Intelligent detection method and device for energy abnormal channel, electronic equipment and storage medium

Abstract

The invention provides an intelligent detection method and device for an energy abnormal channel, electronic equipment and a storage medium, geometric distortion and time difference interference in original data are eliminated through shot point and receiving node distance sorting and first arrival line leveling processing, and the accuracy of subsequent analysis is improved; the method comprises the following steps: extracting local energy indexes, capturing tiny anomalies, enhancing the sensitivity to local energy such as noise and local geological anomalies, extracting global energy indexes, identifying overall energy such as systematic equipment faults or large-scale geological change trend anomalies, avoiding the one-sidedness of local detection, and improving the detection comprehensiveness. A multi-classification deep network model is adopted, training is supervised through energy sample label data, the classification robustness is improved, the detection and quality control effect on node trace gather data energy abnormal traces in seismic data acquisition field quality control is improved, final classification is determined through statistics of the most types of model output appearing, accidental errors are reduced, and the result stability is enhanced.

Description

Technical Field

[0001] This application relates to the field of petroleum seismic exploration and is a method, device, electronic equipment, and storage medium for intelligent detection of energy anomalies in seismic data acquisition and on-site quality control of nodal acquisition data. Background Technology

[0002] In the field of petroleum seismic exploration, nodal acquisition equipment, with its advantages of portability, convenient deployment, economy, and construction efficiency, is gradually replacing traditional equipment and becoming the dominant technology in seismic exploration. It has been widely used in numerous petroleum exploration acquisition projects. As the importance of nodal acquisition in seismic exploration becomes increasingly prominent, the demand for quality control of nodal gather data is also increasing. Among these, the quality control of energy indicators is a crucial aspect. Traceways with excessively high or low energy levels will fail to meet the quality standards for seismic data acquisition, severely impacting data processing and imaging quality. For example, an excessive number of traceways with abnormal energy levels will affect the quality of seismic data acquisition, the construction progress of seismic data acquisition, and data delivery. Therefore, it is essential to conduct rigorous quality control and analysis of the energy indicators of nodal gather data to ensure the reliability and effectiveness of seismic data.

[0003] In the nodal acquisition process of seismic exploration, the acquired seismic data is gathered in gathers organized around each node as the basic unit. These gathers represent seismic signals generated by excitation sources at different times and locations, recording the entire propagation process of seismic waves from the excitation source to the receiving node. Because seismic waves are affected by various factors during propagation, seismic data from different gathers exhibit significant differences in excitation intensity, propagation path, human environment, and geographical location. Especially in onshore acquisition, observation systems typically use short shot lines. This characteristic manifests in nodal gathers as short gathers, poor energy consistency, uneven energy distribution within nodes, short gathers, poor continuity, and poor regularity, making it difficult to detect energy indicators. Furthermore, the range of seismic signal data varies greatly; nodal instruments typically have a resolution of 24 or 32 bits, corresponding to a dynamic range exceeding 120 dB, meaning the difference between the maximum and minimum values ​​can be in the millions or tens of millions. Within the same artillery or node collection, the difference in channel data between channels with extremely high or extremely low energy is not significant, making the determination very difficult.

[0004] Traditional shot gather quality control methods for addressing energy anomalies in trace data employ neighbor-to-neighbor energy difference or neighbor-to-neighbor energy ratio. This involves calculating whether the difference between the current trace's energy and that of its neighboring traces, or its difference from that of a standard trace, exceeds a certain range to determine if an energy anomaly is present. Alternatively, it can be determined by calculating the energy ratio between the current trace and its neighboring traces, or the ratio of the current trace's energy to the average energy of the shot gather traces. These methods are simple and fast, but they rely on fixed thresholds or fixed standard traces, limiting their applicability. Similar methods have been used in the quality control of energy anomaly indicators for nodal trace gathers, facing the same problems and limitations. Summary of the Invention

[0005] To address one of the aforementioned technical shortcomings, this application provides an intelligent detection method for energy anomaly traces. This method solves the problem that the energy range of node trace-level data varies greatly, and the detection standards and methods are fixed and singular, making them unsuitable for widespread application in field seismic data quality control scenarios. In particular, it addresses the energy differences between the characteristics of anomaly traces in different trace sets and those of normal traces, thereby achieving high-quality quality control of the energy indicators of node trace set data.

