Mineral prediction-oriented dynamic adaptive learning method, system, equipment and medium

Abstract

The invention discloses a mineral product prediction-oriented dynamic adaptive learning method, system and device and a medium, and relates to the technical field of geophysical exploration. The method comprises the following steps: acquiring a data set; the data set comprises geological data and corresponding labels; performing feature extraction, parameter initialization and parameter optimization on the geological data in sequence, and performing association training in a dynamic adaptive learning framework by using the label to obtain a mineral product prediction model; algorithms in the association training process comprise a variational expectation algorithm and a variational maximization algorithm; the variational expectation algorithm comprises an unsupervised learning mode, a semi-supervised learning mode and a fully supervised learning mode; and performing prediction by using the mineral product prediction model to obtain a mineral product prediction result. According to the method, the limitation of a traditional machine learning method can be broken through, and a more efficient, universal and stable strategy is provided for geophysical data analysis and mineral resource evaluation.

Description

Technical Field

[0001] The present invention relates to the field of geophysical prospecting technology, and in particular to a dynamic adaptive learning method, system, equipment and medium for mineral prediction. Background Art

[0002] In the field of geological exploration and geophysics, it is crucial to identify and predict target rock masses. Traditional rock mass identification methods mainly rely on the experience of geologists and simple statistical data processing techniques. These methods are often limited in effect when dealing with complex geological data, especially when faced with large-scale and multi-source heterogeneous data. With the advancement of detection technology, the quantity and quality of detection data have been significantly improved. The data are not only multi-source, but also high-dimensional and complex. These data include but are not limited to seismic data, geomagnetic data, gravity data, and various geological and geochemical data. Each data carries unique information about underground structures. However, how to effectively integrate and analyze these multi-source data and extract useful information to accurately predict the location and properties of rock masses is still a huge challenge.

[0003] At present, machine learning technology has shown powerful data processing capabilities in many fields, especially in image recognition and natural language processing. This has inspired researchers to apply machine learning methods to geological data analysis in order to solve problems that traditional methods cannot overcome. Although machine learning technology provides new possibilities for data analysis, in practical applications, how to design an effective learning framework that adapts to the characteristics of geological data is still a technical problem.

[0004] However, although data-driven methods based on machine learning have shown significant advantages in processing complex geological data and identifying potential mineral resources, they often lack versatility and adaptability in multi-data environments of multiple mining areas or different mining sections within complex mining areas. On the one hand, there are significant differences in the data sets collected in different mining areas, and many existing methods mainly focus on the application of a single data scenario. On the other hand, with the increasing complexity of the mining environment and the continuous expansion of the detection range, the data sources collected in different mining sections are relatively diverse. At this time, there will be mismatches in data features. For example, vegetation-covered areas often use aerial or semi-aerial detection methods, while flat terrain areas tend to use ground detection methods. In addition, due to the spatial aggregation and uneven distribution of mineral deposits, the acquisition of data labels will also be uneven. Using a single machine learning method at the mining scale will lead to regional imbalance and information asymmetry in model training, which will in turn affect the generalization performance of the model. Summary of the invention

[0005] The purpose of the present invention is to provide a dynamic adaptive learning method, system, equipment and medium for mineral prediction, which can break through the limitations of traditional machine learning methods and provide a more efficient, universal and stable strategy for geophysical data analysis and mineral resource assessment.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A dynamic adaptive learning method for mineral prediction, comprising:

[0008] Acquire a data set; the data set includes geological data and corresponding labels;

[0009] The geological data is sequentially subjected to feature extraction, parameter initialization and parameter optimization, and the labels are used to perform association training in a dynamic adaptive learning framework to obtain a mineral prediction model; the algorithms in the association training process include a variational expectation algorithm and a variational maximization algorithm; the variational expectation algorithm includes an unsupervised learning mode, a semi-supervised learning mode and a fully supervised learning mode;

[0010] The mineral prediction model is used to perform prediction to obtain a mineral prediction result.

[0011] Optionally, the label includes a number 0 and a non-0 number; the number 0 indicates no lithology; the non-0 number indicates different lithologies.

