A deep learning inversion method and system for magnetotelluric

Abstract

The present application relates to the technical field of geophysical exploration, and discloses a deep learning inversion method and system for magnetotelluric sounding, comprising: constructing a trained magnetotelluric one-dimensional inversion model; converting measured apparent resistivity and measured phase data into interpolated apparent resistivity and interpolated phase data, using a Savitzky-Golay filter to construct a target multi-window Savitzky-Golay filter set to obtain smoothed apparent resistivity and smoothed phase data, and constructing an inversion apparent resistivity-phase data set based on the smoothed apparent resistivity and smoothed phase data; inputting the inversion apparent resistivity-phase data set into the trained magnetotelluric one-dimensional inversion model to obtain a corresponding inversion resistivity geoelectric model set, and obtaining a target resistivity geoelectric model based on the inversion resistivity geoelectric model set; the present application solves the problems of low accuracy and poor practicability in the existing deep learning inversion method for magnetotelluric sounding.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geophysical exploration, and in particular to a deep learning inversion method and system for magnetotelluric. BACKGROUND

[0002] Magnetotelluric method is a geophysical exploration method for studying the electrical distribution of the earth interior by using natural alternating electromagnetic field, and is widely used in the fields of deep electrical structure detection, oil and gas exploration, and metal ore exploration. Magnetotelluric inversion is an important means of magnetotelluric data processing and interpretation, which infers the electrical parameters of underground medium according to the surface observation data. The conventional magnetotelluric inversion method is based on linearization theory, and the electrical parameters of underground medium are obtained by forward iteration optimization, which has problems of dependence on initial model, easy to fall into local optimum and low computational efficiency.

[0003] In recent years, the rapid development of deep learning provides a new solution for geophysical inversion. Deep learning is a data-driven algorithm, and its prediction performance is highly dependent on a large number of representative training sets. In the deep learning inversion of magnetotelluric, due to the difficulty in obtaining high-quality magnetotelluric measured data and the corresponding real geoelectric model of underground, the training set is usually obtained by synthetic data set through forward. However, magnetotelluric exploration has high regional and project-oriented characteristics, and different training sets need to be established accordingly when facing different exploration research areas or projects. At the same time, in the actual exploration scene, the magnetotelluric measured data is usually disturbed by complex noise. Therefore, it is difficult to quickly establish a large number of training sets covering all noise types to meet the actual inversion requirements due to the difficulty of implementation, computational resources and time cost. As a result, although the deep learning method can achieve excellent inversion effect on synthetic magnetotelluric data, it will encounter the dilemma of serious degradation of generalization performance on measured data, which limits its practicality in magnetotelluric inversion. Therefore, the existing magnetotelluric inversion determination method has the problems of low accuracy and poor practicability. SUMMARY

[0004] The present application provides a deep learning inversion method and system for magnetotelluric to solve the problem of low accuracy and poor practicability in the existing deep learning inversion method for magnetotelluric.

[0005] In order to achieve the above purpose, the present application realizes the technical scheme as follows:

[0006] In the first aspect, the present application provides a deep learning inversion method for magnetotelluric, comprising:

[0007] Obtaining geophysical information and petrophysical information of a target exploration area, and constructing a synthetic resistivity geoelectric model set based on the geophysical information and the petrophysical information;

[0008] Calculating the synthetic resistivity geoelectric model by magnetotelluric one-dimensional forward, obtaining corresponding synthetic apparent resistivity and synthetic phase data, and establishing a magnetotelluric synthetic data sample set based on the synthetic apparent resistivity and the synthetic phase data;

[0009] Constructing a magnetotelluric one-dimensional deep learning inversion model based on the synthetic apparent resistivity, the synthetic phase data and the synthetic resistivity geoelectric model, and training the magnetotelluric one-dimensional inversion model by the magnetotelluric synthetic data sample set to obtain a trained magnetotelluric one-dimensional inversion model;

[0010] Obtaining measured apparent resistivity and measured phase data of the target exploration area, converting the measured apparent resistivity and the measured phase data into interpolated apparent resistivity and interpolated phase data by linear interpolation, and constructing a multi-window Savitzky-Golay filter set by using a Savitzky-Golay filter;

[0011] Serially smoothing the interpolated apparent resistivity and the interpolated phase data by using the multi-window Savitzky-Golay filter set to obtain smoothed apparent resistivity and smoothed phase data, and constructing a to-be-inverted apparent resistivity-phase data set based on the smoothed apparent resistivity and the smoothed phase data;

[0012] Inputting the to-be-inverted apparent resistivity-phase data set into the trained magnetotelluric one-dimensional inversion model to obtain a corresponding inversion resistivity geoelectric model set, obtaining a target resistivity geoelectric model based on the inversion resistivity geoelectric model set, and taking the target resistivity geoelectric model as electromagnetic data of the target exploration area.

