Oil and gas reservoir physical property parameter prediction precision improving method, system, equipment and medium

Abstract

The invention discloses an oil and gas reservoir physical property parameter prediction precision improving method, system, equipment and medium, and the method comprises the steps: 1, standardizing logging curve data and actually measured reservoir parameter data, dividing the data into a training set and a test set, inputting the data of the training set into a prediction model for training until convergence, and obtaining a trained prediction model; and finally, predicting test set data by using the model, and outputting a final prediction result. Step 2, constructing an improved VAE-Attention model, introducing Gaussian noise into training set data to enhance model robustness, and then optimizing model hyper-parameters by using a genetic algorithm; inputting the optimal parameters into the model for training to obtain a final model; and finally, taking the prediction result as input, further processing through a final model, and outputting a final operation result. The system, the equipment and the medium are used for implementing the method. The method is high in precision and has good stability, reliability and applicability.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas reservoir physical property parameter prediction, and in particular to a method, system, device and medium for improving the prediction accuracy of oil and gas reservoir physical property parameters based on a VAE-Attention model of an improved variational autoencoder. Background Art

[0002] In the process of oil and gas field development, reservoir physical property parameters are one of the important parameters that affect oil and gas production. High-precision prediction of reservoir physical property parameters can not only improve oil and gas recovery, but also optimize oil field development plans and reduce production costs. Traditional reservoir physical property parameter prediction methods are mainly divided into two categories: physical model-based methods and data-driven methods. Physical model-based methods rely on a detailed understanding of geological structure and physical properties, usually require a large amount of experimental data and complex calculation processes, and have limited applicability. Data-driven methods use machine learning and deep learning technologies to establish a reservoir permeability prediction model by mining and analyzing historical data.

[0003] With the deepening of deep learning technology, many new models have emerged, which have significantly promoted the development of sequence data processing capabilities. Among them, recurrent neural networks (RNNs), as pioneers in processing time series data, have shown unique advantages in sequence modeling due to their recursive characteristics. However, the gradient vanishing and exploding problems faced by RNNs when processing long sequences limit their scope of application. To solve this problem, long short-term memory networks (LSTMs) and their bidirectional variants (BiLSTMs) have been proposed one after another. LSTMs effectively alleviate the gradient problem through a cleverly designed gating mechanism and successfully capture long-term dependencies in the data. BiLSTMs go a step further, taking into account the contextual information of the sequence and improving the model's ability to fully understand the context. In addition, in pursuit of higher computational efficiency, gated recurrent units (GRUs) and their bidirectional versions (BiGRUs) have also emerged. GRUs achieve faster training speeds and more efficient computing performance by simplifying the gating structure of LSTMs. BiGRUs inherit BiLSTM's ability to capture bidirectional dependencies of sequences while maintaining the efficiency of GRUs.

[0004] For the reservoir physical property parameter prediction task, the invention patent application with application number CN202010197833.7 proposes a porosity prediction method based on a multi-layer long short-term memory neural network model. This method has shown good results in reservoir prediction tasks, but like other prediction methods, this method only considers the performance of the constructed model in the prediction task, without further refining the final prediction results, so as to improve its accuracy. Existing data reconstruction technologies using unsupervised learning include generative adversarial networks (GAN) and variational autoencoders (VAE), among which GAN generates realistic data samples through adversarial training, which can be used to improve the diversity and quality of training data sets. However, it is more of an improvement in the quality of feature data, and is not ideal in the processing of subsequent prediction results. VAE is used to extract latent variables from the real data distribution and generate new data points. By reconstructing the data, VAE can capture the complex structure of the data, which helps to improve the performance of the prediction model. Although traditional VAE can extract a distribution of real data, it uses simple encoding and decoding networks, which may be insufficient in capturing the time-dependent features of complex sequence data, resulting in information loss. At the same time, traditional VAE has no attention mechanism and is difficult to focus on important features in the input sequence, resulting in a lack of global consistency in the generated samples. Summary of the invention

