Geothermal resource prediction method and device, system, and storage medium

Abstract

The application discloses a geothermal resource prediction method and device, system and storage medium, and comprises the following steps: constructing a geothermal resource development sample library; using a supervised learning model to extract objective weights of geothermal system features; discretizing continuous features at three scales of fine, medium and coarse while retaining original continuous values; at each scale, performing Bayesian posterior estimation on a mixed representation of the sample based on normal-inverse Wishart prior and Dirichlet-multinomial prior, obtaining a category prototype defined by discrete features and a corresponding target variable posterior prediction distribution, and constructing a lookup table; calculating the similarity of a target area and historical categories within a scale, and performing multi-scale fusion by learning scale weights through data driving to obtain a comprehensive similarity; and weighting and mixing the posterior prediction distribution of each category by using the comprehensive similarity as a weight to generate a prediction distribution of the target variable. The application solves the problems of strong subjectivity, lack of uncertainty quantification and unstable model in the prior art.

Description

Technical Field

[0001] This invention belongs to the field of geothermal resource exploration technology, specifically relating to a geothermal resource prediction method, device, system, and storage medium. Background Technology

[0002] Geothermal resources, as a clean and renewable energy source, are characterized by high risks and high investment in exploration and development. Currently, geothermal resource assessment mainly relies on expert experience and qualitative analogy methods, which suffer from strong subjectivity, poor interpretability, and difficulty in quantifying uncertainty. Although machine learning methods are gradually being applied in resource prediction, under conditions of small samples, high-dimensional features, and strong geological constraints, the stability, interpretability, and uncertainty assessment capabilities of the models still fall short of practical needs. Summary of the Invention

[0003] To address the problems existing in the prior art, this invention provides a geothermal resource prediction method, apparatus, system, and storage medium, aiming to solve the following problems in the prior art:

[0004] (1) The assignment of values ​​is highly subjective;

[0005] (2) Lack of geological interpretability and quantification of uncertainty;

[0006] (3) Organically integrate data-driven methods with expert knowledge;

[0007] (4) The prediction model is unstable under high-dimensional feature conditions.

[0008] To achieve the above objectives, the present invention provides the following solution:

[0009] A method for predicting geothermal resources, comprising:

[0010] Step S1: Construct a geothermal resource development sample library, which includes: geothermal system characteristics and corresponding target variables of multiple historical geothermal wells or geothermal fields;

[0011] Step S2: Use a supervised learning model to extract the importance of geothermal system features, evaluate their stability through cross-validation and resampling, and normalize the evaluation results to obtain objective feature weights.

[0012] Step S3: Discretize the continuous feature at three scales: fine, medium, and coarse. At each scale, divide the continuous value into ordered levels according to a preset threshold to obtain the discrete feature at that scale, while retaining the original continuous value of the continuous feature.

[0013] Step S4: At each scale, perform Bayesian modeling on the sample mixture representation composed of continuous target variables and discrete features to obtain the category prototype defined by the combination of discrete features and its corresponding posterior prediction distribution of the target variable; wherein, a normal-inverse Wissaud prior is used for continuous target variables, a Dirichlet-multinomial prior is used for discrete features, and a queryable posterior prediction distribution lookup table is constructed for the target variable.

[0014] Step S5: Calculate the intra-scale similarity between the target area and each historical category, and perform multi-scale fusion by learning the weights of each scale through data-driven learning to obtain the comprehensive similarity between the target area and each historical category; wherein, the intra-scale similarity between the target area and the historical category is calculated using the posterior prediction distribution; or the intra-scale similarity between the target area and the historical category is calculated using objective feature weights and weighted Mahalanobis distance.

[0015] Step S6: Using comprehensive similarity as the weight, perform weighted mixing of the posterior prediction distributions of each category to generate the prediction distribution of the target variable, wherein the prediction distribution of the target variable includes the predicted mean, variance, and prediction interval of geothermal resources.

[0016] The present invention also provides a geothermal resource prediction device, comprising:

[0017] The first processing module is used to construct a geothermal resource development sample library, which includes: geothermal system characteristics and corresponding target variables of multiple historical geothermal wells or geothermal fields; wherein the geothermal system characteristics are continuous characteristics.

