Mineralization anomaly real-time early warning system and method

Abstract

This invention discloses a real-time early warning system and method for mineralization anomalies, belonging to the field of mineralization material detection technology. The system includes a mineralization material data acquisition module, a positioning module, a data storage module, a material data preprocessing module, a mineralization material anomaly identification module, an early warning classification module, an early warning output module, and a model dynamic update module. This invention comprehensively addresses the shortcomings of traditional mineralization anomaly monitoring technologies from the data layer, algorithm layer, and application layer through a closed-loop design encompassing multi-source data fusion acquisition, standardized preprocessing, high-precision intelligent identification, classified early warning output, and dynamic model updates. It employs simultaneous acquisition of multi-source data (spectral data, elemental content, environmental parameters), combined with BeiDou + GPS dual-mode positioning, to achieve precise binding of mineralization data with spatial location; a timed + triggered dual acquisition mode ensures data real-time performance and integrity, overcoming the limitations of traditional single-data-dimensional approaches.

Description

Technical Field

[0001] This invention relates to the field of mineralization material detection technology, and more specifically, to a real-time early warning system and method for mineralization anomalies. Background Technology

[0002] In the process of mineral resource exploration and safe production, the timely detection and accurate early warning of mineralization anomalies are crucial, directly affecting the rational development of mineral resources, the safe control of the geological environment in mining areas, and the improvement of exploration efficiency. Mineralization anomalies are mainly manifested as abnormal fluctuations in the mineral composition and element content of mineralized materials in mining areas. If they are not identified and warned of in a timely manner, they may lead to problems such as deviations in resource exploration and hidden dangers in mining safety. Therefore, achieving real-time and accurate early warning of mineralization anomalies has important practical significance.

[0003] Currently, mineralization anomaly monitoring in mining areas mainly relies on traditional manual sampling and detection, combined with single-sensor data acquisition and simple analysis. Existing technologies have several shortcomings and cannot meet practical application needs: First, data acquisition methods are relatively limited, often only collecting data on one or a few mineral components or elemental contents, lacking the collection of auxiliary data such as mining area environmental parameters, resulting in insufficient data dimensions and affecting the accuracy of mineralization anomaly identification. Second, data preprocessing methods are simplistic, failing to effectively reduce noise, standardize, and complete missing values ​​for the collected multi-source data, resulting in poor data quality and further reducing the accuracy of anomaly identification. Third, mineralization anomaly identification often uses traditional algorithms, which struggle to effectively extract the spatial and temporal characteristics of mineralization data, leading to low identification accuracy, high false alarm rates, and an inability to accurately quantify anomaly features. Fourth, existing systems lack dynamic update mechanisms; fixed model parameters prevent optimization of identification accuracy based on new mineralization data, making it difficult to adapt to dynamic changes in the mining area's mineralization environment. Fifth, some monitoring systems fail to accurately locate anomaly positions, and the output of early warning information is incomplete, preventing staff from quickly locating anomaly areas and conducting on-site verification, thus affecting the timeliness and effectiveness of early warnings.

[0004] Furthermore, existing mineralization anomaly early warning technologies mostly focus on optimizing single aspects, failing to form a complete closed loop of "data acquisition - preprocessing - anomaly identification - tiered early warning - model update," resulting in low early warning efficiency and limited practicality. To address these shortcomings, there is an urgent need for a real-time mineralization anomaly early warning technology capable of accurate multi-source data acquisition, efficient preprocessing, high-precision anomaly identification, tiered early warning, and dynamic model optimization. This technology would solve problems such as low identification accuracy, untimely early warning, and poor adaptability, meeting the actual needs of mineralization anomaly monitoring in mining areas. Therefore, we propose a real-time mineralization anomaly early warning system and method. Summary of the Invention

[0005] The purpose of this invention is to provide a real-time early warning system and method for mineralization anomalies to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A real-time early warning system for mineralization anomalies includes: The mineralization material data acquisition module is used to collect multi-source characteristic data of mineralization-related materials in the target area of ​​the mining area in real time. The multi-source characteristic data includes mineral composition data, element content data and mining area environmental parameter data. The mineral composition data is obtained through spectral detection and the element content data is obtained through chemical detection methods. The positioning module is used to collect geographical location data of mineralization-related material samples in real time and determine the collection location corresponding to the characteristic data of each material. The data storage module is used to store the collected mineralization-related material characteristic data, geographical location data, historical material analysis data, anomaly identification results, and model parameters; The material data preprocessing module is communicatively connected to the mineralized material data acquisition module, the positioning module, and the data storage module. It is used to perform noise reduction, standardization, and missing value completion processing on the acquired mineralized material characteristic data and geographical location data to obtain a standardized material analysis dataset. The mineralized material anomaly identification module is communicatively connected to the material data preprocessing module and the data storage module. It uses a CNN-LSTM hybrid network with an attention mechanism to extract features and identify anomalies in the standardized material analysis dataset, and outputs the mineralized material anomaly features and anomaly confidence. The early warning and classification module is communicatively connected to the mineralized material anomaly identification module. It is used to determine the anomaly level of the mineralized material based on the anomaly characteristics and anomaly confidence level of the mineralized material, combined with the preset mineralized material anomaly classification standard. The early warning output module is communicatively connected to the early warning classification module, the positioning module, and the data storage module. It is used to output corresponding early warning information, abnormal material collection location, and abnormal feature description in real time according to the abnormality level of the mineralized material, and push the early warning information to the designated terminal. The model dynamic update module is communicatively connected to the mineralized material anomaly identification module, the mineralized material data acquisition module, and the data storage module. It is used to perform incremental training of the model based on newly added mineralized material feature data, geographical location data, and anomaly identification results, combined with transfer learning and incremental learning methods, to optimize the accuracy of material anomaly identification.

