Salt lake mineral available resource quantity prediction method based on multi-source data fusion

Abstract

The invention discloses a salt lake mineral available resource quantity prediction method based on multi-source data fusion, and the method comprises the steps: obtaining multi-source data, carrying out the registration of the multi-source data, and obtaining the registered multi-source data; according to the registered multi-source data, key features related to salt lake mineral resources are extracted, and a multi-dimensional feature vector is obtained; a resource quantity prediction model is constructed, and the resource quantity prediction model is obtained through multi-dimensional feature vector training; obtaining multi-source feature data of the to-be-predicted region; inputting the multi-source feature data of the to-be-predicted region into a resource quantity prediction model, and obtaining an available quantity prediction result of salt lake mineral resources; and carrying out regional division on a prediction result by adopting a clustering analysis method to obtain resource distribution conditions of different regions. According to the method, the problems of multi-source heterogeneous data fusion and salt lake mineral resource quantity prediction are effectively solved, and the prediction precision and reliability are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and in particular relates to a method for predicting the available resource quantity of salt lake minerals based on multi-source data fusion. Background Art

[0002] In the business scenario of salt lake mineral resource exploration and development, there is a unique technical problem: how to accurately predict the available resource quantity of salt lake minerals based on multi-source data fusion. The distribution of salt lake mineral resources is complex, and the data sources are diverse, including geological exploration data, remote sensing image data, geochemical data, hydrological data, etc. Each type of data has different spatio-temporal resolutions and accuracies, and there are also significant differences in its acquisition methods and processing technologies. For example, geological exploration data is obtained through drilling and sampling, which can provide local high-precision resource information but has a limited coverage area; remote sensing image data can cover a large area, but its ability to analyze resource distribution is weak; geochemical data and hydrological data reflect the chemical and physical environments of salt lake mineral formation, but it is difficult to directly convert them into resource quantities.

[0003] When constructing a mineral resource quantity prediction model, how to effectively fuse these multi-source data becomes a key issue. On the one hand, the heterogeneity between data makes direct fusion difficult, and unified feature extraction and data preprocessing methods need to be designed. On the other hand, the spatio-temporal distribution inconsistency of data may lead to biases in the model training process. For example, the weight allocation of local high-precision data and global low-precision data in the model may affect the accuracy of the prediction results. In addition, the formation mechanism of salt lake minerals is complex, involving multiple geological, chemical, and hydrological processes. How to fully reflect these processes in the model is also one of the technical difficulties.

[0004] The selection and application of machine learning algorithms also need to solve multiple technical contradictions. For example, traditional regression models may not be able to capture complex non-linear relationships, while deep learning models can handle complex non-linear relationships but require a large amount of labeled data, and the labeled data of salt lake mineral resources is usually scarce. How to select a suitable model and optimize its performance under limited data volume becomes another key issue. Finally, how to apply the trained prediction model to the area to be predicted and ensure the reliability and interpretability of the prediction results is also a complex technical problem that needs to be studied in depth. Summary of the Invention

[0005] To solve the above technical problems, the present invention proposes a method for predicting the available resource quantity of salt lake minerals based on multi-source data fusion, which can effectively solve the problems of multi-source heterogeneous data fusion and salt lake mineral resource quantity prediction, and improve the prediction accuracy and reliability.

[0006] The present invention provides a method for predicting the exploitable resource quantity of salt lake minerals based on multi-source data fusion, including:

[0007] Obtain multi-source data, and register the multi-source data to obtain the registered multi-source data;

[0008] Extract key features related to salt lake mineral resources from the registered multi-source data to obtain a multi-dimensional feature vector;

[0009] Construct a resource quantity prediction model, wherein the resource quantity prediction model is obtained by training with the multi-dimensional feature vector;

[0010] Obtain the multi-source feature data of the area to be predicted;

[0011] Input the multi-source feature data of the area to be predicted into the resource quantity prediction model to obtain the prediction result of the exploitable quantity of salt lake mineral resources;

[0012] Adopt a clustering analysis method to divide the prediction result into regions and obtain the resource distribution in different regions.

[0013] Optionally, obtaining the multi-source data includes:

[0014] Obtain geological data, remote sensing data, chemical data, and hydrological data;

[0015] Process the geological data, remote sensing data, chemical data, and hydrological data to obtain the multi-source data.

