A method for mineral prediction and exploration risk assessment based on bayesian deep learning

Abstract

The application provides a kind of mineral forecast and exploration risk evaluation method of bayesian deep learning, it is related to mineral exploration technical field, method includes: constructing multi-source geological data set and pre-processing, obtaining deposit and ore-controlling element;Through data preprocessing, construct training data set, and then construct the training model of bayesian neural network based on Metropolis-Hastings sampling;Based on the training model, input all ore-controlling element data set, calculate the probability mean, contingency uncertainty estimate and cognitive uncertainty estimate, and draw the prediction probability graph, contingency uncertainty graph and cognitive uncertainty graph, carry out deposit exploration risk and prospecting potential evaluation.The application adopts the method based on Monte Carlo Metropolis-Hastings sampling, provides more reliable metallogenic probability mapping and uncertainty estimation, and can effectively distinguish cognitive uncertainty and contingency uncertainty.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of mineral exploration, and in particular to a mineral prediction and exploration risk evaluation method based on Bayesian deep learning. BACKGROUND

[0002] With the decrease in the number of outcrop deposits and easily found deposits, it is urgent to develop an advanced, reliable and easily interpreted method to extract and integrate deep mineralization information characteristics and find extremely potential prospecting target areas. However, the mineralization process is a complex physical and chemical process. Due to the variability of geological phenomena, the complexity of deposit genesis, the concealment of ore-controlling factors, the multi-solution of prospecting information, the uneven distribution of exploration data, the incompleteness of data collection, and the cognitive limitations of describing geological body properties and geological processes, all of these factors will lead to uncertainty in deposit prediction results.

[0003] Although the deep learning technology has very superior performance in intelligent prediction and evaluation of mineral resources, it also has serious defects. For example, machine learning is a deterministic model and cannot estimate the uncertainty of the prediction results, which loses the guiding significance for the risk assessment of mineral exploration. Moreover, the machine learning model is not robust, and slight changes in weights can affect the prediction results. Mineral prediction is a work of making the best decision under multiple uncertain conditions. Scientific understanding of these uncertainties is of great significance for reducing the uncertainty of mineral prediction and the risk of deposit exploration. SUMMARY

[0004] The purpose of the present application is to solve the problem of low prediction accuracy of the existing prospecting target prediction method, and to provide a mineral prediction and exploration risk evaluation method based on Bayesian deep learning.

[0005] The above-mentioned purpose of the present application is realized by the following technical solutions:

[0006] S1: Construct a multi-source geological data set and preprocess it to obtain a deposit and ore-controlling elements, and construct an ore-controlling element data set;

[0007] S2: Construct a Bayesian neural network training model based on Metropolis-Hastings sampling;

[0008] S3: Input the ore-controlling element data set into the Bayesian neural network training model to obtain the probability mean, accidental uncertainty estimation and cognitive uncertainty estimation;

[0009] S4: Draw a prediction probability graph, an accidental uncertainty graph and a cognitive uncertainty graph by using the probability mean, the accidental uncertainty estimation and the cognitive uncertainty estimation, and evaluate the deposit exploration risk and the prospecting potential.

[0010] Optionally, step S1 comprises:

[0011] The multi-source geological data set comprises geological data, geochemical data, geographical data and remote sensing data;

[0012] The preprocessing comprises missing value processing, outlier processing and noise processing;

[0013] The preprocessed multi-source address data set is filtered to obtain the ore deposit and ore-controlling elements.

[0014] Optionally, step S2 comprises:

[0015] The network layers, the number of neurons in each layer and the activation function of the Bayesian neural network training model are set .

[0016] The prior distribution of the weight and bias of the Bayesian neural network training model is set, specifically as follows:

[0017] The ore deposit and ore-controlling elements are taken as input data, the input data of the Bayesian neural network training model is defined as , the output label is , and the training data is .

[0018] The neural network parameters are The probability distribution of the prediction output of the Bayesian neural network training model is represented as follows:

[0019]

[0020]

[0021]

[0022] In the formula, is the output prediction of the Bayesian neural network training model, is the activation function, are the weight matrix and bias vector of the i-th layer of the Bayesian neural network training model respectively.