[0007] Preferably, the mineralization material data acquisition module includes a spectral sensor, an elemental analyzer, an environmental sensor, and a data acquisition terminal. The spectral sensor is used to collect mineral spectral data of mineralization materials on the surface and in the shallow underground layer (0-5 meters) of the mining area. The elemental analyzer is used to detect the content of target mineralization elements in soil and rock samples of the mining area. The environmental sensor is used to collect environmental parameters such as temperature, humidity, and topographic slope of the mining area. The environmental parameters are used to assist in correcting the characteristic data of mineralization materials. The data acquisition terminal is used to realize the synchronous acquisition and real-time transmission of characteristic data of multi-source materials.

[0008] Preferably, the processing steps of the material data preprocessing module include: using wavelet denoising algorithm to remove noise interference from mineralization-related material feature data and geographical location data; using Z-score normalization method to map material data of different dimensions to the same order of magnitude; using K-nearest neighbor interpolation method to complete missing values ​​in the data; and finally generating a standardized material analysis dataset with unified structure and complete data.

[0009] Preferably, the training process of the CNN-LSTM hybrid network with attention mechanism includes: acquiring historical mineralization material feature data, corresponding geographical location data, and anomaly annotation information to construct a material analysis training dataset; inputting the training dataset into the CNN-LSTM hybrid network, extracting spatial features of the mineralization material data through CNN, and extracting temporal variation features of the mineralization material data through LSTM; introducing an attention mechanism to assign weights to the extracted spatial and temporal features, prioritizing features related to mineralization material anomalies, and suppressing interference from irrelevant features; training the model using cross-entropy loss function and gradient descent algorithm until the model converges, thus obtaining the trained mineralization material anomaly recognition model.

[0010] Preferably, the pre-set mineralization material anomaly classification standard of the early warning classification module includes three levels: Level 1 anomaly, with an anomaly confidence level of 50% ≤ anomaly confidence level < 70%, indicating that the anomaly in the corresponding mineralized material is initially apparent, with no obvious mineralization enrichment; Level 2 anomaly, with an anomaly confidence level of 70% ≤ anomaly confidence level < 90%, indicates that the mineralized material is significantly abnormal and shows a trend of mineralization enrichment; Level 3 anomaly, with an anomaly confidence level ≥ 90%, indicates significant anomalies in the corresponding mineralized materials, with high concentrations of mineralized enrichment, requiring immediate on-site material verification.

[0011] Preferably, the warning information of the warning output module includes text warnings, sound and light warnings, and visual warnings. The visual warnings mark the location, range, and level of abnormal material collection on a GIS map. The designated terminals include the mine monitoring center terminal and the mobile terminal of the management personnel. The location of the abnormal material collection is determined by the geographical location data collected by the positioning module.

[0012] A real-time early warning method for mineralization anomalies, applied to the above system, includes the following steps: S1. Real-time acquisition of multi-source characteristic data and corresponding geographical location data of mineralized materials in the target area of ​​the mining area. The multi-source characteristic data includes mineral composition data, element content data and mining area environmental parameter data. The mineral composition data is obtained through spectral detection, the element content data is obtained through chemical detection, and the mining area environmental parameter data is used to assist in correcting the characteristic data of mineralized materials and improve the accuracy of material analysis and anomaly identification. S2. The collected mineralization-related material characteristic data and geographical location data are subjected to noise reduction, standardization and missing value completion processing to obtain a standardized material analysis dataset, and the relevant data are stored in the data storage module; S3. Input the standardized material analysis dataset into a CNN-LSTM hybrid network containing an attention mechanism, extract the spatial and temporal features of the mineralized material data through the model, identify the abnormal features of the mineralized material based on the chemical and physical characteristics of the material, and calculate the anomaly confidence. S4. Based on the abnormal characteristics and confidence level of the mineralized materials, and in conjunction with the preset classification standards for abnormal mineralized materials, determine the abnormal level of the mineralized materials; S5. Based on the anomaly level of the mineralized material, and combined with the geographic location data, output the corresponding early warning information, the location of the abnormal material collection, and the description of the abnormal characteristics in real time, and push the early warning information to the designated terminal. S6. Based on the newly added mineralized material feature data, geographical location data and anomaly identification results stored in the data storage module, the CNN-LSTM hybrid network containing the attention mechanism is incrementally trained by combining transfer learning and incremental learning methods to dynamically optimize the material anomaly identification accuracy and achieve continuous and accurate early warning of mineralized material anomalies.

[0013] Preferably, in step S1, the collection of mineralization-related material characteristic data and geographical location data adopts a combination of timed collection and trigger-based collection. The timed collection cycle is 10-30 minutes. When the environmental sensor detects that the rate of change of environmental parameters in the mining area exceeds the preset threshold of 20% / min, the mineralization material data collection module and the positioning module are triggered to collect data in real time.

[0014] Preferably, in step S3, the specific process of mineral material anomaly identification includes: extracting spatial features from mineral spectral data and elemental content data in the standardized material analysis dataset using a CNN network to obtain a mineral material feature map; analyzing the temporal variation data of the mineral material data using an LSTM network to obtain the trend features of the mineral material; dynamically assigning weights to spatial and temporal features using an attention mechanism to highlight the weight proportion of material anomaly features; and inputting the fused features into a fully connected layer to output the mineral material anomaly features and the corresponding anomaly confidence scores.

