Step-down mining capacity prediction method of natural gas hydrate by deep neural network
A deep neural network model addresses the limitations of conventional methods by accurately predicting natural gas hydrate depressurization mining capacity, enhancing mining efficiency and adaptability through data preprocessing and optimization.
Patent Information
- Application Number
- JP2024134872
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2024-01-04
- Filing Date
- 2024-08-13
- Publication Date
- 2025-07-16
- Estimated Expiration
- 2044-08-13
AI Technical Summary
Conventional methods for predicting the production capacity of natural gas hydrate depressurization mining are limited by their accuracy and adaptability, particularly due to complex non-linear relationships and diverse indoor mining test data, which hinders efficient mining strategy adjustment and optimization.
A deep neural network model is constructed with five layers to process multi-scale natural gas hydrate pressure reduction mining test data, incorporating data preprocessing and optimization through the backpropagation algorithm to achieve high accuracy and adaptability in predicting production capacity.
The method significantly improves prediction accuracy and adaptability, reducing resource requirements and costs, enabling real-time, efficient mining strategy optimization and enhancing production efficiency.
Smart Images
Figure 2025106783000001_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of gas hydrate mining technology, and in particular, it is an innovative prediction method based on a deep neural network for predicting the production capacity of gas hydrates by a pressure reduction mining method.
Background Art
[0002] Natural gas hydrate is an ice-like solid compound formed from natural gas and water under high pressure and low temperature conditions. It has many advantages such as wide distribution, abundant reserves, shallow burial, high energy efficiency, and low pollution, and is regarded as one of the most ideal alternative energy sources in the post-oil era. Among many methods for mining natural gas hydrates, the pressure reduction mining method and its improvement plans are the most potential and are widely recognized as one of the routes to achieve commercial mining. Deeply understanding the production capacity situation of mining with different mining parameters under pressure reduction is helpful for adjusting the mining strategy and fully exploiting the potential of hydrate resources. Therefore, accurately predicting the production capacity plays an extremely important role in optimizing the mining process and improving efficiency.
[0003] However, on-site mining projects of natural gas hydrates usually require long-term preparation, high costs, and high risks. Therefore, the current on-site tests are relatively limited, and most research work has to rely on indoor tests. In addition, due to the diversity and complexity of indoor mining test data, it has become difficult to capture the laws and accurately predict the production capacity with conventional analysis methods. At the same time, the complex non-linear relationship between mining parameters and pressure reduction mining production capacity also limits the accuracy and applicability of existing prediction methods. There are certain limitations in both simple linear regression based on empirical formulas and complex calculation methods based on physical models. The deep neural network, a sophisticated machine learning technique, has demonstrated clear advantages in handling complex data and non-linear relationships. Its central advantage lies in the ability to automatically learn and extract complex features in the data by constructing a multi-layer neural network. Therefore, there is an urgent need for a highly efficient and smart method for predicting the production capacity of natural gas hydrate depressurization mining. Compared with conventional prediction methods, the present invention proposes a prediction method based on a deep neural network that can better understand the complex relationship between production capacity and various mining parameters and significantly improve the prediction accuracy. At the same time, because it has the ability to automatically learn features, this method can better adapt to various different natural gas hydrate mining environments and does not need to rely on complex artificial parameter inputs and calculations.
Summary of the Invention
Problems to be Solved by the Invention
[0004] To address the deficiencies in the accuracy and adaptability of conventional methods for predicting the production capacity of natural gas hydrate depressurization mining, the present invention proposes an innovative prediction method based on a deep neural network. This method can achieve accurate prediction of production capacity for multi-scale natural gas hydrate depressurization mining tests under different conditions. This is helpful for us to adjust the mining strategy, optimize the mining process, and significantly improve the mining efficiency.