[0013] Optionally, the obtaining of the measured apparent resistivity and the measured phase data of the target exploration area comprises:

[0014] Realizing exploration of the target exploration area by a magnetotelluric instrument, collecting apparent resistivity and phase data of the target exploration area, and taking the collected apparent resistivity and phase data as the measured apparent resistivity and the measured phase data.

[0015] Optionally, the establishing of the magnetotelluric synthetic data sample set by the synthetic apparent resistivity, the synthetic phase data and the synthetic resistivity geoelectric model set comprises:

[0016] Corresponding the synthetic apparent resistivity, the synthetic phase data and data in the synthetic resistivity geoelectric model set one by one to form a magnetotelluric synthetic data sample set for deep learning model training.

[0017] Optionally, the constructing the magnetotelluric one-dimensional inversion model based on the synthetic apparent resistivity, the synthetic phase data and the synthetic resistivity geoelectric model comprises:

[0018] The synthetic apparent resistivity and the synthetic phase data are taken as double-channel inputs of a model network, and the synthetic resistivity geoelectric model is taken as an expected output of the model network, so as to construct a magnetotelluric one-dimensional inversion model, wherein a dimension of the model network input is , and a dimension of an output of the model network is .

[0019] Optionally, the training the magnetotelluric one-dimensional inversion model through a magnetotelluric synthetic data sample set to obtain a trained magnetotelluric one-dimensional inversion model comprises:

[0020] The magnetotelluric synthetic data sample set is divided into a training set and a verification set, and the apparent resistivity and the phase data in the training set and the verification set are converted into training set apparent resistivity and training set phase data with a length of X through a linear interpolation method;

[0021] The training set apparent resistivity, the training set phase data and the training set resistivity geoelectric model are input into the magnetotelluric one-dimensional inversion model for training, so as to obtain the trained magnetotelluric one-dimensional inversion model.

[0022] Optionally, the constructing the target multi-window Savitzky-Golay filter set by using a Savitzky-Golay filter comprises:

[0023] As an optional implementation manner, a polynomial order of the Savitzky-Golay filter is set to , and a recommended polynomial order is 3, a window set containing windows is set to , wherein is a window length. As an optional implementation manner, the maximum window length is set according to a noise level, and a recommended maximum window length is or ;

[0024] An integer between 1 and is randomly generated, and then windows are randomly selected from the window set , and the selected windows are sorted in ascending order of the window length to form the target multi-window Savitzky-Golay filter set.

[0025] Optionally, after the smoothed apparent resistivity and the smoothed phase data are obtained, before the apparent resistivity-phase data set to be inverted is constructed based on the smoothed apparent resistivity and the smoothed phase data, the method further comprises:

[0026] The operation of constructing the target multi-window Savitzky-Golay filter set by reusing the Savitzky-Golay filter and the operation of obtaining the smoothed apparent resistivity and the smoothed phase data by serially smoothing and interpolating the apparent resistivity and the phase data using the target multi-window Savitzky-Golay filter set are repeated N times, where N is a positive integer.

[0027] Optionally, obtaining the target resistivity geoelectric model based on the inverted resistivity geoelectric model set comprises:

[0028] Summing all the resistivity geoelectric model data in the inverted resistivity geoelectric model set and averaging the sum result;

[0029] Taking the average value calculated as the target resistivity geoelectric model.

[0030] In the second aspect, the embodiments of the present application provide a magnetotelluric deep learning inversion system, comprising a processor, a memory;

[0031] The memory is used to store a computer program.

[0032] The processor is used to execute the program stored on the memory, and realize the method steps of any one of the first aspect.

