Geologic body weight consideration method, device and medium for predicting subsurface fluid

Abstract

The application discloses a kind of underground fluid prediction method, equipment and medium considering geological body weight, method includes: well logging gas content curve is converted from depth domain to time domain and is matched with corresponding well seismic record;Well logging gas content curve is matched with well seismic record and is formed into label set by square wave processing, training sample set is formed by using short-time Fourier transform to well seismic record, the k value of KNN algorithm, distance measurement mode and classification decision rule are determined, wherein, consider geological body weight in the classification decision process of KNN algorithm, give the training sample of more close to the predicted sample greater weight, output the predicted sample and the gas sample in the first k most close training sample of predicted sample with the probability attribute of considering geological body weight;The predicted data is segmented into the predicted data block of equal size with training sample, and the predicted data block is input into KNN algorithm and is predicted.The accuracy of underground fluid distribution prediction is improved.

Description

Technical Field

[0001] This invention belongs to the field of geological exploration technology, and more specifically, relates to a method, equipment and medium for predicting underground fluids that takes into account the weight of geological bodies. Background Technology

[0002] In recent decades, one of the most common techniques for hydrocarbon prediction has been to directly extract one or more sensitive attributes from seismic data using linear or simple nonlinear operations. However, the seismic response of tight sandstone gas reservoirs is influenced by numerous factors, such as reservoir thickness, lithofacies, fluid distribution, pore structure, porosity, and fluid type. Therefore, the relationship between seismic attributes extracted from seismic data and fluids in tight sandstone reservoirs is not one-to-one, and predicting fluids using one or more independent attributes is highly ambiguous. Furthermore, because the pore fluid response and seismic response in tight sandstone reservoirs generally exhibit a highly nonlinear relationship, using linear or simple nonlinear operations to obtain seismic attributes to indicate the reservoir fluids themselves is not accurate enough. Therefore, using complex nonlinear operations to extract multiple interrelated attributes or features holds promise for enriching seismic attribute extraction theory and expanding the types of seismic attributes.

[0003] Besides seismic attributes, using pre-stack inversion to obtain elastic parameters or further transforming them to obtain fluid indicator factors to predict subsurface fluid distribution is currently the mainstream technique for seismic fluid detection. However, the accuracy of pre-stack inversion results depends on the simplification of the wave equation model. The more complex the wave equation used, the higher the accuracy of simulating actual data, but the more parameters are inverted, the more severe the parameter coupling crosstalk becomes, and the more unstable the inversion process becomes. Therefore, pre-stack inversion technology has always sought a balance between "stability" and "accuracy." To achieve industrialization, the linearized and concise expression of the Zoeppritz equation is currently widely used to characterize the relationship between elastic parameters and seismic data, and a series of practical AVO inversion techniques have been developed. However, these practical techniques generally require conditions such as "actual observation data satisfying the plane wave superposition principle, the actual subsurface medium being perfectly elastic and isotropic, and the elastic parameters of adjacent strata having small differences," which limits the application scope of fluid prediction. Moreover, from the perspective of information content, linearization loses the nonlinear seismic information components with large offsets that are more sensitive to fluids.

[0004] Furthermore, the constraints of well logging, geological, or electromagnetic field information that originates from the same source but is heterogeneous as the earthquake are rarely considered during the inversion process. Insufficient effective seismic information is the fundamental reason limiting the accuracy of existing pre-stack inversion results. Therefore, how to fully explore and utilize richer seismic information and integrate more geophysical field information to achieve fluid prediction technology based on multi-geophysical field multi-dimensional seismic information fusion models, using more complex nonlinear relationships and more effective information to achieve a dual improvement in "stability" and "accuracy," has significant scientific and practical value.

[0005] Since the 1950s, machine learning technology has made tremendous progress. Using machine learning methods to analyze seismic data allows for the extraction of reservoir information from massive amounts of seismic data. Currently, machine learning technology is rapidly being applied to seismic data processing, interpretation, and reservoir prediction. However, research on applying artificial intelligence technology to fluid prediction is relatively limited and still in its early stages.

