MMP Prediction Method and Device Based on Conditional Convolutional Generative Adversarial Network

Abstract

An embodiment of the present invention discloses an MMP prediction method and device based on a conditional convolutional generative adversarial network. The method includes: obtaining MMP influencing factor data of a target reservoir; inputting the MMP influencing factor data into a pre-trained convolutional generator to obtain an MMP prediction value of the target reservoir output by the pre-trained convolutional generator. The convolutional generator is built according to a convolutional neural network and does not include random noise input. The pre-trained convolutional generator is obtained by performing multiple iterative trainings on the convolutional generator according to a training sample set. Each training sample in the training sample set includes: the MMP value of the reservoir and the MMP influencing factor data of the reservoir. The present invention achieves the beneficial effect of accurately and efficiently predicting the MMP of the reservoir.

Description

Technical Field

[0001] The present invention relates to the technical field of reservoir development, and in particular, to an MMP prediction method and device based on a conditional convolutional generative adversarial network. Background Art

[0002] CO2 miscible flooding is the most widely used and highest oil recovery method in CO2-EOR of low-permeability reservoirs. During the process of injecting CO2 into the reservoir for oil displacement, there will be an interaction among gas, oil, and water phases in the rock formation, resulting in interphase component transfer, phase change, and other complex phase behaviors. The basic mechanism of miscible flooding is that the displacing agent (CO2 injection gas) and the displaced agent (crude oil) form a stable miscible zone front under reservoir conditions. This front is a single phase, and its movement can effectively push the crude oil forward and finally reach the production well. Due to miscibility, the oil-gas interface disappears, reducing the interfacial tension in the porous medium to zero. Therefore, theoretically, the microscopic displacement efficiency can reach 100%.

[0003] The minimum miscibility pressure (MMP) between CO2 and reservoir crude oil is one of the key parameters in the CO2 displacement process and is the boundary between CO2 miscible flooding and immiscible flooding. Accurately determining the minimum miscibility pressure between CO2 and crude oil is very important for improving the CO2 miscible displacement efficiency, reducing operating costs, and generating social and economic benefits.

[0004] The existing technology usually determines MMP by experimental measurement. Although this method can ensure accuracy, it is complex to operate, time-consuming, and costly. Therefore, the existing technology lacks a more efficient solution for determining the minimum miscibility pressure (MMP) between CO2 and reservoir crude oil. Summary of the Invention

[0005] In order to solve at least one of the above technical problems in the background art, the present invention proposes an MMP prediction method and device based on a conditional convolutional generative adversarial network.

[0006] To achieve the above object, according to one aspect of the present invention, there is provided an MMP prediction method based on a conditional convolutional generative adversarial network, the method comprising:

[0007] Obtaining MMP influencing factor data of a target reservoir;

[0008] Input the MMP influencing factor data into a pre-trained convolutional generator to obtain the MMP prediction value of the target reservoir output by the pre-trained convolutional generator. Among them, the convolutional generator is built based on a convolutional neural network and does not include random noise input. The pre-trained convolutional generator is obtained by iteratively training the convolutional generator multiple times according to a training sample set. Each training sample in the training sample set includes: the MMP value of the reservoir and the MMP influencing factor data of the reservoir.

[0009] Optionally, this MMP prediction method based on a conditional convolutional generative adversarial network further includes:

[0010] Obtain the training sample set;

[0011] Perform H1 times of iterative training according to the training sample set to obtain the pre-trained convolutional generator. Among them, each iterative training is divided into multiple batches of training. When performing each batch of training, first select H2 training samples from the training sample set, then train the network weights of the convolutional discriminator based on the selected training samples, and finally train the network weights of the convolutional generator in a combined model composed of the convolutional discriminator and the convolutional generator based on the selected training samples. The convolutional discriminator is built by combining a convolutional neural network and a fully connected neural network. Both H1 and H2 are positive integers.

[0012] Optionally, the training sample consists of first data and second data. The first data is the MMP influencing factor data of the reservoir, and the second data is the MMP value of the reservoir;

[0013] The training of the network weights of the convolutional discriminator based on the selected training samples specifically includes:

[0014] For each selected training sample, combine the MMP prediction value output by the convolutional generator according to the first data of the training sample with the first data of the training sample to obtain combined data, set the label of the combined data to 0, and smooth the label of the combined data;

[0015] Set the label of each selected training sample to 1, and smooth the label of the training sample;

[0016] Input the combined data with smoothed labels and the training samples with smoothed labels into the convolutional discriminator to train the network weights of the convolutional discriminator.

[0017] Optionally, the training of the network weights of the convolutional generator in a combined model composed of the convolutional discriminator and the convolutional generator based on the selected training samples specifically includes:

[0018] For each selected training sample, input the first data of the training sample into the convolutional generator to obtain the MMP prediction value corresponding to the training sample output by the convolutional generator;

[0019] For each selected training sample, combine the first data of the training sample with the MMP prediction value corresponding to the training sample to obtain combined data, set the label of the combined data to 1, and smooth the label of the combined data;

[0020] Input the combined data with the smoothed label into the convolutional discriminator to obtain the probability that the combined data output by the convolutional discriminator is real data.

[0021] Optionally, the step of performing H1 iterations of training according to the training sample set to obtain the pre-trained convolutional generator includes:

[0022] Use a hyperparameter optimization method to optimize the number of iteration training times H1, the number of training samples H2, the hyperparameters of the convolutional generator, and the hyperparameters of the convolutional discriminator to obtain the best parameter combination, and then perform iterative training according to the best parameter combination to obtain the pre-trained convolutional generator.

