Deep metal mine multi-source geophysical data real-time fusion system

Abstract

The invention relates to the technical field of deep mineral exploration, and discloses a deep metal mine multi-source geophysical data real-time fusion system. The system comprises a data access module which divides detection dimensions according to physical field types so as to receive original data; the space-time registration module is used for establishing mapping between original data and a unified space-time reference system through timestamps and space coordinates; the feature extraction module is used for extracting key features from the registration data, determining weights according to physical meanings and constructing a feature association network; the fusion decision module identifies an abnormal value and adjusts a network topology structure based on the credibility of the network evaluation data; and the real-time updating module is used for optimizing the fusion priority and algorithm parameters and generating a real-time fusion result library. According to the system, real-time fusion of multi-source geophysical data is realized, and systematicness and timeliness of data processing are improved.

Description

Technical Field

[0001] This invention relates to the field of deep mineral exploration technology, specifically a real-time fusion system for multi-source geophysical data of deep metal mines. Background Technology

[0002] Deep metal deposits are characterized by complex burial conditions and varied geological structures, necessitating multi-source geophysical exploration. Different types of geophysical equipment, such as gravimeters, magnetometers, and electromagnetic detectors, can collect data on various physical fields, including gravity, magnetic fields, and electromagnetic fields. These data reflect the distribution characteristics of underground mineral resources from different perspectives. However, processing this multi-source data still faces numerous challenges.

[0003] The different principles and parameters of various detection devices result in variations in the format, accuracy, and resolution of the raw data collected. Existing data processing methods are mostly single-dimensional or simply superimposed, failing to systematically categorize data dimensions according to physical field type. This leads to a mixture of data from different sources, making it difficult to form an orderly processing system. Furthermore, the operating time and spatial positioning benchmarks of different detection devices differ, and the lack of a unified reference for data timestamps and spatial coordinates makes it difficult to directly correlate data from different time periods and locations, affecting horizontal comparisons and comprehensive analysis.

[0004] In the feature extraction stage, existing methods often ignore the physical meaning relationships between features, relying solely on mathematical methods for extraction. This results in extracted features lacking geophysical connotations and failing to reflect the intrinsic connection between mineral resources and the surrounding geological environment. Furthermore, the setting of feature weights often depends on experience, without dynamic adjustment based on the physical meaning of the features, leading to insufficient structural stability of the feature association network.

[0005] During the fusion process, the lack of a data reliability assessment mechanism and the failure to effectively identify outliers make the fusion results susceptible to noise interference. Furthermore, the fixed parameters and data priorities of the fusion algorithm prevent real-time optimization based on data changes during exploration, causing the fusion results to lag behind actual exploration needs and failing to meet the timeliness and accuracy requirements for data processing in deep metal mineral exploration. Summary of the Invention

[0006] The purpose of this invention is to provide a real-time fusion system for multi-source geophysical data of deep metal mines to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a real-time fusion system for multi-source geophysical data of deep metal deposits, the system comprising:

[0008] The data access module is used to receive raw data sent by different types of geophysical exploration equipment and divide the data into exploration dimensions according to the physical field type of the data acquisition.

[0009] The spatiotemporal registration module is used to obtain the timestamps and spatial coordinates of each detection dimension and establish the mapping relationship between the original data and the unified spatiotemporal reference system.

[0010] The feature extraction module is used to extract key features from registered multi-source data, determine feature weights according to the physical meaning of the features, and construct a feature association network based on the feature weights.

[0011] The fusion decision module is used to perform credibility assessment on multi-source data based on the feature association network, identify outliers in the data, and adjust the topology of the feature association network according to the credibility assessment results.

[0012] The real-time update module is used to optimize the priority and fusion algorithm parameters of data fusion based on the adjusted feature association network, and generate a real-time fusion result library.

[0013] Preferably, the data access module can be implemented in the following ways:

[0014] For any raw data sent by any detection device, invoke the data format parsing protocol;

[0015] Use a parsing protocol to decode the original data, at least to obtain the physical field type and numerical sequence of the data;

[0016] Identify noise signals, missing values, and anomalous jumps in the original data under the corresponding physical field type to form a data sample set;

[0017] The noise signal, missing value, and anomalous transition in the data sample set are processed separately to generate noise filtering results, missing value imputation results, and transition correction results in sequence, which are used as preprocessed data corresponding to the original data.

