A method and device for predicting residual oil distribution based on physical constraint neural network
Through the method based on physical constraint neural network, the characteristics of reservoir physical properties and water injection development scheme are extracted, and the residual oil distribution timing prediction model is established, which solves the problems of low computational efficiency and general accuracy of numerical simulation methods, and achieves fast and accurate residual oil Distribution prediction provides dynamic evolution law analysis support for oil field development.
Patent Information
- Application Number
- CN202411165234.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-23
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2044-08-23
AI Technical Summary
The existing numerical simulation methods have low computational efficiency when predicting the distribution of residual oil in oil fields, which is difficult to adapt to the needs of production history fitting and development adjustment. Moreover, the calculation accuracy is average, and a fast and accurate prediction method is urgently needed.
Using the residual oil distribution prediction method based on physical constraint neural network, water injection development plan data is input into the neural network, and features are extracted using physical conditional convolution calculation layer, image convolution calculation layer and data fusion layer to establish water injection Develop a time series prediction model for residual oil distribution to achieve rapid prediction.
It significantly reduces the calculation time, can describe the residual oil change process in detail, provides effective support for the analysis of the dynamic evolution law of reservoirs with water injection development, and improves prediction accuracy.
Smart Images

Figure CN118690922B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of artificial intelligence prediction and oilfield development, and in particular to a method and device for predicting residual oil distribution based on a physical constraint neural network. Background Art
[0002] Water injection development is one of the most common technologies in secondary oil recovery. It increases formation pressure by injecting water into the oil layer to improve oil recovery efficiency. However, after years of exploitation, the oil field has entered a high water content stage with a high average water content, which may even exceed 90%, resulting in a decrease in water drive effect and an increase in ineffective water circulation, making water injection oil extraction ineffective. In addition, the distribution of residual oil in the oil field is very complex, usually showing a combination of widespread dispersion and local concentration. Accurately grasping the distribution of residual oil in the oil field is crucial for further tapping the potential of residual oil and improving water drive recovery efficiency. Therefore, developing a method that can quickly and accurately predict the distribution of residual oil in the reservoir is one of the key factors in achieving efficient oil field exploitation.
[0003] Currently, numerical simulation is the mainstream method for predicting the distribution of remaining oil in reservoirs. However, this method faces challenges in modeling and computational efficiency, especially for the simulation of an entire oil field or a large block. It may take several days or even longer to complete a simulation, which makes it difficult to adapt to the needs of production history matching and development adjustments.
[0004] Recently, with the accumulation of a large amount of actual oilfield production data and the development of neural network proxy models, new approaches have been provided for automated history matching. The proxy model is designed to replace the traditional numerical simulator. It extracts features from the input data and establishes a mapping between the input and output data by analyzing the relationship between geological parameters (such as permeability) and the dynamic response of the model (such as saturation changes). The proxy model can significantly shorten the prediction time and improve the processing speed, however, its calculation accuracy is average. There is an urgent need for a faster and more accurate method for predicting the remaining oil distribution to achieve high-precision prediction of the remaining oil distribution. Summary of the invention
[0005] In order to enrich the process routes and increase the selection space, the embodiment of the present invention provides a method and device for predicting the remaining oil distribution based on a physical constraint neural network, which greatly reduces the calculation time and describes the remaining oil change process in detail, providing effective support for the analysis of the dynamic evolution law of water injection development oil reservoirs.
[0006] In a first aspect, an embodiment of the present invention provides a method for predicting residual oil distribution based on a physical constraint neural network, comprising:
[0007] The reservoir physical property data, water injection development plan data and time step of the target block are input into the remaining oil distribution time series prediction model, and the remaining oil distribution time series data is obtained according to the output result of the model. The remaining oil distribution time series prediction model is pre-established in the following way:
[0008] Based on the geological model of the target block, the time series data of the water injection development scheme and the remaining oil distribution image under different reservoir physical property images are obtained through numerical simulation. For the time series data and reservoir physical property image of the same water injection development scheme, the reservoir physical property image and the time series data of the corresponding remaining oil distribution image are spliced one by one to obtain a spliced time series image, and the new time series data is obtained by combining the time series data of the water injection development scheme as a sample to obtain a sample set;
[0009] The sample set is input into a selected neural network. The neural network is embedded in a data processing network based on the prediction network. The data processing network comprises a physical condition convolution calculation layer, an image convolution calculation layer and a data fusion layer. The physical condition convolution calculation layer is used to perform feature extraction and up-sampling on the time series data of the water injection development plan in the sample to obtain the convolution feature of the water injection development plan. The image convolution calculation layer is used to perform feature extraction on the spliced time series image to obtain the image convolution feature. The data fusion layer is used to obtain the fusion feature of the convolution feature of the water injection development plan and the image convolution feature. The output data of the data processing network is used to train the prediction network to obtain a time series prediction model for the remaining oil distribution.
