Data-mode driven artificial intelligence model construction method for mineral product prediction

Abstract

The invention discloses a data-mode-driven artificial intelligence model construction method for mineral product prediction, and relates to the field of earth science, and the method mainly comprises the steps: determining source, transportation and storage substitution indexes according to a mineralization system; assembling a multi-source earth science data set; constructing a source, transportation and storage branch artificial intelligence model; establishing a power law function between the ore deposit space distribution density and the key ore control element distance; adding the power law function into a hidden layer of the artificial intelligence model, and constructing a metallogenic mode constraint artificial intelligence model; inputting the multi-source earth scientific data set into a source, transport and storage branch artificial intelligence model, and extracting a source, transport and storage alternative index layer; and inputting the layers into a metallogenic mode constraint artificial intelligence model, and delineating a metallogenic prospective area. By implementing the data-mode driven artificial intelligence model construction method for mineral product prediction, the interpretability and generalization ability of the artificial intelligence model for mineral product prediction can be improved, and the precision of predicting the metallogenic prospective area is improved.

Description

Technical Field

[0001] The present invention relates to the field of earth sciences, and in particular to a method for constructing a data-pattern driven artificial intelligence model for mineral prediction. Background Art

[0002] There are many difficulties in prospecting in covered areas and deep deposits. For example, the prospecting information is weak due to the influence of the covering layer and the deep burial of the ore deposits, the mineralization information is mixed due to the superposition of complex geological processes and multiple periods of mineralization, and the prospecting information is missing or incomplete due to changes in the preservation environment of the ore deposits. There is an urgent need to innovate the theories and methods for deep-level prospecting information mining and integration in prospecting in covered areas and deep deposits.

[0003] We have accumulated a large amount of geological, geophysical, geochemical, remote sensing and drilling data, and the quantitative prediction of mineral resources has entered the era of big data. Artificial intelligence can effectively mine complex, nonlinear geospatial big data, extract unknown patterns of geological processes, and has high prediction accuracy. It is becoming an important tool for mineral prediction. For example, graph convolutional neural networks can fully characterize the spatial distribution characteristics of mineralization and the spatial coupling relationship between ore deposits and prospecting information, and take into account the anisotropy of mineralization; recurrent neural networks can capture the content of geochemical elements and element combination characteristics in exploration geochemical data.

[0004] Data-driven mineral prediction represented by big data and artificial intelligence is at the forefront of the current mathematical geology field. However, data-driven mineral prediction mainly focuses on the spatial mapping relationship between geological variables and ore deposits, without fully considering the mineralization pattern, resulting in weak generalization ability and poor interpretability of data-driven mineral prediction models. Therefore, the development of data-pattern-driven mineral prediction artificial intelligence models is at the forefront of current mineral prediction and is of great significance for improving the effectiveness of prospecting. At present, there is no report on how to couple data-driven artificial intelligence algorithms with pattern-driven representative mineralization systems to improve the effectiveness and credibility of mineral prediction artificial intelligence models. Summary of the invention

[0005] The present invention provides a data-pattern driven method for constructing an artificial intelligence model for mineral prediction, so as to solve the problem that the artificial intelligence modeling process for mineral prediction lacks guidance from a metallogenic system, correct the prediction process that is inconsistent with the metallogenic system, and improve the interpretability and generalization ability of the artificial intelligence model for mineral prediction.

[0006] The present invention provides a data-mode driven method for constructing an artificial intelligence model for mineral prediction, comprising the following steps: S1: determining source substitution indicators, transportation substitution indicators and reserve substitution indicators according to a mineralization system; S2: compiling data according to the source substitution indicators, transportation substitution indicators and reserve substitution indicators to obtain a multi-source earth science data set; S3: constructing a source branch artificial intelligence model, a transportation branch artificial intelligence model and a reserve branch artificial intelligence model; S4: establishing a power law function between the spatial distribution density of the ore deposit and the distance to the key ore-controlling elements according to the mineralization system; S5: adding the power law function to the hidden layer of the artificial intelligence model to construct a mineralization mode constraint artificial intelligence model; S6: inputting the multi-source earth science data set into the source branch artificial intelligence model, the transportation branch artificial intelligence model and the reserve branch artificial intelligence model to extract a source substitution indicator layer, a transportation substitution indicator layer and a reserve substitution indicator layer; S7: inputting the source substitution indicator layer, the transportation substitution indicator layer and the reserve substitution indicator layer into the mineralization mode constraint artificial intelligence model to delineate a mineralization prospective area.

