A Multi-Response Joint Prediction Method and System for Geological Carbon Sequestration Based on Secondary Neuron Multilayer Perceptron
By using a quadratic mapping neuron multilayer perceptron to form quadratic and cross terms of input variables within neurons, the problems of high computational cost and decreased prediction performance in geological carbon dioxide sequestration prediction are solved, achieving higher accuracy in predicting gas saturation, pressure, and water production rate.
Patent Information
- Application Number
- CN202611261179.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-19
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies are computationally expensive and prone to overfitting in predicting geological carbon dioxide sequestration, and are difficult to effectively express complex nonlinear relationships, resulting in a decline in prediction performance.
A method based on a second-order neuron multilayer perceptron is adopted to enhance the network's ability to express complex nonlinear relationships by forming square and cross terms of input variables within neurons. Furthermore, a second-order mapping layer is used to simultaneously predict gas saturation, pressure, and water production rate.
It achieves smaller prediction errors and higher coefficients of determination, accurately recovers the spatial distribution of reservoirs, reduces computational costs, and improves prediction accuracy.
Smart Images

Figure CN123117430A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of carbon sequestration response prediction technology in geological exploration, specifically to a method and system for joint prediction of multiple responses to geological carbon sequestration based on a secondary neuron multilayer perceptron. Background Technology
[0002] Geological carbon dioxide sequestration is a technology that injects carbon dioxide into deep geological reservoirs and stores it for extended periods. During injection, factors such as reservoir spatial heterogeneity, porosity, permeability, carbon dioxide injection rate, and injection time collectively influence the fluid transport state within the reservoir. Among these, gas saturation reflects the spatial distribution and plume spread of carbon dioxide within the reservoir, pressure reflects reservoir pressure changes caused by injection, and water production reflects the displacement and flow of the water phase during carbon dioxide injection. Therefore, rapid prediction of gas saturation, pressure, and water production is crucial for comparing different injection sites and analyzing reservoir dynamics.
[0003] Existing technologies typically employ numerical simulation methods to calculate geological carbon dioxide sequestration processes. These methods establish multiphase flow control equations based on the mass conservation equation and Darcy's law, numerically solving for gas saturation, pressure, and corresponding water production rates at different time points for each grid cell. However, this method incurs significantly increased computational costs when analyzing multiple geological models, various injection sites, or long time series.
[0004] To reduce computational costs, existing technologies employ data-driven surrogate models to learn the mapping relationship between reservoir conditions and flow response. Implementations similar to this invention mainly include multilayer perceptrons and long short-term memory networks.
[0005] A Multilayer Perceptron (MLP) is a feedforward neural network consisting of an input layer, several fully connected hidden layers, and an output layer. Its inputs can include grid spatial coordinates, time, permeability, porosity, and carbon dioxide injection rate. For each hidden layer, the MLP neuron uses a set of trainable weights to perform a weighted sum of the inputs, adds a bias term, and then feeds the result into an activation function. After multilayer computation, the outputs are gas saturation, pressure, and water production rate. During training, the network weights and biases are iteratively updated based on the error between the network's predictions and the corresponding response data in the training samples. MLPs are characterized by their simple structure, ease of training, and flexible input formats, and have been used for predicting flow response and carbon dioxide sequestration effectiveness.
[0006] Conventional MLPs and LSTMs do not directly form squared terms and variable interaction terms within a single neuron; their representation of coupling relationships primarily relies on multi-layer first-order transformations and activation functions. To improve the network's ability to represent complex nonlinear relationships, existing methods typically increase the model capacity by increasing the network width or depth. However, this approach increases the number of trainable parameters, training time, and storage requirements; when the number of training samples is limited or the network size is mismatched with the data complexity, it may also increase the risk of overfitting, causing a decline in the model's predictive performance on untrained data. Summary of the Invention
[0007] The purpose of this invention is to provide a method and system for joint prediction of geological carbon sequestration multi-response based on a secondary neuron multilayer perceptron. This method learns the relationship between reservoir spatiotemporal conditions and three types of flow responses, realizes rapid joint prediction of reservoir responses, and solves at least one of the technical problems existing in the background art.