[0018] Preferably, the implementation methods for generating noise filtering results, missing value imputation results, and jump correction results also include:

[0019] Noise signals, missing values, and anomalous jumps in the data sample set are classified according to physical field type, resulting in multiple types of processing units;

[0020] An adaptive filtering algorithm is applied to the noise signal in the type processing unit to obtain the noise filtering result;

[0021] The temporal distribution density of missing values ​​and the amplitude of anomalous jumps in the type processing unit are calculated. The type processing unit is then corrected according to the temporal distribution density and amplitude values ​​to obtain the missing value filling result and jump correction result.

[0022] Preferably, the implementation methods for dividing the data into corresponding detection dimensions also include:

[0023] The detection dimensions are validated by analyzing the sampling frequency and coverage of the data in the detection dimensions. The detection dimensions are then correlated according to their sampling frequency and coverage to construct a mapping relationship between the detection dimensions and the data credibility.

[0024] Preferably, the spatiotemporal registration module can be implemented in the following ways:

[0025] Retrieve the timestamp calibration parameters and spatial reference transformation model corresponding to the original data, and generate multiple initial registration results. The initial registration results represent spatiotemporal coordinate data without error correction.

[0026] Determine whether multiple initial registration results meet the preset registration accuracy threshold. If they meet the registration accuracy threshold, the initial registration results are considered as registered multi-source data.

[0027] Preferably, the method for establishing the mapping relationship between the original data and the unified spatiotemporal reference system includes: using the data characteristics of the noise filtering results, missing value imputation results and jump correction results present in the preprocessed data, the coordinate parameters and time base of the unified spatiotemporal reference system, to construct a mapping relationship between the original data and the unified spatiotemporal reference system.

[0028] Preferably, the feature extraction module is implemented in the following ways:

[0029] The registered multi-source data is processed into layers according to physical field type, data precision, and acquisition depth. The layer with the highest data density after layer processing is set as the key data layer.

[0030] Extract feature vectors from key data layers, calculate the Euclidean distance between each feature vector, and establish a feature sequence related to the Euclidean distance between each feature vector;

[0031] By utilizing the feature sequences related to the Euclidean distance between each feature vector, features with significant physical meaning in the feature sequences are screened, and the correlation strength between each feature is determined. The numerical value of the correlation strength is set as the weight coefficient between each feature.

[0032] The weighting coefficients between each feature are set according to the spatial distribution range of each feature;

[0033] During the layered processing, the data is divided into gravity field data layer, magnetic field data layer, electric field data layer and seismic wave data layer according to the physical field type. Each layer of data is further divided into surface layer, shallow middle layer and deep bottom layer according to the acquisition depth. The selection of key data layers is determined based on the signal-to-noise ratio and data volume of each layer.

[0034] Preferably, the implementation methods of the fusion decision module include:

[0035] Extract the spatial distribution range of each feature from the feature weights; set the influence boundary of the feature weights according to the region corresponding to the spatial distribution range of each feature;

[0036] The influence boundaries of each feature in the feature weights are superimposed to obtain the superimposed influence boundary. The coverage of the superimposed influence boundary in each region is set as the regional contribution of the feature weight.

[0037] The regional contribution of feature weights is compared with the distribution of outliers in multi-source data to mark the amount of deviation. Outliers in multi-source data are then removed according to the amount of deviation, thus completing the identification of outliers.

[0038] The preferred implementation method for the real-time update module is as follows:

[0039] Based on the adjusted feature association network, the fusion requirements of multi-source data are extracted. The fusion requirements are sorted according to their urgency to obtain the data fusion priority. The fusion priority and fusion algorithm parameters are integrated according to the modular design to obtain a real-time fusion result library.

[0040] Preferably, the adaptive filtering algorithm includes a noise decomposition step based on wavelet transform and a noise suppression step based on a threshold function. The noise decomposition step is used to decompose the noise signal into wavelet coefficients of different frequencies, and the noise suppression step is used to apply soft thresholding to the high-frequency wavelet coefficients to preserve the waveform characteristics of the effective signal.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] This system, through the collaborative operation of multiple modules, provides a more efficient solution for processing multi-source geophysical data from deep metal deposits. The data access module divides the detection dimensions according to physical field types, enabling the classification of different types of raw data based on their physical nature, thus avoiding processing chaos caused by mixed data. This dimensional division based on physical fields allows subsequent modules to perform targeted processing on data with similar physical properties, reducing interference between different types of data and making the starting point of data processing more systematic.

