Method and system for predicting recoverable reserves of unconventional oil and gas single well and computer equipment

Abstract

The invention provides an unconventional oil and gas single well recoverable reserve prediction method and system, computer equipment and a storage medium. The method comprises the steps that multiple basic characteristics and recoverable reserves of a sample well are obtained; determining the importance of each basic feature relative to the recoverable reserves; screening the basic features for multiple times according to the importance to obtain a plurality of input variable combinations; training the reserve prediction model by using the corresponding sample data to obtain a reserve prediction model to be selected; and analyzing the accuracy of the to-be-selected reserve prediction models, selecting the to-be-selected reserve prediction model with the highest accuracy as an optimal recoverable reserve prediction model, and selecting the input variable combination input in the to-be-selected reserve prediction model with the highest accuracy as an optimal input variable combination. According to the method, importance evaluation and screening are carried out on the basic features, the input parameters of the reserve prediction model are optimized, and the recoverable reserve prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas resource development, and in particular to a method, system, computer equipment and storage medium for predicting recoverable reserves of unconventional oil and gas single wells. Background Art

[0002] Current methods for predicting recoverable reserves of unconventional oil and gas resources per well can be divided into two categories: formula-based predictions based on theory and experience, and machine learning-based methods. Unconventional oil and gas recoverable reserves are influenced by numerous and complex factors, including geological factors (lithofacies, mineral composition, TOC content, maturity, burial depth, gas content, reservoir brittleness, reservoir porosity, reservoir permeability, and reservoir sensitivity) and engineering factors (horizontal well length, main stratum penetration rate, completion method, number of fracturing stages, fracturing fluid volume, and fracturing sand addition). Due to the complex influence of each factor on recoverable reserves and the complex interactions between these factors, current theoretical understanding and experience have made it difficult to develop a well-adaptable and highly accurate single-well oil and gas recoverable reserves prediction method. In comparison, machine learning methods offer greater adaptability in this regard. Current machine learning-based methods for recoverable reserves primarily involve data preprocessing, establishing labeled datasets, training machine learning algorithms, and utilizing these models for prediction. However, in the process of using machine learning algorithms to predict unconventional oil and gas recoverable reserves, due to the large number of factors affecting unconventional oil and gas recoverable reserves and the complex relationships, there is a lack or inadequacy of the importance analysis and screening process of the input variables, resulting in the prediction accuracy of the final model not reaching the optimal level; in addition, its methods all predict recoverable reserves based on data from a specific development stage, and have poor adaptability to different stages. Summary of the Invention

[0003] Based on this, it is necessary to provide a method, system, computer equipment and storage medium for predicting recoverable reserves of unconventional oil and gas single wells in response to the above technical problems.

[0004] A method for predicting recoverable reserves of unconventional oil and gas single wells, comprising:

[0005] Obtain multiple basic characteristics and recoverable reserves sample data required for sample well recoverable reserves prediction;

[0006] Determining the significance of each of the basic characteristics relative to the recoverable reserves;

[0007] The basic features are screened multiple times according to the importance, and the basic features screened out each time are used as an input variable combination, so as to obtain multiple input variable combinations after multiple screenings;

[0008] Constructing a reserve prediction model based on a machine learning algorithm, using each combination of the input variables as an input of the reserve prediction model, using recoverable reserves as an output of the reserve prediction model, and training the reserve prediction model using corresponding sample data to obtain a candidate reserve prediction model;

[0009] outputting a prediction result of recoverable reserves based on the candidate reserves prediction model, analyzing the accuracy of the candidate reserves prediction model based on the deviation between the prediction result of recoverable reserves and sample data, selecting the candidate reserves prediction model with the highest accuracy as the preferred recoverable reserves prediction model, and selecting the input variable combination inputted into the candidate reserves prediction model with the highest accuracy as the preferred input variable combination;

[0010] The recoverable reserves of the target well are predicted using the preferred recoverable reserves prediction model and the preferred combination of input variables.

[0011] An unconventional oil and gas single well recoverable reserves prediction system, comprising:

[0012] A data server is used to obtain multiple basic characteristics and recoverable reserves sample data required for recoverable reserves prediction of sample wells;

[0013] a computing server configured to determine the importance of each of the basic features relative to the recoverable reserves; screen the basic features multiple times based on the importance, and use the basic features screened out each time as an input variable combination, thereby obtaining multiple input variable combinations after multiple screenings; construct a reserve prediction model based on a machine learning algorithm, use each input variable combination as the input of the reserve prediction model, use recoverable reserves as the output of the reserve prediction model, and train the reserve prediction model using corresponding sample data to obtain a candidate reserve prediction model; output a prediction result of recoverable reserves based on the candidate reserve prediction model, analyze the accuracy of the candidate reserve prediction model based on the deviation between the prediction result of recoverable reserves and the sample data, select the candidate reserve prediction model with the highest accuracy as the preferred recoverable reserves prediction model, and select the input variable combination input into the candidate reserve prediction model with the highest accuracy as the preferred input variable combination; and predict the recoverable reserves of the target well using the preferred recoverable reserves prediction model and the preferred input variable combination;

[0014] The storage server is used to store the preferred recoverable reserves prediction model, the preferred input variable combination, and the predicted recoverable reserves of the target well.

[0015] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0016] Obtain multiple basic characteristics and recoverable reserves sample data required for sample well recoverable reserves prediction;

[0017] Determining the significance of each of the basic characteristics relative to the recoverable reserves;

[0018] The basic features are screened multiple times according to the importance, and the basic features screened out each time are used as an input variable combination, so as to obtain multiple input variable combinations after multiple screenings;

[0019] Constructing a reserve prediction model based on a machine learning algorithm, using each combination of the input variables as an input of the reserve prediction model, using recoverable reserves as an output of the reserve prediction model, and training the reserve prediction model using corresponding sample data to obtain a candidate reserve prediction model;

[0020] outputting a prediction result of recoverable reserves based on the candidate reserves prediction model, analyzing the accuracy of the candidate reserves prediction model based on the deviation between the prediction result of recoverable reserves and sample data, selecting the candidate reserves prediction model with the highest accuracy as the preferred recoverable reserves prediction model, and selecting the input variable combination inputted into the candidate reserves prediction model with the highest accuracy as the preferred input variable combination;

[0021] The recoverable reserves of the target well are predicted using the preferred recoverable reserves prediction model and the preferred combination of input variables.