[0016] Optionally, processing the geological data, remote sensing data, chemical data, and hydrological data includes:

[0017] Adopt a normalization method to map the geological data, remote sensing data, chemical data, and hydrological data to the same numerical range;

[0018] Adopt a feature extraction method to process the geological data, remote sensing data, chemical data, and hydrological data into a unified feature vector;

[0019] Adopt a matrix construction technique to process the geological data, remote sensing data, chemical data, and hydrological data into a standardized feature matrix.

[0020] Optionally, obtaining the registered multi-source data includes:

[0021] Adopt a spatio-temporal alignment algorithm to register the multi-source data to obtain initial registration data;

[0022] Judge whether the deviation of the initial registration data is within a preset range. If it exceeds the range, re-perform spatio-temporal registration.

[0023] Optionally, the key features related to salt lake mineral resources include: geological structure, chemical element distribution, and hydrogeological conditions.

[0024] Optionally, constructing the resource quantity prediction model includes:

[0025] Using the random forest algorithm, obtaining a training set from the multi-dimensional feature vectors, and training the resource quantity prediction model through the training set;

[0026] If the prediction result of the resource quantity prediction model does not reach the expected value, then use the semi-supervised learning method to update the training set, and use the updated training set to train the resource quantity prediction model until the prediction result reaches the expected value to obtain the resource quantity prediction model.

[0027] Optionally, using the semi-supervised learning method to update the training set includes:

[0028] Obtaining a small amount of labeled data and a large amount of unlabeled data, and constructing an initial data set;

[0029] Using the semi-supervised learning method to extract feature information from the initial data set, and updating the training set according to the extracted feature information.

[0030] Optionally, inputting the multi-source feature data of the area to be predicted into the resource quantity prediction model, and obtaining the prediction result of the available quantity of salt lake mineral resources includes:

[0031] Using the principal component analysis method to perform dimensionality reduction processing on the multi-source feature data, and extracting key features;

[0032] Inputting the key features into the resource quantity prediction model, running the model for prediction. During the running of the model, if a certain feature data is missing, then use the data of the adjacent area for interpolation to obtain the prediction result of the available quantity of salt lake mineral resources output by the model.

[0033] Compared with the prior art, the present invention has the following advantages and technical effects:

[0034] The present invention aims at multi-source heterogeneous data such as geology, remote sensing, chemistry, and hydrogeology, and realizes standardized conversion and consistency registration through data preprocessing and spatio-temporal alignment. Using the feature extraction algorithm to extract key features from the standardized data to form multi-dimensional feature vectors. Using the random forest algorithm to train the resource quantity prediction model, and introducing the semi-supervised learning method when the labeled data is scarce. Optimizing the feature weights and model parameters according to the training results to ensure the reasonable allocation of high and low precision data. Finally, applying the optimized model to the area to be predicted to achieve high-precision prediction of the available quantity of salt lake mineral resources. The present invention effectively solves the problems of multi-source heterogeneous data fusion and salt lake mineral resource quantity prediction, and improves the prediction accuracy and reliability. Brief Description of the Drawings

[0035] The accompanying drawings, which form a part of this application, are used to provide a further understanding of this application. The schematic embodiments and descriptions thereof of this application are used to explain this application and do not constitute an improper limitation of this application. In the accompanying drawings:

[0036] Figure 1 It is a flow chart of a method for predicting the available resource quantity of salt lake minerals based on multi-source data fusion according to an embodiment of the present invention. Detailed implementation manners

[0037] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments may be combined with each other. The following will refer to the accompanying drawings and combine the embodiments to detail this application.

[0038] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that here.

[0039] The present invention proposes a method for predicting the available resource quantity of salt lake minerals based on multi-source data fusion, as Figure 1 shown, specifically including the following steps:

[0040] Obtain multi-source data, and register the multi-source data to obtain the registered multi-source data;

[0041] According to the registered multi-source data, extract the key features related to salt lake mineral resources to obtain a multi-dimensional feature vector;

[0042] Construct a resource quantity prediction model, where the resource quantity prediction model is obtained by training with the multi-dimensional feature vector;

[0043] Obtain the multi-source feature data of the area to be predicted;

[0044] Input the multi-source feature data of the area to be predicted into the resource quantity prediction model to obtain the prediction result of the available quantity of salt lake mineral resources;

[0045] Adopt a clustering analysis method to divide the prediction result into regions to obtain the resource distribution in different regions.