[0023] Optionally, the input of the convolutional discriminator is MMP influencing factor data and MMP values, where the MMP values include the MMP prediction values output by the convolutional generator, and the output of the convolutional discriminator is the probability that the data is real data; the network structure of the convolutional discriminator specifically includes: a convolutional neural network layer, a splicing layer, and a fully connected neural network layer. The convolutional neural network layer is used to preprocess the MMP influencing factor data, the splicing layer is used to splice the preprocessed data output by the convolutional neural network layer with the MMP values, and the fully connected neural network layer is used to process the spliced data output by the splicing layer to output the probability that the data is real data.

[0024] Optionally, the hyperparameters of the convolutional generator specifically include: the number of layers of the convolutional neural network layer, the number of convolutional kernels in each convolutional neural network layer, the size of the convolutional kernels, and the initial learning rate of the optimizer in the convolutional generator;

[0025] The hyperparameters of the convolutional discriminator specifically include: the number of layers of the convolutional neural network layer, the number of convolutional kernels in each convolutional neural network layer, the size of the convolutional kernels, the number of layers of the fully connected neural network layer, the number of neurons in each fully connected neural network layer, the dropout rate of each fully connected neural network layer, and the initial learning rate of the optimizer in the convolutional discriminator.

[0026] To achieve the above object, according to another aspect of the present invention, there is provided an MMP prediction device based on a conditional convolutional generative adversarial network, the device comprising:

[0027] A data acquisition unit, configured to acquire MMP influencing factor data of a target reservoir;

[0028] A prediction unit, configured to input the MMP influencing factor data into a pre-trained convolutional generator, to obtain an MMP prediction value of the target reservoir output by the pre-trained convolutional generator, wherein the convolutional generator is constructed according to a convolutional neural network, the convolutional generator does not include a random noise input, the pre-trained convolutional generator is obtained by performing multiple iterative trainings on the convolutional generator according to a training sample set, and each training sample in the training sample set includes: an MMP value of a reservoir and MMP influencing factor data of the reservoir.

[0029] To achieve the above object, according to another aspect of the present invention, there is also provided a computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the computer program, the steps of the above MMP prediction method based on a conditional convolutional generative adversarial network are implemented.

[0030] To achieve the above object, according to another aspect of the present invention, there is also provided a computer program product, comprising a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the above MMP prediction method based on a conditional convolutional generative adversarial network are implemented.

[0031] The beneficial effects of the present invention are as follows:

[0032] The present invention combines a conditional generative adversarial network with the prediction of the minimum miscibility pressure (MMP) between CO2 and reservoir crude oil, and constructs a generator of the conditional generative adversarial network based on a convolutional neural network, and then trains a convolutional generator of the conditional generative adversarial network as an MMP prediction model, achieving the beneficial effect of being able to accurately and efficiently predict the MMP of a reservoir. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In the drawings:

[0034] Figure 1 is the first flowchart of the MMP prediction method based on a conditional convolutional generative adversarial network in an embodiment of the present invention;

[0035] Figure 2 It is the second flowchart of the MMP prediction method based on the conditional convolutional generative adversarial network in the embodiments of the present invention;

[0036] Figure 3 It is the training flowchart of the convolutional discriminator in the embodiments of the present invention;

[0037] Figure 4 It is the training flowchart of the convolutional generator in the embodiments of the present invention;

[0038] Figure 5 It is the schematic diagram of the training sample set in the embodiments of the present invention;

[0039] Figure 6 It is the schematic diagram of the network structure of the convolutional generator in the embodiments of the present invention;

[0040] Figure 7 It is the schematic diagram of the network structure of the convolutional discriminator in the embodiments of the present invention;

[0041] Figure 8 It is the schematic diagram of the combined model in the embodiments of the present invention;

[0042] Figure 9 It is the relationship curve of MMP varying with temperature based on the conditional fully-connected generative adversarial network model;

[0043] Figure 10 It is the relationship curve of MMP varying with temperature based on the conditional convolutional generative adversarial network model;

[0044] Figure 11 It is the relationship between MMP and the mole fraction of N2 in CO2 based on the conditional convolutional generative adversarial network model;

[0045] Figure 12 It is the relationship between MMP and the mole fraction of H2S in CO2 based on the conditional convolutional generative adversarial network model;

[0046] Figure 13 It is the first structural block diagram of the MMP prediction device based on the conditional convolutional generative adversarial network in the embodiments of the present invention;

[0047] Figure 14 It is the second structural block diagram of the MMP prediction device based on the conditional convolutional generative adversarial network in the embodiments of the present invention;

[0048] Figure 15 It is the schematic diagram of the computer device in the embodiments of the present invention. Detailed implementation manners

[0049] To enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0050] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0051] It should be noted that the terms "including" and "having" in the description and claims of the present invention and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0052] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the drawings and in combination with the embodiments.

[0053] It should be noted that MMP in the present invention refers to the minimum miscibility pressure between CO2 and crude oil.

[0054] Generative Adversarial Networks (GANs) is a deep learning model and one of the most promising methods for unsupervised learning on complex distributions in recent years. The model generates quite good outputs through the mutual game learning of (at least) two modules in the framework: the generative model (also known as the generator) and the discriminative model (also known as the discriminator).