[0012] Optionally, the training process of the dynamic adaptive learning framework is:

[0013] Performing feature extraction, parameter initialization and parameter optimization on the geological data in sequence to obtain training parameters, and inputting the training parameters into a dynamic adaptive learning framework to construct a training model;

[0014] The geological data is input into the training model, supervised learning training is performed with the goal of minimizing the loss between the model output result and the label, and the trained model is determined as a mineral prediction model.

[0015] Optionally, when the labels are all 0, an unsupervised learning mode is adopted; when the labels are not all 0, a semi-supervised learning mode is adopted; when the labels are not all 0, a fully supervised learning mode is adopted.

[0016] Optionally, after obtaining the mineral prediction model, it also includes: performing model evaluation and geological interpretation; wherein the model evaluation is specifically: for the unsupervised learning mode, using the silhouette coefficient distribution for evaluation; for the semi-supervised learning mode and the fully supervised learning mode, using the calculation accuracy, the mixed matrix function and the f1 function for evaluation.

[0017] The present invention also provides a dynamic adaptive learning system for mineral prediction, comprising:

[0018] A data set generation unit, used to obtain a data set; the data set includes geological data and corresponding labels;

[0019] A model training unit is used to perform feature extraction, parameter initialization and parameter optimization on the geological data in sequence, and to perform association training in a dynamic adaptive learning framework using the labels to obtain a mineral prediction model; the algorithms in the association training process include a variational expectation algorithm and a variational maximization algorithm; the variational expectation algorithm includes an unsupervised mode, a semi-supervised mode and a fully supervised mode;

[0020] The model prediction unit is used to use the mineral prediction model to perform prediction and obtain mineral prediction results.

[0021] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned dynamic adaptive learning method for mineral prediction.

[0022] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the dynamic adaptive learning method for mineral prediction as described above.

[0023] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0024] The present invention discloses a dynamic adaptive learning method, system, device and medium for mineral prediction, the method comprising acquiring a data set; the data set comprising geological data and corresponding labels; performing feature extraction, parameter initialization and parameter optimization on the geological data in sequence, and performing association training in a dynamic adaptive learning framework using the labels to obtain a mineral prediction model; the algorithms in the association training process include a variational expectation algorithm and a variational maximization algorithm; the variational expectation algorithm includes an unsupervised learning mode, a semi-supervised learning mode and a fully supervised learning mode; and performing prediction using the mineral prediction model to obtain a mineral prediction result. The present invention can break through the limitations of traditional machine learning methods and provide a more efficient, universal and stable strategy for geophysical data analysis and mineral resource assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0026] Figure 1 The present invention is a flow chart of a dynamic adaptive learning algorithm for mineral prediction;

[0027] Figure 2 3D display diagram of mesh density and resistivity in this embodiment; wherein part (a) is a schematic diagram of mesh density, and part (b) is a schematic diagram of resistivity;

[0028] Figure 3 This is a schematic diagram of the drilling verification result in this embodiment;

[0029] Figure 4 Schematic diagram of the prediction model and geological interpretation results in this embodiment; wherein, part (a) is the 3D model diagram, and part (b) is the target perspective diagram. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] The purpose of the present invention is to provide a dynamic adaptive learning method, system, equipment and medium for mineral prediction, which can break through the limitations of traditional machine learning methods and provide a more efficient, universal and stable strategy for geophysical data analysis and mineral resource assessment.

[0032] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] like Figure 1 As shown, the present invention provides a dynamic adaptive learning method for mineral prediction, comprising:

[0034] Step 100: Acquire a data set; the data set includes geological data and corresponding labels;

[0035] Step 200: extracting features, initializing parameters and optimizing parameters on the geological data in sequence, and performing association training in a dynamic adaptive learning framework using the labels to obtain a mineral prediction model; the algorithms in the association training process include a variational expectation algorithm and a variational maximization algorithm; the variational expectation algorithm includes an unsupervised learning mode, a semi-supervised learning mode and a fully supervised learning mode;

[0036] Step 300: Use the mineral prediction model to perform prediction and obtain mineral prediction results.

