Method for constructing three-dimensional geophysical field in open well-to-mining area

Abstract

The invention relates to the field of mining area three-dimensional geologic modeling, and discloses an open well-to-mining area three-dimensional geophysical field construction method, which comprises the following steps of: 1, acquiring multi-physical field data, and acquiring seismic data, gravity data, magnetic data, electrical method data and remote sensing data of a mining area; 2, data preprocessing: carrying out normalization, multi-scale decomposition, noise processing and feature extraction on the multi-physical field data; step 3, constructing a multi-modal deep fusion model, and carrying out adaptive attention weight distribution on the features of different physical field data to realize weighted fusion of fusion features; and 4, acquiring new data in real time through an Internet of Things sensor, and dynamically updating the three-dimensional geophysical field model. When the multi-physics field data is fused, an attention mechanism and a self-adaptive weight distribution method are introduced, the weight of each physics field data is automatically adjusted according to the geologic features of different areas of the mining area, and the data fusion precision and the model robustness are improved.

Description

Technical Field

[0001] The invention relates to the technical field of three-dimensional geological modeling of mining areas, and in particular to a method for constructing a three-dimensional geophysical field in an open-pit mining area. Background Art

[0002] With the continuous exploitation of mineral resources, there are more and more cases of open-pit mining being converted to underground mining and underground mining being converted to open-pit mining. In the process of open-pit mining, disasters such as slope landslides, goaf collapses, mudslides and water seepage caused by fault fracture zones, goafs, water-bearing bodies and other disaster-causing bodies are becoming increasingly serious, posing new challenges to the accurate detection of disaster-causing bodies, and a single geophysical detection method cannot meet its requirements.

[0003] In order to improve the mining efficiency and safety of mining areas, it is crucial to accurately construct a high-precision three-dimensional geological model. Traditional geological exploration methods usually rely on single physical field data, or improve the model through manual correction and empirical judgment, but these methods often have great limitations in complex geological environments. To this end, the development of a three-dimensional geophysical field construction method that integrates multiple physical field data, can be updated in real time, and has intelligent optimization capabilities has become an important issue that needs to be urgently solved in the field of mining geological exploration.

[0004] In the prior art, the construction of a three-dimensional geological model of a mining area generally relies on the collection and analysis of multiple physical field data such as earthquakes, gravity, magnetism, electrical methods, and remote sensing. However, these traditional methods mostly rely on experience and manual intervention, and when fusing multi-physical field data, a weighted average method with fixed weights is usually used, which cannot fully reflect the changes in geological characteristics of different areas in the mining area. In addition, the prior art usually lacks a real-time data update mechanism. Once the model is generated, it is no longer updated and cannot adapt to changes in the mining area environment in a timely manner, resulting in the accuracy and practicality of the model being difficult to guarantee during long-term exploration and mining.

[0005] Existing technologies have significant deficiencies in multi-physics field data fusion and model updating, especially in different mining geological environments. They fail to adjust the weight of physical field data in a targeted manner, resulting in low accuracy of the constructed three-dimensional geological model in deep anomaly identification and complex geological structure analysis. The lack of a dynamic update mechanism makes it impossible for the model to reflect real-time changes in the mining area, resulting in a lag in exploration data, which in turn affects the safety of mining and resource utilization efficiency. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides a method for constructing a three-dimensional geophysical field in an open-pit mining area, which solves the problem that the existing methods fail to adjust the weights of physical field data in a targeted manner under different mining geological environments, resulting in low accuracy of the constructed three-dimensional geological model in deep anomaly identification and complex geological structure analysis.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for constructing a three-dimensional geophysical field in an open-pit to underground mining area, comprising the following steps:

[0008] Step 1, multi-physical field data acquisition, obtaining seismic data, gravity data, magnetic data, electrical method data and remote sensing data of the mining area;

[0009] Step 2, data preprocessing, performing normalization, multi-scale decomposition, noise processing and feature extraction on the multi-physical field data;

[0010] Step 3, constructing a multi-modal deep fusion model, adaptively allocating attention weights to the features of different physical field data, and realizing weighted fusion of the fusion features;

[0011] Step 4, real-time collecting new data through Internet of Things sensors, and performing incremental learning in combination with expert feedback to dynamically update the three-dimensional geophysical field model.

[0012] Preferably, the multi-modal deep fusion model includes:

[0013] Establishing an independent convolutional neural network branch for each type of physical field data, and extracting features through convolutional operations;

[0014] Introducing an adaptive attention mechanism in the fusion layer, and adjusting its weight according to the sensitivity of each type of physical field data to the geological structure of the mining area.

[0015] Preferably, the weight α of the adaptive attention mechanism i The calculation formula is:

[0016]

[0017] Among them, W α Is a trainable parameter, and F i Is the feature of different physical field data.