[0046] Specifically, adopt data preprocessing technology. For the heterogeneity of geological data, remote sensing data, chemical data and hydrological data, uniformly convert them into a standardized feature matrix to eliminate the differences in data formats and scales;

[0047] Through a spatio-temporal alignment algorithm, register the data collected at different times and spaces to ensure the consistency of the data in the spatio-temporal dimension and reduce the bias in model training;

[0048] Based on the feature extraction algorithm, key features related to salt lake mineral resources are extracted from the standardized data, including geological structures, chemical element distributions, hydrological conditions, etc., to form a multi-dimensional feature vector;

[0049] The random forest algorithm is used to train a resource quantity prediction model with the multi-dimensional feature vector, dealing with complex non-linear relationships through ensemble learning and reducing the risk of overfitting;

[0050] If the labeled data is scarce, a semi-supervised learning method is introduced, combining a small amount of labeled data and a large amount of unlabeled data to enhance the model's prediction ability for resource quantity;

[0051] According to the model training results, the feature weights are adjusted and the model parameters are optimized to ensure the reasonable allocation of high-precision geological data and low-precision remote sensing data in the prediction;

[0052] The optimized model is applied to the area to be predicted, the multi-source feature data of this area is input, and the prediction result of the available quantity of salt lake mineral resources is output.

[0053] Furthermore, the multi-source data obtained includes:

[0054] Geological data, remote sensing data, chemical data and hydrological data are obtained;

[0055] The geological data, remote sensing data, chemical data and hydrological data are processed to obtain multi-source data.

[0056] Furthermore, the processing of the geological data, remote sensing data, chemical data and hydrological data includes:

[0057] The normalization method is adopted to map the geological data, remote sensing data, chemical data and hydrological data to the same numerical range;

[0058] The feature extraction method is adopted to process the geological data, remote sensing data, chemical data and hydrological data into a unified feature vector;

[0059] The matrix construction technology is adopted to process the geological data, remote sensing data, chemical data and hydrological data into a standardized feature matrix.

[0060] Specifically, obtain geological data, remote sensing data, chemical data, and hydrological data. For differences in data formats, adopt data parsing techniques to extract structured information. According to differences in data scales, use normalization techniques to map all data to the same numerical range. For heterogeneity, adopt feature extraction techniques to generate unified feature vectors from different types of data. Combine the feature vectors and use matrix construction techniques to generate a standardized feature matrix. If there are missing values in the feature matrix, use interpolation techniques to fill in the missing parts. According to the dimensions of the feature matrix, use dimensionality reduction techniques to reduce redundant features and retain key information. Use clustering algorithms to classify the feature matrix and identify potential patterns in the data.

[0061] Further, the obtained registered multi-source data includes:

[0062] Use a spatio-temporal alignment algorithm to register the multi-source data and obtain the initial registered data;

[0063] Judge whether the deviation of the initial registered data is within the preset range. If it exceeds the range, re-perform spatio-temporal registration.

[0064] Specifically, obtain data collected at different times and spaces. For the differences in the collection time and collection space, use a spatio-temporal alignment algorithm for registration. Through spatio-temporal registration, eliminate the differences in the time dimension and space dimension to obtain consistent data. Based on the consistent data, judge whether the data deviation is within the preset range. If it exceeds the range, re-perform spatio-temporal registration. Use a noise reduction algorithm to process the registered data to remove noise interference. Use a clustering algorithm to classify the processed data and identify potential patterns. According to the classification results, extract key features to generate a feature matrix. Use matrix decomposition techniques to reduce the dimension of the feature matrix and retain key information.

[0065] Further, the key features related to salt lake mineral resources include: geological structure, chemical element distribution, and hydrological conditions.

