application is based on residual neural network and realizes quick inversion of tunnel TEM data, i.e., semi-quantitative evaluation of whether there is a low-resistance anomaly body and determination of its approximate position. The core protection of the present application is as follows: 1) semi-quantitative ResNet prediction idea: based on residual neural network, using TEM data to semi-quantitatively predict whether there is a low-resistance anomaly in front of the tunnel and the direction of the anomaly; 2) generation method of resistivity model training set, processing of input data and design scheme of label; 3) design of loss function.

[0038] The transient electromagnetic two-dimensional fast inversion technology provided by the application has the following advantages: on the one hand, the inversion model is obtained based on a three-dimensional data set, more three-dimensional information is used compared with one-dimensional inversion, the constraint on the inversion model is strengthened, and the non-uniqueness of one-dimensional inversion is reduced; on the other hand, after the neural network model is trained, the prediction result can be obtained within 1 second given the measurement data, the calculation efficiency is greatly improved, and real-time monitoring of whether a dangerous anomaly exists in front of the tunnel and the approximate direction of the existing anomaly is realized.

[0039] Secondly, from the perspective of the product or as a whole, the technical effects and advantages of the technical solution to be protected by the application are described as follows:

[0040] Compared with the existing transient electromagnetic tunnel advance prediction technology, the main advantages of the application are as follows:

[0041] 1) More accurate, the possibility of false anomalies in inversion is reduced;

[0042] 2) More efficient, in the traditional method, one-dimensional inversion usually takes about 1 hour, and two-dimensional or three-dimensional inversion usually takes several hours to several tens of hours, and the method of the application can complete inversion within 1 second, which lays a foundation for real-time tunnel advance prediction.

[0043] Thirdly, the inventiveness of the claims of the application is also reflected in the following important aspects:

[0044] (1) The expected income and commercial value of the technical solution of the application after transformation are as follows:

[0045] Due to the high calculation efficiency of the application, fast real-time inversion can be realized, and the tunnel construction efficiency is improved.

[0046] (2) The technical solution of the application fills the technical gap in the industry at home and abroad:

[0047] Deep learning is applied to TEM advance prediction for the first time, realizing qualitative inversion and real-time inversion of two-dimensional or three-dimensional TEM advance prediction in tunnels.

[0048] (3) Does the technical solution of the application solve the technical problems that people have been eager to solve but have failed to solve successfully:

[0049] To some extent, the problems of large one-dimensional uncertainty of tunnel transient electromagnetic and low two-dimensional or three-dimensional inversion efficiency are solved. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments of the present application will be briefly introduced as follows. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0051] Figure 1 is a flow chart of a transient electromagnetic quick advance prediction method provided by the embodiments of the present application;

[0052] Figure 2 is a structure diagram of an observation system and input data provided by the embodiments of the present application;

[0053] Figure 3 is a resistivity model setting method and a part of a model diagram provided by the embodiments of the present application;

[0054] Figure 4 is a ResNet framework diagram for advance prediction provided by the embodiments of the present application, the numbers above the feature maps represent the number of layers of the feature maps, the length and the width; the number of layers corresponds to the number of filters; the pooling layer represents the maximum pooling operation of down-sampling the feature maps;

[0055] Figure 5A is an evolution diagram of a loss function in a training process provided by the embodiments of the present application;

[0056] Figure 5B is an evolution diagram of accuracy in a training process provided by the embodiments of the present application;

[0057] Figure 6A is a schematic diagram of a synthetic model anomaly above a tunnel provided by the embodiments of the present application;

[0058] Figure 6B is a probability distribution of a ResNet prediction result provided by the embodiments of the present application Figure 1 ;

[0059] Figure 6C is a converted prediction anomaly body space distribution provided by the embodiments of the present application Figure 1 , the darker the color, the greater the possibility of the existence of low resistance body, and the dashed line represents the specific position of the anomaly;

[0060] Figure 6D is a schematic diagram of a synthetic model anomaly in front of a tunnel provided by the embodiments of the present application;

[0061] Figure 6E is a probability distribution of a ResNet prediction result provided by the embodiments of the present application Figure 2 ;

[0062] Figure 6FThe converted predicted abnormal body spatial distribution provided by the embodiment of the present application Figure 2 The darker the color, the greater the possibility of the existence of low-resistance bodies, and the dashed line represents the specific location of the abnormality.