[0023] Optionally, step S2 further comprises:

[0024] The likelihood function of the Bayesian neural network training model is set, and the specific steps are as follows:

[0025] The occurrence of the ore deposit is marked as 1, and the non-occurrence of the ore deposit is marked as 0, and each sample belongs to one of the two categories;

[0026] The label of each sample is set as ​0 or 1, the prediction probability output by the Bayesian neural network training model representing a sample probability of belonging to class 1

[0027] The likelihood function of the binomial distribution is defined as:

[0028]

[0029]

[0030] wherein, is the true label of the sample ; represents the probability of taking 0 is 0, the probability of taking 1 is ; the probability of taking 0 is , the probability of taking 1 is 0;

[0031] The likelihood function of the binomial distribution is converted into a log-likelihood function:

[0032]

[0033] Taking the negative of the log-likelihood, the negative log-likelihood, i.e. the loss function, is obtained , as follows:

[0034] .

[0035] Optionally, step S3 comprises:

[0036] S31: setting the initial neural network parameters of the Bayesian neural network training model ;

[0037] Selecting a proposal distribution , specifically, adopting a Gaussian distribution as the prior distribution;

[0038]

[0039] wherein, is the covariance matrix, and is a diagonal matrix; represents a Gaussian distribution function; represents a generated parameter sample; represents the parameter of the i-th iteration ;

[0040] S32: performing Metropolis-Hastings sampling to generate a parameter sample;

[0041] For each iteration of the sampling process, the generated parameter sample is as follows:

[0042]

[0043]

[0044] wherein is the total number of iterations of the sampling process;

[0045] The acceptance rate is calculated as as follows:

[0046]

[0047] wherein, represents the likelihood function of the label data under the condition ; represents the prior probability distribution of the parameter ; represents the probability density of generating from the proposal distribution , given ; represents the likelihood function of the label data under the condition , based on the sampled parameter in the last iteration; represents the prior probability distribution of the parameter ; represents the probability density of generating from the proposal distribution , given ;

[0048] The acceptance of the parameter sample is determined according to the following steps:

[0049] A uniformly distributed random number is generated, and if , the parameter sample is accepted, i.e.:

[0050]

[0051] If , the parameter sample is rejected, i.e.:

[0052]

[0053] S33: Approximate the posterior probability distribution using the generated parameter sample: ​​​

[0054]

[0055] wherein, denotes the training data; denotes the given parameter under the condition that the observation data is the likelihood function; denotes the prior probability of the parameter ;

[0056] S34: save the parameter sample set obtained by sampling ;

[0057] S35: make model prediction through the parameter sample set, calculate the probability mean of the prediction probability and the accidental uncertainty estimate;

[0058] S36: calculate the variance of the prediction probability under different parameter samples to obtain the cognitive uncertainty estimate.

[0059] Optionally, step S35 includes:

[0060] The calculation steps of the probability mean are as follows:

[0061] For new input data , use all the sampled parameter samples to perform forward propagation to obtain a series of prediction probabilities , and the probability mean after M times of sampling is :

[0062]

[0063] The calculation steps of the accidental uncertainty estimate are as follows:

[0064] By predicting the variance at the output layer of the Bayesian neural network training model, the accidental uncertainty estimate is obtained as follows:

[0065]

[0066] wherein, the variance reflects the noise of the input data itself; denotes the probability obtained from the input data ; denotes the average probability obtained from the input data .

[0067] Optionally, step S36 includes:

[0068] Calculate the variance of the prediction probability under different parameter samples:

[0069]

[0070] wherein, denotes the parameter sample at the t-th sampling; denotes the probability of the average.

[0071] Optionally, the step S4 comprises:

[0072] drawing a prediction probability map based on the probability mean value, to form a mineral potential distribution map;

[0073] drawing a chance uncertainty map based on the chance uncertainty estimate, to identify areas with relatively large data noise;

[0074] drawing a cognitive uncertainty map based on the cognitive uncertainty estimate, to identify areas with insufficient knowledge or data scarcity;

[0075] combining the mineral potential distribution map, the areas with relatively large data noise and the areas with insufficient knowledge or data scarcity, to assist in mineral exploration decision-making, to determine priority exploration areas and areas requiring further data acquisition, and to complete deposit exploration risk and prospecting potential evaluation.

