Intelligent positioning method for deep geothermal target areas based on machine learning algorithm

Abstract

The present invention discloses a method for intelligent positioning of deep geothermal target areas based on a machine learning algorithm, and the present invention relates to the field of geothermal resource exploration technology. The method comprises the following steps: S1, multi-source heterogeneous data fusion acquisition and preprocessing; S2, geological feature entropy quantification and spatial autocorrelation analysis; S3, deep feature extraction and multimodal feature fusion; S4, multi-model adaptive integration and dynamic weight optimization; S5, reinforcement learning dynamic adjustment to participate in abnormal threshold determination; S6, three-dimensional geothermal target area intelligent positioning and risk assessment. This positioning technology integrates geological structure, geophysical field, geochemistry and remote sensing data through a machine learning algorithm to construct a high-dimensional feature vector, thus solving the limitation of the existing technology that relies solely on physical detection. At the same time, it introduces reinforcement learning to dynamically optimize model parameters, combined with real-time geothermal well data updates, so that the model can adapt to changes in geological conditions, and the prediction accuracy is greatly improved compared with traditional methods.

Description

Technical Field

[0001] The present invention relates to the technical field of geothermal resource exploration, and specifically to a method for intelligent positioning of deep geothermal target areas based on a machine learning algorithm. Background Art

[0002] Geothermal resource exploration involves conducting geothermal resource surveys. The process includes data collection, geothermal geological surveys, geothermal field surveys, geophysical and geochemical surveys, geothermal drilling, pumping tests, and geothermal reinjection. The process also involves calculating geothermal reserves based on geothermal geological parameters.

[0003] The invention with publication number CN118897328A discloses an exploration method and exploration device for medium-deep geothermal resource reserves. The exploration method for medium-deep geothermal resource reserves includes the following steps: obtaining a first parameter of cosmic ray muons reaching the surface of a target area, the first parameter including intensity and incident direction; obtaining a second parameter of cosmic ray muons underground, the second parameter including transmission information and scattering information; and generating geothermal resource information of the target area based on the first parameter and the second parameter.

[0004] As shown in the above invention, existing exploration methods rely on cosmic rays Muon detection devices are expensive to deploy and complex to operate, and they rely mainly on cosmic rays. The physical detection data of muons is limited in popularity and difficult to apply to rapid exploration under complex geological conditions. In addition, the existing exploration methods mainly rely on cosmic rays. The physical detection data of mesons have not fully integrated multi-dimensional data such as geological structure, geochemistry, remote sensing thermal infrared, and lack of quantitative analysis of geological characteristics, resulting in an incomplete characterization of the geothermal system and difficulty in capturing the control of complex geological structures on the distribution of heat reservoirs. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides an intelligent positioning method for deep geothermal target areas based on machine learning algorithms, which solves the existing problems.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a deep geothermal target area intelligent positioning method based on machine learning algorithm, comprising the following steps:

[0007] S1. Multi-source heterogeneous data fusion acquisition and preprocessing: Collect geological structure data, geophysical field data, geochemical data, and remote sensing thermal infrared data to construct a multi-source data set and eliminate dimensional differences through standardization;

[0008] S2. Quantification of geological characteristic entropy and spatial autocorrelation analysis: Calculate the entropy of geological structural complexity to quantify data uncertainty, and use the Moran index to analyze spatial autocorrelation;

[0009] S3. Deep feature extraction and multimodal feature fusion: Convolutional neural networks are used to extract texture features from remote sensing thermal infrared images, long short-term memory networks are used to extract geophysical field time series features, and multimodal fusion is achieved through feature splicing.

[0010] S4. Multi-model adaptive integration and dynamic weight optimization: Build a model group including random forest, gradient boosting machine, and Transformer, and dynamically adjust the model contribution based on the adaptive weight mechanism;

[0011] S5. Dynamically tune the participation of reinforcement learning in determining anomaly thresholds: Introduce reinforcement learning algorithms to optimize model parameters, using prediction error as the reward function; use the Otsu algorithm to determine geothermal anomaly thresholds;

[0012] S6. 3D geothermal target area intelligent positioning and risk assessment: Based on 3D geological modeling technology, a geothermal resource probability body is constructed, and the confidence level is calculated in combination with drilling verification data to delineate target areas of different levels.

[0013] Preferably, the normalization formula in step S1 is:

[0014] ;

[0015] Where, is the original eigenvalue, is the characteristic mean, is the standard deviation.

