CN114399119A - MMP prediction method and device based on conditional convolution generative adversarial network - Google Patents

Abstract

The embodiment of the invention discloses a method and a device for predicting MMP (matrix metalloproteinase) of a countermeasure network based on a conditional convolution generation mode, wherein the method comprises the following steps: acquiring MMP influence factor data of a target oil reservoir; inputting the MMP influencing factor data into a pre-trained convolution generator to obtain an MMP predicted value of the target oil reservoir output by the pre-trained convolution generator, wherein the convolution generator is built according to a convolution neural network, the convolution generator does not contain random noise input, the pre-trained convolution generator is obtained by carrying out multiple iterative training on the convolution generator according to a training sample set, and each training sample in the training sample set comprises: MMP values of the reservoir and MMP influential factor data of the reservoir. The method has the beneficial effect of accurately and efficiently predicting the MMP of the oil reservoir.

Description

MMP prediction method and device based on conditional convolution generative adversarial network

technical field

The invention relates to the technical field of oil reservoir development, in particular, to an MMP prediction method and device based on a conditional convolution generative confrontation network.

Background technique

CO ₂ miscible flooding is the most widely used and highest recovery method in CO ₂ -EOR in low permeability reservoirs. In the process of injecting _CO2 into the oil reservoir, the three-phase interaction of gas, oil and water will occur in the rock formation. Produce interphase composition transfer, phase transitions, and other complex phase behaviors. The basic mechanism of miscible flooding is that the displacing agent (CO ₂ injection gas) and the displaced agent (crude oil) form a stable front of miscible zone under reservoir conditions. The front is a single phase, and its movement can effectively push the crude oil toward flow and eventually reach the production well. Due to the miscibility, the oil-gas interface disappears and the interfacial tension in the porous medium is reduced to zero, so the microscopic displacement efficiency can theoretically reach 100%.

The minimum miscibility pressure (MMP) between CO ₂ and reservoir crude oil is one of the key parameters in the process of CO ₂ flooding, and it is the boundary between CO ₂ miscible flooding and immiscible flooding. Accurately determining the minimum miscible pressure of CO ₂ and crude oil is very important to improve the CO ₂ miscible displacement efficiency, reduce operating costs, and generate social and economic benefits.

In the prior art, an experimental measurement method is usually used to determine the MMP. Although this method can ensure the accuracy of measurement, this method is complicated to operate, takes a long time and costs a lot. Therefore, the prior art lacks a more efficient solution for determining the minimum miscibility pressure (MMP) between CO ₂ and the crude oil in the reservoir.

SUMMARY OF THE INVENTION

In order to solve at least one technical problem in the above background technology, the present invention proposes an MMP prediction method and device based on conditional convolution generative adversarial network.

In order to achieve the above object, according to one aspect of the present invention, there is provided an MMP prediction method based on conditional convolution generative adversarial network, the method comprising:

Obtain the MMP influencing factor data of the target reservoir;

Inputting the MMP influencing factor data into a pre-trained convolution generator to obtain the MMP predicted value of the target oil reservoir output by the pre-trained convolution generator, wherein the convolution generator is based on The convolutional neural network is constructed, the convolutional generator does not contain random noise input, and the pre-trained convolutional generator is obtained by performing multiple iterative training on the convolutional generator according to the training sample set. Each training sample of contains: the MMP value of the reservoir and the MMP influencing factor data of the reservoir.

Optionally, the MMP prediction method based on conditional convolution generative adversarial network further includes:

obtain the training sample set;

Perform H1 iterations of training according to the training sample set to obtain the pre-trained convolution generator, wherein each iteration training is divided into multiple batches of training. 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 use the selected training samples in the convolutional discriminator and the convolutional generator. In the combined model, the network weight of the convolution generator is trained, the convolution discriminator is constructed according to the combination of a convolutional neural network and a fully connected neural network, and both H1 and H2 are positive integers.

Optionally, the training sample consists of first data and second data, the first data is MMP influencing factor data of the oil reservoir, and the second data is the MMP value of the oil reservoir;

The network weights of the convolution discriminator are trained based on the selected training samples, specifically including:

For each selected training sample, combine the MMP predicted value output by the convolution generator according to the first data of the training sample and the first data of the training sample to obtain combined data, and set the label of the combined data to 0 , and smooth the labels of the combined data;

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

The combined data after label smoothing and the training samples after label smoothing are input into the convolution discriminator, and the network weight of the convolution discriminator is trained.

Optionally, the network weights of the convolution generator are trained in the combined model composed of the convolution discriminator and the convolution generator based on the selected training samples, specifically including:

For each selected training sample, input the first data of the training sample into the convolution generator, and obtain the MMP prediction value corresponding to the training sample output by the convolution 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 the combined data, set the label of the combined data to 1, and perform an analysis on the label of the combined data. smooth processing;

The combined data after label smoothing is input into the convolution discriminator, and the probability that the combined data output by the convolution discriminator is real data is obtained.