[0076] An electronic device comprising a processor, a memory, a user interface and a network interface, the memory being configured to store instructions, the user interface and the network interface being configured to communicate with other devices, and the processor being configured to execute the instructions stored in the memory to cause the electronic device to perform a method for mineral prediction and exploration risk evaluation based on Bayesian deep learning.

[0077] A computer-readable storage medium storing instructions that, when executed, perform a method for mineral prediction and exploration risk evaluation based on Bayesian deep learning.

[0078] The technical scheme provided by the present application has the beneficial effects of:

[0079] 1. Using the Metropolis-Hastings method based on Monte Carlo, the posterior probability distribution is approximated by multiple sampling, and then the model parameters and related uncertainty are estimated, and the mineral prediction provides a robust uncertainty estimate. Specifically, under the Bayesian framework, the neural network structure and prior distribution are constructed based on the knowledge of the field experts and the characteristics of the input data; based on Bayesian inference, the prior probability and the likelihood probability are combined to construct the posterior distribution function; based on the Metropolis-Hastings sampling result, the probability mean, accidental uncertainty and cognitive uncertainty are calculated; finally, the probability mean graph, accidental uncertainty graph and cognitive uncertainty graph are drawn and explained. The technical scheme of the present application combines deep learning technology and Bayesian inference idea, realizes the uncertainty measurement and simulation of the prediction result, and provides support for the risk assessment of mineral exploration.

[0080] 2. The hidden layer in the Bayesian neural network training model structure adopts three full connection layers, the prior distribution is defined as a normal distribution to simplify the model calculation complexity, and the likelihood function is defined as a binomial distribution, which is consistent with the characteristics of deposit prediction and classification.

[0081] 3. Different posterior probability approximation methods have a great influence on the uncertainty of the prediction result. The present application adopts the Monte Carlo Metropolis-Hastings sampling method based on Monte Carlo, provides more reliable uncertainty estimation, and can effectively distinguish cognitive uncertainty and accidental uncertainty. BRIEF DESCRIPTION OF DRAWINGS

[0082] The present application will be further described below in combination with the drawings and examples, wherein:

[0083] Figure 1 is the Bayesian neural network architecture in the embodiment of the present application;

[0084] Figure 2 is the average probability graph in the embodiment of the present application;

[0085] Figure 3 is the cognitive uncertainty graph in the embodiment of the present application;

[0086] Figure 4 is the cognitive uncertainty graph in the embodiment of the present application;

[0087] Figure 5 is the performance evaluation graph in the embodiment of the present application;

[0088] Figure 6 is the schematic diagram of the electronic device structure in the embodiment of the present application. DETAILED DESCRIPTION

[0089] In order to make the technical features, objectives and effects of the present application clearer, the specific embodiments of the present application will be described in detail with reference to the drawings.

[0090] The embodiment of the present application provides a mineral prediction and exploration risk evaluation method based on Bayesian deep learning.

[0091] Please refer to Figure 1 , Figure 1 FIG. 1 is a Bayesian neural network architecture diagram of a mineral prediction and exploration risk evaluation method based on Bayesian deep learning in the embodiment of the present application, comprising:

[0092] S1: Construct a multi-source geological data set and perform preprocessing to obtain a deposit and ore-controlling elements, and construct an ore-controlling element data set;

[0093] S2: Construct a Bayesian neural network training model based on Metropolis-Hastings sampling;

[0094] S3: Input the ore-controlling element data set into the Bayesian neural network training model to obtain a probability mean value, an accidental uncertainty estimate and a cognitive uncertainty estimate;

[0095] S4: Draw a prediction probability graph, an accidental uncertainty graph and a cognitive uncertainty graph through the probability mean value, the accidental uncertainty estimate and the cognitive uncertainty estimate, and perform deposit exploration risk and prospecting potential evaluation.

