Exploration electronic measurement data analysis system and method based on cloud computing

Abstract

The invention relates to the technical field of exploration, and discloses an exploration electronic measurement data analysis system and method based on cloud computing, and the system comprises a preprocessing module which is used for collecting multi-source exploration electronic measurement data, carrying out the unified format conversion and data index establishment of different types of data through a cloud computing environment, carrying out the denoising, and separating effective frequency band sub-bands; pre-processed electronic measurement data are obtained; the classification module is used for inputting the preprocessed electronic measurement data into the CNN-LSTM hybrid model, dynamically adjusting a classification threshold according to signal distribution by adopting an adaptive particle swarm optimization algorithm, and outputting a classification result; the dynamic allocation module is used for constructing a PPO-DRL model and dynamically allocating computing resources; the result generation module is used for analyzing the classified data based on the classification result output by the CNN-LSTM hybrid model, generating an exploration result and presenting the exploration result in a visual mode; the overall analysis efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of exploration technology, and specifically to a cloud computing-based electronic survey data analysis system and method for exploration. Background Technology

[0002] In the exploration of resources such as oil and gas and minerals, the analysis of exploration electronic survey data is a key link in obtaining information on underground geological structures. Existing exploration electronic survey data analysis systems mainly rely on cloud platforms to centrally process electronic survey data generated at the exploration site. In oil and gas exploration, when simultaneously acquiring magnetotelluric signals and seismic reflection signals, existing systems usually use a uniform preprocessing and filtering framework, ignoring the essential differences between the two types of signals in terms of propagation mechanisms and frequency response characteristics. For example, if the same high-pass filter is used to remove low-frequency noise, it may excessively attenuate the effective low-frequency electromagnetic components. Since the low-frequency components in seismic signals are crucial to the resolution of tomographic imaging, this can lead to systematic deviations in the inversion results. Existing parallel computing architectures often distribute computing resources evenly across various types of signals. For example, when processing high-frequency sound signals and low-frequency gravity signals, the system still uses the same slice size and number of iterations, resulting in insufficient analysis of key signals and excessive computation of secondary signals, leading to low overall energy efficiency. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing a cloud computing-based exploration electronic measurement data analysis system and method.

[0004] This invention provides a cloud computing-based exploration electronic survey data analysis system, the system comprising: The preprocessing module is used to collect multi-source electronic measurement data, including electromagnetic wave signals, seismic wave signals, and resistivity signals, through exploration equipment. It performs unified format conversion and data indexing on different types of data through a cloud computing environment, and uses the WPT decomposition strategy to denoise different signals and separate effective frequency band sub-bands to obtain preprocessed electronic measurement data. The classification module is used to input the preprocessed electronic measurement data into the CNN-LSTM hybrid model, and to dynamically adjust the classification threshold according to the signal distribution using the adaptive particle swarm optimization algorithm, to extract features and classify the preprocessed electronic measurement data, and output the classification results. The dynamic allocation module is used to build the PPO-DRL model, which uses signal type, data volume, and node load as states and resource allocation strategy as actions to dynamically allocate computing resources. The results generation module is used to analyze the classified data based on the classification results output by the CNN-LSTM hybrid model, generate exploration results, and present them in a visualization manner.

[0005] Optionally, in a first implementation of the present invention, the preprocessing module includes: The transmission submodule is used to collect multi-source electronic measurement data, including electromagnetic wave signals, seismic wave signals, and resistivity signals, through exploration equipment, store them in their original format, and transmit them to the distributed storage nodes in the cloud computing environment. The format conversion submodule is used to start the format conversion engine in the cloud computing environment, define mapping rules for different original format data, and perform unified format conversion for different types of data; The index building submodule is used to extract metadata based on the converted unified format data, build a multi-level index using the B+ tree indexing algorithm, and store the index information on the index server in the cloud computing environment. The metadata includes the collection time, latitude and longitude of the measurement point, signal type, and data volume. The wavelet packet decomposition submodule is used to configure wavelet packet decomposition parameters for different signal types after unification. Electromagnetic wave signals are decomposed using the db6 wavelet basis with 5 layers, seismic wave signals are decomposed using the sym8 wavelet basis with 3 layers, and resistivity signals are decomposed using the db4 wavelet basis with 2 layers. The filtering submodule is used to perform threshold denoising on each frequency band obtained by decomposition, filter effective subbands according to signal characteristics, and output preprocessed electronic measurement data. The effective subbands are the low-frequency subbands of 0.001–62.5Hz for electromagnetic wave signals, the main frequency subbands of 12.5–100Hz for seismic wave signals, and the subbands containing 50 or 60Hz power frequency interference for resistivity signals.