Optionally, performing H1 iterations of training according to the training sample set to obtain the pre-trained convolution generator, including:

The hyperparameter optimization method is used to optimize the iterative training times H1, the number of training samples H2, the hyperparameters of the convolution generator and the hyperparameters of the convolution discriminator to obtain the optimal parameter combination, and then according to the optimal parameter combination The parameter combination is iteratively trained to obtain the pre-trained convolution generator.

Optionally, the input of the convolution discriminator is MMP influencing factor data and an MMP value, the MMP value includes the MMP predicted value output by the convolution generator, and the output of the convolution discriminator is that the data is real data. The network structure of the convolution 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, and the splicing layer For splicing the preprocessed data output by the convolutional neural network layer with the MMP value, and the fully connected neural network layer is used to process the spliced data output by the splicing layer. The output data is real data The probability.

Optionally, the hyperparameters of the convolution generator specifically include: the number of layers of the convolutional neural network layer, the number of convolution kernels of each convolutional neural network layer, the size of the convolution kernel, and the optimization in the convolutional generator. the initial learning rate of the machine;

The hyperparameters of the convolution discriminator specifically include: the number of layers of the convolutional neural network layer, the number of convolution kernels of each convolutional neural network layer, the size of the convolution kernel, the number of layers of the fully connected neural network layer, the number of layers of each The number of neurons in the 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.

In order to achieve the above object, according to another aspect of the present invention, there is provided an MMP prediction device based on conditional convolution generative adversarial network, the device comprising:

The data acquisition unit is used to acquire the MMP influencing factor data of the target oil reservoir;

A prediction unit, configured to input the MMP influencing factor data into a pre-trained convolution generator to obtain the MMP predicted value of the target oil reservoir output by the pre-trained convolution generator, wherein the volume The product generator is constructed according to the convolutional neural network, the convolutional generator does not contain random noise input, and the pre-trained convolutional generator is obtained by performing multiple iterative training on the convolutional generator according to the training sample set, Each training sample in the training sample set includes: the MMP value of the oil reservoir and the MMP influencing factor data of the oil reservoir.

In order to achieve the above object, according to another aspect of the present invention, a computer device is also provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the computer Procedure to implement the above-mentioned MMP prediction method based on conditional convolutional generative adversarial network.

In order to achieve the above object, according to another aspect of the present invention, a computer program product is also provided, comprising a computer program/instruction, when the computer program/instruction is executed by a processor, the above-mentioned MMP based on conditional convolution generative adversarial network is implemented The steps of the prediction method.

The beneficial effects of the present invention are:

The invention combines the conditional generative adversarial network with the minimum miscibility pressure (MMP) prediction between CO ₂ and oil in the reservoir, and builds the conditional generative adversarial network generator based on the convolutional neural network, and then trains the conditional generative confrontation As the MMP prediction model, the convolutional generator of the network achieves the beneficial effect of accurately and efficiently predicting the MMP of the reservoir.

Description of drawings

In order to illustrate the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are For some embodiments of the present invention, for those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts. In the attached image:

Fig. 1 is the first flow chart of the MMP prediction method based on conditional convolution generative adversarial network according to an embodiment of the present invention;

2 is a second flowchart of an MMP prediction method based on a conditional convolution generative adversarial network according to an embodiment of the present invention;

Fig. 3 is the training flow chart of the convolution discriminator of the embodiment of the present invention;

Fig. 4 is the training flow chart of the convolution generator of the embodiment of the present invention;

5 is a schematic diagram of a training sample set according to an embodiment of the present invention;

6 is a schematic diagram of a network structure of a convolution generator according to an embodiment of the present invention;

7 is a schematic diagram of a network structure of a convolution discriminator according to an embodiment of the present invention;

8 is a schematic diagram of a combination model according to an embodiment of the present invention;

Fig. 9 is the relation curve that the MMP of the conditional fully connected generative adversarial network model changes with the change of temperature;

Fig. 10 is the relationship curve of the MMP based on the conditional convolution generative adversarial network model as a function of temperature;

Figure 11 is the relationship between the MMP and the mole fraction of _N2 in _CO2 based on the conditional convolutional generative adversarial network model;

Figure 12 is the relationship between the MMP and the mole fraction of H ₂ S in CO ₂ based on the conditional convolution generative adversarial network model;

13 is a first structural block diagram of an MMP prediction apparatus based on a conditional convolution generative adversarial network according to an embodiment of the present invention;

14 is a second structural block diagram of an MMP prediction apparatus based on a conditional convolution generative adversarial network according to an embodiment of the present invention;

FIG. 15 is a schematic diagram of a computer device according to an embodiment of the present invention.

Detailed ways

In order to make those skilled in the art better understand the solutions 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 Embodiments are part of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

It should be noted that the terms "comprising" and "having" in the description and claims of the present invention and the above-mentioned drawings, as well as any variations thereof, are intended to cover non-exclusive inclusion, for example, including a series of steps or units The processes, methods, systems, products or devices are not necessarily limited to those steps or units expressly listed, but may include other steps or units not expressly listed or inherent to such processes, methods, products or devices.