[0022] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0023] Obtain multiple basic characteristics and recoverable reserves sample data required for sample well recoverable reserves prediction;

[0024] Determining the significance of each of the basic characteristics relative to the recoverable reserves;

[0025] The basic features are screened multiple times according to the importance, and the basic features screened out each time are used as an input variable combination, so as to obtain multiple input variable combinations after multiple screenings;

[0026] Constructing a reserve prediction model based on a machine learning algorithm, using each combination of the input variables as an input of the reserve prediction model, using recoverable reserves as an output of the reserve prediction model, and training the reserve prediction model using corresponding sample data to obtain a candidate reserve prediction model;

[0027] outputting a prediction result of recoverable reserves based on the candidate reserves prediction model, analyzing the accuracy of the candidate reserves prediction model based on the deviation between the prediction result of recoverable reserves and sample data, selecting the candidate reserves prediction model with the highest accuracy as the preferred recoverable reserves prediction model, and selecting the input variable combination inputted into the candidate reserves prediction model with the highest accuracy as the preferred input variable combination;

[0028] The recoverable reserves of the target well are predicted using the preferred recoverable reserves prediction model and the preferred combination of input variables.

[0029] The above-mentioned unconventional oil and gas single-well recoverable reserves prediction method, system, computer equipment and storage medium, through the importance evaluation and screening of basic characteristics, select the best combination from the limited available input basic characteristics, that is, it can optimize the input variable combination with the greatest influence, achieve the optimal combination of input variables of the reserve prediction model, and improve the accuracy of unconventional oil and gas recoverable reserves prediction; at the same time, the prediction method does not require fixed input variables for the data, can adapt to the recoverable reserves prediction needs of different unconventional oil and gas exploration and development stages, break through the limitations of input parameters in different exploration and development stages, and can be applied to all exploration and development stages. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A schematic flow chart of a method for predicting recoverable reserves of unconventional oil and gas single wells in one embodiment;

[0031] Figure 2 A schematic diagram of ranking single well input variables by importance in one embodiment;

[0032] Figure 3 is a graph showing the number of input variables and relative model error in the first stage in one embodiment;

[0033] Figure 4 A comparison chart of the predicted recoverable reserves and the actual values ​​in the first stage in one embodiment;

[0034] Figure 5 is a graph showing the number of input variables and relative model error in the second stage in one embodiment;

[0035] Figure 6 A comparison chart of the predicted and actual recoverable reserves in the second phase of an embodiment;

[0036] Figure 7 is a graph of the number of input variables and the relative error of the model in the third stage in one embodiment;

[0037] Figure 8 A comparison chart of the predicted and actual recoverable reserves in the third stage in one embodiment;

[0038] Figure 9 is a graph of the number of input variables versus model relative error in another embodiment;

[0039] Figure 10 A comparison chart of the predicted recoverable reserves and the actual values ​​in another embodiment;

[0040] Figure 11 is a graph of the number of input variables and the relative error of the model in yet another embodiment;

[0041] Figure 12 A comparison chart of the predicted recoverable reserves and the actual recoverable reserves in another embodiment;

[0042] Figure 13 The implementation process of the method for predicting recoverable reserves of unconventional oil and gas single wells in one embodiment is as follows: Figure 1 ;

[0043] Figure 14 The implementation process of the method for predicting recoverable reserves of unconventional oil and gas single wells in one embodiment is as follows: Figure 2 ;

[0044] Figure 15 A schematic flow chart of a method for predicting recoverable reserves of unconventional oil and gas single wells in another embodiment;

[0045] Figure 16 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0047] Example 1

[0048] In this embodiment, Figure 1 As shown, a method for predicting recoverable reserves of unconventional oil and gas single wells is provided, which includes:

[0049] Step 110: Acquire sample data of multiple basic characteristics and recoverable reserves required for recoverable reserves prediction of sample wells.

[0050] In this embodiment, the basic characteristics required for predicting recoverable reserves of sample wells may include geological parameters, well logging or logging interpretation parameters, extracted seismic attributes, or seismic inversion data parameters, etc., and this embodiment does not impose any limitation thereto. The sample data for recoverable reserves includes production-verified unconventional oil and gas recoverable reserves or calculated recoverable reserves information.

[0051] Step 120: Determine the importance of each of the basic characteristics relative to the recoverable reserves.

[0052] In this embodiment, a main controlling factor analysis of recoverable reserves is performed on each of the basic characteristics, and the importance of each basic characteristic relative to recoverable reserves is calculated.

[0053] Step 130 , screening the basic features multiple times according to the importance, and using the basic features screened out each time as an input variable combination, to obtain multiple input variable combinations after multiple screenings.

[0054] In this embodiment, the importance of each basic feature relative to recoverable reserves is calculated, and the basic features are screened multiple times according to the importance; wherein, the basic features are screened multiple times according to the importance, and the basic features can be sorted from high to low according to the importance, and the last ones in the sorting results are eliminated to obtain multiple input variable combinations; or a preset number of basic features with relatively high importance are preferentially selected according to the importance, and the remaining basic features are randomly combined to obtain multiple quantitative variable combinations, etc., which is not limited in this article.

[0055] Step 140: construct a reserve prediction model based on a machine learning algorithm, use each of the input variable combinations as the input of the reserve prediction model, use the recoverable reserves as the output of the reserve prediction model, and use the corresponding sample data to train the reserve prediction model to obtain a candidate reserve prediction model.

[0056] In one embodiment, a reserve prediction model is constructed based on the same type of machine learning algorithm. Different reserve prediction models can be constructed based on the same type of machine learning algorithm according to different combinations of input variables.

[0057] In this embodiment, a reserve prediction model is constructed based on a machine learning algorithm, each input variable combination is used as the input of the reserve prediction model, the recoverable reserves are used as the output of the reserve prediction model, and the reserve prediction model is trained using corresponding sample data to obtain a candidate reserve prediction model.

[0058] Step 150: Based on the candidate reserve prediction model, output the prediction result of recoverable reserves; analyze the accuracy of the candidate reserve prediction model according to the deviation between the prediction result of recoverable reserves and the sample data; select the candidate reserve prediction model with the highest accuracy as the preferred recoverable reserves prediction model; and select the input variable combination input into the candidate reserve prediction model with the highest accuracy as the preferred input variable combination.

[0059] In this embodiment, the candidate reserve prediction model with the highest accuracy is determined as the preferred recoverable reserves prediction model based on its accuracy among multiple candidate reserve prediction models. Furthermore, the input variable combination input into the candidate reserve prediction model with the highest accuracy is determined as the preferred input variable combination. This allows the optimal combination to be selected from a limited number of available input basic features, specifically the input variable combination with the greatest impact to be selected, thereby optimizing the preferred recoverable reserves prediction model input variable combination and ensuring the accuracy of subsequent recoverable reserves prediction for the target well.

