Method for constructing geoelectric structure prediction model based on time-frequency electromagnetism

Abstract

The invention provides a method for constructing a geoelectric structure prediction model based on time-frequency electromagnetism, and belongs to the field of oil exploration. The method comprises the following steps: firstly, establishing an initial reservoir model based on a model-driven geophysical inversion process; then geophysical data and a corresponding reservoir model data set are collected, classified and standardized, and construction of the data set is completed; establishing a reservoir geoelectric structure prediction model based on the deep learning model; and finally training the reservoir geoelectric structure prediction model by using the data set to obtain an optimized reservoir geoelectric structure prediction model. Wherein the reservoir comprises an ultra-deep carbonate rock reservoir; the invention further provides a method for constructing the model for predicting the geoelectric structure of the ultra-deep carbonate reservoir. Compared with the prior art, according to the geoelectric structure prediction model constructed by the method, when observation data is input, an inversion result can be obtained without executing an iteration process, and the reservoir oil-water relationship can be further analyzed according to the inversion result.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of oil exploration, and particularly relates to a method for constructing a geoelectric structure prediction model based on time-frequency electromagnetic. BACKGROUND

[0002] The problems of complex reservoir space, large burial depth, reservoir connectivity and oil-water identification in ultra-deep carbonate reservoirs have become the main difficulties in the exploration and development of fault-dissolution body reservoirs. At present, there is no effective method for oil-water identification in ultra-deep carbonate reservoirs, while time-frequency electromagnetic has good resolution ability for oil-water identification. Time-frequency electromagnetic (TFEM) is a newly emerging oil exploration technology, which uses a working mode similar to large offset seismic exploration, provides strong current to excite oil and gas exploration targets, and measures the secondary electromagnetic field and electromagnetic field spectrum formed by the discharge of oil and gas reservoir pore media.

[0003] Time-frequency electromagnetic method mainly studies the vertical magnetic induction signal and horizontal electric field signal in long wire source electromagnetic method, and evaluates the oil and gas content of traps according to the resistivity and polarization rate information. This technology can obtain time domain and frequency domain signals at the same time, accurately reconstruct the underground physical property model through joint processing of time domain and frequency domain signals, and obtain the resistivity and polarization rate anomalies of oil and gas exploration targets. It can be seen that time-frequency electromagnetic method has important significance in oil and gas exploration, which can help technical personnel better understand the underground physical property model and the distribution of oil and gas reservoirs, and improve the accuracy and efficiency of oil and gas exploration.

[0004] Chinese invention patent CN115047531A discloses a transient electromagnetic data inversion method based on ELM network, which is characterized in that: synthetic sample data set is generated according to the mapping relationship of geoelectric model parameters containing resistivity and layer thickness and observation response data; the ELM network structure is established according to the synthetic data set and the network training is completed; the theoretical TEM data and the actual observation TEM data are input into the trained network to realize the rapid interpretation of transient electromagnetic data, and the resistivity distribution of the earth model is obtained by using the transient electromagnetic data inversion modeling; the invention mainly solves the problem of large data amount and strong multi-solution of transient electromagnetic inversion, and does not apply to the specific application of oil-water identification in ultra-deep carbonate reservoirs. Since the aforementioned reservoirs usually have high temperature, high pressure and complex lithology characteristics, it is difficult to directly apply the existing time-frequency electromagnetic technology.

[0005] The Chinese invention patent CN114048823A discloses a method for establishing a resistivity inversion model based on a full convolution network, comprising: establishing a magnetotelluric sounding model sample data set with different resistivity sizes and shapes, and converting the corresponding apparent resistivity and phase data by using Maxwell equation forward transformation; taking the apparent resistivity and phase data as model input, and taking the corresponding magnetotelluric sounding model as output to construct a full convolution neural network U-net network, and performing learning mapping, and training and adjusting the network to obtain the best inversion network weight and hyperparameter model; although the invention does not use time-frequency electromagnetic method to construct the geoelectric model, it still completes the positioning and imaging of the geoelectric model; the application of the invention in the process of oil-water identification of super-deep carbonate reservoirs still needs research and test optimization of technical personnel, and at present, it cannot meet the needs of actual production of super-deep carbonate reservoir development.