It should be noted that the embodiments of the present invention and the features of the embodiments may be combined with each other under the condition of no conflict. The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

It should be noted that MMP in the present invention refers to the minimum miscibility pressure between CO ₂ and crude oil.

Generative Adversarial Networks (GANs), a deep learning model, is one of the most promising approaches for unsupervised learning on complex distributions in recent years. The model produces reasonably good outputs through mutual game learning of (at least) two modules in the framework: a generative model (aka generator) and a discriminative model (aka discriminator).

Generative adversarial network is an unsupervised machine learning method, which makes this method generally applied to data enhancement, that is, when there are few training samples provided to machine learning, generative adversarial network can be used to generate some data samples, for machine learning. Due to the late birth of generative adversarial network, it has only been used by a few oil workers for data enhancement in recent years. However, the effect of the data generated by this method is different from various researchers.

The most primitive generative adversarial network only needs to input a random vector and get a generated object, but we have no control over what kind of object is generated. Therefore, the researchers proposed the conditional generative adversarial network theory, which added constraints to the original GAN, introduced the conditional variable y (Conditional variable y) in the generative model and the discriminant model, introduced additional information to the model, and guided the generation of data. Theoretically, y can make meaningful various information, such as class labels, which can turn the unsupervised learning method of GAN into supervised.

The emergence of conditional generative adversarial networks has changed generative adversarial networks from unsupervised learning to supervised learning, which means that this method can be used in oil parameter prediction and has great application prospects in oil. When using conditional generative adversarial network to predict MMP, although the MMP prediction model based on conditional fully connected generative adversarial network has high prediction accuracy, due to the existence of random noise input in its generator, it is difficult to explore the relationship between MMP and its influence. When the relationship between the factors is changed, the MMP will show different laws from the physical experimental phenomena with the changes of N ₂ , H ₂ S and other factors, which is not conducive to the practical application of the MMP prediction model.

Figure 9 shows that in the conditional fully-connected neural network, the MMP caused by the random noise of the generating network presents an unstable upward change with the increase of temperature. It can be seen that this is inconsistent with the actual physical laws and is not conducive to the practical application of the MMP prediction model. need to be improved.

In order to solve the deficiencies in the prior art in predicting MMP through conditional fully-connected neural networks, embodiments of the present invention provide a CO ₂ and crude oil minimum miscibility pressure (MMP) prediction scheme based on an improved conditional convolutional generative adversarial network.

FIG. 1 is a first flowchart of an MMP prediction method based on a conditional convolution generative adversarial network according to an embodiment of the present invention. As shown in FIG. 1 , in an embodiment of the present invention, the conditional convolution generative adversarial network of the present invention The MMP prediction method includes step S101 and step S102.

Step S101, acquiring MMP influencing factor data of the target oil reservoir.

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 C ₂ -C ₄ components in crude oil (X _C2 - ₄ ), the mole fraction of C ₅ -C ₆ components in crude oil (X _C5-6 ), the molecular weight of C ₇₊ components in crude oil (MW _C7+ ), the mole fraction of CO ₂ and four impurities in the injected gas (ie y _{CO 2} , y _C1 , y _N2 , y _H2S and y _HC ) and the like.

Step S102: Input the MMP influencing factor data into a pre-trained convolution generator to obtain the MMP predicted value of the target oil reservoir output by the pre-trained convolution generator, wherein the convolution generates The convolution generator is built according to the convolutional neural network, the convolution generator does not contain random noise input, and the pre-trained convolution generator is obtained by performing multiple iterative training on the convolution generator according to the training sample set. Each training sample in the training sample set contains: the MMP value of the reservoir and the data of the MMP influencing factors of the reservoir.

By improving the generator model, the present invention abandons the prior art scheme of using a fully connected neural network to build a generator, and creatively uses a convolutional neural network to build a generator to form a convolution generator. The input of the convolution generator of the present invention does not contain random noise input, so that MMP prediction is more accurate, and when exploring the relationship between MMP and influencing factors, MMP will not change with N ₂ , H ₂ S and other factors However, it shows different laws from the physical experimental phenomena, which is helpful for the practical application of the MMP prediction model.

FIG. 2 is a second flowchart of the MMP prediction method based on the conditional convolution generative adversarial network according to the embodiment of the present invention. As shown in FIG. 2 , the pre-trained convolution generator in the above step S102 is specifically composed of steps It is generated by training in S201 and step S202.

Step S201, acquiring the training sample set.

In an embodiment of the present invention, the present invention collects a certain number of MMP values of existing oil 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 proportion.

Fig. 5 shows the MMP values of 105 oil reservoirs and the corresponding MMP influencing factor data collected in an embodiment of the present invention. Specifically, the present invention divides all data into training sample set, verification sample set and test sample set according to the ratio of 6:2:2, then the training sample set has 63 sets of data, the verification sample set has 21 sets of data, and the test sample set has 21 sets of data. There are 21 sets of data.

In an 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 and minimum normalization processing on the training sample set, and then uses the maximum and minimum values of the training sample set data. The values in the validation sample set and the data in the test sample set are treated the same.