[0060] Step 160: Predict the recoverable reserves of the target well using the preferred recoverable reserves prediction model and the preferred combination of input variables.

[0061] In this embodiment, the basic characteristics of the target well are screened and combined according to the preferred input variable combination determined above to obtain a preferred input variable combination, and the preferred input variable combination is input into the preferred recoverable reserves prediction model established above to predict recoverable reserves.

[0062] In the above embodiment, by evaluating and screening the importance of basic features, the best combination is selected from the limited available input basic features, that is, the input variable combination with the greatest influence can be selected, thereby achieving the optimal input variable combination of the recoverable reserves prediction model and improving the accuracy of unconventional oil and gas recoverable reserves prediction; at the same time, the prediction method does not require fixed input variables for the data, and can adapt to the recoverable reserves prediction needs of different unconventional oil and gas exploration and development stages, breaking through the limitations of input parameters in different exploration and development stages, and can be applied to all exploration and development stages.

[0063] In one embodiment, the basic features are screened multiple times according to the importance, and the basic features screened out each time are used as an input variable combination. The step of obtaining multiple input variable combinations after the multiple screenings includes: sorting the basic features from high to low according to the importance, obtaining the sorted results of the basic features, and using them as the first input variable combination; eliminating the last-placed input variable combinations of the previous input variable combinations, performing multiple eliminations, and obtaining multiple input variable combinations based on the basic features retained each time.

[0064] In this embodiment, the basic features are sorted from high to low according to their importance to obtain a sorted result of the basic features, which is used as the first input variable combination; the last-place elimination is performed on the previous input variable combination, and elimination is performed multiple times. Multiple input variable combinations are obtained based on the basic features retained in each round, that is, the basic features are sorted from high to low according to their importance to obtain a sorted result of the basic features, which is used as the first input variable combination, and a basic feature with the lowest importance is deleted from the end of the first input variable combination to obtain a second input variable combination; a basic feature with the lowest importance is deleted from the end of the second input variable combination to obtain a third input variable combination, and so on, the last-place elimination is performed on the previous input variable combination, and elimination is performed multiple times to obtain multiple input variable combinations, wherein the number of basic features in the multiple input variable combinations decreases successively until there is only one basic feature in the input variable combination, and this basic feature is the most important basic feature.

[0065] In one embodiment, the step of determining the importance of each of the basic characteristics relative to the recoverable reserves includes: analyzing the importance of each of the basic characteristics relative to the recoverable reserves using a distance correlation coefficient method.

[0066] In this example, the distance correlation coefficient method is used to measure the degree of correlation between each basic feature and recoverable reserves, thereby determining the importance of each basic feature relative to recoverable reserves. Based on the importance of each basic feature relative to recoverable reserves, basic features are screened and combined to determine the optimal combination of input variables for the reserves prediction model, thereby improving the accuracy of the reserves prediction model in predicting recoverable unconventional oil and gas reserves.

[0067] In one embodiment, the step of analyzing the importance of each basic feature relative to the recoverable reserves using the distance correlation coefficient method includes: calculating the distance between each basic feature and the recoverable reserves and the distance between every two basic features to determine the distance correlation coefficient of each basic feature; determining the importance of each basic feature relative to the recoverable reserves based on the distance correlation coefficient of each basic feature;

[0068] Wherein, the distance correlation coefficient is expressed as:

[0069]

[0070] Where x is the basic feature; y is the recoverable reserves of the sample; dcor is the distance correlation coefficient; dcov(x,y) is the distance covariance between variables x and y, dvar(x,x) is the distance variance of variable x, and dvar(y,y) is the distance variance of variable y.

[0071] In one embodiment, before determining the importance of each basic feature relative to the recoverable reserves, the method further includes: normalizing each basic feature.

[0072] In one embodiment, the acquired basic characteristics and recoverable reserves are collated. In this embodiment, the acquired data is collated in a one-record format per well, where each record includes an input variable X and a corresponding target variable (or output variable) Y. The basic characteristics are defined as the input variable X, where X includes x1, x2, x3, ..., determined based on the actual data collected. Recoverable reserves information is defined as the target variable or output variable Y.

[0073] In one embodiment, after data sorting, X is normalized, and the parameter settings for normalization of each variable are recorded.

[0074] In this embodiment, the main controlling factors of recoverable reserves are analyzed using the distance correlation coefficient method on the sorted and normalized data.

[0075] Specifically, the distance between each basic feature and recoverable reserves and the distance between every two basic features are calculated, and the distance correlation coefficient of each basic feature is determined;

[0076] The importance of each basic feature relative to recoverable reserves is determined based on the distance correlation coefficient of each basic feature.

[0077] The distance correlation coefficient of each basic feature is the importance of each basic feature relative to recoverable reserves; the larger the distance correlation coefficient of a basic feature is, the greater the importance of the basic feature relative to recoverable reserves.

[0078] In one embodiment, the distance correlation coefficient is expressed as:

[0079]

[0080] Where x includes x1, x2, x3, ..., x represents each basic characteristic; y represents recoverable reserves; dcor represents the distance correlation coefficient; dcov(x,y) represents the distance covariance between variables x and y, dvar(x,x) represents the distance variance of variable x, and dvar(y,y) represents the distance variance of variable y.

[0081] The formula for calculating the variable distance is as follows (taking variable x as an example):

[0082] dx=||x i -x j ||,i,j=1,2,3,...,n

[0083] dy=||y i -yj ||,i,j=1,2,3,...,n

[0084]

[0085]

[0086]

[0087] dvar(x,x)=dcov(x,x)

[0088] dvar(y,y)=dcov(y,y)

[0089] In this embodiment, the distance correlation coefficient method is used to analyze the main controlling factors of recoverable reserves for the basic characteristics to obtain the importance of each basic characteristic.

[0090] In one embodiment, the step of constructing a reserve prediction model based on a machine learning algorithm, taking each input variable combination as the input of the reserve prediction model, taking recoverable reserves as the output of the reserve prediction model, and training the reserve prediction model using corresponding sample data to obtain a candidate reserve prediction model includes: constructing an objective function of the reserve prediction model based on a relative error between a prediction result of recoverable reserves output by the reserve prediction model and sample data of recoverable reserves; iterating the reserve prediction model based on the objective function to adjust parameters of the reserve prediction model; when the objective function reaches a minimum value, determining the corresponding parameter as the target parameter of the reserve prediction model; and determining the candidate reserve prediction model based on the target parameter.