[0006] Optionally, in a second implementation of the present invention, the format conversion submodule includes: Convert SEG-Y format seismic wave signals into JSON format containing trace head information and sampling point values; Convert EDIS format electromagnetic wave signals into JSON format containing frequency, amplitude, and phase; Convert the resistivity signal in CSV format to JSON format, which includes the coordinates of the measurement point and the resistivity value.

[0007] Optionally, in a third implementation of the present invention, the index construction submodule includes: The primary index is divided according to the time interval of data collection, the secondary index is divided according to the geographical area of ​​the measurement point, and the tertiary index is divided according to the signal type.

[0008] Optionally, in a fourth implementation of the present invention, the classification module includes: The first convolutional submodule is used to input the preprocessed electronic measurement data into the CNN-LSTM hybrid model. The first convolutional layer uses 32 3×3 convolutional kernels to perform sliding convolution and extracts local frequency domain features through the ReLU activation function. The second convolutional submodule is used in the second convolutional layer to deepen feature extraction with 64 3×3 convolutional kernels, enhancing the capture of signal detail patterns; The capture submodule is used to perform 2×2 max pooling on the feature map output by the convolutional layer, and input the pooled features into an LSTM layer with 128 hidden units to learn the dynamic changes of the signal in the time dimension and capture the temporal dependencies. The weight allocation submodule is used to introduce the Bahdanau attention mechanism to assign weights to the temporal features output by the LSTM layer, highlighting the feature contribution of time nodes. The classification submodule is used to output classification results and corresponding confidence scores for electromagnetic signals, seismic signals, and resistivity signals through a fully connected layer and a softmax classifier.

[0009] Optionally, in a fifth implementation of the present invention, the classification module further includes: The initialization submodule is used to set 50 particles as threshold candidate solutions. Each particle contains 3 dimensions and the position and velocity of the particles are randomly initialized. The first calculation submodule is used to apply the threshold of the current particle to the classification result with the classification error rate as the fitness function, and to count and calculate the fitness value of each particle. The update submodule is used for each particle to compare its current fitness value with the historical best value. If the current fitness value is better, the individual best position is updated. The group compares the fitness values ​​of all particles and updates the global best position. The iterative submodule is used to update velocity and position based on the individual optimal and global optimal positions, guiding the particles to move towards a better solution. It stops after 50 iterations and uses the threshold corresponding to the global optimal position as the final classification threshold, which is used to determine the confidence level for real-time signal classification.

[0010] Optionally, in a sixth implementation of the present invention, the dynamic allocation module includes: The integration submodule is used to integrate signal type, data volume, real-time load rate of computing nodes, and signal importance weight into a state vector, which serves as the input to the PPO-DRL model. The sub-modules are used to divide the action into three dimensions: computing node selection, CPU core allocation, and memory allocation. The reward value of the reward function is calculated by weighting the load bias rate, latency rate, and classification error rate. The second computational submodule is used to calculate the action advantage value using generalized advantage estimation and update the policy network and value network through the reward function. The output submodule is used to input the state vector of the signal to be processed into the PPO-DRL model and output the optimal action. The cloud computing environment schedules resources according to the plan to realize dynamic allocation of computing resources.

[0011] Optionally, in a seventh implementation of the present invention, the result generation module includes: The distribution submodule is used to distribute the classification results output by the CNN-LSTM hybrid model to the corresponding computing nodes according to the category through the task scheduler in the cloud computing environment. The generation submodule is used by electromagnetic signal nodes to calculate resistivity values ​​at different depths underground through inversion algorithms, seismic signal nodes to construct seismic wave velocity models of underground media, and resistivity signal nodes to identify strata lithology based on resistivity differences, generating preliminary analysis results. The summary submodule is used to summarize the preliminary results of each node, perform spatial coordinate matching based on the GPS information of the exploration points, and generate a multi-dimensional exploration report. The conversion submodule is used to convert the aggregated multi-dimensional exploration report into a WebGL-compatible format, store it as mesh data in depth layers, and convert attribute information into texture maps. The display submodule is used to load the transformed data on a 3D platform built with WebGL and display signal classification logs and resource allocation records to visualize the exploration results.