[0063] Figure 7A The schematic diagram of the case where the synthetic model abnormality is under the tunnel provided by the embodiment of the present application

[0064] Figure 7B The probability distribution of the ResNet prediction result provided by the embodiment of the present application Figure 1 ;

[0065] Figure 7C The converted predicted abnormal body spatial distribution provided by the embodiment of the present application Figure 1 The darker the color, the greater the possibility of the existence of low-resistance bodies, and the dashed line represents the specific location of the abnormality.

[0066] Figure 7D The schematic diagram of the case where the synthetic model abnormality does not exist provided by the embodiment of the present application

[0067] Figure 7E The probability distribution of the ResNet prediction result provided by the embodiment of the present application Figure 2 ;

[0068] Figure 7F The converted predicted abnormal body spatial distribution provided by the embodiment of the present application Figure 2 The darker the color, the greater the possibility of the existence of low-resistance bodies, and the dashed line represents the specific location of the abnormality. DETAILED DESCRIPTION

[0069] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application.

[0070] In view of the problems in the prior art, the present application provides a transient electromagnetic rapid advance prediction method, which will be described in detail below with reference to the accompanying drawings.

[0071] In order to enable those skilled in the art to fully understand how the present application is specifically implemented, this part is an explanatory embodiment for explaining and describing the technical solution of the claims.

[0072] As shown in Figure 1 The transient electromagnetic rapid advance prediction method provided by the embodiment of the present application includes the following steps:

[0073] S101, setting an observation system based on actual exploration needs;

[0074] S102, a large number of theoretical tunnel resistivity models are randomly generated;

[0075] S103, based on the observation system and the numerical simulation algorithm, the tunnel resistivity model is numerically calculated to obtain transient electromagnetic field data corresponding to each model;

[0076] S104, a residual neural network (ResNet) model is constructed, a loss function is designed, and the obtained transient electromagnetic field data corresponding to each model is used to train the ResNet;

[0077] S105, the trained model is used to predict new observation data to obtain an inversion result.

[0078] As a preferred embodiment, the transient electromagnetic rapid advance prediction method provided by the embodiment of the application specifically includes the following steps:

[0079] (1) Observation system and data preparation

[0080] In tunnel advance detection, a central loop device is used to scan around the tunnel and directly in front of the tunnel, and the observation system is arranged as shown in Figure 2 . It is assumed that 38 data acquisition channels are set, 9 above the tunnel, 9 below the tunnel, and 19 in front of the tunnel; the acquisition points above and below the tunnel only acquire Z-direction induced electromotive force, and the acquisition points in front of the tunnel are in the detection direction of the receiving coil normal direction, and are rotated within a range of 180°, a range of [-90°, 90°], and an interval of 10°, and a total of 19 channels can be obtained. There are 30 time sampling points in each channel, and the time sampling points are logarithmically spaced within 10 -5 s to 10 -3 s.

[0081] The application uses the finite volume method to numerically simulate the three-dimensional transient electromagnetic field in the tunnel space, and adds 5% Gaussian noise to each simulation data to obtain a large number of training data sets. It is assumed that the tunnel environment is an isotropic non-uniform medium in the whole space, there is a tunnel space with a resistivity of 1e8 (the resistivity of air), and a low-resistance anomaly body is located at an arbitrary position around the tunnel. In order to improve the generalization ability of the data, the background resistivity, the size of the low-resistance anomaly body, the spatial position, and the low-resistance anomaly resistivity are randomly changed within a certain range (see Figure 3 ). The background resistivity changes in the range of [1e2, 1e5] Ω·m, and the low-resistance anomaly body resistivity changes in the range of [0.1, 10] Ω·m. Since the transient electromagnetic field is not sensitive to the shape of the anomaly body, a square shape is used instead of any possible anomaly shape, and the side length ranges from 10 to 30 m.