[0006] Therefore, the person skilled in the art expects to provide a method for constructing a geoelectric structure prediction model of a super-deep carbonate reservoir based on time-frequency electromagnetic (TFEM) to provide help for oil-water identification and later oilfield exploration and development of the super-deep carbonate reservoir. SUMMARY

[0007] The present application provides a method for constructing a geoelectric structure prediction model based on time-frequency electromagnetic to solve the problem of lack of geoelectric model construction method for oil-water identification of super-deep carbonate reservoir in the prior art, combines logging, VSP (Vertical Seismic Profiling) and other data to constrain the initial model, optimizes and updates the model parameters, collects electromagnetic field data, performs time-frequency analysis, extracts frequency spectrum features as input features, combines geoelectric structure data to construct training data set and test data set, trains and establishes a prediction model through neural network deep learning to realize prediction of the geoelectric structure, provides ideas and references for construction of super-deep carbonate reservoir geoelectric model in Shunbei area, and further provides a applicable method for oil-water identification and description of super-deep carbonate reservoir.

[0008] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows:

[0009] In one aspect, the present application provides a method for constructing a geoelectric structure prediction model based on time-frequency electromagnetic, comprising the following steps:

[0010] S1. Establishing an initial reservoir model based on a model-driven geophysical inversion process;

[0011] S2. Collecting geophysical data and corresponding reservoir model data set, classifying, performing standardization processing, and completing construction of the data set;

[0012] S3. Establishing a reservoir geoelectric structure prediction model based on a deep learning model;

[0013] S4. training the reservoir geoelectric structure prediction model obtained in step S3 using the data set constructed in step S2 to obtain an optimized reservoir geoelectric structure prediction model;

[0014] The reservoir in step S3 comprises an ultra-deep carbonate reservoir.

[0015] Preferably, the method for establishing the initial reservoir model in step S1 comprises the following steps: establishing an initial ultra-deep carbonate reservoir model based on a model-driven geophysical inversion process.

[0016] Further preferably, the ultra-deep carbonate reservoir model is iterated; the parameters involved in the iteration include well logging data and VSP logging data.

[0017] Preferably, the geophysical data in step S2 comprises seismic data, electromagnetic data, etc. and reservoir model data.

[0018] Further preferably, the reservoir model data comprises lithology, porosity and saturation.

[0019] The sources of the geophysical data include seismic data, well logging data and VSP logging data.

[0020] Preferably, the categories of the classification in step S2 include training data set, validation data set and test data set.

[0021] Preferably, the standardization processing in step S2 is to eliminate the dimensional differences between data.

[0022] Preferably, the deep learning model in step S3 comprises data input layer, intermediate layer and output layer.

[0023] Further preferably, the intermediate layer comprises regularization term design, loss function design and iteration process design.

[0024] Further preferably, the deep learning model is selected from at least one of a backpropagation (BP) neural network model, a convolutional neural network (CNN) model and a deep neural network (DNN) model.

[0025] Preferably, the data set constructed in step S2 in step S4 comprises training data set and validation data set.

[0026] Preferably, the training method in step S4 is:

[0027] (1) using self-compiled code to perform single-point time-domain electromagnetic method forward for each of the earth structure model sample to obtain a forward response curve sample set;

[0028] (2) constructing an MLP neural network for multi-classification task, inputting the forward curve sample set data obtained in step (1) into the MLP neural network for training, and establishing a mapping relationship between the earth structure model and the single-point time-domain electromagnetic method response curve;

[0029] (3) inputting the data set into the multilayer perceptron (MLP, Multilayer Perceptron) neural network, and obtaining the output predicted earth structure model after mapping.

[0030] On the other hand, the application provides a method for constructing a predicted earth structure model of an ultra-deep carbonate reservoir, comprising the following steps:

[0031] ① using the time-frequency electromagnetic based earth structure prediction model to predict the earth structure model of the test data to obtain the output earth structure model;

[0032] ② comparing and analyzing the output earth structure model obtained in step ① with the label earth model in the test data to obtain the predicted earth structure model of the ultra-deep carbonate reservoir.

[0033] Preferably, the test data in step ① is time-domain electromagnetic observation data.

[0034] The time-domain electromagnetic observation data is single-point one-dimensional time-domain data, the receiving parameter is located 20 meters above the ground, the receiving coil is a horizontal ring coil with a radius of 6 meters, and the receiver measurement parameter is the vertical component of the magnetic flux at the coil center; the transmitting parameter is to use a horizontal ring transmitting source, the current amplitude of the transmitting peak value is 1A, and the transmitting waveform is a step signal.