[0007] Furthermore, the above-mentioned mineralization system is a magmatic-hydrothermal mineralization system; in the magmatic-hydrothermal mineralization system, the source substitution indicator is the intermediate-acidic granite body, the transport substitution indicator is the fault, and the reservoir substitution indicator is the geochemical anomaly.

[0008] Furthermore, the above-mentioned multi-source earth science dataset includes an intermediate-acidic granite dataset, a fracture dataset and a geochemical anomaly dataset; the intermediate-acidic granite dataset includes geochemical major and trace elements, geophysical data, a digital elevation model and an intermediate-acidic granite dataset; the fracture dataset includes geochemical major and trace elements, geophysical data, a digital elevation model and fractures; and the geochemical anomaly dataset includes geochemical major and trace elements.

[0009] Furthermore, the basic model of the above-mentioned source branch artificial intelligence model and operation branch artificial intelligence model is a graph convolutional neural network, which is used to learn the spatial structural characteristics of intermediate-acidic granite bodies and faults; the reservoir branch artificial intelligence model is a dual-branch model, and the reservoir branch artificial intelligence model includes a first model and a second model. The basic model of the first model is a graph convolutional neural network, which is used to extract geochemical spatial patterns; the basic model of the second model is a recurrent neural network, which is used to identify geochemical element content and element combination characteristics.

[0010] Furthermore, the power law function between the spatial distribution density of the above-mentioned ore deposits and the distance to the key ore-controlling elements is: ,in, is the spatial distribution density of the mineral deposits, is a constant, is the distance to the key ore-controlling elements, is the singularity index.

[0011] Furthermore, the power law function is added to the hidden layer of the artificial intelligence model, as shown in the formula: ,in, [∙] is the activation function, which is used to improve the nonlinear expression ability of the hidden layer. is a power law function, which characterizes the relationship between the spatial distribution density of mineral deposits and the distance to key mineral-controlling factors. is the input feature vector, and are the trainable weight and bias parameters respectively.

[0012] Furthermore, step S6 specifically includes: inputting the intermediate-acidic granite body data set into the source branch artificial intelligence model, inputting the fracture data set into the transport branch artificial intelligence model, and inputting the geochemical anomaly data set into the reservoir branch artificial intelligence model, minimizing the gap between the predicted probability and the previous geological survey results to tune the model, extracting the source substitute indicator layer, the transport substitute indicator layer and the reservoir substitute indicator layer, the source substitute indicator layer is the intermediate-acidic granite body prediction probability map, the transport substitute indicator layer is the fracture prediction probability map, and the reservoir substitute indicator layer is the geochemical anomaly map.

[0013] Furthermore, step S7 specifically includes: inputting the acidic granite body prediction probability map in the source substitute indicator layer, the fracture prediction probability map in the transport substitute indicator layer, and the geochemical anomaly map in the reservoir substitute indicator layer into the mineralization pattern constraint artificial intelligence model, taking into account the mineralization pattern and data characteristics, integrating the substitute indicator layers, and delineating the prospective mineralization areas.