[0055] Generative adversarial network is an unsupervised machine learning method, which makes this method generally applied to data augmentation. That is, when there are few training samples provided for machine learning, generative adversarial network can be used to generate some data samples for the machine to learn. Since generative adversarial network emerged relatively late, it has only started to be used by a few oil workers for data augmentation in recent years. However, regarding the effects of the data generated by this method, different researchers have different opinions.

[0056] For the most primitive generative adversarial network, as long as a random vector is input, and then a generated object can be obtained, but we cannot control what kind of object is generated. Therefore, researchers proposed the theory of conditional generative adversarial network, adding constraints to the original GAN and introducing a conditional variable y (Conditional variable y) into the generative model and the discriminative model, introducing additional information for the model to generate data in a guiding way. In theory, y can be various meaningful information, such as class labels, which can turn this unsupervised learning method of GAN into a supervised one.

[0057] The emergence of conditional generative adversarial network has changed generative adversarial network from unsupervised learning to supervised learning, meaning that this method can be used for parameter prediction in the oil field and has great application prospects in the oil field. When using conditional generative adversarial network to predict MMP, although the MMP prediction model based on conditional fully connected generative adversarial network has a relatively high prediction accuracy, due to the existence of random noise input in its generator, when exploring the relationship between MMP and influencing factors, MMP shows different laws from physical experimental phenomena with the changes of factors such as N2 and H2S, which is not conducive to the practical application of the MMP prediction model.

[0058] Figure 9 In the conditional fully connected neural network, the random noise of the generative network causes MMP to show an unstable upward change with the increase of temperature. It can be seen that this does not conform to the actual physical law and is not conducive to the practical application of the MMP prediction model, and needs to be improved.

[0059] In order to solve the defects existing in the prediction of MMP by the existing technology through conditional fully connected neural network, the embodiment of the present invention provides a prediction scheme for the minimum miscibility pressure (MMP) of CO2 and crude oil based on an improved conditional convolutional generative adversarial network.

[0060] Figure 1 It is the first flowchart of the MMP prediction method based on conditional convolutional generative adversarial network in the embodiment of the present invention. As Figure 1 shown, in an embodiment of the present invention, the MMP prediction method based on conditional convolutional generative adversarial network of the present invention includes step S101 and step S102.

[0061] Step S101, obtain the MMP influencing factor data of the target reservoir.

[0062] In an embodiment of the present invention, the MMP influencing factor data specifically includes: reservoir temperature (T R ), mole fraction of volatile components in crude oil (X vol ), mole fraction of C2-C4 components in crude oil (X C2-4 ), mole fraction of C5-C6 components in crude oil (X C5-6 ), molecular weight of C 7+ components in crude oil (MW C7+ ), mole fractions of CO2 and four impurities in the injection gas (i.e., y CO2 , y C1 , y N2 , y H2S and y HC ), etc.

[0063] Step S102, input the MMP influencing factor data into the pre-trained convolutional generator to obtain the MMP prediction value of the target reservoir output by the pre-trained convolutional generator, where the convolutional generator is built based on a convolutional neural network, the convolutional generator does not include random noise input, the pre-trained convolutional generator is obtained by iteratively training the convolutional generator according to a training sample set, and each training sample in the training sample set includes: the MMP value of the reservoir and the MMP influencing factor data of the reservoir.

[0064] By improving the generator model, the present invention abandons the solution of using a fully connected neural network to build a generator in the prior art, and creatively uses a convolutional neural network to build a generator to form a convolutional generator. The input of the convolutional generator of the present invention does not include random noise input, making the MMP prediction more accurate, and when exploring the relationship between MMP and influencing factors, MMP will not show a law different from the physical experimental phenomenon with the change of factors such as N2 and H2S, which is helpful for the practical application of the MMP prediction model.

[0065] Figure 2 is the second flow chart of the MMP prediction method based on a conditional convolutional generative adversarial network in an embodiment of the present invention. As Figure 2 shown, the pre-trained convolutional generator in the above step S102 is specifically generated by step S201 and step S202.

[0066] Step S201, obtain the training sample set.

[0067] In one embodiment of the present invention, the present invention collects MMP values of a certain number of existing reservoirs and corresponding MMP influencing factor data, and divides them into a training sample set, a verification sample set, and a test sample set according to a certain ratio.

[0068] Figure 5 The MMP values of 105 reservoirs collected in one embodiment of the present invention and the corresponding MMP influencing factor data are shown. Specifically, the present invention divides all the data into a training sample set, a verification sample set, and a test sample set according to a ratio of 6:2:2. Then, the training sample set has 63 groups of data, the verification sample set has 21 groups of data, and the test sample set has 21 groups of data.

[0069] In one embodiment of the present invention, after obtaining the training sample set, the verification sample set, and the test sample set, the present invention first performs maximum-minimum normalization processing on the training sample set, and then uses the maximum and minimum values of the training sample set data to perform the same processing on the data in the verification sample set and the test sample set.

[0070] In one embodiment of the present invention, the maximum-minimum normalization formula can be as follows:

[0071]

[0072] Step S202: Perform H1 iterations of training according to the training sample set to obtain the pre-trained convolutional generator. Wherein, each iteration of training is divided into multiple batches of training. When performing each batch of training, first select H2 training samples from the training sample set, then train the network weights of the convolutional discriminator based on the selected training samples, and finally train the network weights of the convolutional generator in a combined model composed of the convolutional discriminator and the convolutional generator based on the selected training samples. The convolutional discriminator is built by combining a convolutional neural network and a fully connected neural network, and both H1 and H2 are positive integers.