[0006] According to a first aspect of the embodiments of this application, an intelligent detection method for energy anomaly traces is provided, comprising the following steps: acquiring node gather data; reordering each trace data based on the distance between each shot point and the seismic receiving node in the node gather data and performing leveling processing on the node gather data with the first arrival line as a reference to complete the preprocessing of the node gather data; dividing the preprocessed node gather data into multiple data units with a preset size, calculating the energy nonlinear quantization classification data of each data unit to generate local energy index data; calculating the overall energy nonlinear quantization classification data of the entire gather data on a trace-by-trace basis to generate global energy index data; generating energy sample label data on a trace-by-trace basis for the preprocessed node gather data; using the local and global energy index data and the energy sample label data as a multi-class deep network training model to process the input data to achieve multi-class detection of energy anomaly traces; and, based on the output results of the multi-class deep network training model, statistically analyzing the type that appears most frequently in each trace as the predicted type of the corresponding trace, and generating the classification detection results of each trace data in the node gather data.

[0007] In some optional embodiments, the step of calculating the energy nonlinear quantization classification data of each data unit includes calculating the nonlinear quantization classification index of the mean amplitude energy, maximum amplitude energy, minimum amplitude energy, and median amplitude energy of each data unit.

[0008] In some optional embodiments, the step of calculating the energy nonlinear quantization classification data of the entire gather includes calculating the nonlinear quantization classification index of the mean amplitude energy, the maximum amplitude energy, the minimum amplitude energy, and the median amplitude energy on a per-trace basis.

[0009] In some optional embodiments, the multi-class deep network training model is a dual-input branch-code-decode network model. The dual-input branch-code-decode network model is used to perform the following steps: extracting nonlinear quantization hierarchical features containing local energy; extracting nonlinear quantization hierarchical features containing all local energy; extracting anomaly index features from local energy features and global energy features; performing attention weighting on local energy anomaly index features and global energy anomaly index features respectively; and concatenating and fusing the local energy anomaly index features and the global energy anomaly index features before outputting the result.

[0010] In some optional embodiments, the multi-class deep network training model is a multi-class loss function model, wherein the multi-class loss function model is,

[0011]

[0012]

[0013] in, For the total loss error, C represents the weighting factor for the classification results, where C is the total number of categories. Error loss function oriented towards classification results, The ratio in which the model correctly predicts abnormal channels as abnormal and normal channels as normal. This refers to the percentage of normal channels that the model predicts as abnormal channels. The ratio of energy anomalies to normal channels; The penalty factor for predicting a normal channel as an abnormal channel. The penalty factor for predicting an abnormal channel as a normal channel; This is a multi-class cross-entropy calculation method for energy anomalies. For the nth point to be of class c, this is the actual situation. Let n be the probability that the nth point belongs to the cth class. The ratio of normal lanes predicted to be normal lanes.

[0014] In some optional embodiments, the energy sample tag data includes ultra-high energy channels, ultra-low energy channels, normal energy channels, and zero-value channels.

[0015] According to a second aspect of the embodiments of this application, an intelligent detection device for energy anomalies is provided, used to execute the intelligent detection method for energy anomalies as mentioned in any of the preceding claims, comprising: a preprocessing module, used to acquire node gather data, reorder each gather data based on the distance between each shot point and the seismic receiving node in the node gather data, and perform leveling processing on the node gather data based on the first arrival line, so as to complete the preprocessing of the node gather data; a local energy index calculation unit, used to divide the preprocessed node gather data into multiple data units with a preset size, calculate the energy nonlinear quantization classification data of each data unit, so as to generate local energy index data; and a global energy index calculation unit; The system comprises: a module for calculating the nonlinear quantization and grading of energy across the entire trace set on a trace-by-trace basis after preprocessing the node trace set data; a sample label generation unit for generating energy sample label data on a trace-by-trace basis after preprocessing the node trace set data; a multi-class deep network training model for processing the local and global energy index data and the energy sample label data as input data to achieve multi-class detection of energy anomaly traces; and a classification detection unit for calculating the most frequent type of each trace based on the output of the multi-class deep network training model, using it as the predicted type for the corresponding trace, and generating classification detection results for each trace in the node trace set data.

[0016] In some optional embodiments, the dual-input branch coding-decoding network model includes: a multi-scale preprocessing module for extracting nonlinear quantization hierarchical features containing local energy; a coding network for extracting nonlinear quantization hierarchical features encompassing all local energy; a multi-path data attention enhancement module for extracting anomalous indicator features from local energy features and global energy features, and performing attention weighting on the local energy anomalous indicator features and global energy anomalous indicator features respectively; and a local-global dual-branch fusion module for concatenating and fusing the local energy anomalous indicator features and the global energy anomalous indicator features before outputting the result.