[0018] Preferably, the dynamic update includes the following steps:

[0019] Using Internet of Things sensors to collect new data of the open-pit to underground mining area in real time;

[0020] Based on incremental learning, online update the model to adapt to the dynamic changes of the geological structure of the mining area.

[0021] Preferably, the incremental learning updates the model parameters to minimize the incremental loss function, and the incremental loss function is:

[0022]

[0023] Among them: Indicates the model's response to the newly input data The predicted value, is the true label value in the new dataset, N new is the number of newly collected data samples.

[0024] Preferably, in the fourth step, an expert feedback mechanism is further included, and the expert feedback mechanism includes:

[0025] Geological experts manually correct the three-dimensional geological model generated by the model;

[0026] The corrected data is re-input into the model for training to improve the accuracy and adaptability of the model.

[0027] Preferably, the data preprocessing includes:

[0028] Data standardization processing, normalizing different physical field data;

[0029] Performing multi-scale decomposition on the data through wavelet decomposition to extract different scale features;

[0030] Using the wavelet soft threshold denoising method for noise processing to improve the signal-to-noise ratio of the data.

[0031] Preferably, the loss function of the multi-modal depth fusion model is a multi-task loss function, including feature similarity loss and boundary consistency loss which is defined as:

[0032]

[0033] where λ 1 and λ 2 are hyperparameters.

[0034] Preferably, the boundary consistency loss is used to constrain the accuracy of abnormal body boundary recognition, and its calculation formula is:

[0035]

[0036] where: represents the gradient value of the fused feature map F fused at the pixel or point (x, y), ||.|| 1 represents taking the absolute value of the gradient, and the calculation result is used to measure the change amplitude of the feature map at the boundary.

[0037] Preferably, the three-dimensional geophysical field model is used for geological structure modeling, abnormal body boundary recognition, resource estimation, and mining path planning in open-pit to underground mining areas to support the dynamic management and real-time decision-making of the mining area.

[0038] The present invention provides a method for constructing a three-dimensional geophysical field in an open-pit to underground mining area. It has the following

[0039] Advantages:

[0040] 1. When fusing multi-physical field data, the present invention introduces an attention mechanism and an adaptive weight allocation method, automatically adjusting the weights of each physical field data according to the geological characteristics of different areas in the mining area, making the model more adaptable to the requirements of different depths and different geological structures in the mining area, and improving the accuracy of data fusion and the robustness of the model.

[0041] 2. The present invention uses Internet of Things sensors to collect the latest data in the mining area in real time and online updates the model through incremental learning, enabling the model to timely reflect the dynamic changes in the mining area and ensuring the real-time nature and accuracy of the geological model during the mining process. This dynamic update ability significantly improves the practicality of the model in long-term exploration and mining in the mining area.

[0042] 3. Through the collection and fusion of multi-physical field data, the present invention integrates various data sources such as seismic, gravity, magnetic, electrical and remote sensing, and uses a deep learning model to extract and fuse various physical field features, effectively improving the resolution and accuracy of the three-dimensional geological model in the mining area, especially having obvious advantages in the fine identification of complex geological structures and deep abnormal bodies.

[0043] 4. The present invention introduces a feedback mechanism of geological experts to manually correct the geological structure generated by the model and retrain it, enabling the model to automatically optimize in combination with expert experience, thereby further improving the intelligence and prediction accuracy of the model, making it more in line with the geological conditions of the actual mining area. The constructed high-precision three-dimensional geophysical field model can be applied to geological exploration, abnormal body identification, resource estimation and mining path planning in the open-pit to underground mining area, providing comprehensive geological structure information for the mining area, effectively reducing the mining risk, improving the resource utilization rate, and providing reliable data support for mining area management and decision-making. Description of the Drawings

[0044] Figure 1 It is a flowchart of the method of the present invention. Detailed Embodiments

[0045] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0046] Embodiment 1:

[0047] Please refer to the attached Figure 1, the embodiment of the present invention provides a method for constructing a three-dimensional geophysical field in an open-pit to underground mining area, including the following steps:

[0048] Step 1: Collect multi-physical field data to obtain seismic data, gravity data, magnetic data, electrical method data, and remote sensing data of the mining area;

[0049] Step 2: Perform data preprocessing, including normalizing, multi-scale decomposition, noise processing, and feature extraction on the multi-physical field data;

[0050] Step 3: Construct a multi-modal deep fusion model, adaptively assign attention weights to the features of different physical field data, and achieve weighted fusion of the fused features;

[0051] Step 4: Collect new data in real time through Internet of Things sensors, and perform incremental learning in combination with expert feedback to dynamically update the three-dimensional geophysical field model.