[0006] Because the actual distribution of underground fluids is unknown, most artificial intelligence (AI) methods suffer from the black-box problem, making it difficult to evaluate the effectiveness of AI methods for fluid prediction and hindering their application. Statistical machine learning algorithms (such as neural networks) build global models based on all training samples, resulting in high sample acquisition costs and sample imbalance. The underground distribution of tight sandstone reservoirs is complex and variable; building a global model may lead to overly complex models and increased training costs. Traditional machine learning classification algorithms make independent predictions in each instance, failing to consider the continuity of underground structures. Classification tasks often use "voting" decision rules, which cannot fully utilize the geophysical information contained in the differences between samples.

[0007] The KNN algorithm is widely used in pattern recognition and data mining. KNN is also an instance-based lazy learning algorithm. Such algorithms directly store training samples, and the KNN learning algorithm can create a local function within a certain nearest neighbor range of the sample point to be predicted. This local simulation function may be more suitable for the characteristics of fluid prediction problems. Applying simple intelligent methods first to the field of geophysics not only facilitates an objective and comprehensive evaluation of intelligent methods but also conforms to Occam's razor principle.

[0008] Traditional KNN uses a manual determination of the k-value, employing a "voting" classification decision rule. A drawback of this algorithm is its heavy reliance on the training sample database, resulting in generally low prediction accuracy. Furthermore, the distances between samples in the task that traditional KNN addresses often lack definite physical meaning. "Voting" is a commonly used decision rule in machine learning methods for solving classification tasks. However, when applying KNN to fluid prediction, considering the geological significance of the algorithm may further improve its performance. Summary of the Invention

[0009] The purpose of this invention is to propose a method, device, and medium for predicting underground fluids that takes into account the weight of geological bodies, so as to make full use of seismic, well logging, and geological information and improve the accuracy of underground fluid distribution prediction based on multi-dimensional and multi-scale fluid-sensitive information.

[0010] In a first aspect, the present invention proposes a method for predicting underground fluids that considers the weight of geological bodies, comprising:

[0011] The well logging gas content curve is converted from the depth domain to the time domain and matched with the corresponding wellside seismic record to complete well-seismic matching.

[0012] The well logging gas content curves that have completed well-seismic matching are processed into square waves to form a tag set. The wellside seismic records are then processed using short-time Fourier transform to form a training sample set, with each training sample corresponding to a tag.

[0013] Determine the value of k, distance metric, and classification decision rules for the KNN algorithm. In the classification decision process of the KNN algorithm, geological body weights are considered, and training samples that are closer to the sample to be predicted are given greater weights. The probability attribute of the sample to be predicted and the gas-bearing samples among the k closest training samples is output, taking into account the geological body weights.

[0014] The data to be predicted is divided into blocks of the same size as the training samples. Each block is then input into the KNN algorithm for prediction. This yields the probability attribute of the location corresponding to the center sampling point of each block, taking into account the weight of the geological body. Finally, the probability attribute of the entire data to be predicted, taking into account the weight of the geological body, is obtained. Combined with prior geological information, the distribution of underground fluids is predicted.

[0015] Optionally, the step of square-wave processing the well logging gas-bearing curves that have completed well-vibration matching to form a tag set includes:

[0016] Each sampling point in the well logging gas content curve is classified into two categories, where sampling points containing gas are represented by 1 and sampling points not containing gas are represented by 0. The processed well logging gas content curve forms a label set.

[0017] Optionally, the process of using short-time Fourier transform to process the seismic records along the well to form a training sample set includes:

[0018] The well-side seismic records are segmented using a fixed-length sliding time window to form a training sample set.

[0019] Optionally, the number of rows in each training sample in the training sample set corresponds to the length of the sliding window, the number of columns corresponds to the number of offset channels, and the label of each training sample is determined by the label corresponding to the center sampling point of the training sample.

[0020] Optionally, determining the distance metric for the KNN algorithm includes:

[0021] The L1 distance between the training samples and the data to be predicted is chosen as the distance metric to obtain the geological similarity between the samples to be predicted and the training samples.