In an embodiment of the present invention, the maximum-minimum normalization formula may be as follows:

Step S202, performing H1 iterations of training according to the training sample set to obtain the pre-trained convolution generator, wherein each iteration training is divided into multiple batches of training, and each batch of training is performed , first select H2 training samples from the training sample set, then train the network weights of the convolution discriminator based on the selected training samples, and finally generate the convolution discriminator and convolution based on the selected training samples. The network weight of the convolution generator is trained in the combined model composed of the convolution generator, the convolution discriminator is constructed according to the combination of the convolution neural network and the fully connected neural network, and both H1 and H2 are positive integers.

In the present invention, each iterative training is divided into multiple batches of training, and when performing each batch of training, first select H2 training samples from the training sample set, and for multiple batches of the same iterative training The training samples selected for training are all different. If the number of remaining samples in the training sample set during training of a certain batch is less than H2, all the remaining samples are selected for training in this batch. Therefore, after all samples in the training sample set are trained once, an iterative training process of the conditional generative adversarial network is realized, also known as a training cycle. The performance of the convolution generator and the convolution discriminator will increase with the iteration. The training times (H1) increase gradually.

In the present invention, the present invention performs H1 iteration training according to the above process. When the iterative training reaches a certain number of times, the predicted value of MMP under the corresponding conditions generated by the convolution generator will be very close to the real data, which realizes the prediction function of MMP. In an embodiment of the present invention, the present invention simultaneously monitors the prediction error of the convolution generator on the validation set after each iteration training is completed during the iterative training process, and saves the convolution generator after each iteration training separately. , and finally select the convolution generator with the smallest error on the validation set during the iterative training process, which is the MMP prediction model.

In the present invention, the present invention firstly trains the network weights of the convolution discriminator based on the training samples, and then, based on the training samples, sets the network weights of the convolution generator in the combined model composed of the convolution discriminator and the convolution generator based on the training samples. to train. FIG. 8 is a schematic diagram of a combined model according to an embodiment of the present invention. As shown in FIG. 8 , when the network weight of the convolution generator is trained in the combined model, the network weight of the convolution discriminator does not change, and the network weight of the convolution generator does not change. The weights will change as the data is trained.

In an embodiment of the present invention, the training sample consists of first data and second data, the first data is MMP influencing factor data of the oil reservoir, and the second data is the MMP value of the oil reservoir.

FIG. 3 is a training flow chart of a convolution discriminator according to an embodiment of the present invention. As shown in FIG. 3 , in an embodiment of the present invention, the network weights of the convolution discriminator are trained based on the selected training samples in the above step S202 , which specifically includes steps S301 to S303.

Step S301, for each selected training sample, combine the MMP predicted value output by the convolution generator according to the first data of the training sample and the first data of the training sample to obtain combined data, and the label of the combined data is obtained. Set to 0 and smooth the labels for this combined data.

In step S302, the label of each selected training sample is set to 1, and the label of the training sample is smoothed.

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

Step S303 , the combined data after label smoothing and the training samples after label smoothing are input into the convolution discriminator, and the network weight of the convolution discriminator is trained.

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, and use the established convolution generator , the first data in the batch of H2 training samples is the conditional input of the convolution generator, and the MMP prediction value under the corresponding conditions generated by the convolution generator for the first time is output; Combine the MMP predicted values output by the product generator to obtain the combined data, set the label of the combined data to 0, then smooth the label and input it into the convolution discriminator; at the same time, combine the first data and the corresponding real MMP value The combination of , that is, the training data, set the label to 1, then smooth the label, and also send it to the convolution discriminator, so that the convolution discriminator performs the first learning, that is, the first learning of true and false data . Then, according to the above process, the convolution discriminator is trained in multiple batches and then H1 iterations, and the convolution discriminator obtained after iterative training can more accurately identify the real data.

FIG. 4 is a training flow chart of a convolution generator according to an embodiment of the present invention. As shown in FIG. 4 , in an embodiment of the present invention, the training samples based on the selected training samples in the above step S202 are processed by the convolution discriminator and the volume The network weight of the convolution generator is trained in the combined model composed of the product generator, which specifically includes steps S401 to S403.

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

Step S402, for each selected training sample, combine the first data of the training sample and the MMP predicted value corresponding to the training sample to obtain combined data, set the label of the combined data to 1, and set the combined data to 1. labels are smoothed.

Step S403: Input the combined data after label smoothing into the convolution discriminator to obtain the probability that the combined data output by the convolution discriminator is real data.

In the present invention, after the convolution discriminator is trained, the weight of the convolution discriminator is kept unchanged, and then the network weight of the convolution generator is set in the combined model composed of the convolution discriminator and the convolution generator. to train. Specifically, the present invention combines the conditional input of the convolution generator (that is, the first data in the training data) with the MMP predicted value output by the convolution generator, sets the label to 1, then smoothes the label, and inputs it into the combination In the model, the separate training and learning of the convolution generator is realized without affecting the convolution discriminator, which improves the accuracy of the data generated by the convolution generator.