[0017] According to a third aspect of the embodiments of this application, an intelligent detection device for an energy anomaly channel is provided, comprising: a memory; a processor; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method mentioned in any of the above.

[0018] According to a fourth aspect of the embodiments of this application, a computer-readable storage medium is provided having a computer program stored thereon; the computer program is executed by a processor to implement the method as described in any of the preceding embodiments.

[0019] The intelligent detection method for energy anomalies provided in this application eliminates geometric distortions and time-difference interference in the original data by sorting the distances between the shot points and receiver nodes and calibrating the first arrival lines, thereby improving the accuracy of subsequent analysis. Local energy index extraction captures minute anomalies and enhances sensitivity to local energy anomalies such as noise and sudden geological changes. Global energy index extraction identifies overall energy anomalies such as systemic equipment failures or large-scale geological trend anomalies, avoiding the limitations of local detection. The combination of local and global energy indices covers anomaly scenarios from micro to macro levels, improving detection comprehensiveness. A multi-classification deep network model is employed, trained under supervised conditions using energy sample label data. It automatically learns complex nonlinear characteristics of differentiated patterns such as noise, weak signals, and strong interference. Compared to traditional threshold methods, this significantly improves classification robustness and enhances the detection and quality control of energy anomalies in node gather data during seismic data acquisition. The model output determines the final classification by statistically analyzing the most frequent type, reducing random errors and enhancing result stability. The entire process, from preprocessing and feature extraction to model classification, is automated, reducing manual intervention. It is suitable for rapid analysis of massive seismic data, especially for large-scale node gather detection in oil and gas exploration. It can accurately classify anomalous traces such as noise traces, low signal-to-noise ratio traces, and instrument malfunction traces, assisting in the screening of effective signals, improving the quality of subsequent seismic imaging and inversion, and reducing exploration costs. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A schematic diagram of the energy anomaly channel multi-classification network model provided in the embodiments of this application; Figure 2 A schematic diagram of the multi-scale preprocessing module provided in an embodiment of this application; Figure 3 A schematic diagram of a local-to-global dual-branch fusion module provided in an embodiment of this application; Figure 4 A schematic diagram of a multi-path attention enhancement module provided in an embodiment of this application; Figure 5 A flowchart of an intelligent detection method for energy anomalies provided in an embodiment of this application. Detailed Implementation

[0021] To make the technical solutions and advantages of the embodiments of this application clearer, the exemplary embodiments of this application will be described in further detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not an exhaustive list of all embodiments. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other.

[0022] To address the aforementioned problems, this application provides an intelligent detection method for energy anomaly channels, such as... Figure 5 As shown, the process includes the following steps: acquiring node gather data; reordering each trace data based on the distance between each shot point and the seismic receiving node in the node gather data; and performing leveling processing on the node gather data based on the first arrival line to complete the preprocessing of the node gather data; dividing the preprocessed node gather data into multiple data units with a preset size, and calculating the energy nonlinear quantization classification data of each data unit to generate local energy index data; calculating the overall energy nonlinear quantization classification data of the entire gather data on a trace-by-trace basis to generate global energy index data; generating energy sample label data on a trace-by-trace basis for the preprocessed node gather data; using the local and global energy index data and the energy sample label data as a multi-class deep network training model to process the input data to achieve multi-class detection of energy anomaly traces; and based on the output results of the multi-class deep network training model, statistically analyzing the type that appears most frequently in each trace as the predicted type of the corresponding trace, and generating the classification detection results of each trace data in the node gather data.