[0096] In an embodiment of the present application, the classification performance of the Bayesian neural network training model is evaluated using an ROC curve, an AUC value and the like.

[0097]

[0098] Wherein, TPR is the true positive rate, and FPR is the false positive rate.

[0099] Step S1 comprises:

[0100] The multi-source geological data set comprises geological data, geochemical data, geographical data and remote sensing data;

[0101] The preprocessing comprises missing value processing, abnormal value processing and noise processing;

[0102] The multi-source address data set after preprocessing is filtered to obtain the deposit and ore-controlling elements.

[0103] Step S2 comprises:

[0104] The network layer number, the number of neurons in each layer and the activation function of the Bayesian neural network training model are set ;

[0105] The prior distributions of the weights and biases for training the Bayesian neural network model are defined as follows:

[0106] Using ore deposits and ore-controlling factors as input data, the input data for training the Bayesian neural network model is defined as follows: The output labels are The training data is ;

[0107] The neural network parameters are The probability distribution of the predicted output of the Bayesian neural network training model is represented as follows:

[0108]

[0109]

[0110]

[0111] In the formula, To predict the output of a Bayesian neural network training model. For activation function, These are the first and second training models of the Bayesian neural network. The weight matrix and bias vector of the layer.

[0112] Step S2 also includes:

[0113] The likelihood function for training a Bayesian neural network model is defined by the following steps:

[0114] The presence of a mineral deposit is marked as 1, and the absence of a mineral deposit is marked as 0. Each sample belongs to one of the two categories.

[0115] Let each sample tags The value can be 0 or 1, representing the predicted probability output by the Bayesian neural network training model. Indicates sample The probability of belonging to category 1;

[0116] The likelihood function of the binomial distribution is defined as:

[0117]

[0118]

[0119] in, For the sample The true label; express The probability of taking the value 0 is 0. The probability of taking the value 1 is ; The probability of taking 0 is , The probability of taking 1 is 0;

[0120] Convert the likelihood function of the binomial distribution into a log-likelihood function:

[0121]

[0122] Take the negative of the log-likelihood to get the negative log-likelihood, which is the loss function , as follows:

[0123] .

[0124] As an embodiment, by minimizing the loss function, the difference between the probability distribution predicted by the Bayesian neural network training model and the real label distribution is measured to obtain a loss value; by the loss value, the neural network parameters of the Bayesian neural network training model are adjusted.

[0125] Step S3 includes:

[0126] S31: Set the initial neural network parameters of the Bayesian neural network training model ;

[0127] Select the proposal distribution , specifically: adopt a Gaussian distribution as the prior distribution;

[0128]

[0129] wherein, is a covariance matrix, and is a diagonal matrix; denotes a Gaussian distribution function; denotes the generated parameter sample; denotes the parameter of the th iteration;

[0130] S32: Perform Metropolis-Hastings sampling to generate a parameter sample;

[0131] For each iteration sampling process, the generated parameter sample is as follows:

[0132]

[0133]

[0134] wherein is the total number of iteration sampling processes;

[0135] Calculate the acceptance rate , as follows:

[0136]

[0137] where, denotes the likelihood function of the label data under the condition ; denotes the prior probability distribution of the parameter ; denotes the probability density of generating from the proposal distribution with as the condition; denotes the likelihood function of the label data under the condition based on the sampled parameter in the last iteration; denotes the prior probability distribution of the parameter ; denotes the probability density of generating from the proposal distribution with as the condition; denotes the likelihood function of the label data under the condition

[0138] based on the sampled parameter in the last iteration;

[0139] denotes the prior probability distribution of the parameter ; denotes the probability density of generating from the proposal distribution with

[0140] as the condition; if , accept

[0141] , i.e.

[0142]

[0143] S33: Approximate the posterior probability distribution with the generated parameter sample:

[0144]

[0145] where, denotes the training data; denotes the likelihood function of the observed data under the condition of the given parameter ; denotes the prior probability of the parameter ;

[0146] S34: Save the parameter sample set obtained by sampling ;

[0147] S35: Model prediction is performed through the parameter sample set, a probability mean of the prediction probability and a contingency uncertainty estimate are calculated;

[0148] S36: The variance of the prediction probability under different parameter samples is calculated to obtain a cognitive uncertainty estimate.