[0091] In one embodiment, based on expert experience and different geological characteristics, a type of machine learning algorithm model suitable for a local single well can be selected to construct a reserve prediction model.

[0092] The objective function of the reserve prediction model plays a crucial role in the training process. This function defines the goal that the model needs to optimize. By minimizing or maximizing this objective function, the model parameters are adjusted to better fit the data and achieve the desired results.

[0093] In this embodiment, an objective function of the reserves prediction model is constructed based on the relative error between the prediction result of the recoverable reserves output by the reserves prediction model and the sample data of the recoverable reserves; the parameters of the reserves prediction model are optimized through the objective function to ensure that the final candidate reserves prediction model can better predict the recoverable reserves of the target well.

[0094] In one embodiment, in the order of the number of basic features in each input variable combination from large to small, each input variable combination is used as the input of the reserve prediction model, and the recoverable reserves are used as the output of the reserve prediction model, which are successively input into the reserve prediction model for training to obtain multiple candidate reserve prediction models.

[0095] In this embodiment, a reserve prediction model is constructed based on a machine learning algorithm, and the parameters in the reserve prediction model are set using the algorithm's default parameters. First, the input variable combination with the largest number of basic features among the input variable combinations is used as the input of the reserve prediction model, and the recoverable reserves are used as the output of the reserve prediction model. The reserve prediction model is then trained using the corresponding sample data to obtain the first candidate reserve prediction model. Then, in descending order of the number of basic features in the input variable combination, each input variable combination is used as the input of the reserve prediction model, and the recoverable reserves are used as the output of the reserve prediction model. The reserve prediction model is then trained using the corresponding sample data to obtain at least one candidate reserve prediction model. This continues until the input variable combination with the smallest number of basic features, i.e., the input variable combination with only one basic feature, is input into the reserve prediction model for training, obtaining the final candidate reserve prediction model.

[0096] In this embodiment, based on each candidate reserve prediction model, the prediction result of recoverable reserves is output, and according to the deviation between the prediction result of recoverable reserves and the sample data, the accuracy of each candidate reserve prediction model is analyzed, and the candidate reserve prediction model with the highest accuracy is selected as the preferred recoverable reserves prediction model, and the input variable combination inputted into the candidate reserve prediction model with the highest accuracy is selected as the preferred input variable combination.

[0097] In machine learning, accuracy is one of the metrics used to evaluate model performance. It represents the proportion of samples correctly predicted by the model to the total number of samples. In this embodiment, the accuracy of the reserve prediction model and the input variable combinations were screened using accuracy to effectively improve the accuracy of single-well recoverable reserves prediction using machine learning algorithms.

[0098] In one embodiment, the step of iterating the reserve prediction model based on the objective function and adjusting the parameters of the reserve prediction model includes: adjusting the parameters of the reserve prediction model by using a particle swarm algorithm or a grid search algorithm.

[0099] In this embodiment, the parameters of the reserve prediction model are adjusted by using a particle swarm algorithm to obtain target parameters; and a candidate reserve prediction model is determined based on the target parameters.

[0100] Specifically, the calculation formula for model parameter tuning by the particle swarm algorithm is:

[0101]

[0102]

[0103] Where: n is the number of particles, c1 is the individual acceleration factor of the particle, c2 is the social acceleration factor of the particle, w is the inertia weight, vi is the velocity of the particle, xi is the position of the particle, pbest is the best position passed by the i-th particle, and gbest is the best position passed by all particles.

[0104] Alternatively, the parameters of the reserve prediction model are adjusted through a grid search algorithm to obtain target parameters; and the candidate reserve prediction model is determined based on the target parameters.

[0105] In this embodiment, a particle swarm algorithm or a grid search algorithm is introduced to perform parameter optimization training on the reserve prediction model to ensure that the optimal solution is found within a limited number of iterations, so as to more efficiently obtain the candidate reserve prediction model.

[0106] In one embodiment, the step of constructing a reserve prediction model based on a machine learning algorithm, taking each input variable combination as the input of the reserve prediction model, taking recoverable reserves as the output of the reserve prediction model, and training the reserve prediction model using corresponding sample data to obtain a candidate reserve prediction model also includes: performing model evaluation on the reserve prediction model through a cross-validation method.

[0107] In this embodiment, K-fold cross validation is used to evaluate the reserve prediction model. Specifically, the dataset is first divided into K subsets, which are usually subsets of approximately equal size.

[0108] Among them, the calculation formula of K-fold cross validation is:

[0109]

[0110]

[0111] Among them, Err k is the number of classification errors on the k-th test set.

[0112] Example 2

[0113] In this embodiment, a method for predicting recoverable reserves of unconventional oil and gas single wells is provided, which includes:

[0114] Step 11, obtaining multiple basic characteristics and recoverable reserves sample data required for recoverable reserves prediction of sample wells;

[0115] Step 12, determining the importance of each of the basic characteristics relative to the recoverable reserves;

[0116] The basic features are screened multiple times according to the importance, and the basic features screened out each time are used as an input variable combination, and multiple input variable combinations are obtained after multiple screenings;

[0117] Step 13: construct different types of reserve prediction models based on different types of machine learning algorithms. For any type of reserve prediction model, use each input variable combination as the input of the reserve prediction model, use recoverable reserves as the output of the reserve prediction model, and train the reserve prediction model using corresponding sample data to obtain a candidate reserve prediction model.

[0118] Step 14: outputting a prediction result of recoverable reserves based on the candidate reserve prediction model; analyzing the accuracy of the reserve prediction model based on the deviation between the prediction result of recoverable reserves and the sample data; and selecting the reserve prediction model with the highest accuracy as the preferred reserve prediction model of this type;

[0119] Step 15: comparing different types of preferred reserve prediction models, selecting the preferred reserve prediction model that meets preset requirements as the final recoverable reserve prediction model, and selecting the input variable combination input into the final recoverable reserve prediction model with the highest accuracy as the preferred input variable combination;

[0120] Step 16: Use the final recoverable reserves prediction model and the preferred input variable combination to predict the recoverable reserves of the target well.

[0121] In an embodiment of the present invention, different types of reserve prediction models can be constructed based on different types of machine learning algorithms.

[0122] Specifically, different types of machine learning algorithms may be, for example, random forests, gradient boosting trees, feedforward neural networks, or convolutional neural networks, etc. The present invention is not limited thereto.

[0123] Random Forest is a machine learning algorithm that constructs multiple decision trees to improve prediction accuracy and generalization ability.

[0124] The expression of random forest is:

[0125]

[0126] Gradient Boosting Tree is a machine learning algorithm that iteratively trains decision trees to gradually improve the accuracy of the model.