[0015] Preferably, in step S6, the specific process of dynamic model updating includes: collecting newly added mineralized material feature data, geographical location data, and corresponding anomaly verification results from the data storage module; preprocessing the data and supplementing it to the material analysis training dataset; using the original model parameters as initial weights based on the transfer learning method, and combining the incremental learning method to incrementally train the supplemented training dataset; verifying the model performance through the validation set, wherein the recognition accuracy is evaluated by the accuracy rate. If the model accuracy rate is ≥5% higher than that of the original model, the model parameters are updated and stored in the data storage module before being put into use; otherwise, the model hyperparameters are adjusted, labeled samples are supplemented, or the loss function is optimized before continuing training.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This invention comprehensively addresses the shortcomings of traditional mineralization anomaly monitoring technologies by employing a closed-loop design encompassing multi-source data fusion acquisition, standardized preprocessing, high-precision intelligent identification, hierarchical early warning output, and dynamic model updates, covering the data layer, algorithm layer, and application layer. It utilizes simultaneous acquisition of multi-source data (spectral, elemental content, environmental parameters) combined with BeiDou + GPS dual-mode positioning to achieve precise binding of mineralization data with spatial location; a timed + triggered dual acquisition mode ensures data real-time performance and integrity, compensating for the shortcomings of traditional single-data-dimensional approaches. Through three-step preprocessing—wavelet denoising, Z-score standardization, and K-nearest neighbor interpolation—noise, dimensional differences, and data gaps are effectively eliminated, generating a high-quality standardized dataset and significantly reducing the impact of data interference on anomaly identification. An attention mechanism + CNN-LSTM hybrid network is used to simultaneously extract the spatial distribution and temporal variation features of mineralization data. The attention mechanism strengthens the weight of anomaly-sensitive features, achieving an accuracy rate ≥95% and a false alarm rate ≤3%.

[0017] (2) This invention divides the early warning into three levels according to the anomaly confidence level, and matches the handling strategies for different risk levels to avoid the waste of resources caused by unified alarms; combined with multiple forms of output such as text, sound and light, and GIS visualization, it can accurately locate abnormal areas with a response time of ≤1 second. Based on transfer learning and incremental learning, the model can be automatically iterated and updated without the need for full retraining, and the recognition accuracy can be continuously optimized; it can be dynamically adjusted according to the mineralization environment of the mining area, solving the problems of fixed parameters and poor adaptability of traditional models. It constructs a complete closed loop of data collection-preprocessing-anomaly identification-tiered early warning-verification and update, taking into account the needs of resource exploration and safe production. The hardware deployment is flexible and the software operation is simple, and it can be directly adapted to various actual monitoring scenarios in mining areas. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Example 1: Please see Figure 1 A real-time early warning system for mineralization anomalies includes: The mineralization material data acquisition module is used to collect multi-source characteristic data of mineralization-related materials (soil and rock) in the target area of ​​the mining area in real time. It adopts multi-source data fusion acquisition to make up for the shortcomings of single data dimensions. The multi-source characteristic data includes mineral composition data, element content data, and mining area environmental parameter data. Mineral composition data is obtained through spectral detection (optical analysis method), element content data is obtained through chemical detection method, and mining area environmental parameter data is used to assist in correcting the characteristic data of mineralization materials, improve the accuracy of material analysis and anomaly identification, and provide basic data for the analysis of mineralization-related materials.

[0021] Specifically, the mineralization material data acquisition module includes a spectral sensor, an elemental analyzer, an environmental sensor, and a data acquisition terminal. The spectral sensor is used to collect mineral spectral data of mineralization materials on the surface and in the shallow underground layer (0-5 meters) of the mining area (optical analysis of material characteristics), and to collect the reflectance and absorption spectra of mineralization materials in real time. The elemental analyzer is used to detect the content of target mineralization elements in soil and rock samples of the mining area (chemical analysis of material characteristics). The elemental analyzer uses an X-ray fluorescence analyzer to perform in-situ detection on soil and rock samples of the mining area and output the content data of target ore-forming elements. The environmental sensor integrates temperature, humidity, and tilt angle sensing units to collect environmental temperature, relative humidity, and topographic slope parameters of the mining area in real time. Environmental parameters are used to assist in correcting the characteristic data of mineralization materials and improve the accuracy of material analysis and anomaly identification. The data acquisition terminal is used to realize the synchronous acquisition and real-time transmission of multi-source material characteristic data. The data acquisition terminal adopts an embedded control unit, is configured with a synchronous clock and data interface, and realizes the synchronous acquisition, unified encoding, and real-time transmission of spectral data, elemental content data, and environmental parameters. The sampling frequency is adaptively adjusted according to the monitoring density of the mining area.

[0022] Spectral sensor: Detection band of 350~2500nm, used to collect reflectance and absorption spectra of mineralized materials on the surface and in the shallow underground layer of 0-5 meters in mining areas, with a spectral resolution of ≤3nm; Elemental analyzer: Employs an X-ray fluorescence analyzer, capable of detecting target ore-forming elements such as Cu, Pb, Zn, and Fe, with an elemental detection accuracy of ≤0.01% and a detection range of 0~100%. Environmental sensors: Temperature measurement range -40~85℃, error ±0.5℃; Humidity measurement range 0~100% RH, error ±3% RH; Terrain slope measurement range 0~90°, error ±1°. Data acquisition terminal: It adopts an ARM embedded control unit, supports 4G / 5G / LoRa tri-mode communication, and has an adaptive sampling frequency adjustment range of 1~5 times / minute, realizing synchronous acquisition, unified encoding and real-time transmission of multi-source data.