[0005] In view of the above problems, the purpose of the present invention is to provide a method, system, device and medium for improving the prediction accuracy of oil and gas reservoir physical property parameters. The method improves the traditional variational autoencoder, replaces the original encoding layer and decoding layer in the VAE structure with a BiGRU layer that is better capable of extracting sequence features, and adds an Attention mechanism to focus on the global input, dynamically calculates the attention weight of each time step, so as to better capture the distribution characteristics of data under long-term changes, and further improve the accuracy of the prediction effect of the prediction model itself.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] The method for improving the prediction accuracy of oil and gas reservoir physical property parameters comprises the following steps:

[0008] Step 1: Prediction model and operation:

[0009] After the well logging curve data and measured reservoir parameter data are standardized, they are divided into training set and test set according to a predetermined ratio; then, the training set data is input into the prediction model for training until the model converges, and a trained prediction model is obtained; finally, the model is used to predict the test set data and the final prediction result is output;

[0010] Step 2: Improved VAE-Attention model and operation:

[0011] An improved VAE-Attention model is constructed, and the measured reservoir parameter data is standardized. The training set and the test set are also divided according to the proportion in step one. Gaussian noise is introduced into the training set data to enhance the robustness of the model. Then, the genetic algorithm is used to optimize the model hyperparameters to determine the optimal parameter combination. The optimal parameters are input into the model for training until the model is stable to obtain the final model. Finally, the prediction results in step one are used as input, further processed by the final model, and the final operation results of the model are output.

[0012] The detailed steps in step 1 are as follows: standardize the collected logging curve data and measured reservoir parameter data, select Z-Score standardization in the data standardization method, and the Z-Score standardization performs data standardization based on the mean and standard deviation of each feature. For each feature, the mean and standard deviation of all samples of the feature are calculated, and then the feature is standardized so that the mean of each feature is 0 and the standard deviation is 1, that is, the converted feature conforms to the standard normal distribution, and its expression is:

[0013]

[0014] Among them, x is the original data, μ is the mean of the data, and σ is the standard deviation of the data;

[0015] The standardized data set is divided into a training set and a test set. Then, the training set data is input into four prediction models, namely BiGRU, LSTM, BiLSTM and RNN, for training. After each model is trained, its performance is evaluated using the test set data, and the final prediction results are saved. In order to quantify the prediction accuracy, the mean square error (MSE) is used as the error statistical indicator, and its expression is:

[0016]

[0017] Where n is the number of samples, y i is the true value, is the predicted value of the model.

[0018] The detailed steps of step 2 are as follows:

[0019] (2.1) Construct a VAE-Attention model, wherein the VAE-Attention model includes an encoder and a decoder;

[0020] (2.1.1), the encoder (q(z|x)) maps the reservoir parameter input data to the latent space and outputs the mean μ(x) and variance σ of the latent variable z2 (x), the encoder adopts a BiGRU structure, in which each GRU unit contains an update gate and a reset gate, and the corresponding mathematical expression is as follows:

[0021] Reset Gate t :

[0022] r t =σ(W r ·[h t-1 ,x t ])

[0023] Candidate hidden states

[0024]

[0025] Update gate z t :

[0026] z t =σ(W z ·[h t-1 ,x t ])

[0027] Update hidden state h t :

[0028]

[0029] Among them, x t represents the reservoir permeability data at the tth time step of the input sequence, h t-1 represents the hidden state of the previous time step, W r , W z and W represent the weight parameters of the reset gate, update gate, and candidate hidden state, respectively, σ is the sigmoid function, and ⊙ represents element-by-element multiplication;

[0030] After extracting the features of the permeability data using the BiGRU structure, it is distributed to the corresponding latent space through the linear layer, namely:

[0031] q(z|x)=N(z;μ(x),σ 2 (x));

[0032] The mean μ(x) and variance σ of the output latent variable z are 2 (x), in order to enable the gradient to be passed through the random variable, the reparameterization technique is used to represent the encoder's encoded data as a function of the mean and variance, that is:

[0033] z=μ(x)+σ(x)⊙∈

[0034] Among them, ∈~N(0,1), represents the noise of standard normal distribution, ⊙ represents element-by-element multiplication;