[0052] In one embodiment, in this embodiment, the specific steps of the method for constructing a three-dimensional geophysical field in an open-pit to underground mining area include:

[0053] Multi-physical field data collection:

[0054] In this embodiment, various geophysical technical means such as seismic, gravity, magnetic, electrical method, and remote sensing are used to comprehensively collect data in the mining area, ensuring that the data input covers different depths and formation characteristics, and providing comprehensive and accurate basic data for the subsequent geological model.

[0055] Seismic data collection: By deploying a seismic observation network, obtain the propagation, reflection, and refraction data of seismic waves, and capture the characteristic information of the layered structure and main geological structures inside the mining area. Seismic data can reveal the boundaries of geological horizons and the change trends of underground structures, providing accurate underground structure data for the three-dimensional model.

[0056] Gravity and magnetic data collection: Use gravity measurement to obtain density change data of the mining area, and use magnetic measurement to obtain magnetic anomaly data. Combining gravity anomaly and magnetic anomaly data can identify areas with ore body density and magnetic differences, which helps to identify the distribution and possible depth range of ore bodies.

[0057] Electrical method data collection: Adopt transient electromagnetic method and high-density resistivity method to collect resistivity distribution data of the mining area. This resistivity information can help identify the electrical differences between abnormal bodies and surrounding rocks, provide the electrical characteristics of the boundaries of abnormal bodies, and further enhance the refined expression ability of the model.

[0058] Remote Sensing and LiDAR Data Acquisition: Three-dimensional structural data of the mining area surface is collected by drones equipped with LiDAR and high-resolution cameras. Remote sensing and LiDAR data are used to construct surface boundary conditions, capture the morphological characteristics of the surface, and provide accurate surface structures for three-dimensional geological models.

[0059] Data Preprocessing and Multi-scale Feature Extraction:

[0060] In this embodiment, the collected multi-physical field data is preprocessed to ensure data compatibility and effectiveness, and multi-scale geological features are extracted to provide rich input information for the model.

[0061] Data Standardization: Standardize data sources such as seismic, gravity, magnetic, and electrical methods to address differences in numerical scales and achieve compatibility between data sources. The standardized data ensures the consistency of model input and the balance of subsequent feature fusion.

[0062] Multi-scale Decomposition: Introduce a multi-scale decomposition method in data preprocessing to decompose data from different physical fields into feature hierarchies at multiple scales, enabling the simultaneous capture of local fine features and overall trend features during model training. The extraction of multi-scale features provides multi-level support for the accurate modeling of complex geological structures.

[0063] Noise Processing: Environmental noise is inevitably introduced during data collection. In this embodiment, advanced methods such as wavelet denoising are used to process the data of each physical field to improve the signal-to-noise ratio of the data. Through efficient noise filtering, the accuracy of subsequent feature extraction is ensured, providing high-quality data input for the three-dimensional geological model.

[0064] Feature Extraction: Use a convolutional neural network (CNN) to extract features from data of different physical fields. Extract waveform features from seismic data, density and magnetic anomaly features from gravity and magnetic data, resistivity change features from electrical method data, etc. The multi-level features extracted from each physical field data by the CNN reflect the geological properties inside the mining area, providing diverse information input for the fusion model.

[0065] Construction of a Multi-modal Deep Fusion Model:

[0066] In this embodiment, a fusion model based on multi-modal deep learning is constructed to fuse data from different physical fields through an adaptive attention mechanism to generate comprehensive features of the three-dimensional geophysical field.

[0067] Multi-input deep neural network structure: Independent convolutional neural network branches are designed for each type of physical field data, enabling the features of seismic, gravity, magnetic, and electrical method data to be processed separately. Each convolutional branch extracts features from different data sources, forming a feature representation containing various geological information and providing diverse information support for the fusion process.

[0068] Adaptive attention mechanism: An adaptive attention mechanism is introduced during the feature fusion process. According to the contribution of each type of physical field data to the geological structure of the mining area, the weights of each data source are dynamically adjusted. This mechanism can enhance the recognition of key geological features in the mining area, enabling the model to highlight the key points and suppress non-primary information in different strata, and improving the expression ability of the fused features.

[0069] Fusion feature generation: The fusion model weights and integrates the feature information of each data source to generate a comprehensive three-dimensional geological feature representation. The fused features not only contain the advantages of multi-physical field data but also retain the relative importance of different data sources, providing accuracy and integrity for the three-dimensional geophysical field model.

[0070] Model training and optimization:

[0071] In this embodiment, a large amount of mining area data is used to train and optimize the fusion model to ensure the prediction accuracy and generalization ability of the model in a complex mining area environment.

[0072] Training data preparation: Based on the actual mining area data, data augmentation techniques are combined to generate diverse samples and enhance the robustness of the model. Data augmentation methods include randomly adding noise, data rotation, translation, and scaling to simulate the complex changes in the mining area environment and improve the adaptability of the model.

[0073] Optimization algorithm: The Adam optimization algorithm is used for model training. By iteratively updating the parameters of the multi-modal feature fusion model, the model prediction error is continuously reduced, and the convergence effect and accuracy of the model are improved, enabling it to accurately identify the geological features of the mining area.