[0043] The spatiotemporal registration module establishes a mapping between the original data and a unified spatiotemporal reference system by using timestamps and spatial coordinates, breaking down the barriers between different detection devices in terms of time and spatial reference. Data that were originally scattered across different spatiotemporal dimensions can now be placed under the same reference system through this mapping, making it possible to directly compare and correlate data collected at different times and locations, thus clearing away spatiotemporal obstacles for the deep fusion of multi-source data.

[0044] The feature extraction module extracts key features from the registered data and constructs a feature association network based on weights determined by physical meaning, making the relationships between features more consistent with geological and physical laws. This weight setting based on physical meaning avoids biases caused by empiricism, enabling the feature association network to truly reflect the intrinsic relationships between different features in mineral exploration. This makes the network structure more closely resemble actual geological scenarios, and clearly presents the interactions and influences between features.

[0045] The fusion decision module uses a feature association network as a foundation to conduct credibility assessment and outlier identification, and adjusts the network topology accordingly to make the fusion process more adaptive. Credibility assessment distinguishes the reliability of data, reducing the impact of unreliable data on the fusion results; outlier identification prevents extreme data from interfering with the overall trend. The dynamic adjustment of the network topology allows the feature association network to optimize as data quality and distribution change, ensuring that the fusion logic always remains consistent with the actual data characteristics.

[0046] The real-time update module generates a real-time fusion result library based on the adjusted network optimization fusion priority and algorithm parameters, ensuring that data fusion keeps pace with the exploration process. Dynamic priority optimization ensures that critical data receives priority processing, while real-time adjustment of algorithm parameters allows the fusion method to adapt to constantly changing data characteristics. The resulting real-time result library promptly reflects the overall situation of current exploration data, meeting the timeliness requirements for data processing in deep metal mineral exploration and enabling exploration personnel to conduct analysis and judgment based on the latest fusion results.

[0047] The synergistic effect of each module allows the entire process of multi-source geophysical data from access to fusion to form an organic whole. This not only ensures the systematic nature of data processing but also improves the relevance and timeliness of the fusion results, enabling it to better meet the exploration data processing needs under complex geological conditions in deep metal mines. Attached Figure Description

[0048] Figure 1 This is a schematic diagram illustrating the working principle of the real-time fusion system for multi-source geophysical data of deep metal mines as described in this invention.

[0049] Figure 2 This is a flowchart of the data access module;

[0050] Figure 3 A flowchart for verifying the detection dimension;

[0051] Figure 4 This is a flowchart of the feature extraction module. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Please see Figure 1 This invention provides a real-time fusion system for multi-source geophysical data of deep metal deposits. The system includes: a data access module, a spatiotemporal registration module, a feature extraction module, a fusion decision module, and a real-time update module. The specific implementation steps are as follows:

[0054] The data access module receives raw data from different types of geophysical exploration equipment and divides the data into corresponding exploration dimensions according to the physical field type of the data acquisition. The spatiotemporal registration module obtains the timestamps and spatial coordinates of each exploration dimension and establishes a mapping relationship between the raw data and a unified spatiotemporal reference system. The feature extraction module extracts key features from the registered multi-source data, determines feature weights according to the physical meaning of the features, and constructs a feature association network based on the feature weights. The fusion decision module uses the feature association network as a basis to perform credibility assessment on the multi-source data, identifies outliers in the data, and adjusts the topology of the feature association network according to the credibility assessment results. The real-time update module optimizes the data fusion priority and fusion algorithm parameters based on the adjusted feature association network and generates a real-time fusion result library.

[0055] Example 1: See Figure 2 When the data access module receives raw data from any detection device, it first invokes a preset data format parsing protocol. These parsing protocols are pre-configured for the raw data formats generated by different types of geophysical detection devices, and can adapt to various data transmission protocols and encoding methods. After invoking the parsing protocol, the raw data is decoded. In this process, at least the physical field type and numerical sequence of the data are extracted. The physical field type includes gravity field, magnetic field, electric field, seismic wave field, etc., while the numerical sequence is the set of specific measurement values ​​that vary with time or space under the corresponding physical field type.

