A method and system for predicting permeability of complex reservoirs

Abstract

This invention, applicable to the field of reservoir permeability prediction, provides a method and system for predicting permeability in complex reservoirs. The method comprises the following steps: extracting stratigraphic characteristic data from complex unconventional reservoirs to establish a dataset; building a Bayesian optimization-bidirectional long-short-term memory network model based on the dataset, training the model, and performing preliminary predictions; and verifying the model's reservoir permeability predictions. This invention proposes a Bayesian-bidirectional long-short-term memory network, combining the strong adaptability and efficiency of Bayesian optimization with the bidirectional processing mechanism of BiLSTM to explore the complex relationship between different stratigraphic characteristics and permeability, thereby achieving accurate permeability prediction for complex unconventional reservoirs.

Description

Technical Field

[0001] The present invention belongs to the technical field of reservoir permeability prediction, and in particular relates to a method and system for predicting the permeability of a complex reservoir. Background Art

[0002] Unconventional reservoirs hold considerable oil and gas reserves and possess significant economic value. However, due to their complex lithology, mineral composition, and pore structure, conventional permeability evaluation models have become ineffective. Establishing permeability equations for different regions is also difficult with limited data. For example, altered igneous reservoirs exhibit significant lithology and pore structure variation, resulting in significant heterogeneity. Under the influence of alteration, not only does the mineral composition change, but the seepage pathways in the formation are also complexly affected. Therefore, empirical permeability calculation methods based on homogeneous formations are difficult to accurately determine the permeability of altered igneous reservoirs. Furthermore, the numerous influencing factors in complex reservoirs, the complex relationship between porosity and permeability, and the scattered distribution of data make it difficult to directly establish empirical formulas for regional porosity-permeability conversion, further posing a challenge to permeability prediction in these complex reservoirs. The rapid development of artificial intelligence algorithms has provided new insights into permeability prediction. However, most AI methods require large amounts of data, making them inapplicable to unconventional reservoirs characterized by diverse lithofacies, diverse lithologies, complex reservoir space structures, and strong heterogeneity.

[0003] LSTM (Long Short-Term Memory) is a time-recurrent neural network capable of handling complex problems. It overcomes the long-term dependency issues and the vanishing and exploding gradient problems of general neural networks when processing long sequences, but its ability to learn complex relationships is somewhat limited. BiLSTM (Bidirectional Long Short-Term Memory) is an optimized LSTM that can combine the forward and backward outputs to learn bidirectional dependencies in sequence data and process sequence data. It has a stronger ability to capture complex nonlinear relationships. However, BiLSTM's hyperparameters are difficult to tune, and manually selecting appropriate hyperparameters often requires a huge amount of time. Summary of the Invention

[0004] The purpose of the embodiments of the present invention is to provide a method for predicting the permeability of a complex reservoir, aiming to solve the problems raised in the above background technology.

[0005] The embodiment of the present invention is implemented as follows: a method for predicting the permeability of a complex reservoir comprises the following steps:

[0006] Extract stratigraphic characteristic data of complex unconventional reservoirs and build data sets;

[0007] Establish a Bayesian optimization-bidirectional long short-term memory network model based on the data set and perform model training and preliminary result prediction;

[0008] Verify the model's reservoir permeability prediction results.

[0009] Preferably, the steps of extracting stratigraphic characteristic data of complex unconventional reservoirs and establishing a data set specifically include:

[0010] The inversion results in a continuous curve used to evaluate the distribution of reservoir minerals and porosity characteristics;

[0011] Characteristic data of corresponding depths are extracted based on core physical property data to establish a permeability prediction data set for different lithologies, which includes a training set and a test set.

[0012] Preferably, the step of inverting to obtain a continuous curve for evaluating the distribution of reservoir minerals and porosity characteristics specifically includes:

[0013] Based on conventional logging curves and element curves, the continuous curves for evaluating reservoir pore and porosity characteristic distribution are inverted. The integrated optimization inversion of logging and well logging is adopted. The basic mathematical model is:

[0014]

[0015] Where, is the objective function in the optimization inversion process, are the unknown reservoir parameters and the volume content of the mineral components to be determined, The number of input curves selected for the optimal logging interpretation method, For each regional explanatory parameter in the response equation, is a set of actual logging value vectors corrected for environmental effects, The first The logging response equation is is the equality constraint, is the inequality constraint, and are measurement error and response equation error, respectively;

[0016] After establishing the logging and well logging response equations, the damped least squares optimization method is applied to minimize the corresponding objective function value and obtain theoretical logging values ​​close to the actual logging values.