[0010] In a second aspect, an embodiment of the present invention provides a method for establishing a remaining oil distribution prediction model based on a physical constraint neural network, comprising:
[0011] Based on the geological model of the target area, numerical simulation is used to obtain the time series data of water injection development schemes and remaining oil distribution images under different reservoir physical property images;
[0012] For the time series data and reservoir physical property images of the same water injection development plan, the reservoir physical property images and the time series data of the corresponding remaining oil distribution images are spliced one by one to obtain a spliced time series image, and the new time series data is obtained by combining the time series data of the water injection development plan as a sample to obtain a sample set;
[0013] The sample set is input into a selected neural network, the neural network is embedded in a data processing network based on the prediction network, the data processing network comprises a physical condition convolution calculation layer, an image convolution calculation layer and a data fusion layer, the physical condition convolution calculation layer is used to perform feature extraction and up-sampling on the time series data of the water injection development plan in the sample to obtain the convolution feature of the water injection development plan, the image convolution calculation layer is used to perform feature extraction on the spliced time series image to obtain the image convolution feature, and the data fusion layer is used to obtain the fusion feature of the convolution feature of the water injection development plan and the image convolution feature;
[0014] The prediction network is trained using the output data of the data processing network to obtain a residual oil distribution time series prediction model.
[0015] In a third aspect, an embodiment of the present invention provides a device for predicting the remaining oil distribution based on a physical constraint neural network, the device comprising: a memory and a processor; the memory is used for a program for predicting the remaining oil distribution based on a physical constraint neural network, and the processor is used to read and execute the program to execute the above-mentioned method for predicting the remaining oil distribution based on a physical constraint neural network.
[0016] In a fourth aspect, an embodiment of the present invention provides a device for establishing a remaining oil distribution prediction model based on a physical constraint neural network, the device comprising: a memory and a processor; the memory is used for a program for establishing a remaining oil distribution prediction model based on a physical constraint neural network, and the processor is used to read and execute the program to execute the above-mentioned method for establishing a remaining oil distribution prediction model based on a physical constraint neural network.
[0017] In a fifth aspect, an embodiment of the present invention provides a computer storage medium, in which computer executable instructions are stored. When the computer executable instructions are executed by a processor, the above-mentioned remaining oil distribution prediction method based on a physical constraint neural network is implemented, or the above-mentioned remaining oil distribution prediction model establishment method based on a physical constraint neural network is implemented.
[0018] The beneficial effects of the above technical solution provided by the embodiment of the present invention include at least:
[0019] The remaining oil distribution prediction method based on physical constraint neural network provided by the embodiment of the present invention takes reservoir physical properties and water injection development scheme as physical constraint conditions of the dynamic change process of the reservoir, adopts time series image prediction method to establish a time series prediction model of remaining oil distribution in water injection development, uses physical condition convolution calculation layer to extract the main features of water injection development scheme, uses image convolution calculation layer to extract the main features of reservoir attribute field and remaining oil distribution, uses data fusion layer to realize different dimensional data feature fusion, establishes multi-channel feature fusion calculation network, and realizes rapid prediction of remaining oil distribution in water injection development. Compared with traditional numerical simulation methods and other neural network remaining oil prediction models, this method greatly reduces the calculation time consumption, and can describe the remaining oil change process in detail, providing effective support for the analysis of dynamic evolution law of water injection development reservoir.
[0020] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.
[0021] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:
[0023] Figure 1 This is a flow chart of a method for establishing a remaining oil distribution prediction model based on a physical constraint neural network in Embodiment 1 of the present invention;
[0024] Figure 2 for Figure 1 Specific implementation flow chart of step S11;
[0025] Figure 3 This is a network structure diagram of ConvLSTM in Embodiment 2 of the present invention;
[0026] Figure 4 This is a technical flow chart of a method for predicting remaining oil distribution based on a physical constraint neural network in Embodiment 2 of the present invention;
[0027] Figure 5 This is a specific implementation flow chart of the remaining oil distribution prediction method based on the physical constraint neural network in the second embodiment of the present invention. DETAILED DESCRIPTION
[0028] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.