[0035] (2.1.2) The decoder maps the low-dimensional potential representation back to the original data space, thereby realizing data reconstruction; the decoding layer also adopts the BiGRU structure, and reconstructs the re-parameterized latent variable z back to the original data through the BiGRU and Attention mechanism, specifically:

[0036] First, map the reparameterized latent variable z to the initial hidden state h0:

[0037] h0=f c (z)

[0038] Then use GRU to generate data sequences:

[0039] o t ,h t =GRU(o {t-1} ,h {t-1} )

[0040] Among them t ,h t Respectively represent the output and hidden state at t time step;

[0041] The current output of the decoder is o t and the hidden state h t Associated with the calculation of the attention weight, an attention layer is then used to convert the current decoder output o t and the previous hidden state h t-1 Connect it up:

[0042] score(o t ,h t-1 )=attention(o t ,h t-1 )

[0043] Then normalize the attention score through the Softmax function to get the attention weight a t :

[0044]

[0045] Then use the attention weight a t The output of the encoder is weighted and summed to obtain the context vector c t :

[0046]

[0047] Among them, i is the output sequence of the encoder, at,i is the attention weight;

[0048] Finally, reconstruct the output using the context vector c t Combined with the decoder's current hidden state h t , through a fully connected layer f to generate the final reconstruction output

[0049]

[0050] The goal of the VAE model is to maximize the marginal likelihood p(x) of the data, that is:

[0051] logp(x)=log(x|z)p(z)dz

[0052] Using the variational inference method, we introduce an approximate posterior distribution q(z|x), and obtain the following lower bound through Jensen's inequality:

[0053]

[0054] Among them, D KL is the KL divergence, which measures the difference between two distributions;

[0055] The loss function of the improved VAE-Attention model consists of reconstruction loss and KL divergence loss:

[0056]

[0057] Among them, the reconstruction loss is to measure the reconstruction data The difference between the original data x, usually using the mean square error (MSE); KL divergence loss measures the difference between the approximate posterior distribution of the encoder output and the prior distribution, which is usually a standard normal distribution;

[0058] (2.2) Model training

[0059] (2.2.1) Data standardization and division

[0060] Standardize the collected measured reservoir parameter data, select Z-Score standardization in the data standardization method, and divide the standardized data into training data and test data with a ratio of 9:1;

[0061] (2.2.2) Introducing noise

[0062] The introduced noise is Gaussian noise, and its expression is:

[0063] x noisy =x+∈,

[0064] Among them, x is the reservoir permeability data, ∈ is subject to N(0,σ 2 ) random variables;

[0065] (2.2.3) Hyperparameter Tuning

[0066] Select the genetic algorithm for hyperparameter tuning, build the genetic algorithm framework, and tune the hyperparameters in the model. The tuning range is shown in the following table:

[0067] Hyperparameters Experimental setup Hidden layer dimension 32-512 Latent space vector dimension 16-256 Learning Rate 0.001-0.01 Batch size 32-512 Noise variance 0.0-0.1 Iterations 50-200

[0068] (2.2.4) Model training and application

[0069] The optimal parameters are input into the model for training until the model is stable and the final model is obtained; finally, the prediction results in step one are used as input, further processed by the final model, and the final operation results of the model are output.

[0070] The system for improving the accuracy of predicting oil and gas reservoir physical property parameters is applied in the process of oil and gas field development to realize the above-mentioned method for improving the accuracy of predicting oil and gas reservoir physical property parameters, including a processing instruction module for executing each step in the method.

[0071] Equipment for improving the accuracy of oil and gas reservoir physical property parameter prediction, including:

[0072] Memory: a computer program for storing the method for improving the prediction accuracy of oil and gas reservoir physical property parameters, which is a computer-readable device;

[0073] Processor: used to implement the method for improving the prediction accuracy of oil and gas reservoir physical property parameters when executing the computer program.

[0074] A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, can implement the method for improving the prediction accuracy of oil and gas reservoir physical property parameters.

[0075] The present invention adopts the above technical solution and has the following beneficial effects:

[0076] (1) Step 2 of the present invention combines the BiGRU model's better ability to extract sequence data features with the Attention mechanism's global attention characteristics, and modifies the structure of the original VAE model, utilizing its unsupervised learning characteristics and its data reconstruction characteristics to achieve the effect of improving prediction accuracy.