FIG. 6 is a schematic diagram of the network structure of a convolution generator according to an embodiment of the present invention. As shown in FIG. 6 , the present invention uses a convolutional neural network (CNN) to build a convolution generator in a conditional convolution generative adversarial network, and convolution The random noise input in the product generator is removed to obtain an improved convolutional generator model. The convolution generator accepts only one input, the MMP influencer data, and the output of the convolution generator is the MMP prediction value.

As shown in FIG. 6 , in an embodiment of the present invention, the convolution generator is specifically composed of multi-layer convolutional neural network layers.

In an embodiment of the present invention, before inputting the MMP influencing factor data into the convolutional neural network layer of the convolution generator, the present invention also needs to convert the one-dimensional MMP influencing factor data into a convolutional neural network layer that can receive 2D matrix form.

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

In an embodiment of the present invention, the initial learning rate of the optimizer in the convolution generator is specifically the initial learning rate of the Adam optimizer in the convolution generator.

FIG. 7 is a schematic diagram of a network structure of a convolution discriminator according to an embodiment of the present invention. As shown in FIG. 7 , the input of the convolution discriminator is MMP influencing factor data and an MMP value, and the MMP value includes the output of the convolution generator. The MMP predicted value of , the output of the convolution discriminator is the probability that the data is real data; the network structure of the convolution discriminator specifically includes: a convolutional neural network layer, a splicing layer and a fully connected neural network layer, the convolutional The neural network layer is used for preprocessing the MMP influencing factor data, the splicing layer is used for splicing the preprocessed data output by the convolutional neural network layer and the MMP value, and the fully connected neural network layer is used for splicing. The spliced data output by the splicing layer is processed and the probability that the output data is real data.

As shown in Figure 7, the convolution discriminator accepts two inputs, the first input is the condition X, that is, the normalized MMP influencing factor data, which is preprocessed, and the second is the real MMP value corresponding to the condition X Y or the MMP predicted value Y' generated by the convolution generator under condition X.

As shown in FIG. 7 , in one embodiment of the present invention, when the convolution discriminator is set, the convolutional neural network layer is first set to preprocess the condition X, that is, the normalized MMP influencing factor data; then, Concatenate the data preprocessed by the convolutional neural network layer with the input MMP value (the real MMP value Y corresponding to the condition X or the MMP predicted value Y' generated by the convolution generator under the condition X); finally, again Set the fully connected neural network layer 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 output current input data is real data. Probability, if the output probability is > 0.5, it belongs to real data, otherwise, it is fake data, that is, the data generated by the convolution generator.

In an embodiment of the present invention, the hyperparameters of the convolutional discriminator specifically include: the number of layers of the convolutional neural network layer A2, the number of convolution kernels B2 of each convolutional neural network layer, the size of the convolutional kernel C2, the full The number of layers D2 connected to the neural network layer, the number of neurons E2 of each fully connected neural network layer, the drop rate F2 of each fully connected neural network layer, and the initial learning rate G2 of the optimizer in the convolutional discriminator.

In an embodiment of the present invention, the initial learning rate of the optimizer in the convolution discriminator is specifically the initial learning rate of the Adam optimizer in the convolution generator.

In an embodiment of the present invention, when the iterative training is performed in the above step S202, the present invention also adopts a hyperparameter optimization method to adjust the iterative training times H1, the number of training samples H2, the hyperparameters of the convolution generator, and the hyperparameters of the convolution discriminator. The parameters are optimized to obtain the best parameter combination. Then, iterative training is performed according to the optimal parameter combination to obtain the pre-trained convolution generator, that is, the MMP prediction model.

In a specific embodiment of the present invention, the present invention utilizes the Bayesian hyperparameter optimization method to optimize the hyperparameters of the convolution generator, the hyperparameters of the convolution discriminator, and the training in each batch in each iteration training process. The number of samples (H2) and iterative training times (H1) are optimized to find the parameter combination that makes the model predict the best in the validation set, and the parameter combination with the best prediction effect in the validation set is used as the best parameter combination.

Among them, during Bayesian optimization, several sets of parameter combinations will be randomly used to perform trial calculations. The number of trial calculations can be set manually. Here, the number of trial calculations is set to 10. Then, the real Bayesian optimization is performed. Each Bayesian optimization after the trial calculation will refer to the results of the previous calculation, that is, the performance of the model on the validation set, so as to select the hypervisor that should be used in the next calculation. Parameters and the number of iterative training, here the number of steps to actually perform Bayesian optimization is set to 40.

In one embodiment of the present invention, the present invention uses the optimal parameter combination obtained through Bayesian hyperparameter optimization to establish an MMP prediction model (ie, a convolution generator). This model is the model we will use to predict the MMP of the new reservoir. Input the MMP influencing factor data of the new reservoir, and then the corresponding MMP value can be predicted.

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

In the convolution generator network, the number of layers A1 of the convolutional neural network layer is set to 1 layer, the number of convolution kernels B1 of each convolutional neural network layer is set to 26, and the size of the convolution kernel C1 is set to 2. The activation functions are all relu. The initial learning rate G1 of the optimizer in the convolutional generator is set to 0.0002190.