[0073] In the present invention, each iteration of training is divided into multiple batches of training. When performing each batch of training, first select H2 training samples from the training sample set. The training samples selected for multiple batches of training in the same iteration of training are all different. If the number of remaining samples in the training sample set is less than H2 during a certain batch of training, then select all the remaining samples for this batch of training. After all the samples in the training sample set have been trained once, that is, one iteration training process of the conditional generative adversarial network is realized, which is also called a training cycle. The performance of the convolutional generator and the convolutional discriminator will gradually improve as the number of iteration training times (H1) increases.

[0074] In the present invention, the present invention performs H1 - time iterative training according to the above - mentioned process. When the iterative training reaches a certain number of times, the MMP prediction value generated by the convolutional generator under the corresponding conditions will be very close to the real data, thus realizing the prediction function of MMP. In an embodiment of the present invention, during the iterative training process, the present invention simultaneously monitors the prediction error of the convolutional generator on the validation set after each iterative training is completed, and saves each convolutional generator after each iterative training separately. Finally, the convolutional generator with the smallest error on the validation set during the iterative training process is selected, and this convolutional generator is the MMP prediction model.

[0075] In the present invention, the present invention first trains the network weights of the convolutional discriminator based on the training samples, and then trains the network weights of the convolutional generator in a combined model composed of the convolutional discriminator and the convolutional generator based on the training samples. Figure 8 is a schematic diagram of the combined model in an embodiment of the present invention, as Figure 8 shown, when training the network weights of the convolutional generator in the combined model, the network weights of the convolutional discriminator do not change, while the network weights of the convolutional generator will change with the training of the data.

[0076] In an embodiment of the present invention, the training samples are composed of first data and second data. The first data is the MMP influencing factor data of the reservoir, and the second data is the MMP value of the reservoir.

[0077] Figure 3 is a training flow chart of the convolutional discriminator in an embodiment of the present invention, as Figure 3 shown, in an embodiment of the present invention, the training of the network weights of the convolutional discriminator based on the selected training samples in step S202 specifically includes steps S301 to S303.

[0078] Step S301, for each selected training sample, combine the MMP prediction value output by the convolutional generator according to the first data of the training sample with the first data of the training sample to obtain combined data, set the label of the combined data to 0, and smooth the label of the combined data.

[0079] Step S302, set the label of each selected training sample to 1, and smooth the label of the training sample.

[0080] In the present invention, there are many ways to smooth the labels. In a specific embodiment of the present invention, the label smoothing method of the present invention is: replace the label 1 with a random number within a first preset numerical range (preferably 0.8 to 1.0), and replace the label 0 with a random number within a second preset numerical range (preferably 0 to 0.2).

[0081] Step S303: Input the combined data after label smoothing and the training samples after label smoothing into the convolutional discriminator to train the network weights of the convolutional discriminator.

[0082] In an embodiment of the present invention, when starting an iterative training, first select the first batch of training samples from the training sample set. The number of training samples in this batch is H2. Using the established convolutional generator, use the first data in these H2 training samples of this batch as the conditional input of the convolutional generator, and output the MMP prediction value corresponding to the condition generated by the convolutional generator for the first time. Then, combine the first data with the MMP prediction value output by the convolutional generator to obtain combined data. Set the label of the combined data to 0, then smooth the label, and input it into the convolutional discriminator. At the same time, combine the first data and the corresponding real MMP value, that is, the training data, set the label to 1, then smooth the label, and also send it into the convolutional discriminator, so that the convolutional discriminator conducts the first learning, that is, the first learning of true and false data. Furthermore, according to the above process, perform H1 iterations of training on the convolutional discriminator in multiple batches. The convolutional discriminator trained through iteration can more accurately identify real data.

[0083] Figure 4 is the training flow chart of the convolutional generator in the embodiment of the present invention. As Figure 4 shown, in an embodiment of the present invention, training the network weights of the convolutional generator in the combined model composed of the convolutional discriminator and the convolutional generator in step S202 specifically includes steps S401 to S403.

[0084] Step S401: For each selected training sample, input the first data of the training sample into the convolutional generator to obtain the MMP prediction value corresponding to the training sample output by the convolutional generator.

[0085] Step S402: For each selected training sample, combine the first data of the training sample with the MMP prediction value corresponding to the training sample to obtain combined data. Set the label of the combined data to 1, and smooth the label of the combined data.

[0086] Step S403: Input the combined data after label smoothing into the convolutional discriminator to obtain the probability that the convolutional discriminator outputs that the combined data is real data.

[0087] In the present invention, after training the convolutional discriminator, the weights of the convolutional discriminator are kept unchanged, and then the network weights of the convolutional generator are trained in the combined model composed of the convolutional discriminator and the convolutional generator. Specifically, the present invention combines the conditional input of the convolutional generator (i.e., the first data in the training data) with the MMP prediction value output by the convolutional generator, sets the label to 1, then performs smoothing processing on the label, and inputs it into the combined model to realize the independent training and learning of the convolutional generator without affecting the convolutional discriminator, thereby improving the accurate ability of the convolutional generator to generate data.

[0088] Figure 6 is the schematic diagram of the network structure of the convolutional generator in the embodiment of the present invention. As Figure 6 shown, the present invention uses a convolutional neural network (CNN) to build the convolutional generator in the conditional convolutional generative adversarial network, and removes the random noise input in the convolutional generator to obtain an improved convolutional generator model. The convolutional generator only accepts one input, that is, the MMP influencing factor data, and the output of the convolutional generator is the MMP prediction value.