[0016] Preferably, the calculation formula of the geological structure complexity entropy value in step S2 is:

[0017] ;

[0018] Where, For the Probability density of geological features.

[0019] Preferably, the Moran index calculation formula in step S2 is:

[0020] ;

[0021] Where, is the number of samples, is the spatial weight matrix, is the sum of weights, 、 For location 、 The eigenvalues ​​of is the global mean.

[0022] Preferably, the feature splicing formula in step S3 is:

[0023] ;

[0024] Where, represents the remote sensing thermal infrared texture features extracted by CNN, represents the temporal characteristics of the geophysical field extracted by LSTM, represents the fault fractal dimension, Indicates the earthquake intensity, It represents the ratio of the earth's heat flow value to the Moho surface depth, reflecting the intensity of deep heat sources.

[0025] Preferably, the adaptive weight calculation formula in step S4 is:

[0026] ;

[0027] Where, For the model Accuracy, is the area under the curve.

[0028] Preferably, the reinforcement learning reward function formula in step S5 is:

[0029] ;

[0030] Where, represents the number of samples, represents the predicted value, Represents the true value.

[0031] Preferably, the Otsu algorithm threshold calculation formula in step S5 is:

[0032] ;

[0033] Where, represents the candidate threshold, Indicates the threshold The corresponding two types of sample weights, represents the mean of the two types of samples.

[0034] Preferably, the confidence calculation formula in step S6 is:

[0035] ;

[0036] Where, represents the mean absolute error between the predicted value and the drill hole verification value, Indicates the maximum value (theoretical optimal value) of the geothermal probability body.

[0037] Preferably, the method further includes step S7: real-time data access and model update, which accesses geothermal well monitoring data in real time through edge computing nodes, updates model parameters every hour, and realizes dynamic target area correction.

[0038] This invention provides a method for intelligently locating deep geothermal target areas based on a machine learning algorithm. Compared with existing technologies, it has the following advantages:

[0039] 1. This intelligent positioning method for deep geothermal target areas based on machine learning algorithms integrates geological structure, geophysical field, geochemistry and remote sensing data through machine learning algorithms to construct high-dimensional feature vectors, which overcomes the limitations of existing technologies that rely solely on physical detection. At the same time, it introduces reinforcement learning to dynamically optimize model parameters, combined with real-time geothermal well data updates, so that the model can adapt to changes in geological conditions, and the prediction accuracy is greatly improved compared with traditional methods.

[0040] 2. This intelligent positioning method for deep geothermal targets based on machine learning algorithms establishes a complete quantitative analysis chain, from quantifying the entropy value of geological complexity to three-dimensional target confidence assessment. By integrating multiple models with adaptive weights and calculating the cost-effectiveness index, it comprehensively considers resource volume, drilling costs, and prediction reliability to form a scientific risk assessment system. Compared with the "anomaly superposition" strategy of existing technologies, the drilling hit rate is greatly improved and the exploration cost is significantly reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0043] See for example Figure 1 , the present invention provides the following two technical solutions:

[0044] The first implementation method: a method for intelligent positioning of deep geothermal target areas based on a machine learning algorithm, comprising the following steps:

[0045] S1. Multi-source heterogeneous data fusion acquisition and preprocessing: Collect geological structure data (fault distribution, lithology, etc.), geophysical field data (gravity, magnetism, geothermal flow), geochemical data (hot spring water chemical composition) and remote sensing thermal infrared data to construct a multi-source data set and eliminate dimensional differences through standardization.

[0046] The standardization formula is:

[0047] ;

[0048] Where, is the original eigenvalue, is the characteristic mean, is the standard deviation.

[0049] S2. Quantification of Geological Entropy and Spatial Autocorrelation Analysis: Calculate the entropy of geological structural complexity to quantify data uncertainty. Use the Moran index to analyze spatial autocorrelation and identify the spatial distribution patterns of geothermal anomalies. Dynamically adjust sampling density based on entropy (increasing to 50m in high-entropy areas) to optimize data acquisition efficiency.

[0050] The calculation formula of geological structure complexity entropy is:

[0051] ;

[0052] Where, For the Probability density of geological features.

[0053] The Moran Index calculation formula is:

[0054] ;

[0055] Where, is the number of samples, is the spatial weight matrix, is the sum of weights, 、 For location 、 The eigenvalues ​​of is the global mean.