[0056] After decoding, the raw data is further analyzed to identify noise signals, missing values, and anomalous jumps under the corresponding physical field types. This raw data, containing various data issues, is then integrated to form a data sample set. Noise signals manifest as irregular fluctuations in the data, with frequency components differing from the valid signal; missing values ​​refer to values ​​not recorded at certain time points or spatial locations in the data sequence; anomalous jumps are characterized by large numerical variations between adjacent data points that deviate from normal patterns.

[0057] Before processing noise signals, missing values, and anomalous jumps in the data sample set individually, these data issues are categorized according to physical field type, forming multiple processing units. Data under the same physical field type often have similar characteristics and interference patterns; after categorization, more suitable processing methods can be adopted for the characteristics of different physical fields.

[0058] For noise signals in the type processing unit, an adaptive filtering algorithm is used. The adaptive filtering algorithm includes a noise decomposition step based on wavelet transform and a noise suppression step based on a threshold function. In the noise decomposition step, the noise signal is decomposed into wavelet coefficients of different frequencies. Through multi-scale decomposition, the high-frequency and low-frequency components of the signal are separated. The high-frequency components often contain more noise information, while the low-frequency components correspond more to the trends of the effective signal. In the noise suppression step, a soft threshold is applied to the high-frequency wavelet coefficients obtained from the decomposition. That is, when the absolute value of a high-frequency wavelet coefficient is less than a set threshold, it is set to zero; when the absolute value is greater than or equal to the threshold, the threshold is subtracted from the coefficient. Through this processing, noise can be suppressed while preserving the waveform characteristics of the effective signal, avoiding the loss of effective information due to over-filtering, and finally obtaining the noise filtering result.

[0059] For missing values ​​in the type processing unit, its temporal distribution density is calculated. Temporal distribution density refers to the frequency or proportion of missing values ​​occurring within a unit of time. By statistically analyzing the distribution of missing values ​​across different time periods, the degree of concentration or dispersion of missing values ​​can be understood. For anomalous jumps, their amplitude values ​​are calculated, i.e., the numerical difference between data points before and after the jump, to measure the severity of the jump. Based on the calculated temporal distribution density and amplitude values, the type processing unit is corrected. For missing values, based on their temporal distribution density, if the missing values ​​are relatively concentrated, they can be filled using the mean or trend fitting of adjacent time periods; if the missing values ​​are relatively dispersed, interpolation can be used to fill them, thus obtaining the missing value filling result. For anomalous jumps, based on the magnitude of their amplitude values ​​and combined with the normal variation range of data under this physical field type, the values ​​of the jump points are adjusted to conform to the variation trend of surrounding data, obtaining the jump correction result.

[0060] After the above processing, the noise filtering results, missing value imputation results, and jump correction results together constitute the preprocessed data corresponding to the original data. Compared with the original data, this preprocessed data reduces noise interference, supplements missing information, and corrects abnormal fluctuations, providing a more regular and reliable data foundation for subsequent data processing stages. Throughout the process, the data access module continuously receives data from different detection devices and processes it in real time according to the above process, ensuring that various types of geophysical data can enter the next processing stage of the system in a unified preprocessed form. The processing units for different physical field types work independently yet collaboratively during operation, each completing its corresponding data processing tasks, together forming the complete processing flow of the data access module.

[0061] Example 2: See Figure 3 When dividing the data into corresponding detection dimensions, the data access module verifies these dimensions. During verification, the sampling frequency and coverage area of ​​the data within each detection dimension are analyzed first. Sampling frequency refers to the number of times data is collected per unit time. Different detection devices have different sampling frequencies due to differences in their principles and applications. Some devices collect data at a higher frequency to capture rapidly changing physical field characteristics, while others collect data at a lower frequency over longer periods to reflect long-term trends in the physical field. Coverage area refers to the spatial region involved in data collection, which may be a specific depth range, a planar area, or a portion of three-dimensional space.

[0062] The detection dimensions are correlated according to sampling frequency and coverage to construct a mapping relationship between detection dimensions and data reliability. Sampling frequency is related to the data's ability to characterize the details of the physical field; under the same conditions, data whose sampling frequency matches the rate of change of the physical field reflects information that is more consistent with reality. Coverage relates to the spatial scale that the data can reflect; when the coverage matches the spatial distribution characteristics of the detected target, the reference value of the data changes accordingly. By comprehensively evaluating the adaptability of the sampling frequency and the fit of the coverage, reliability indicators corresponding to different detection dimensions are formed. These indicators are used to characterize the reference value of the data in that dimension during the overall fusion process.