[0012] Optionally, in the eighth implementation of the present invention, the electromagnetic signal is distributed to the resistivity inversion node, the seismic signal is distributed to the velocity modeling node, and the resistivity signal is distributed to the lithology identification node.

[0013] Optionally, in the ninth implementation of the present invention, a method for implementing a cloud computing-based exploration electronic survey data analysis system includes the following steps: Multi-source electronic measurement data, including electromagnetic wave signals, seismic wave signals, and resistivity signals, are collected by exploration equipment. Different types of data are converted into a unified format and indexed using a cloud computing environment. The WPT decomposition strategy is used to denoise different signals and separate effective frequency band sub-bands to obtain preprocessed electronic measurement data. The preprocessed electronic measurement data is input into the CNN-LSTM hybrid model, and the adaptive particle swarm optimization algorithm is used to dynamically adjust the classification threshold according to the signal distribution. The preprocessed electronic measurement data is then used for feature extraction and classification, and the classification results are output. Construct a PPO-DRL model, with signal type, data volume, and node load as states and resource allocation strategy as actions, to dynamically allocate computing resources; The classification results are analyzed based on the output of the CNN-LSTM hybrid model to generate exploration results, which are then presented in a visualization manner.

[0014] The technical solution provided by this invention involves collecting multi-source electronic measurement data, including electromagnetic wave signals, seismic wave signals, and resistivity signals, using exploration equipment. A cloud computing environment is used to perform unified format conversion and data indexing for different data types. A WPT decomposition strategy is employed to denoise different signals and separate effective frequency band sub-bands, resulting in preprocessed electronic measurement data. This preprocessed electronic measurement data is then input into a CNN-LSTM hybrid model, and an adaptive particle swarm optimization algorithm is used to dynamically adjust the classification threshold based on signal distribution. Feature extraction and classification are performed on the preprocessed electronic measurement data, and the classification results are output. Finally, a PPO-DRL model is constructed, taking into account signal type, data volume, and... Node load is the state, resource allocation strategy is the action, and computing resources are dynamically allocated. The classification results output by the CNN-LSTM hybrid model are used to analyze the classified data, generate exploration results, and present them in a visualization manner. This invention can effectively distinguish different types of exploration signals through the CNN-LSTM hybrid model, avoiding the signal aliasing problem in traditional methods, improving classification accuracy. The adaptive computing resource scheduling algorithm ensures that key signals obtain sufficient computing resources, improving overall analysis efficiency, reducing system energy consumption, and shortening data processing time through cloud computing, making exploration analysis more timely and efficient, and significantly improving the accuracy and reliability of exploration results. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 A schematic diagram of the structure of a cloud-based electronic survey data analysis system for exploration provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the preprocessing module provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of the dynamic allocation module provided in an embodiment of the present invention. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 A schematic diagram of the structure of the cloud-based exploration electronic survey data analysis system provided in this embodiment of the invention. The system includes: The preprocessing module is used to collect multi-source electronic measurement data, including electromagnetic wave signals, seismic wave signals, and resistivity signals, through exploration equipment. It performs unified format conversion and data indexing on different types of data through a cloud computing environment, and uses the WPT decomposition strategy to denoise different signals and separate effective frequency band sub-bands to obtain preprocessed electronic measurement data. The classification module is used to input the preprocessed electronic measurement data into the CNN-LSTM hybrid model, and to dynamically adjust the classification threshold according to the signal distribution using the adaptive particle swarm optimization algorithm, to extract features and classify the preprocessed electronic measurement data, and output the classification results. The dynamic allocation module is used to build the PPO-DRL model, which uses signal type, data volume, and node load as states and resource allocation strategy as actions to dynamically allocate computing resources. The results generation module is used to analyze the classified data based on the classification results output by the CNN-LSTM hybrid model, generate exploration results, and present them in a visualization manner.