[0035] Preferably, the method for constructing the label earth model in step ② is:

[0036] According to the transmitting parameter and the receiving parameter, using self-compiled code to perform single-point time-domain electromagnetic method forward for each of the earth structure model sample to obtain a forward response curve sample set;

[0037] An MLP neural network for multi-classification task is constructed, the forward response curve sample set is input into the multilayer perceptron neural network for training, and after the training is completed, a mapping relationship between the earth structure model and the single-point time-domain electromagnetic method response curve is established; the observation data is input into the multilayer perceptron neural network, and the output predicted earth structure model, i.e. the label earth model, is obtained after mapping.

[0038] Preferably, the comparison and analysis method in step ② comprises the following steps:

[0039] A sample set of 10,000 stratified geoelectric structure models is randomly generated by an algorithm, with a depth of 0-250 meters underground, a layer thickness of 25 meters, and each layer with an electrical conductivity randomly generated by an algorithm in a range of 10 -3 S / m to 10 1 S / m; and the random generation result is compared with an actual geoelectric structure model.

[0040] Compared with the prior art, the present application has the following beneficial effects:

[0041] The geoelectric structure prediction model provided by the present application can train a deep neural network for mapping from observation data to a prediction model, so that the deep neural network can predict a corresponding predicted geoelectric structure model according to new, unidentified observation data.

[0042] The geoelectric structure prediction model constructed by the present application has a pre-completed iteration process in the data preprocessing stage, so that when the observation data is input, the iteration process does not need to be performed, and only one step is needed to obtain the inversion result, and the reservoir oil-water relationship can be further analyzed according to the inversion result. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 A flowchart of the method for constructing a geoelectric structure prediction model based on time-frequency electromagnetic of embodiment 1 of the present application.

[0044] Figure 2 A structure diagram of the deep learning model described in S3 of embodiment 1 of the present application.

[0045] Figure 3 A comparison diagram of an actual geoelectric structure model, a geoelectric structure model obtained by deep learning inversion, and a geoelectric structure model obtained by L2 norm regularization inversion.

[0046] Figure 4 A comparison diagram between observation data, forward response data corresponding to a deep learning inversion method, and predicted response data corresponding to an L2 norm regularization inversion method. DETAILED DESCRIPTION

[0047] Embodiment 1: A method for constructing a geoelectric structure prediction model based on time-frequency electromagnetic

[0048] S1. Establish an initial reservoir model based on a model-driven geophysical inversion process;

[0049] The establishment of the initial reservoir model is based on the collection of electromagnetic data of ultra-deep carbonate reservoirs, which can be obtained by surface or well time-frequency electromagnetic technology methods. The collected electromagnetic data is processed, and combined with the geological condition analysis of the work area:

[0050] An initial ultra-deep carbonate reservoir model is established based on a model-driven geophysical inversion process, as shown in Figure 1 The model will be updated and iterated by combining geological data such as logging data and VSP logging data.

[0051] S2. Collect geophysical data and corresponding reservoir model data sets, classify, and perform standardization processing to complete the construction of the data set;

[0052] The geophysical data includes: collecting geophysical data including seismic data, electromagnetic data, and reservoir model data;

[0053] The source of the geophysical data includes: seismic data, logging data, and VSP logging data;

[0054] The classification categories include: training data set, validation data set, and test data set;

[0055] The standardization processing is to eliminate the dimensional differences between the data.

[0056] S3. Establish an ultra-deep carbonate reservoir geoelectric structure prediction model based on a deep learning model;

[0057] The deep learning model in step S3 includes a data input layer, an intermediate layer, and an output layer;

[0058] The intermediate layer includes: regularization term design, loss function design, and iteration process design;

[0059] The deep learning model is selected from a BP neural network model, a CNN model, and a DNN model;

[0060] S4. Train the ultra-deep carbonate reservoir geoelectric structure prediction model obtained in step S3 using the data set constructed in step S2 to obtain an optimized reservoir geoelectric structure prediction model;

[0061] The data set constructed in step S2 in step S4 includes a training data set and a validation data set;

[0062] The training method in step S4 is:

[0063] (1) Use self-coded code to perform single-point time-domain electromagnetic method forward for each geoelectric structure model sample to obtain a forward response curve sample set;