[0063] When processing data, the spatiotemporal registration module retrieves the timestamp calibration parameters and spatial reference transformation model corresponding to the original data. The timestamp calibration parameters are determined by the internal clock characteristics of the detection equipment. Different equipment has different clock accuracies, which may cause drift. The calibration parameters are used to correct these drifts, ensuring that the time records of all equipment maintain a consistent reference. The spatial reference transformation model is used to unify the spatial coordinate systems used by different detection equipment under the same reference frame. Different equipment may acquire data based on different coordinate systems; some may use local coordinate systems, while others may use national or internationally accepted coordinate systems. The transformation model achieves the conversion between different coordinate systems through coordinate transformation parameters.

[0064] The raw data is processed using timestamp calibration parameters and a spatial reference transformation model to generate multiple initial registration results. These results are uncorrected spatiotemporal coordinate data, containing the corresponding time points and spatial location information. Then, it is determined whether these initial registration results meet a preset registration accuracy threshold. The registration accuracy threshold is set according to the specific requirements of the detection mission, involving accuracy standards in both time and space dimensions. In the time dimension, the timestamp deviation of data from different devices must be controlled within a certain range; in the spatial dimension, the deviation between the transformed spatial coordinates and the target reference system must not exceed a set value.

[0065] If the initial registration result meets the registration accuracy threshold, it is considered as registered multi-source data. If the initial registration result does not meet the threshold, the applicability of the timestamp calibration parameters and spatial reference transformation model is rechecked. If necessary, the parameters or model are adjusted and the registration calculation is performed again until a registration result that meets the accuracy requirements is obtained. Through this process, data originally from different devices and based on different spatiotemporal references can be integrated within a unified spatiotemporal framework. The temporal and spatial correspondences of data from different detection dimensions are clarified, and subsequent feature extraction and fusion decisions can be based on this data with clear spatiotemporal relationships.

[0066] Example 3: See Figure 4 When establishing the mapping relationship between the original data and the unified spatiotemporal reference system, the spatiotemporal registration module relies on the data characteristics of the preprocessed data, including noise filtering, missing value imputation, and jump correction results, and operates in conjunction with the coordinate parameters and time base of the unified spatiotemporal reference system. Noise filtering removes irregular interference from the data, making the overall trend and key features clearer; missing value imputation fills in gaps in the data sequence, ensuring the continuity of the data in time and space; and jump correction eliminates abrupt changes that do not conform to normal variation patterns, making the data change process smoother. These data characteristics presented by the preprocessed data, including the magnitude, frequency, and spatial distribution characteristics of data changes, provide the basic information for establishing the mapping relationship. The coordinate parameters of the unified spatiotemporal reference system include the reference origin, coordinate axis directions, and coordinate units, while the time base includes the time origin and time units. By mapping the temporal and spatial information of the preprocessed data to these parameters, the mapping between the original data and the unified spatiotemporal reference system is achieved.

[0067] When processing registered multi-source data, the feature extraction module performs layered processing based on physical field type, data precision, and acquisition depth. The physical field type is divided according to the geophysical field properties reflected in the data, specifically into a gravity field data layer, a magnetic field data layer, an electric field data layer, and a seismic wave data layer. The gravity field data layer contains data related to changes in the Earth's gravity field; the magnetic field data layer covers data such as geomagnetic field strength and direction; the electric field data layer involves information on the electric field distribution of the subsurface medium; and the seismic wave data layer includes data such as seismic wave propagation velocity and amplitude. Each layer is further divided into surface layer, shallow-middle layer, and deep-bottom layer based on acquisition depth. The surface layer corresponds to data collected near the Earth's surface; the shallow-middle layer involves data within a certain depth range; and the deep-bottom layer contains data from deeper underground locations. Data precision, as another criterion for layering, reflects the degree to which measured values ​​approximate true values; data with different precision are assigned to different sublayers.