[0018] The second processing module is used to extract the importance of geothermal system features using a supervised learning model, evaluate its stability through cross-validation and resampling, and normalize the evaluation results to obtain objective feature weights.

[0019] The third processing module is used to discretize the continuous features at three scales: fine, medium, and coarse. At each scale, the continuous values ​​are divided into ordered levels according to a preset threshold to obtain the discrete features at that scale, while retaining the original continuous values ​​of the continuous features.

[0020] The fourth processing module performs Bayesian modeling on the sample mixture representation composed of continuous target variables and discrete features at each scale to obtain the category prototype defined by the combination of discrete features and its corresponding posterior prediction distribution of the target variable; wherein, a normal-inverse Wissaud prior is used for continuous features and a Dirichlet-multinomial prior is used for discrete features, and a queryable posterior prediction distribution lookup table is constructed for the target variable.

[0021] The fifth processing module is used to calculate the intra-scale similarity between the target area and each historical category, and to perform multi-scale fusion by learning the weights of each scale through data-driven learning to obtain the comprehensive similarity between the target area and each historical category; wherein, the intra-scale similarity between the target area and the historical category is calculated using the posterior prediction distribution; or the intra-scale similarity between the target area and the historical category is calculated using objective feature weights and weighted Mahalanobis distance.

[0022] The sixth processing module is used to perform weighted mixing of the posterior prediction distributions of each category with the comprehensive similarity as the weight, and generate the prediction distribution of the target variable, wherein the prediction distribution of the target variable includes the predicted mean, variance and prediction interval of geothermal resources.

[0023] The present invention also provides a geothermal resource prediction system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes the above-described geothermal resource prediction method when executed by the processor.

[0024] The present invention also provides a storage medium storing a computer program, which executes the above-described geothermal resource prediction method when running.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0026] By objectively extracting feature weights through machine learning, subjective human assignment is avoided; a hybrid framework of Bayesian prototype learning and predictive distribution is introduced to achieve the organic integration of data-driven and expert knowledge; the output results have good probabilistic interpretability, can clearly identify the dominant heat-controlling factors and quantify uncertainty; it is suitable for geothermal resource evaluation scenarios with small samples, high-dimensional features and strong geological constraints, and significantly improves prediction accuracy and decision reliability. Attached Figure Description

[0027] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart of the geothermal resource prediction method according to an embodiment of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] Example 1

[0032] like Figure 1 As shown, the present invention provides a geothermal resource prediction method, comprising:

[0033] Step S1: Construct a geothermal resource development sample library. The sample library includes geothermal system characteristics and corresponding target variables (resource quantity) from multiple historical geothermal wells or geothermal fields. The geothermal system characteristics, such as heat source, reservoir, caprock, and other related parameters, are represented as continuous numerical vectors during input to facilitate unified processing in subsequent models. For features that are inherently discrete (such as lithology), numerical encoding is required first.

[0034] Step S2: Use a supervised learning model to extract the importance of geothermal system features, evaluate stability through cross-validation and resampling, and then normalize to obtain objective feature weights;

[0035] Step S3: Discretize the continuous geothermal system features obtained in Step S1 into three scales: fine, medium, and coarse. By dividing the continuous values ​​into ordered levels according to a preset threshold at each scale, the discrete features at that scale are obtained. At the same time, the original continuous values ​​of the continuous features are retained.

[0036] Step S4: At each scale, perform Bayesian posterior estimation on the mixture representation of the samples to obtain the category prototype defined by the combination of discrete features and its corresponding posterior prediction distribution of the target variable; wherein, the category prototype refers to a group of samples with similar feature combinations identified from historical samples through clustering or unsupervised learning; the mixture representation of the samples includes: continuous target variable and discrete features; for continuous target variable, a normal-inverse Wissaud prior is used to obtain the posterior prediction distribution; for discrete features, a Dirichlet-multinomial prior is used to obtain the posterior distribution; a queryable posterior prediction distribution lookup table is constructed for the target variable for fast querying during final prediction to improve computational efficiency;

[0037] Step S5: Calculate the intra-scale similarity between the target area and the historical categories, and perform multi-scale fusion by learning scale weights through data-driven learning to obtain a comprehensive similarity; wherein, the target area refers to a new area with geothermal system characteristics but unknown target variables; the historical categories are the category prototypes mentioned in step S4; calculate the intra-scale similarity between the target area and the historical categories using the posterior prediction distribution; or calculate the intra-scale similarity between the target area and the historical categories using objective feature weights and weighted Mahalanobis distance;

[0038] Step S6: Using the comprehensive similarity as the weight, perform a weighted mixing of the posterior prediction distributions of each historical category to generate the prediction distribution of the target variable. The prediction distribution of the target variable includes the predicted mean, variance, and prediction interval of geothermal resources.