[0082] The transient electromagnetic field is essentially the decay signal of the electromagnetic field. Generally, the induced electromotive force is collected, and the order of magnitude of the obtained induced electromotive force is generally low and has a large difference, generally in the range of 10 -5 to 10 -18 , which is not conducive to the transmission of neurons. Therefore, the obtained value is taken as log10, and then the absolute value is taken (see Figure 2 ). The order of magnitude of the input data can be controlled within 20 positive numbers, which is a reasonable neural network input value. The input data is a two-dimensional structure, and the horizontal coordinate is the distance or the number of scanning angles, and the vertical coordinate is the number of time channels. In order to obtain the spatial structure of the three positions, the present application combines three data sets (upper, lower, and front) according to the spatial position into a graph as the input of ResNet.

[0083] In order to semi-quantitatively evaluate whether there is a low-resistivity anomaly body near the tunnel and its spatial position, it is particularly important to reasonably set the data label. The present application projects the center position coordinates of the low-resistivity anomaly body in each model to one-dimensional space, the spatial range x [0, 90] m, and the interval is 0.2 m, so the number of neurons in the output layer is 450. In actual situations, there is a certain uncertainty in the center point of the model, so the present application sets the label value near the coordinates to be a Gaussian distribution, the model center point is located at the maximum value point of the Gaussian function, and the variance of the Gaussian distribution is set to 5 m. This setting not only reduces the influence of the data set error and speeds up the convergence of the network, but also makes the anomaly bodies with relatively close distances partially overlap in the label, showing a certain correlation, so that the prediction result of the network is more stable.

[0084] Based on the above settings, a total of 4000 theoretical models and corresponding simulation data sets are generated, of which 90% are used for training and 10% are used for testing.

[0085] (2) Residual neural network construction

[0086] The deep learning framework used for tunnel TEM data training is as shown in Figure 4As shown. Network left and right each 4 layers, a total of 10 convolutional layers, 2 pooling layers. Convolutional layer has the effect of local perception, parameter sharing, compared with fully connected layer can greatly reduce the number of weights. Pooling layer has the effect of reducing information redundancy, improving the scale invariance of the model, rotation invariance and preventing overfitting. The RE LU activation function is selected to act on the output of the convolutional layer, which improves the nonlinearity of the neural network and improves the efficiency of backpropagation gradient calculation. In addition, ReLu will make a part of the output of the neuron to 0, which will cause the sparsity of the network and reduce the interdependence of the parameters, and alleviate the occurrence of overfitting. After convolution operation, Batch_Normilization is used to standardize the data to further prevent gradient disappearance or gradient explosion phenomenon, and can increase the regularization effect. ShortCut is used to directly pass the information of the shallow network to the deep layer to prevent network degradation phenomenon with the increase of network layers. In order to obtain the predicted probability distribution, the sigmoid function is used on the last convolutional layer to make the output value range in [0, 1].

[0087] The application evaluates the performance of ResNet by predicting the cross entropy between the variable q i and the label p i , and the specific expression is:

[0088]

[0089] Where i=0,1 respectively represent the class of abnormality and the class of non-low resistance abnormality, and the sum of the two is always 1, x∈[0, 90] represents the horizontal coordinate of the output distribution. The hyperparameters of the residual network can be continuously updated by backpropagation, and the training can be completed.

[0090] Figure 5 shows the evolution of the loss function and accuracy during the training process. It can be seen that the errors of the training set and the test set during the training process gradually decrease with the increase of the number of iterations, and finally both are lower than 0.05, which shows that the model can well learn the features in the training set. In order to better reveal the degree of agreement between the prediction result and the label, the prediction area is divided into ten regions (1-9 in order, each region represents a class, and 0 class without anomaly, see Figure 6C ), according to the coordinates corresponding to the maximum value of the predicted label, the predicted classification result can be obtained, and then the accuracy is calculated with the label as follows:

[0091]

[0092] Where K is the total number of partitions 9, and I is the distance between the predicted classification and the label classification.

[0093] Figure 5BThe accuracy evolution results of the test set are shown in the middle, and with the increase of the number of iterations, the accuracy of the test set and the training set gradually approaches 99%, indicating that ResNet has learned the electromagnetic field characteristics of TEM well and can be used to effectively predict the low-resistivity anomaly body in front of the tunnel.