In the convolutional discriminator network, the number of layers A2 of the convolutional neural network layer is set to 2, the number of convolutional kernels B2 of each convolutional neural network layer is set to 88, and the size of the convolutional kernel C2 is set to 4. The activation functions are all relu; the number of layers D2 of the fully connected neural network layer is set to 3 layers, the number of neurons in each layer of the first 2 layers is set to 91, the drop rate of each layer of the first 2 layers of the network is 0.2021, and the number of neurons in the last layer Set to 1, the activation functions of the first two layers are relu, and the activation function of the last layer is sigmoid. The initial learning rate G2 of the optimizer in the convolutional discriminator is set to 0.0009755.

The number of iterative training (H1) is set to 482 after Bayesian optimization, and the number of training samples (H2) in each batch in each iteration training process is set to 45 after Bayesian optimization.

Further, according to the same training set data, the present invention uses three machine learning methods, such as fully connected neural network, support vector machine, and conditional fully connected neural network, to establish an MMP prediction model, and combines Bayesian algorithm and verification data set. The structure of each model has been optimized. Finally, the prediction accuracy of each optimized model is evaluated using the same untrained test set data.

Table 1 The mean absolute percentage error of each MMP prediction model in the test sample set

As can be seen from Table 1, compared with the prediction results of MMP models established by machine learning methods such as fully connected neural networks, support vector machines, and conditional fully connected neural networks, the present invention is based on the improved conditional convolution generative adversarial network. The error of the MMP prediction model in the test set is increased by 3, 10, and 4 percentage points, respectively, which is the highest accuracy among the four machine learning methods. This reflects the powerful fitting ability of the improved conditional convolution generative adversarial network of the present invention, which has higher prediction accuracy and stronger generalization ability than the fully connected neural network, the support vector machine and the conditional fully connected neural network.

Figure 10, Figure 11 and Figure 12, the MMP prediction model obtained by using the improved conditional convolutional generative adversarial network with temperature, the mole fraction of N ₂ in CO ₂ and the mole fraction of H ₂ S in CO ₂ changes with the change of the curve. It can be seen that the predicted MMP increases with temperature as well as the mole fraction of _N2 in _CO2 , and decreases with the increase of the mole fraction of _H2S in _CO2 , which shows that the law is consistent with the actual physical law, It is proved that the model of the invention is reliable and effective, and can be used for MMP prediction and influencing factor analysis.

It can be seen from the above embodiments that the MMP prediction method based on the conditional convolution generative adversarial network of the present invention at least achieves the following beneficial effects:

1. The present invention combines the conditional convolution generative adversarial network in the machine learning method with the oil reservoir MMP prediction, which is a new MMP prediction idea and method, and opens up a precedent for the conditional convolution generative adversarial network in MMP prediction. , which is of great significance to reservoir MMP prediction and reservoir development plan design;

2. The present invention adaptively improves the conditional convolution generative adversarial network according to the prediction scene of MMP. First, the generator is improved, and the random noise input of the generator is deleted. The labels are smoothed to jointly improve the prediction accuracy of the model, resulting in an improved conditional convolutional generative adversarial network. The experimental results show that the improved conditional convolutional generative adversarial network not only improves the prediction accuracy of MMP, but also It can accurately reflect the changing relationship of MMP with various influencing factors, and is more applicable.

On the whole, the model of the invention has the advantages of simple process of establishing the model, high calculation efficiency, high prediction accuracy, strong comprehensiveness and applicability, and has a wide application prospect.

It should be noted that the steps shown in the flowcharts of the accompanying drawings may be executed in a computer system, such as a set of computer-executable instructions, and, although a logical sequence is shown in the flowcharts, in some cases, Steps shown or described may be performed in an order different from that herein.

Based on the same inventive concept, an embodiment of the present invention also provides an MMP prediction device based on a conditional convolution generative adversarial network, which can be used to implement the MMP prediction method based on a conditional convolution generative adversarial network described in the above embodiments, as described in the examples below. Since the principle of solving the problem of the MMP prediction device based on the conditional convolution generative adversarial network is similar to the MMP prediction method based on the conditional convolution generative adversarial network, the embodiment of the MMP prediction device based on the conditional convolution generative adversarial network can refer to The embodiment of the MMP prediction method based on the conditional convolution generative adversarial network will not be repeated here. As used below, the term "unit" or "module" may be a combination of software and/or hardware that implements a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, implementations in hardware, or a combination of software and hardware, are also possible and contemplated.

FIG. 13 is a first structural block diagram of an MMP prediction apparatus based on a conditional convolution generative adversarial network according to an embodiment of the present invention. As shown in FIG. 13 , in an embodiment of the present invention, the conditional convolution generative adversarial network-based The MMP predictor includes:

The data acquisition unit 1 is used to acquire the MMP influencing factor data of the target oil reservoir;

A prediction unit 2, configured to input the MMP influencing factor data into a pre-trained convolution generator to obtain the MMP predicted value of the target oil reservoir output by the pre-trained convolution generator, Among them, the convolution generator is constructed according to the convolutional neural network, the convolution generator does not contain random noise input, and the pre-trained convolution generator is to perform multiple iterative training on the convolution generator according to the training sample set It is obtained that each training sample in the training sample set includes: the MMP value of the oil reservoir and the data of the MMP influencing factors of the oil reservoir.