[0014] The implementation of the data-mode driven mineral prediction artificial intelligence model construction method provided by the present invention has the following beneficial effects: The present invention utilizes the characteristics that pattern-driven mineral prediction relies on easy-to-understand metallogenic systems and expert knowledge but has poor adaptability to dynamically changing exploration scenarios, and that data-driven mineral prediction can discover potential patterns in geological prospecting big data but has poor interpretability of decision-making logic. Combining data-driven and pattern-driven, the current frontier paradigm of earth science, the present invention determines source substitution indicators, transportation substitution indicators and storage substitution indicators based on the source, transportation and storage components of the metallogenic system, and compiles multi-source earth science data sets; designs source branch artificial intelligence models, transportation branch artificial intelligence models and storage branch artificial intelligence models based on the different characteristics of the components of the metallogenic system; extracts the metallogenic pattern according to the metallogenic system, and establishes the relationship between the spatial distribution density of ore deposits and the distance to key ore-controlling factors. The power law function between the two is added to the hidden layer of the artificial intelligence model, and the power law function is linked to the trainable parameters of the hidden layer in the artificial intelligence model to construct an artificial intelligence model constrained by a metallogenic pattern, so that the metallogenic prospecting area delineated by the model is consistent with the metallogenic system; the multi-source earth science data set is input into the source branch artificial intelligence model, the transportation branch artificial intelligence model and the reservoir branch artificial intelligence model, the difference between the predicted probability and the previous geological survey results is minimized to tune the model, and the source substitute indicator layer, the transportation substitute indicator layer and the reservoir substitute indicator layer are extracted to reduce the uncertainty of generating the substitute indicator layer; the extracted source substitute indicator layer, the transportation substitute indicator layer and the reservoir substitute indicator layer are input into the artificial intelligence model constrained by a metallogenic pattern In the model, the metallogenic pattern and data characteristics are taken into account, and the alternative indicator layers are integrated to delineate the prospective metallogenic areas. The present invention constructs a data-pattern driven artificial intelligence model for mineral prediction from the perspectives of data set construction, alternative indicator extraction and model operation, which fully combines the advantages of data-driven artificial intelligence algorithms and pattern-driven representation of metallogenic systems. It can compile earth science data under the guidance of the source, transportation and storage components of the metallogenic system to improve the rationality of data selection; it can also design source, transportation and storage branches to extract alternative indicator layers according to the different characteristics of the components of the metallogenic system, so that underground ore-controlling elements can be detected, reducing the uncertainty of generating alternative indicator layers; at the same time, the metallogenic pattern extracted from the metallogenic system is embedded into the implicit In the containing layer, the model learns the coupling relationship between the mineralization pattern and the geological prospecting data during operation, and continuously adjusts the trainable parameters in the hidden layer, thereby improving the accuracy of mineral prediction; the mineralization system guides the construction of the mineral prediction artificial intelligence model data set, the extraction of alternative indicators and the operation of the model, which can not only improve the application effectiveness of the mineral prediction artificial intelligence, but also obtain results consistent with the mineralization system, and improve the interpretability of the prediction results; in short, the present invention can solve the problem that the mineral prediction artificial intelligence model lacks the guidance of the mineralization system, corrects the prediction process inconsistent with the mineralization system, improves the interpretability and generalization ability of the mineral prediction artificial intelligence model, and promotes the deployment of the mineral prediction artificial intelligence model in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which: Figure 1 It is a flow chart of a data-mode driven mineral prediction artificial intelligence model construction method provided by the present invention; Figure 2 The power law function relationship diagram between the spatial distribution density of the ore deposit and the distance from the ore-bearing stratum provided by the present invention is as an example; Figure 3 It is a schematic diagram of the structure of a data-mode driven mineral prediction artificial intelligence model provided by the present invention; DETAILED DESCRIPTION In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0016] Figure 1 A flow chart of a data-mode driven mineral prediction artificial intelligence model construction method of this embodiment is shown. In this embodiment, a data-mode driven mineral prediction artificial intelligence model construction method includes the following steps: S1: According to the metallogenic system, determine the source substitution index, transport substitution index and storage substitution index; Specifically, in step S1, the mineralization system is a magmatic-hydrothermal mineralization system; in the magmatic-hydrothermal mineralization system, the source substitution indicator is an intermediate-acidic granite body, the transport substitution indicator is a fault, and the reservoir substitution indicator is a geochemical anomaly, that is, the source substitution indicator, the transport substitution indicator, and the reservoir substitution indicator in the magmatic-hydrothermal mineralization system are an intermediate-acidic granite body, a fault, and a geochemical anomaly, respectively; S2: Data compilation is performed based on source substitution indicators, transport substitution indicators and storage substitution indicators to obtain a multi-source earth science dataset; Specifically, the multi-source geoscience dataset includes intermediate-acidic granite dataset, fracture dataset and geochemical anomaly dataset; the intermediate-acidic granite dataset includes geochemical major and trace elements, geophysical data, digital elevation model and intermediate-acidic granite; the fracture dataset includes geochemical major and trace elements, geophysical data, digital elevation model and fracture; the geochemical anomaly dataset includes geochemical major and trace elements; It should be noted that in step S2, based on the intermediate-acid granite body as the source substitute indicator, the fault as the transport substitute indicator, and the geochemical anomaly as the reservoir substitute indicator in the metallogenic system, a multi-source earth science data set is compiled, including an intermediate-acid granite body data set, a fault data set, and a geochemical anomaly data set; the intermediate-acid granite body data set includes geochemical major and trace elements, geophysical data, a digital elevation model, and an intermediate-acid granite body; the fault data set includes geochemical major and trace elements, geophysical data, a digital elevation model, and a fault; the geochemical anomaly data set includes geochemical major and trace elements; and the multi-source earth science data are pre-processed by spatial interpolation and GIS analysis; S3: Build the source branch AI model, operation branch AI model, and storage branch AI model; Specifically, the basic models of the source branch artificial intelligence model and the transport branch artificial intelligence model are graph convolutional neural networks, which are used to learn