[0023] Specifically, this application provides an intelligent detection method for energy anomalies. It captures subtle anomalies such as local noise and weak signal distortion through local energy indicators, avoiding the omission of key anomalies due to global averaging in traditional methods. By combining global energy indicators to analyze the overall energy trend, it identifies systemic anomalies such as instrument drift and large-scale signal attenuation, overcoming the limitations of single-scale detection and thus achieving multi-scale energy feature extraction. First-arrival line leveling and data reordering eliminate time difference interference caused by the distance difference between the shot point and the receiver node, improving the accuracy of energy feature extraction and making the input data more suitable for deep learning models. Multi-scale energy feature extraction and first-arrival line leveling and data reordering optimize input data extraction and enhance the adaptability of anomaly detection. Based on the characteristics of energy anomalies, such as non-stationarity and local abrupt changes, a multi-branch feature fusion network is designed to enhance the ability to capture complex anomaly patterns. An attention mechanism is introduced to strengthen the feature weights of key energy regions and suppress irrelevant noise interference. In the training process of multi-class deep network training models, energy anomaly detection performance is incorporated as an optimization objective. This allows the model to focus more on key quality control indicators such as classification accuracy for low signal-to-noise ratio channels and distortion channels. From targeted optimization of the network structure to improvement of the loss function, the deep neural network design is optimized, enhancing the model's generalization ability. The entire process from preprocessing to classification output is automated, and a single processing run can cover massive amounts of node and channel data, improving efficiency by tens of times compared to manual quality inspection. It is suitable for real-time quality control in the field, achieving automated and efficient detection. This method achieves an accuracy of >90% in detecting energy anomaly channels, and reduces the false positive rate by more than 60% compared to traditional methods.

[0024] Further, a shot point is the location where artificially generated seismic waves are produced. A receiving node is a geophone or nodal device deployed on the surface or seabed to record seismic wave signals. A single shot gather is a collection of multiple seismic data channels recorded by all receiving nodes (or geophones) from seismic waves generated at the same source point (shot point). It represents the seismic traces recorded synchronously by all receiving nodes after a single excitation. It reflects the path differences of seismic waves propagating from the same shot point to different receiving nodes. A nodal gather is a collection of all seismic trace data from different shot points recorded by a single seismic receiving node. It represents seismic signals generated by multiple shot points received by the same node at different times. A node is an independent seismic data acquisition unit, containing a geophone, recording module, and power supply, typically deployed wirelessly.

[0025] Optionally, the preprocessing steps for acquiring node gather data, reordering each trace based on the distance between each shot point and the seismic receiving node in the node gather data, and leveling the node gather data using the first-arrival line as a baseline include: reading the trace head information of each trace from the node gather data, including the shot point coordinates of each trace; calculating the distance from each shot point to the current node; rearranging the trace data in ascending order of distance; and leveling the seismic data using the first-arrival line as a baseline.

[0026] Optionally, the step of dividing the preprocessed node gather data into multiple data units of a preset size, and calculating the energy nonlinear quantization and hierarchical data of each data unit to generate local energy index data includes, where the size of each data unit is n. r *n c , where n r n is the number of rows. c Let t be the number of columns. The seismic data has t columns, and each column has s sampling points. The seismic data is divided into nr groups by row, and different energy quantification classification indices are calculated for each row group. The column dimension is based on the number of columns per trace, and the data column size is n according to the depth model. c Grouping is performed, padding with zeros if necessary, to divide a gather into several small deep network model input data blocks. In the z-direction, data are grouped according to quantized grading indices of different energy levels, forming multi-channel data. The positional information of each data block within the original gather is preserved during the data organization process.

[0027] Optionally, the step of generating energy sample label data for the preprocessed node gather data by channel includes: labeling the gather data with energy categories according to energy quality control standards, identifying channels such as ultra-strong energy channels, ultra-weak energy channels, normal energy channels, and zero-value channels by channel. Based on the channel category, a two-dimensional data set with the same row and column size as the input data is generated, and each column of data is set with its category value according to its channel category in the node gather data.

[0028] During the training process, the multi-class deep network training model is trained using local and global energy index data and corresponding channel energy sample label data as input data until the convergence error reaches the specified error requirement.

[0029] In some optional embodiments, the step of calculating the energy nonlinear quantization classification data of each data unit includes calculating the nonlinear quantization classification index of the mean amplitude energy, maximum amplitude energy, minimum amplitude energy, and median amplitude energy of each data unit.

[0030] Specifically, when calculating the nonlinear quantization and grading of energy data units, multi-dimensional statistical indicators such as the mean, maximum, minimum, and median of amplitude energy are extracted as the basis for nonlinear quantization and grading. The mean amplitude energy reflects the overall energy level of the data unit, identifying overall deviations such as persistent low or high energy anomalies; the maximum amplitude energy can capture energy mutation points such as sudden noise and impulse interference, avoiding anomaly omissions caused by mean smoothing; the minimum amplitude energy detects energy voids or signal loss of zero-value data caused by node failures; the median amplitude energy suppresses outlier interference and provides a more robust energy distribution estimate, especially suitable for non-Gaussian distributed data. These indicators are mapped to a unified standard through nonlinear transformations such as logarithmic scaling and quantile grading, enhancing the ability to distinguish small energy differences. Through nonlinear quantization and grading using multiple statistics, this method significantly outperforms traditional single-indicator analysis in terms of anomaly sensitivity, anti-interference, and interpretability, providing high-information-density input features for subsequent deep learning models and serving as the core technical support for intelligent detection of energy anomalies.