[0008] To achieve the above objectives, the present invention adopts the following technical solution:
[0009] In a first aspect, the present invention provides a multi-response joint prediction method for geological carbon sequestration based on a secondary neuron multilayer perceptron, comprising:
[0010] Acquire reservoir data under the geological carbon-sealed reservoir area to be predicted. The reservoir data includes grid coordinates, time, permeability, porosity, and carbon dioxide injection rate. Among them, grid spatial coordinates are used to represent the location of the sample in the reservoir; time is used to represent the moment of the reservoir's dynamic response.
[0011] The reservoir data is processed using a pre-trained quadratic mapping multilayer perceptron. After feedforward calculations through the quadratic mapping layers, normalized predicted values of gas saturation, pressure, and water production rate are obtained. The quadratic mapping multilayer perceptron includes a feedforward neural network consisting of an input layer, multiple hidden layers, and an output layer. The neurons in the feedforward neural network are quadratic mapping neurons, which connect multiple quadratic mapping layers sequentially. The output of the previous layer serves as the input of the next layer until the output layer obtains the three predicted values: gas saturation, pressure, and water production rate.
[0012] As a further limitation of the first aspect of the present invention, for any quadratic mapping layer, assuming the input of the layer is h, the calculation is performed using three sets of independent trainable weights and biases:
[0013] u1=W1h+b1
[0014] u2=W2h+b2
[0015] u3=W3h+b3
[0016] Among them, W1, W2 and W3 are three sets of trainable weight matrices, and b1, b2 and b3 are the corresponding trainable bias vectors;
[0017] Multiply the first two sets of calculation results by corresponding elements, and then add them to the third set of calculation results to obtain the pre-activation value of this layer: h'=u1⊙u2+u3; where ⊙ represents the multiplication of corresponding elements;
[0018] The pre-activated value is input into the activation function to obtain the output of the layer.
[0019] As a further limitation of the first aspect of the present invention, after multiplying the first two sets of weighted calculation results, the square term and cross term of the input variable can be formed in the same neuron; the third set of calculation results retains the first-order term of the input variable, so that the quadratic mapping neuron has a quadratic polynomial form before the activation function.
[0020] As a further limitation of the first aspect of the present invention, the training samples are input into the quadratic mapping multilayer perceptron, and the predicted values of gas saturation, pressure and water production rate are obtained by forward calculation layer by layer; a joint loss function is constructed based on the error between the three predicted values and the corresponding response data in the training samples; the gradient of the loss function with respect to the weights and biases of each layer is calculated by backpropagation, and the network parameters are updated using an optimization algorithm; the sample input, forward calculation, loss calculation, backpropagation and parameter update are repeated until a preset number of training rounds are reached.
[0021] As a further limitation of the first aspect of the present invention, for a mini-batch of data consisting of B training samples, the joint loss function is constructed using mean squared error as follows: Where L is the joint loss function, B is the number of training samples in each batch, and M is the number of output response variables. In this method, M=3. Indicates the first The first sample One predicted response value, Indicates the first The first sample There are three target response values, and the three response variables are gas saturation. ,pressure and water production rate .
[0022] As a further limitation of the first aspect of the present invention, the gradient of the loss function with respect to the weights and biases of each layer is calculated through backpropagation, and the network parameters are updated using the Adam optimization algorithm, including: assuming the input of the l-th layer is... Three sets of linear transformations are performed on the three predicted values to determine the gradients of the three sets of weights and biases in the l-th layer. At the same time, the error signal is propagated backward to the previous layer. After obtaining the gradients of each layer, the Adam optimization algorithm is used to update the network parameters.
[0023] Secondly, the present invention provides a multi-response joint prediction system for geological carbon sequestration based on a secondary neuron multilayer perceptron, comprising:
[0024] The acquisition module is used to acquire reservoir data under the geological carbon-sealed reservoir area to be predicted. The reservoir data includes grid coordinates, time, permeability, porosity, and carbon dioxide injection rate. Among them, the grid spatial coordinates are used to represent the position of the sample in the reservoir; the time is used to represent the moment of the reservoir dynamic response.