the spatial structural characteristics of intermediate-acidic granite bodies and faults; the reservoir branch artificial intelligence model is a dual-branch model, which includes a first model and a second model. The basic model of the first model is a graph convolutional neural network, which is used to extract geochemical spatial patterns; the basic model of the second model is a recurrent neural network, which is used to identify geochemical element content and element combination characteristics; It should be noted that in step S3, a source branch artificial intelligence model, a transport branch artificial intelligence model and a reservoir branch artificial intelligence model are designed based on the different characteristics of the components of the mineralization system. Among them, the acidic granite bodies and faults in the source substitute indicators and transport substitute indicators are usually irregular morphological structures with complex spatial patterns. Therefore, a graph convolutional neural network is used as the basic model of the source branch artificial intelligence model and the transport branch artificial intelligence model, and the high-level spatial patterns are extracted layer by layer with the help of stacked graph convolution layers; the reservoir substitute indicators are geochemical anomalies, with geochemical spatial patterns and geochemical element content and element combination characteristics, so a double-branch model is designed as the basic model of the reservoir branch artificial intelligence model. The reservoir branch artificial intelligence model includes a first model and a second model. The basic model of the first model is a graph convolutional neural network, which uses stacked graph convolution layers to extract geochemical spatial patterns. The basic model of the second model is a recurrent neural network, which uses stacked recurrent units to identify geochemical element content and element combination characteristics; in one embodiment, the three branches run in parallel, please refer to Figure 3 ; S4: According to the metallogenic system, a power law function is established between the spatial distribution density of mineral deposits and the distance to key ore-controlling factors; Specifically, in step S4, the mineralization pattern is extracted according to the mineralization system, such as the key mineralization-controlling elements. The distances from the key mineralization-controlling elements are different, and the spatial distribution density of the ore deposits is different. The distances from the key mineralization-controlling elements are different, and the distances from the key mineralization-controlling elements are different. The number of deposits and the cumulative number of deposits and spatial distribution density of mineral deposits , , based on the distance of the controlling elements is the horizontal axis, and the spatial distribution density of the deposit Draw a double logarithmic scatter plot for the ordinate, and use the least squares method to fit the straight line to obtain the power law function ,in is a constant, is the distance to the key ore-controlling elements, is the singularity index; In one embodiment, taking the mineralized stratum as an example, Figure 2 As shown, the fitted straight line expression in the double logarithmic graph is ; S5: Add the power law function to the hidden layer of the artificial intelligence model to construct an artificial intelligence model constrained by the mineralization pattern; Specifically, in step S5, the power law function (obtained according to S4) is added to the hidden layer of the artificial intelligence model, as shown in the formula: ,in, is the hidden layer of the artificial intelligence model, [∙] is the activation function, which is used to improve the nonlinear expression ability of the hidden layer. is a power law function, which characterizes the relationship between the spatial distribution density of mineral deposits and the distance to key mineral-controlling factors. is the input feature vector, and are trainable weights and bias parameters respectively; It should be noted that during the operation of the model, it can not only learn the spatial coupling relationship between key mining control factors and input feature vectors in forward propagation, but also adjust the trainable weight parameters in back propagation. and bias parameters , guiding the model to couple data and mineralization patterns, and improving the interpretability and generalization ability of artificial intelligence models for mineral prediction; S6: input the multi-source earth science dataset into the source branch artificial intelligence model, the operation branch artificial intelligence model and the storage branch artificial intelligence model to extract the source substitute indicator layer, the operation substitute indicator layer and the storage substitute indicator layer; Specifically, step S6 includes: inputting the intermediate-acidic granite body data set into the source branch artificial intelligence model, inputting the fracture data set into the transport branch artificial intelligence model, and inputting the geochemical anomaly data set into the reservoir branch artificial intelligence model, minimizing the gap between the predicted probability and the previous geological survey results to tune the model, extracting the source substitute indicator layer, the transport substitute indicator layer and the reservoir substitute indicator layer, the source substitute indicator layer is the intermediate-acidic granite body prediction probability map, the transport substitute indicator layer is the fracture prediction probability map, and the reservoir substitute indicator layer is the geochemical anomaly map; It should be noted that, in step S6, the compiled multi-source earth science data set (obtained according to S2) is input into the source branch artificial intelligence model (obtained according to S3), the transport branch artificial intelligence model (obtained according to S3) and the storage branch artificial intelligence model (obtained according to S3), and the source substitute indicator layer, the transport substitute indicator layer and the storage substitute indicator layer are extracted; In one embodiment, please refer to Figure 3 , input the intermediate-acidic granite body data set (obtained from S2), the fracture data set (obtained from S2) and the geochemical anomaly data set (obtained from S2) into the constructed source branch artificial intelligence model (obtained from S3), the transport branch artificial intelligence model (obtained from S3) and the reservoir branch artificial intelligence model (obtained from S3), respectively, minimize the gap between the predicted probability and the previous geological survey results to tune the model, extract the intermediate-acidic granite body prediction probability map of the source substitute indicator layer, the fracture prediction probability map of the transport substitute indicator layer and the geochemical anomaly map of the reservoir substitute indicator layer; S7: input the source substitute index layer, the transport substitute index layer and the storage substitute index layer into the mineralization pattern constraint artificial intelligence model to delineate the mineralization prospective area; Specifically, step S7 specifically includes: inputting the acidic granite prediction probability map in the source substitute indicator layer, the fracture prediction probability map in the transport substitute indicator layer, and the geochemical anomaly map in the reservoir substitute indicator layer into the mineralization pattern constraint artificial intelligence model, taking into account the mineralization pattern and data characteristics, integrating the substitute indicator layers, and delineating the mineralization prospective area; It should be noted that in step S7, the source substitute indicator layer (obtained according to S6), the transport substitute indicator layer (obtained according to S6) and the storage substitute indicator layer (obtained according to S6) are input into the mineralization pattern constraint artificial intelligence model (obtained according to S5), taking into account the mineralization pattern and data characteristics, integrating the substitute indicator layers, and delineating the prospective mineralization areas.