[0017] The reservoir logging curves are input and inversion is performed to obtain a continuous curve that can describe the distribution of mineral and porosity characteristics of reservoirs with different lithologies.

[0018] Preferably, the steps of extracting characteristic data of corresponding depths based on core physical property data and establishing a permeability prediction data set for different lithologies are specifically as follows: using core porosity and permeability as labels, extracting corresponding characteristic values ​​from the reservoir characteristic curve according to the core sampling depth, and establishing a permeability prediction data set for complex unconventional reservoirs, wherein the reservoir characteristic curve includes conventional logging data, radioactive logging data, elemental data, mineral curves and porosity curves.

[0019] Preferably, the steps of establishing a Bayesian optimization-bidirectional long short-term memory network model based on the data set and performing model training and preliminary result prediction specifically include:

[0020] The Bayesian optimization algorithm is used to automatically select the optimal parameters of the bidirectional long short-term memory network;

[0021] The permeability training sets of different lithologies are input into the Bayesian optimization bidirectional long short-term memory network model for training and learning;

[0022] Input the test set to obtain the laboratory permeability prediction results.

[0023] Preferably, the step of verifying the reservoir permeability prediction result of the model specifically includes:

[0024] Use different intelligent algorithm evaluation indicators to analyze the prediction effect of the permeability prediction model of each lithology;

[0025] The corresponding characteristic curves of the actual formation are input into the prediction model and the core permeability is used to verify the reservoir permeability prediction results.

[0026] Another object of the present invention is to provide a complex reservoir permeability prediction system for implementing the above-mentioned complex reservoir permeability prediction method, comprising:

[0027] Data extraction unit, used to extract stratigraphic characteristic data of complex unconventional reservoirs and establish data sets;

[0028] A model building unit is used to establish a Bayesian optimization-bidirectional long short-term memory network model based on the data set and perform model training and preliminary result prediction;

[0029] Verification unit, used to verify the reservoir permeability prediction results of the model.

[0030] The present invention provides a method for predicting the permeability of complex reservoirs. This method classifies complex data sets based on lithology, effectively preventing differences in complex characteristics such as mineral composition and pore structure of different lithologies from affecting the model prediction results.

[0031] A Bayesian-bidirectional long short-term memory network (BO-BiLSTM) was proposed to overcome the difficulty of hyperparameter debugging faced by the BiLSTM method. By combining the strong adaptability and efficiency of Bayesian optimization with the bidirectional processing mechanism of BiLSTM, it explores the complex relationship between different formation characteristics and permeability, improves the efficiency of reservoir permeability prediction, and realizes accurate permeability prediction of complex unconventional reservoirs. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A flowchart of a complex reservoir permeability prediction method provided by an embodiment of the present invention;

[0033] Figure 2 A schematic diagram of a permeability prediction process is provided in an embodiment of the present invention, taking an altered diorite reservoir in an altered igneous rock reservoir as an example;

[0034] Figure 3 Comparison results between the complex unconventional reservoir mineral inversion results and core results provided by the embodiment of the present invention;

[0035] Figure 4 A schematic diagram of the Bayesian optimization-bidirectional long short-term memory network permeability prediction principle provided by an embodiment of the present invention;

[0036] Figure 5 A graph showing the relationship between the Bayesian optimization-bidirectional long short-term memory network penetration rate and actual data provided by an embodiment of the present invention;

[0037] Figure 6 The permeability prediction result of the altered igneous rock reservoir in Well X in a certain area provided by the embodiment of the present invention;

[0038] Figure 7 A structural block diagram of a complex reservoir permeability prediction system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0040] The specific implementation of the present invention is described in detail below with reference to specific embodiments.

[0041] like Figure 1 FIG. 1 is a flowchart of a method for predicting permeability of a complex reservoir provided by an embodiment of the present invention, comprising the following steps:

[0042] S1. Extract stratigraphic characteristics of complex unconventional reservoirs and establish a data set: First, invert to obtain a continuous curve that can evaluate the distribution of reservoir minerals and porosity characteristics. Based on the core physical property data, characteristic data corresponding to the depth are extracted. The data is classified according to lithology to establish a permeability prediction data set for different lithologies. The permeability prediction data set for each lithology is divided into a training set and a test set according to a certain ratio.