[0149] Step S35 includes:

[0150] The calculation step of the probability mean is as follows:

[0151] For new input data , forward propagation is performed using all sampled parameter samples to obtain a series of prediction probabilities , and the probability mean after M times of sampling is :

[0152]

[0153] The calculation step of the contingency uncertainty estimate is as follows:

[0154] The contingency uncertainty estimate is obtained by predicting the variance of the output layer of the Bayesian neural network training model , as follows:

[0155]

[0156] Wherein, the variance reflects the noise of the input data itself; represents the probability obtained from the input data ; represents the average probability obtained from the input data .

[0157] Step S36 includes:

[0158] The variance of the prediction probability under different parameter samples is calculated as follows:

[0159]

[0160] Wherein, represents the parameter sample of the t-th sampling; represents the average of the probability .

[0161] Step S4 includes:

[0162] A prediction probability graph based on the probability mean is drawn to form a mineral potential distribution map;

[0163] Draw an accidental uncertainty map based on accidental uncertainty estimation, identify areas with high data noise;

[0164] Draw a cognitive uncertainty map based on cognitive uncertainty estimation, identify areas with insufficient knowledge or data scarcity;

[0165] Combine the mineral potential distribution map, the area with high data noise, and the area with insufficient knowledge or data scarcity to assist in mineral exploration decision-making, determine the priority exploration area and the area that needs further data collection, and complete the deposit exploration risk and prospecting potential evaluation.

[0166] In an embodiment of the present application, based on the multi-source geoscience data set of the Nanling area, taking tungsten polymetallic metallogenic prediction and its uncertainty evaluation as an example, the implementation process of the technical method and its efficiency and practicality in practical application are illustrated. The present application will be further described below with reference to the accompanying drawings.

[0167] (1) Select 11 ore-controlling elements in the Nanling area as input data X, including northwest faults, northeast faults, east-west faults, south-north faults, fault density, granite, and 5 geochemical variables W, Sn, Mo, Bi and Ag.

[0168] (2) Collect the coordinates of the confirmed deposits in the study area as positive sample data, and randomly select the corresponding number of data points from the non-deposit area as negative sample data, which together constitute the input label data Y.

[0169] (3) Based on GIS technology, rasterize the data of ore-controlling elements and label data, with a grid unit of 1km x 1km, which is convenient for model processing and calculation;

[0170] (4) Set the number of network layers, the number of neurons, and the activation function, and record these learnable weights and biases as θ; give the prior distribution of θ , use the likelihood function of the binomial distribution , and construct the structure of the Bayesian neural network, the schematic diagram of which is shown in Figure 1 .

[0171] (5) Use Metropolis-Hastings for sampling, initialize the parameters , generate candidate parameters in the iteration process, calculate the acceptance rate , and decide whether to accept, and the generated sequence approximately obeys the target posterior distribution.

[0172] (6) Use the parameter sampling set to do forward propagation on the new sample , get the prediction probability distribution, take its mean value, and get the average probability distribution map of tungsten polymetallic ore prediction in the Nanling area, as shown inFigure 2 The variance is calculated to quantify the cognitive uncertainty of the tungsten polymetallic ore prediction result, as shown in Figure 3 The output noise parameter can represent the accidental uncertainty of the tungsten polymetallic ore prediction result, as shown in Figure 4 .

[0173] (7) The classification performance of the Bayesian neural network training model based on Metropolis-Hastings sampling is evaluated using ROC curve, AUC value, etc., as shown in Figure 5 The model has excellent prediction ability.

[0174] The application also discloses an electronic device. Referring to Figure 6 , Figure 6 is a structural schematic diagram of an electronic device disclosed by the embodiment of the application. The electronic device 500 can include at least one processor 501, at least one network interface 504, a user interface 503, a memory 505, and at least one communication bus 502.

[0175] The communication bus 502 is used to realize the connection and communication between the components.

[0176] The user interface 503 can include a display screen, and the optional user interface 503 can further include a standard wired interface and a wireless interface.