[0017] In one embodiment, please refer to Figure 3 The extracted intermediate-acidic granite prediction probability map, fracture prediction probability map and geochemical anomaly map are input into the mineralization pattern constraint artificial intelligence model, taking into account the mineralization pattern and data characteristics, integrating the source substitution indicators, transport substitution indicators and reservoir substitution indicators of the magma-hydrothermal mineralization system, and delineating the mineralization prospective areas to improve the reliability, interpretability and prediction accuracy of the mineral prediction artificial intelligence model.

[0018] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A data-mode driven method for constructing an artificial intelligence model for mineral prediction, characterized in that: The following steps are involved: S1: According to the metallogenic system, determine the source substitution index, transport substitution index and storage substitution index; S2: compile data based on the source substitute index, transportation substitute index and storage substitute index to obtain a multi-source earth science data set; S3: Build the source branch AI model, operation branch AI model, and storage branch AI model; S4: according to the mineralization system, a power law function is established, wherein the power law function represents the relationship between the spatial distribution density of the mineral deposit and the distance of the key mineral-controlling elements; S5: Add the power law function to the hidden layer of the artificial intelligence model to construct an artificial intelligence model constrained by the mineralization pattern; S6: input the multi-source earth science dataset into the source branch artificial intelligence model, the operation branch artificial intelligence model and the storage branch artificial intelligence model to extract the source substitute indicator layer, the operation substitute indicator layer and the storage substitute indicator layer; S7: Input the source substitution indicator layer, transport substitution indicator layer and storage substitution indicator layer into the mineralization pattern constraint artificial intelligence model to delineate the prospective mineralization area.