[0039] As one embodiment of the present invention, in step S1, the geothermal system characteristics include: heat source and thermal background characteristics, thermal reservoir characteristics, caprock characteristics, fluid and channel characteristics, and geophysical and geochemical response characteristics; the target variable is the amount of resources.

[0040] As one embodiment of the present invention, step S2 includes:

[0041] Step S21: Calculate the importance of random forest for each geothermal system feature j. absolute value of LASSO coefficient The average value across folds was obtained through cross-validation (K-fold, R-fold resampling);

[0042] Step S22: Perform robust summaries after standardization by z-scores:

[0043] ;

[0044] in, For the first The preliminary importance score of a geothermal system feature is a scalar value that combines the evaluation results of two models. The fusion weight coefficient is a hyperparameter between 0 and 1, used to adjust the relative weight of the random forest importance and the LASSO coefficient in the final score; The first one calculated using the random forest model The importance of each feature (such as the amount of Gini impurity reduction or the decrease in accuracy). The first result obtained by LASSO (L1 regularized linear model) regression is the... The absolute value of the coefficient of each feature reflects the strength of the feature's contribution to the target variable; This is the Z-score standardization function. It transforms the input values ​​into a standard normally distributed variable with a mean of 0 and a standard deviation of 1, aiming to eliminate... and The difference in dimensions makes them comparable.

[0045] Step S23: Select frequency based on features With coefficient of variation As a stability constraint, a stability-weighted score is constructed: and to The unstable characteristics were eliminated;

[0046] in, The score is calculated by weighting the stability of the j-th feature; In order to achieve the desired result in multiple resampling operations during cross-validation, the first... The frequency with which a feature is selected by the model; A minimum frequency threshold is selected for the features; features with frequencies below this threshold are considered unstable and are removed. For the first Feature importance score The coefficient of variation between each fold of cross-validation is calculated by dividing the standard deviation by the mean. This is the maximum allowable threshold for the coefficient of variation; features exceeding this value are considered unstable and are discarded. This is an indicator function; its value is 1 when the condition inside the parentheses is true, and 0 otherwise.

[0047] Step S24, for Softmax or L1 normalization is performed to obtain for

[0048] ;

[0049] or ;

[0050] in, For the first The final objective feature weight of a geothermal system feature is a normalized value, and the sum of the weights of all features is 1. The temperature parameter is the one used in the Softmax function. This is used to control the "sharpness" of the weight distribution. The larger the value, the higher the weight of the important features, and the more concentrated their distribution. To find the sum index, iterate through all features.

[0051] As one embodiment of the present invention, in step S3, the scale... s The above is based on industry / national standard thresholds or data distribution thresholds. Divide into ordered levels and define level mappings:

[0052] ;

[0053] in, In scale Above, regarding the first Mapping function for each feature; For the first The original values ​​of a continuous feature; The discrete ordered levels obtained after mapping (e.g., 1, 2, 3, …); , These are predefined thresholds used to divide different grade ranges; The expression "if and only if" indicates that the conditions on both sides of the equation are equivalent.

[0054] In one embodiment of the present invention, in step S4...

[0055] Continuous features employ the normal-inverse Wissaud prior NIW (Normal-Inverse Wissaud Prior). Its posterior sufficient statistics are:

[0056] ;

[0057] ;

[0058] ;

[0059] in, This represents the number of samples belonging to category prototype C; This is the mean vector of all samples in category C on continuous features; Let C be the within-class scatter matrix. This measures the degree of dispersion of samples within a class. Let be the hyperparameters of the normal-inverse Wissaud prior, and represent the prior mean, prior equivalent sample size, prior degrees of freedom, and prior scale matrix, respectively. It is a sufficient statistic for the posterior distribution; after observation After a certain number of samples, the prior parameters are updated to these posterior parameters.