[0019] In this embodiment, please refer to Figure 2 The preprocessing module includes: The transmission submodule is used to collect multi-source electronic measurement data, including electromagnetic wave signals, seismic wave signals, and resistivity signals, through exploration equipment, store them in their original format, and transmit them to the distributed storage nodes in the cloud computing environment. The format conversion submodule is used to start the format conversion engine in the cloud computing environment, define mapping rules for different original format data, and perform unified format conversion for different types of data; The index building submodule is used to extract metadata based on the converted unified format data, build a multi-level index using the B+ tree indexing algorithm, and store the index information on the index server in the cloud computing environment. The metadata includes the collection time, latitude and longitude of the measurement point, signal type, and data volume. The wavelet packet decomposition submodule is used to configure wavelet packet decomposition parameters for different signal types after unification. Electromagnetic wave signals are decomposed using the db6 wavelet basis with 5 layers, seismic wave signals are decomposed using the sym8 wavelet basis with 3 layers, and resistivity signals are decomposed using the db4 wavelet basis with 2 layers. The filtering submodule is used to perform threshold denoising on each frequency band obtained by decomposition, filter effective subbands according to signal characteristics, and output preprocessed electronic measurement data. The effective subbands are the low-frequency subbands of 0.001–62.5Hz for electromagnetic wave signals, the main frequency subbands of 12.5–100Hz for seismic wave signals, and the subbands containing 50 or 60Hz power frequency interference for resistivity signals.

[0020] In this embodiment, the format conversion submodule includes: converting SEG-Y format seismic wave signals into JSON format containing trace head information and sampling point values; converting EDIS format electromagnetic wave signals into JSON format containing frequency, amplitude, and phase; and converting CSV format resistivity signals into JSON format containing measurement point coordinates and resistivity values.

[0021] In this embodiment, the index construction submodule includes: a first-level index divided by the collection time interval, a second-level index divided by the geographical area of ​​the measurement point, and a third-level index divided by the signal type.

[0022] In this embodiment, the classification module includes: The first convolutional submodule is used to input the preprocessed electronic measurement data into the CNN-LSTM hybrid model. The first convolutional layer uses 32 3×3 convolutional kernels to perform sliding convolution and extracts local frequency domain features through the ReLU activation function. The second convolutional submodule is used in the second convolutional layer to deepen feature extraction with 64 3×3 convolutional kernels, enhancing the capture of signal detail patterns; The capture submodule is used to perform 2×2 max pooling on the feature map output by the convolutional layer, and input the pooled features into an LSTM layer with 128 hidden units to learn the dynamic changes of the signal in the time dimension and capture the temporal dependencies. The weight allocation submodule is used to introduce the Bahdanau attention mechanism to assign weights to the temporal features output by the LSTM layer, highlighting the feature contribution of time nodes. The classification submodule is used to output classification results and corresponding confidence scores for electromagnetic signals, seismic signals, and resistivity signals through a fully connected layer and a softmax classifier.

[0023] In this embodiment, the classification module further includes: The initialization submodule is used to set 50 particles as threshold candidate solutions. Each particle contains 3 dimensions and the position and velocity of the particles are randomly initialized. The first calculation submodule is used to apply the threshold of the current particle to the classification result with the classification error rate as the fitness function, and to count and calculate the fitness value of each particle. The update submodule is used for each particle to compare its current fitness value with the historical best value. If the current fitness value is better, the individual best position is updated. The group compares the fitness values ​​of all particles and updates the global best position. The iterative submodule is used to update velocity and position based on the individual optimal and global optimal positions, guiding the particles to move towards a better solution. It stops after 50 iterations and uses the threshold corresponding to the global optimal position as the final classification threshold, which is used to determine the confidence level for real-time signal classification.