[0031] Optionally, the formula for calculating the average amplitude energy of each data unit is:

[0032] in, This represents the a-th mean energy nonlinear quantization classification result of the i-th row group. This indicates the number of valid data points in the i-th row group. This represents the absolute value of the j-th number in the group, i.e., the amplitude energy. These represent the nonlinear quantization classification results of the maximum and minimum amplitude energy values ​​of the node devices, respectively.

[0033] Optionally, the formula for calculating the maximum amplitude energy of each data unit is as follows:

[0034] in This represents the b-th mean energy quantization classification result of the i-th row group. This indicates the number of valid data points in the i-th row group. This represents the absolute value of the j-th number in the group, i.e., the amplitude energy. , These represent the nonlinear quantization classification results of the maximum and minimum amplitude energy values ​​of the node devices, respectively.

[0035] Optionally, the formula for calculating the mean amplitude energy of each data unit is as follows: sort all non-zero points in each data unit according to the magnitude of amplitude energy, and take the g-th data point as the median of the largest amplitude, starting from the largest end. Take the s-th data point starting from the smallest end as the median of the small amplitude. ,

[0036]

[0037] in This represents the quantization and classification result of the median large amplitude energy in the i-th row group. , The range of maximum and minimum energy values ​​used when calculating the median energy of large amplitude vibrations. This represents the quantization and grading result of the median small amplitude energy in the i-th row group. For its corresponding quantization unit level parameter, , The range parameter for the maximum and minimum energy values ​​used when calculating the median energy of small amplitude vibrations.

[0038] In some optional embodiments, the step of calculating the energy nonlinear quantization classification data for the entire gather includes calculating nonlinear quantization classification indices for the mean amplitude energy, maximum amplitude energy, minimum amplitude energy, and median amplitude energy on a per-trace basis.

[0039] Specifically, when calculating global energy nonlinear quantization and classification data on a trace-by-trace basis, multi-dimensional statistical indicators such as the mean, maximum, minimum, and median of amplitude energy are extracted and nonlinearly quantized and classified. The mean amplitude energy reflects the overall energy level of a single trace, identifying persistent energy anomalies such as low or high energy drift across the entire trace, avoiding fragmentation problems in local detection. The maximum amplitude energy captures extreme energy events within a single trace, such as sudden strong noise or impulse interference, helping to distinguish between local and systemic anomalies. The minimum amplitude energy indicates the loss of amplitude energy detection signals or hardware failures such as dead traces or zero-value data, avoiding missed detections of traces without signals. The median amplitude energy suppresses outlier interference and provides a robust energy benchmark, especially suitable for non-stationary seismic data. Through nonlinear classification methods such as logarithmic transformation and percentile ranking, the original energy values ​​are mapped to standardized levels, compressing high dynamic range data, such as strong amplitudes differing from weak amplitudes by tens of thousands of times, making it easier for multi-class deep network training models to learn effective features. Highlighting the significance of energy differences, such as attenuating weak but critical signals, like median downgrading, can separate them from background noise.

[0040] The method for calculating the average amplitude energy per channel is the same as the method for calculating the average amplitude energy of each data unit, and will not be repeated here.

[0041] The method for calculating the maximum amplitude energy per channel is the same as the method for calculating the maximum amplitude energy of each data unit, and will not be repeated here.

[0042] The method for calculating the minimum amplitude energy on a channel basis is the same as the method for calculating the minimum amplitude energy of each data unit, and will not be repeated here.

[0043] The method for calculating the median amplitude energy per channel is the same as the method for calculating the median amplitude energy of each data unit, and will not be repeated here.

[0044] In some optional embodiments, the multi-class deep network training model is a dual-input branch-code-decode network model. The dual-input branch-code-decode network model is used to perform the following steps: extracting nonlinear quantization hierarchical features containing local energy; extracting nonlinear quantization hierarchical features containing all local energy; extracting anomaly index features from local energy features and global energy features; performing attention weighting on local energy anomaly index features and global energy anomaly index features respectively; and concatenating and fusing the local energy anomaly index features and the global energy anomaly index features before outputting the result.