[0022] Optionally, dividing the data to be predicted into blocks of data to be predicted that are the same size as the training samples includes:

[0023] The data to be predicted is divided into blocks of the same size as the training samples using a fixed-length sliding window.

[0024] Optionally, the formula for calculating the probabilistic attribute considering the geological body weight is:

[0025]

[0026] In the formula, k z D is the number of training samples that are closest to the sample to be predicted. mi (i = 1, 2, ..., k) z ) is k z The distance between the nearest training sample and the sample to be predicted is given by n, where n is k. z The number of classes that appear most frequently in each training sample is class c, where n ≤ k. z D cj (j=1,2,…,n) represents the distance between the n training samples of class c and the sample to be predicted, and s is the probability attribute of the sample to be predicted being classified as c when considering the weight of geological bodies.

[0027] Optionally, the geological prior information includes geological body continuity information.

[0028] Secondly, the present invention provides an electronic device, the electronic device comprising:

[0029] At least one processor; and,

[0030] A memory communicatively connected to the at least one processor; wherein,

[0031] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the subsurface fluid prediction method considering geological body weights as described in the first aspect.

[0032] Thirdly, the present invention provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the subsurface fluid prediction method considering geological body weights as described in the first aspect.

[0033] The beneficial effects of this invention are as follows:

[0034] This invention first transforms the well logging gas-bearing curve from the depth domain to the time domain and matches it with the corresponding well-side seismic records. Then, the well logging gas-bearing curves with completed well-seismic matching are processed into square waves to form a label set. Short-time Fourier transform is used to process the well-side seismic records to form a training sample set. Next, the value of k, the distance metric, and the classification decision rules for the KNN algorithm are determined. In the classification decision process of the KNN algorithm, geological body weights are considered, giving greater weight to training samples closer to the sample to be predicted. The probability attribute considering geological body weights is output for the sample to be predicted and the gas-bearing samples among the k closest training samples. Then, the data to be predicted is divided into data blocks of the same size as the training samples, and each data block is input into the KNN algorithm for prediction, obtaining the corresponding center sampling point of each data block. The invention considers the probabilistic attributes of geological body weights in the location-based prediction process, ultimately obtaining the probabilistic attribute volume of the entire data to be predicted, taking into account the geological body weights. Combined with prior geological information, it predicts the distribution of underground fluids. This invention obtains high-dimensional seismic information through time-frequency analysis, which is combined with low-dimensional seismic amplitude information and well logging information as training data for the method. It makes full use of well logging, seismic, and geological information, and fully utilizes the fusion of high-dimensional seismic frequency domain information with low-dimensional seismic information (amplitude information, etc.) and well logging information at different scales to extract multi-dimensional and multi-scale fluid-sensitive information for underground fluid distribution prediction. By considering geological body weights in the KNN classification decision process, higher decision weights are given to training samples with underground geological information similar to the samples to be predicted, improving the accuracy of fluid prediction and fully utilizing the geophysical significance of machine learning methods in the problem of underground fluid prediction.

[0035] The system of the present invention has other features and advantages that will be apparent from or will be set forth in detail in the accompanying drawings and following detailed description, which together serve to explain the particular principles of the invention. Attached Figure Description

[0036] The above and other objects, features and advantages of the present invention will become more apparent from the accompanying drawings, in which like reference numerals generally denote like parts.

[0037] Figure 1 A flowchart illustrating the steps of a subsurface fluid prediction method considering geological body weights according to Embodiment 1 of the present invention is shown.

[0038] Figure 2a This diagram shows the actual fluid distribution in a tight sandstone reservoir.

[0039] Figure 2b The results of fluid distribution prediction in tight sandstone reservoirs using the traditional KNN classification method are shown.

[0040] Figure 2c The results of fluid distribution prediction in tight sandstone reservoirs are shown using a subsurface fluid prediction method that takes into account the weight of geological bodies, as described in Example 1.

[0041] Figure 3 The diagram shows a cross-sectional view of the well sections from actual data in a certain work area.