[0024] In this embodiment, the core parameter configuration of the particle swarm is clearly defined: 50 particles are set as candidate solutions for the classification threshold. The three dimensions of each particle correspond to the classification thresholds of electromagnetic wave signals, seismic wave signals, and resistivity signals, respectively. That is, the value of each dimension represents the confidence judgment standard of the corresponding signal. Then, the position of each particle is randomly initialized, and the position value is limited to the range of [0.5, 1.0]. At the same time, the velocity of each particle is randomly initialized, and the velocity value is limited to the range of [-0.1, 0.1]. Finally, an initial particle swarm containing 50 three-dimensional particles is formed. Using the classification error rate as the sole fitness function, the signal classification results output by the CNN-LSTM hybrid model are first obtained from the classification module. Then, the thresholds corresponding to the three dimensions of the current particle are applied to the confidence judgment of the three types of signals respectively: if the confidence of a signal is higher than the threshold of the corresponding dimension, it is judged as a correct classification; if it is lower than the threshold, it is judged as a misclassification. Subsequently, the number of misclassified samples under the threshold of the particle is counted to the total number of samples, and the ratio of the two is calculated. This ratio is used as the fitness value of the current particle. After calculating the fitness values ​​of all 50 particles, the fitness dataset of the particle swarm is formed. The optimization is performed in the order of individual update → swarm update: For each particle, its historical best fitness value is retrieved, initially the first fitness value after the particle is initialized. The current calculated fitness value is compared with the historical best value. If the current fitness value is smaller, that is, the classification error rate is lower, the current position of the particle is updated to the new individual best position, and its historical best fitness value is updated simultaneously. After all particles have completed the individual best update, the current fitness values ​​of all particles in the entire particle swarm are collected, the position of the particle corresponding to the minimum fitness value is selected, and the position is updated to the new global best position, ensuring that the particle swarm always converges in the direction of lower classification error rate. First, the particle velocity is updated based on the particle swarm optimization rules: inertia weight, cognitive factor, and social factor are introduced, and the new velocity of each particle is calculated by combining the current velocity of the particle, the difference between the individual optimal position and the current position, and the difference between the global optimal position and the current position, so as to avoid the velocity being too large or too small affecting the convergence efficiency. Then, the new position of each particle is calculated based on the new velocity, ensuring that the new position is still within a reasonable threshold range of [0.5, 1.0]. Subsequently, the first calculation submodule is called to recalculate the fitness value of all particles, and the update submodule is called to update the individual optimal and global optimal positions. This process is repeated iteratively until 50 iterations are completed and then the operation stops. Finally, the global optimal position at the end of the iteration is extracted. The three dimensions of this position are used as the final classification thresholds for electromagnetic waves, seismic waves, and resistivity signals, respectively. In the real-time signal classification process, the signal type is determined by comparing the signal confidence with the corresponding final classification threshold, and the confidence verification is completed.

[0025] In this embodiment, please refer to Figure 3 The dynamic allocation module includes: The integration submodule is used to integrate signal type, data volume, real-time load rate of computing nodes, and signal importance weight into a state vector, which serves as the input to the PPO-DRL model. The sub-modules are used to divide the action into three dimensions: computing node selection, CPU core allocation, and memory allocation. The reward value of the reward function is calculated by weighting the load bias rate, latency rate, and classification error rate. The second computational submodule is used to calculate the action advantage value using generalized advantage estimation and update the policy network and value network through the reward function. The output submodule is used to input the state vector of the signal to be processed into the PPO-DRL model and output the optimal action. The cloud computing environment schedules resources according to the plan to realize dynamic allocation of computing resources.

[0026] In this embodiment, the signal type is encoded according to preset rules, such as electromagnetic signal as 1, seismic signal as 2, and resistivity signal as 3. The data volume is converted into a unified unit and normalized and mapped to the [0,1] interval. The load percentage of each computing node is collected in real time through the monitoring interface of the cloud computing environment. The signal weights are preset according to the importance of the exploration target, such as resistivity 0.4, seismic 0.35, and electromagnetic 0.25. Then, the processed signal type encoding, normalized data volume, node load rate, and signal weight are arranged in sequence and combined into a fixed-dimensional state vector, such as a four-dimensional vector. The vector is then standardized so that the mean of each dimension is 0 and the variance is 1. Finally, this state vector is used as the input data of the PPO-DRL model. The computing node selection dimension includes all available computing nodes in the cloud, such as 5 nodes numbered 1 to 5, from which the model can select one. The CPU core allocation dimension sets discrete optional values ​​to cover different processing needs. The memory allocation dimension also sets discrete optional values ​​to match the data volume. Simultaneously, this submodule defines the calculation method for the reward function: the load deviation rate is 1 minus the absolute deviation ratio between the actual load and the target load of the computing node; the latency rate is 1 minus the absolute deviation ratio between the actual processing time and the theoretical processing time; and the classification error rate is 1 minus the ratio of the number of misclassified samples to the total number of samples. Finally, the above three indicators are weighted and summed in a 6:3:1 ratio to obtain the reward value that measures the quality of the action, providing feedback signals for model training. The trajectory data generated by the interaction between the PPO-DRL model and the environment includes state, action, reward, and next state. A generalized advantage estimation algorithm is used to calculate the action advantage value: based on the reward value of the current state and the estimated value of the next state, a decay coefficient λ (e.g., 0.95) is introduced to weight and sum the advantages of multiple steps, obtaining an advantage estimate of each action relative to the average level. Subsequently, this advantage value is used to update the policy network: the truncation objective function of PPO limits the policy update magnitude to ensure training stability, keeping the probability ratio of the new policy to the old policy within a reasonable range. Simultaneously, the value network is updated based on the mean squared error loss function, making the value network's estimation of state value more accurate. By alternately updating the policy network and the value network, the model's decision-making ability is continuously optimized. After receiving the relevant information of the signal to be processed, the integration submodule is invoked to generate the corresponding state vector, which is then input into the trained and converged PPO-DRL model. The model processes the state vector through the policy network, outputs the probability distribution of each action, and selects the action combination with the highest probability as the optimal action, including the specific computing node number, number of CPU cores, and memory size. Subsequently, the optimal action is converted into a resource scheduling instruction executable in the cloud computing environment. The cloud resource manager allocates the specified number of CPU cores and memory to the selected computing node and transmits the data of the signal to be processed to that node. The resource manager monitors the node resource usage in real time to ensure accurate execution of the scheduling instructions, ultimately realizing the dynamic allocation of computing resources to meet the processing needs of different signals.