[0025] The processing module is used to process the reservoir data using a pre-trained quadratic mapping multilayer perceptron. After feedforward calculation by the quadratic mapping layer, normalized predicted values of gas saturation, pressure, and water production rate are obtained. The quadratic mapping multilayer perceptron includes a feedforward neural network composed of an input layer, multiple hidden layers, and an output layer. The neurons in the feedforward neural network are quadratic mapping neurons, which connect multiple quadratic mapping layers sequentially. The output of the previous layer is used as the input of the next layer until the output layer obtains the three predicted values: gas saturation, pressure, and water production rate.
[0026] Thirdly, the present invention provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the geological carbon sequestration multi-response joint prediction method based on a secondary neuron multilayer perceptron as described in the first aspect.
[0027] Fourthly, the present invention provides a computer device including a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the geological carbon sequestration multi-response joint prediction method based on a secondary neuron multilayer perceptron as described in the first aspect.
[0028] Fifthly, the present invention provides an electronic device, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the geological carbon sequestration multi-response joint prediction method based on a secondary neuron multilayer perceptron as described in the first aspect.
[0029] Terminology Explanation: A secondary neuron computes on the same input from a higher-order neuron, that is, it uses three sets of trainable weights to sum the input in a weighted manner, and adds the corresponding bias terms to each. Then, it multiplies the results of the first two sets of calculations, adds them to the result of the third set of calculations, and finally passes them through an activation function to obtain the neuron's output; a secondary neuron multilayer perceptron is a feedforward neural network composed of the above-mentioned neurons.
[0030] The beneficial effects of this invention are as follows: By using quadratic mapping neurons in a feedforward multilayer perceptron, the network's ability to express complex nonlinear relationships is enhanced by introducing quadratic terms into the input-output mapping of neurons, and the same network can be used to simultaneously predict gas saturation, pressure, and water production rate; quadratic terms and variable interaction terms are directly included in a single computational layer; feedforward computation is used, eliminating the need to pass cyclic states step by step; smaller prediction errors and higher coefficients of determination are obtained for gas saturation, pressure, and water production rate, and spatial distribution can be recovered more accurately.
[0031] The advantages of additional aspects of the invention will be set forth more clearly in the following description or will be learned by practice of the invention. Attached Figure Description
[0032] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0033] Figure 1 This is a functional network framework diagram of the secondary neuron multilayer perceptron described in an embodiment of the present invention.
[0034] Figure 2 This is a flowchart of the training method for a secondary neuron multilayer perceptron according to an embodiment of the present invention.
[0035] Figure 3 This is a schematic diagram of the true value of gas saturation under the same test conditions described in the embodiments of the present invention.
[0036] Figure 4 This diagram illustrates a comparison of the gas saturation prediction results and absolute errors of the method of this invention with those of LSTM and conventional MLP. Detailed Implementation
[0037] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0038] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0039] It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as here.
[0040] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, and / or groups thereof.
[0041] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.
[0042] To facilitate understanding of the present invention, the present invention will be further explained and described below with reference to the accompanying drawings and specific embodiments. However, the specific embodiments do not constitute a limitation on the embodiments of the present invention.
[0043] Those skilled in the art should understand that the accompanying drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily essential for implementing the present invention.
[0044] Example 1
[0045] In this embodiment 1, a multi-response joint prediction system for geological carbon sequestration is first provided, including: an acquisition module, used to acquire reservoir data under the geological carbon sequestration reservoir area to be predicted, the reservoir data including grid coordinates, time, permeability, porosity and carbon dioxide injection rate; wherein, grid spatial coordinates are used to represent the position of the sample in the reservoir; time is used to represent the moment of the reservoir dynamic response. A processing module is used to process the reservoir data using a pre-trained quadratic mapping multilayer perceptron, and after feedforward calculation by the quadratic mapping layer, normalized predicted values of gas saturation, pressure and water production rate are obtained; wherein, the quadratic mapping multilayer perceptron includes a feedforward neural network composed of an input layer, multiple hidden layers and an output layer; the neurons in the feedforward neural network are quadratic mapping neurons, and multiple quadratic mapping layers are connected sequentially, with the output of the previous layer serving as the input of the next layer, until the output layer obtains three predicted values: gas saturation, pressure and water production rate.