[0089] As Figure 6 shown, in an embodiment of the present invention, the convolutional generator is specifically composed of multiple convolutional neural network layers.

[0090] In an embodiment of the present invention, before inputting the MMP influencing factor data into the convolutional neural network layer of the convolutional generator, it is also necessary to convert the one-dimensional MMP influencing factor data into a two-dimensional matrix form that can be received by the convolutional neural network layer.

[0091] In an embodiment of the present invention, the hyperparameters of the convolutional generator specifically include: the number of layers A1 of the convolutional neural network layer, the number of convolutional kernels B1 in each convolutional neural network layer, the size C1 of the convolutional kernel, and the initial learning rate G1 of the optimizer in the convolutional generator.

[0092] In an embodiment of the present invention, the initial learning rate of the optimizer in the convolutional generator is specifically the initial learning rate of the Adam optimizer in the convolutional generator.

[0093] Figure 7 is the schematic diagram of the network structure of the convolutional discriminator in the embodiment of the present invention. As Figure 7As shown, the input of the convolutional discriminator is MMP influencing factor data and MMP values, and the MMP values include the MMP prediction values output by the convolutional generator. The output of the convolutional discriminator is the probability that the data is real data. The network structure of the convolutional discriminator specifically includes: a convolutional neural network layer, a splicing layer, and a fully connected neural network layer. The convolutional neural network layer is used to preprocess the MMP influencing factor data, the splicing layer is used to splice the preprocessed data output by the convolutional neural network layer with the MMP values, and the fully connected neural network layer is used to process the spliced data output by the splicing layer to output the probability that the data is real data.

[0094] As Figure 7 shown, the convolutional discriminator receives two inputs. The first input is condition X, that is, the normalized MMP influencing factor data, for preprocessing. The second is the true MMP value Y corresponding to condition X or the MMP prediction value Y' generated by the convolutional generator under condition X.

[0095] As Figure 7 shown, in an embodiment of the present invention, when setting the convolutional discriminator, first set the convolutional neural network layer to preprocess condition X, that is, the normalized MMP influencing factor data. Then, splice the data preprocessed by the convolutional neural network layer with the input MMP value (the true MMP value Y corresponding to condition X or the MMP prediction value Y' generated by the convolutional generator under condition X). Finally, set the fully connected neural network layer again to process the spliced data. And the number of neurons in the last layer of this fully connected neural network layer is 1, the activation function is sigmoid, and the probability that the currently input data is real data is output. If the output probability > 0.5, it belongs to real data, otherwise, it is fake data, that is, the data generated by the convolutional generator.

[0096] In an embodiment of the present invention, the hyperparameters of the convolutional discriminator specifically include: the number of layers A2 of the convolutional neural network layer, the number of convolutional kernels B2 in each convolutional neural network layer, the size C2 of the convolutional kernels, the number of layers D2 of the fully connected neural network layer, the number of neurons E2 in each fully connected neural network layer, the dropout rate F2 in each fully connected neural network layer, and the initial learning rate G2 of the optimizer in the convolutional discriminator.

[0097] In an embodiment of the present invention, the initial learning rate of the optimizer in the convolutional discriminator is specifically the initial learning rate of the Adam optimizer in the convolutional generator.

[0098] In an embodiment of the present invention, when performing iterative training in the above step S202, the present invention also uses a hyperparameter optimization method to optimize the number of iterative training times H1, the number of training samples H2, the hyperparameters of the convolutional generator, and the hyperparameters of the convolutional discriminator, so as to obtain an optimal parameter combination. Then, iterative training is performed according to the optimal parameter combination to obtain the pre-trained convolutional generator, that is, the MMP prediction model.

[0099] In a specific embodiment of the present invention, the present invention uses the Bayesian hyperparameter optimization method to optimize the hyperparameters of the convolutional generator and the convolutional discriminator, the number of training samples (H2) in each batch during each iterative training process, and the number of iterative training times (H1), and searches for the parameter combination that makes the model have the best prediction effect in the validation set, and uses the parameter combination with the best prediction effect in the validation set as the optimal parameter combination.

[0100] Among them, during Bayesian optimization, several parameter combinations will be randomly used for trial calculation first, and the number of trial calculations can be set artificially. Here, the number of trial calculations is set to 10. Then, real Bayesian optimization is performed. After each Bayesian optimization after the trial calculation, the results of the previous calculations, that is, the performance of the model on the validation set, will be referred to, so as to select the hyperparameters and the number of iterative training times that should be used in the next calculation. Here, the number of steps for performing real Bayesian optimization is set to 40.

[0101] In an embodiment of the present invention, the present invention uses the optimal parameter combination obtained by Bayesian hyperparameter optimization to establish an MMP prediction model (that is, a convolutional generator). This model is the model that we will use to predict the MMP of a new oil reservoir. By inputting the MMP influencing factor data of the new oil reservoir, the corresponding MMP value can be predicted.

[0102] In an embodiment of the present invention, the optimal parameter combination obtained by using Bayesian hyperparameter optimization can be as follows:

[0103] In the convolutional generator network, the number of convolutional neural network layers A1 is set to 1 layer, the number of convolutional kernels B1 in each convolutional neural network layer is set to 26, the size C1 of the convolutional kernels is set to 2, and the activation function is relu. The initial learning rate G1 of the optimizer in the convolutional generator is set to 0.0002190.