[0043] S2. Establish a Bayesian optimization-bidirectional long short-term memory network model and perform model training and preliminary result prediction: Use the Bayesian optimization algorithm to automatically select the optimal parameters of the bidirectional long short-term memory network. Then, input the permeability training set of different lithologies into the model for training and learning, and input the test set to obtain permeability prediction results;

[0044] S3. Verify the effectiveness of the permeability prediction model for complex unconventional reservoirs and actual formation evaluation: Use different intelligent algorithm evaluation indicators to analyze the prediction effect of the permeability prediction model for each lithology, input the corresponding characteristic curve of the actual formation into the prediction model, and use the core permeability to verify the reservoir permeability prediction results.

[0045] Taking the altered diorite reservoir in the altered igneous rock reservoir as an example, the permeability prediction process is as follows: Figure 2 The specific steps are as follows:

[0046] S1. Extract stratigraphic characteristics of complex unconventional reservoirs and establish a data set:

[0047] S11. Inversion to obtain a curve that can be used to evaluate reservoir mineral and porosity distribution:

[0048] The basic data describing reservoir characteristics include well logging curves and mud logging curves. Curves for evaluating reservoir pore and mineral characteristics, such as porosity and mineral component distribution, can be inverted based on conventional well logging curves and element curves. The commonly used method is the comprehensive optimization inversion of logging and mud logging. The basic mathematical model is:

[0049]

[0050] Where, is the objective function in the optimization inversion process, are the unknown reservoir parameters and the volume content of the mineral components to be determined, The number of input curves selected for the optimal logging interpretation method, For each regional explanatory parameter in the response equation, is a set of actual logging value vectors corrected for environmental effects, The first The logging response equation is is the equality constraint, is the inequality constraint, and are measurement error and response equation error, respectively;

[0051] After establishing the logging and well logging response equations, the damped least squares (LM) optimization method is applied to minimize the corresponding objective function value, and the theoretical logging value that is sufficiently close to the actual logging value can be obtained. It is the optimal solution for the volume content of different components. When the input data is the reservoir logging curve, it can be inverted to obtain a continuous curve that can describe the mineral distribution and porosity of reservoirs with different lithologies.

[0052] In order to verify the accuracy of the inversion results, core mineral data is used for verification. However, the core mineral content is in weight percentage. Therefore, the mineral inversion module of the logging software can be used to invert the complex unconventional reservoir minerals, and the output mineral weight content curve is compared with the core test results. Taking the altered igneous rock reservoir as an example, Figure 3 As shown in the figure, the inversion results are basically consistent with the core mineral data and can be used to evaluate the reservoir mineral characteristics. However, in the permeability prediction, the volume content mineral results in the inversion results in the same system as the porosity are selected;

[0053] S12. Extract characteristic data of corresponding depth based on core physical property data and establish permeability prediction data sets for different lithologies:

[0054] Laboratory core physical property analysis data is an important basis for evaluating reservoir porosity and permeability. Therefore, using core porosity and permeability as labels and extracting corresponding characteristic values ​​from the reservoir characteristic curve according to the core sampling depth, a permeability prediction data set for complex unconventional reservoirs can be established. Commonly used reservoir characteristic curves include conventional logging data such as natural gamma (GR), acoustic transit time (AC), compensated neutron (CNL), compensated density (DEN), radioactive logging data such as radioactive potassium (40K), radioactive uranium (U), radioactive thorium (Th), uranium-free gamma (KTh), and sodium (Na). , magnesium (Mg), aluminum (Al), silicon (Si), phosphorus (P), potassium (K), calcium (Ca), iron (Ti), manganese (Mn) and other element data (element logging or element logging), as well as the mineral curves and porosity curves obtained by inversion; Since the pore structures of different lithologies vary greatly, after extracting the above data, it is necessary to classify the data sets according to lithology and establish permeability prediction data sets for different lithologies; Then, according to the commonly used data set allocation ratio (8:2), the permeability prediction data sets of each lithology are divided into training sets and test sets;

[0055] S2. Establish a Bayesian optimization-bidirectional long short-term memory network model and perform model training and preliminary result prediction:

[0056] S21. Use the Bayesian optimization algorithm to automatically select the optimal parameters of the bidirectional long short-term memory network:

[0057] BiLSTM is based on LSTM, which is based on a recurrent neural network (RNN) and adds input, forget, and output gates to eliminate interference information while selectively recursively processing data based on time or spatial sequences. This solves the gradient vanishing or exploding problem faced by RNN. The relevant functions and key parameters of the processing process are as follows:

[0058] Data input:

[0059] Input Gate:

[0060] Forget Gate:

[0061] unit:

[0062] Output gate:

[0063] Result output:

[0064] In the above formula, and is the activation function, is the weight, for deviation;

