A geothermal exploration method based on machine learning

Abstract

The present invention belongs to the field of geothermal resource exploration technology, and in particular, relates to a geothermal exploration method based on machine learning. First, regional adaptive feature engineering is carried out, and geological, climate, historical exploration data and satellite images are integrated, weighted averaged and multi-scale fused to extract regional adaptive features. Then a multi-level integrated learning model is designed, which is divided into three layers: basic, advanced, and meta-model. Information is transmitted by weighted voting between layers, and model performance is improved through dynamic weight adjustment, collaborative optimization, and weighted combined output. Finally, a deviation correction mechanism is introduced to correct the final output based on the model output error. The present invention effectively integrates multi-source data, overcomes the limitations of traditional exploration methods, improves the efficiency and accuracy of geothermal resource exploration, reduces costs, and provides a reliable basis for geothermal resource development. It has important application value and broad market prospects.

Description

Technical Field

[0001] The present invention belongs to the technical field of geothermal resource exploration, and in particular relates to a geothermal exploration method based on machine learning. Background Art

[0002] As an important clean energy, geothermal resources have the advantages of sustainability, stability and wide distribution, and have great potential in addressing the global energy crisis and environmental problems. In recent years, with the continuous growth of global demand for clean energy, the development and utilization of geothermal resources has received more and more attention.

[0003] Existing geothermal exploration methods mainly include geological surveys, geophysical explorations, and drilling. Geological surveys infer the distribution of underground geothermal resources by observing and analyzing surface geological features, such as rock types, stratigraphic structures, and structural features. However, this method often relies on the experience of geological experts, is highly subjective, and has limited ability to obtain information about deep strata. Geophysical exploration methods, such as gravity exploration, magnetic exploration, and seismic exploration, use the differences in the physical properties of different strata to detect underground geological structures, but these methods have high equipment costs, complex operations, and multiple interpretation results. Drilling is the most direct way to obtain underground geothermal resources, but drilling is expensive and has a certain degree of blindness. If the drilling location is not properly selected, it may not be possible to accurately obtain geothermal resource information, resulting in a waste of resources.

[0004] With the rapid development of information technology, machine learning technology has achieved remarkable results in many fields. Machine learning algorithms can process large-scale and complex data and extract valuable information from them. They have the advantages of high efficiency, accuracy, and high degree of automation. In the field of geothermal exploration, combining machine learning technology with traditional exploration methods is expected to overcome the limitations of traditional methods and improve the efficiency and accuracy of geothermal resource exploration. For example, through machine learning analysis of geological, climate, and historical exploration data, the distribution and reserves of geothermal resources can be predicted more comprehensively and accurately, providing a more reliable basis for the development of geothermal resources. However, there are still some challenges in applying machine learning to geothermal exploration, such as how to effectively integrate multi-source heterogeneous data, build a suitable machine learning model, and improve the generalization ability of the model. These problems limit the widespread application and in-depth development of machine learning technology in geothermal exploration. Summary of the invention

[0005] In view of the technical problems existing in the above-mentioned background technology, the present invention proposes a geothermal exploration method based on machine learning.

[0006] In order to achieve the above object, the technical solution adopted by the present invention comprises the following steps:

[0007] S1. First, regional adaptive feature engineering is performed to extract regional adaptive features by weighted averaging and multi-scale fusion of multiple geological, climate and exploration data;

[0008] S2. Design a multi-level integrated learning model, use multiple machine learning models for training, and adjust the weight coefficients of each model according to the dynamic optimization mechanism. The specific implementation is as follows:

[0009] S21. First, the multi-layer ensemble learning model is divided into three layers, namely the basic model layer, the advanced model layer and the meta-model layer. Information is transmitted between each layer by weighted voting.