[0042] Figure 4 The paper presents the gas-bearing distribution results obtained by predicting the underground fluid distribution based on actual data and well profiles using a groundwater prediction method that considers the weight of geological bodies according to Example 1.

[0043] Figure 5 The image shows a slice plot along the layer 4ms below layer l1, obtained using a subsurface fluid prediction method that considers the weight of geological bodies according to Example 1.

[0044] Figure 6 The image shows a slice plot along the stratum 27ms above layer l2, obtained using a subsurface fluid prediction method that takes into account the weight of geological bodies, based on Example 1. Detailed Implementation

[0045] k-Nearest Neighbor (kNN) learning is a commonly used supervised learning method. Its working mechanism is very simple: given a test sample, it finds the k nearest training samples in the training set based on a certain distance metric, and then makes a prediction based on these k training samples. In classification tasks, a "voting method" can be used, where the class label that appears most frequently among the k nearest samples is used as the prediction result; in regression tasks, an "averaging method" can be used, where the average of the real-valued output labels of these k samples is used as the prediction result; weighted voting or weighted averaging can also be used based on distance, with closer samples having higher weights. The algorithm for classification tasks based on kNN learning is simple and intuitive. Its algorithm description is as follows:

[0046] 1) Calculate the distance between the test data and each training data point;

[0047] 2) Sort according to increasing distance;

[0048] 3) Select the k points with the smallest distance;

[0049] 4) Determine the frequency of occurrence of the category of the first k points;

[0050] 5) Return the category with the highest frequency among the top k points as the predicted category for the test data.

[0051] As can be seen, the kNN algorithm has three key elements: the selection of the k value, the distance metric, and the classification decision rule. kNN is a "lazy learning" algorithm, meaning that during the training phase, the samples are simply stored, resulting in zero training time. Processing is only performed after receiving test samples. Conversely, methods that process samples during the training phase are called "eager learning." The kNN algorithm has advantages such as simplicity, ease of understanding, ease of implementation, no need for parameter estimation or training, and suitability for classifying rare events.

[0052] This invention extracts high-dimensional frequency domain information from seismic data using time-frequency analysis technology, and fuses this information with low-dimensional information (waveform, amplitude, etc.) from the seismic data, well logging information, and geological information to obtain multi-dimensional, multi-scale fluid information. An improved KNN algorithm is used to predict underground fluid distribution based on fluid-sensitive information. The physical meaning of the KNN algorithm's measurement of "distance" between samples is explored, and the classification decision rules are improved to address the characteristics of fluid prediction problems. The similarity of geological body information is considered. Training samples with similar geological body information to the samples to be predicted are given greater decision weights. The prediction results are output as a probability attribute indicating "considering geological body weights." By fully utilizing prior geological information to constrain the classification decision process, the accuracy of fluid prediction is further improved.

[0053] The invention will now be described in more detail with reference to the accompanying drawings. While preferred embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0054] Example 1

[0055] Figure 1 A flowchart illustrating the steps of a subsurface fluid prediction method that considers the weight of geological bodies according to the present invention is shown.

[0056] like Figure 1 As shown, a method for predicting subsurface fluids that considers the weight of geological bodies includes:

[0057] Step S1: Convert the well logging gas content curve from the depth domain to the time domain and match it with the corresponding wellside seismic record to complete well-seismic matching;

[0058] Specifically, this step involves processing the well-seismic distribution consistency. Using well-seismic matching technology, the gas-bearing curves from the well logging are converted from the depth domain to the time domain and matched with the corresponding wellside seismic records.

[0059] Step S2: The gas-bearing curves of the well logging that have completed well-seismic matching are processed into square waves to form a tag set. The well-side seismic records are processed using short-time Fourier transform to form a training sample set, with each training sample corresponding to a tag.

[0060] In this embodiment, each sampling point in the well logging gas-bearing curve is binary-classified, with gas-bearing sampling points represented by 1 and non-gas-bearing sampling points represented by 0. The processed well logging gas-bearing curve forms a label set. The wellside seismic records are segmented using a fixed-length sliding time window to form a training sample set. The number of rows in each training sample in the training sample set corresponds to the length of the sliding time window, and the number of columns corresponds to the number of offset traces. The label of each training sample is determined by the label corresponding to the central sampling point of the training sample.