Means for Solving the Problems
[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for predicting the pressure reduction mining capacity of natural gas hydrate by a deep neural network is proposed. The core of this method is to construct a neural network model with five layers of depth, and use the existing multi-scale natural gas hydrate pressure reduction mining test data to train, optimize, test and evaluate the model. Furthermore, based on this model, a production capacity prediction module is developed to realize the accurate prediction of the production capacity of multi-scale natural gas hydrate pressure reduction mining tests under various different conditions. This process not only helps to adjust the mining strategy and optimize the mining process, but also can effectively improve the mining efficiency. The specific implementation procedures include the following: S1: Collect multi-scale natural gas hydrate pressure reduction mining test data. These test data cover discrete features and continuous features. Among them, the discrete features include sample scale information and mining wellbore information, and the continuous features include mining parameters and production capacity data. S2: Pretreatment of the test data. S3: Construct a deep neural network model including one input layer, three hidden layers and one output layer. S4: Use the pressure reduction mining test data of multi-scale and different mining parameters to train the deep neural network model to accurately capture the complex modes and non-linear relationships in the test data, and optimize the network parameters through the backpropagation algorithm. S5: Evaluate the calculation accuracy of the model by using two parameters, the coefficient of determination and the mean relative error, so as to obtain a production capacity prediction model with high accuracy and small error. S6: Install the production capacity prediction model to design a production capacity prediction module, and use the production capacity prediction module to predict the production capacity of natural gas hydrate pressure reduction mining tests under different scales and mining parameters. Furthermore, in step S1, the sample scale information includes one-dimensional, two-dimensional, and three-dimensional three-scale sample scale information, the mining well information includes horizontal well, vertical well, and horizontal well + vertical well mining well information, the mining parameters include sample volume, porosity, gas hydrate saturation, pre-mining pressure, post-mining pressure, initial mining temperature, and pressure drop rate, and the production capacity data includes cumulative gas production and average gas production rate. Furthermore, the method for data preprocessing in step S2 is as follows: S2.1: Clean the collected data, remove potential outliers and noise, and obtain standardized test data. S2.2: Set 5% of each mining parameter as the floating range, expand the number of test samples, and perform sampling using the stratified random sampling method. Let the total number of samples be N. S2.3: Perform normalization processing on the test data to obtain nine types of features, and divide the normalized dataset into a training set, a validation set, and a test set. S2.4: Divide the normalized dataset into a 70% training set, a 15% validation set, and a 15% test set according to the splitting ratio. Furthermore, the method for test data normalization processing in step 2.3 is as follows: S2.3.1: For discrete features, formally normalize the sample scale information and the mining well information using one-hot encoding: Sample scale information: one-dimensional = (1, 0, 0), two-dimensional = (0, 1, 0), three-dimensional = (0, 0, 1); Mining well information: horizontal well = (1, 0, 0), vertical well = (0, 1, 0), horizontal well + vertical well = (0, 0, 1); S2.3.2: In the case of continuous features, perform normalization processing by scaling the data of the mining parameters to the range of [0, 1]. The formula is as follows:
[0006]
Number
[0007] Here, x represents the seven types of mining parameters in step S1, min(x) represents the minimum value of the corresponding mining parameter in all the test samples extracted in step S2.2, max(x) represents the maximum value of the corresponding mining parameter in all the test samples extracted in step S2.2, and xn represents the normalized value. Furthermore, step S3 specifically includes the following steps: S3.1: Construct a five-layer deep neural network model including one input layer, three hidden layers, and one output layer. S3.2: The number of input layer nodes is nine, which exactly corresponds to a total of nine types of features including sample scale information, mining well information, and seven mining parameters respectively. S3.3: The number of output layer nodes is two, which respectively corresponds to the two types of production capacity data in step S1. S3.4: Set the number of nodes per hidden layer in the deep neural network model to 60. Furthermore, step S4 specifically includes the following steps: S4.1: Initialize the deviation of the model to zero and randomly initialize the weights of the model using a normal distribution. The formula is as follows: W~N(0,0.012) In the mathematical formula, W is the weight matrix, and N(0,0.012) is a normal distribution with a mean value of 0 and a standard deviation of 0.01. S4.2: Select Leaky ReLU as the activation function. The formula is as follows:
[0008]
Number
[0009]
Number
[0010]
Number
[0011]
Number
[0012]
Number
[0013] Furthermore, the detailed procedure of step S6 is as follows: In step S5, a production capacity prediction module is designed by installing a production capacity prediction model with high fitting accuracy and few errors, real-time acquisition of gas hydrate pressure reduction mining test data is performed, and the test data is input into the production capacity prediction module according to the data format normalized in S2.3 to obtain a production capacity prediction value, thereby providing accurate multi-scale gas hydrate pressure reduction mining test production capacity information in real time and further providing data reference for on-site applications.