[0029] It should be understood that the terms described in the present invention are only for describing special embodiments and are not intended to limit the present invention. In addition, for the numerical range in the present invention, it should be understood that each intermediate value between the upper and lower limits of the scope is also specifically disclosed. Each smaller range between the intermediate value in any stated value or stated range and any other stated value or intermediate value in the described range is also included in the present invention. The upper and lower limits of these smaller ranges can be independently included or excluded in the scope.
[0030] Unless otherwise indicated, all technical and scientific terms used herein have the same meanings as commonly understood by those skilled in the art to which the invention belongs. Although the present invention describes only preferred methods and materials, any methods and materials similar or equivalent to those described herein may also be used in the implementation or testing of the present invention. All documents mentioned in this specification are incorporated by reference to disclose and describe the methods and / or materials related to the documents. In the event of a conflict with any incorporated document, the content of this specification shall prevail.
[0031] The embodiment of the present invention provides a method and device for predicting the distribution of remaining oil based on a physical constraint neural network, which is shorter in calculation time than numerical simulation and describes the remaining oil change process in detail, providing effective support for the analysis of the dynamic evolution law of water injection development oil reservoirs.
[0032] Embodiment 1
[0033] Embodiment 1 of the present invention provides a method for establishing a residual oil distribution prediction model based on a physical constraint neural network. Figure 1 As shown, the following steps are included:
[0034] Step S11: Based on the geological model of the target block, the time series data of the water injection development scheme and the remaining oil distribution image under different reservoir physical property images are obtained through numerical simulation.
[0035] For details, see Figure 2 As shown, the following steps are included:
[0036] Step S111: Based on the geological model of the target block, determine the reservoir attribute data set used for reservoir numerical simulation, including the area, reserves, thickness, number of layers, porosity, permeability and relative permeability curve of the reservoir.
[0037] Step S112: Determine the time series data of the water injection development plan based on the water well injection volume, oil well production volume, bottom hole flowing pressure and production days.
[0038] Furthermore, in the time series data of the water injection development plan, the water injection development plans at different times may be the same or different.
[0039] The time series data of the water injection development plan is saved as one-dimensional vector format data.
[0040] Step S113: sequentially change the data of a parameter in the reservoir attribute data set, or change the time series data of the water injection development plan, use a numerical simulator to perform reservoir numerical simulation, and obtain the time series data of the water injection development plan and the remaining oil distribution image under different reservoir physical property images.
[0041] Taking reservoir properties and water injection development scheme as the physical constraints of the dynamic change process of the reservoir, numerical simulation is used to obtain the time series images of the remaining oil distribution under different physical constraints. Furthermore, the reservoir property image is a layered reservoir property image, that is, the reservoir property data of each layer is obtained, and its display form can be a plane distribution diagram of the reservoir property of the layer, for example, using color to represent the size of the value.
[0042] It should be noted that the reservoir physical property image here is not an image of all physical properties of the reservoir, but an image of the changing physical properties. Taking the study of the influence of permeability changes on the distribution of remaining oil in the process of water injection development of oil reservoirs as an example, the reservoir physical property image can be specific to the layered permeability image.
[0043] Correspondingly, the remaining oil distribution image is a layered remaining oil distribution image.
[0044] Step S12: For the time series data and reservoir property images of the same water injection development plan, the reservoir property images and the corresponding remaining oil distribution image time series data are spliced one by one to obtain a spliced time series image, and the new time series data is obtained by combining the time series data of the water injection development plan as a sample to obtain a sample set.
[0045] Furthermore, before step S12, the acquired data is usually preprocessed as follows:
[0046] (1) The MinMaxScaler function is used to process the different parameter data in the time series data of the water injection development plan to eliminate the adverse effects caused by singular sample data.
[0047] Since there are generally no abnormalities in production days, the following processing is usually performed on the water well injection volume, oil well production volume and bottom hole flow pressure:
[0048]
[0049] In the formula, The parameter data of the water injection development plan, i.e. injection volume, production volume or bottom hole pressure, and are the minimum and maximum values of the corresponding parameter data, is the processed parameter data.