[0023] The positioning module is used to collect real-time geographic location data of mineralization-related material samples, determining the collection location corresponding to each material characteristic data, and providing support for spatial traceability of material analysis results. Specifically, the positioning module adopts a BeiDou + GPS dual-mode positioning unit, integrated with the data acquisition terminal. While collecting mineralization characteristic data, it simultaneously acquires the latitude, longitude coordinates and elevation information of the corresponding sampling points, binding the geographic location information with each set of mineralization characteristic data to form sample data with spatial tags. This achieves precise matching between mineralization data and spatial location, providing basic support for subsequent anomaly area location and GIS visualization, facilitating staff to quickly locate anomaly areas. The positioning module uses a BeiDou + GPS dual-mode positioning unit with a positioning accuracy of ≤1 meter, and can simultaneously acquire the latitude, longitude coordinates and elevation information of sampling points, binding the geographic location data with the mineralization characteristic data one by one.

[0024] The data storage module stores collected mineralization-related material characteristic data, geographic location data, historical material analysis data, anomaly identification results, and model parameters. This provides data support for material analysis and model optimization, enabling unified management and long-term storage of multi-source heterogeneous data. It also provides complete data support for data preprocessing, model training, and incremental updates, ensuring system data traceability and reusability. Specifically, it uses a combination of time-series and relational databases to store data. The time-series database stores real-time collected mineralization characteristic data, geographic location data, and time-series change data; the relational database stores anomaly identification results, model weight parameters, early warning records, and historical labeled data. Data is categorized and stored according to collection time, location, and type, supporting incremental writing and fast retrieval.

[0025] Time-series database: Stores real-time collected mineralization characteristics, geographical location, and time-series change data, supports millisecond-level incremental writing, and has a data retrieval response time of ≤1 second; Relational database: Stores anomaly identification results, model weights, early warning records, and historical labeled data. Data is categorized and stored by time, location, and type, and supports long-term retention and traceability.

[0026] The material data preprocessing module communicates with the mineralized material data acquisition module, positioning module, and data storage module. It is used to perform noise reduction, standardization, and missing value completion processing on the acquired mineralized material characteristic data and geographical location data to obtain a standardized material analysis dataset, ensuring the accuracy of material analysis; improving data quality, eliminating the interference of noise, dimensions, and missing values ​​on subsequent model recognition, and improving the stability and accuracy of anomaly recognition.

[0027] In this application, the material data preprocessing module includes the following steps: 1. Using wavelet denoising algorithm to remove noise interference from mineralization-related material feature data and geographical location data. Wavelet denoising: Selecting the db4 wavelet basis function, performing multi-level wavelet decomposition on the original data (performing 5-level wavelet decomposition), reconstructing the signal after removing high-frequency noise components, and removing sensor drift and environmental electromagnetic interference; 2. Mapping material data of different dimensions to the same order of magnitude using the Z-score normalization method. Z-score normalization: Mapping spectral data, elemental content data, and environmental parameters of different dimensions and numerical ranges to a standard normal distribution, eliminating dimensional differences, according to the formula Z=(x The multi-dimensional data is mapped to a standard normal distribution using μ) / σ (x is the original data, μ is the mean, and σ is the standard deviation). 3. K-nearest neighbor interpolation is used to complete missing values ​​in the data, ultimately generating a standardized material analysis dataset with a unified structure and complete data, meeting the accuracy requirements of material analysis. K-nearest neighbor interpolation: using adjacent valid sampling point data as a reference, it completes missing values ​​caused by data transmission packet loss or instantaneous sensor failure; setting K=5, it completes missing values ​​using the data of 5 adjacent valid sampling points as a reference. After processing, a standardized material analysis dataset with a unified format, no noise, and no missing values ​​is output.

[0028] The mineralized material anomaly identification module communicates with the material data preprocessing module and the data storage module. It uses a CNN-LSTM hybrid network with an attention mechanism to extract features and identify anomalies in a standardized material analysis dataset. Based on the chemical and physical characteristics of the materials, it outputs the anomaly features and anomaly confidence levels of the mineralized materials. It mines spatial distribution features and temporal variation patterns, and its feature extraction is more comprehensive than that of traditional algorithms. The attention mechanism enhances the anomaly sensitivity features, significantly reducing the false positive rate and improving the identification accuracy.

[0029] Specifically, the training process of the CNN-LSTM hybrid network incorporating an attention mechanism includes: acquiring historical mineralization material feature data, corresponding geographical location data, and anomaly annotation information to construct a material analysis training dataset; inputting the training dataset into the CNN-LSTM hybrid network, using CNN to extract spatial features of the mineralization material data and LSTM to extract temporal variation features of the mineralization material data; introducing an attention mechanism to assign weights to the extracted spatial and temporal features, prioritizing features related to mineralization material anomalies, suppressing interference from irrelevant features, and enhancing the ability to identify material anomalies; training the model using the cross-entropy loss function and gradient descent algorithm until the model converges, resulting in a trained mineralization material anomaly recognition model. Among them, (1) CNN branch: Convolution and pooling operations are performed on the spectral data and element content data in the standardized data to extract the spatial features of mineral composition and element distribution; (2) LSTM branch: Modeling the time series data of the same monitoring point to extract the trend features of mineralization data changing over time; (3) Attention layer: Automatically assigning weights to spatial features and time series features, increasing the weight of features related to mineralization anomalies, and suppressing invalid features; (4) Fully connected layer output: Mineralization anomaly feature vector and anomaly confidence in the range of 0-100%. The model adopts the cross-entropy loss function and is optimized by Adam gradient descent until the loss function converges.