[0094] In order to illustrate the effectiveness and applicability of the proposed inversion scheme, the present application shows a plurality of synthetic data inversion results from the test data set (see Figures 6, 7). For different resistivity models, ResNet can accurately restore the orientation of the anomaly body; although the predicted waveform is relatively shorter than the label, the maximum value corresponds well, which can better reflect the position of the anomaly body. When the anomaly body does not exist, the predicted probability distribution value is very low, which corresponds well with the label, indicating that ResNet can effectively distinguish whether a low-resistivity anomaly body exists in front of the tunnel and within a range of 100m around it, and also illustrates the effectiveness of the inversion method of the present application.

[0095] The transient electromagnetic rapid advance prediction system provided by the embodiment of the present application comprises:

[0096] The tunnel resistivity model generation module is used to set an observation system based on actual exploration needs, and randomly generate a large number of theoretical tunnel resistivity models;

[0097] The model numerical calculation module is used to perform numerical calculation on the tunnel resistivity model based on the observation system and a numerical simulation algorithm, to obtain transient electromagnetic field data corresponding to each model;

[0098] The model training module is used to construct a residual neural network model, design a loss function, and train ResNet using the transient electromagnetic field data corresponding to each model obtained;

[0099] The observation data prediction module is used to predict new observation data using the trained ResNet model to obtain the inversion result of the tunnel TEM data.

[0100] The embodiment of the present application has achieved some positive effects during research and development or use, and indeed has great advantages compared with the prior art, which will be described below in combination with the data, charts and the like during the test process.

[0101] In order to illustrate the effectiveness and applicability of the proposed inversion scheme, the present application shows a plurality of synthetic data inversion results from a test data set (see Figures 6, 7). For different resistivity models, ResNet can accurately restore the orientation of the anomaly body; although the predicted waveform is relatively shorter than the label, the maximum value corresponds well, which can better reflect the position of the anomaly body. When the anomaly body does not exist, the predicted probability distribution value is very low, which corresponds well with the label, indicating that ResNet can effectively distinguish whether a low-resistivity anomaly body exists in front of the tunnel and within a range of 100m around it, and also illustrates the effectiveness of the inversion method of the present application.

[0102] It should be noted that the embodiments of the present application can be realized by hardware, software or a combination of software and hardware. The hardware part can be realized by special logic; the software part can be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or a specially designed hardware. Those skilled in the art can understand that the above-mentioned devices and methods can be realized by computer executable instructions and / or included in processor control code, such as provided on a carrier medium, such as a magnetic disk, CD or DVD-ROM, a programmable memory, such as a read-only memory (firmware), or a data carrier, such as an optical or electronic signal carrier. The devices of the present application and their modules can be realized by hardware circuits, such as very large scale integrated circuits or gate arrays, semiconductors, such as logic chips, transistors, etc., or programmable hardware devices, such as field programmable gate arrays, programmable logic devices, etc., by software executed by various types of processors, or by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0103] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any modification, equivalent replacement and improvement made by those skilled in the art within the technical scope disclosed by the present application, within the spirit and principles of the present application, should be covered within the protection scope of the present application.