[0033] Advantages:

[0034] The deep learning inversion method of magnetotellurics provided by the present application, before the measured magnetotelluric data is input into the trained deep learning model for inversion and prediction, the measured apparent resistivity and phase data of magnetotellurics are first smoothed by using the multi-window Savitzky-Golay filter proposed in the present application, and then the smoothed apparent resistivity and phase data are input into the trained deep learning model for inversion and prediction, so as to improve the inversion effect of the deep learning method on the measured magnetotelluric data, and promote the practicality of the deep learning inversion of magnetotellurics.

[0035] It is further worth mentioning that the method makes the measured apparent resistivity and phase data to be inverted close to the synthetic apparent resistivity and synthetic phase data in the synthetic data sample set established for network training, instead of striving to establish a large number of detailed data sample sets covering the target solution of the current exploration research area, which can reduce the dependence of the deep learning inversion method on the large number of training sets, thereby reducing the construction time of the data sample set for training, the parameter adjustment and training time of the deep learning model, improving the deep learning inversion accuracy of the magnetotelluric method, and promoting the intelligent level of the magnetotelluric inversion. BRIEF DESCRIPTION OF DRAWINGS

[0036] Fig. 1 Flowchart of the deep learning inversion method of the magnetotelluric method of the preferred embodiment of the present application;

[0037] Fig. 2 Flowchart of the smoothing operation of the multi-window Savitzky-Golay filter of the preferred embodiment of the present application;

[0038] Fig. 3 Deep learning inversion result provided by the preferred embodiment of the present application. DETAILED DESCRIPTION

[0039] The technical solutions of the present application will be described in detail below. Obviously, the described embodiments are only part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0040] Unless otherwise defined, the technical terms or scientific terms used in the present application shall have the usual meanings understood by those skilled in the art to which the present application belongs. The terms "first", "second" and similar words used in the present application do not represent any order, quantity or importance, but are used to distinguish different components. Similarly, "one" or "a" and similar words do not represent a quantity limit, but represent the existence of at least one. The words "connected" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to represent relative positional relationships, and when the absolute position of the described object changes, the relative positional relationship also changes accordingly.

[0041] Embodiment 1

[0042] See Figs. 1-3 The embodiment of the present application provides a deep learning inversion method of the magnetotelluric method, which comprises:

[0043] Obtain geophysical information and petrophysical information of a target exploration area, and construct a synthetic resistivity geoelectric model set based on the geophysical information and the petrophysical information;

[0044] Calculate the synthetic resistivity geoelectric model by magnetotelluric one-dimensional forward calculation to obtain corresponding synthetic apparent resistivity and synthetic phase data, and establish a magnetotelluric synthetic data sample set based on the synthetic apparent resistivity and the synthetic phase data;

[0045] Construct a magnetotelluric one-dimensional deep learning inversion model based on the synthetic apparent resistivity, the synthetic phase data and the synthetic resistivity geoelectric model, and train the magnetotelluric one-dimensional inversion model by the magnetotelluric synthetic data sample set to obtain a trained magnetotelluric one-dimensional inversion model;

[0046] Obtain measured apparent resistivity and measured phase data of the target exploration area, convert the measured apparent resistivity and the measured phase data into interpolated apparent resistivity and interpolated phase data by linear interpolation, and construct a multi-window Savitzky-Golay filter set by using a Savitzky-Golay filter;

[0047] Smooth the interpolated apparent resistivity and the interpolated phase data in series by using the multi-window Savitzky-Golay filter set to obtain smoothed apparent resistivity and smoothed phase data, and construct a to-be-inverted apparent resistivity-phase data set based on the smoothed apparent resistivity and the smoothed phase data;

[0048] Input the to-be-inverted apparent resistivity-phase data set into the trained magnetotelluric one-dimensional inversion model to obtain a corresponding inversion resistivity geoelectric model set, obtain a target resistivity geoelectric model based on the inversion resistivity geoelectric model set, and use the target resistivity geoelectric model as electromagnetic data of the target exploration area.

[0049] In the above embodiment, before the magnetotelluric measured data is input into the trained deep learning model for inversion prediction, the multi-window Savitzky-Golay filter proposed in the present application is used to smooth the magnetotelluric measured apparent resistivity and phase data, and then the smoothed apparent resistivity and phase data are input into the trained deep learning model for inversion prediction, thereby improving the inversion effect of the deep learning method on the magnetotelluric measured data and promoting the practicality of the magnetotelluric deep learning inversion.