[0064] (2) Construct an MLP neural network for multi-classification tasks, input the forward curve sample set data obtained in step (1) into the MLP neural network for training, and establish a mapping relationship between the geoelectric structure model and the single-point time-domain electromagnetic method response curve;

[0065] (3) inputting the data set into the MLP neural network, and obtaining an output of a predicted geoelectric structure model after mapping;

[0066] The embodiment also provides a method for constructing a predicted geoelectric structure model of an ultra-deep carbonate reservoir, which adopts the method for constructing a predicted geoelectric structure model based on time-frequency electromagnetic construction to construct the predicted geoelectric structure model, and includes the following steps:

[0067] ① further adopting the method for constructing a predicted geoelectric structure model based on time-frequency electromagnetic construction to predict the test data to obtain an output of a predicted geoelectric structure model;

[0068] The test data in step ① include time-domain electromagnetic observation data; the time-domain electromagnetic observation data are single-point one-dimensional time-domain data, the receiving parameter is 20 meters above the ground, the receiving coil is a horizontal ring coil with a radius of 6 meters, and the receiver measurement parameter is the magnetic flux vertical component at the center of the coil; the transmitting parameter is a horizontal ring transmitting source, the transmitting peak current amplitude is 1 A, and the transmitting waveform is a step signal.

[0069] ② comparing and analyzing the output of the predicted geoelectric structure model obtained in step ① with a label geoelectric model in the test data to obtain a predicted geoelectric structure model of an ultra-deep carbonate reservoir;

[0070] The method for constructing the label geoelectric model in step ② is as follows: according to the transmitting parameter and the receiving parameter, a self-compiled code is used to perform single-point time-domain electromagnetic method forward modeling on the geoelectric structure model samples one by one to obtain a forward response curve sample set. An MLP neural network for a multi-classification task is constructed, the forward curve sample set data is input into the MLP neural network for training, and after the training is completed, a mapping relationship between the geoelectric structure model and the single-point time-domain electromagnetic method response curve is established, and the observation data is input into the MLP neural network.

[0071] The method for comparing and analyzing in step ② includes the following steps: first, 10,000 layered geoelectric structure model samples are randomly generated by an algorithm, the depth is 0-250 meters underground, the layer thickness is 25 meters, and the electrical conductivity of each layer is randomly generated by an algorithm in the range of 10 -3 S / m to 10 1 S / m.

[0072] Figure 3 The comparison chart of the inverted geoelectric structure model and the actual model is given in Table 3, which is specifically: the comparison of the actual geoelectric structure model, the geoelectric structure model obtained by deep learning inversion (step S3 in the embodiment), and the geoelectric structure model obtained by L2 norm regularization inversion (the final predicted geoelectric structure model obtained in the embodiment).

[0073] Figure 4The inversion corresponding to the comparison chart of the predicted response data and the actual observation data is given in the middle; specifically, the comparison between the observation data, the forward response data corresponding to the deep learning inversion method and the predicted response data corresponding to the L2 norm regularization inversion method. From Figure 4 It can be seen from the middle that the predicted response data corresponding to the L2 norm regularization inversion method partially coincides with the actual model.

[0074] At present, the super-deep carbonate fracture-cave type reservoir in Tahe has entered the stage of large-scale water injection and gas injection for enhanced oil recovery, and the oil-water relationship has become more complex. The development well deployment, infilling adjustment and sidetracking water breakthrough risk are increasing; the understanding of fluid rules and the unclear distribution of remaining oil in Tahe area have become the bottleneck restricting the deepening of exploration and development in Tahe oilfield; in view of the application status of time-frequency electromagnetic technology in super-deep oil and gas reservoirs, the key technologies and operation standards of time-frequency electromagnetic processing for super-deep carbonate oil and gas reservoirs are established to improve the inversion accuracy. The collected geophysical data including seismic data, electromagnetic data and corresponding reservoir model data are sorted and classified according to the characteristics and attributes.

[0075] Finally, it should be noted that the above content is only used to illustrate the technical solutions of the present application, and is not a limitation on the protection scope of the present application. Simple modifications or equivalent replacements of the technical solutions of the present application made by those skilled in the art do not deviate from the essence and scope of the technical solutions of the present application.