[0037] Based on the above technical solution, the following embodiments are provided.

[0038] 1) Dataset generation: Unify geophysical, geochemical, remote sensing, hyperspectral and other data into the same coordinate system to form a feature set (Features) of data samples, use the data provided by prior geological information (lithology) such as drilling holes as labels (Labels), and match them to the same position in the feature set. Design corresponding labels for different lithologies, represented by non-zero numbers; labels without lithology representation are defined as 0. The data set is shown in Table 1. After the data set is generated, the Features are Gaussian-like and standardized, and the data set is updated.

[0039] Table 1 Dataset

[0040]

[0041] 2) Feature extraction: Install the Matlab dimensionality reduction toolbox and independently select the most suitable dimensionality reduction tool in the Matlab dimensionality reduction toolbox to provide optimized data input for subsequent model training.

[0042] 3) Parameter initialization: Before the model training process begins, a preliminary estimate of the model parameters is required, including the mean μ of each Gaussian component. k , covariance Λ k and the mixing weight π k wait.

[0043] 4) Parameter optimization: The purpose of this step is to optimize global parameters and improve the efficiency of parameter adjustment. k and β as optimization targets, design the genetic algorithm function. When the algorithm is running, first set π k The interval range of the two parameters π and β, the number of iterations of the function, and the convergence condition. k is the mixing weight, β is the precision parameter of the Gaussian distribution, and the covariance matrix Λ k GA starts with a set of randomly generated candidate solutions, and in each generation, the most adapted individuals are selected for genetic operations to generate a new generation. This process is repeated until the termination condition is reached and the optimal π is output.k and the optimal β.

[0044] 5) Dynamic adaptive model training: The data set generated by step 1), the optimal parameters obtained in step 4), and other initial parameters in step 3) are input into the designed dynamic adaptive learning framework. By associating labels with learning patterns, dynamic adaptive learning of data is achieved to improve the prediction ability of the model. The algorithm alternately executes two steps: the variational expectation step (E step) and the variational maximization step (M step).

[0045] Variational E step: First, based on the optimal parameter π obtained in step 3) k and β and other parameters in step 2) are calculated for each data point x i The posterior probability (i.e., responsibility γ) of each Gaussian component ij ), see formula (1):

[0046]

[0047] Among them, π k , μ k and Λ k are the mixture weight, mean and covariance of the kth Gaussian component, and r ij is the responsibility of data point i to belong to distribution j.

[0048] However, different label situations correspond to different learning modes. When all label data are 0, the unsupervised learning mode is selected; when the label data are not all 0, the semi-supervised learning mode is selected; when the label data are not all 0, the fully supervised learning mode is selected. Later, according to the learning mode, the responsibility matrix γ is flexibly processed. ij .

[0049] A Unsupervised learning model: only the responsibility of each data point γ is required ij Normalize (Formula (2)) to ensure that the sum of the responsibility of each data point is 1:

[0050]

[0051] B Semi-supervised learning model: the degree of responsibility r for each data point ij After normalization, we get To perform the “Cannot-link” label constraint, first for each labeled data point x i , find its most likely cluster j:

[0052] j * (i) = argmax j γ 1 ij (3)

[0053] Here * (i) is the data point x i The most likely cluster index. For each cluster j and each data point x i , if x i The actual label l(i) and the label l(j) of the representative data point of cluster j * ) is different, then reduce x i The degree of responsibility of j:

[0054]

[0055] Normalize the updated responsibility to ensure that the sum of the responsibility of each data point to all clusters is 1, and finally update the responsibility matrix of the model

[0056] C Fully supervised learning mode: Directly set the responsibility corresponding to the data point according to formula (5) to ensure that it is fully assigned to the cluster specified by its label.

[0057]

[0058] here, is the data point x i The responsibility assigned to cluster j. l(i) is the responsibility of data point x i The actual label.

[0059] Variational M-step: Use the responsibility γ calculated by variational E-step new ij , this embodiment updates the parameter mean μ of each Gaussian component k , covariance Λ k and the mixing weight π k Parameter updating can be achieved by maximizing the expected complete data log-likelihood obtained in the E step.