[0077] (2)) The present invention selects a genetic algorithm in the model hyperparameter tuning method and performs a comprehensive screening of the model's hyperparameters, thereby obtaining the model operation effect under the optimal parameters.

[0078] In summary, the present invention is different from the traditional VAE model which mainly focuses on anomaly recognition. Instead, it is based on the improved VAE-Attention model of variational autoencoder, which utilizes its unsupervised learning characteristics to learn the distribution characteristics within the data, thereby better improving the prediction accuracy of the original model. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 Flowchart for the improved VAE-Attention application.

[0080] Figure 2 This is the improved VAE-Attention structure diagram.

[0081] Figure 3 This is the GRU unit structure diagram.

[0082] Figure 4 is the improved VAE-Attention training loss graph, where Figure 4 (a) is the training loss graph of offspring 1, (b) is the training loss graph of offspring 2, (c) is the training loss graph of offspring 3, (d) is the training loss graph of offspring 4, and (e) is the training loss graph of offspring 5.

[0083] Figure 5 This is a comparison chart of the improved VAE-Attention application effect, where: Figure 5 (a) is the reconstruction value and true value curve of BiGRU, (b) is the prediction value and true value curve of BiGRU, (c) is the reconstruction value and true value curve of LSTM, (d) is the prediction value and true value curve of LSTM, (e) is the reconstruction value and true value curve of BiLSTM, (f) is the prediction value and true value curve of BiLSTM, (g) is the reconstruction value and true value curve of RNN, and (h) is the prediction value and true value curve of RNN. DETAILED DESCRIPTION

[0084] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0085] The present invention designs a VAE-Attention model based on an improved variational autoencoder, aiming to improve the prediction accuracy of the prediction model. The encoder and decoder in the VAE are replaced with the BiGRU layer with the feature of extracted sequence data, and then combined with the Attention mechanism to dynamically calculate the attention weight of each time step according to the output of the encoder and the current state of the decoder. In this way, the model can more accurately focus on the input part related to the current generation position, improving the utilization efficiency of information at different positions in the input sequence.

[0086] like Figure 1As shown, the method for improving the prediction accuracy of oil and gas reservoir physical property parameters includes the following steps:

[0087] Step 1: Prediction model and operation:

[0088] After the well logging curve data and measured reservoir parameter data are standardized, they are divided into training set and test set according to the established ratio. Subsequently, the training set data is input into the prediction model for training until the model converges and a trained prediction model is obtained. Finally, the model is used to predict the test set data and output the final prediction results.

[0089] The detailed steps in step 1 are as follows:

[0090] The collected logging curve data and measured reservoir parameter data are standardized. In the data standardization method, Z-Score standardization is selected. The Z-Score standardization performs data standardization based on the mean and standard deviation of each feature. For each feature, the mean and standard deviation of all samples of the feature are calculated, and then the feature is standardized so that the mean of each feature is 0 and the standard deviation is 1, that is, the converted feature conforms to the standard normal distribution. The expression is:

[0091]

[0092] Among them, x is the original data, μ is the mean of the data, and σ is the standard deviation of the data.

[0093] The standardized data set is divided into a training set and a test set at a ratio of 9:1. Subsequently, the training set data is input into four prediction models (BiGRU, LSTM, BiLSTM, RNN) for training. After each model is trained, its performance is evaluated using the test set data, and the final prediction results are saved. In order to quantify the prediction accuracy, the mean square error (MSE) is used as the error statistical indicator, and its expression is:

[0094]

[0095] Where n is the number of samples, y i is the true value, is the predicted value of the model. The prediction results are shown in Table 2. It can be seen that the BiGRU model shows better results in prediction accuracy.

[0096] Table 2 Mean square error of different prediction models

[0097] Neural Network Model Mean square error LSTM 1.2116 BiLSTM 1.043 BiGRU 0.7911 RNN 1.4372 .