[0027] In this embodiment, the result generation module includes: The distribution submodule is used to distribute the classification results output by the CNN-LSTM hybrid model to the corresponding computing nodes according to the category through the task scheduler in the cloud computing environment. Electromagnetic signals are distributed to the resistivity inversion node, seismic signals are distributed to the velocity modeling node, and resistivity signals are distributed to the lithology identification node. The generation submodule is used by electromagnetic signal nodes to calculate resistivity values ​​at different depths underground through inversion algorithms, seismic signal nodes to construct seismic wave velocity models of underground media, and resistivity signal nodes to identify strata lithology based on resistivity differences, generating preliminary analysis results. The summary submodule is used to summarize the preliminary results of each node, perform spatial coordinate matching based on the GPS information of the exploration points, and generate a multi-dimensional exploration report. The conversion submodule is used to convert the aggregated multi-dimensional exploration report into a WebGL-compatible format, store it as mesh data in depth layers, and convert attribute information into texture maps. The display submodule is used to load the transformed data on a 3D platform built with WebGL and display signal classification logs and resource allocation records to visualize the exploration results.

[0028] In this embodiment, addressing the challenges of classification and aggregation in cross-domain parallel statistics of multidimensional exploration signals, existing methods typically rely on fixed thresholds or manual intervention for signal differentiation. These methods struggle to handle issues such as signal aliasing, frequency band overlap, and strong time-varying characteristics in the exploration environment, leading to large statistical biases and significant cross-domain interference. To address this, this paper proposes an adaptive classification statistical model based on the physical hierarchy of signals. This model achieves hierarchical classification based on the physical properties of different signal sources, such as electromagnetic waves, seismic waves, and resistivity, fundamentally overcoming the statistical bias caused by signal mixing in traditional methods. During classification, the system intelligently allocates various signals to dedicated statistical modules based on high-dimensional parameters such as the original characteristics, frequency band distribution, and temporal characteristics of the signals. The adaptive mechanism dynamically optimizes the classification threshold based on real-time signal distribution and automatically schedules data resources in the parallel architecture, ensuring accurate and independent statistical processing for each type of signal. This method significantly improves cross-domain computational efficiency and effectively suppresses mutual interference between signal types, thereby greatly enhancing the accuracy and reliability of exploration data processing.

[0029] In this embodiment, a hierarchical adaptive signal classification mechanism is constructed to address the fundamental differences in physical properties, frequency bands, and temporal characteristics of different types of signals, such as electromagnetic waves, seismic waves, and resistivity signals. This effectively solves the problem of feature confusion and error accumulation caused by mixed signal processing in traditional methods. The model achieves refined classification based on the physical hierarchy of signals, ensuring that each type of signal is processed in an independent module. This suppresses mutual interference between multimodal signals at the source and significantly improves the accuracy of data statistics. In a cross-domain parallel computing architecture, the adaptive classification mechanism can dynamically optimize thresholds based on real-time signal distribution and automatically schedule computing resources to achieve differentiated processing for different signal types. For example, appropriate computing nodes and iteration strategies are allocated to high-frequency seismic signals and low-frequency electromagnetic signals respectively, avoiding redundant calculations and improving resource utilization efficiency. This system not only significantly accelerates the processing speed of large-scale exploration data but also improves the accuracy and reliability of inversion and interpretation, providing key technical support for high-precision data analysis in complex geophysical exploration environments.