2. A data-mode driven mineral prediction artificial intelligence model construction method according to claim 1, characterized in that: The mineralization system is a magmatic-hydrothermal mineralization system; in the magmatic-hydrothermal mineralization system, the source substitution indicator is the intermediate-acidic granite body, the transport substitution indicator is the fault, and the reservoir substitution indicator is the geochemical anomaly.

3. A data-mode driven mineral prediction artificial intelligence model construction method according to claim 1, characterized in that: The multi-source earth science data set includes an intermediate-acidic granite data set, a fracture data set and a geochemical anomaly data set; the intermediate-acidic granite data set includes geochemical major and trace elements, geophysical data, a digital elevation model and an intermediate-acidic granite; The fracture data set includes geochemical major and trace elements, geophysical data, digital elevation models and fractures; the geochemical anomaly data set includes geochemical major and trace elements.

4. A data-mode driven mineral prediction artificial intelligence model construction method according to claim 1, characterized in that: The basic model of the source branch artificial intelligence model and the operation branch artificial intelligence model is a graph convolutional neural network, which is used to learn the spatial structural characteristics of intermediate-acidic granite bodies and fractures; The reservoir-branch artificial intelligence model is a dual-branch model, which includes a first model and a second model. The basic model of the first model is a graph convolutional neural network, which is used to extract geochemical spatial patterns; the basic model of the second model is a recurrent neural network, which is used to identify geochemical element content and element combination characteristics.

5. A data-mode driven mineral prediction artificial intelligence model construction method according to claim 1, characterized in that: The power law function is: ,in, is the spatial distribution density of the mineral deposits, is a constant, is the distance to the key ore-controlling elements, is the singularity index; the power law function characterizes the relationship between the spatial distribution density of the mineral deposit and the distance to the key mineral-controlling elements.

6. A data-mode driven mineral prediction artificial intelligence model construction method according to claim 1, characterized in that: The power law function is added to the hidden layer of the artificial intelligence model, as shown in the formula: ， in, [∙] is the activation function, which is used to improve the nonlinear expression ability of the hidden layer. is a power law function, which characterizes the relationship between the spatial distribution density of mineral deposits and the distance to key mineral-controlling factors. is the input feature vector, and are the trainable weight and bias parameters respectively.

7. A data-mode driven mineral prediction artificial intelligence model construction method according to claim 1, characterized in that: Step S6 specifically includes: inputting the intermediate-acidic granite body data set into the source branch artificial intelligence model, inputting the fracture data set into the transport branch artificial intelligence model, and inputting the geochemical anomaly data set into the reservoir branch artificial intelligence model, minimizing the gap between the predicted probability and the previous geological survey results to tune the model, extracting the source substitute indicator layer, the transport substitute indicator layer and the reservoir substitute indicator layer, the source substitute indicator layer is the intermediate-acidic granite body prediction probability map, the transport substitute indicator layer is the fracture prediction probability map, and the reservoir substitute indicator layer is the geochemical anomaly map.

8. A data-mode driven mineral prediction artificial intelligence model construction method according to claim 7, characterized in that: Step S7 specifically includes: inputting the acidic granite prediction probability map in the source substitute indicator layer, the fracture prediction probability map in the transport substitute indicator layer, and the geochemical anomaly map in the reservoir substitute indicator layer into the mineralization pattern constraint artificial intelligence model, taking into account the mineralization pattern and data characteristics, integrating the substitute indicator layers, and delineating the mineralization prospective area.

Images