[0098] Step 2: Improved VAE-Attention model and operation:

[0099] An improved VAE-Attention model is constructed, and the measured reservoir parameter data is standardized. The training set and the test set are also divided according to the proportion in step 1. Gaussian noise is introduced into the training set data to enhance the robustness of the model. Then the genetic algorithm is used to optimize the model hyperparameters and determine the optimal parameter combination. The optimal parameters are input into the model for training until the model is stable and the final model is obtained. Finally, the prediction results in step 1 are used as input, further processed by the final model, and the final operation results of the model are output.

[0100] The detailed steps of step 2 are as follows, refer to Figure 2 :

[0101] (2.1) Construct a VAE-Attention model, wherein the VAE-Attention model includes an encoder and a decoder;

[0102] (2.1.1), the encoder (q(z|x)) maps the reservoir parameter input data to the latent space and outputs the mean μ(x) and variance σ of the latent variable z 2 (x), the encoder adopts BiGRU structure, the structure refers to Figure 3 , where each GRU unit contains an update gate and a reset gate, and the corresponding mathematical expression is as follows:

[0103] Reset Gate t :

[0104] r t =σ(W r ·[h t-1 ,x t ])

[0105] Candidate hidden states

[0106]

[0107] Update gate z t :

[0108] z t =σ(W z ·[h t-1 ,x t ])

[0109] Update hidden state h t :

[0110]

[0111] Among them, x t represents the reservoir permeability data at the tth time step of the input sequence, ht-1 represents the hidden state of the previous time step, W r , W z and W represent the weight parameters of the reset gate, update gate, and candidate hidden state, respectively, σ is the sigmoid function, and ⊙ represents element-wise multiplication.

[0112] After extracting the features of the permeability data using the BiGRU structure, it is distributed to the corresponding latent space through the linear layer. That is:

[0113] q(z|x)=N(z;μ(x),σ 2 (x));

[0114] The mean μ(x) and variance σ of the output latent variable z are 2 (x), in order to enable the gradient to be passed through the random variable, the reparameterization technique is used to represent the encoder's encoded data as a function of the mean and variance, that is:

[0115] z=μ(x)+σ(x)⊙∈

[0116] Among them, ∈~N(0,1) represents the noise of standard normal distribution, and ⊙ represents element-by-element multiplication.

[0117] (2.1.2), the decoder maps the low-dimensional latent representation back to the original data space, thereby realizing data reconstruction. This patent replaces the decoding layer with the BiGRU structure, and reconstructs the re-parameterized latent variable z back to the original data through the BiGRU and Attention mechanism. The Attention mechanism mainly acts on the decoder process, and is used to enhance the accuracy and expressiveness of the model when generating data. In the decoder, the Attention mechanism helps the decoder to dynamically adjust the attention to the input sequence according to the latent variable z and the current output state, thereby generating a more accurate output. Specifically:

[0118] First, map the reparameterized latent variable z to the initial hidden state h0:

[0119] h0=f c (z)

[0120] Then use GRU to generate data sequences:

[0121] o t ,h t =GRU(o {t-1} ,h {t-1} )

[0122] Among them, t ,h t They represent the output and hidden state at time step t respectively.

[0123] The current output of the decoder is o t and the hidden state h t Associated with the calculation of attention weights. Then use an attention layer to convert the current decoder output o t and the previous hidden state h t-1 Connect it up:

[0124] score(o t ,h t-1 )=attention(o t ,h t-1 )

[0125] Then normalize the attention score through the Softmax function to get the attention weight a t :

[0126]

[0127] Then use the attention weight a t The output of the encoder is weighted and summed to obtain the context vector c t :

[0128]

[0129] Among them, i is the output sequence of the encoder, a t,i is the attention weight.

[0130] Finally, reconstruct the output using the context vector c t Combined with the decoder's current hidden state h t , through a fully connected layer f to generate the final reconstruction output

[0131]

[0132] It should be noted that the goal of the VAE model is to maximize the marginal likelihood p(x) of the data, that is:

[0133] log p(x)=log(x|z)p(z)dz

[0134] Since it is difficult to directly calculate this integral, we use the variational inference method to introduce an approximate posterior distribution q(z|x). Through Jensen's inequality, we can get the following lower bound:

[0135]

[0136] Among them, D KL is the KL divergence, which measures the difference between two distributions.