[0066] Specifically, use a standardization algorithm to process the original data to eliminate the difference in dimensions and obtain standardized data. Based on feature extraction algorithms, obtain key features such as geological structure, chemical element distribution, and hydrological conditions from the standardized data. According to the extracted key features, construct a multi-dimensional feature vector to form structured data. If there are outliers in the multi-dimensional feature vector, use a noise reduction algorithm to remove the abnormal interference and obtain the noise-reduced data. Use a clustering algorithm to classify the noise-reduced data and identify the distribution patterns of mineral resources in different regions. According to the classification results, use the principal component analysis algorithm to reduce the dimension of the multi-dimensional feature vector and retain key information. Based on the dimension-reduced data, generate a mineral resource distribution feature matrix for subsequent analysis.

[0067] More specifically, the denoised data is classified by a clustering algorithm to identify the distribution patterns of mineral resources in different regions.

[0068] The regional classification of the denoised data is carried out by using a clustering algorithm to obtain the distribution pattern of mineral resources. If there are outliers in the regional class, the abnormal interference is removed based on the denoised data. For the distribution pattern, a feature vector is constructed to obtain the key information of geological structures, chemical elements, and hydrography. Through the dimensionality-reduced data, a feature matrix is generated to retain the core features of mineral resources. According to the feature matrix, the distribution laws of mineral resources in different regions are determined. If there is redundant information in the feature vector, a dimensionality reduction algorithm is used to extract the key features. Based on the extracted key features, a distribution model of mineral resources is constructed.

[0069] Furthermore, the construction of the resource volume prediction model includes:

[0070] The random forest algorithm is used to obtain a training set from the multi-dimensional feature vector, and the resource volume prediction model is trained through the training set.

[0071] If the prediction result of the resource volume prediction model does not reach the expected value, the semi-supervised learning method is used to update the training set, and the resource volume prediction model is trained by using the updated training set until the prediction result reaches the expected value to obtain the resource volume prediction model.

[0072] Specifically, the random forest algorithm is used to obtain a training set from the multi-dimensional feature vector and construct a resource volume prediction model. The integrated learning method is used to handle the non-linear relationships in the feature quantities. If the model shows overfitting, a risk reduction strategy is adopted to optimize the performance of the prediction model. According to the feature distribution of the training set, the influence degree of the complex relationship is judged, and the way of processing ability is adjusted. Through the evaluation results of the model stability, the prediction range of the resource quantity is determined to verify the accuracy of the prediction model. If the prediction range exceeds the preset threshold, the weights of the feature quantities are readjusted to optimize the structure of the training set. According to the final prediction result, a resource volume distribution map is generated for subsequent analysis and decision-making.

[0073] More specifically, the random forest algorithm constructs a prediction model by combining multiple decision trees, and each decision tree is independently trained and predicted. Taking the prediction of salt lake mineral resources as an example, the content of elements such as potassium, lithium, and boron, geological structure characteristics, and hydrogeological conditions can be used as input features, and the historical resource reserves can be used as output labels. Suppose there are five years of historical data in a certain salt lake area, where the potassium element content fluctuates between 3% and 8%, and the lithium element content is distributed between 0.1% and 0.3%. These data can be used as training samples. The ensemble learning method integrates the advantages of multiple base learners to handle the non-linear relationship between features. Taking the relationship between geological structure and chemical element distribution as an example, the distribution of fault zones may affect the enrichment degree of certain elements, and the two show a complex non-linear relationship. By establishing multiple sub-models to learn different feature combinations respectively, this complex relationship can be better captured. In the case of model overfitting, the cross-validation method can be used to evaluate the model performance. Suppose the dataset is divided into a training set and a test set in a ratio of 8:2. If the accuracy of the training set reaches 95% while the test set is only 70%, it indicates that the model is overfitting. At this time, the risk of overfitting can be reduced by adjusting parameters such as the depth of the tree and increasing the minimum number of samples. In the evaluation of the complexity of feature relationships, the correlation coefficient between features can be calculated. For example, in a certain salt lake, the correlation coefficient between the groundwater level and the lithium element content is 0.8, indicating a strong correlation between the two, and the model needs to focus on this correlation. By adjusting the feature weights, the influence of important features can be highlighted. The model stability is evaluated by multiple random sampling validations. Suppose ten cross-validations are performed. If the standard deviation of the prediction results is less than the preset threshold of 0.1, the model is considered to have good stability. The rationality of the prediction range needs to be judged in combination with the actual geological conditions. If the predicted reserves of lithium ore in a certain area exceed three times the average value of the adjacent area, the feature weight configuration needs to be re-examined. The finally generated resource quantity distribution map can be displayed in the form of a heat map. Different colors represent different reserve levels, helping decision-makers intuitively grasp the resource distribution law. By overlaying auxiliary information such as geological structure and hydrogeology, the key factors affecting resource distribution can be analyzed in depth, providing a scientific basis for exploration and development. Regarding the reliability of the prediction results, different confidence intervals can be set. For example, an interval estimate of the resource reserves can be given at a 95% confidence level, providing more comprehensive reference information for decision-making.