[0061] Specifically, to reduce classification difficulty, the gas-bearing curves in well logging are processed into square waves, and each sampling point is binary-classified—gas-bearing (represented by 1) and non-gas-bearing (represented by 0). The processed gas-bearing curves form a label set. A sample set is formed by segmenting the wellside seismic records using a fixed-length sliding time window, thereby extracting local seismic response features near the sampling points. The number of rows in the sample corresponds to the length of the sliding time window, and the number of columns corresponds to the number of offset traces. The label of each sample is determined by the label corresponding to the sampling point at the center of the sample.

[0062] Furthermore, as seismic waves propagate downwards, their energy attenuates over time. Different lithologies absorb and attenuate frequencies differently, and the frequency attenuation changes after passing through oil, gas, and water layers differ from those after passing through non-reservoir strata. High frequencies attenuate rapidly, while low frequencies attenuate slowly, and the dominant frequency gradually decreases over time. Time-frequency analysis methods can provide valuable high-dimensional frequency domain information for reservoir prediction, effectively demonstrating the relationship between signal frequency and time. A commonly used tool in seismic data analysis is the Fourier Transform, which maps seismic signals from the time domain to the frequency domain.

[0063] To characterize the local features of the signal, this embodiment uses STFT (Short Time Fourier Transform), which uses a fixed-width window function to capture the spectrum of the signal, and then continuously moves the window function and records the time to obtain the time-frequency distribution of the entire signal.

[0064] The STFT of signal s(t) is defined as follows:

[0065]

[0066] In the formula, h(t) is a window function.

[0067] Signal s(t) can be reconstructed using inverse STFT:

[0068]

[0069] In this embodiment, the well-side seismic records processed by short-time Fourier transform are high-dimensional and can reflect high-dimensional seismic frequency domain information. High-dimensional seismic information obtained through time-frequency analysis is combined with low-dimensional seismic amplitude information from well logging gas-bearing curves and well logging information as training data for the method. When the KNN method measures the samples, it actually reflects the similarity of geological body information between samples. Step S3: Determine the value of k, distance measurement method, and classification decision rule for the KNN algorithm. In the classification decision process of the KNN algorithm, geological body weights are considered, giving greater weight to training samples closer to the sample to be predicted. The output is the probability attribute considering geological body weights between the sample to be predicted and the gas-bearing samples among the k closest training samples.

[0070] Specifically, the appropriate value of k, the distance metric, and the classification decision rule are determined. This embodiment uses the L1 distance between the sample and the data to be predicted as the distance metric to obtain the geological similarity between the sample to be predicted and the training samples. Furthermore, based on the characteristics of underground gas content, the classification decision rule for kNN is determined. Here, it is assumed that if any of the k most similar samples to the data to be predicted contains a gas-bearing sample, then the data to be predicted carries gas-bearing information to some extent. To retain this key information to the greatest extent, the probability attribute considering the geological body weights of the gas-bearing samples among the k "closest" training samples is output for the sample to be predicted.

[0071] In this embodiment, the distance between the sample to be predicted and the training sample is measured by calculating their Euclidean distance. The optimal distance measurement method is determined by comparing the prediction performance when using L1 distance and L2 distance respectively.

[0072] The kNN algorithm using L1 distance calculates the sum of the absolute values ​​of the differences between each feature of the sample to be predicted and the training samples.

[0073]

[0074] in, This represents the i-th feature of the training sample; represents the i-th feature of the sample to be predicted; n represents the number of features in the training samples and the data to be predicted.

[0075] The kNN algorithm using L2 distance calculates the square root of the sum of the absolute values ​​of the differences between each feature of the sample to be predicted and the training samples.

[0076]

[0077] in, This represents the i-th feature of the training sample; represents the i-th feature of the sample to be predicted; n represents the number of features in the training samples and the data to be predicted.

[0078] Preferably, in this embodiment, the L1 distance between the training sample and the data to be predicted is used as the distance metric to obtain the geological similarity between the sample to be predicted and the training sample.