[0056] S3. Deep feature extraction and multimodal feature fusion: Convolutional neural networks (CNNs) are used to extract texture features from remote sensing thermal infrared images, and long short-term memory networks (LSTMs) are used to extract time series features from geophysical fields. Deep learning features are then combined with geophysical indicators (fault fractal dimension, earthquake intensity, etc.) to achieve multimodal fusion and construct high-dimensional feature vectors.

[0057] The feature splicing formula is:

[0058] ;

[0059] Where, represents the remote sensing thermal infrared texture features extracted by CNN, represents the temporal characteristics of the geophysical field extracted by LSTM, represents the fault fractal dimension, Indicates the earthquake intensity, It represents the ratio of the earth's heat flow value to the Moho surface depth, reflecting the intensity of deep heat sources.

[0060] S4. Multi-model adaptive integration and dynamic weight optimization: Build a model group including random forest (RF), gradient boosting machine (GBM), and Transformer, and dynamically adjust the model contribution based on the adaptive weight mechanism. The Transformer model uses a multi-head attention mechanism to capture the complex correlation between data.

[0061] The adaptive weight calculation formula is:

[0062] ;

[0063] Where, For the model Accuracy, is the area under the curve.

[0064] S5. Dynamic adjustment of reinforcement learning to determine anomaly thresholds: Introduce reinforcement learning algorithms to optimize model parameters, using prediction error as the reward function; use the Otsu algorithm to automatically determine geothermal anomaly thresholds and achieve unsupervised segmentation of anomaly areas.

[0065] The reinforcement learning reward function formula is:

[0066] ;

[0067] Where, represents the number of samples, represents the predicted value, Represents the true value.

[0068] The Otsu algorithm threshold calculation formula is:

[0069] ;

[0070] Where, represents the candidate threshold, Indicates the threshold The corresponding two types of sample weights, represents the mean of the two types of samples.

[0071] S6. 3D geothermal target area intelligent positioning and risk assessment: Based on 3D geological modeling technology, a geothermal resource probability body is constructed. The confidence level is calculated in combination with drilling verification data. Target areas of different levels are delineated (Level I: C ≥ 0.7; Level II: 0.5 ≤ C < 0.7). The cost-effectiveness index is calculated to guide drilling decisions.

[0072] The confidence calculation formula is:

[0073] ;

[0074] Where, represents the mean absolute error between the predicted value and the drill hole verification value, Indicates the maximum value (theoretical optimal value) of the geothermal probability body.

[0075] Step S7: Real-time data access and model update: geothermal well monitoring data is accessed in real time through edge computing nodes, and model parameters are updated every hour to achieve dynamic target area correction and adapt to the time-varying characteristics of the geothermal system.

[0076] The second implementation method:

[0077] This example uses a sedimentary basin in North my country as the research area. The target horizon is Mesoproterozoic carbonate rocks, buried at a depth of approximately 3,000-4,000 meters. The study area has complex geological structures and well-developed faults. The geothermal resource potential is large, but exploration is difficult.

[0078] Step 1: Multi-source heterogeneous data fusion collection and preprocessing (S1)

[0079] Data collection:

[0080] Geological structure data: 1:50,000 geological map, fault distribution map, lithology histogram;

[0081] Geophysical data: gravity anomaly data (resolution 1km), magnetic anomaly data (resolution 1km), and geothermal heat flow data (30 measurement points);

[0082] Geochemical data: chemical composition analysis (pH, TDS, trace elements, etc.) of 20 hot spring water samples;

[0083] Remote sensing data: Landsat 8 thermal infrared band data (resolution 30m).

[0084] Data preprocessing:

[0085] Use standardized processing formulas to eliminate dimensional differences;

[0086] Kriging interpolation is used to fill missing values;

[0087] Principal component analysis (PCA) was used for dimensionality reduction, and five principal components were extracted (cumulative contribution rate > 85%).

[0088] Step 2: Quantification of geological characteristic entropy and spatial autocorrelation analysis (S2)

[0089] Geological complexity calculation:

[0090] Calculate the fault density, lithologic diversity and other parameters, and substitute them into the geological structure complexity entropy calculation formula to calculate the geological structure complexity entropy value Hg;

[0091] The results showed that Hg=0.92 (high complexity) in the southwest of the study area and Hg=0.56 (low complexity) in the northeast.