[0050] Optionally, the measured apparent resistivity and the measured phase data of the target exploration area include:

[0051] The target exploration area is actually explored by a magnetotelluric instrument, the apparent resistivity and the phase data of the target exploration area are collected, and the collected apparent resistivity and phase data are used as the measured apparent resistivity and the measured phase data.

[0052] Optionally, a synthetic MT data sample set is established by the synthetic apparent resistivity, synthetic phase data and synthetic resistivity geoelectric model set, including:

[0053] The synthetic apparent resistivity, synthetic phase data and data in the synthetic resistivity geoelectric model set are one-to-one corresponding, and constitute a synthetic MT data sample set for deep learning model training.

[0054] Optionally, a synthetic MT one-dimensional inversion model is constructed based on the synthetic apparent resistivity, synthetic phase data and synthetic resistivity geoelectric model, including:

[0055] The synthetic apparent resistivity and synthetic phase data are taken as double-channel inputs of the model network, and the synthetic resistivity geoelectric model is taken as expected output of the model network, and a synthetic MT one-dimensional inversion model is constructed, wherein the input dimension of the model network is , and the output dimension of the model network is .

[0056] Optionally, the synthetic MT one-dimensional inversion model is trained by the synthetic MT data sample set to obtain a trained synthetic MT one-dimensional inversion model, including:

[0057] The synthetic MT data sample set is divided into a training set and a validation set, and the apparent resistivity and phase data in the training set and the validation set are converted into training set apparent resistivity and training set phase data with a length of X by a linear interpolation method;

[0058] The training set apparent resistivity, training set phase data and training set resistivity geoelectric model are input into the synthetic MT one-dimensional inversion model for training to obtain a trained synthetic MT one-dimensional inversion model.

[0059] Optionally, a target multi-window Savitzky-Golay filter set is constructed by a Savitzky-Golay filter, including:

[0060] As an optional implementation manner, the polynomial order of the Savitzky-Golay filter is set to , and the recommended polynomial order is 3, the window set containing windows is set to , wherein is the window length. As an optional implementation manner, the maximum window length is set according to the noise level, and the recommended maximum window length is or ;

[0061] An integer between 1 and is randomly generated, and then a window is randomly selected from the window set Selected from A window will select The windows are sorted in ascending order of window length to form the target multi-window Savitzky-Golay filter set.

[0062] Optionally, after obtaining the smoothed apparent resistivity and smoothed phase data, and before constructing the apparent resistivity-phase dataset to be inverted based on the smoothed apparent resistivity and smoothed phase data, the method further includes:

[0063] The process involves repeating the two operations N times: constructing a target multi-window Savitzky-Golay filter set using the Savitzky-Golay filter set and serially smoothing the apparent resistivity and phase data using the target multi-window Savitzky-Golay filter set to obtain smoothed apparent resistivity and smoothed phase data. Here, N is a positive integer.

[0064] Optionally, the target resistivity geoelectric model is obtained based on the inverted resistivity geoelectric model set, including:

[0065] The data of all resistivity geoelectric models in the inverted resistivity geoelectric model set are summed, and the average value of the summation results is taken.

[0066] The calculated average value is used as the target resistivity geoelectric model.

[0067] Example 2

[0068] like Fig. 2 , 3 As shown, the present invention provides a deep learning inversion method for magnetotellurics, which includes the following specific steps:

[0069] (1) In this embodiment, the cubic spline interpolation method is used to generate a composite resistivity geoelectric model with a smooth and gradual change in layered resistivity. The number of layers is 50, and the resistivity ranges from 1 to 10000. The model is 1km in size and 1km in depth. With a length of 5km, a set of resistivity geoelectric models containing 100,000 synthetic models is established.

[0070] (2) The corresponding synthetic magnetotelluric apparent resistivity was calculated by one-dimensional forward modeling of magnetotellurics. and phase Data was collected at 40 frequency points within the range of 0.001Hz to 1000Hz, at logarithmic intervals. The calculated composite apparent resistivity and composite phase data were then mapped one-to-one with the composite resistivity geoelectric model. This constitutes the magnetotelluric composite data sample set containing 100,000 sets of sample data used for training the deep learning model in this embodiment.