[0060] The variational EM algorithm iteratively executes the variational E step and variational M step until convergence. Convergence is determined based on the change in the variational lower bound (ELBO). When the increase in ELBO is less than a preset threshold, the algorithm is considered to have converged.

[0061] 6) Output of prediction results: First, determine whether the algorithm has converged based on the ELBO situation. If the ELBO no longer increases, the algorithm is considered stable and converged. Subsequently, determine the output of the prediction results. For the unsupervised learning mode, the output result is the prediction result. For the semi-supervised learning mode, the output result of the data with the original label of 0 is the prediction result. However, for supervised learning, after the model training is completed, only the prediction model is obtained. It is necessary to prepare the training set to input into the model, calculate the probability of the training set, and determine the prediction category of each data point by finding the clustering index with the highest probability for each data point.

[0062] 7) Model evaluation and geological interpretation: For the unsupervised learning model, the silhouette coefficient distribution is used for evaluation; for the semi-supervised learning and fully supervised learning models, the accuracy, mixed matrix function, and f1 function are used to evaluate the model effect. Finally, the prediction results are geologically interpreted and compared with other prior data.

[0063] Practical Examples

[0064] Experimental data This embodiment uses the geophysical and drilling data in the "density-resistivity-efb.intermediate" in the example library of the EarthVolumetric Studio (EVS) 3D geological modeling software system. This data body contains the density and resistivity of the formation within a certain depth range of the survey area, as well as 22 drilling data. The main goal of this data body is to find coal seams based on the density and resistivity characteristics of the formation. This embodiment first uses the Kriging grid algorithm to grid the density and resistivity data, with the dimensions of 5m×5m×5m in three directions, and uses surfer software to image it, such as Figure 2 As shown in the figure, a part of it is cut out in this embodiment to better show the distribution of internal data. For the drilling data, the position coordinates and the revealed stratum information are listed in Table 2 in this embodiment.

[0065] First, the geophysical data and drilling data are unified into the same coordinate system to generate the required data set. There are 34,125 data in total, of which 897 are labeled data. In the data set, this embodiment uses boreholes BC-725 and BC-719 as verification boreholes and does not use them for model training. This mining area is a typical coal seam mining area, and it can be seen from the drilling data that there are obvious stratifications between the media in this area, which are mainly divided into 4 types of media, among which medium 3 is the target body of this embodiment, namely the coal seam. This embodiment marks the known labeled data with Arabic numerals 1-4 according to the medium type, and marks the unlabeled data with "0".

[0066] After the data set is generated, this embodiment uses depth, density, and resistivity as features for model training in this embodiment, and performs preprocessing operations such as normalization on the features. Considering that the labeled data only accounts for 2% of the total data, in order to ensure the reliability of the prediction effect, this embodiment inputs the labeled data and the unlabeled data into the model together. First, the entire data set is input into MS-DALF. At this time, the label contains 0, so semi-supervised learning is automatically performed. The search range of the parameter is set to 0.01 to 10, PopulationSize sets the population size to 50, and MaxGenerations sets the maximum iteration number to 50. At this time, the optimal Alpha and kappa parameter values ​​found by the genetic algorithm are 0.1 and 0.096, respectively. The optimal parameters are then input into SsL-VGMM for training. The model converges after 216 iterations. In the prediction results obtained, no new categories are found to be generated.

[0067] After the model training, the classification accuracy of the label data was 100%. In addition, the results of the two verification boreholes BC-725 and BC-719 were compared with the prediction results of this embodiment. Figure 3 As shown, it is found that the results of different lithologies are 100% consistent with the results predicted by this embodiment, which further illustrates the reliability of the method proposed in this embodiment.

[0068] Table 2 Stratigraphic information list

[0069]

[0070] Figure 4 What is shown is the prediction result of the entire mining area. In the figure, a part of it is removed in this embodiment to better show the distribution of internal data. It can be found that there are 4 relatively obvious stratifications in the mining area. Among them, the third layer medium is the coal seam, which is also the target layer of this study area. The coal seam is mainly distributed in the elevation range of 480m to 510m, with a thickness of about 30m. This result is consistent with the actual situation of the survey area and is more scientific and reliable than the traditional method based on the estimation of physical property parameter range.