[0074] Multi-task loss function: A multi-task loss function is introduced during model training, including feature similarity loss and anomaly boundary consistency loss, ensuring that the model can not only accurately fuse the features of different physical field data but also clearly identify the anomaly boundary features. The design of the loss function effectively constrains the feature fusion and boundary accuracy during the model training process, making the three-dimensional geological model more reliable.

[0075] Dynamic update and expert feedback mechanism

[0076] In this embodiment, a dynamic update and expert feedback mechanism is adopted to ensure that the model can adapt to the changes in the mining area geological environment in real time and improve the real-time performance and accuracy of the three-dimensional geological model.

[0077] Data stream monitoring and incremental learning: By deploying Internet of Things sensors within the mining area, newly generated data such as seismic, gravity, and electrical method data are collected in real-time. These data are input into the model for incremental learning. Incremental learning enables the model to quickly adjust to new data, achieving dynamic updates and self-adaptation of the 3D geological model.

[0078] Expert feedback mechanism: Based on the dynamic update of the model, an expert feedback mechanism is introduced. Geological experts can correct the 3D geological prediction results of the model according to the actual situation of the mining area. By re-inputting the expert-corrected data into the model for retraining, it is ensured that the model can maintain high accuracy and adaptability under the dynamically changing mining area conditions, improving the reliability of the model in practical applications.

[0079] The multi-modal deep fusion model includes:

[0080] Establish independent convolutional neural network branches for each type of physical field data, and extract features through convolutional operations;

[0081] Introduce an adaptive attention mechanism in the fusion layer to adjust the weights according to the sensitivity of each type of physical field data to the geological structure of the mining area.

[0082] In one embodiment, feature extraction of independent convolutional neural network branches:

[0083] Each type of physical field data (such as seismic, gravity, magnetic, electrical method, etc.) has different physical characteristics and data distribution characteristics. Therefore, the present invention designs an independent CNN branch for each data type. Each CNN branch performs convolutional operations according to the characteristic structure of a specific data type, and gradually extracts and simplifies the complex information of the original data through multiple layers of convolution and pooling to form a high-level feature representation. For example:

[0084] Seismic data extracts waveform features through convolutional operations, reflecting the formation structure and the location of abnormal bodies.

[0085] Gravity data extracts density distribution features through convolutional operations, which helps to identify ore bodies with large density differences.

[0086] Magnetic data extracts magnetic anomaly information through convolutional operations to help identify magnetic ore bodies within the mining area.

[0087] Electrical method data extracts resistivity change features through convolutional operations, reflecting the electrical property differences between abnormal bodies and surrounding rocks.

[0088] Each CNN branch performs deep encoding on the feature information of different physical field data through an independent feature extraction process and generates its own data feature representation. This independent extraction mechanism ensures the complete expression of each physical field feature and avoids feature confusion between different data sources.

[0089] Weight adjustment of the adaptive attention mechanism in the fusion layer:

[0090] An adaptive attention mechanism is introduced in the fusion layer of the multi-modal deep fusion model to adjust its weight in the overall feature representation according to the sensitivity of each physical field data to the geological structure of the mining area.

[0091] Adaptive adjustment of weights: The adaptive attention mechanism dynamically adjusts the feature importance of different data sources by calculating the attention weights of each physical field feature. In practical applications, the sensitivities of different physical field data to different geological features are not the same. The attention mechanism can learn and determine the weights of each data feature based on the feedback during the training process, so that important feature information is amplified and unimportant or interfering information is suppressed.

[0092] Dynamic feature fusion: In the fusion layer, according to the weights calculated by the attention mechanism, the features of each physical field are weighted and then fused to generate a more representative comprehensive feature representation. The adaptive attention mechanism allows the model to spontaneously adjust the weights of each data source in practical applications, so that the features of different physical field data are balanced and reasonably expressed in the fused feature representation.

[0093] For example, in some specific geological structures, seismic data is more sensitive to structural changes, while electrical method data can reflect the boundaries of abnormal bodies more accurately. The adaptive attention mechanism can automatically enhance the weights of seismic features and electrical method features in this case, so as to ensure the accurate characterization of the boundaries between ore bodies and abnormal bodies and the accurate reflection of geological structures in the fused features.

[0094] Generation of comprehensive features of 3D geological structures:

[0095] The feature representation after adaptive weighted fusion contains comprehensive information from multi-physical field data, which can comprehensively reflect the geological structure and the distribution of abnormal bodies in the mining area. This comprehensive feature has higher expressiveness and can capture complex geological details during model training. After being optimized by model training, this 3D geological structure feature will be used as the final output of the 3D geophysical field model of the mining area, meeting the actual needs of mining area exploitation and exploration.