Advantages of the Invention
[0014] In connection with all the above-described technical aspects, the beneficial effects of the present invention include the following: (1) The production capacity prediction method for natural gas hydrate pressure reduction mining based on the deep neural network provided by the present invention greatly reduces the large amount of human and material resources required for conventional natural gas hydrate pressure reduction mining tests, and also effectively reduces the burden of laboratory work and data analysis by the automatic learning and pattern recognition capabilities of the deep learning model, providing more efficient experimental means for researchers. (2) The production capacity prediction method for pressure reduction mining based on the deep neural network of natural gas hydrate pressure reduction mining provides support for the optimal mining strategy for the pressure reduction mining project, and production capacity prediction information can be obtained immediately due to the generalization ability and real-time nature of the model, thereby adjusting and optimizing the pressure reduction mining operation, improving production efficiency, and reducing costs. (3) The production capacity prediction method for natural gas hydrate pressure reduction mining based on the deep neural network provided by the present invention has higher prediction accuracy compared to conventional methods, and the model can better capture the complex relationships in natural gas hydrate pressure reduction mining tests, providing highly reliable production capacity prediction support for the pressure reduction mining project and serving scientific decision-making and resource planning.
Brief Description of the Drawings
[0015]
Figure 1
Figure 2
Figure 3
Embodiments for Carrying out the Invention
[0016] Hereinafter, the construction process of the embodiments of the present invention will be described in detail in connection with the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art and the protection scope of the present invention can be more clearly and precisely defined. As shown in FIG. 1, the present invention is an example of a method for predicting the production capacity of natural gas hydrate depressurization mining based on a deep neural network, which collects engineering multi-scale natural gas hydrate depressurization mining test data and performs preprocessing, and constructs a deep neural network model including one input layer, three hidden layers, and one output layer. Through the depressurization mining test data of different mining parameters at three scales, the deep neural network model is trained to accurately capture the complex modes and non-linear relationships in the test data, and the network parameters are optimized through the backpropagation algorithm. The calculation accuracy of the model is evaluated using two parameters, the coefficient of determination and the mean relative error; a production capacity prediction module is designed to predict the production capacity of the natural gas hydrate depressurization mining test at different scales and mining parameters. The main steps of this method are as follows: TIFF2025106783000012.tif42166S2: Preprocessing of data S2.1: Clean the collected data, remove potential outliers and noise, obtain standard test data, and merge the data with data fluctuations caused by test operations, etc. within the error tolerance range, as shown in Table 1.
[0017]
Table 1
[0018]
Equation
[0019]
Equation
[0020]
Number
[0021]
Number
[0022]
Number
[0023]
Number
[0024]
Number
[0025] Of course, the above description does not limit the present invention, and the application scope of the present invention is not limited to the above specific examples. Changes, deformations, additions or substitutions made by experts in the technical field are regarded as part of the protection scope of the invention within the basic principles and essential scope of the invention.
Claims
1. A method for predicting the production capacity of natural gas hydrate by pressure reduction mining based on a deep neural network, comprising the following steps: The following steps: S1: Collect multi-scale experimental data of natural gas hydrate pressure reduction mining, covering discrete and continuous features. The discrete features include sample scale information and mining wellbore information, and the continuous features include mining parameters and production capacity data. In the method for predicting the production capacity of natural gas hydrate by pressure reduction mining based on the described deep neural network, in step S1, the sample scale information includes one-dimensional, two-dimensional, and three-dimensional three-scale sample scale information. The mining well information includes horizontal well, vertical well, and horizontal well + vertical well mining well information. The mining parameters include sample volume, porosity, gas hydrate saturation, pre-mining pressure, post-mining pressure, initial mining temperature, and pressure reduction rate. The production capacity data includes cumulative gas production and average gas production rate. S2: Preprocess the test data; the following steps: S2.1: Clean the collected data, remove potential outliers and noise, and obtain standardized test data. S2.2: Set 5% of each mining parameter as the floating range, expand the number of test samples, and perform sampling using the stratified random sampling method, with the total number of samples being N. S2.3: Perform normalization processing on the test data to obtain nine types of features, and divide the normalized dataset into a training set, a validation set, and a test set: S2.3.1: For discrete features, formally normalize the sample scale information and mining well information using one-hot encoding. Sample scale information: One-dimensional = (1, 0, 0), two-dimensional = (0, 