[0030] Training hyperparameters for a CNN-LSTM hybrid network with attention mechanism: CNN branches: 32 / 64 / 128 convolutional kernels, 3×3 kernel size, and max pooling. LSTM branch: 128 hidden layer nodes, time step set to 10; Training parameters: batch size=32, learning rate=0.001, number of training epochs=100, loss function is cross-entropy loss, optimized by Adam gradient descent algorithm, and the loss convergence threshold is ≤0.001; Dataset partitioning: Training set: Validation set: Test set = 7:2:1.

[0031] The early warning and classification module communicates with the mineral material anomaly identification module. It is used to determine the anomaly level of mineral materials based on the anomaly characteristics and confidence level of the mineral materials, combined with the preset mineral material anomaly classification standards. This avoids redundant alarms caused by unified early warning and facilitates classified handling and reasonable allocation of resources by management personnel.

[0032] Specifically, the pre-set anomaly classification standards for mineralized materials in the early warning grading module include three levels: Level 1 (low risk), with an anomaly confidence level of 50% ≤ anomaly confidence level < 70%, indicating initial signs of anomaly in the mineralized materials, with no obvious mineralization enrichment; Level 2 (medium risk), with an anomaly confidence level of 70% ≤ anomaly confidence level < 90%, indicating significant anomaly in the mineralized materials, with a tendency for mineralization enrichment; and Level 3 (high risk), with an anomaly confidence level ≥ 90%, indicating significant anomaly in the mineralized materials, with high concentrations of mineralization enrichment, requiring immediate on-site material verification. The module automatically matches the corresponding level based on the confidence level.

[0033] The early warning output module, which communicates with the early warning classification module, the positioning module, and the data storage module, is used to output corresponding early warning information, the location of abnormal material collection, and a description of abnormal characteristics in real time according to the anomaly level of mineralized materials. Simultaneously, the early warning information is pushed to designated terminals. In this application, the early warning information from the early warning output module includes text warnings, audible and visual warnings, and visual warnings. Visual warnings use GIS maps to mark the location of abnormal material collection, the anomaly range, and the anomaly level. Designated terminals include the mine monitoring center terminal and mobile terminals of management personnel. The location of abnormal material collection is determined by the geographic location data collected by the positioning module. Multiple forms and multiple terminals simultaneously output early warning information, providing a visually intuitive display of the anomaly location, thus improving response speed and handling efficiency.

[0034] Text alert template: [Level XX Alert] Longitude XX°, Latitude XX°, XX element content abnormal, anomaly confidence level XX%; Audible and visual warnings: Level 1 warning (slow flashing green light), Level 2 warning (fast flashing yellow light), Level 3 warning (constant red light + buzzer alarm); GIS visualization: Anomalies are marked with colors according to their level (level 1: green, level 2: yellow, level 3: red), the scope of the anomaly is automatically rendered, and the data is simultaneously pushed to the monitoring center's large screen and the mobile terminals of management personnel.

[0035] The model dynamic update module connects the mineralization material anomaly identification module, mineralization material data acquisition module, and data storage module. It is used to incrementally train the model based on newly added mineralization material feature data, geographical location data, and anomaly identification results, combining transfer learning and incremental learning methods to dynamically optimize the accuracy of material anomaly identification. Specifically, it periodically collects newly added mineralization data, geographical location data, and annotation results after on-site verification; using the parameters of the previously trained model as initial weights, it employs a transfer learning + incremental learning strategy, training only the newly added data and fully connected layers, while retaining the underlying feature extraction capabilities; using classification accuracy as the evaluation metric, if the accuracy of the new version model is at least 5% higher than the original model, it is replaced and deployed; otherwise, hyperparameters such as learning rate and batch size are adjusted, or the model is retrained after supplementing labeled samples.

[0036] The model incremental training cycle is once a week, and the update process is as follows: Collect newly added mineralization data, geographic location data, and on-site verification and annotation results this week; Using the original model parameters as initial weights, transfer learning and incremental learning are used to train the new data. Using classification accuracy as an indicator, if the accuracy of the new model is ≥5% higher than that of the original model, it will be replaced and put into operation; otherwise, the learning rate will be adjusted, labeled samples will be added, and the model will be retrained.

[0037] A real-time early warning method for mineralization anomalies, applied to the above system, includes the following steps: S1. Real-time acquisition of multi-source characteristic data and corresponding geographical location data of mineralization-related materials (soil, rock) in the target area of ​​the mining area. The multi-source characteristic data includes mineral composition data, element content data, and mining area environmental parameter data. Mineral composition data is obtained through spectral detection (optical analysis), element content data is obtained through chemical detection, and mining area environmental parameter data is used to assist in correcting the characteristic data of mineralization materials and improve the accuracy of material analysis and anomaly identification. Among them, the acquisition of mineralization-related material characteristic data and geographical location data adopts a combination of timed acquisition and trigger-based acquisition. The timed acquisition cycle is 10-30 minutes. When the environmental sensor detects that the rate of change of the mining area environmental parameters exceeds the preset threshold (20% / min), the mineralization material data acquisition module and the positioning module are triggered to carry out real-time acquisition to ensure the real-time and comprehensiveness of material data acquisition.