[0050] (2) Adjusting the acquired reservoir physical property image and the corresponding remaining oil distribution image to images of a set size.
[0051] In order to achieve data unification, the image data needs to be adjusted to images of the same size; taking into account both the amount of calculation and the calculation accuracy, the preferred solution determined in this embodiment is to adjust the reservoir physical property image and the corresponding remaining oil distribution image to images of a size of 256×256.
[0052] Accordingly, the reservoir physical property image and the corresponding residual oil distribution image in the time series data of the residual oil distribution image are spliced one by one to obtain a matrix with a dimension of 2×256×256, where the first channel represents the residual oil distribution and the second channel represents the reservoir physical property.
[0053] Step S13: Input the sample set into the selected neural network, which is embedded in a data processing network based on the prediction network. The data processing network includes a physical condition convolution calculation layer, an image convolution calculation layer and a data fusion layer. The physical condition convolution calculation layer is used to extract and upsample the time series data of the water injection development plan in the sample to obtain the convolution features of the water injection development plan. The image convolution calculation layer is used to extract features from the spliced time series images to obtain the image convolution features. The data fusion layer is used to obtain the fusion features of the convolution features of the water injection development plan and the image convolution features.
[0054] The selected network may be a ConvLSTM network.
[0055] The workflow of a data processing network may include:
[0056] (1) The physical condition convolution calculation layer is used to extract features from the time series data of the water injection development plan in the sample, and the extracted features are upsampled by the bilinear interpolation algorithm to obtain the convolution features of the water injection development plan with a dimension of 1×256×256. The process of physical condition convolution feature extraction is expressed as:
[0057]
[0058] in, Represents the physical condition convolution calculation; represents upsampling; Developing solutions for water injection; It is the feature obtained by the physical condition convolution calculation layer.
[0059] (2) Use the image convolution calculation layer to extract features from the spliced time series images to obtain image convolution features.
[0060] The process of image convolution feature extraction is expressed as:
[0061]
[0062] in, Represents image convolution calculation; The image matrix with the dimension of 2×256×256 obtained in step 12; It is the feature obtained by the image convolution layer.
[0063] (3) The data fusion layer is used to obtain the fusion features of the water injection development plan convolution features and the image convolution features.
[0064] Through the data fusion layer, the physical condition convolution layer features obtained in the above step (1) and the image convolution layer features obtained in step (2) are fused, and the dimension of the fused features is 3×256×256:
[0065]
[0066] in, For data fusion calculation, Represents fusion features.
[0067] Step S14: using the output data of the data processing network to train the prediction network to obtain a remaining oil distribution time series prediction model.
[0068] Before using the output data of the data processing network to train the prediction network, the network training parameters are set first: the remaining oil distribution image of the previous production time is used to predict the remaining oil distribution image of the next production time, the number of iterations is 500 times, the optimizer is the Adam optimizer, the learning rate is 0.0003, and the learning rate decrease rate is 0.91.
[0069] The resulting remaining oil distribution time series prediction model is used to predict the remaining oil distribution time series data based on the input reservoir physical property data, water injection development plan data and time step of the target block.
[0070] Based on the inventive concept of the present invention, an embodiment of the present invention further provides a method for predicting residual oil distribution based on a physical constraint neural network, comprising:
[0071] The reservoir property data, water injection development plan data and time step of the target block are input into the remaining oil distribution time series prediction model, and the remaining oil distribution time series image is obtained according to the output results of the model.
[0072] The residual oil distribution time series prediction model is pre-established by the method in the above-mentioned embodiment 1.
[0073] The remaining oil distribution prediction method based on physical constraint neural network provided by the embodiment of the present invention takes reservoir physical properties and water injection development scheme as physical constraint conditions of the dynamic change process of the reservoir, adopts time series image prediction method to establish a time series prediction model of remaining oil distribution in water injection development, uses physical condition convolution calculation layer to extract the main features of water injection development scheme, uses image convolution calculation layer to extract the main features of reservoir attribute field and remaining oil distribution, uses data fusion layer to realize different dimensional data feature fusion, establishes multi-channel feature fusion calculation network, and realizes rapid prediction of remaining oil distribution in water injection development. Compared with traditional numerical simulation methods and other neural network remaining oil prediction models, this method greatly reduces the calculation time consumption, and can describe the remaining oil change process in detail, providing effective support for the analysis of dynamic evolution law of water injection development reservoir.