1, 0), three-dimensional = (0, 0, 1); Mining well information: Horizontal well = (1, 0, 0), vertical well = (0, 1, 0), horizontal well + vertical well = (0, 0, 1); S2.3.2: In the case of continuous features, perform normalization processing by scaling the data of the mining parameters to the range of [0, 1]. The formula is as follows: 【Number 1】 Here, x represents the seven types of mining parameters in step S1, min(x) represents the minimum value of the corresponding mining parameter in all the test samples extracted in step S2.2, max(x) represents the maximum value of the corresponding mining parameter in all the test samples extracted in step S2.2, and x ｎ is the normalized value; S2.4: Divide the normalized dataset into a 70% training set, a 15% validation set, and a 15% test set according to the splitting ratio. S3: Construct a deep neural network model including one input layer, three hidden layers, and one output layer; as follows: S3.1: Construct a five-layer deep neural network model including one input layer, three hidden layers, and one output layer. S3.2: The number of input layer nodes is 9, which exactly corresponds to a total of 9 types of features including sample scale information, mining well information, and 7 mining parameters; S3.3: The number of nodes in the output layer is 2, which respectively corresponds to the 2 types of production capacity data in step S1; S3.4: Set the number of nodes per hidden layer in the deep neural network model to 60; S4: Utilize the pressure reduction mining test data of different mining parameters at multiple scales to train the deep neural network model to accurately capture the complex patterns and non-linear relationships in the test data, and optimize the network parameters through the backpropagation algorithm; S4.1: Initialize the deviation of the model to zero, and randomly initialize the weights of the model using a normal distribution. The formula is as follows: 【Number 2】 In the formula, W is the weight matrix, and N(0, 0.01 ２ ) is a normal distribution with a mean of 0 and a standard deviation of 0.01; S4.2: Select Leaky ReLU as the activation function. The formula is as follows: 【Mathematics 3】 In the formula, α is 0.01, and X ｉ is the nine types of features in step S2.3; S4.3: Use MSE as the loss function to measure the square of the difference between the predicted production capacity value and the actual value obtained using the training set in step S2.
4. The formula is as follows: 【Number 4】 【Number】 S4.4: Utilize the chain rule to inversely calculate the gradient of the loss with respect to the weights and biases of the model parameters starting from the output layer, and use the optimization algorithm of batch gradient descent to realize the update of the parameters. The formula is as follows: 【Number 5】 In the formula, ∇ θ J(θ) is the gradient calculation value of the loss function with respect to the model parameters, θ is the parameter of the deep neural network model, β is the learning rate, and the initial value is 0.01; S4.5: Apply a penalty to the weights of the model using L2 regularization technology to prevent the weights from being too large and the model from overfitting the training data. The formula is as follows: 【Number 6】 where, J Ｌ２ is the loss function after adding L2 for regularization, λ is the strength of regularization, the initial value is 0.1, and Σ i ∥ω i ∥ 2 is the sum of the squares of all the weights of the model; S4.6: Evaluate the performance of the model on the validation set and adjust the learning rate and regularization strength based on the performance on the validation set; S4.7: Repeat steps S4.3 to S4.6 multiple times until the model achieves satisfactory performance; S5: Evaluate the calculation accuracy of the model using two parameters, the coefficient of determination and the mean relative error; S6: Install the production capacity prediction model to design a production capacity prediction module, and use the production capacity prediction module to predict the production capacity of the natural gas hydrate pressure reduction mining test at different scales and mining parameters; having A method for predicting the pressure reduction mining production capacity of natural gas hydrate based on a deep neural network, characterized by the above.
2. In step S5, the following procedure: Evaluate the generalization performance of the model trained and optimized using the test set for unknown data, and determine the coefficient of determination R ２ Evaluate the calculation accuracy of the model using the criteria that the coefficient of determination R ２ is at least 0.8 or more and the average relative error E is less than 10% to obtain a model with high fitting accuracy and low error; the expressions for the coefficient of determination R 【Number 7】 【Number】 R ２ As R approaches 1, the fitting accuracy of the model increases, and as E approaches 0, the error between the predicted production capacity value of the model and the actual production capacity value decreases. having A method for predicting the production capacity of natural gas hydrate by pressure reduction mining based on the deep neural network according to claim 1.
3. In step S6, the following procedure: Install a model with high fitting accuracy and low error in step S5 to design a production capacity prediction module, obtain gas hydrate pressure reduction mining test data in real time, input it into the production capacity prediction module according to the data format after normalization in S2.3 to obtain a production capacity prediction value, provide accurate multi-scale gas hydrate pressure reduction mining test production capacity information in real time, and further provide data reference for on-site application having A method for predicting the production capacity of natural gas hydrate by pressure reduction mining based on the deep neural network according to claim 2.