[0060] Using the prior scale matrix Dimensional scaling is performed based on the weights of objective features:

[0061] ;

[0062] here Based on objective feature weights It is a diagonal matrix with diagonal elements. It is a basic scale matrix.

[0063] The posterior prediction distribution of continuous features is a multivariate student Distribution, that is:

[0064] ;

[0065] in, For a given category prototype At that time, the posterior prediction distribution of the new continuous feature vector; For diverse students Degrees of freedom of the distribution; degrees of freedom This ensures that the distribution has a finite covariance matrix; The dimension of continuous features.

[0066] Discrete features employ a Dirichlet-polynomial prior, and the size of the value space for the m-th discrete feature is denoted as . Intra-class count is Then the posterior prediction distribution of the discrete features is:

[0067] ;

[0068] in, For a given category prototype , No. The discrete feature takes values ​​of The posterior probability; For the first A discrete feature; For the first The number of all possible values ​​for a discrete feature; For the category prototype In the middle, the first The discrete feature takes values ​​of Sample count; The hyperparameters of the Dirichlet prior can be understood as the counts. The “pseudo-counting” or a priori belief.

[0069] As one embodiment of the present invention, step S5 includes:

[0070] Step S51: Calculate the intra-scale similarity between the target region and historical categories using the posterior prediction distribution as follows:

[0071] ;

[0072] Intra-scale normalized similarity:

[0073] ;

[0074] in, In scale Above, target area With the A historical category prototype The non-normalized scale similarity; its value is the target region feature in The probability density or quality under the posterior predictive distribution; In order to scale superior, The posterior prediction distribution function; For the target area In scale The feature representation below (which may be a discretized hierarchy or a raw continuous value); In order to scale Above, target area With the indivual The normalized intra-scale similarity is such that its sum across all historical categories is 1, facilitating cross-scale comparison and fusion.

[0075] Step S52: Calculate the intra-scale similarity between the target region and historical categories using objective feature weights and weighted Mahalanobis distance as follows:

[0076] ;

[0077] ;

[0078] In scale Above, target area With the Each category prototype The weighted Mahalanobis distance; the smaller the distance, the higher the similarity. For target area With History Category The feature intersection; For category prototype In the The posterior mean over each feature; For category prototype In the Posterior variance over each feature; For scale Temperature parameters; used to control the flatness of the similarity distribution; The smaller the value, the sharper the similarity distribution, meaning the more prominent the maximum similarity. For the target region X and the category prototype In scale The similarity within the scale is calculated; the summation index k in the denominator iterates through all historical class prototypes.

[0079] Step S53: Multi-scale fusion is performed by learning scale weights through data-driven learning to obtain the comprehensive similarity as follows:

[0080] ;

[0081] Among them, scale weight ≥0, and .

[0082] As one embodiment of the present invention, step S6 includes:

[0083] Step S61: Convert the overall similarity S(X,i) into a mixed weight using the sharpening parameter β.

[0084] ;

[0085] satisfy ;

[0086] in, For the first The weights of the posterior predictions of each historical category prototype distributed in the final mixture model; These are the sharpening parameters; Sometimes, the weight of highly similar categories is amplified, making the final prediction more dependent on the most similar categories.

[0087] Step S62: Construct a similarity-weighted prediction distribution:

[0088] ;

[0089] in, The posterior predicted distribution of resource quantity for the class to which historical sample i belongs;

[0090] Step S63: Output the prediction statistics:

[0091] Mean:

[0092] ;

[0093] in, The posterior distribution of the target variable for the i-th category prototype. The mean, =E[ ];

[0094] variance:

[0095] ;

[0096] in, The posterior distribution of the target variable for the i-th category prototype. The variance.

[0097] The prediction interval (the central (1-α) interval) is [ , ],in, =F (-1) (α / 2), = F (-1) (1-α / 2).

[0098] The cumulative distribution function of the predicted distribution is obtained by weighted mixing of the cumulative distribution functions of the prototype distributions of each category.