[0079] Furthermore, when using the KNN method to solve machine learning tasks, the "distance metric" between the sample to be predicted and the training samples often lacks clear physical meaning and has poor interpretability. However, in geophysics, if underground geological conditions are similar, their seismic responses generally also exhibit similarities. Therefore, in the problem of underground fluid prediction, the "distance metric" between samples reflects, to some extent, the similarity of the geological information they carry. Simultaneously, in the classification decision-making process using the kNN method, the k value is manually assigned. A large k value may introduce training samples with low correlation to the sample to be predicted, leading the method to make incorrect decisions.

[0080] Therefore, this invention considers geological body weights in the classification decision-making process, that is, assigning greater weights to training samples that are "closer" to the sample to be predicted, which can effectively utilize prior geological information to constrain the classification decision-making process. The weighted decision-making process considering geological body weights in this embodiment is as follows:

[0081] Suppose we have a sample z to be predicted, and based on some distance metric, we select the top k samples from the training sample library. z The k training samples that are closest to the sample to be predicted. z The distance between the closest training sample and the sample to be predicted is D. mi (i = 1, 2, ..., k) z ). Where k z The class that appears most frequently in the training samples is class c, and the number of such classes is n (n≤k). z The distance between these n training samples of class c and the sample to be predicted is D. cj (j = 1, 2, ..., n). Considering the weights of geological bodies, the probability attribute of the sample to be classified as c is:

[0082]

[0083] Considering the weights of geological bodies, the probabilistic attribute of the sample to be predicted and the training samples of class c can, to some extent, characterize the comprehensive similarity of the sample to be predicted and all class c samples in the training sample library in carrying geological body information. Functionally, it can serve as the probability of classifying the sample to be predicted into class c.

[0084] It should be noted that the probability attribute output by the KNN algorithm is not actually a probability value. After outputting the probability attribute of each sample to be predicted in the prediction results of this method, further geological prior information (such as the continuity of geological bodies) can be used to determine the distribution of hydrocarbon content.

[0085] Step S4: Divide the data to be predicted into data blocks of the same size as the training samples, and input each data block into the KNN algorithm for prediction to obtain the probability attribute of the location of the central sampling point of each data block considering the geological body weight. Finally, obtain the probability attribute volume of the entire data to be predicted considering the geological body weight. Combined with geological prior information, predict the distribution of underground fluids.

[0086] Specifically, the data to be predicted is divided into data blocks of the same size as the samples using the same sliding window as in step S2, and then input into the KNN algorithm for prediction. The predicted results are used as the probabilistic attribute of the geological body weights at the corresponding locations of the central sampling points of this data block, and finally the probabilistic attribute volume of the geological body weights of the entire data to be predicted is obtained. Combined with geological prior information, fluid is predicted based on the weighted geological body similarity results.

[0087] To verify the application effect of the method in this embodiment, a numerical model and an actual four-dimensional pre-stack data volume were used for testing in a specific application scenario.

[0088] First, a numerical model is used to illustrate the effectiveness and superiority of the underground fluid prediction method in this embodiment.

[0089] Figure 2a This figure shows the actual gas-bearing distribution profile of the numerical model. The light-colored bowl-shaped and strip-shaped areas represent gas-bearing reservoirs, labeled 1, while the dark-colored areas represent non-gas-bearing areas, labeled 0. To simulate actual conditions, four channels are randomly selected from the profile as "pseudo-wells" to provide well logging gas-bearing curves. The corresponding synthetic four-dimensional pre-stack seismic gathers are used as well-side seismic data. Training samples are created using the method described in this invention based on the "pseudo-well" data and the corresponding well-side pre-stack gather data. Gas-bearing prediction is then performed based on the pre-stack four-dimensional seismic data of the entire numerical model.

[0090] Figure 2b The display shows the gas-bearing distribution prediction results obtained using the traditional KNN method with k=5. It can be observed that the predicted reservoir morphology is incomplete. This is because the k value is fixed, and the number of training samples in the non-gas-bearing category is significantly greater than the number of training samples in the gas-bearing category. During the classification decision process, a voting mechanism is used, and the category that appears most frequently among the k closest training samples is chosen as the category of the sample to be predicted.