[0074] Furthermore, the method for updating the training set using the semi-supervised learning method includes:

[0075] Obtain a small amount of labeled data and a large amount of unlabeled data to construct an initial dataset;

[0076] Adopt the semi-supervised learning method to extract feature information from the initial dataset, and update the training set according to the extracted feature information.

[0077] Specifically, a small amount of labeled data and a large amount of unlabeled data are obtained to construct an initial data set. A semi-supervised learning method is used to extract feature information from the initial data set. Through the feature information, an initial version of the resource quantity prediction model is trained. If the prediction performance of the model does not reach the preset threshold, the parameters of the semi-supervised learning method are adjusted. According to the adjusted parameters, the model is retrained to obtain an optimized version. Through the optimized version of the model, the distribution of the resource quantity is predicted. According to the prediction results, an analysis report of the resource quantity is generated.

[0078] More specifically, the semi-supervised learning method is an effective solution for dealing with the situation of scarce labeled data. In the resource quantity prediction model, first, some labeled resource data can be obtained, such as the real records of equipment usage and personnel demand at different times in a certain area. At the same time, a large amount of unlabeled relevant data is collected, such as equipment operation status and personnel flow records. These data form the initial training set, and the labeled data may only account for about 20% of the total data volume. Through feature extraction technology, feature information in multiple dimensions can be obtained from the data. For example, the equipment usage features can include usage frequency, duration, idle time, etc., and the personnel demand features can include workload, skill requirements, time distribution, etc. These feature information form feature vectors for training the initial prediction model. The finally generated analysis report needs to include key information such as the confidence interval of the predicted value, the demand distribution at different times, and the potential resource shortage risk. For example, the report can point out that under specific conditions, the utilization rate of a certain type of resource may exceed 95%, and it is recommended to make resource allocation or supplementation in advance. These analysis results can provide a scientific decision-making basis for managers and improve resource utilization efficiency.

[0079] Furthermore, inputting the multi-source feature data of the area to be predicted into the resource quantity prediction model, the prediction results of the available quantity of salt lake mineral resources obtained include:

[0080] Using the principal component analysis method, the multi-source feature data is dimensionally reduced to extract key features;

[0081] Input the key features into the resource quantity prediction model and run the model for prediction. During the running process of the model, if a certain feature data is missing, the data of the adjacent area is used for interpolation to obtain the prediction results of the available quantity of salt lake mineral resources output by the model.

[0082] Specifically, multi-source feature data of the area to be predicted are obtained, including geological, remote sensing, climate and other information. The principal component analysis method is used to reduce the dimension of the multi-source feature data and extract key features. The dimension-reduced feature data are input into a pre-trained optimization model, and the model is run for prediction. During the model operation, if a certain feature data is missing, the data of adjacent areas are used for interpolation. The prediction result of the available quantity of salt lake mineral resources output by the model is obtained. According to the prediction result, the data are normalized to ensure that the data are in the same dimension. The clustering analysis method is used to divide the prediction result into regions to obtain the resource distribution of different regions.