[0099] ;

[0100] in, The final predicted cumulative distribution function represents the distribution of features known. and sample library Under these conditions, resource quantity Less than or equal to The probability of; For the first A mixed weighting of each category prototype (historical analogy scenario); For the first The posterior cumulative distribution function of the resource quantity corresponding to each category prototype is derived from the lookup table or posterior prediction distribution constructed in step S4.

[0101] As one embodiment of the present invention, multi-scale fusion weights are learned by minimizing the continuous rank probability score (CRPS) and / or negative log-likelihood (NLL) through cross-validation. Specifically:

[0102] (1) Define the objective function as the mean of the Continuous Rank Probability Score (CRPS) as...

[0103] ;

[0104] or

[0105] ;

[0106] in, For use in learning scale weights The objective function (loss function); The number of samples in the validation set or training set; The continuous rank probability score for the i-th sample is an indicator for evaluating the accuracy of probability prediction; a smaller value is better.

[0107] ;

[0108] in, and These are two independent and identically distributed samples whose predicted distribution is given. The number of samples used for cross-validation; Let be the observed actual resource quantity for the i-th sample.

[0109] To in a given feature Sample library D and scale weights Below, observed values The log-likelihood is such that a larger value indicates a better prediction, therefore the negative log-likelihood (NLL) needs to be minimized.

[0110] The solution obtained through grid search, gradient descent, or Bayesian optimization is as follows:

[0111] ;

[0112] in, For scale weights; loss function Optimization solver.

[0113] This invention first constructs a sample library containing historical geothermal well / field features and target variable (resource quantity); it then objectively extracts feature weights using random forest or LASSO models; it discretizes continuous features at multiple scales to integrate expert knowledge while retaining the original continuous values; based on Bayesian prototype learning, it obtains category prototypes defined by combinations of discrete features and their corresponding posterior prediction distributions of target variables; it calculates the intra-scale similarity between the target area and historical categories using the posterior prediction distributions or weighted Mahalanobis distance, and performs multi-scale fusion through data-driven learning of scale weights to obtain a comprehensive similarity; finally, using this similarity as weight, it weights and mixes the posterior prediction distributions of each category to generate a complete prediction distribution of the target variable. This invention outputs the prediction mean, interval, and success probability, achieving an organic integration of data and knowledge, possessing high interpretability, strong generalization ability, and uncertainty assessment capability, and is suitable for geothermal resource evaluation under conditions of small samples and high-dimensional features.

[0114] Example 2

[0115] The present invention also provides a geothermal resource prediction device, comprising:

[0116] The first processing module is used to construct a geothermal resource development sample library, which includes: geothermal system characteristics and corresponding target variables of multiple historical geothermal wells or geothermal fields; wherein the geothermal system characteristics are continuous characteristics.

[0117] The second processing module is used to extract the importance of geothermal system features using a supervised learning model, evaluate its stability through cross-validation and resampling, and normalize the evaluation results to obtain objective feature weights.

[0118] The third processing module is used to discretize the continuous features at three scales: fine, medium, and coarse. At each scale, the continuous values ​​are divided into ordered levels according to a preset threshold to obtain the discrete features at that scale, while retaining the original continuous values ​​of the continuous features.

[0119] The fourth processing module is used to perform Bayesian posterior estimation on the sample mixture representation composed of continuous target variables and discrete features at each scale, to obtain the category prototype defined by the combination of discrete features and its corresponding posterior prediction distribution of the target variable; wherein, a normal-inverse Wissaud prior is used for continuous features, and a Dirichlet-multinomial prior is used for discrete features, and a queryable posterior prediction distribution lookup table is constructed for the target variable;

[0120] The fifth processing module is used to calculate the intra-scale similarity between the target area and each historical category, and to perform multi-scale fusion by learning the weights of each scale through data-driven learning to obtain the comprehensive similarity between the target area and each historical category; wherein, the intra-scale similarity between the target area and the historical category is calculated using the posterior prediction distribution; or the intra-scale similarity between the target area and the historical category is calculated using objective feature weights and weighted Mahalanobis distance.

[0121] The sixth processing module is used to perform weighted mixing of the posterior prediction distributions of each category with the comprehensive similarity as the weight, and generate the prediction distribution of the target variable, wherein the prediction distribution of the target variable includes the predicted mean, variance and prediction interval of geothermal resources.