[0038] S2. The collected mineralization-related material characteristic data and geographical location data are subjected to noise reduction, standardization and missing value completion processing to obtain a standardized material analysis dataset, and the relevant data are stored in the data storage module; S3. Input the standardized materials analysis dataset into a CNN-LSTM hybrid network with an attention mechanism. The model extracts spatial and temporal features of the mineralized material data, identifies anomalous features of the mineralized materials based on their chemical and physical characteristics, and calculates the anomalous confidence scores. The specific process of mineralized material anomaly identification includes: extracting spatial features from the mineral spectral data and elemental content data (material chemical and physical characteristics) in the standardized materials analysis dataset using a CNN network to obtain mineralized material feature maps; analyzing the temporal variation data of the mineralized material data using an LSTM network to obtain the trend features of the mineralized materials; dynamically assigning weights to the spatial and temporal features using an attention mechanism to highlight the weight proportion of anomalous material features; and inputting the fused features into a fully connected layer to output the anomalous features of the mineralized materials and their corresponding anomalous confidence scores.

[0039] S4. Based on the abnormal characteristics and confidence level of the mineralized materials, and in conjunction with the preset classification standards for abnormal mineralized materials, determine the abnormal level of the mineralized materials; S5. Based on the anomaly level of the mineralized material, and combined with the geographic location data, output the corresponding early warning information, the location of the abnormal material collection, and the description of the abnormal characteristics in real time, and push the early warning information to the designated terminal. S6. Based on the newly added mineralized material feature data, geographical location data and anomaly identification results stored in the data storage module, the CNN-LSTM hybrid network containing the attention mechanism is incrementally trained by combining transfer learning and incremental learning methods to dynamically optimize the material anomaly identification accuracy and achieve continuous and accurate early warning of mineralized material anomalies.

[0040] The specific process of dynamic model updates includes: collecting newly added mineralized material feature data, geographical location data, and corresponding anomaly verification results from the data storage module; preprocessing the data and adding it to the material analysis training dataset; using the original model parameters as initial weights based on transfer learning, and combining incremental learning methods to incrementally train the supplemented training dataset; verifying model performance through the validation set, with accuracy as the evaluation metric. If the model accuracy is ≥5% higher than the original model, the model parameters are updated and stored in the data storage module before use; otherwise, the model hyperparameters are adjusted, labeled samples are added, or the loss function is optimized before continuing training.

[0041] Mineralization Anomaly Verification - Updated Closed-Loop Implementation Process: Level 3 warning triggered → Management personnel receive information → Go to the site for sampling and verification; Verification results and measured data are entered into the data storage module; The model dynamic update module retrieves labeled data to complete incremental training; The optimized model improves recognition accuracy and forms a complete closed loop of data collection, recognition, early warning, verification, and update.

[0042] Example 2 This embodiment provides a real-time early warning method for mineralization anomalies, applied to the system described in Embodiment 1. The specific steps are as follows: S1. Data Acquisition: A combination of 20-minute timed acquisition and trigger-based acquisition is adopted. When the rate of change of environmental parameters is ≥20% / min, acquisition is triggered immediately; spectral data, elemental content, environmental parameters, and location data are acquired simultaneously. S2. Data preprocessing: Perform wavelet denoising, Z-score normalization, and K-nearest neighbor interpolation as described in Example 1 to generate a standardized dataset; S3. Anomaly Detection: Input the dataset into the CNN-LSTM + Attention model. The CNN extracts spatial features, the LSTM extracts temporal features, and the attention mechanism outputs the anomaly features and confidence scores after weighting. S4. Level Determination: Match the anomaly level according to the three-level standard; S5. Early Warning Output: Output text, audio-visual, and GIS visual early warnings according to their severity levels and push them to designated terminals; S6. Model Update: Dynamically optimize model accuracy according to the incremental training rules described in Example 1.

[0043] Example 3 This embodiment provides a complete example of on-site operation in a mining area: Select monitoring point #10 in a certain metal mining area and execute the following procedure: Data collection: Spectral data of this point were collected every 20 minutes, with Fe element content of 6.2%, ambient temperature of 25℃, and slope of 15°. The location coordinates are 116°E, 39°N; Preprocessing: After noise reduction, the data is free of noise and missing data, and is then standardized before being input into the model; Identification: The model outputs abnormal features of the Fe element, with an anomaly confidence level of 93%. Grading: Determined as Level III abnormal; Warning: The point is marked as a red warning zone on the GIS map, a text warning is pushed to the management personnel's terminal, and an audible and visual alarm is triggered; Verification: On-site sampling and verification confirmed high concentration of Fe element enrichment, and the verification results were entered into the system; Update: The labeled data has been incorporated into the incremental training set, improving the model accuracy by 6%. Model update complete.

[0044] Example 4 This embodiment provides system deployment, debugging, and maintenance specifications: Hardware deployment: Sensors are deployed in a 50m x 50m grid. Surface sensors are installed in the open air, and shallow underground sensors are buried at a depth of 0.5 to 5 meters, with proper waterproofing and dustproofing measures. System debugging: Complete sensor calibration, data transmission testing, model pre-training, and early warning function integration testing in sequence to ensure a data transmission success rate of ≥99%; Fault handling: Automatic alarm for sensor failure, automatic reconnection for data interruption, and automatic rollback to a previous version for model anomalies; Routine maintenance: Daily data backup, monthly equipment calibration, and quarterly full model training and optimization.

[0045] Example 5 The performance comparison data between the present invention and traditional technologies in this embodiment is shown in Table 1 below: Table 1 Data Comparison

[0046] In comparison, the system of the present invention is significantly superior to traditional technologies in terms of recognition accuracy, response speed, and adaptability.