Claims

1. A method for constructing a geoelectric structure prediction model based on time-frequency electromagnetics, characterized in that, Includes the following steps: S1. Establish an initial reservoir model based on a model-driven geophysical inversion process; S2. Collect geophysical data and corresponding reservoir model datasets, classify them, perform standardization processing, and complete the construction of the dataset; S3. Establish a reservoir geoelectric structure prediction model based on a deep learning model; S4. The reservoir geoelectric structure prediction model obtained in step S3 is trained using the dataset constructed in step S2 to obtain an optimized reservoir geoelectric structure prediction model. The reservoir mentioned in step S3 includes ultra-deep carbonate rock reservoirs.

2. The method according to claim 1, characterized in that, The method for establishing the initial reservoir model described in step S1 includes the following steps: establishing an initial ultra-deep carbonate reservoir model based on a model-driven geophysical inversion process.

3. The method according to claim 2, characterized in that, The ultra-deep carbonate reservoir model is iterated; the parameters involved in the iteration include well logging data and VSP well logging data.

4. The method according to claim 1, characterized in that, The geophysical data mentioned in step S2 includes: seismic data, electromagnetic data, and reservoir model data; the reservoir model data includes: lithology, porosity, and saturation.

5. The method according to claim 1, characterized in that, The categories described in step S2 include: training dataset, validation dataset, and test dataset.

6. The method according to claim 1, characterized in that, The standardization process described in step S2 is to eliminate dimensional differences between data.

7. The method according to claim 1, characterized in that, The deep learning model described in step S3 includes a data input layer, an intermediate layer, and an output layer.

8. The method according to claim 7, characterized in that, The intermediate layer includes: regularization term design, loss function design, and iterative process design.

9. The method according to claim 7, characterized in that, The deep learning model is selected from at least one of the following: backpropagation neural network model, convolutional neural network model, and deep neural network model.

10. The method according to claim 1, characterized in that, In step S4, the dataset constructed in step S2 includes a training dataset and a validation dataset.

11. The method according to claim 1, characterized in that, In step S4, the training method is as follows: (1) Using self-written code, the geoelectric structure model samples were subjected to single-point time-domain electromagnetic forward modeling to obtain a sample set of forward modeling response curves; (2) Construct a multilayer perceptron neural network for multi-classification tasks, input the forward curve sample set data obtained in step (1) into the multilayer perceptron neural network for training, and establish the mapping relationship between the geoelectric structure model and the time-domain electromagnetic response curve of a single measurement point. (3) Input the dataset into the multilayer perceptron neural network and obtain the output predicted geoelectric structure model after mapping.

12. A method for constructing a predictive geoelectric structure model for ultra-deep carbonate reservoirs, characterized in that, The method described in any one of claims 1-11 is used to construct the geoelectric structure prediction model.

13. The method for constructing a predictive geoelectric structure model for ultra-deep carbonate reservoirs according to claim 12, characterized in that, Includes the following steps: ① The geoelectric structure prediction model is used to predict the geoelectric structure model of the test data to obtain the output geoelectric structure model; ② Compare and analyze the output geoelectric structure model obtained in step ① with the labeled geoelectric model in the test data to obtain the predicted geoelectric structure model of ultra-deep carbonate reservoirs.

14. The method for constructing a predictive geoelectric structure model for ultra-deep carbonate reservoirs according to claim 13, characterized in that, The test data mentioned in step ① includes: time-domain electromagnetic observation data; the time-domain electromagnetic observation data is one-dimensional time-domain data at a single measurement point, the receiving parameters are located 20 meters above the ground, the receiving coil is a horizontal loop coil with a radius of 6 meters, the receiver measurement parameters are the vertical component of the magnetic flux at the center of the coil; the transmission parameters are a horizontal loop transmitter source, the peak current amplitude of the transmission is 1A, and the transmission waveform is a step signal.

15. The method for constructing a predictive geoelectric structure model for ultra-deep carbonate reservoirs according to claim 13, characterized in that, The method for constructing the tag-based geoelectric model in step ② is as follows: Based on the transmission and reception parameters, a self-written code is used to perform forward modeling of each geoelectric structure model sample using the time-domain electromagnetic method at a single measurement point, resulting in a sample set of forward modeling response curves; a multilayer perceptron neural network for multi-classification tasks is constructed, and the forward modeling curve sample set data is input into the multilayer perceptron neural network for training. After training, a mapping relationship between the geoelectric structure model and the single-point time-domain electromagnetic response curve is established, and the observation data is input into the multilayer perceptron neural network to obtain the result.

Images