[0071] (3) This embodiment uses residual neural network unit and U-Net network architecture to construct a deep neural network model suitable for magnetotelluric one-dimensional inversion, the input of the network is double-channel apparent resistivity and phase, the dimension is , the output of the network is resistivity geoelectric model data, the dimension is , the loss function calculation formula is , wherein, and are the predicted and expected resistivity geoelectric model, is the number of samples for training.

[0072] (4) The magnetotelluric synthetic data sample set is randomly divided into training set and validation set according to the ratio of 8:2, and the network model established above is trained. Before the training starts, the initial apparent resistivity and phase data in the training set and the validation set are converted into apparent resistivity and phase data with a length of by linear interpolation method; during the training process, the interpolated apparent resistivity and phase data are used as network input data, and the resistivity geoelectric model is used as network expected output data; after the training is completed, the trained magnetotelluric deep learning inversion model is saved. This embodiment builds the network based on the Tensorflow platform, and the related hyperparameter settings in other training processes are: the training optimizer is Adam algorithm, the batch size is 128, the initial learning rate is 0.01, when the validation error does not decrease for three times in a row, the learning rate is reduced by a factor of 0.8, and the number of training times (epoch) is 200.

[0073] (5) This embodiment adds 0%~10% random disturbance to the synthetic magnetotelluric apparent resistivity and phase data to simulate the measured data, which is used as the inversion example data of this embodiment. The measured magnetotelluric measured apparent resistivity and phase data are converted into interpolated apparent resistivity and interpolated phase data with a length of by linear interpolation.

[0074] (6) The polynomial order of the Savitzky-Golay filter is set to 3, and the window set containing 8 windows is set to , wherein is the window length, and the maximum window length is 33.

[0075] (7) A random integer between 1 and 4 is generated Then randomly select a window from the window set Sort the selected windows according to the window length from low to high to form a target window set .

[0076] (8) Use the target window set to serially smooth the interpolated apparent resistivity and phase data to obtain smoothed apparent resistivity and phase data .

[0077] (9) Repeat steps (7) and (8) 100 times to finally obtain a set of inverted resistivity and phase data .

[0078] (10) Input the set of inverted resistivity and phase data into the trained magnetotelluric deep learning inversion model to obtain a set of corresponding inverted resistivity geoelectric models , and then average the 100 sets of resistivity geoelectric model values in the set of inverted resistivity geoelectric models to obtain as the final inversion result. Fig. 3 The expected and inverted resistivity geoelectric models shown in the figure show that when inverting magnetotelluric noisy data, the inversion method provided by the present application achieves excellent inversion effect, and the inverted resistivity geoelectric model is highly consistent with the true resistivity geoelectric model.

[0079] The present application also provides a magnetotelluric deep learning inversion system, comprising a processor, a memory;

[0080] The memory is used to store a computer program;

[0081] The processor is used to execute the program stored on the memory to realize any method step in the magnetotelluric determination method.

[0082] The above magnetotelluric deep learning inversion system can realize each embodiment of the above magnetotelluric deep learning inversion method and achieve the same beneficial effects, and thus will not be described here.

[0083] The above detailed the preferred embodiments of the present application. It should be understood that those skilled in the art can make many modifications and changes without creative labor according to the concept of the present application. Therefore, any technical solution obtained by logical analysis, reasoning or limited experiment based on the prior art according to the concept of the present application shall be within the protection scope determined by the claims.​​