[0047] Example 6 (1) Implementation of early warning for mineralization anomalies in non-metallic mining areas (quartz / kaolin) In a selected monitoring area #2 of a quartz mine, system parameters were optimized based on non-metallic mineralization characteristics. Data acquisition: The spectral sensor focuses on the 2000~2500nm quartz characteristic band, and the elemental analyzer detects the Si and Al content. The timed acquisition cycle is set to 15 minutes. Preprocessing: To address the spectral noise characteristics of non-metallic minerals, the wavelet decomposition layer was adjusted to 3 layers to improve data purity; Anomaly identification: The model output shows an anomaly confidence level of 82% for the Si element, classifying it as a level 2 anomaly; Early warning output: Yellow warning areas are marked on the GIS map and pushed to the quality inspection terminal in the mining area; Results: The accuracy rate of non-metallic mineralization anomaly identification reached 94%, meeting the exploration needs of non-metallic mining areas.

[0048] (2) Deploy the system in a mining area at an altitude of 3800m and a low temperature of -15℃ to verify its adaptability to extreme environments: Hardware adaptation: The sensor is set to low-temperature compensation mode, and the data acquisition terminal is heated and kept warm. Data collection strategy: The threshold for the rate of change of environmental parameters was adjusted to 15% / min to adapt to sudden changes in the high-altitude environment; Data processing: The preprocessing module enhances low-temperature noise filtering, maintaining a data effectiveness rate of 98.7%. Recognition performance: The anomaly recognition accuracy remains ≥93% even in extreme environments, and the system operates stably.

[0049] Example 7 1000 newly labeled datasets were selected for model update validation: The original model achieved an accuracy of 92.3%, while the accuracy after incremental training reached 97.8%, representing an improvement of 5.5%. The new model is launched once the "improvement rate ≥ 5%" update rule is met. After four consecutive incremental updates, the model's ability to identify rare mineralization anomalies was significantly enhanced. Verification: Incremental training takes only 1 / 10 of the time of full training, significantly improving efficiency.

[0050] Example 8 Using a portable data collection terminal (72-hour battery life) for emergency field monitoring: Deployment: Manually carried to remote mining areas without power supply for rapid deployment at a single point; Data Acquisition: Primarily trigger-based acquisition, with a timing period set to 30 minutes to reduce power consumption; Data transmission: 4G / 5G backup mode, offline data caching, and automatic data upload upon network connection; Application: Successfully discovered two hidden mineralization anomalies, suitable for emergency field exploration scenarios.

[0051] 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 real-time early warning system for mineralization anomalies, characterized in that, include: The mineralization material data acquisition module is used to collect multi-source characteristic data of mineralization-related materials in the target area of ​​the mining area in real time. The multi-source characteristic data includes mineral composition data, element content data and mining area environmental parameter data. The mineral composition data is obtained through spectral detection and the element content data is obtained through chemical detection methods. The positioning module is used to collect geographical location data of mineralization-related material samples in real time and determine the collection location corresponding to the characteristic data of each material. The data storage module is used to store the collected mineralization-related material characteristic data, geographical location data, historical material analysis data, anomaly identification results, and model parameters; The material data preprocessing module is communicatively connected to the mineralized material data acquisition module, the positioning module, and the data storage module. It is used to perform noise reduction, standardization, and missing value completion processing on the acquired mineralized material characteristic data and geographical location data to obtain a standardized material analysis dataset. The mineralized material anomaly identification module is communicatively connected to the material data preprocessing module and the data storage module. It uses a CNN-LSTM hybrid network with an attention mechanism to extract features and identify anomalies in the standardized material analysis dataset, and outputs the mineralized material anomaly features and anomaly confidence. The early warning and classification module is communicatively connected to the mineralized material anomaly identification module. It is used to determine the anomaly level of the mineralized material based on the anomaly characteristics and anomaly confidence level of the mineralized material, combined with the preset mineralized material anomaly classification standard. The early warning output module is communicatively connected to the early warning classification module, the positioning module, and the data storage module. It is used to output corresponding early warning information, abnormal material collection location, and abnormal feature description in real time according to the abnormality level of the mineralized material, and push the early warning information to the designated terminal. The model dynamic update module is communicatively connected to the mineralized material anomaly identification module, the mineralized material data acquisition module, and the data storage module. It is used to perform incremental training of the model based on newly added mineralized material feature data, geographical location data, and anomaly identification results, combined with transfer learning and incremental learning methods, to optimize the accuracy of material anomaly identification.

2. The real-time early warning system for mineralization anomalies according to claim 1, characterized in that: The mineralization material data acquisition module includes a spectral sensor, an elemental analyzer, an environmental sensor, and a data acquisition terminal. The spectral sensor is used to collect mineral spectral data of mineralization materials on the surface and in the shallow underground layer (0-5 meters) of the mining area. The elemental analyzer is used to detect the content of target mineralization elements in soil and rock samples from the mining area. The environmental sensor is used to collect environmental parameters such as temperature, humidity, and topographic slope of the mining area. These environmental parameters are used to assist in correcting the characteristic data of the mineralization materials. The data acquisition terminal is used to realize the synchronous acquisition and real-time transmission of characteristic data of multiple materials.

3. The real-time early warning system for mineralization anomalies according to claim 1, characterized in that: The processing steps of the material data preprocessing module include: using wavelet denoising algorithm to remove noise interference from mineralization-related material feature data and geographical location data; using Z-score normalization method to map material data of different dimensions to the same order of magnitude; using K-nearest neighbor interpolation method to complete missing values ​​in the data; and finally generating a standardized material analysis dataset with unified structure and complete data.