[0083] More specifically, taking the Qaidam Basin in Qinghai as an example, the multi-source feature data mainly include lithology data and drilling data obtained from geological exploration, surface spectral information and topographic data obtained by satellite remote sensing, and data such as precipitation and evaporation from climate monitoring stations. By processing these data with the principal component analysis method, the original dozens of features can be reduced to five to six key principal components, which can explain more than 85% of the total variance of the data. In the practice of the Qarhan Salt Lake, the features input into the model include formation salinity, groundwater salinity, evaporation, etc. When the data of some monitoring stations are missing, Kriging interpolation can be performed based on the historical data of surrounding monitoring stations to achieve data integrity. For example, the groundwater salinity data of a monitoring point in the eastern part of the Qarhan Salt Lake is missing, and the estimated value of this point is obtained as 200 grams per liter through interpolation of the data of five surrounding monitoring points. When predicting the resources of the Xitieshan Salt Lake, the dimension differences of different feature data are significant. For example, the unit of formation salinity is percentage, the unit of annual average precipitation is millimeter, and the unit of evaporation is cubic meter. Through the maximum-minimum normalization process, all data are uniformly mapped to the interval from zero to one, ensuring the comparability of different features in the prediction. In the Dachaidan Salt Lake mining area, the clustering analysis method is used to divide the mining area into three mining sub-areas based on features such as resource grade and burial depth. Among them, the average lithium ion content in the northern sub-area reaches 0.8 grams per liter, which is suitable as a priority mining area; the potassium salt grade in the central sub-area is relatively high and the magnesium-lithium ratio is relatively low, which is suitable as a potassium fertilizer production base; the southern sub-area is temporarily used as a reserve area due to its deep burial. The resource prediction of the Chaka Salt Lake shows that under the existing mining technology and economic conditions, the recoverable reserves of the mining area are about 300 million tons. Through the spatial analysis of the prediction result, it is found that the resource distribution shows an obvious circular feature, with lower salt mineral content in the outer area and higher brine grade in the central area, which is consistent with the geological law of salt lake evolution and verifies the reliability of the prediction result. The prediction result of the Golmud Salt Lake group shows that there is a hydraulic connection of the groundwater system between different salt lakes. When formulating the mining plan, the influence between adjacent salt lakes needs to be considered to avoid over-mining leading to damage to the regional ecological environment. By establishing an ecological safety threshold for salt lake resource development, the balance between resource development and environmental protection is ensured.

[0084] Further, during the operation of the model, if a certain feature data is missing, the data in the adjacent area is used for interpolation.

[0085] Specifically, the original feature data of the area to be predicted is obtained. For the missing values in the feature data, the data in the adjacent area is used for interpolation. The interpolated feature data is input into the pre-established dimensionality reduction model to extract key features. According to the key features, the optimization model is run for resource quantity prediction. The predicted values output by the model are obtained and normalized. The clustering analysis method is used to divide the normalized predicted values into regions. The regional division result is obtained and a resource distribution map is generated.

[0086] More specifically, obtaining the original feature data of the area to be predicted is the basic link of salt lake mineral resource assessment. Taking a certain salt lake in the Qaidam Basin as an example, it is necessary to collect geological feature data such as geological structure, lithology distribution, and stratigraphic sequence. At the same time, hydrochemical data such as brine ion concentration and salinity in this area, as well as climate data such as surface temperature and precipitation, are obtained. These multi-source data constitute the data basis for assessment. For the problem of missing feature data, spatial interpolation methods can be used. For example, if the potassium ion concentration data is missing at a certain sampling point, spatial interpolation can be carried out through the Kriging interpolation method based on the measured data of surrounding sampling points. This method takes into account spatial autocorrelation and can better restore the feature distribution in the missing area. Feature dimension reduction is an important means of dealing with high-dimensional data. Taking salt lake mineral assessment as an example, the original features may include dozens of indicators, and through the principal component analysis method, it can be reduced to several principal components. For example, reducing the original twenty features to five principal components, and the cumulative explained variance of these five principal components reaches more than 80%, which not only retains the main information of the data but also reduces the complexity of subsequent modeling. In the stage of resource quantity prediction, an optimized random forest model can be used. This model can handle the non-linear relationship between features and has strong generalization ability. By inputting the dimension-reduced feature data, the model will output the predicted value of the resource reserve in the predicted area. The normalization processing of the prediction results helps in the comparative analysis between different regions. The maximum-minimum normalization method can be used to map the predicted values uniformly to the interval from zero to one. The processed data is comparable and convenient for subsequent regional division. The regional division uses the fuzzy clustering analysis method, which can divide the predicted area into high, medium, and low resource enrichment areas. In practice, if the area of a certain salt lake is 100 square kilometers, through clustering analysis, it may be divided into a high enrichment area of about 20 square kilometers, a medium enrichment area of 50 square kilometers, and a low enrichment area of 30 square kilometers. The finally generated resource distribution map visually displays the resource endowments of different regions and can provide an important reference for development decisions. Based on the resource distribution map, factors such as geological structure and hydrogeological conditions can be combined to further analyze the genetic mechanism of resource enrichment. For example, the high enrichment area is often closely related to the distribution of fault zones and groundwater recharge conditions. This genetic analysis can provide guidance for prospecting exploration in surrounding areas. Through the above technical route, both the quantitative prediction of resource quantity and the revelation of the spatial characteristics and formation mechanism of resource distribution are achieved.