[0177] The network interface 504 can optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0178] The application also discloses a computer readable storage medium, which stores a plurality of instructions adapted to be loaded by a processor to execute the above-mentioned mineral prediction and exploration risk evaluation method based on Bayesian deep learning.

[0179] The above are only exemplary embodiments of the present disclosure, and cannot limit the scope of the present disclosure. That is, any equivalent changes and modifications made in accordance with the teachings of the present disclosure are still within the scope of the present disclosure.

[0180] The present application is intended to cover any variations, uses, or adaptive changes of the present disclosure that follow the general principles of the present disclosure and include common knowledge or conventional technical means in the technical field not described in the present disclosure. The scope and spirit of the present disclosure are defined by the claims. The specification and examples are only considered as exemplary, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A method for mineral prospectivity and exploration risk assessment using Bayesian deep learning, characterized in that, The method comprises the following steps: S1: constructing a multi-source geological data set and preprocessing to obtain a deposit and ore-controlling elements, and constructing an ore-controlling element data set; S2: constructing a Bayesian neural network training model based on Metropolis-Hastings sampling; Step S2 comprises: setting the number of network layers, the number of neurons in each layer, and the activation function softmax of the Bayesian neural network training model; setting the prior distribution of the weight and bias of the Bayesian neural network training model, specifically as follows: taking the deposit and ore-controlling elements as input data, defining the input data of the Bayesian neural network training model as x, the output label as y, and the training data as D={x, y}; the probability distribution of the prediction output of the Bayesian neural network training model with neural network parameters θ is represented as follows: p(y | x, 0) = softmax(f θ (x)) f θ (x) = w L g(w L-1 g(…g(w 1 x + b 1 ))+ b L-1 )+ b L where f θ (x) is the output prediction of the Bayesian neural network training model, g(·) is an activation function, w l ,b l are the weight matrix and bias vector of the l-th layer of the Bayesian neural network training model, respectively. Step S2 further comprises: setting the likelihood function of the Bayesian neural network training model, and the specific steps are as follows: marking the occurrence of the deposit as 1 and the non-occurrence of the deposit as 0, and each sample belongs to one of the two categories; Let each sample x i have a label y i that takes on the value 0 or 1, and let p i be the predicted probability output by a Bayesian neural network trained model that represents the probability that the sample x i belongs to class 1; the likelihood function of the binomial distribution is defined as: p i = p(y i | x i , θ) where y i ∈ {0,1} is the true label of sample i; denotes y i 0 with probability 0, y i 1 with probability p i ; 0 with probability (1-p i ), y i 1 with probability 0. the likelihood function of the binomial distribution is converted into a log-likelihood function: log[p(y i |x i , θ)] = y i ·log(p i ) + (1 - y i )·log(p i ) taking the negative of the log-likelihood to obtain the negative log-likelihood, that is, the loss function L, as follows: l = -log[p(y i |x i , θ)] = -y i ·log(p i ) - (1 - y i )·log(p i ) S3: inputting the ore-controlling element data set into the Bayesian neural network training model to obtain a probability mean, an accidental uncertainty estimate, and a cognitive uncertainty estimate; S4: drawing a prediction probability graph, an accidental uncertainty graph, and a cognitive uncertainty graph through the probability mean, the accidental uncertainty estimate, and the cognitive uncertainty estimate to evaluate the deposit exploration risk and the prospecting potential.

2. The method of claim 1, wherein the method is a Bayesian deep learning method for mineral prospectivity and exploration risk assessment. Step S1 comprises: The multi-source geological data set comprises: geological data, geochemical data, geographic data, and remote sensing data; The preprocessing comprises: missing value processing, abnormal value processing, and noise processing; The preprocessed multi-source address data set is filtered to obtain the deposit and ore-controlling elements.