[0091] and Figure 2cThe results shown are the predictions obtained using the method of this embodiment. The reservoir morphology prediction is relatively accurate. The numerical model illustrates that the method of this invention can effectively improve the prediction of reservoir fluid distribution.

[0092] To demonstrate the practical application potential of this invention, the underground fluid prediction method that considers the weight of geological bodies was applied to a specific work area, and certain results were achieved.

[0093] The actual work area covers approximately 100 square kilometers, with a temporal sampling interval of 1 ms and a spatial sampling interval of 25 m. The seismic records are pre-stack offset gathers, with offsets ranging from 500 m to 4100 m. The work area includes logging data from 5 wells, and the reservoir type is tight sandstone reservoir. The reservoir area is characterized by thin interbedded sandstone and mudstone layers, which presents certain exploration challenges.

[0094] Figure 3 This image shows a post-stack well profile based on actual data. The vertical broken line represents the well logging gas-bearing curve, and the horizontally extending curve represents two seismic layers, numbered l1 and l2 from top to bottom. The target reservoir is located between these two layers. Observation reveals that there is very little seismic information between the two layers in the post-stack well profile, with almost no seismic phase axes.

[0095] Figure 4 The gas-bearing prediction results obtained by the method of this embodiment are shown. In the figure, the probability of gas-bearing in the area decreases from light to dark colors. Three thin gas-bearing layers were observed near stratigraphic level l2, and they showed good agreement with the well logging gas-bearing curves. The existing geological understanding of the work area is that braided rivers are developed near stratigraphic level l1, with reservoirs distributed along paleochannels and developing in the central shoal, with relatively low gas content. Alluvial fans are developed near stratigraphic level l2, with gas layers distributed in a continuous manner and rich in gas content.

[0096] Figure 5 The figure shows the gas content distribution along a slice 4 ms below layer l1 obtained by the method of this embodiment. The curves in the figure mark two paleochannels. The gas content distribution is relatively scattered, which is consistent with the existing geological understanding of this area.

[0097] Figure 6 The results of the gas-bearing distribution along a slice 27ms above layer l2 obtained by the method of this embodiment are shown. The gas-bearing distribution is continuous and consistent with the existing geological knowledge of the work area.

[0098] In summary, the method of the present invention has the following technical effects:

[0099] 1) This invention makes full use of well logging, seismic and geological information, and fully utilizes the fusion of high-dimensional seismic frequency domain information with low-dimensional seismic information (amplitude information, etc.) and well logging information of different scales to extract multi-dimensional and multi-scale fluid-sensitive information for underground fluid distribution prediction.

[0100] 2) This invention utilizes an improved kNN learning method to establish a "scale" between multi-dimensional and multi-scale fluid-sensitive information and underground fluid distribution. In the kNN classification decision process, geological body weights are considered, and training samples that are similar to the underground geological information of the sample to be predicted are given higher decision weights, thereby improving the accuracy of fluid prediction and making full use of the geophysical significance of machine learning methods in the problem of underground fluid prediction.

[0101] 3) This invention serves as an exploration of applying intelligent methods to solve the problem of predicting underground fluid distribution. Considering the difficulty in verifying geophysical problems, this invention emphasizes the interpretability and evaluability of the methods when using machine learning to predict underground fluid distribution, rather than being limited to the advanced nature of intelligent methods. This approach is innovative and has guiding significance in the research of applying intelligent methods to geophysical data processing, inversion, and interpretation.

[0102] Example 2

[0103] An electronic device, the electronic device comprising:

[0104] At least one processor; and,

[0105] A memory communicatively connected to the at least one processor; wherein,

[0106] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the subsurface fluid prediction method considering geological body weights as described in Embodiment 1.

[0107] An electronic device according to an embodiment of the present disclosure includes a memory and a processor.