[0127] The expression of gradient boosting tree is:

[0128]

[0129]

[0130] Where M is the number of trees, T is the decision tree, and Θ is the decision tree parameter.

[0131] Feedforward Neural Network (FNN) is a type of artificial neural network that adopts a single-layer and multi-layer structure.

[0132] The expression of the feedforward neural network is:

[0133] z (l) =W (l) f l-1 (z (l-1) )+b (l)

[0134] a (l) =f l (W (l) a (l-1) +b (l) )

[0135]

[0136] Where z is the input of the neuron, a is the output of the neuron, b is the bias, and f is the activation function.

[0137] Convolutional Neural Networks (CNN) is a type of feedforward neural network algorithm that includes convolution calculations and has a deep structure.

[0138] The expression of convolutional neural network is:

[0139]

[0140]

[0141] Where x is the network input, k is the convolution kernel, and b is the bias term.

[0142] Output variable importance, using v to represent variable importance:

[0143]

[0144] In addition, in an embodiment of the present invention, different types of reserve prediction models can be constructed based on different types of machine learning algorithms. The above model optimization process is then demonstrated for each type of reserve prediction model. That is, for each type of reserve prediction model, the accuracy of the model is analyzed under different input variable combinations to determine the optimal reserve prediction model for each type of machine learning algorithm. These different types of optimal reserve prediction models are then compared, and the optimal reserve prediction model that meets preset requirements (e.g., the one with the shortest computational time) is selected as the final recoverable reserve prediction model to predict the recoverable reserves of the target well.

[0145] The specific definition of the method for predicting recoverable reserves of unconventional oil and gas single wells in this embodiment can be found in the definition of the method for predicting recoverable reserves of unconventional oil and gas single wells in the above embodiment 1, and will not be repeated here.

[0146] Example 3

[0147] In this embodiment, Figures 2 to 8 As shown in the figure, a method for predicting recoverable reserves of unconventional oil and gas single wells is provided for the actual reservoir development layer of a shale gas field.

[0148] The method includes:

[0149] Step 161: Acquire sample data of multiple basic characteristics and recoverable reserves required for recoverable reserves prediction of sample wells.

[0150] In this example, parameter data such as well number, burial depth, pressure coefficient, brittleness index, brittle mineral content, porosity, clay pores, organic pores, organic matter content, gas saturation, horizontal well section length, number of fracturing stages, main layer penetration length, sand addition amount, fracturing fluid amount, and single-well technically recoverable reserves were first collected. The data were organized as a record for each well, as shown in Table 1.

[0151] Table 1 Basic characteristics and recoverable reserves of sample wells

[0152]

[0153] Step 162: Determine the importance of each of the basic characteristics relative to the recoverable reserves.

[0154] In this embodiment, the well number in Table 1 is the well name, the technically recoverable reserves of a single well is the target variable (output variable) Y, and the remaining variables are the input variables to be input X. After normalization of each input variable, the importance of each input variable is calculated and ranked using the distance coefficient. The results are shown in a bar chart as follows: Figure 2As shown in the figure, the order of importance of input variables to output variables is pressure coefficient, horizontal well section length, sand addition amount, organic pores, number of fracturing stages, fracturing fluid volume, brittleness index, burial depth, organic matter content, porosity, length of main layer penetrated, brittle mineral content, gas saturation, and clay pores.

[0155] Step 163: Screen the basic features multiple times according to the importance, and use the basic features screened out each time as an input variable combination, to obtain multiple input variable combinations after multiple screenings; construct a reserve prediction model based on a machine learning algorithm, use each input variable combination as the input of the reserve prediction model, use recoverable reserves as the output of the reserve prediction model, and train the reserve prediction model using corresponding sample data to obtain a candidate reserve prediction model; output a prediction result of recoverable reserves based on the candidate reserve prediction model, analyze the accuracy of the candidate reserve prediction model based on the deviation between the prediction result of recoverable reserves and the sample data, select the candidate reserve prediction model with the highest accuracy as the preferred recoverable reserves prediction model, and select the input variable combination input into the candidate reserve prediction model with the highest accuracy as the preferred input variable combination.

[0156] In this embodiment, an input variable combination is obtained based on each of the basic features. The basic features with the lowest importance are eliminated from the input variable combination one by one. After multiple eliminations, multiple input variable combinations are obtained based on the input variable combinations retained each time. The multiple input variable combinations are successively input into the reserves prediction model for training to obtain a candidate reserves prediction model, and the model accuracy of the candidate reserves prediction model is calculated.

[0157] In this embodiment, the shale gas development stage is divided into three stages according to the actual shale gas production process:

[0158] (1) The first stage, geological exploration, uses only geological features;

[0159] (2) The second stage, the fracturing production stage, uses geological characteristics and fracturing engineering characteristics;

[0160] (3) The third stage, the gas testing stage, uses geological characteristics, fracturing engineering characteristics and the open flow rate of the gas test.

[0161] The input variables corresponding to geological characteristics include pressure coefficient, organic pores, brittleness index, burial depth, organic matter content, porosity, brittle mineral content, gas saturation, and clay pores;

[0162] The input variables corresponding to the fracturing engineering characteristics include the length of the horizontal well section, the amount of sand added, the number of fracturing stages, the amount of fracturing fluid, and the length of the main layer penetrated.

[0163] In the first stage, only geological features are used to train the model. A machine learning algorithm model is selected. This time, the XGBoost algorithm model is used. According to the importance of the input variables, the input variables are input into the XGBoost model one by one from the most to the least. The relative error index is used to evaluate the impact of the number of input variables on the model algorithm results. The results of this process are plotted as a line graph (such as Figure 3 As shown in the figure, find the point with the smallest relative error from the graph. The number of variables corresponding to this point is 5, which is the optimal number of input parameters for the XGBoost algorithm model. Combined with the results of the importance ranking of the input variables, the optimal input parameter combination can be determined as pressure coefficient, organic pores, brittleness index, burial depth, and organic matter content.

[0164] In this process, the grid search algorithm was used to optimize the model parameters and finally the XGBoost algorithm model was obtained to predict the recoverable reserves of shale gas single well technology. The predicted results were compared with the actual values ​​(such as Figure 4 The overall prediction results are good, with a relative error of 15.7%.

[0165] In the second stage, the model is trained using geological and engineering features. The same variable selection and model training methods are adopted, and the relative error index is used to evaluate the impact of the number of input variables on the model algorithm results (such as Figure 5 As shown in the figure, the number of variables corresponding to the point with the smallest relative error and no overfitting is 8. Combined with the results of the importance ranking of the input variables, the optimal input parameter combination can be determined as pressure coefficient, organic pores, brittleness index, burial depth, sand addition amount, organic matter content, number of fracturing stages, and porosity.