[0010] S22, then introduce a dynamic weight adjustment mechanism to calculate the weight of each model according to the performance of each model i , the weight dynamic adjustment formula is: Where ΔScore i =Accuracy i -MeanAccuracy, where Accuracy i is the accuracy of model i, MeanAccuracy is the average accuracy of all models, σ i is the standard deviation of the model’s performance, is the weight of the tth round;

[0011] S23. In addition, for each model M in the multi-model combination i , by introducing the loss function to optimize the synergy between multiple models, the synergy optimization objective function is: in is the loss function of model i, Represents the difference in prediction results between models i and j;

[0012] S24. Finally, the prediction results of each model are weighted and merged, and the weighted average method is used to fuse the output of each level;

[0013] S3. Finally, a bias correction mechanism based on the model output error is introduced to correct the final output of the integrated model, reduce the prediction error caused by the deviation in the model training process, and obtain the final geothermal resource prediction results.

[0014] Preferably, the extraction of regional adaptive features in step S1 is based on geological distribution, climate conditions, geothermal fluid flow model and historical exploration data. Given a sample set X = {x1, x2, x3, ..., x n}, where each data point x i Contains multiple features f i =[f i1 ,f i2,...,f im ], the regional features are obtained by weighted averaging and multi-scale transformation: where ω i is the weight coefficient; then the features are fused, the multiple features of satellite images and historical data are fused, and the feature conversion function is designed through a multi-scale filter to obtain the fused feature F combined .

[0015] Preferably, the weight coefficient ω i The specific calculation method is: where f region , g region ,h region are the measures of the geological, climatic and historical exploration data of the region, and the weight ω i Adjustments are made in each region based on actual geological characteristics.

[0016] Preferably, in step S21, the basic model layer uses decision trees and random forests, each model independently predicts the input data, the advanced model layer uses XGBoost to weightedly merge the output results of the first layer model to generate more refined predictions, and the meta-learning layer uses a neural network to make a final decision based on the output of the second layer to further optimize the prediction results.

[0017] Preferably, the implementation of the fusion output by the weighted average method in step S24 is as follows: Where L is the number of model levels, β l is the weight of this level, Where ΔScore l is the performance change of each layer model in the current training cycle, σ l is the performance standard deviation of the model at this layer.

[0018] Preferably, the specific implementation of the deviation correction mechanism is as follows: first, the deviation of the model is estimated by comparing the prediction errors of the training set and the validation set, and then the correction coefficient ξ is calculated based on the deviation, and the correction coefficient is applied to the final output of the model. The formula is: Where Bias is the model bias.

[0019] Preferably, the model bias Bias=E val -E train , where E val , E train They are the error on one side of the training set and the prediction error on the validation set, and the correction coefficient for the deviation calculation Where g is the number of samples in the validation set.

[0020] Compared with the prior art, the advantages and positive effects of the present invention are that, through weighted averaging and multi-scale fusion, the value of multi-source data is fully explored to obtain more representative regional adaptive features, which can more comprehensively reflect the status of geothermal resources than traditional methods. The multi-level integrated learning model combines a variety of machine learning algorithms to improve the accuracy and generalization ability of the model through dynamic weight adjustment, collaborative optimization and other mechanisms. The deviation correction mechanism effectively reduces the model training deviation, improves the prediction accuracy, provides a more scientific basis for geothermal resource development decision-making, and reduces exploration risks and costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 It is a structural flow chart of the present invention;

[0023] Figure 2 This is the structure diagram of the multi-level integrated learning model. DETAILED DESCRIPTION