[0045] Specifically, the local and global energy index input data are processed using a multi-class backbone network with a dual-input branch encoder-decoder network, such as... Figure 1 As shown, a multi-scale preprocessing module is designed for local energy anomaly data, such as... Figure 2 As shown, energy anomaly features at different scales are extracted from data from each channel. Simultaneously, a local and global dual-branch fusion module is designed at the highest abstraction level to achieve high-level filtering and fusion of energy features from the two types of data, as illustrated below. Figure 3 As shown, a multi-channel data attention enhancement module is used in the data fusion section of the upsampling and encoding / decoding modules, such as... Figure 4 As shown, multi-path data attention enhancement fusion is performed on the decoded data of the previous layer and the global and local energy index data of the same coding layer.

[0046] Optionally, for a multi-classification network model with a dual-input branch encoding-decoding backbone, and considering the characteristics of global and local channel energy index data, this invention designs an encoding-decoding network model using a dual-input branch, such as... Figure 1 As shown, this is used to achieve multi-class detection of energy anomalies. Figure 2 As shown, local energy index data undergoes multi-scale local index feature extraction through a multi-scale preprocessing module, followed by energy feature extraction through a residual enhancement module, and then feature extraction through downsampling. Each feature extraction layer is a cohesive unit consisting of convolutional layers, pooling layers, and data normalization layers. After feature extraction at each high level through a multi-layer coding network, the data reaches a local-to-global dual-branch fusion module. Global energy index data directly passes through a coding network composed of multiple residual enhancement modules and downsampling. Global energy features undergo layer-by-layer filtering to reach a local-to-global feature fusion module, as shown below. Figure 3As shown, the two data streams are fused through attention weighting in this module, preserving the main energy features at higher levels in both global and local data. This enables the filtering and fusion of high-level features from different types of data. Then, upsampling is performed to restore detailed features. The restored data, along with the data from the corresponding coding layers in the dual input branches, undergoes multi-attention fusion through a multi-attention enhancement module. Figure 4 As shown, the data is then fused with the upsampled data. Each upsampled layer consists of a deconvolution layer, a data merging layer, a convolutional layer, and a data normalization layer. Finally, the output of the activation discriminant layer is compared with the label results to calculate the error between the predicted category and the sample target, thereby correcting the weights in reverse.

[0047] Optionally, for the multi-scale preprocessing module, since local energy index data mainly focuses on local features, a multi-scale preprocessing module is used for features of a larger scale or wider range, such as... Figure 2 As shown, energy anomaly features at different scales are extracted from data from each channel. This module first normalizes the input data, then performs feature extraction at different scales in four paths. Each path involves convolution, using 1*1 convolution, 3*1 convolution, 5*1 dilated convolution, 7*1 dilated convolution, ReLU, and max pooling respectively. The results are then concatenated, and after further convolution, ReLU, and max pooling calculations, the optimal features are obtained from the features at different scales.

[0048] Optionally, for the local and global dual-branch fusion module, this module performs data feature fusion on the local energy index channel and the global overall energy index channel, such as... Figure 3 As shown, the data from the two channels are first concatenated, and then divided into two branches: a global attention channel and a local attention channel. The global attention channel includes convolution (1*1), GeLU, convolution (5*5), batch normalization, GeLU, and average pooling. The local attention channel includes convolution (1*1), GeLU, convolution (3*3), batch normalization, GeLU, and max pooling. Then, the weights of each attention channel are multiplied by the data from each channel and added together. The results are then superimposed and rearranged with the original data from each channel. Finally, convolution, Sigmoid, and max pooling calculations are performed to calculate the final output.

[0049] Optionally, for multi-channel data attention enhancement modules, such as Figure 4As shown, this method processes energy data features and location correlations across different sensing ranges. Before data fusion between the decoder and the corresponding encoders of the global and local energy indices, a multi-level attention module is employed to select and control important energy anomaly indices at different spatial scales and locations. This module has three input data sets: the local energy index channel encoding layer data, the global energy index channel encoding layer data, and the next layer data in the corresponding decoding layer. First, each input is convolved with a 1x1 layer, then element-wise summed, followed by LeakReLU, 1x1 convolution, and a sigmoid function before resampling to restore the original data size, yielding attention coefficients. Finally, these coefficients are multiplied by the sum of the global and local energy index channel data to obtain the final output data. This multi-level attention module can filter target signal region data from energy index data at different levels, especially regions where energy anomalies are observed at different levels, and enhance these regions using attention coefficients.