3. The method of claim 1, wherein the method is a Bayesian deep learning method for mineral prospectivity and exploration risk assessment. Step S3 comprises: S31: set initial neural network parameters θ of the Bayesian neural network training model (0) ; Selecting the proposal distribution q(θ' | θ (n) ), specifically: using a Gaussian distribution N(0, σ 2 ) as the prior distribution; wherein∑ is a covariance matrix, and I is a diagonal matrix; denotes a Gaussian distribution function; θ' denotes generated parameter samples; θ (n) denotes the parameter of the n-th iteration; S32: Metropolis-Hastings sampling is performed to generate parameter samples; For each iteration sampling process, the generated parameter samples are as follows: θ' ~ q(θ' | θ (n-1) ) n=1,2,…,M Where M is the total number of iteration sampling processes; The acceptance rate α is calculated as follows: Where p(y|x,θ′) represents the likelihood function of the label data y under condition x with parameter θ′; p(θ′) represents the prior probability distribution of parameter θ′; q(θ (n-1) |θ′) represents generating θ from the proposal distribution q with θ′ as a condition. (n-1) The probability density of p(y|x,θ) (n-1) ) indicates that the sampling parameters θ were obtained in the previous iteration. (n-1) Based on this, the likelihood function of the labeled data y under condition x; p(θ) (n-1) ) represents the parameter θ (n-1) The prior probability distribution of q(θ′|θ (n-1) ) represents the number of proposals from the proposal distribution q, with θ (n-1) Generate the probability density of θ′ conditionally; determine whether to accept the parameter sample, and the specific steps are as follows: Generating a uniformly distributed random number If μ < α, accept θ', i.e. θ (n) = θ' if μ>α, then reject θ', that is: θ (n) = θ (n-1) S33: approximating the posterior probability distribution p(θ|D) through the generated parameter samples: p(θ|D)∝p(D|θ)p(θ) Where D represents the training data; p(D|θ) represents the likelihood function of the observed data D given the parameter θ; p(θ) represents the prior probability of the parameter θ; S34: Save the parameter sample set {θ} obtained from sampling. (1) ,θ (2) ,…,θ (M) }; S35: model prediction is performed through the parameter sample set to calculate the probability mean of the prediction probability and the accidental uncertainty estimate; S36: the variance of the prediction probability under different parameter samples is calculated to obtain the cognitive uncertainty estimate.

4. The method of claim 3, wherein the method is characterized by, Step S35 comprises: The calculation steps of the probability mean are as follows: For new input data x * , forward propagation is performed with all sampled parameter samples {θ (1) , θ (2) ,..., θ (M)} to obtain a series of prediction probabilities p(y * | x * , θ (t) ). The probability mean after M times of sampling is: The calculation steps of the accidental uncertainty estimate are as follows: by predicting the variance σ at the output layer of the Bayesian neural network trained model 2 (x * ), resulting in an aleatoric uncertainty estimate L(x * ,y * ), as follows: where the variance σ 2 (x * ) reflects the noise in the input data x * itself; y * represents the probability resulting from the input data x * ; represents the average probability resulting from the input data x * .

5. The method of claim 3, wherein the method is characterized by, Step S36 comprises: calculating the variance of the prediction probability under different parameter samples: where θ (t) denotes the parameter sample at the t-th sampling; denotes the probability y * of the average.

6. The method of claim 1, wherein the method is a Bayesian deep learning method for mineral prospectivity and exploration risk assessment. Step S4 comprises: drawing a prediction probability graph based on the probability mean to form a mineral potential distribution map; drawing an aleatory uncertainty map based on the aleatory uncertainty estimation, identifying areas with high data noise; drawing a cognitive uncertainty map based on the cognitive uncertainty estimation, identifying areas with insufficient knowledge or data scarcity; combining the mineral potential distribution map, the areas with high data noise and the areas with insufficient knowledge or data scarcity, assisting the decision of mineral exploration, determining the priority exploration areas and the areas requiring further data acquisition, and completing the evaluation of the deposit exploration risk and the prospecting potential.

7. An electronic device, comprising: The electronic device comprises a processor, a memory, a user interface and a network interface. The memory is configured to store instructions. The user interface and the network interface are configured to communicate with other devices. The processor is configured to execute the instructions stored in the memory, so that the electronic device performs the method according to any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores instructions, and when the instructions are executed by a computer, the method according to any one of claims 1-6 is performed.

Images