[0068] After layering, the layer with the highest data density is designated as the key data layer. Data density is represented by the number of data points per unit time or unit space; higher data density means more data information is contained within the same area. The selection of the key data layer is also determined based on the signal-to-noise ratio (SNR) and data volume of each layer. The SNR is the ratio of signal strength to noise strength, and the data volume is the total number of data points contained in that layer. By comprehensively considering these two factors, the layer that best represents the characteristics of the physical field is selected as the key data layer.

[0069] Extract feature vectors from the key data layer. Each feature vector consists of multiple parameters reflecting the data characteristics, including the average, maximum, minimum, and rate of change. Calculate the Euclidean distance between each feature vector. The Euclidean distance is calculated as follows: for two feature vectors X = (x1, x2, ..., x...)... n ) and Y = (y1, y2, ..., y n The Euclidean distance d between them is the square root of the sum of squares of the differences of all corresponding components, where x i and y i These are the i-th components of two feature vectors, where n is the dimension of the feature vectors. Based on the calculated Euclidean distance, a feature sequence related to each feature vector is established, and the feature sequences are sorted or grouped according to the magnitude of the Euclidean distance.

[0070] Feature sequences are used to screen features with significant physical meaning. Significant physical meaning refers to features that are directly or indirectly related to the existence, distribution, or properties of underground metal deposits; for example, changes in certain features may indicate the presence of underground ore bodies. The strength of the correlation between each feature is determined by analyzing co-occurrence frequency and consistency of change trends. Features with high co-occurrence frequency and consistent change trends have a strong correlation. The numerical value of the correlation strength is set as a weighting coefficient between each feature, and the magnitude of the weighting coefficient reflects the degree of mutual influence between features. Finally, feature weights are set according to the spatial distribution range of each feature. The weight of features with a wide spatial distribution range in the overall fusion may be adjusted according to the detection area they cover, so that the feature weights can better reflect the spatial importance of the features.

[0071] Example 4: When processing feature weights, the fusion decision module first extracts the spatial distribution range of each feature from the feature weights. The spatial distribution range of the features is determined based on the geographical location information of the data collection, covering information such as the latitude and longitude range, altitude range, and underground exploration depth involved in the data. Different features correspond to different data collection areas; some features may be concentrated in a local area of ​​a specific mining area, while others cover a large area of ​​the entire exploration area.

[0072] The influence boundaries of feature weights are set according to the regions corresponding to the spatial distribution range of each feature. The influence boundary is a spatial limit defined by the physical properties of the feature and the data distribution characteristics. Regions within the boundary are significantly affected by the feature, while the influence gradually weakens outside the boundary. The delineation of the influence boundary needs to consider the propagation characteristics of the geophysical field reflected by the feature. For example, the influence boundary of gravity field data may cover a wide area, while the influence boundary of electric field data may be relatively concentrated near specific geological structures.

[0073] The influence boundaries of each feature weight are superimposed to obtain the superimposed influence boundary. During the superposition process, for the overlapping areas of the influence boundaries of different feature weights, the weight value of each feature in that area is recorded. The superimposed influence boundary can intuitively present the spatial range of the combined effect of multiple features, as well as the comprehensive influence of features on different areas. The coverage of the superimposed influence boundary in each region is set as the regional contribution of the feature weight. The coverage is calculated as the ratio of the area covered by the influence boundaries of each feature in that region to the total area of ​​that region. The larger the ratio, the more significant the comprehensive influence of the feature weights on that region.

[0074] The regional contribution of feature weights is compared with the distribution of outliers in multi-source data. The distribution of outliers is obtained by statistically analyzing the frequency and density of outliers in each region. The frequency refers to the ratio of the number of outliers in a region to the total data volume in that region, while the density is the number of outliers per unit area. During the comparison, special attention is paid to regions where the regional contribution and outlier distribution do not match. For example, a region may have a high regional contribution but a significantly lower outlier frequency than other regions; or a region may have a low regional contribution but an abnormally high outlier frequency.

[0075] The deviation amount of these mismatched regions is marked. This deviation is calculated as the difference between the region's contribution and the outlier distribution frequency; the magnitude of the difference reflects the degree of mismatch. Outliers in the multi-source data are then removed according to the marked deviation amount. For regions with deviations exceeding a certain range, outliers within that region are considered invalid data and removed. For regions with smaller deviations, outliers are retained, and their location information is recorded for further analysis. This method completes the identification of outliers in the data.