[0092] Spatial autocorrelation analysis:

[0093] The Moran index I calculated using the Moran index formula for the terrestrial heat flow is 0.63 (significant positive correlation), indicating that the geothermal anomaly is distributed in a clustered manner.

[0094] The sampling density is adjusted according to the entropy value. The sampling interval in high entropy areas is increased to 50m, and that in low entropy areas is increased to 200m.

[0095] Step 3: Deep feature extraction and multimodal feature fusion (S3)

[0096] CNN feature extraction:

[0097] Construct a 5-layer CNN model to process remote sensing thermal infrared images and extract texture features Ft;

[0098] The convolution kernel size decreases from 7×7 to 3×3, and the pooling layer adopts maximum pooling.

[0099] LSTM feature extraction:

[0100] To process geophysical field time series data (such as annual changes in geothermal flux), the number of LSTM hidden layer nodes is calculated according to the formula Determined to be 8;

[0101] Extracting time series features , reflecting the dynamic changes of the geothermal system.

[0102] Feature fusion:

[0103] Will 、 Combined with geophysical indicators (fault fractal dimension, earthquake intensity, etc.) to form a 50-dimensional feature vector .

[0104] Step 4: Multi-model adaptive integration and dynamic weight optimization (S4)

[0105] Model construction:

[0106] Random Forest (RF): number of decision trees 100, maximum depth 8;

[0107] Gradient Boosting Machine (GBM): learning rate 0.1, number of trees 500;

[0108] Transformer: embedding dimension 64, number of attention heads 8.

[0109] Weight optimization:

[0110] The adaptive weight calculation formula is used to calculate the weight of each model. In the initial stage, the RF weight is 0.4, the GBM weight is 0.3, and the Transformer weight is 0.3;

[0111] As the training iterates, the Transformer weight gradually increases to 0.5, indicating that it has a stronger ability to represent complex geothermal systems.

[0112] Step 5: Reinforcement learning dynamic tuning to participate in abnormal threshold determination (S5)

[0113] Parameter optimization:

[0114] State space S: model parameters (such as learning rate, tree depth) and historical error Et−1;

[0115] Action space A: parameter adjustment step size ;

[0116] Reward function: Adopts the reinforcement learning reward function formula, with the goal of maximizing the reward (minimizing the prediction error);

[0117] After 1000 iterations of Q-learning, the model MAE dropped from the initial 0.85°C to 0.52°C.

[0118] Abnormal threshold determination:

[0119] The Otsu threshold value Th = 75 mW / m² was calculated using the Otsu algorithm threshold calculation formula, and the area with a terrestrial heat flow value > 75 mW / m² was divided into a geothermal anomaly area.

[0120] Step 6: 3D geothermal target area intelligent positioning and risk assessment (S6)

[0121] 3D Modeling:

[0122] Kriging interpolation was used to construct a three-dimensional geological model with an interpolation accuracy of RMSE = 0.32 km;

[0123] Combined with drilling data, the geothermal resource probability body P(x,y,z) is constructed.

[0124] Confidence calculation:

[0125] Substituting the confidence calculation formula to calculate the confidence C, C=0.82 in the northwest of the study area (Level I target area), and C=0.65 in the southeast (Level II target area).

[0126] risk assessment:

[0127] Calculate the cost-effectiveness index R, R=2.3 for level I target area and R=1.5 for level II target area;

[0128] Level I target areas are recommended for drilling verification.

[0129] Step 7: Real-time data access and model update (S7)

[0130] Edge computing deployment:

[0131] Deploy five edge computing nodes in the study area to collect geothermal well temperature and pressure data in real time;

[0132] The data transmission frequency is once every hour.

[0133] Model Update:

[0134] Adopting an incremental learning strategy, the model parameters are updated every 24 hours;

[0135] After six months of continuous operation, the model prediction error decreased by 12%.

[0136] Three verification wells were deployed in the study area, and the results are as follows:

[0137] Well V1 in the Level I target area: bottomhole temperature is 152°C, geothermal gradient is 4.8°C / 100m, and the error with the predicted value is less than 5%;

[0138] Well V2 in the Level II target area: bottomhole temperature 128°C, geothermal gradient 4.2°C / 100m, with an error of <8% from the predicted value;

[0139] Well V3 in the non-target area: bottomhole temperature was 95°C, and geothermal gradient was 3.1°C / 100m, which verified the reliability of the model.