[0030] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A cloud computing-based exploration electronic survey data analysis system, characterized in that, The system includes: The preprocessing module is used to collect multi-source electronic measurement data, including electromagnetic wave signals, seismic wave signals, and resistivity signals, through exploration equipment. It performs unified format conversion and data indexing on different types of data through a cloud computing environment, and uses the WPT decomposition strategy to denoise different signals and separate effective frequency band sub-bands to obtain preprocessed electronic measurement data. The classification module is used to input the preprocessed electronic measurement data into the CNN-LSTM hybrid model, and to dynamically adjust the classification threshold according to the signal distribution using the adaptive particle swarm optimization algorithm, to extract features and classify the preprocessed electronic measurement data, and output the classification results. The dynamic allocation module is used to build the PPO-DRL model, which uses signal type, data volume, and node load as states and resource allocation strategy as actions to dynamically allocate computing resources. The results generation module is used to analyze the classified data based on the classification results output by the CNN-LSTM hybrid model, generate exploration results, and present them in a visualization manner.

2. The cloud computing-based exploration electronic survey data analysis system as described in claim 1, characterized in that, The preprocessing module includes: The transmission submodule is used to collect multi-source electronic measurement data, including electromagnetic wave signals, seismic wave signals, and resistivity signals, through exploration equipment, store them in their original format, and transmit them to the distributed storage nodes in the cloud computing environment. The format conversion submodule is used to start the format conversion engine in the cloud computing environment, define mapping rules for different original format data, and perform unified format conversion for different types of data; The index building submodule is used to extract metadata based on the converted unified format data, build a multi-level index using the B+ tree indexing algorithm, and store the index information on the index server in the cloud computing environment. The metadata includes the collection time, latitude and longitude of the measurement point, signal type, and data volume. The wavelet packet decomposition submodule is used to configure wavelet packet decomposition parameters for different signal types after unification. Electromagnetic wave signals are decomposed using the db6 wavelet basis with 5 layers, seismic wave signals are decomposed using the sym8 wavelet basis with 3 layers, and resistivity signals are decomposed using the db4 wavelet basis with 2 layers. The filtering submodule is used to perform threshold denoising on each frequency band obtained by decomposition, filter effective subbands according to signal characteristics, and output preprocessed electronic measurement data. The effective subbands are the low-frequency subbands of 0.001–62.5Hz for electromagnetic wave signals, the main frequency subbands of 12.5–100Hz for seismic wave signals, and the subbands containing 50 or 60Hz power frequency interference for resistivity signals.

3. The cloud computing-based exploration electronic survey data analysis system as described in claim 2, characterized in that, The format conversion submodule includes: Convert SEG-Y format seismic wave signals into JSON format containing trace head information and sampling point values; Convert EDIS format electromagnetic wave signals into JSON format containing frequency, amplitude, and phase; Convert the resistivity signal in CSV format to JSON format, which includes the coordinates of the measurement point and the resistivity value.

4. The cloud computing-based exploration electronic survey data analysis system as described in claim 2, characterized in that, The index building submodule includes: The primary index is divided according to the time interval of data collection, the secondary index is divided according to the geographical area of ​​the measurement point, and the tertiary index is divided according to the signal type.

5. The cloud computing-based exploration electronic survey data analysis system as described in claim 1, characterized in that, The classification module includes: The first convolutional submodule is used to input the preprocessed electronic measurement data into the CNN-LSTM hybrid model. The first convolutional layer uses 32 3×3 convolutional kernels to perform sliding convolution and extracts local frequency domain features through the ReLU activation function. The second convolutional submodule is used in the second convolutional layer to deepen feature extraction with 64 3×3 convolutional kernels, enhancing the capture of signal detail patterns; The capture submodule is used to perform 2×2 max pooling on the feature map output by the convolutional layer, and input the pooled features into an LSTM layer with 128 hidden units to learn the dynamic changes of the signal in the time dimension and capture the temporal dependencies. The weight allocation submodule is used to introduce the Bahdanau attention mechanism to assign weights to the temporal features output by the LSTM layer, highlighting the feature contribution of time nodes. The classification submodule is used to output classification results and corresponding confidence scores for electromagnetic signals, seismic signals, and resistivity signals through a fully connected layer and a softmax classifier.