[0096] The weight α of the adaptive attention mechanism i The calculation formula is:

[0097]

[0098] Among them, W α is a trainable parameter, and F i is the feature of different physical field data.

[0099] The dynamic update includes the following steps:

[0100] Utilize Internet of Things sensors to collect new data of the open-pit to underground mining area in real time;

[0101] Based on incremental learning, perform online update on the model to adapt to the dynamic changes of the mining area geological structure.

[0102] In one embodiment, the real-time data collection of Internet of Things sensors

[0103] Deploy multiple Internet of Things sensors inside the mining area for real-time monitoring of physical field data such as seismic, gravity, and electrical methods. The Internet of Things sensors collect data at regular intervals and transmit the data to the central server through a wireless network to ensure that the three-dimensional geophysical field model can receive the latest geological information of the mining area.

[0104] Seismic data collection: The sensor records the propagation changes of seismic waves in real time, reflecting the minute changes in the formation structure, and providing updated formation structure information for the model.

[0105] Gravity and electrical method data collection: Real-time collection of data such as the gravity field and resistivity of the mining area, reflecting the changes in ore body density and electrical properties, and ensuring that the model can respond in a timely manner when abnormal changes are detected.

[0106] Through real-time monitoring and data transmission, the model continuously receives the latest mining area information, so that it can quickly capture the changes in the internal structure of the mining area and keep the model always up-to-date with information.

[0107] Application of incremental learning in the model:

[0108] Incremental learning is the core of the dynamic update mechanism of the present invention. Through incremental learning, the model can quickly adjust to new data without retraining the entire network.

[0109] Local adjustment of the model: Incremental learning is based on the original trained model and uses new data for local update. Compared with retraining, incremental learning only needs to partially optimize the weights of the model, greatly reducing the training time and ensuring that the model can be updated in a short time.

[0110] Minimization of the incremental loss function: During the incremental learning process, the model calculates the loss of new data and uses optimization methods such as gradient descent to minimize the error of the new data, ensuring the adaptability of the model to new information.

[0111] Incremental learning enables adaptive and rapid adjustment in the model, allowing the three-dimensional geophysical field model to dynamically and accurately represent the latest geological structure changes.

[0112] Real-time update and gradual optimization of the model;

[0113] Based on incremental learning, the model can continuously optimize under the drive of real-time data:

[0114] Continuous optimization process: As the real-time data collected from the mining area flows into the model, incremental learning continuously updates the model weights to ensure that the model continuously adapts to the dynamic changes in the geological structure of the mining area.

[0115] Rolling data update: The mechanism of incremental learning supports rolling data update, enabling the model to gradually optimize its prediction ability based on each new data, thereby improving the model's adaptability to future data and its ability to capture geological structure changes.

[0116] This rolling update and continuous optimization process enable the model to always be based on the latest mining area information, making the prediction of the three-dimensional geophysical field more accurate and real-time.

[0117] Incremental learning updates the model parameters to minimize the incremental loss function, and the incremental loss function is:

[0118]

[0119] Where: represents the predicted value of the model for the new input data of, is the true label value in the new dataset, and N new is the number of newly collected data samples.

[0120] In step four, an expert feedback mechanism is also included, and the expert feedback mechanism includes:

[0121] Geological experts manually correct the three-dimensional geological model generated by the model;

[0122] The corrected data is re-input into the model for training to improve the accuracy and adaptability of the model.

[0123] In one embodiment, the manual correction of the three-dimensional geological model by geological experts

[0124] After the model generates a preliminary three-dimensional geological model, geological experts will evaluate and correct the key features such as the geological structure and the boundary of abnormal bodies generated by the model according to the actual geological exploration data and the characteristics of the mining area.

[0125] Main content of manual correction: Experts check the detailed features such as the internal structure of the mining area (such as faults and folds), the boundaries and thicknesses of anomalies, identify the parts that the model fails to fully reflect or misjudges in the complex geological environment, and make manual adjustments based on the actual data. For example, experts can correct the strike, depth, boundary shape, etc. of anomalies according to the new exploration data to ensure that the model is more accurate in key geological features.

[0126] Feedback and correction process: After discovering the deviation, experts will adjust the relevant features to generate a model that better conforms to the geological actual situation. This corrected model will be used as new data for the retraining of the model.

[0127] Re-input of corrected data and retraining of the model:

[0128] The data corrected by experts is re-input into the model to further optimize the prediction accuracy and adaptability of the model.

[0129] Generation of retraining data: The data corrected by experts is used as high-value samples and added to the training dataset of the model. During the training process, the model will adjust the key parameters according to the new corrected data, thereby improving the recognition ability of geological structure features and the overall accuracy of the model.