[0137] The loss function of the improved VAE-Attention model consists of reconstruction loss and KL divergence loss:

[0138]

[0139] The reconstruction loss is used to measure the reconstruction data The difference between the original data x, usually using the mean square error (MSE). KL divergence loss measures the difference between the approximate posterior distribution of the encoder output and the prior distribution, which is usually a standard normal distribution.

[0140] (2.2) Model training

[0141] (2.2.1) Data standardization and division

[0142] The collected measured reservoir parameter data are standardized, and Z-Score standardization is selected in the data standardization method. The standardized data are divided into training data and test data with a ratio of 9:1.

[0143] (2.2.2) Introducing noise

[0144] The purpose of introducing noise is to better increase the generalization ability of the model. The noise introduced by the present invention is Gaussian noise.

[0145] Then its expression is:

[0146] x noisy =x+∈,

[0147] Among them, x is the reservoir permeability data, ∈ is subject to N(0,σ 2 ) is a random variable.

[0148] (2.2.3) Hyperparameter Tuning

[0149] Select the genetic algorithm for hyperparameter tuning. Construct the genetic algorithm framework and tune the hyperparameters in the model. The tuning range is shown in Table 1. In the genetic algorithm framework, the population size is set to 20, the number of offspring is set to 5, the crossover probability is set to 0.7, and the mutation probability is set to 0.2. The optimizer used in the model training and parameter tuning process is the Adam optimizer. The optimal parameters of the five offspring after final tuning are shown in Table 3.

[0150] Table 3 Optimal parameters of five offspring of genetic algorithm

[0151] Offspring 1 2 3 4 5 Hidden layer dimension 42 372 42 421 372 Latent space vector dimension 40 115 40 153 115 Learning Rate 0.008 0.001 0.008 0.001 0.001 Batch size 427 169 427 133 169 Noise variance 0.04 0.08 0.04 0.02 0.08 Iterations 111 138 111 166 111

[0152] The parameters corresponding to the five offspring are input into the improved VAE-Attention model for training, and the corresponding improved VAE-Attention models are obtained. Then, the prediction results in step 1 are put into the trained prediction model, the corresponding mean square error is calculated (Table 4), and the optimal parameters are finally selected. The model training process is referred to Figure 4 , it can be seen that the model has finally converged.

[0153] Table 4. Errors in prediction results of reservoir parameters of the prediction model

[0154]

[0155]

[0156] From the analysis of Table 4, we can see that the optimal offspring corresponding to the BiGRU model is offspring 1, the optimal offspring corresponding to the LSTM is offspring 2, the optimal offspring corresponding to the BiLSTM model is offspring 1, and the optimal offspring corresponding to the RNN model is offspring 2.

[0157] 2.2.4 Model training and application

[0158] The optimal parameters are input into the model for training until the model is stable and the final model is obtained. Finally, the prediction results in step 1 are used as input and further processed by the final model to output the final operation results of the model.

[0159] Comparison of the effects of the model using the present invention:

[0160] The prediction results of the prediction model are compared and analyzed with the operating results of the improved VAE-Attention model.

[0161] Reference Figure 5 , draw a comparison image of the true value and the predicted value of the prediction model, and a comparison image of the result after using the improved VAE-Attention model and the true value. Figure 5 It can be found that for the BiGRU model with good prediction effect, the results of using the improved VAE-Attention model are significantly improved compared with the direct prediction results, making the predicted value closer to the actual value. At the same time, observing other models, the RNN model with the worst prediction effect has achieved a greater improvement, making it closer to the actual value.

[0162] Analysis of the precision improvement effect of the present invention:

[0163] Calculate the mean square error between the prediction results of the prediction model and the running results of the improved VAE-Attention model and the true value, and analyze the improvement effect of the running results of the improved VAE-Attention model on the prediction results.

[0164] Table 5 is obtained from the statistics of Table 2 and Table 4. It can be seen from Table 5 that the improved VAE-Attention model has a significant effect on improving accuracy, especially when the prediction effect is poor, the improvement is more obvious, which proves that it has good applicability. Finally, in order to prove the improvement effect of the experiment, the improvement rate of the improved VAE-Attention model in the prediction results is calculated respectively.