[0076] Throughout the process, the extraction of the spatial distribution range of feature weights provides the basis for setting the influence boundary. The superposition of the influence boundary and the calculation of regional contribution realize the comprehensive consideration of the spatial features of multi-source data. The comparison with the distribution of outliers and the processing of deviation ensure the pertinence and accuracy of outlier identification.

[0077] Example 5: During runtime, the real-time update module first operates based on the adjusted feature association network. After adjustment by the fusion decision module, the topology of the feature association network more accurately reflects the intrinsic relationships between features in multi-source data. The connection strength and connection method between different feature nodes have been optimized based on the reliability assessment results. The fusion requirements for multi-source data are extracted from this feature association network. These requirements are determined based on the activity level of each feature node in the network, the tightness of the relationships between nodes, and the frequency of data flow. Feature data corresponding to nodes with high activity levels often require more frequent fusion processing, while closely related sets of nodes have a need for collaborative fusion. Paths with high data flow frequency reflect the urgency of data fusion.

[0078] The extracted fusion needs are prioritized according to their urgency, resulting in a data fusion priority. The urgency is determined based on factors including the real-time requirements of the data, the importance of feature nodes within the network, and the phased objectives of the exploration mission. Data with high real-time requirements has its corresponding fusion needs prioritized; fusion needs related to feature nodes at the core of the network are processed first; and fusion needs related to the geological areas of focus in the current exploration phase are also given higher priority. During the prioritization process, a comprehensive consideration of different urgency indicators is used to form a continuous priority sequence, ensuring that each fusion need has a clear processing order.

[0079] The fusion priority and fusion algorithm parameters are integrated according to a modular design. The modular design divides the data fusion process into multiple independent functional modules, each responsible for a specific type of data processing or fusion computation. Determined fusion priorities are assigned to the corresponding functional modules, enabling each module to process its assigned fusion task in priority order. Fusion algorithm parameters include filter window size, feature matching threshold, and weight update rate. These parameters are adjusted based on the current feature association network topology and data characteristics. For example, for closely related feature node sets, a higher feature matching threshold is used to reduce false matches; for features with large data fluctuations, the filter window size is adjusted to adapt to their changing characteristics.

[0080] Through the above integration process, a real-time fusion result library is generated. This library is categorized and stored according to the physical field type, acquisition time, and spatial location of the data. Each category contains the fused feature data, records of the algorithm parameters used in the data fusion, and timestamps of the fusion process. The library can receive new fusion results in real time and update the corresponding stored content based on the temporal and spatial attributes of the data, ensuring that the data in the library always reflects the latest fusion status. Indexes are established between data in different categories to facilitate quick retrieval and access to relevant fusion results according to detection needs, supporting centralized viewing and analysis of data from specific regions, time periods, or physical field types.

[0081] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0082] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A real-time fusion system for multi-source geophysical data of deep metal deposits, characterized in that, include: The data access module is used to receive raw data sent by different types of geophysical exploration equipment and divide the data into exploration dimensions according to the physical field type of the data acquisition. The spatiotemporal registration module is used to obtain the timestamps and spatial coordinates of each detection dimension and establish the mapping relationship between the original data and the unified spatiotemporal reference system. The feature extraction module is used to extract key features from registered multi-source data, determine feature weights according to the physical meaning of the features, and construct a feature association network based on the feature weights. The fusion decision module is used to perform credibility assessment on multi-source data based on the feature association network, identify outliers in the data, and adjust the topology of the feature association network according to the credibility assessment results. The real-time update module is used to optimize the priority and fusion algorithm parameters of data fusion based on the adjusted feature association network, and generate a real-time fusion result library.

2. The real-time fusion system for multi-source geophysical data of deep metal mines according to claim 1, characterized in that, The data access module can be implemented in the following ways: For any raw data sent by any detection device, invoke the data format parsing protocol; Use a parsing protocol to decode the original data, at least to obtain the physical field type and numerical sequence of the data; Identify noise signals, missing values, and anomalous jumps in the original data under the corresponding physical field type to form a data sample set; The noise signal, missing value, and anomalous transition in the data sample set are processed separately to generate noise filtering results, missing value imputation results, and transition correction results in sequence, which are used as preprocessed data corresponding to the original data.