[0166] The obtained XGBoost algorithm model is used to predict the recoverable reserves of shale gas single well technology. The predicted results are compared with the actual values ​​(such as Figure 6 The overall prediction results are good, with a relative error of 14.5%.

[0167] In the third stage, the model is trained using geological characteristics, engineering characteristics and production characteristics. The same variable selection and model training methods are adopted. The impact of the number of input variables on the model algorithm results (such as Figure 7 As shown in the figure, the number of variables corresponding to the point with the smallest relative error and no overfitting is 9. Combined with the results of the input variable importance ranking, the optimal input parameter combination can be determined as the open-flow rate of test gas, pressure coefficient, organic pores, brittleness index, burial depth, sand addition amount, organic matter content, number of fracturing stages, and porosity. The obtained XGBoost algorithm model is used to predict the technical recoverable reserves of shale gas single wells. The predicted results are compared with the actual values ​​(as shown in the figure). Figure 8 The overall prediction results are good, with a relative error of 12.1%.

[0168] The methods for selecting input variables and model hyperparameters in the three stages are the same, which is the degree of importance of the input variables. The input variables are input into the XGBoost model one by one from most to least for training. The data used in the three stages are gradually improved, and the relative error of the model is reduced from 15.7% to 12.1%.

[0169] Example 4

[0170] In this embodiment, Figures 9 and 10 As shown in the figure, a method for predicting recoverable reserves of unconventional oil and gas single wells is provided for sandstone tight gas formations.

[0171] The method includes:

[0172] Step 261: Acquire sample data of multiple basic characteristics and recoverable reserves required for recoverable reserves prediction of sample wells.

[0173] In this embodiment, data is first collected. In this example, parameter data such as well number, buried depth of the gas reservoir middle layer, pressure coefficient, brittleness index, clay mineral content, porosity, permeability, gas saturation, thickness of Class I reservoir, length of the stimulation section, sand addition intensity, number of perforation clusters, and technically recoverable reserves of a single well are collected. Each well is organized as a record, with the well number as the index, the technically recoverable reserves of a single well as the target variable (output variable) Y, and the remaining variables as the input variables X, to form a data set.

[0174] Step 262: Determine the importance of each of the basic characteristics relative to the recoverable reserves.

[0175] In this embodiment, after normalizing each input variable, the importance of each input variable is calculated and ranked using the distance correlation coefficient method. The order is pressure coefficient, sand injection intensity, porosity, stimulation section length, brittleness index, gas saturation, clay mineral content, thickness of Class I reservoir, permeability, burial depth of the middle layer of the gas reservoir, and number of perforation clusters.

[0176] Step 263: Screen the basic features multiple times according to the importance, and use the basic features screened out each time as an input variable combination, to obtain multiple input variable combinations after multiple screenings; construct multiple reserve prediction models based on different machine learning algorithms, use each input variable combination as the input of the reserve prediction model, use recoverable reserves as the output of the reserve prediction model, and train the reserve prediction model using corresponding sample data to obtain a candidate reserve prediction model; output a prediction result of recoverable reserves based on the candidate reserve prediction model, analyze the accuracy of the candidate reserve prediction model based on the deviation between the prediction result of recoverable reserves and the sample data, select the candidate reserve prediction model with the highest accuracy as the preferred recoverable reserves prediction model, and select the input variable combination input into the candidate reserve prediction model with the highest accuracy as the preferred input variable combination.

[0177] In this embodiment, a machine algorithm model is selected (you can also perform preliminary predictions on multiple machine learning algorithms based on experience and select the one with better results). This time, the Adaboost algorithm model is used. According to the importance of the input variables, the input variables are input into the AdaBoost model from most to least to complete the training and record the relative error of the model under different variables (such as Figure 9 As shown in the figure, find the point with the smallest relative error. The number of variables corresponding to this point is the optimal number of input parameters for the most suitable algorithm model in this area. In this example, it is 7, that is, the optimal input parameter combination is pressure coefficient, sand addition intensity, porosity, reforming section length, brittleness index, gas saturation, and clay mineral content.

[0178] The AdaBoost algorithm model obtained is used to predict the single-well technical recoverable reserves of tight gas. The predicted results are compared with the actual values ​​(such as Figure 10 The overall prediction results are good, with a relative error of 15.5%.

[0179] Example 5

[0180] In this embodiment, Figures 11 to 12 As shown in FIG, a method for predicting recoverable reserves of unconventional oil and gas single wells is provided for single shale oil wells.

[0181] The method includes:

[0182] Step 361: Acquire sample data of multiple basic characteristics and recoverable reserves required for recoverable reserves prediction of sample wells.

[0183] In this embodiment, data is first collected. In this example, parameter data such as well number, burial depth, pressure coefficient, total organic matter content, organic matter maturity, oil content, brittle mineral content, total porosity, organic matter porosity, horizontal well section length, number of fracturing stages, sand addition amount, fracturing fluid amount, and single-well technically recoverable reserves are collected. Each well is organized as a record, with the well number as the index, single-well technically recoverable reserves as the target variable (output variable) Y, and the remaining variables as the input variables X, to form a data set.

[0184] Step 362: Determine the importance of each of the basic characteristics relative to the recoverable reserves.

[0185] In this embodiment, after normalizing each input variable, the distance correlation coefficient method is used to calculate the importance of each input variable and rank them in the following order: organic matter maturity, pressure coefficient, total organic matter content, sand addition amount, brittle mineral content, total porosity, fracturing fluid volume, oil content, horizontal well section length, organic porosity, burial depth, and number of fracturing stages.

[0186] Step 363: Screen the basic features multiple times according to the importance, and use the basic features screened out each time as an input variable combination, to obtain multiple input variable combinations after multiple screenings; construct multiple reserve prediction models based on different machine learning algorithms, use each input variable combination as the input of the reserve prediction model, use recoverable reserves as the output of the reserve prediction model, and train the reserve prediction model using corresponding sample data to obtain a candidate reserve prediction model; output a prediction result of recoverable reserves based on the candidate reserve prediction model, analyze the accuracy of the candidate reserve prediction model based on the deviation between the prediction result of recoverable reserves and the sample data, select the candidate reserve prediction model with the highest accuracy as the preferred recoverable reserves prediction model, and select the input variable combination input into the candidate reserve prediction model with the highest accuracy as the preferred input variable combination.