[0074] Embodiment 2
[0075] Embodiment 2 of the present invention provides a specific implementation process of a remaining oil distribution prediction method based on a physical constraint neural network.
[0076] The convolution-long short-term memory network ConvLSTM is selected. Its network structure is as follows Figure 3 shown.
[0077] ConvLSTM embeds the data processing network on the basis of the prediction network. The embedding of the data processing network realizes the feature extraction and feature reshaping of multi-dimensional time series data to capture more complex spatiotemporal relationships.
[0078] Among them, the prediction network contains input convolution layer, long short-term memory network and output convolution layer:
[0079] (1) The input convolution layer is configured with 3 input channels, 128 output channels, a 5×5 convolution kernel, a stride of 1, and a padding of 2;
[0080] (2) The LSTM network contains four hidden layers, each of which has forget gate, input gate, output gate and update gate functions;
[0081] (3) The output convolution layer is set to have 128 input channels, 1 output channel, 1×1 kernel size, 1 stride, and no padding.
[0082] The data processing network includes:
[0083] (1) Physical condition convolution calculation layer, including two convolution processing layers, each layer has 1 input and output channel, the convolution kernel size is 3×3, the step size is set to 1, and the padding size is 1;
[0084] (2) The configuration of the image convolution calculation layer is: input channel 2, output channel 2, convolution kernel size is 1×1, stride is 1, and there is no padding.
[0085] (3) Data fusion layer.
[0086] The core formula of ConvLSTM is as follows:
[0087] Input Gate:
[0088] Forget Gate:
[0089] Candidate memory cells:
[0090] Memory unit update:
[0091] Output Gate:
[0092] Hidden state:
[0093] in, Represents the convolution operation; Represents element-wise product operation; Represents tanh or sigmoid function; express Fusion features of neuron input at each moment; , Control the input gate and forget gate respectively; is the output gate; for Moment candidate memory unit; for The neuronal state at each moment; Represents the weight of the input to each gate; Represents the weights of the hidden state to each gate; Represents the weight of the unit state to each gate; Indicates bias; express The output of the neuron at the moment. Among the three gated units, the forget gate controls how much information is discarded / inherited from the previous memory; the input gate determines how much input information at the current moment is added to the memory information stream; the output gate determines the amount of information transmitted to the hidden state from the memory cell at the current moment. Each gate is controlled by the hidden state of the previous moment. ; Cell state and the current input It is composed of a linear combination of and activation function.
[0094] The remaining oil distribution prediction method based on physical constraint neural network provided in the second embodiment of the present invention takes the change of reservoir physical property data as an example, and its technical process is shown in FIG. Figure 4 shown.
[0095] For the specific implementation process, see Figure 5 As shown, it includes the following steps:
[0096] Step S51: Based on the geological model of the target block, water injection development plan data and time series data of the remaining oil distribution image under different reservoir permeability images are obtained through numerical simulation.
[0097] Step S52: splicing the permeability image with the remaining oil distribution image in the time series data one by one, obtaining new time series data as a sample, and obtaining a sample set.
[0098] (1) The MinMaxScaler function is used to process the injection volume, production volume and bottom hole pressure in the water injection development plan data.
[0099] (2) The acquired permeability image and remaining oil distribution image are resized to an image of 256 × 256.
[0100] (3) The permeability image and the remaining oil distribution image in the time series data are concatenated one by one to form a matrix with a dimension of 2×256×256. The new time series data is taken as a sample to obtain a sample set.
[0101] Step S53: Divide the sample set into a training set and a validation set in a ratio of 8:2.
[0102] Step S54: Input the training set into the ConvLSTM network, use the physical condition convolution calculation layer to extract features and upsample the water injection development plan data in the sample, and obtain the water injection development plan convolution features with a dimension of 1×256×256. Use the image convolution calculation layer to extract features from the spliced image to obtain image convolution features. Use the data fusion layer to fuse the water injection development plan convolution features and the image convolution features to obtain fused features with a size of 3×256×256.
[0103] Step S55: using the output data of the data processing network to train the prediction network to obtain a remaining oil distribution time series prediction model.