[0024] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0025] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0026] Embodiment, in geothermal resource exploration, accurate prediction of the distribution and reserves of geothermal resources is crucial for the rational development and utilization of geothermal energy. Existing exploration methods such as geological surveys, geophysical exploration and drilling face many problems in practical applications. In order to solve these problems, the present invention proposes a geothermal exploration method based on machine learning, which aims to improve the efficiency and accuracy of geothermal resource exploration, reduce exploration costs, and reduce the blindness of exploration. First, in order to comprehensively and accurately extract feature information related to geothermal resources and overcome the problem of incomplete description of geological characteristics by traditional methods, the present invention adopts regional adaptive feature engineering. Extraction is based on geological distribution, climatic conditions, geothermal fluid flow model and historical exploration data. Given a sample set X = {x1, x2, x3, ..., x n}, where each data point x iContains multiple features f i =[f i1 ,f i2 ,...,f im ], the regional features are obtained by weighted averaging and multi-scale transformation: where ω i is the weight coefficient, and the specific calculation method is: where f region , g region ,h region are the measures of the geological, climatic and historical exploration data of the region, and the weight ω i Adjustments are made in each area according to the actual geological characteristics; then feature fusion is performed, integrating multiple features of satellite images and historical data, and the feature conversion function is designed through a multi-scale filter to obtain the fused feature F combined . The multiple features of satellite images and historical data are integrated, and the fused features are obtained by designing a feature conversion function through a multi-scale filter. In this embodiment, the satellite image provides information such as the topography and landforms of the surface, and the historical exploration data contains information such as previous drilling results and thermal displays. Through the multi-scale filter, these features of different scales are integrated, for example, the macroscopic terrain features in the satellite image are combined with the local thermal display information in the historical data, to obtain a richer and more comprehensive feature representation.

[0027] In order to improve the prediction accuracy and generalization ability of the model and overcome the limitations of a single model or a simple combination model, the present invention constructs a multi-level integrated learning model. The multi-layer integrated learning model is divided into three layers, namely, a basic model layer, an advanced model layer and a meta-model layer, and information is transmitted between each layer by weighted voting. The basic model layer uses decision trees and random forests, and each model independently predicts the input data. In this embodiment, the decision tree model can quickly classify the data, and is more effective for judging the relationship between some simple geological features and geothermal resources; the random forest improves the stability and accuracy of the model by integrating multiple decision trees. The advanced model layer uses XGBoost to weightedly merge the output results of the first layer model to generate a more refined prediction. The XGBoost algorithm has advantages in processing large-scale data and complex relationships, and can further mine the potential information in the output of the basic model layer and improve the prediction accuracy. The meta-learning layer uses a neural network to make a final decision based on the output of the second layer to further optimize the prediction results. The neural network has a strong nonlinear fitting ability and can comprehensively analyze the output of the advanced model layer to obtain a more accurate final prediction. The specific implementation is that the multi-layer integrated learning model is first divided into three layers, namely the basic model layer, the advanced model layer and the meta-model layer. Information is transmitted between each layer by weighted voting; secondly, a dynamic weight adjustment mechanism is introduced to calculate the weight δ of each model according to the performance of each model. i , the weight dynamic adjustment formula is: Where ΔScore i =Accuracy i -MeanAccuracy, where Accuracy i is the accuracy of model i, MeanAccuracy is the average accuracy of all models, σ i is the standard deviation of the model’s performance, is the weight of the tth round; in addition, for each model M in the multi-model combination i , by introducing the loss function to optimize the synergy between multiple models, the synergy optimization objective function is: in is the loss function of model i, Represents the difference in prediction results between models i and j; finally, the prediction results of each model are weighted and merged, and the weighted average method is used to fuse the output of each level. The implementation of the weighted average method to fuse the output is: Where L is the number of model levels, β l is the weight of this level, in ΔScorel is the performance change of each layer model in the current training cycle, σ l is the performance standard deviation of the model at this layer.

[0028] Finally, in order to reduce the prediction error caused by the deviation in the model training process and improve the reliability of the prediction results, the present invention introduces a deviation correction mechanism based on the model output error. First, the model deviation is estimated by comparing the prediction errors of the training set and the validation set, and then the correction coefficient ξ is calculated based on this deviation, and the correction coefficient is applied to the final output of the model. The formula is: Where Bias is the model bias. Model bias Bias = E val -E train , where E val , E train They are the error on one side of the training set and the prediction error on the validation set, and the correction coefficient for the deviation calculation Where g is the number of samples in the validation set. The geothermal exploration method based on machine learning of the present invention has significant advantages in improving the efficiency of geothermal resource exploration, reducing costs, and improving prediction accuracy, effectively overcoming many limitations of traditional geothermal exploration methods and providing strong support for the rational development and utilization of geothermal resources.