[0187] In this embodiment, a machine algorithm model is selected (you can also make preliminary predictions based on experience using multiple machine learning algorithms and select the one with better results). This time, the XGBoost algorithm model is used. According to the importance of the input variables, the input variables are input into the XGBoost model from most to least to complete the training and record the relative error of the model under different variables (such as Figure 11 As shown in the figure, find the point with the smallest relative error from the graph. The number of variables corresponding to this point is the optimal number of input parameters for the most suitable algorithm model in this area. In this example, it is 6, that is, the optimal input parameter combination is organic matter maturity, pressure coefficient, total organic matter content, sand addition amount, brittle mineral content, and total porosity.

[0188] The obtained XGBoost algorithm model is used to predict the single well technical recoverable reserves of tight gas. The predicted results are compared with the actual values ​​(such as Figure 12 The overall prediction result is good, with a relative error of 11.96%.

[0189] Example 6

[0190] In this embodiment, Figures 13 to 15 As shown in the figure, a system for predicting recoverable reserves of unconventional oil and gas single well is provided. Figures 13 to 14 As shown, it includes:

[0191] A data server is used to obtain multiple basic characteristics and recoverable reserves sample data required for recoverable reserves prediction of sample wells;

[0192] a computing server configured to determine the importance of each of the basic features relative to the recoverable reserves; screen the basic features multiple times based on the importance, and use the basic features screened out each time as an input variable combination, thereby obtaining multiple input variable combinations after multiple screenings; construct a reserve prediction model based on a machine learning algorithm, use each input variable combination as the input of the reserve prediction model, use recoverable reserves as the output of the reserve prediction model, and train the reserve prediction model using corresponding sample data to obtain a candidate reserve prediction model; output a prediction result of recoverable reserves based on the candidate reserve prediction model, analyze the accuracy of the candidate reserve prediction model based on the deviation between the prediction result of recoverable reserves and the sample data, select the candidate reserve prediction model with the highest accuracy as the preferred recoverable reserves prediction model, and select the input variable combination input into the candidate reserve prediction model with the highest accuracy as the preferred input variable combination; and predict the recoverable reserves of the target well using the preferred recoverable reserves prediction model and the preferred input variable combination;

[0193] The storage server is used to store the preferred recoverable reserves prediction model, the preferred input variable combination, and the predicted recoverable reserves of the target well.

[0194] In this embodiment, basic characteristics for unconventional oil and gas recoverable reserves prediction can be collected through measuring instruments and sent to a data server.

[0195] In this embodiment, the basic features for predicting unconventional oil and gas recoverable reserves can be geological data, well logging and well logging interpretation results data, extracted seismic attributes or seismic inversion data, etc., which need to include unconventional oil and gas recoverable reserves or calculated recoverable reserves data that have been verified by production.

[0196] Correspondingly, see Figure 15As shown, the data server is used to execute step 1: obtaining multiple basic information and sample data of recoverable reserves required for single well recoverable reserves prediction.

[0197] The data server is also used to perform step 2: receive data, organize and preprocess the data, and send the processed data to the computing server: organize the data into a format corresponding to the input multidimensional vector x and the output y, and normalize the input data.

[0198] In this embodiment, the data collected from the data server is organized into a single record format, one per well. Each record includes the input variable X and the corresponding target variable (output variable) Y. X includes x1, x2, x3, ..., determined based on the actual data collected. Y consists of a single vector, y1, representing the recoverable or calculated recoverable reserves of unconventional oil and gas in a single well. After data organization, X is normalized, and the normalization parameter settings for each variable are recorded. After preprocessing, the data is packaged and sent to the computation server.

[0199] The computing server is used to execute step 3: receiving the data processed by the data server, performing the analysis of the main controlling factors of unconventional oil and gas recoverable reserves, and ranking the importance of the input vectors in terms of their impact on recoverable reserves.

[0200] In this embodiment, the computing server is used to analyze the main controlling factors of recoverable reserves using the distance correlation coefficient method on the data collated and processed by the data server, and the analysis results are automatically sorted by the input variables X from high to low according to the importance.

[0201] The computing server is also used to perform step 4: optimal input parameter combination selection, and perform last-place elimination machine learning algorithm training based on the importance of the input vector to determine the optimal input parameter combination.

[0202] The computing server selects a machine learning algorithm and uses the default algorithm parameters for initial parameter settings. First, the machine learning algorithm model is trained and its accuracy is evaluated using all input variables. The model iteratively minimizes the objective function (Obj) and the resulting accuracy is recorded.

[0203] Then, following the X importance ranking results from step 3, remove the least important input variable and retrain the machine learning algorithm model and evaluate its accuracy, recording the resulting accuracy. Repeat this process until only one input variable has completed the machine learning algorithm model training and accuracy evaluation, recording the resulting accuracy. The input variable with the highest accuracy is selected as the optimal input variable combination. If multiple machine learning methods are used to predict unconventional oil and gas recoverable reserves, different machine learning algorithm models can be used.

[0204] Repeat the above steps in step 4. Note that throughout the entire process, the model evaluation method uses a unified cross-validation (CV) method. Taking k-fold cross-validation as an example, the idea is to divide the dataset into k parts, alternately use k-1 parts for training and 1 part for validation, and use the average of the k results as an estimate of the algorithm accuracy.

[0205] The computing server is also used in step 4 to optimize the model using a particle swarm optimization algorithm or a grid search algorithm. This algorithm identifies the most accurate model parameter combination and training results from the model evaluation results, which serve as the final machine-learning recoverable reserves prediction model. If multiple machine learning algorithm models are used, the computing server will repeat the above steps, switching between different models, and finally comparing and selecting the optimal model. After training is complete, the computing server saves the optimal model to the storage server.

[0206] The computing server is also used to execute step 5: predicting unconventional oil and gas recoverable reserves: the computing server loads the optimal model determined in step 4 into the computing server, loads the data to be predicted into the data server, normalizes the data according to the normalization parameters recorded in step 2, and sends the data to the computing server after processing. The optimal input parameter combination determined in step 4 is selected as the input parameter, and is input into the optimal model for automatic prediction. The computing server saves the final unconventional oil and gas recoverable reserves prediction result to the storage server.

[0207] Compared to existing machine learning algorithms for predicting unconventional oil and gas recoverable reserves, this application uses the importance of input variables to predict recoverable reserves to select the optimal input variable combination, eliminate potential interference variables, and improve the accuracy of unconventional oil and gas recoverable reserves prediction of the machine learning algorithm model; this application uses cross-validation within the computing server for training and evaluation, and does not divide the sample data set into training sets and test sets, thereby reducing the random interference of sample sampling during model evaluation. In addition, this application does not require special restrictions on input variables and can be data from different exploration and development stages, thereby achieving efficient unconventional oil and gas recoverable reserves prediction using this application at different exploration and development stages.