[0104] Step S56: Use the verification set to determine whether the prediction accuracy of the current remaining oil distribution time series prediction model meets the requirements.
[0105] If the judgment result of step S56 is no, execute step S57; if the judgment result of step S56 is yes, execute step S58.
[0106] Step S57: modify the hyperparameters of the current remaining oil distribution time series prediction model.
[0107] After step S57, the process returns to step S54.
[0108] Step S58: input the reservoir permeability data, water injection development plan data and time step of the target block into the remaining oil distribution time series prediction model, and obtain the remaining oil distribution time series image according to the output result of the model.
[0109] Based on the inventive concept of the present invention, an embodiment of the present invention also provides a device for predicting the remaining oil distribution based on a physical constraint neural network, the device comprising: a memory and a processor; the memory is used for a program for predicting the remaining oil distribution based on a physical constraint neural network, the processor is used to read and execute the program, and execute the above-mentioned method for predicting the remaining oil distribution based on a physical constraint neural network.
[0110] Based on the inventive concept of the present invention, an embodiment of the present invention also provides a device for establishing a remaining oil distribution prediction model based on a physical constraint neural network, the device comprising: a memory and a processor; the memory is used for a program for establishing a remaining oil distribution prediction model based on a physical constraint neural network, and the processor is used to read and execute the program to execute the above-mentioned method for establishing a remaining oil distribution prediction model based on a physical constraint neural network.
[0111] Based on the inventive concept of the present invention, an embodiment of the present invention further provides a computer storage medium, in which computer executable instructions are stored. When the computer executable instructions are executed by a processor, the above-mentioned remaining oil distribution prediction method based on a physical constraint neural network is implemented, or the above-mentioned remaining oil distribution prediction model establishment method based on a physical constraint neural network is implemented.
[0112] Unless otherwise specifically stated, terms such as processing, computing, calculating, determining, displaying, etc. may refer to the actions and / or processes of one or more processing or computing systems, or similar devices, which operate and convert data represented as physical (e.g., electronic) quantities within registers or memories of a processing system into other data similarly represented as physical quantities within memories, registers, or other such information storage, transmission, or display devices of the processing system. Information and signals may be represented using any of a variety of different techniques and methods. For example, data, instructions, commands, information, signals, bits, symbols, and chips mentioned throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, light fields or particles, or any combination thereof.
[0113] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of protection of the present disclosure. The attached method claims present the elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.
[0114] In the above detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that the embodiments of the claimed subject matter require more features than those set forth in each claim. On the contrary, as reflected in the appended claims, the invention is in a state of having less than all the features of the disclosed individual embodiments. Therefore, the appended claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.
[0115] Those skilled in the art will also appreciate that the various illustrative logic blocks, modules, circuits, and algorithmic steps described in conjunction with the embodiments herein can all be implemented as electronic hardware, computer software, or a combination thereof. In order to clearly illustrate the interchangeability between hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above around their functions. Whether such functions are implemented as hardware or software depends on specific applications and the design constraints imposed on the entire system. A skilled person can implement the described functions in an alternative manner for each specific application, but such implementation decisions should not be interpreted as departing from the scope of protection of the present disclosure.
[0116] The steps of the method or algorithm described in conjunction with the embodiments herein may be directly embodied as hardware, a software module executed by a processor, or a combination thereof. The software module may be located in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a mobile disk, a CD-ROM, or any other form of storage medium well known in the art. An exemplary storage medium is connected to the processor so that the processor can read information from the storage medium and can write information to the storage medium. Of course, the storage medium may also be an integral part of the processor. The processor and the storage medium may be located in an ASIC. The ASIC may be located in a user terminal. Of course, the processor and the storage medium may also be present in a user terminal as discrete components.
[0117] For software implementation, the techniques described in this application can be implemented with modules (e.g., procedures, functions, etc.) that perform the functions described in this application. These software codes can be stored in a memory unit and executed by a processor. The memory unit can be implemented within the processor or outside the processor. In the latter case, it is coupled to the processor in a communication manner via various means, which are well known in the art.