[0130] Gradually optimized training process: The process of retraining the model is an incremental optimization process. Each time the corrected data provided by experts will gradually affect the weights and structure of the model, enabling the model to achieve continuous learning in multiple rounds of feedback and retraining and gradually adapt to the changes in the geological structure. This feedback mechanism enables the model to better understand the knowledge provided by geological experts and reduce the same type of errors in future data processing.

[0131] Strengthening of the model's prediction ability by expert feedback:

[0132] Through the feedback and retraining process of experts, the model's recognition ability for complex geological structures and anomaly features gradually improves.

[0133] Accumulation of feedback information: Each piece of feedback information from experts will be "remembered" by the model and applied to future predictions. With the continuous accumulation of feedback information, the model's prediction ability will be gradually strengthened, and the judgment of future geological features will be more accurate, especially when dealing with complex geological structures and subtle anomaly features.

[0134] Reduction of model errors and biases: The expert feedback process effectively controls the model errors. By integrating expert knowledge into the model training, the model's prediction of anomaly boundaries, shapes, and formation changes will be more accurate, and the model errors will gradually decrease, thereby improving the reliability of the model.

[0135] Dynamic Feedback Optimization Model Adaptability and Practicality:

[0136] The expert feedback mechanism can not only improve the static prediction ability of the model but also optimize the adaptability of the model in a real-time environment.

[0137] Continuous Learning: The model retrained after each expert feedback has the ability to "learn" new geological structures. This enables the model to continuously update its knowledge when encountering new geological phenomena, achieve adaptive adjustment, and enhance the long-term applicability of the model.

[0138] Adapting to Different Geological Environments: Expert feedback not only enables the model to adapt to the geological conditions of the current mining area but also enhances the application potential of the model in future similar mining area environments. The model can quickly respond to the feedback of geological experts in future mining area applications, thus possessing strong environmental migration ability.

[0139] Data preprocessing includes:

[0140] Data standardization processing, normalizing different physical field data;

[0141] Performing multi-scale decomposition on the data through wavelet decomposition to extract features at different scales;

[0142] Using the wavelet soft threshold denoising method for noise processing to improve the signal-to-noise ratio of the data.

[0143] In one embodiment, data standardization processing:

[0144] The multi-physical field data in the mining area (such as seismic, gravity, magnetic, and electrical method data) usually has different numerical ranges and scales. The differences between different data sources may lead to weight imbalance during model training and prediction, affecting the fusion effect. In this embodiment, through data standardization processing, the physical field data is normalized to the same numerical range, enabling the subsequent multi-modal fusion model to maintain consistency when inputting different physical field data.

[0145] Processing process: For each data type, calculate its minimum value and maximum value respectively, and scale the data to the interval between 0 and 1. The standardized data eliminates the differences in the numerical ranges of each physical field, ensuring that the model will not cause unbalanced feature weights due to different numerical scales when processing multi-source data.

[0146] Effect: Data standardization processing enhances the compatibility of different data sources, enabling the deep learning model to fairly utilize the information of each data source, which helps to improve the overall accuracy and performance of the fusion model.

[0147] Multi-scale Feature Extraction by Wavelet Decomposition:

[0148] Geological data usually contains characteristic information at different spatial scales (such as local anomalies and overall trends). Direct fusion may cause the model to ignore subtle features or lose important hierarchical information. In this embodiment, wavelet decomposition technology is used to decompose the data into features at multiple scales, enabling the capture of geological structure information at different scales.

[0149] Decomposition process: Wavelet decomposition is performed on each physical field data to obtain a multi-level feature representation, which includes high-frequency and low-frequency features respectively. High-frequency features represent local detailed features (such as the boundaries of abnormal bodies or geological faults), while low-frequency features represent the overall geological structure or trend.

[0150] Effect: Multi-scale features enable the model to identify and express the geological information of the mining area at different levels, not only retaining the overall structural information but also being able to magnify local features. This multi-scale decomposition technology improves the model's ability to capture details and its adaptability to geological structures.

[0151] Wavelet soft threshold denoising processing:

[0152] Environmental noise is inevitably introduced during the data acquisition process, such as measurement equipment errors and external interferences. To ensure the accuracy of subsequent feature extraction, the present invention adopts the wavelet soft threshold denoising method to effectively remove the noise signal and improve the purity of the data.

[0153] Denoising process: After wavelet decomposition of the data, soft threshold processing is applied to the decomposed high-frequency coefficients, only retaining important geological signals while suppressing irrelevant high-frequency noise. Soft threshold processing can reduce the high-frequency coefficients below the set threshold to zero, reducing the interference of noise on model training.

[0154] Effect: The wavelet soft threshold denoising method effectively filters out the noise signal, retains important geological information, makes the input data of the model clearer and more accurate, is suitable for feature extraction and model training, and thus improves the stability and accuracy of the 3D geological model.

[0155] Effective docking of data preprocessing and deep learning models:

[0156] Through data standardization processing, multi-scale decomposition, and denoising processing, data features are clearly stratified, and noise is effectively suppressed, making the information after data preprocessing more in line with the input requirements of deep learning models.