[0065] BiLSTM learns by combining bidirectional dependencies with LSTM. It consists of an input layer responsible for receiving input sequence data, two LSTM layers (a forward LSTM layer that processes the input sequence from front to back, and a reverse LSTM layer that processes the input sequence from back to front), and an output layer that concatenates the outputs of the two LSTM layers to obtain the final output vector. During operation, after the input sequence data is input into the BiLSTM network, it is first processed by the forward LSTM to generate a forward hidden state sequence. Subsequently, the reverse LSTM is processed in the reverse order of the input sequence to generate a reverse hidden state sequence. The hidden state sequences of the forward and reverse LSTM layers are then concatenated to obtain the final output vector. However, in BiLSTM, the window size must be considered, that is, the range to which the current prediction target is affected by the previous and next data. A reasonable window selection is required to obtain reasonable prediction results. Bayesian optimization is needed to solve this problem.

[0066] Bayesian Optimization (BO) is a black-box algorithm used to optimize the objective function. It is suitable for high-dimensional and high-noise problems that are computationally expensive or difficult to calculate. Bayesian optimization approximates the objective function by constructing a surrogate model (such as a Gaussian process, random forest, or expected improvement) and uses Bayesian inference to select the next sampling point to efficiently find the global optimal solution. In this embodiment of the present invention, a Gaussian process commonly used in programming software is used to construct the surrogate model, and the expected improvement method is used to select the sampling point, thereby establishing a Bayesian optimized bidirectional long short-term memory network model BO-BiLSTM;

[0067] S22. Input the permeability training set of different lithologies into the Bayesian optimization bidirectional long short-term memory network model for training and learning:

[0068] After selecting appropriate formation features and prediction targets for input into the network, a convolutional neural network (CNN) is used, which includes convolution calculations and a feedforward neural network with a deep structure. The relevant data is processed through a convolution layer (BN is the convolution kernel and LeakRelu is the activation function) and a pooling layer to complete feature extraction and flattening. Subsequently, the flattened features are input into the BO-BiLSTM model for training, which can realize the permeability prediction of altered igneous rock reservoirs. Figure 4 As shown;

[0069] S23. Input the test set to obtain the laboratory permeability prediction results:

[0070] In order to analyze the reliability of the BO-BiLSTM permeability prediction model, the reservoir permeability test set that has not participated in the training is input into the prediction model, and then the laboratory permeability prediction results can be obtained. Taking the altered diorite reservoir in the altered igneous rock reservoir as an example, the input features include conventional logging data such as GR, AC, CNL, and DEN. 40 Radioactive logging data such as K, U, Th, KTh, standardized element logging data such as Na, Mg, Al, Si, K, Ga, Fe, Mn, and porosity curves obtained by mineral inversion, and prediction results such as Figure 5 As shown in the figure, it shows that the predicted permeability and the actual data have an obvious positive correlation in the double logarithmic coordinate system, and the order of magnitude is basically consistent, and the numerical prediction results are reliable;

[0071] S3. Verification of the effectiveness of the permeability prediction model for complex unconventional reservoirs and actual formation evaluation:

[0072] S31. Use different intelligent algorithm evaluation indicators to analyze the prediction effect of the permeability prediction model for each lithology:

[0073] The parameters used to evaluate the reliability of the prediction model include mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE) and coefficient of determination ( ), the formulas for each evaluation parameter are as follows:

[0074]

[0075] In the above formula, represents the sample point number, is the total number of samples, For the The actual data value of the sample points, For the The prediction results of sample points, is the average value of the prediction results;

[0076] Among them, MSE, RMSE and MAE are used to measure the error of model prediction. The smaller the value, the higher the model prediction accuracy. The determination coefficient is used to reflect the model's ability to explain the variables. The larger the value, the higher the degree of model fitting. Taking the permeability prediction results of altered diorite reservoir as an example, the BO-BiLSTM predicted permeability, MSE, RMSE, MAE and The evaluation indices are 0.27, 0.52, 0.41 and 0.74, respectively. Among them, the mean square error, root mean square error and mean absolute error are no higher than 0.6, and the coefficient of determination is greater than 0.7, indicating that the model has a good prediction effect in complex unconventional reservoirs;

[0077] S32. Input the corresponding characteristic curve of the actual formation into the prediction model and use the core permeability to verify the reservoir permeability prediction result:

[0078] In actual reservoirs, formation permeability can be predicted based on well data such as logging and mud logging, and then the results can be verified based on core data at the corresponding depth. Taking Well X in a certain area as an example, the input data for permeability prediction include conventional logging data such as natural gamma, acoustic transit time, compensated neutron, and compensated density, logging data such as potassium, uranium, thorium, and uranium-free gamma, element curves such as sodium, magnesium, aluminum, silicon, potassium, calcium, iron, and manganese, as well as inverted mineral curves and porosity curves. The comparison results between the predicted permeability results and the actual core data are shown in the figure below. Figure 6As shown in the figure, it can be seen that the permeability predicted by BO-BiLSTM is basically consistent with the core permeability in the logarithmic coordinate system, but the errors of individual data points are large, which may be caused by the systematic error of the logging resolution. Overall, the predicted permeability results in the altered igneous rock reservoir of Well X meet the exploration and development needs of complex unconventional reservoirs. The results are reliable and can provide strong support for reservoir evaluation.