[0050] In some optional embodiments, to address the problems of extremely small number of energy anomaly samples, their irregular and uncertain location, time, and data, and uneven distribution of samples of different types, the multi-class deep network training model uses a multi-class loss function model. The multi-class loss function model is as follows:

[0051]

[0052]

[0053] in, For the total loss error, C represents the weighting factor for the classification results, where C is the total number of categories. Error loss function oriented towards classification results, The ratio at which the model correctly predicts abnormal channels as abnormal and normal channels as normal. This refers to the percentage of normal channels that the model predicts as abnormal channels. The ratio of energy anomalies to normal channels; The penalty factor for predicting a normal channel as an abnormal channel. The penalty factor for predicting an abnormal channel as a normal channel; This is a multi-class cross-entropy calculation method for energy anomalies. For the nth point to be of class c, this is the actual situation. Let n be the probability that the nth point belongs to the cth class. The ratio of normal lanes predicted to be normal lanes.

[0054] In some optional embodiments, the energy sample tag data includes ultra-high energy channels, ultra-low energy channels, normal energy channels, and zero-value channels.

[0055] On the other hand, this application also provides an intelligent detection device for energy anomalies, such as... Figure 1 As shown, an intelligent detection method for energy anomaly traces as described in any of the above embodiments includes: a preprocessing module for acquiring node gather data, reordering the trace data based on the distance between each shot point and the seismic receiving node in the node gather data, and performing leveling processing on the node gather data based on the first arrival line to complete the preprocessing of the node gather data; a local energy index calculation unit for dividing the preprocessed node gather data into multiple data units with a preset size, calculating the energy nonlinear quantization classification data of each data unit to generate local energy index data; and a global energy index calculation unit for processing the preprocessed node gather data. The system calculates the nonlinear quantization and hierarchical data of the entire channel set on a channel-by-channel basis to generate global energy index data. A sample label generation unit generates energy sample label data on a channel-by-channel basis from the preprocessed node channel set data. A multi-class deep network training model processes the local and global energy index data and the energy sample label data as input data to achieve multi-class detection of energy anomaly channels. A classification detection unit, based on the output of the multi-class deep network training model, calculates the most frequently occurring type for each channel as the predicted category for that channel, generating classification detection results for each channel in the node channel set data.

[0056] In some optional embodiments, the dual-input branch coding-decoding network model includes: a multi-scale preprocessing module for extracting nonlinear quantization hierarchical features containing local energy; a coding network for extracting nonlinear quantization hierarchical features encompassing all local energy; a multi-path data attention enhancement module for extracting anomalous indicator features from local energy features and global energy features, and performing attention weighting on the local energy anomalous indicator features and global energy anomalous indicator features respectively; and a local-global dual-branch fusion module for concatenating and fusing the local energy anomalous indicator features and the global energy anomalous indicator features before outputting the result.

[0057] On the other hand, this application also provides an intelligent detection device for an energy anomaly channel, comprising: a memory; a processor; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement any of the methods mentioned above.

[0058] In another aspect, this application also provides a computer-readable storage medium having a computer program stored thereon; the computer program is executed by a processor to implement the methods mentioned in any of the above.

[0059] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as C, VHDL, Verilog, the object-oriented programming language Java, and the interpreted scripting language JavaScript.

[0060] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0061] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0062] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0063] In the description of this application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0064] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0065] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection, an electrical connection, or a connection that allows communication between them; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication between two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0066] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0067] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for intelligent detection of energy anomalies, characterized in that, Includes the following steps: Obtain node gather data, reorder each trace data based on the distance between each shot point and the seismic receiving node in the node gather data, and perform leveling processing on the node gather data with the first arrival line as the reference to complete the preprocessing of the node gather data. The preprocessed node gather data is divided into multiple data units with a preset size, and the energy nonlinear quantization classification data of each data unit is calculated to generate local energy index data. The preprocessed node gather data is used to calculate the overall energy nonlinearity quantization and hierarchical data of the entire gather on a gather-by-trace basis, in order to generate global energy index data. Energy sample label data is generated from the preprocessed node gather data, with each channel as the unit. The local and global energy index data and the energy sample label data are used as training models for multi-class deep networks to process the input data, so as to achieve multi-class detection of energy anomalies. Based on the output of the multi-class deep network training model, the type that appears most frequently in each channel is counted as the predicted type of the corresponding channel, and the classification detection results of each channel data in the node channel set data are generated.