[0104] In the convolutional discriminator network, the number of convolutional neural network layers (A2) was set to 2, the number of convolution kernels (B2) in each convolutional neural network layer was set to 88, the kernel size (C2) was set to 4, and the activation function was ReLU. The number of fully connected neural network layers (D2) was set to 3, the number of neurons in each of the first two layers was set to 91, the dropout rate for each of the first two layers was 0.2021, the number of neurons in the last layer was set to 1, the activation function for the first two layers was ReLU, and the activation function for the last layer was sigmoid. The initial learning rate G2 of the optimizer in the convolutional discriminator was set to 0.0009755.

[0105] The number of training iterations (H1) was set to 482 after Bayesian optimization, and the number of training samples in each batch during each training iteration (H2) was set to 45 after Bayesian optimization.

[0106] Furthermore, based on the same training data, the present invention established MMP prediction models using three machine learning methods: a fully connected neural network, a support vector machine, and a conditional fully connected neural network. The structure of each model was optimized using a Bayesian algorithm and a validation dataset. Finally, the prediction accuracy of each optimized model was evaluated using the same untrained test data.

[0107] Table 1 Average absolute percentage errors of various MMP prediction models in the test sample set

[0108]

[0109] As can be seen from Table 1, compared with the prediction results of MMP models established using machine learning methods such as fully connected neural networks, support vector machines, and conditional fully connected neural networks, the MMP prediction model based on the improved conditional convolutional generative adversarial network of the present invention achieved improvements of 3, 10, and 4 percentage points in the test set, respectively, achieving the highest accuracy among the four machine learning methods. This demonstrates the powerful fitting capability of the improved conditional convolutional generative adversarial network of the present invention, which has higher prediction accuracy than fully connected neural networks, support vector machines, and conditional fully connected neural networks, and possesses strong generalization capabilities.

[0110] Figure 10 、 Figure 11 and Figure 12The curve shows the variation of MMP obtained from the MMP prediction model established using the improved conditional convolutional generative adversarial network with respect to temperature, the molar fraction of N2 in CO2, and the molar fraction of H2S in CO2. It can be seen that the predicted MMP increases with the increase in temperature and the molar fraction of N2 in CO2, and decreases with the increase in the molar fraction of H2S in CO2. The displayed pattern conforms to the actual physical law, demonstrating the reliability and effectiveness of the model of the present invention, and it can be used for MMP prediction and analysis of influencing factors.

[0111] As can be seen from the above embodiments, the MMP prediction method based on conditional convolutional generative adversarial network of the present invention has at least achieved the following beneficial effects:

[0112] 1. The present invention combines the conditional convolutional generative adversarial network in machine learning methods with reservoir MMP prediction, which is a new MMP prediction idea and method, opening up a precedent for the application of conditional convolutional generative adversarial network in MMP prediction, and is of great significance for reservoir MMP prediction and reservoir development plan design;

[0113] 2. The present invention adaptively improves the conditional convolutional generative adversarial network according to the MMP prediction scenario. First, the generator is improved by deleting the random noise input of the generator. Then, the labels of the generated real and fake data are smoothed to jointly improve the model prediction accuracy, thus obtaining the improved conditional convolutional generative adversarial network. Experimental results show that the improved conditional convolutional generative adversarial network not only improves the prediction accuracy of MMP, but also can accurately reflect the variation relationship of MMP with various influencing factors, and has stronger applicability.

[0114] Generally speaking, the model of the present invention is simple in establishment process, high in calculation efficiency, high in prediction accuracy, comprehensive and applicable, and has broad application prospects.

[0115] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0116] Based on the same inventive concept, an embodiment of the present invention further provides an MMP prediction device based on a conditional convolutional generative adversarial network, which can be used to implement the MMP prediction method based on a conditional convolutional generative adversarial network described in the above embodiments, as described in the following embodiments. Since the principle of the MMP prediction device based on a conditional convolutional generative adversarial network to solve problems is similar to that of the MMP prediction method based on a conditional convolutional generative adversarial network, the embodiments of the MMP prediction device based on a conditional convolutional generative adversarial network can refer to the embodiments of the MMP prediction method based on a conditional convolutional generative adversarial network, and the repeated parts will not be elaborated. Hereinafter, the term "unit" or "module" can be a combination of software and / or hardware that can implement a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0117] Figure 13 is the first structural block diagram of the MMP prediction device based on a conditional convolutional generative adversarial network according to an embodiment of the present invention, as Figure 13 shown. In an embodiment of the present invention, the MMP prediction device based on a conditional convolutional generative adversarial network of the present invention includes:

[0118] A data acquisition unit 1, configured to acquire MMP influencing factor data of a target reservoir;

[0119] A prediction unit 2, configured to input the MMP influencing factor data into a pre-trained convolutional generator, and obtain an MMP prediction value of the target reservoir output by the pre-trained convolutional generator, wherein the convolutional generator is built according to a convolutional neural network, the convolutional generator does not include a random noise input, and the pre-trained convolutional generator is obtained by performing multiple iterative trainings on the convolutional generator according to a training sample set, and each training sample in the training sample set includes: the MMP value of the reservoir and the MMP influencing factor data of the reservoir.