[0029] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A geothermal exploration method based on machine learning, characterized in that: The following steps are involved: S1. First, regional adaptive feature engineering is performed to extract regional adaptive features by weighted averaging and multi-scale fusion of multiple geological, climate and exploration data; S2. Design a multi-level integrated learning model, use multiple machine learning models for training, and adjust the weight coefficients of each model according to the dynamic optimization mechanism. The specific implementation is as follows: S21. First, the multi-layer ensemble learning model is divided into three layers, namely the basic model layer, the advanced model layer and the meta-model layer. Information is transmitted between each layer by weighted voting. S22, then introduce a dynamic weight adjustment mechanism to calculate the weight of each model according to the performance of each model i , the weight dynamic adjustment formula is: Where ΔScore i =Accuracy i -MeanAccuracy, where Accuracy i is the accuracy of model i, MeanAccuracy is the average accuracy of all models, σ i is the standard deviation of the model’s performance, is the weight of the tth round; S23. In addition, for each model M in the multi-model combination i , by introducing the loss function to optimize the synergy between multiple models, the synergy optimization objective function is: in is the loss function of model i, Represents the difference in prediction results between models i and j; S24. Finally, the prediction results of each model are weighted and merged, and the weighted average method is used to fuse the output of each level; S3. Finally, a bias correction mechanism based on the model output error is introduced to correct the final output of the integrated model, reduce the prediction error caused by the deviation in the model training process, and obtain the final geothermal resource prediction result.

2. A geothermal exploration method based on machine learning according to claim 1, characterized in that: The extraction of regional adaptive features in step S1 is based on geological distribution, climate conditions, geothermal fluid flow model and historical exploration data. Given a sample set X = {x1, x2, x3, ..., x n }, where each data point x i Contains multiple features f i =[f i1 ,f i2 ,...,f im ], the regional features are obtained by weighted averaging and multi-scale transformation: where ω i is the weight coefficient; then the features are fused, the multiple features of satellite images and historical data are fused, and the feature conversion function is designed through a multi-scale filter to obtain the fused feature F combined .

3. A geothermal exploration method based on machine learning according to claim 2, characterized in that: The weight coefficient ω i The specific calculation method is: where f region , g region ,h regi on are the measures of the geological, climate and historical exploration data of the region, and the weight ω i Adjustments are made in each region based on actual geological features.

4. The geothermal exploration method based on machine learning according to claim 1, characterized in that: In step S21, the basic model layer uses decision trees and random forests, and each model independently predicts the input data. The advanced model layer uses XGBoost to weightedly merge the output results of the first layer model to generate more refined predictions. The meta-learning layer uses a neural network to make the final decision based on the output of the second layer to further optimize the prediction results.

5. The geothermal exploration method based on machine learning according to claim 1, characterized in that: The implementation of the weighted average method in step S24 to fuse and output is as follows: Where L is the number of model levels, β l is the weight of this level, in ΔScorel is the performance change of each layer model in the current training cycle, σ l is the performance standard deviation of the model at this layer.

6. The geothermal exploration method based on machine learning according to claim 1, characterized in that: The specific implementation of the deviation correction mechanism is as follows: first, the deviation of the model is estimated by comparing the prediction errors of the training set and the validation set, and then the correction coefficient ξ is calculated based on this deviation, and the correction coefficient is applied to the final output of the model. The formula is: Where Bias is the model bias.

7. The geothermal exploration method based on machine learning according to claim 6, characterized in that: The model bias Bias = E val -E train , where E val , E train They are the error on one side of the training set and the prediction error on the validation set, and the correction coefficient for the deviation calculation Where g is the number of samples in the validation set.

Images