[0157] Enhancing the model's feature expression ability: The data after standardization, decomposition, and denoising enables the model to have a clearer feature expression when processing multi-source data, ensuring the efficient extraction and utilization of features from different physical fields during model fusion.

[0158] Improve the fusion effect and prediction accuracy: The preprocessed data has achieved a high degree of consistency in numerical scale and feature level, which is conducive to the unified processing and fusion of multi-physical field data by the model. Finally, these data provide stable and accurate inputs for the deep learning model, improving the prediction accuracy and effect of the 3D geological model.

[0159] The loss function of the multi-modal deep fusion model is a multi-task loss function, including feature similarity loss and boundary consistency loss which is defined as:

[0160]

[0161] where λ 1 and λ 2 are hyperparameters.

[0162] The boundary consistency loss is used to constrain the accuracy of abnormal body boundary recognition, and its calculation formula is:

[0163]

[0164] where: represents the gradient value of the fused feature map F fused at the pixel or point position (x, y), ||.|| 1 represents taking the absolute value of the gradient, and the calculation result is used to measure the change amplitude of the feature map at the boundary.

[0165] The 3D geophysical field model is used for geological structure modeling, abnormal body boundary recognition, resource estimation, and mining path planning in open-pit to underground mining areas to support the dynamic management and real-time decision-making of the mining area.

[0166] Preferred Embodiment 1: Adaptive Standardization and Feature Scaling:

[0167] In data standardization processing, there are large differences in the numerical ranges of different physical field data, and the geological characteristics of each area in the mining area are uneven. To further improve the adaptability of data preprocessing, this preferred embodiment introduces adaptive standardization and feature scaling techniques to dynamically adjust the scale range of data during model training, thereby enhancing the sensitivity of the model to geological feature changes and prediction accuracy.

[0168] Adaptive Standardization: Dynamically adjust the standardization range according to the geological characteristics of different regions. For example, perform standardization processing on data in ore body dense areas with a smaller range to highlight local dense features; perform standardization processing on background areas with a larger range to balance the overall data scale. This adaptive standardization ensures the adaptability of the model under different geological conditions.

[0169] Feature Scaling: Further scale the data features after standardization and assign weights to different physical field data. For example, according to the sensitivity of seismic data to structural changes, increase its feature weight; for gravity data, adjust its feature scaling coefficient according to the importance of density changes. Through feature scaling, key features can be amplified in the model to further improve the multi-modal data fusion effect.

[0170] Preferred Embodiment Two: Dynamic Update Mechanism Based on Feedback Priority:

[0171] During the dynamic update and expert feedback process, different feedback information and data update contents have different priorities for model optimization. This preferred embodiment introduces a dynamic update mechanism based on feedback priority to focus on updating high-priority feedback contents to improve the adaptability of the model to key geological features and changes in abnormal bodies.

[0172] Feedback Priority Division: According to the real-time data changes in the mining area and the expert feedback contents, the data and feedback are divided into high priority, medium priority, and low priority. For example, changes in the boundary of abnormal bodies, discovery of new ore layers, etc. are regarded as high-priority feedback, local detail adjustments are medium priority, and minor changes in the background area are low priority.

[0173] Key Update Strategy: Focus on incrementally updating high-priority feedback contents and preferentially adjust the model weights and parameters. For low-priority feedback contents, adopt delayed update or batch update methods to reduce the computational burden and ensure that resources are preferentially used for optimizing key features.

[0174] Real-time Monitoring and Dynamic Adjustment: During each feedback update process, continuously monitor the changes in the model in key areas and perform multiple iterative updates on high-priority contents according to further feedback from geological experts to ensure that the model can accurately adapt to sudden changes in geological features and major changes in the mining area environment.

[0175] Comparative Example One:

[0176] 3D Modeling Based on Inversion Algorithm: Use the inversion algorithm to process geophysical data and gradually optimize the 3D model to make the simulated data match the actual measurement data. Commonly used inversion algorithms include full waveform inversion, joint inversion, etc., which can integrate and process various geophysical data to improve the accuracy of the model.

[0177] Data Acquisition: Obtain seismic, gravity, magnetic, electrical method, and remote sensing data of the open-pit to underground mining area and collect them by region to ensure the diversity and integrity of the data.

[0178] Experimental Group and Control Group:

[0179] Experimental Group: Adopt the multi-modal deep fusion model in Embodiment One.

[0180] Control group: A three-dimensional modeling method based on traditional inversion algorithms (such as full waveform inversion, joint inversion, etc.) was adopted.

[0181] Data preprocessing: All data were standardized, noise processed, and feature extracted to ensure that the data preprocessing processes of the experimental group and the control group were the same.