[0120] Figure 14 is the second structural block diagram of the MMP prediction device based on a conditional convolutional generative adversarial network according to an embodiment of the present invention, as Figure 14 shown. In an embodiment of the present invention, the MMP prediction device based on a conditional convolutional generative adversarial network of the present invention further includes:

[0121] A training sample set acquisition unit 3, configured to acquire the training sample set;

[0122] A model training unit 4, configured to perform H1 iterations of training according to the training sample set to obtain the pre-trained convolutional generator, where each iteration of training is divided into multiple batches of training. When performing each batch of training, first select H2 training samples from the training sample set, then train the network weights of the convolutional discriminator based on the selected training samples, and finally train the network weights of the convolutional generator in a combined model composed of the convolutional discriminator and the convolutional generator based on the selected training samples. The convolutional discriminator is constructed by combining a convolutional neural network and a fully connected neural network, and both H1 and H2 are positive integers.

[0123] In an embodiment of the present invention, the training sample is composed of first data and second data. The first data is the MMP influencing factor data of the reservoir, and the second data is the MMP value of the reservoir. In an embodiment of the present invention, the model training unit specifically includes:

[0124] A first label setting module, configured to, for each selected training sample, combine the MMP prediction value output by the convolutional generator according to the first data of the training sample with the first data of the training sample to obtain combined data, set the label of the combined data to 0, and smooth the label of the combined data;

[0125] A second label setting module, configured to set the label of each selected training sample to 1 and smooth the label of the training sample;

[0126] A convolutional discriminator training module, configured to input the combined data with smoothed labels and the training samples with smoothed labels into the convolutional discriminator to train the network weights of the convolutional discriminator.

[0127] In an embodiment of the present invention, the model training unit specifically includes:

[0128] A predicted value obtaining module, configured to, for each selected training sample, input the first data of the training sample into the convolutional generator to obtain the MMP predicted value corresponding to the training sample output by the convolutional generator;

[0129] A combined data obtaining module, configured to, for each selected training sample, combine the first data of the training sample with the MMP predicted value corresponding to the training sample to obtain combined data, set the label of the combined data to 1, and smooth the label of the combined data;

[0130] A convolutional generator training module, configured to input the combined data with smoothed labels into the convolutional discriminator to obtain the probability that the convolutional discriminator outputs that the combined data is real data.

[0131] Optionally, the model training unit further includes:

[0132] A hyperparameter optimization module for optimizing the number of iterative training times H1, the number of training samples H2, the hyperparameters of the convolutional generator, and the hyperparameters of the convolutional discriminator by using a hyperparameter optimization method to obtain an optimal parameter combination.

[0133] To achieve the above object, according to another aspect of the present application, a computer device is further provided. As Figure 15 shown, the computer device includes a memory, a processor, a communication interface, and a communication bus. A computer program that can run on the processor is stored in the memory. When the processor executes the computer program, the steps in the method of the above embodiment are implemented.

[0134] The processor may be a central processing unit (CPU). The processor may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. chips, or combinations of the above types of chips.

[0135] The memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and units, such as the corresponding program units in the method embodiment of the present invention above. The processor executes various functional applications and work data processing of the processor by running the non-transitory software programs, instructions, and modules stored in the memory, that is, the method in the method embodiment above is implemented.

[0136] The memory may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created by the processor, etc. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory may optionally include a memory remotely set relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above network include but are not limited to the Internet, an enterprise internal network, a local area network, a mobile communication network, and combinations thereof.

[0137] The one or more units are stored in the memory and, when executed by the processor, perform the methods in the above embodiments.

[0138] Specific details of the above computer device can be understood by referring to the corresponding relevant descriptions and effects in the above embodiments, and will not be elaborated here.

[0139] To achieve the above object, according to another aspect of the present application, there is also provided a computer-readable storage medium storing a computer program, which when executed in a computer processor, implements the steps in the above MMP prediction method based on conditional convolutional generative adversarial network. Those skilled in the art can understand that to implement all or part of the processes in the above method embodiments, it can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the above method embodiments. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above types of memories.

[0140] To achieve the above object, according to another aspect of the present application, there is also provided a computer program product including a computer program / instructions, which when executed by a processor, implements the steps of the above MMP prediction method based on conditional convolutional generative adversarial network.

[0141] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device. They can be concentrated on a single computing device or distributed on a network composed of multiple computing devices. Optionally, they can be implemented by program codes executable by the computing device, so that they can be stored in a storage device and executed by the computing device, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module to implement. In this way, the present invention is not limited to any specific combination of hardware and software.

[0142] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A MMP prediction method based on a conditional convolutional generative adversarial network, characterized in that Including: Obtain the MMP influencing factor data of the target reservoir, where the MMP influencing factor data includes: reservoir temperature T R , mole fraction X of volatile components in crude oil vol , mole fraction X of C2-C4 components in crude oil C2-4 , mole fraction X of C5-C6 components in crude oil C5-6 , C in crude oil 7+ component molecular weight MW C7+ , mole fraction y of CO2 in the injection gas CO2 and mole fractions y of four impurities in the injection gas C1 , y N2 , y H2S and y HC ; Input the MMP influencing factor data into a pre-trained convolutional generator to obtain the MMP prediction value of the target reservoir output by the pre-trained convolutional generator. The convolutional generator is built based on a convolutional neural network and does not include random noise input. The pre-trained convolutional generator is obtained by performing multiple iterative trainings on the convolutional generator according to a training sample set. During training, first train the network weights of the convolutional discriminator based on the training sample set, and then train the network weights of the convolutional generator in a combined model composed of the convolutional discriminator and the convolutional generator based on the training sample set. The convolutional discriminator is built by combining a convolutional neural network and a fully connected neural network. Each training sample in the training sample set includes the MMP value of the reservoir and the MMP influencing factor data of the reservoir.