[0079] like Figure 7 FIG. 1 is a block diagram of a complex reservoir permeability prediction system according to an embodiment of the present invention, comprising:

[0080] The data extraction unit S100 is used to extract stratigraphic characteristic data of complex unconventional reservoirs and establish a data set;

[0081] The model building unit S200 is used to establish a Bayesian optimization-bidirectional long short-term memory network model and perform model training and preliminary result prediction;

[0082] The verification unit S300 is used to verify the reservoir permeability prediction result of the model.

[0083] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting the permeability of a complex reservoir, characterized in that: The following steps are involved: Extract stratigraphic characteristic data of complex unconventional reservoirs and build data sets; Establish a Bayesian optimization-bidirectional long short-term memory network model based on the data set and perform model training and preliminary result prediction; Verify the model's reservoir permeability prediction results; The steps of extracting stratigraphic characteristic data of complex unconventional reservoirs and establishing a data set specifically include: The inversion results in a continuous curve used to evaluate the distribution of reservoir minerals and porosity characteristics; Extract characteristic data of corresponding depths based on core physical property data and establish a permeability prediction dataset for different lithologies, the dataset including a training set and a test set; The steps of extracting characteristic data corresponding to the depth based on the core physical property data and establishing a permeability prediction data set for different lithologies are specifically as follows: using the core porosity and permeability as labels, extracting corresponding characteristic values ​​from the reservoir characteristic curve according to the core sampling depth, and establishing a permeability prediction data set for complex unconventional reservoirs, wherein the reservoir characteristic curve includes conventional well logging data, radioactive well logging data, elemental data, mineral curves, and porosity curves; The steps of establishing a Bayesian optimization-bidirectional long short-term memory network model based on the data set and performing model training and preliminary result prediction include: The Bayesian optimization algorithm is used to automatically select the optimal parameters of the bidirectional long short-term memory network; The permeability training sets of different lithologies are input into the Bayesian optimization bidirectional long short-term memory network model for training and learning; Input the test set to obtain the laboratory permeability prediction results.

2. The method for predicting the permeability of a complex reservoir according to claim 1, wherein: The steps of inverting to obtain a continuous curve for evaluating the distribution of reservoir minerals and porosity characteristics include: Based on conventional logging curves and element curves, the continuous curves for evaluating reservoir pore and porosity characteristic distribution are inverted. The comprehensive optimization inversion of logging and well logging is adopted. The basic mathematical model is: ； Where, is the objective function in the optimization inversion process, are the unknown reservoir parameters and the volume content of the mineral components to be determined, The number of input curves selected for the optimal logging interpretation method, For each regional explanatory parameter in the response equation, is a set of actual logging value vectors corrected for environmental effects, The first The logging response equation is is the equality constraint, is the inequality constraint, and are measurement error and response equation error, respectively; After establishing the logging and well logging response equations, the damped least squares optimization method is applied to minimize the corresponding objective function value and obtain theoretical logging values ​​close to the actual logging values. The reservoir logging curves are input and inversion is performed to obtain a continuous curve that can describe the distribution of mineral and porosity characteristics of reservoirs with different lithologies.

3. The method for predicting the permeability of a complex reservoir according to claim 1, wherein: The steps to verify the model's reservoir permeability prediction results include: Use different intelligent algorithm evaluation indicators to analyze the prediction effect of the permeability prediction model of each lithology; The corresponding characteristic curves of the actual formation are input into the prediction model and the core permeability is used to verify the reservoir permeability prediction results.

4. A complex reservoir permeability prediction system, used to implement the complex reservoir permeability prediction method according to any one of claims 1 to 3, characterized in that: include: Data extraction unit, used to extract stratigraphic characteristic data of complex unconventional reservoirs and establish data sets; A model building unit is used to establish a Bayesian optimization-bidirectional long short-term memory network model based on the data set and perform model training and preliminary result prediction; Verification unit, used to verify the reservoir permeability prediction results of the model.

Images