[0165] The formula for calculating the improvement rate is:

[0166]

[0167] Among them, the improvement effect on the RNN model is about 60.53%; the improvement effect on the LSTM model is about 44.96%; the improvement effect on the BiLSTM model is about 36.08%; and the improvement effect on the BiGRU model is about 17.28%.

[0168] Table 5 Comparison of reservoir parameter error results of different models

[0169]

[0170] The above describes the specific embodiments of the present invention in detail. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work.

[0171] The present invention also includes a system for improving the accuracy of predicting oil and gas reservoir physical property parameters, which is applied in the process of oil and gas field development to implement the above-mentioned method for improving the accuracy of predicting oil and gas reservoir physical property parameters, including a processing instruction module for executing each step in the method.

[0172] The present invention also includes a device for improving the accuracy of predicting oil and gas reservoir physical property parameters, including:

[0173] Memory: a computer program for storing the method for improving the prediction accuracy of oil and gas reservoir physical property parameters, which is a computer-readable device;

[0174] Processor: used to implement the method for improving the prediction accuracy of oil and gas reservoir physical property parameters when executing the computer program.

[0175] The present invention also includes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the method for improving the prediction accuracy of oil and gas reservoir physical property parameters can be implemented.

Claims

1. A method for improving the prediction accuracy of oil and gas reservoir physical property parameters, characterized in that: The following steps are involved: Step 1: Prediction model and operation: After the well logging curve data and measured reservoir parameter data are standardized, they are divided into training set and test set according to a predetermined ratio; then, the training set data is input into the prediction model for training until the model converges, and a trained prediction model is obtained; finally, the model is used to predict the test set data and the final prediction result is output; Step 2: Improved VAE-Attention model and operation: An improved VAE-Attention model is constructed, and the measured reservoir parameter data is standardized. The training set and the test set are also divided according to the proportion in step one. Gaussian noise is introduced into the training set data to enhance the robustness of the model. Then, the genetic algorithm is used to optimize the model hyperparameters to determine the optimal parameter combination. The optimal parameters are input into the model for training until the model is stable to obtain the final model. Finally, the prediction results in step one are used as input, further processed by the final model, and the final operation results of the model are output.

2. The method for improving the prediction accuracy of oil and gas reservoir physical property parameters according to claim 1, characterized in that: The detailed steps in step 1 are as follows: standardize the collected logging curve data and measured reservoir parameter data, select Z-Score standardization in the data standardization method, and the Z-Score standardization performs data standardization based on the mean and standard deviation of each feature. For each feature, the mean and standard deviation of all samples of the feature are calculated, and then the feature is standardized so that the mean of each feature is 0 and the standard deviation is 1, that is, the converted feature conforms to the standard normal distribution, and its expression is: Among them, x is the original data, μ is the mean of the data, and σ is the standard deviation of the data; The standardized data set is divided into a training set and a test set. Then, the training set data is input into four prediction models, namely BiGRU, LSTM, BiLSTM and RNN, for training. After each model is trained, its performance is evaluated using the test set data, and the final prediction results are saved. In order to quantify the prediction accuracy, the mean square error (MSE) is used as the error statistical indicator, and its expression is: Where n is the number of samples, yi is the true value, is the predicted value of the model.