3. The real-time fusion system for multi-source geophysical data of deep metal mines according to claim 2, characterized in that, The methods for generating noise filtering results, missing value imputation results, and jump correction results also include: Noise signals, missing values, and anomalous jumps in the data sample set are classified according to physical field type, resulting in multiple types of processing units; An adaptive filtering algorithm is applied to the noise signal in the type processing unit to obtain the noise filtering result; The temporal distribution density of missing values ​​and the amplitude of anomalous jumps in the type processing unit are calculated. The type processing unit is then corrected according to the temporal distribution density and amplitude values ​​to obtain the missing value filling result and jump correction result.

4. The real-time fusion system for multi-source geophysical data of deep metal mines according to claim 1, characterized in that, Other methods for implementing the detection dimensions corresponding to data partitioning include: The detection dimensions are validated by analyzing the sampling frequency and coverage of the data in the detection dimensions. The detection dimensions are then correlated according to their sampling frequency and coverage to construct a mapping relationship between the detection dimensions and the data credibility.

5. The real-time fusion system for multi-source geophysical data of deep metal mines according to claim 1, characterized in that, The spatiotemporal registration module can be implemented in the following ways: Retrieve the timestamp calibration parameters and spatial reference transformation model corresponding to the original data, and generate multiple initial registration results. The initial registration results represent spatiotemporal coordinate data without error correction. Determine whether multiple initial registration results meet the preset registration accuracy threshold. If they meet the registration accuracy threshold, the initial registration results are considered as registered multi-source data.

6. The real-time fusion system for multi-source geophysical data of deep metal mines according to claim 3, characterized in that, The methods for establishing a mapping relationship between the original data and the unified spatiotemporal reference system include: using the data characteristics of the noise filtering results, missing value imputation results, and jump correction results present in the preprocessed data, the coordinate parameters and time base of the unified spatiotemporal reference system, to construct a mapping relationship between the original data and the unified spatiotemporal reference system.

7. The real-time fusion system for multi-source geophysical data of deep metal mines according to claim 1, characterized in that, The feature extraction module can be implemented in the following ways: The registered multi-source data is processed into layers according to physical field type, data precision, and acquisition depth. The layer with the highest data density after layer processing is set as the key data layer. Extract feature vectors from key data layers, calculate the Euclidean distance between each feature vector, and establish a feature sequence related to the Euclidean distance between each feature vector; By utilizing the feature sequences related to the Euclidean distance between each feature vector, features with significant physical meaning in the feature sequences are screened, and the correlation strength between each feature is determined. The numerical value of the correlation strength is set as the weight coefficient between each feature. The weighting coefficients between each feature are set according to the spatial distribution range of each feature; During the layered processing, the data is divided into gravity field data layer, magnetic field data layer, electric field data layer and seismic wave data layer according to the physical field type. Each layer of data is further divided into surface layer, shallow middle layer and deep bottom layer according to the acquisition depth. The selection of key data layers is determined based on the signal-to-noise ratio and data volume of each layer.

8. The real-time fusion system for multi-source geophysical data of deep metal mines according to claim 7, characterized in that, The implementation methods of the fusion decision-making module include: Extract the spatial distribution range of each feature from the feature weights; set the influence boundary of the feature weights according to the region corresponding to the spatial distribution range of each feature; The influence boundaries of each feature in the feature weights are superimposed to obtain the superimposed influence boundary. The coverage of the superimposed influence boundary in each region is set as the regional contribution of the feature weight. The regional contribution of feature weights is compared with the distribution of outliers in multi-source data to mark the amount of deviation. Outliers in multi-source data are then removed according to the amount of deviation, thus completing the identification of outliers.

9. The real-time fusion system for multi-source geophysical data of deep metal mines according to claim 1, characterized in that, The real-time update module is implemented as follows: Based on the adjusted feature association network, the fusion requirements of multi-source data are extracted. The fusion requirements are sorted according to their urgency to obtain the data fusion priority. The fusion priority and fusion algorithm parameters are integrated according to the modular design to obtain a real-time fusion result library.

10. The real-time fusion system for multi-source geophysical data of deep metal mines according to claim 3, characterized in that, The adaptive filtering algorithm includes a noise decomposition step based on wavelet transform and a noise suppression step based on a threshold function. The noise decomposition step is used to decompose the noise signal into wavelet coefficients of different frequencies, and the noise suppression step is used to apply soft thresholding to the high-frequency wavelet coefficients to preserve the waveform characteristics of the effective signal.

Images