[0140] Through the deep integration of machine learning algorithms and geological knowledge, this invention can improve the accuracy and efficiency of deep geothermal target positioning, providing strong technical support for the efficient development of geothermal resources.

[0141] At the same time, the contents not described in detail in this specification belong to the existing technology known to those skilled in the art, and the model parameters of each electrical appliance are not specifically limited, and conventional equipment can be used.

[0142] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0143] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A deep geothermal target area intelligent positioning method based on machine learning algorithm, characterized in that: The following steps are involved: S1. Multi-source heterogeneous data fusion acquisition and preprocessing: Collect geological structure data, geophysical field data, geochemical data, and remote sensing thermal infrared data to construct a multi-source data set and eliminate dimensional differences through standardization; S2. Quantification of geological characteristic entropy and spatial autocorrelation analysis: Calculate the entropy of geological structural complexity to quantify data uncertainty, and use the Moran index to analyze spatial autocorrelation; The calculation formula for the geological structure complexity entropy value in step S2 is: Where p i is the probability density of the i-th type of geological features; S3. Deep feature extraction and multimodal feature fusion: Convolutional neural networks are used to extract texture features from remote sensing thermal infrared images, long short-term memory networks are used to extract geophysical field time series features, and multimodal fusion is achieved through feature splicing. S4. Multi-model adaptive integration and dynamic weight optimization: Build a model group including random forest, gradient boosting machine, and Transformer, and dynamically adjust the model contribution based on the adaptive weight mechanism; The adaptive weight calculation formula in step S4 is: Where, ACC i is the accuracy of model i, AUC i is the area under the curve; S5. Dynamically tune the participation of reinforcement learning in determining anomaly thresholds: Introduce reinforcement learning algorithms to optimize model parameters, using prediction error as the reward function; use the Otsu algorithm to determine geothermal anomaly thresholds; S6. 3D geothermal target area intelligent positioning and risk assessment: Based on 3D geological modeling technology, a geothermal resource probability body is constructed, and the confidence level is calculated in combination with drilling verification data to delineate target areas of different levels.

2. The deep geothermal target area intelligent positioning method based on machine learning algorithm according to claim 1 is characterized by: The normalization formula in step S1 is: Where X is the original eigenvalue, μ is the eigenvalue mean, and σ is the standard deviation.

3. The deep geothermal target area intelligent positioning method based on machine learning algorithm according to claim 1 is characterized by: The calculation formula of the Moran index in step S2 is: Where n is the number of samples, w is ij is the spatial weight matrix, S0 is the sum of weights, X i 、X j is the eigenvalue of position i and j, is the global mean.

4. The deep geothermal target area intelligent positioning method based on machine learning algorithm according to claim 1 is characterized by: The feature splicing formula in step S3 is: Where, F t represents the remote sensing thermal infrared texture features extracted by CNN, F s represents the temporal features of the geophysical field extracted by LSTM, D s represents the fault fractal dimension, v represents the earthquake intensity, It represents the ratio of the earth's heat flow value to the Moho surface depth, reflecting the intensity of deep heat sources.

5. The deep geothermal target area intelligent positioning method based on machine learning algorithm according to claim 1 is characterized by: The reinforcement learning reward function formula in step S5 is: Where N represents the number of samples, Represents the predicted value, y i Represents the true value.

6. The deep geothermal target area intelligent positioning method based on machine learning algorithm according to claim 1 is characterized by: The Otsu algorithm threshold calculation formula in step S5 is: T h =argmax t [ω0(t)·μ0(t)+ω1(t)·μ1(t)]; Where t represents the candidate threshold, ω0(t) and μ0(t) represent the weights of the two types of samples corresponding to the threshold t, and ω1(t) and μ1(t) represent the means of the two types of samples.

7. The deep geothermal target area intelligent positioning method based on machine learning algorithm according to claim 1 is characterized by: The confidence calculation formula in step S6 is: Where MAE represents the mean absolute error between the predicted value and the borehole verification value, and max(P(x, y, z)) represents the maximum value of the geothermal probability body.

8. The deep geothermal target area intelligent positioning method based on machine learning algorithm according to claim 1 is characterized by: It also includes step S7: real-time data access and model update, which accesses geothermal well monitoring data in real time through edge computing nodes, updates model parameters every hour, and realizes dynamic target area correction.

Images