[0182] Model training and optimization: The multi-modal deep fusion model of the experimental group was trained, and dynamic updates were carried out using incremental learning and expert feedback mechanisms. The traditional inversion model of the control group was trained according to a fixed algorithm.

[0183] Experimental data recording: Key indicators such as the accuracy of each group of models, the recognition accuracy of the boundaries of abnormal bodies, the calculation speed, and the adaptability under different geological change scenarios were recorded.

[0184]

[0185] Table 1

[0186] The multi-modal deep fusion model of Example 1 demonstrated significant advantages in the construction of the three-dimensional geophysical field in the mining area. By using the adaptive attention mechanism to perform weighted fusion on different physical field data and combining incremental learning and expert feedback mechanisms, the model can adapt to the dynamic changes of the geological structure in real time, improving the recognition accuracy of the boundaries of abnormal bodies and the overall accuracy of the model.

[0187] Compared with the traditional inversion algorithm, the fusion model of Example 1 not only showed excellent performance in terms of recognition accuracy and the characterization of the boundaries of abnormal bodies, but also had obvious advantages in calculation speed and dynamic adaptability. The real-time monitoring and dynamic adjustment functions of this model make it more flexible and reliable in practical applications and suitable for complex mining area environments. Generally speaking, Example 1 provides an efficient and accurate method for constructing a three-dimensional geophysical field, with good application prospects and expansion potential.

[0188] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a three-dimensional geophysical field in an open-pit mining area, characterized in that: The following steps are involved: Step 1: Multi-physics field data acquisition, obtaining seismic data, gravity data, magnetic data, electrical data and remote sensing data of the mining area; Step 2: Data preprocessing: normalization, multi-scale decomposition, noise processing and feature extraction of multi-physics field data; Step 3: Build a multimodal deep fusion model, perform adaptive attention weight allocation on the features of different physical field data, and realize weighted fusion of fusion features; Step 4: Collect new data in real time through IoT sensors and perform incremental learning based on expert feedback to dynamically update the three-dimensional geophysical field model.

2. The method for constructing a three-dimensional geophysical field in an open-pit mining area according to claim 1, characterized in that: The multimodal deep fusion model includes: Establish an independent convolutional neural network branch for each physical field data and extract features through convolution operation; An adaptive attention mechanism is introduced in the fusion layer to adjust the weight of each physical field data according to its sensitivity to the geological structure of the mining area.

3. The method for constructing a three-dimensional geophysical field in an open-pit mining area according to claim 1, characterized in that: The weight α of the adaptive attention mechanism i The calculation formula is: Among them, W α is a trainable parameter, F i are the characteristics of different physical field data.

4. The method for constructing a three-dimensional geophysical field in an open-pit mining area according to claim 1, characterized in that: The dynamic update comprises the following steps: Use IoT sensors to collect new data in real time from open-pit mining areas; Based on incremental learning, the model is updated online to adapt to the dynamic changes in the geological structure of the mining area.

5. The method for constructing a three-dimensional geophysical field in an open-pit mining area according to claim 4, characterized in that: The incremental learning updates the model parameters to minimize its incremental loss function, which is: in: Represents the model's response to new input data The predicted value of is the true label value in the new dataset, N new is the number of newly collected data samples.

6. The method for constructing a three-dimensional geophysical field in an open-pit mining area according to claim 1, characterized in that: The step 4 also includes an expert feedback mechanism, which includes: Geological experts manually modify the three-dimensional geological model generated by the model; The corrected data is re-input into the model for training to improve the accuracy and adaptability of the model.

7. The method for constructing a three-dimensional geophysical field in an open-pit mining area according to claim 1, characterized in that: The data preprocessing includes: Data standardization processing, normalizing data from different physical fields; Decompose the data into multiple scales through wavelet decomposition to extract features of different scales; The wavelet soft threshold denoising method is used to process noise and improve the signal-to-noise ratio of the data.

8. The method for constructing a three-dimensional geophysical field in an open-pit mining area according to claim 1, characterized in that: The loss function of the multimodal deep fusion model is a multi-task loss function. Including feature similarity loss and boundary consistency loss It is defined as: Among them, λ1 and λ2 are hyperparameters.

9. The method for constructing a three-dimensional geophysical field in an open-pit mining area according to claim 8, characterized in that: The boundary consistency loss It is used to constrain the accuracy of abnormal body boundary recognition. The calculation formula is: in: Indicates that at the pixel or point (x, y), the fusion feature map F fused The gradient value of ||.||1 means taking the absolute value of the gradient. The calculation result is used to measure the change amplitude of the feature map at the boundary.

10. The method for constructing a three-dimensional geophysical field in an open-pit mining area according to claim 1, characterized in that: The three-dimensional geophysical field model is used for geological structure modeling, abnormal body boundary identification, resource estimation and mining path planning in the open-pit mining area to support dynamic management and real-time decision-making of the mining area.

Images