[0118] The above description includes examples of one or more embodiments. Of course, it is impossible to describe all possible combinations of components or methods for the purpose of describing the above embodiments, but it should be recognized by those skilled in the art that the various embodiments may be further combined and arranged. Therefore, the embodiments described herein are intended to cover all such changes, modifications and variations that fall within the scope of protection of the appended claims. In addition, with respect to the term "comprising" used in the specification or claims, the word is covered in a manner similar to the term "including", just as "including," is explained as a transitional word in the claims. In addition, any term "or" used in the specification of the claims is intended to mean "non-exclusive or".
Claims
1. A method for predicting residual oil distribution based on physical constraint neural network, characterized in that: include: The reservoir physical property data, water injection development plan data and time step of the target block are input into the remaining oil distribution time series prediction model, and the remaining oil distribution time series data is obtained according to the output result of the model. The remaining oil distribution time series prediction model is pre-established in the following way: Based on the geological model of the target block, a reservoir attribute data set for numerical simulation of the reservoir is determined, including the area, reserves, thickness, number of layers, porosity, permeability and relative permeability curve of the reservoir; based on the water well injection volume, oil well production volume, bottom hole flow pressure and production days, the time series data of the water injection development plan is determined; the data of a parameter in the reservoir attribute data set is sequentially changed, or the time series data of the water injection development plan is changed, and the numerical simulation of the reservoir is performed using a numerical simulator to obtain the time series data of the water injection development plan and the remaining oil distribution image under different reservoir physical property images; For the time series data and reservoir physical property images of the same water injection development plan, the reservoir physical property images and the corresponding remaining oil distribution images are adjusted to images of the same size, and the adjusted reservoir physical property images and the time series data of the corresponding remaining oil distribution images are spliced one by one to obtain a spliced time series image, and the new time series data is obtained by combining the time series data of the water injection development plan as a sample to obtain a sample set; The sample set is input into the ConvLSTM network, and the ConvLSTM network is embedded in the data processing network on the basis of the prediction network, and the data processing network includes a physical condition convolution calculation layer, an image convolution calculation layer and a data fusion layer; the physical condition convolution calculation layer includes two layers of convolution processing layers, and the number of input and output channels of each layer is 1, the convolution kernel size is 3×3, the step size is set to 1, and the padding size is 1; the configuration of the image convolution calculation layer is 2 input channels, 2 output channels, the convolution kernel size is 1×1, the step size is 1, and there is no padding; the prediction network includes an input convolution layer, a long short-term memory network and an output convolution layer; the input convolution layer is configured as 3 input channels, 128 output channels, a 5×5 convolution kernel is used, the step size is 1, and the padding is 2; the long short-term memory network includes 4 hidden layers, each layer has forget gate, input gate, output gate and update gate functions; the output convolution layer is set as 128 input channels, 1 output channel, the convolution kernel size is 1×1, the step size is 1, and there is no padding; The physical condition convolution calculation layer is used to perform feature extraction and up-sampling on the time series data of the water injection development plan in the sample to obtain the convolution features of the water injection development plan. The image convolution calculation layer is used to perform feature extraction on the spliced time series image to obtain the image convolution features. The data fusion layer is used to obtain the fusion features of the convolution features of the water injection development plan and the image convolution features. The output data of the data processing network is used to train the prediction network to obtain the remaining oil distribution time series prediction model.
2. The method according to claim 1, characterized in that The step of adjusting the reservoir physical property image and the corresponding remaining oil distribution image to images of the same size includes: The reservoir physical property image and the corresponding remaining oil distribution image are adjusted to an image size of 256×256; accordingly, The step of splicing the adjusted reservoir physical property image with the corresponding time series data of the remaining oil distribution image one by one includes: The adjusted reservoir physical property images are spliced one by one with the images in the time series data of the corresponding remaining oil distribution image to obtain matrix data with a dimension of 2×256×256.
3. The method according to claim 2, characterized in that The physical condition convolution calculation layer is used to extract features and upsample the time series data of the water injection development plan in the sample to obtain the convolution features of the water injection development plan, including: The physical condition convolution calculation layer is used to extract features from the time series data of the water injection development plan in the sample, and the extracted features are upsampled through a bilinear interpolation algorithm to obtain a convolution feature of the water injection development plan with a dimension of 1×256×256.
4. The method according to claim 3, characterized in that The obtained fusion feature is a fusion feature with a dimension of 3×256×256.