Claims

1. A deep learning inversion method for magnetotellurics, characterized in that, The method comprises the following steps: obtaining geophysical information and rock physical information of a target exploration area, and constructing a set of synthetic resistivity geoelectric models based on the geophysical information and the rock physical information; calculating the set of synthetic resistivity geoelectric models by magnetotelluric one-dimensional forward modeling to obtain corresponding synthetic apparent resistivity and synthetic phase data, and establishing a set of magnetotelluric synthetic data samples based on the synthetic apparent resistivity and the synthetic phase data; constructing a magnetotelluric one-dimensional inversion model based on the synthetic apparent resistivity, the synthetic phase data and the set of synthetic resistivity geoelectric models, and training the magnetotelluric one-dimensional inversion model by the set of magnetotelluric synthetic data samples to obtain a trained magnetotelluric one-dimensional inversion model; obtaining measured apparent resistivity and measured phase data of the target exploration area, converting the measured apparent resistivity and the measured phase data into interpolated apparent resistivity and interpolated phase data by linear interpolation, and constructing a set of multi-window Savitzky-Golay filters by using a Savitzky-Golay filter; smoothing the interpolated apparent resistivity and the interpolated phase data in series by using the set of multi-window Savitzky-Golay filters to obtain smoothed apparent resistivity and smoothed phase data, and constructing a set of to-be-inverted resistivity-phase data based on the smoothed apparent resistivity and the smoothed phase data; inputting the set of to-be-inverted resistivity-phase data into the trained magnetotelluric one-dimensional inversion model to obtain a set of corresponding inversion resistivity geoelectric models, obtaining a target resistivity geoelectric model based on the set of inversion resistivity geoelectric models, and taking the target resistivity geoelectric model as electromagnetic data of the target exploration area.

2. The deep learning inversion method of magnetotellurics according to claim 1, characterized in that, The method comprises the following steps: carrying out real exploration on the target exploration area by a magnetotelluric instrument, collecting apparent resistivity and phase data of the target exploration area, and taking the collected apparent resistivity and phase data as measured apparent resistivity and measured phase data.

3. The deep learning inversion method of magnetotellurics according to claim 1, characterized in that, The method comprises the following steps: corresponding the synthetic apparent resistivity, the synthetic phase data and the data in the set of synthetic resistivity geoelectric models one by one to form a set of magnetotelluric synthetic data samples for training of a deep learning model.

4. The deep learning inversion method of magnetotellurics according to claim 1, characterized in that, The method comprises the following steps: The synthetic apparent resistivity and the synthetic phase data are taken as double-channel inputs of a model network, and the synthetic resistivity geoelectric model is taken as an expected output of the model network to construct a magnetotelluric one-dimensional inversion model, wherein a dimension of the model network input is , and a dimension of an output of the model network is .

5. The method of claim 1, wherein, The method comprises the following steps: dividing the set of magnetotelluric synthetic data samples into a training set and a validation set, and converting the apparent resistivity and phase data in the training set and the validation set into training set apparent resistivity and training set phase data with a length of X by a linear interpolation method; training the magnetotelluric one-dimensional inversion model by inputting the training set apparent resistivity, the training set phase data and the training set resistivity geoelectric model to obtain a trained magnetotelluric one-dimensional inversion model.

6. The method of deep learning inversion of magnetotelluric according to claim 1, wherein, The method for constructing the target multi-window Savitzky-Golay filter set by using the Savitzky-Golay filter comprises the following steps: The polynomial order of the Savitzky-Golay filter is set to , the set of windows containing windows is set to , where is the window length, the maximum window length is set according to the noise level; Randomly generate a value between 1 and Integers between Then randomly select from the window set Selected from A window will select The windows are sorted in ascending order of window length to form the target multi-window Savitzky-Golay filter set.

7. The method of claim 1, wherein, After the smoothed apparent resistivity and the smoothed phase data are obtained, before the target apparent resistivity-phase data set is constructed based on the smoothed apparent resistivity and the smoothed phase data, the method further comprises the following steps: The operation of constructing the target multi-window Savitzky-Golay filter set by using the Savitzky-Golay filter and the operation of obtaining the smoothed apparent resistivity and the smoothed phase data by using the target multi-window Savitzky-Golay filter set to serially smooth and interpolate the apparent resistivity and the phase data are repeated for N times, wherein N is a positive integer.

8. The deep learning inversion method of magnetotellurics according to claim 1, characterized in that, The target resistivity geoelectric model is obtained based on the set of inversion resistivity geoelectric models, which comprises the following steps: All the resistivity geoelectric model data in the set of inversion resistivity geoelectric models are summed up, and the average value of the summed result is calculated; The average value calculated is taken as the target resistivity geoelectric model.

9. A deep learning inversion system for magnetotellurics, characterized in that, The device comprises a processor and a memory. The memory is used for storing a computer program. The processor is used for executing the program stored in the memory, so as to realize the method steps in any one of claims 1-8. The device comprises a processor and a memory. The memory is used for storing a computer program. The processor is used for executing the program stored in the memory, so as to realize the method steps in any one of claims 1-8.

Images