[0122] Example 3

[0123] The present invention also provides a geothermal resource prediction system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program performs a geothermal resource prediction method when executed by the processor.

[0124] Example 4

[0125] The present invention also provides a storage medium storing a computer program, which executes a geothermal resource prediction method when running.

[0126] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for predicting geothermal resources, characterized in that, include: Step S1: Construct a geothermal resource development sample library, which includes: geothermal system characteristics and corresponding target variables of multiple historical geothermal wells or geothermal fields; Step S2: Use a supervised learning model to extract the importance of geothermal system features, evaluate their stability through cross-validation and resampling, and normalize the evaluation results to obtain objective feature weights. Step S3: Discretize the continuous features at three scales: fine, medium, and coarse. At each scale, the continuous values ​​are divided into ordered levels according to a preset threshold to obtain the discrete features at that scale, while retaining the original continuous values ​​of the continuous features. Step S4: At each scale, perform Bayesian modeling on the sample mixture representation composed of continuous target variables and discrete features to obtain the category prototype defined by the combination of discrete features and its corresponding posterior prediction distribution of the target variable; wherein, a normal-inverse Wissaud prior is used for continuous target variables, a Dirichlet-multinomial prior is used for discrete features, and a queryable posterior prediction distribution lookup table is constructed for the target variable. Step S5: Calculate the intra-scale similarity between the target region and each historical category, and perform multi-scale fusion by learning the weights of each scale through data-driven learning to obtain the comprehensive similarity between the target region and each historical category; wherein, the intra-scale similarity between the target region and the historical category is calculated using the posterior prediction distribution; or the intra-scale similarity between the target region and the historical category is calculated using objective feature weights and weighted Mahalanobis distance. Step S6: Using comprehensive similarity as the weight, perform weighted mixing of the posterior prediction distributions of each category to generate the prediction distribution of the target variable, wherein the prediction distribution of the target variable includes the predicted mean, variance, and prediction interval of geothermal resources.

2. A geothermal resource prediction device, characterized in that, include: The first processing module is used to construct a geothermal resource development sample library, which includes: geothermal system characteristics and corresponding target variables of multiple historical geothermal wells or geothermal fields; wherein the geothermal system characteristics are continuous characteristics. The second processing module is used to extract the importance of geothermal system features using a supervised learning model, evaluate its stability through cross-validation and resampling, and normalize the evaluation results to obtain objective feature weights. The third processing module is used to discretize the continuous features at three scales: fine, medium, and coarse. At each scale, the continuous values ​​are divided into ordered levels according to a preset threshold to obtain the discrete features at that scale, while retaining the original continuous values ​​of the continuous features. The fourth processing module performs Bayesian modeling on the sample mixture representation composed of continuous target variables and discrete features at each scale to obtain the category prototype defined by the combination of discrete features and its corresponding posterior prediction distribution of the target variable; wherein, a normal-inverse Wissaud prior is used for continuous target variables, and a Dirichlet-multinomial prior is used for discrete features, and a queryable posterior prediction distribution lookup table is constructed for the target variable; The fifth processing module is used to calculate the intra-scale similarity between the target area and each historical category, and to perform multi-scale fusion by learning the weights of each scale through data-driven learning to obtain the comprehensive similarity between the target area and each historical category; wherein, the intra-scale similarity between the target area and the historical category is calculated using the posterior prediction distribution; or the intra-scale similarity between the target area and the historical category is calculated using objective feature weights and weighted Mahalanobis distance. The sixth processing module is used to perform weighted mixing of the posterior prediction distributions of each category with the comprehensive similarity as the weight, and generate the prediction distribution of the target variable, wherein the prediction distribution of the target variable includes the predicted mean, variance and prediction interval of geothermal resources.

3. A geothermal resource prediction system, characterized in that, include: A memory and a processor, wherein the memory stores a computer program executed by the processor, the computer program performing the geothermal resource prediction method as described in claim 1 when executed by the processor.

4. A storage medium, characterized in that, The storage medium stores a computer program, which executes the geothermal resource prediction method as described in claim 1 when it runs.

Images