Claims

1. A transient electromagnetic quick-look prediction method, characterized in that, The transient electromagnetic rapid advance prediction method comprises the following steps: The transient electromagnetic rapid advance prediction method comprises the following steps: Step one, set the observation system and use the finite volume method to numerically simulate the three-dimensional transient electromagnetic field in the tunnel space to obtain the training data set; In step one, the finite volume method is used to numerically simulate the three-dimensional transient electromagnetic field in the tunnel space, and 5% Gaussian noise is added to each simulation data to obtain the training data set; The transient electromagnetic field is essentially an attenuation signal of the electromagnetic field, the induced electromotive force is collected and log10 is taken, and the absolute value is taken, so that the order of magnitude of the input data is controlled within 20 positive numbers; the input data is a two-dimensional structure, the horizontal coordinate is the number of channels of distance or scanning angle, and the vertical coordinate is the number of channels of time; the upper, lower and front three data sets are combined into a graph according to the spatial position as the ResNet input, and the spatial structure of the three positions is obtained; When the tunnel environment is a full-space isotropic inhomogeneous medium, there is a tunnel space with a resistivity of 1e8, and a low-resistivity anomaly body is located at any position around the tunnel; the background resistivity, the size of the low-resistivity anomaly body, the spatial position, and the low-resistivity anomaly resistivity are randomly changed within a certain range; the background resistivity changes in the range of [1e2, 1e5] , the low-resistivity anomaly resistivity changes in the range of [0.1, 10] ; a square shape is used instead of any possible anomaly shape, and the side length ranges from 10m to 30m; In step two, the prediction area is divided into ten regions, which are sequentially divided into 1~9, and each region represents a class, and 0 class without anomaly; according to the coordinates corresponding to the maximum value of the prediction label, the classification result of the prediction is obtained, and the accuracy of the test set and the training set is calculated according to the following formula and the label: The center position coordinates of the low-resistance abnormal body in each model are projected to a one-dimensional space, and the space range is m, interval 0.2m, so the number of neurons in the output layer is 450; the label value near the coordinates is set as a Gaussian distribution, the center point of the model is located at the maximum value point of the Gaussian function, and the variance of the Gaussian distribution is set as 5m; a theoretical model and a corresponding simulation data set are generated, wherein 90% is used for training and 10% is used for testing.

2. The transient electromagnetic quick-look prediction method of claim 1, wherein, In step one, in the tunnel advanced detection, the central loop device is used to scan the tunnel surroundings and the front, and the observation system is arranged; n data acquisition channels are set, the Z direction induced electromotive force is only collected at the upper and lower collection points of the tunnel, the collection points in the front are detected in the normal direction of the receiving coil, and the rotation is performed within the range of 180° , interval ; m time sampling points are set in each channel, and the number interval is collected within to s.

3. The transient electromagnetic quick-look prediction method of claim 1, wherein, In step two, the deep learning network for tunnel TEM data training has 4 layers on the left and 4 layers on the right, a total of 10 convolutional layers and 2 pooling layers; the RELU activation function is selected to act on the output of the convolutional layer, and Batch_Normilization is used to standardize the data; ShortCut is used to directly pass the information of the shallow network to the deep layer; the sigmoid function is used on the last convolutional layer to make the output value range in ; The performance of ResNet is evaluated by cross-entropy between predicted variables and labels , expressed as: ； wherein, respectively represent the class with abnormality and the class without low-resistivity abnormality, and the sum of the class with abnormality and the class without low-resistivity abnormality is always 1; represents the abscissa of the output distribution; the hyperparameters of the residual network are constantly updated through back propagation, and the training of the residual neural network model is completed.

4. The transient electromagnetic quick-look prediction method of claim 1, wherein, Wherein, K is the total number of partitions 9, and I is the distance between the predicted classification and the label classification. ； The transient electromagnetic rapid advance prediction system comprises:

5. A transient electromagnetic quick-look prediction system using the transient electromagnetic quick-look prediction method according to any one of claims 1 to 4, characterized by The tunnel resistivity model generation module is used for setting the observation system based on the actual exploration needs, and randomly generating a large number of theoretical tunnel resistivity models; The model numerical calculation module is used for numerically calculating the tunnel resistivity model based on the observation system and the numerical simulation algorithm to obtain the transient electromagnetic field data corresponding to each model; The model training module is used for constructing a residual neural network model, designing a loss function, and training ResNet using the transient electromagnetic field data corresponding to each model obtained; The observation data prediction module is used for predicting new observation data using the trained ResNet model to obtain the inversion result of the tunnel TEM data. The computer device comprises a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to make the processor execute the steps of the transient electromagnetic rapid advance prediction method according to any one of claims 1-4.

6. A computer device, comprising:

7. A computer readable storage medium storing a computer program, the computer program being executed by a processor to make the processor execute the steps of the transient electromagnetic rapid advance prediction method according to any one of claims 1-4. ​ 8. An information data processing terminal, characterized by The information data processing terminal is used to realize the transient electromagnetic quick advance prediction system as claimed in claim 5.

Images