4. The real-time early warning system for mineralization anomalies according to claim 1, characterized in that: The training process of the CNN-LSTM hybrid network with an attention mechanism includes: acquiring historical mineralization material feature data, corresponding geographical location data, and anomaly annotation information to construct a material analysis training dataset; inputting the training dataset into the CNN-LSTM hybrid network, extracting spatial features of the mineralization material data through CNN, and extracting temporal variation features of the mineralization material data through LSTM; introducing an attention mechanism to assign weights to the extracted spatial and temporal features, prioritizing features related to mineralization material anomalies, and suppressing interference from irrelevant features; training the model using the cross-entropy loss function and gradient descent algorithm until the model converges, thus obtaining the trained mineralization material anomaly recognition model.

5. The real-time early warning system for mineralization anomalies according to claim 1, characterized in that: The pre-set abnormal classification standard for mineralized materials in the early warning classification module includes three levels: Level 1 anomaly, with an anomaly confidence level of 50% ≤ anomaly confidence level < 70%, indicating that the anomaly in the corresponding mineralized material is initially apparent, with no obvious mineralization enrichment; Level 2 anomaly, with an anomaly confidence level of 70% ≤ anomaly confidence level < 90%, indicates that the mineralized material is significantly abnormal and shows a trend of mineralization enrichment; Level 3 anomaly, with an anomaly confidence level ≥ 90%, indicates significant anomalies in the corresponding mineralized materials, with high concentrations of mineralized enrichment, requiring immediate on-site material verification.

6. The real-time early warning system for mineralization anomalies according to claim 1, characterized in that: The warning information from the warning output module includes text warnings, audible and visual warnings, and visual warnings. The visual warnings use GIS maps to mark the location, range, and level of abnormal material collection. The designated terminals include mine monitoring center terminals and mobile terminals for management personnel. The location of abnormal material collection is determined by the geographic location data collected by the positioning module.

7. A real-time early warning method for mineralization anomalies, applied to the system described in any one of claims 1-6, characterized in that, Includes the following steps: S1. Real-time acquisition of multi-source characteristic data and corresponding geographical location data of mineralized materials in the target area of ​​the mining area. The multi-source characteristic data includes mineral composition data, element content data and mining area environmental parameter data. The mineral composition data is obtained through spectral detection, the element content data is obtained through chemical detection, and the mining area environmental parameter data is used to assist in correcting the characteristic data of mineralized materials and improve the accuracy of material analysis and anomaly identification. S2. The collected mineralization-related material characteristic data and geographical location data are subjected to noise reduction, standardization and missing value completion processing to obtain a standardized material analysis dataset, and the relevant data are stored in the data storage module; S3. Input the standardized material analysis dataset into a CNN-LSTM hybrid network containing an attention mechanism, extract the spatial and temporal features of the mineralized material data through the model, identify the abnormal features of the mineralized material based on the chemical and physical characteristics of the material, and calculate the anomaly confidence. S4. Based on the abnormal characteristics and confidence level of the mineralized materials, and in conjunction with the preset classification standards for abnormal mineralized materials, determine the abnormal level of the mineralized materials; S5. Based on the anomaly level of the mineralized material, and combined with the geographic location data, output the corresponding early warning information, the location of the abnormal material collection, and the description of the abnormal characteristics in real time, and push the early warning information to the designated terminal. S6. Based on the newly added mineralized material feature data, geographical location data and anomaly identification results stored in the data storage module, the CNN-LSTM hybrid network containing the attention mechanism is incrementally trained by combining transfer learning and incremental learning methods to dynamically optimize the material anomaly identification accuracy and achieve continuous and accurate early warning of mineralized material anomalies.

8. The real-time early warning method for mineralization anomalies according to claim 7, characterized in that: In step S1, the collection of mineralization-related material characteristic data and geographical location data adopts a combination of timed collection and trigger-based collection. The timed collection cycle is 10-30 minutes. When the environmental sensor detects that the rate of change of environmental parameters in the mining area exceeds the preset threshold of 20% / min, the mineralization material data collection module and the positioning module are triggered to collect data in real time.

9. The real-time early warning method for mineralization anomalies according to claim 7, characterized in that: In step S3, the specific process of mineral material anomaly identification includes: extracting spatial features from mineral spectral data and elemental content data in the standardized material analysis dataset using a CNN network to obtain mineral material feature maps; analyzing the temporal variation data of mineral material data using an LSTM network to obtain mineral material trend features; dynamically assigning weights to spatial and temporal features using an attention mechanism to highlight the weight proportion of material anomaly features; and inputting the fused features into a fully connected layer to output mineral material anomaly features and corresponding anomaly confidence scores.

10. The real-time early warning method for mineralization anomalies according to claim 7, characterized in that: In step S6, the specific process of dynamic model update includes: collecting newly added mineralized material feature data, geographical location data, and corresponding anomaly verification results from the data storage module; preprocessing the data and supplementing it to the material analysis training dataset; using the original model parameters as initial weights based on the transfer learning method, and combining the incremental learning method to incrementally train the supplemented training dataset; verifying the model performance through the validation set, where the recognition accuracy is evaluated by the accuracy rate. If the model accuracy rate is ≥5% higher than the original model, the model parameters are updated and stored in the data storage module before being put into use; otherwise, the model hyperparameters are adjusted, labeled samples are supplemented, or the loss function is optimized before continuing training.

Images