3. The method for improving the prediction accuracy of oil and gas reservoir physical property parameters according to claim 1, characterized in that: The detailed steps of step 2 are as follows: (2.1) Construct a VAE-Attention model, wherein the VAE-Attention model includes an encoder and a decoder; (2.1.1), the encoder (q(z|x)) maps the reservoir parameter input data to the latent space and outputs the mean μ(x) and variance σ of the latent variable z 2 (x), the encoder adopts a BiGRU structure, where each GRU unit contains an update gate and a reset gate, and the corresponding mathematical expression is as follows: Reset Gate t : r t =σ(W r ·[h t-1 ,x t ]) Candidate hidden states Update gate z t : z t =σ(W z ·[h t-1 ,x t ]) Update hidden state h t : Among them, x t represents the reservoir permeability data at the tth time step of the input sequence, h t-1 represents the hidden state of the previous time step, W r , W z and W represent the weight parameters of the reset gate, update gate, and candidate hidden state, respectively, σ is the sigmoid function, and ⊙ represents element-by-element multiplication; After extracting the features of the permeability data using the BiGRU structure, it is distributed to the corresponding latent space through the linear layer, namely: q(z|x)=N(z;μ(x),σ 2 (x)); The mean μ(x) and variance σ of the output latent variable z are 2 (x), in order to enable the gradient to be passed through the random variable, the reparameterization technique is used to represent the encoder's encoded data as a function of the mean and variance, that is: z=μ(x)+σ(x)⊙∈ Among them, ∈~N(0,1), represents the noise of standard normal distribution, ⊙ represents element-by-element multiplication; (2.1.2) The decoder maps the low-dimensional potential representation back to the original data space, thereby realizing data reconstruction; the decoding layer also adopts the BiGRU structure, and reconstructs the re-parameterized latent variable z back to the original data through the BiGRU and Attention mechanism, specifically: First, map the reparameterized latent variable z to the initial hidden state h0: h0=f c (With) Then use GRU to generate data sequences: the t ,h t =GRU(o {t-1} ,h {t-1} ) Among them, t ,h t Respectively represent the output and hidden state at t time step; The current output of the decoder is o t and the hidden state h t Associated with the calculation of the attention weight, an attention layer is then used to convert the current decoder output o t and the previous hidden state h t-1 Connect it up: score(o t ,h t-1 )=attention(o t ,h t-1 ) Then normalize the attention score through the Softmax function to get the attention weight a t : Then use the attention weight a t The output of the encoder is weighted and summed to obtain the context vector c t : Among them, i is the output sequence of the encoder, a t，i is the attention weight; Finally, reconstruct the output using the context vector c t Combined with the decoder's current hidden state h t , through a fully connected layer f to generate the final reconstruction output The goal of the VAE model is to maximize the marginal likelihood p(x) of the data, that is: log p(x)=log(x|z)p(z)dz Using the variational inference method, we introduce an approximate posterior distribution q(z|x), and obtain the following lower bound through Jensen's inequality: Among them, D KL is the KL divergence, which measures the difference between two distributions; The loss function of the improved VAE-Attention model consists of reconstruction loss and KL divergence loss: Among them, the reconstruction loss is to measure the reconstruction data The difference between the original data x, usually using the mean square error (MSE); KL divergence loss measures the difference between the approximate posterior distribution of the encoder output and the prior distribution, which is usually a standard normal distribution; (2.2) Model training (2.2.1) Data standardization and division Standardize the collected measured reservoir parameter data, select Z-Score standardization in the data standardization method, and divide the standardized data into training data and test data with a ratio of 9:1; (2.2.2) Introducing noise The introduced noise is Gaussian noise, and its expression is: x noisy =x+∈, Among them, x is the reservoir permeability data, ∈ is subject to N(0, σ 2 ) random variables; (2.2.3) Hyperparameter Tuning Select the genetic algorithm for hyperparameter tuning, build the genetic algorithm framework, and tune the hyperparameters in the model. The tuning range is shown in the following table: (2.2.4) Model training and application The optimal parameters are input into the model for training until the model is stable and the final model is obtained; finally, the prediction results in step one are used as input, further processed by the final model, and the final operation results of the model are output.

4. A system for improving the accuracy of oil and gas reservoir physical property parameter prediction, characterized in that: The invention relates to a method for improving the accuracy of predicting gas reservoir physical property parameters according to any one of claims 1 to 3, which is applied in the process of oil and gas field development, and comprises a processing instruction module for executing each step in the method.

5. A device for improving the accuracy of oil and gas reservoir physical property parameter prediction, characterized in that: include: Memory: a computer program storing the method for improving the prediction accuracy of oil and gas reservoir physical property parameters according to any one of claims 1 to 3, which is a computer-readable device; Processor: used to implement the method for improving the prediction accuracy of oil and gas reservoir physical property parameters as described in any one of claims 1 to 3 when executing the computer program.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it can implement the method for improving the prediction accuracy of oil and gas reservoir physical property parameters as described in any one of claims 1-3.

Images