2. The intelligent detection method for energy anomaly channels according to claim 1, characterized in that, The steps for calculating the nonlinear quantization classification data of energy for each data unit include calculating the nonlinear quantization classification index of the mean amplitude energy, maximum amplitude energy, minimum amplitude energy, and median amplitude energy for each data unit.

3. The intelligent detection method for energy anomaly channels according to claim 1, characterized in that, The steps for calculating the nonlinear quantization classification data of the entire gather include calculating the nonlinear quantization classification index of the mean amplitude energy, maximum amplitude energy, minimum amplitude energy, and median amplitude energy on a per-trace basis.

4. The intelligent detection method for energy anomaly channels according to claim 1, characterized in that, The multi-class deep network training model is a dual-input branch encoder-decoder network model, which is used to perform the following steps. Extract nonlinear quantized hierarchical features containing local energy; Extracting the nonlinear quantization hierarchical characteristics of the whole local energy of the package; Extract anomalous indicator features from local and global energy features, and then apply attention weighting to the local and global energy anomalous indicator features respectively; The local energy anomaly index features and the global energy anomaly index features are spliced ​​and fused together before being output.

5. The intelligent detection method for energy anomaly channels according to claim 1, characterized in that, The multi-class deep network training model is a multi-class loss function model, and the multi-class loss function model is as follows: in, For the total loss error, C represents the weighting factor for the classification results, where C is the total number of categories. Error loss function oriented towards classification results, The ratio at which the model correctly predicts abnormal channels as abnormal and normal channels as normal. This refers to the percentage of normal channels that the model predicts as abnormal channels. The ratio of energy anomalies to normal channels; The penalty factor for predicting a normal channel as an abnormal channel. The penalty factor for predicting an abnormal channel as a normal channel; This is a multi-class cross-entropy calculation method for energy anomalies. For the nth point to be of class c, this is the actual situation. Let n be the probability that the nth point belongs to the cth class. The ratio of normal lanes predicted to be normal lanes.

6. The intelligent detection method for energy anomaly channels according to claim 1, characterized in that, The energy sample label data includes ultra-high energy channels, ultra-low energy channels, normal energy channels, and 0-value channels.

7. An intelligent detection device for an energy anomaly channel, characterized in that, The intelligent detection method for performing the energy anomaly channel as described in any one of claims 1-5 includes: The preprocessing module is used to acquire node gather data, reorder each trace data based on the distance between each shot point and the seismic receiving node in the node gather data, and perform leveling processing on the node gather data based on the first arrival line to complete the preprocessing of the node gather data. The local energy index calculation unit is used to divide the preprocessed node gather data into multiple data units with a preset size, calculate the energy nonlinear quantization and classification data of each data unit, and generate local energy index data. The global energy index calculation unit is used to calculate the overall energy nonlinear quantization and classification data of the entire gather on a gather-by-trace basis after preprocessing the node gather data, so as to generate global energy index data. The sample label generation unit is used to generate energy sample label data in units of channels from the preprocessed node gather data. A multi-class deep network training model is used to process the local and global energy index data and the energy sample label data as input data to achieve multi-class detection of energy anomalies. The classification and detection unit is used to calculate the most frequent type of each channel based on the output of the multi-class deep network training model, and generate the classification and detection results of each channel data in the node channel set data.

8. The intelligent detection device for energy anomaly channels according to claim 7, characterized in that, The multi-class deep network training model is a dual-input branch encoder-decoder network model, which includes... A multi-scale preprocessing module is used to extract nonlinear quantized hierarchical features containing local energies; Encoding network used to extract nonlinear quantized hierarchical features of packet whole local energy; The multi-channel data attention enhancement module is used to extract abnormal indicator features from local energy features and global energy features, and to perform attention weighting on the local energy abnormal indicator features and global energy abnormal indicator features respectively; The local-global dual-branch fusion module is used to splice and fuse the local energy anomaly index features and the global energy anomaly index features before outputting the result.

9. An intelligent detection device for an energy anomaly channel, characterized in that, include: Memory; processor; as well as Computer programs; The computer program is stored in the memory and configured to be executed by the processor to implement the intelligent detection method for energy anomalies as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, It stores a computer program; the computer program is executed by a processor to implement the intelligent detection method for energy anomalies as described in any one of claims 1-6.

Images