2. The MMP prediction method based on conditional convolutional generative adversarial network according to claim 1, wherein Also including: Obtain the training sample set; Perform H1 times of iterative training according to the training sample set to obtain the pre-trained convolutional generator. Each iterative training is divided into multiple batches of training. When performing each batch of training, first select H2 training samples from the training sample set, then train the network weights of the convolutional discriminator based on the selected training samples, and finally train the network weights of the convolutional generator in a combined model composed of the convolutional discriminator and the convolutional generator based on the selected training samples. Both H1 and H2 are positive integers.

3. The MMP prediction method based on conditional convolutional generative adversarial network according to claim 2, wherein The training sample is composed of first data and second data. The first data is the MMP influencing factor data of the reservoir, and the second data is the MMP value of the reservoir; The training of the network weights of the convolutional discriminator based on the selected training samples specifically includes: For each selected training sample, combine the MMP prediction value output by the convolutional generator according to the first data of the training sample with the first data of the training sample to obtain combined data, set the label of the combined data to 0, and smooth the label of the combined data; Set the label of each selected training sample to 1, and smooth the label of the training sample; Input the combined data with smoothed label and the training sample with smoothed label into the convolutional discriminator to train the network weights of the convolutional discriminator.

4. The MMP prediction method based on conditional convolutional generative adversarial network according to claim 3, characterized in that, The training of the network weights of the convolutional generator in a combined model composed of the convolutional discriminator and the convolutional generator based on the selected training samples specifically includes: For each selected training sample, input the first data of the training sample into the convolutional generator to obtain the MMP prediction value corresponding to the training sample output by the convolutional generator; For each selected training sample, combine the first data of the training sample with the MMP prediction value corresponding to the training sample to obtain combined data, set the label of the combined data to 1, and smooth the label of the combined data; The combined data after label smoothing is input into the convolutional discriminator to obtain the probability that the combined data output by the convolutional discriminator is real data.

5. The MMP prediction method based on conditional convolutional generative adversarial network according to claim 2, wherein, The pre-trained convolutional generator obtained by performing H1 iterations of training according to the training sample set includes: Using a hyperparameter optimization method to optimize the number of iteration training times H1, the number of training samples H2, the hyperparameters of the convolutional generator, and the hyperparameters of the convolutional discriminator to obtain an optimal parameter combination, and then performing iterative training according to the optimal parameter combination to obtain the pre-trained convolutional generator.

6. The MMP prediction method based on conditional convolutional generative adversarial network according to claim 2, characterized in that The input of the convolutional discriminator is MMP influencing factor data and MMP values, and the MMP values include the MMP prediction values output by the convolutional generator. The output of the convolutional discriminator is the probability that the data is real data. The network structure of the convolutional discriminator specifically includes: a convolutional neural network layer, a splicing layer, and a fully connected neural network layer. The convolutional neural network layer is used to preprocess the MMP influencing factor data, the splicing layer is used to splice the preprocessed data output by the convolutional neural network layer with the MMP values, and the fully connected neural network layer is used to process the spliced data output by the splicing layer to output the probability that the data is real data.

7. The MMP prediction method based on conditional convolutional generative adversarial network according to claim 5, characterized in that The hyperparameters of the convolutional generator specifically include: the number of layers of the convolutional neural network layer, the number of convolutional kernels in each convolutional neural network layer, the size of the convolutional kernels, and the initial learning rate of the optimizer in the convolutional generator. The hyperparameters of the convolutional discriminator specifically include: the number of layers of the convolutional neural network layer, the number of convolutional kernels in each convolutional neural network layer, the size of the convolutional kernels, the number of layers of the fully connected neural network layer, the number of neurons in each fully connected neural network layer, the dropout rate of each fully connected neural network layer, and the initial learning rate of the optimizer in the convolutional discriminator.

8. An MMP prediction device based on a conditional convolutional generative adversarial network, characterized in that, It includes: A data acquisition unit for acquiring MMP influencing factor data of a target reservoir, where the MMP influencing factor data includes: reservoir temperature T R , mole fraction X of volatile components in crude oil vol , mole fraction X of C2-C4 components in crude oil C2-4 , mole fraction X of C5-C6 components in crude oil C5-6 , C in crude oil 7+ component molecular weight MW C7+ , mole fraction y of CO2 in the injection gas CO2 and mole fractions y of four impurities in the injection gas C1 , y N2 , y H2S and y HC ; A prediction unit for inputting the MMP influencing factor data into the pre-trained convolutional generator to obtain the MMP prediction value of the target reservoir output by the pre-trained convolutional generator. Among them, the convolutional generator is built according to a convolutional neural network and does not include random noise input. The pre-trained convolutional generator is obtained by performing multiple iterations of training on the convolutional generator according to the training sample set. During training, first train the network weights of the convolutional discriminator based on the training sample set, and then train the network weights of the convolutional generator in a combined model composed of the convolutional discriminator and the convolutional generator based on the training sample set. The convolutional discriminator is built by combining a convolutional neural network and a fully connected neural network. Each training sample in the training sample set includes: the MMP value of the reservoir and the MMP influencing factor data of the reservoir.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, it implements the steps of the method according to any one of claims 1 to 7.

Images