[0071] Therefore, in order to solve the problems in the prior art, this embodiment develops an innovative dynamic adaptive learning framework, MS-DALF, which is characterized by high modularity and can more flexibly adapt to the differences between mines, such as different geological backgrounds and detection technologies, thereby improving the accuracy of prediction and analysis. This learning framework uses the data set label as a "handle" for dynamic conversion of data scenes, and uses the variational Gaussian mixture model as a shared model. It designs proprietary learning methods and evaluation criteria for different data scenes to ensure the high efficiency and accuracy of model training and prediction in different tasks. In addition, the feature dimension reduction module of the MS-DALF framework can flexibly select feature dimension reduction tools, which can effectively alleviate the problem of feature set mismatch between different mining sections in the mining area. The present invention aims to break through the limitations of traditional machine learning methods and provide a more efficient, universal and stable strategy for geophysical data analysis and mineral resource assessment. With the help of this algorithm, mineral resources can be more accurately identified and located, and the allocation and application of exploration resources can be optimized, which is particularly important and valuable in future exploration.

[0072] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0073] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A dynamic adaptive learning method for mineral prediction, It is characterized in that include: Acquire a data set; the data set includes geological data and corresponding labels; The geological data is sequentially subjected to feature extraction, parameter initialization and parameter optimization, and the labels are used to perform association training in a dynamic adaptive learning framework to obtain a mineral prediction model; the algorithms in the association training process include a variational expectation algorithm and a variational maximization algorithm; the variational expectation algorithm includes an unsupervised learning mode, a semi-supervised learning mode and a fully supervised learning mode; The mineral prediction model is used to perform prediction to obtain a mineral prediction result.

2. The dynamic adaptive learning method for mineral prediction according to claim 1, It is characterized in that The label includes a number 0 and a non-zero number; the number 0 indicates no lithology; the non-zero number indicates different lithologies.

3. The dynamic adaptive learning method for mineral prediction according to claim 1, It is characterized in that The training process of the dynamic adaptive learning framework is: Performing feature extraction, parameter initialization and parameter optimization on the geological data in sequence to obtain training parameters, and inputting the training parameters into a dynamic adaptive learning framework to construct a training model; The geological data is input into the training model, supervised learning training is performed with the goal of minimizing the loss between the model output result and the label, and the trained model is determined as a mineral prediction model.

4. The dynamic adaptive learning method for mineral prediction according to claim 2, It is characterized in that When the labels are all 0, an unsupervised learning mode is adopted; when the labels are not all 0, a semi-supervised learning mode is adopted; when the labels are not all 0, a fully supervised learning mode is adopted.

5. The dynamic adaptive learning method for mineral prediction according to claim 1, It is characterized in that After obtaining the mineral prediction model, it also includes: model evaluation and geological interpretation; The model evaluation is specifically as follows: for the unsupervised learning mode, the silhouette coefficient distribution is used for evaluation; for the semi-supervised learning mode and the fully supervised learning mode, the calculation accuracy, the mixing matrix function and the f1 function are used for evaluation.

6. A dynamic adaptive learning system for mineral prediction, It is characterized in that include: A data set generation unit, used to obtain a data set; the data set includes geological data and corresponding labels; A model training unit is used to perform feature extraction, parameter initialization and parameter optimization on the geological data in sequence, and to perform association training in a dynamic adaptive learning framework using the labels to obtain a mineral prediction model; the algorithms in the association training process include a variational expectation algorithm and a variational maximization algorithm; the variational expectation algorithm includes an unsupervised mode, a semi-supervised mode and a fully supervised mode; The model prediction unit is used to use the mineral prediction model to perform prediction and obtain mineral prediction results.

7. An electronic device, It is characterized in that It comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the dynamic adaptive learning method for mineral prediction according to any one of claims 1-5.

8. A computer-readable storage medium, It is characterized in that It stores a computer program, which, when executed by a processor, implements the dynamic adaptive learning method for mineral prediction as described in any one of claims 1-5.

Images