FIG. 14 is a second structural block diagram of the MMP prediction apparatus based on the conditional convolution generative adversarial network according to an embodiment of the present invention. As shown in FIG. 14 , in an embodiment of the present invention, the conditional convolution generative adversarial network-based The MMP predictor also includes:

A training sample set obtaining unit 3, for obtaining the training sample set;

A model training unit 4, configured to perform H1 iteration training according to the training sample set, to obtain the pre-trained convolution generator, wherein each iteration training is divided into multiple batches of training, and each iteration is performed During batch training, first select H2 training samples from the training sample set, then train the network weights of the convolution discriminator based on the selected training samples, and finally use the selected training samples in the convolution discriminator. In the combined model composed of the convolution generator and the convolution generator, the network weights of the convolution generator are trained, and the convolution discriminator is constructed according to the combination of the convolutional neural network and the fully connected neural network, and both H1 and H2 are positive. Integer.

In an embodiment of the present invention, the training sample consists of first data and second data, the first data is MMP influencing factor data of the oil reservoir, and the second data is the MMP value of the oil reservoir. In an embodiment of the present invention, the model training unit specifically includes:

The first label setting module is used to combine the MMP predicted value output by the convolution generator according to the first data of the training sample and the first data of the training sample for each selected training sample, to obtain the combined data, and the The label of the combined data is set to 0, and the label of the combined data is smoothed;

The second label setting module is used to set the label of each selected training sample to 1, and smooth the label of the training sample;

The convolution discriminator training module is used to input the combined data after label smoothing and the training samples after label smoothing into the convolution discriminator, and train the network weight of the convolution discriminator.

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

The predicted value acquisition module is used to input the first data of the training sample into the convolution generator for each selected training sample, and obtain the MMP predicted value corresponding to the training sample output by the convolution generator ;

The combined data acquisition module is used to combine the first data of the training sample and the MMP predicted value corresponding to the training sample for each selected training sample to obtain combined data, set the label of the combined data to 1, and Smooth the labels of the combined data;

The convolution generator training module is used to input the combined data after label smoothing processing into the convolution discriminator, and obtain the probability that the combined data output by the convolution discriminator is real data.

Optionally, the model training unit further includes:

The hyperparameter optimization module is used to optimize the iterative training times H1, the number of training samples H2, the hyperparameters of the convolution generator and the hyperparameters of the convolution discriminator using a hyperparameter optimization method to obtain the best parameter combination .

In order to achieve the above object, according to another aspect of the present application, a computer device is also provided. As shown in FIG. 15 , the computer device includes a memory, a processor, a communication interface and a communication bus, and a computer program that can be run on the processor is stored in the memory, and the processor implements the above embodiments when executing the computer program steps in the method.

The processor may be a central processing unit (Central Processing Unit, CPU). The processor may also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuit (Application Specific Integrated Circuit, ASIC), Field-Programmable Gate Array (Field-Programmable Gate Array, FPGA) or other Chips such as programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above types of chips.

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

The memory may include a storage program area and a storage data area, wherein the storage program area may store an operating system and an application program required by at least one function; the storage data area may store data created by the processor, and the like. Additionally, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory may optionally include memory located remotely from the processor, such remote memory being connectable to the processor via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

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

The specific details of the above computer equipment can be understood by referring to the corresponding related descriptions and effects in the above embodiments, which will not be repeated here.

In order to achieve the above object, according to another aspect of the present application, a computer-readable storage medium is also provided, where the computer-readable storage medium stores a computer program, and when the computer program is executed in a computer processor, the above-mentioned based on Steps in the MMP prediction method of conditional convolutional generative adversarial networks. Those skilled in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. During execution, the processes of the embodiments of the above-mentioned methods may be included. Wherein, the storage medium may be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a flash memory (Flash Memory), a hard disk (Hard Disk Drive, Abbreviation: HDD) or solid-state drive (Solid-State Drive, SSD), etc.; the storage medium may also include a combination of the above-mentioned types of memories.

In order to achieve the above object, according to another aspect of the present application, a computer program product is also provided, comprising a computer program/instruction, when the computer program/instruction is executed by a processor, the above-mentioned MMP based on conditional convolution generative adversarial network is implemented The steps of the prediction method.

Obviously, those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computing device, which can be centralized on a single computing device, or distributed in a network composed of multiple computing devices Alternatively, they can be implemented with program codes executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or they can be integrated into The multiple modules or steps are fabricated into a single integrated circuit module. As such, the present invention is not limited to any particular combination of hardware and software.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims (10)

1. An MMP prediction method for a countermeasure network based on a conditional convolution generation mode is characterized by comprising the following steps:

acquiring MMP influence factor data of a target oil reservoir;

inputting the MMP influencing factor data into a pre-trained convolution generator to obtain an MMP predicted value of the target oil reservoir output by the pre-trained convolution generator, wherein the convolution generator is built according to a convolution neural network, the convolution generator does not contain random noise input, the pre-trained convolution generator is obtained by carrying out multiple iterative training on the convolution generator according to a training sample set, and each training sample in the training sample set comprises: MMP values of the reservoir and MMP influential factor data of the reservoir.