6. The exploration electronic survey data analysis system based on cloud computing as described in claim 1, characterized in that, The classification module also includes: The initialization submodule is used to set 50 particles as threshold candidate solutions. Each particle contains 3 dimensions and the position and velocity of the particles are randomly initialized. The first calculation submodule is used to apply the threshold of the current particle to the classification result with the classification error rate as the fitness function, and to count and calculate the fitness value of each particle. The update submodule is used for each particle to compare its current fitness value with the historical best value. If the current fitness value is better, the individual best position is updated. The group compares the fitness values ​​of all particles and updates the global best position. The iterative submodule is used to update velocity and position based on the individual optimal and global optimal positions, guiding the particles to move towards a better solution. It stops after 50 iterations and uses the threshold corresponding to the global optimal position as the final classification threshold, which is used to determine the confidence level for real-time signal classification.

7. The cloud computing-based exploration electronic survey data analysis system as described in claim 1, characterized in that, The dynamic allocation module includes: The integration submodule is used to integrate signal type, data volume, real-time load rate of computing nodes, and signal importance weight into a state vector, which serves as the input to the PPO-DRL model. The sub-modules are used to divide the action into three dimensions: computing node selection, CPU core allocation, and memory allocation. The reward value of the reward function is calculated by weighting the load bias rate, latency rate, and classification error rate. The second computational submodule is used to calculate the action advantage value using generalized advantage estimation and update the policy network and value network through the reward function. The output submodule is used to input the state vector of the signal to be processed into the PPO-DRL model and output the optimal action. The cloud computing environment schedules resources according to the plan to realize dynamic allocation of computing resources.

8. The cloud computing-based exploration electronic survey data analysis system as described in claim 1, characterized in that, The result generation module includes: The distribution submodule is used to distribute the classification results output by the CNN-LSTM hybrid model to the corresponding computing nodes according to the category through the task scheduler in the cloud computing environment. The generation submodule is used by electromagnetic signal nodes to calculate resistivity values ​​at different depths underground through inversion algorithms, seismic signal nodes to construct seismic wave velocity models of underground media, and resistivity signal nodes to identify strata lithology based on resistivity differences, generating preliminary analysis results. The summary submodule is used to summarize the preliminary results of each node, perform spatial coordinate matching based on the GPS information of the exploration points, and generate a multi-dimensional exploration report. The conversion submodule is used to convert the aggregated multi-dimensional exploration report into a WebGL-compatible format, store it as mesh data in depth layers, and convert attribute information into texture maps. The display submodule is used to load the transformed data on a 3D platform built with WebGL and display signal classification logs and resource allocation records to visualize the exploration results.

9. The cloud computing-based electronic survey data analysis system as described in claim 8, characterized in that, Electromagnetic signals are distributed to the resistivity inversion node, seismic signals are distributed to the velocity modeling node, and resistivity signals are distributed to the lithology identification node.

10. A method for implementing a cloud computing-based exploration electronic survey data analysis system as described in claim 1, characterized in that, The method includes the following steps: Multi-source electronic measurement data, including electromagnetic wave signals, seismic wave signals, and resistivity signals, are collected by exploration equipment. Different types of data are converted into a unified format and indexed using a cloud computing environment. The WPT decomposition strategy is used to denoise different signals and separate effective frequency band sub-bands to obtain preprocessed electronic measurement data. The preprocessed electronic measurement data is input into the CNN-LSTM hybrid model, and the adaptive particle swarm optimization algorithm is used to dynamically adjust the classification threshold according to the signal distribution. The preprocessed electronic measurement data is then used for feature extraction and classification, and the classification results are output. Construct a PPO-DRL model, with signal type, data volume, and node load as states and resource allocation strategy as actions, to dynamically allocate computing resources; The classification results are analyzed based on the output of the CNN-LSTM hybrid model to generate exploration results, which are then presented in a visualization manner.

Images