[0208] For the specific limitations of the unconventional oil and gas single-well recoverable reserves prediction system, please refer to the limitations of the unconventional oil and gas single-well recoverable reserves prediction method above, which will not be repeated here.

[0209] Example 7

[0210] In this embodiment, a computer device is provided. Its internal structure diagram can be shown as follows: Figure 16As shown. The computer device includes a processor, memory, network interface, display screen, and input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program, and the non-volatile storage medium is deployed with a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with other computer devices deployed with application software. When the computer program is executed by the processor, a method for predicting the recoverable reserves of unconventional oil and gas single wells is implemented. The display screen of the computer device can be a liquid crystal display or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a key, trackball, or touchpad provided on the computer device housing, or an external keyboard, touchpad, or mouse.

[0211] Those skilled in the art will understand that Figure 16 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0212] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method for predicting recoverable reserves of unconventional oil and gas single wells described in any of the above embodiments are implemented.

[0213] Example 8

[0214] In this embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for predicting recoverable reserves of unconventional oil and gas single wells described in any of the above embodiments are implemented.

[0215] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0216] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0217] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for predicting recoverable reserves of unconventional oil and gas single wells, characterized by: include: Obtain multiple basic characteristics and recoverable reserves sample data required for sample well recoverable reserves prediction; Determining the significance of each of the basic characteristics relative to the recoverable reserves; The basic features are screened multiple times according to the importance, and the basic features screened out each time are used as an input variable combination, so as to obtain multiple input variable combinations after multiple screenings; Constructing a reserve prediction model based on a machine learning algorithm, using each combination of the input variables as an input of the reserve prediction model, using recoverable reserves as an output of the reserve prediction model, and training the reserve prediction model using corresponding sample data to obtain a candidate reserve prediction model; outputting a prediction result of recoverable reserves based on the candidate reserves prediction model, analyzing the accuracy of the candidate reserves prediction model based on the deviation between the prediction result of recoverable reserves and sample data, selecting the candidate reserves prediction model with the highest accuracy as the preferred recoverable reserves prediction model, and selecting the input variable combination inputted into the candidate reserves prediction model with the highest accuracy as the preferred input variable combination; The recoverable reserves of the target well are predicted using the preferred recoverable reserves prediction model and the preferred combination of input variables.

2. The method for predicting recoverable reserves of unconventional oil and gas single wells according to claim 1, characterized in that: The step of screening the basic features multiple times according to the importance and using the basic features screened out each time as an input variable combination to obtain multiple input variable combinations after multiple screenings includes: Sort the basic features from high to low according to the importance, obtain the sorting result of the basic features, and use it as the first input variable combination; The last digit of the input variable combination is eliminated multiple times, and multiple input variable combinations are obtained based on the basic features retained in each round.

3. The method for predicting recoverable reserves of unconventional oil and gas single wells according to claim 1, characterized in that: The step of determining the importance of each of the basic characteristics relative to the recoverable reserves comprises: The importance of each of the basic characteristics relative to the recoverable reserves is analyzed using the distance correlation coefficient method.

4. The method for predicting recoverable reserves of unconventional oil and gas in a single well according to claim 3, characterized in that: The step of analyzing the importance of each of the basic characteristics relative to the recoverable reserves using the distance correlation coefficient method includes: Calculating the distance between each basic feature and the recoverable reserves and the distance between every two basic features, and determining the distance correlation coefficient of each basic feature; Determining the importance of each of the basic features relative to the recoverable reserves based on a distance correlation coefficient of each of the basic features; Wherein, the distance correlation coefficient is expressed as: Where x is the basic feature; y is the recoverable reserves of the sample; dcor is the distance correlation coefficient; dcov(x,y) is the distance covariance between variables x and y, dvar(x,x) is the distance variance of variable x, and dvar(y,y) is the distance variance of variable y.

5. The method for predicting recoverable reserves of unconventional oil and gas in a single well according to claim 1, characterized in that: The steps of constructing a reserve prediction model based on a machine learning algorithm, using each input variable combination as the input of the reserve prediction model, using recoverable reserves as the output of the reserve prediction model, and training the reserve prediction model using corresponding sample data to obtain a candidate reserve prediction model include: constructing an objective function of the reserves prediction model according to a relative error between a prediction result of recoverable reserves output by the reserves prediction model and sample data of recoverable reserves; Iterating the reserve prediction model based on the objective function to adjust parameters of the reserve prediction model; When the objective function reaches a minimum value, determining the corresponding parameter as the target parameter of the reserve prediction model; The candidate reserve prediction model is determined based on the target parameters.

6. The method for predicting recoverable reserves of unconventional oil and gas in a single well according to claim 5, characterized in that: The step of iterating the reserve prediction model based on the objective function and adjusting the parameters of the reserve prediction model includes: The parameters of the reserve prediction model are adjusted by using a particle swarm algorithm or a grid search algorithm.

7. The method for predicting recoverable reserves of unconventional oil and gas in a single well according to claim 1, characterized in that: Before determining the importance of each of the basic characteristics relative to the recoverable reserves, the method further includes: Normalization processing is performed on each of the basic features.

8. A system for predicting recoverable reserves of unconventional oil and gas single wells, characterized by: include: A data server is used to obtain multiple basic characteristics and recoverable reserves sample data required for recoverable reserves prediction of sample wells; a computing server configured to determine the importance of each of the basic features relative to the recoverable reserves; screen the basic features multiple times according to the importance, and use the basic features screened out each time as an input variable combination, thereby obtaining multiple input variable combinations after multiple screenings; A reserve prediction model is constructed based on a machine learning algorithm, each input variable combination is used as an input of the reserve prediction model, recoverable reserves are used as an output of the reserve prediction model, and the reserve prediction model is trained using corresponding sample data to obtain a candidate reserve prediction model; based on the candidate reserve prediction model, a prediction result of recoverable reserves is output; based on the deviation between the prediction result of recoverable reserves and the sample data, the accuracy of the candidate reserve prediction model is analyzed, and the candidate reserve prediction model with the highest accuracy is selected as the preferred recoverable reserves prediction model, and the input variable combination input into the candidate reserve prediction model with the highest accuracy is selected as the preferred input variable combination; and recoverable reserves of a target well are predicted using the preferred recoverable reserves prediction model and the preferred input variable combination; The storage server is used to store the preferred recoverable reserves prediction model, the preferred input variable combination, and the predicted recoverable reserves of the target well.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images