[0087] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for predicting the exploitable resource volume of salt lake minerals based on multi-source data fusion, characterized in that, Including: Obtain multi-source data, register the multi-source data, and obtain the registered multi-source data; Extract key features related to salt lake mineral resources from the registered multi-source data to obtain a multi-dimensional feature vector; Construct a resource quantity prediction model, where the resource quantity prediction model is obtained by training with the multi-dimensional feature vector; Obtain the multi-source feature data of the area to be predicted; Input the multi-source feature data of the area to be predicted into the resource quantity prediction model to obtain the prediction result of the available quantity of salt lake mineral resources; Adopt a clustering analysis method to divide the prediction results into regions to obtain the resource distribution in different regions.

2. The prediction method for the available resource volume of salt lake minerals based on multi-source data fusion according to claim 1, wherein Obtaining multi-source data includes: Obtain geological data, remote sensing data, chemical data, and hydrological data; Process the geological data, remote sensing data, chemical data, and hydrological data to obtain the multi-source data.

3. A method for predicting the available resource volume of salt lake minerals based on multi-source data fusion according to claim 2, characterized in that, Processing the geological data, remote sensing data, chemical data, and hydrological data includes: Adopt the normalization method to map the geological data, remote sensing data, chemical data, and hydrological data to the same numerical range; Adopt the feature extraction method to process the geological data, remote sensing data, chemical data, and hydrological data into a unified feature vector; Adopt the matrix construction technology to process the geological data, remote sensing data, chemical data, and hydrological data into a standardized feature matrix.

4. A method for predicting the available resource volume of salt lake minerals based on multi-source data fusion according to claim 1, characterized in that, Obtaining the registered multi-source data includes: Adopt a spatio-temporal alignment algorithm to register the multi-source data to obtain initial registration data; Judge whether the deviation of the initial registration data is within the preset range. If it exceeds the range, re-perform spatio-temporal registration.

5. A method for predicting the available resource volume of salt lake minerals based on multi-source data fusion according to claim 1, characterized in that, The key features related to salt lake mineral resources include: geological structure, chemical element distribution, and hydrological conditions.

6. A method for predicting the available resource volume of salt lake minerals based on multi-source data fusion according to claim 1, characterized in that, Constructing the resource quantity prediction model includes: Adopt the random forest algorithm to obtain a training set from the multi-dimensional feature vector, and train the resource quantity prediction model with the training set; If the prediction result of the resource quantity prediction model does not reach the expected value, adopt the semi-supervised learning method to update the training set, and use the updated training set to train the resource quantity prediction model until the prediction result reaches the expected value to obtain the resource quantity prediction model.

7. A method for predicting the available resource quantity of salt lake minerals based on multi-source data fusion according to claim 6, characterized in that, Updating the training set by adopting the semi-supervised learning method includes: Obtain a small amount of labeled data and a large amount of unlabeled data to construct an initial data set; Adopt the semi-supervised learning method to extract feature information from the initial data set, and update the training set according to the extracted feature information.

8. A method for predicting the available resource quantity of salt lake minerals based on multi-source data fusion according to claim 1, characterized in that, Inputting the multi-source feature data of the area to be predicted into the resource quantity prediction model to obtain the prediction result of the available quantity of salt lake mineral resources includes: Adopt the principal component analysis method to reduce the dimension of the multi-source feature data and extract key features; Input the key features into the resource quantity prediction model, run the model for prediction. During the running of the model, if a certain feature data is missing, use the data of the neighboring area for interpolation to obtain the prediction result of the available quantity of salt lake mineral resources output by the model.

Images