5. The method according to claim 1, characterized in that Before the prediction network is trained using the output data of the data processing network, the method further comprises: Set the network training parameters: use the remaining oil distribution image of the previous production time to predict the remaining oil distribution image of the next production time, the number of iterations is 500 times, the optimizer is the Adam optimizer, the learning rate is 0.0003, and the learning rate decrease rate is 0.
91.
6. The method according to any one of claims 1 to 5, characterized in that: The reservoir physical property images in the sample set are stratified reservoir physical property images, and the remaining oil distribution images in the sample set are stratified remaining oil distribution images.
7. A method for establishing a residual oil distribution prediction model based on a physical constraint neural network, characterized in that: include: Based on the geological model of the target block, determine the reservoir attribute data set used for numerical simulation of the reservoir, including the area, reserves, thickness, number of layers, porosity, permeability and relative permeability curve of the reservoir; determine the time series data of the water injection development plan based on the water well injection volume, oil well production volume, bottom hole flow pressure and production days; sequentially change the data of one parameter in the reservoir attribute data set, or change the time series data of the water injection development plan, use the numerical simulator to perform numerical simulation of the reservoir, and obtain the time series data of the water injection development plan and the remaining oil distribution image under different reservoir physical property images; For the time series data and reservoir physical property images of the same water injection development plan, the reservoir physical property images and the corresponding remaining oil distribution images are adjusted to images of the same size, and the adjusted reservoir physical property images and the time series data of the corresponding remaining oil distribution images are spliced one by one to obtain a spliced time series image, and the new time series data is obtained by combining the time series data of the water injection development plan as a sample to obtain a sample set; The sample set is input into the ConvLSTM network, and the ConvLSTM network is embedded in the data processing network on the basis of the prediction network, and the data processing network includes a physical condition convolution calculation layer, an image convolution calculation layer and a data fusion layer; the physical condition convolution calculation layer includes two layers of convolution processing layers, and the number of input and output channels of each layer is 1, the convolution kernel size is 3×3, the step size is set to 1, and the padding size is 1; the configuration of the image convolution calculation layer is 2 input channels, 2 output channels, the convolution kernel size is 1×1, the step size is 1, and there is no padding; the prediction network includes an input convolution layer, a long short-term memory network and an output convolution layer; the input convolution layer is configured as 3 input channels, 128 output channels, a 5×5 convolution kernel is used, the step size is 1, and the padding is 2; the long short-term memory network includes 4 hidden layers, each layer has forget gate, input gate, output gate and update gate functions; the output convolution layer is set as 128 input channels, 1 output channel, the convolution kernel size is 1×1, the step size is 1, and there is no padding; The physical condition convolution calculation layer is used to perform feature extraction and up-sampling on the time series data of the water injection development plan in the sample to obtain the convolution feature of the water injection development plan, the image convolution calculation layer is used to perform feature extraction on the spliced time series image to obtain the image convolution feature, and the data fusion layer is used to obtain the fusion feature of the convolution feature of the water injection development plan and the image convolution feature; The prediction network is trained using the output data of the data processing network to obtain a residual oil distribution time series prediction model.
8. A residual oil distribution prediction device based on physical constraint neural network, characterized in that: The device comprises: a memory and a processor; the memory is used for a program for predicting the remaining oil distribution based on a physical constraint neural network, and the processor is used to read and execute the program to execute the method for predicting the remaining oil distribution based on a physical constraint neural network as described in any one of claims 1 to 6.
9. A device for establishing a residual oil distribution prediction model based on a physical constraint neural network, characterized in that: The device comprises: a memory and a processor; the memory is used for a program for establishing a remaining oil distribution prediction model based on a physical constraint neural network, and the processor is used for reading and executing the program to execute the method for establishing a remaining oil distribution prediction model based on a physical constraint neural network as described in claim 7.
10. A computer storage medium, characterized in that: The computer storage medium stores computer executable instructions, and when the computer executable instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.
Citation Information
Patent Citations
A prediction method for remaining oil distribution of a water flooding development oil field based on deep learning
CN109948841A
Residual oil distribution prediction method based on convolutional encoder-decoder network
CN112541572A
Residual oil distribution prediction method and device based on deep convolutional neural network
CN113052371A
Remaining oil distribution prediction method based on autoregressive network model
CN114492211A
Fracturing crack propagation rapid prediction method fusing physical condition constraint
CN117408168A
Cited By
Residual oil distribution prediction method based on geological data
CN120725226A