2. The MMP prediction method of the conditional convolution-based generative countermeasure network of claim 1, further comprising:

acquiring the training sample set;

performing H1 times of iterative training according to the training sample set to obtain the pre-trained convolution generator, wherein each time of iterative training is divided into multiple batches of training, when each batch of training is performed, firstly selecting H2 training samples from the training sample set, then training the network weight of a convolution discriminator based on the selected training samples, and finally training the network weight of the convolution generator in a combined model composed of the convolution discriminator and the convolution generator based on the selected training samples, wherein the convolution discriminator is built according to the combination of a convolution neural network and a fully-connected neural network, and H1 and H2 are both positive integers.

3. The MMP prediction method based on conditional convolution generation countermeasure network of claim 2, wherein the training sample is composed of a first data and a second data, the first data is MMP influence factor data of the oil reservoir, the second data is MMP value of the oil reservoir;

the training of the network weight of the convolution discriminator based on the selected training sample specifically comprises:

respectively combining the MMP predicted value output by the convolution generator according to the first data of the training sample with the first data of the training sample aiming at each selected training sample to obtain combined data, setting the label of the combined data to be 0, and smoothing the label of the combined data;

setting the label of each selected training sample as 1, and smoothing the label of the training sample;

and inputting the combined data subjected to the label smoothing processing and the training sample subjected to the label smoothing processing into the convolution discriminator, and training the network weight of the convolution discriminator.

4. The MMP prediction method based on conditional convolution generation countermeasure network of claim 3, wherein the training of the network weights of the convolution generator in the combined model composed of the convolution discriminator and convolution generator based on the selected training samples specifically includes:

respectively inputting first data of the training samples into the convolution generator aiming at each selected training sample to obtain an MMP predicted value which is output by the convolution generator and corresponds to the training sample;

respectively combining the first data of the training samples with the MMP predicted values corresponding to the training samples aiming at each selected training sample to obtain combined data, setting the label of the combined data to be 1, and smoothing the label of the combined data;

and inputting the combined data subjected to the label smoothing into the convolution discriminator to obtain the probability that the combined data output by the convolution discriminator is real data.

5. The MMP prediction method based on conditional convolution generation countermeasure network of claim 2, wherein said H1 times of iterative training according to said training sample set to obtain said pre-trained convolution generator includes:

and optimizing the iterative training times H1, the training sample number H2, the hyper-parameters of the convolution generator and the hyper-parameters of the convolution discriminator by adopting a hyper-parameter optimization method to obtain an optimal parameter combination, and then performing iterative training according to the optimal parameter combination to obtain the pre-trained convolution generator.

6. The MMP prediction method based on conditional convolution generation countermeasure network of claim 2, wherein the input of the convolution discriminator is MMP influence factor data and MMP values including MMP prediction values output from the convolution generator, the output of the convolution discriminator is a probability that data is real data; the network structure of the convolution discriminator specifically includes: convolutional neural network layer, concatenation layer and full-connection neural network layer, convolutional neural network layer is used for carrying out the preliminary treatment to MMP influence factor data, the concatenation layer is used for with data after the preliminary treatment of convolutional neural network layer output and MMP value splice, full-connection neural network layer is used for right data after the concatenation of concatenation layer output handles the probability that output data is true data.

7. The MMP prediction method of the conditional convolution-based generative countermeasure network of claim 5, wherein the hyper-parameters of the convolution generator specifically include: the number of layers of the convolutional neural network layer, the number of convolutional kernels of each layer of the convolutional neural network layer, the size of the convolutional kernels and the initial learning rate of an optimizer in the convolutional generator;

the hyper-parameters of the convolution discriminator specifically include: the number of layers of the convolutional neural network layer, the number of convolutional kernels of 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 of each fully-connected neural network layer, the discarding rate of each fully-connected neural network layer, and the initial learning rate of an optimizer in a convolutional discriminator.

8. An MMP prediction apparatus for a countermeasure network based on conditional convolution generation, comprising:

the data acquisition unit is used for acquiring MMP (matrix metalloproteinases) influence factor data of the target oil reservoir;

the prediction unit is configured to input the MMP influence factor data into a pre-trained convolution generator to obtain an MMP prediction value of the target oil reservoir output by the pre-trained convolution generator, where the convolution generator is built according to a convolutional neural network, the convolution generator does not include random noise input, the pre-trained convolution generator is obtained by performing multiple iterative training on the convolution generator according to a training sample set, and each training sample in the training sample set includes: MMP values of the reservoir and MMP influential factor data of the reservoir.

9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the steps of the method according to any of claims 1 to 7 are implemented when the computer program is executed by the processor.

10. A computer program product comprising computer program/instructions, characterized in that the computer program/instructions, when executed by a processor, implement the steps of the method of any one of claims 1 to 7.

Images