[0108] This memory is used to store non-transitory computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0109] The processor may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. In one embodiment of this disclosure, the processor is used to execute computer-readable instructions stored in the memory.

[0110] Those skilled in the art will understand that, in order to solve the technical problem of how to achieve a good user experience, this embodiment may also include well-known structures such as communication buses and interfaces, and these well-known structures should also be included within the protection scope of this disclosure.

[0111] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0112] Example 3

[0113] The present invention proposes a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to execute the subsurface fluid prediction method considering geological body weights as described in Example 1.

[0114] A computer-readable storage medium according to embodiments of the present disclosure stores non-transitory computer-readable instructions. When these non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the methods described in the foregoing embodiments of the present disclosure are performed.

[0115] The aforementioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or portable hard drive), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge).

[0116] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A method for predicting underground fluids that considers the weight of geological bodies, characterized in that, include: The well logging gas content curve is converted from the depth domain to the time domain and matched with the corresponding wellside seismic record to complete well-seismic matching. The well logging gas content curves that have completed well-seismic matching are processed into square waves to form a tag set. The wellside seismic records are then processed using short-time Fourier transform to form a training sample set, with each training sample corresponding to a tag. Determine the value of k, distance metric, and classification decision rules for the KNN algorithm. In the classification decision process of the KNN algorithm, geological body weights are considered, and training samples that are closer to the sample to be predicted are given greater weights. The probability attribute of the sample to be predicted and the gas-bearing samples among the k closest training samples is output, taking into account the geological body weights. The data to be predicted is divided into blocks of the same size as the training samples. Each block is then input into the KNN algorithm for prediction. This yields the probability attribute of the location corresponding to the center sampling point of each block, taking into account the weight of the geological body. Finally, the probability attribute of the entire data to be predicted, taking into account the weight of the geological body, is obtained. Combined with prior geological information, the distribution of underground fluids is predicted.

2. The underground fluid prediction method according to claim 1, characterized in that, The step of converting the well logging gas-bearing curves that have completed well-seismic matching into square waves to form a tag set includes: Each sampling point in the well logging gas content curve is classified into two categories, where sampling points containing gas are represented by 1 and sampling points not containing gas are represented by 0. The processed well logging gas content curve forms a label set.

3. The underground fluid prediction method according to claim 1, characterized in that, The process of using short-time Fourier transform to process the seismic records along the well track to form a training sample set includes: The well-side seismic records are segmented using a fixed-length sliding time window to form a training sample set.

4. The underground fluid prediction method according to claim 3, characterized in that, The number of rows in each training sample in the training sample set corresponds to the length of the sliding window, the number of columns corresponds to the number of offset channels, and the label of each training sample is determined by the label corresponding to the center sampling point of the training sample.

5. The underground fluid prediction method according to claim 1, characterized in that, The distance metric used in the KNN algorithm includes: The L1 distance between the training samples and the data to be predicted is chosen as the distance metric to obtain the geological similarity between the samples to be predicted and the training samples.

6. The underground fluid prediction method according to claim 3, characterized in that, The step of dividing the data to be predicted into blocks of data to be predicted, which are the same size as the training samples, includes: The data to be predicted is divided into blocks of the same size as the training samples using a fixed-length sliding window.

7. The underground fluid prediction method according to claim 1, characterized in that, The formula for calculating the probabilistic attribute that considers the weight of geological bodies is as follows: In the formula, k z The number of training samples that are closest to the sample to be predicted. i =1,2,…, k z , D mi for k z The distance between the nearest training sample and the sample to be predicted. n for k z The category that appears most frequently in the training samples is c The number of classes n≤k z , j =1,2,…, n , D cj Let be the distance between n training samples of class c and the sample to be predicted. s This is the probability attribute for classifying a sample to be predicted as c when considering the weight of geological bodies.

8. The underground fluid prediction method according to claim 1, characterized in that, The geological prior information includes information on the continuity of geological bodies.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the subsurface fluid prediction method considering geological body weights as described in any one of claims 1-8.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions for causing a computer to perform the subsurface fluid prediction method considering the weight of geological bodies as described in any one of claims 1-8.

Images