[0046] In this embodiment, a multi-response joint prediction method for geological carbon sequestration is implemented using the above-described system. The method includes: acquiring reservoir data under the geological carbon sequestration zone to be predicted using an acquisition module. The reservoir data includes grid coordinates, time, permeability, porosity, and carbon dioxide injection rate. The grid spatial coordinates represent the location of the sample within the reservoir, and the time represents the moment of the reservoir's dynamic response. A processing module processes the reservoir data using a pre-trained quadratic mapping multilayer perceptron. After feedforward calculations by the quadratic mapping layers, normalized predicted values for gas saturation, pressure, and water production are obtained. The quadratic mapping multilayer perceptron includes a feedforward neural network composed of an input layer, multiple hidden layers, and an output layer. The neurons in the feedforward neural network are quadratic mapping neurons, and multiple quadratic mapping layers are connected sequentially, with the output of the previous layer serving as the input of the next layer, until the output layer obtains the three predicted values: gas saturation, pressure, and water production.
[0047] In this embodiment, for any quadratic mapping layer, let the input of the layer be h, and calculate using three sets of independent trainable weights and biases respectively:
[0048] u1=W1h+b1
[0049] u2=W2h+b2
[0050] u3=W3h+b3
[0051] Among them, W1, W2 and W3 are three sets of trainable weight matrices, and b1, b2 and b3 are the corresponding trainable bias vectors;
[0052] Multiply the first two sets of calculation results by corresponding elements, and then add them to the third set of calculation results to obtain the pre-activation value of this layer: h'=u1⊙u2+u3; where ⊙ represents the multiplication of corresponding elements;
[0053] The pre-activated value is input into the activation function to obtain the output of the layer.
[0054] After multiplying the first two sets of weighted calculation results, the square term and cross term of the input variable can be formed within the same neuron; the third set of calculation results retains the first-order term of the input variable. Thus, the quadratic mapping neuron has a quadratic polynomial form before the activation function.
[0055] The training samples are input into the quadratic mapping multilayer perceptron, and the predicted values of gas saturation, pressure, and water production rate are obtained through layer-by-layer forward computation. A joint loss function is constructed based on the error between the three predicted values and the corresponding response data in the training samples. The gradient of the loss function with respect to the weights and biases of each layer is calculated through backpropagation, and the network parameters are updated using an optimization algorithm. The process of sample input, forward computation, loss calculation, backpropagation, and parameter update is repeated until the preset number of training rounds is reached.
[0056] A joint loss function is constructed based on the errors between the three predicted values and the corresponding response data in the training samples, including: for the... For a mini-batch of training samples, the joint loss function is constructed using mean squared error:
[0057] ;
[0058] in, For the joint loss function, This represents the number of training samples in each batch. In this method, the number of output response variables is specified. , Indicates the first The first sample One predicted response value, Indicates the first The first sample There are three target response values, and the three response variables are gas saturation. ,pressure and water production rate .
[0059] The gradient of the loss function with respect to the weights and biases of each layer is calculated through backpropagation, and the network parameters are updated using an optimization algorithm, including: assuming the... The input of the layer is The results of the three linear transformations are as follows:
[0060]
[0061]
[0062]
[0063] The pre-activation value is: ;
[0064] The layer output is: ;
[0065] definition Then the first The gradients of the three sets of weights and biases in layer three are as follows:
[0066]
[0067]
[0068]
[0069]
[0070]
[0071]
[0072] Meanwhile, the error signal propagates backward to the previous layer as follows:
[0073] ;
[0074] After obtaining the gradients of each layer, the Adam optimization algorithm is used to update the network parameters. Let any parameter to be optimized be denoted as . In its first The gradient at the next iteration is ;
[0075] The first moment estimate is: ;
[0076] The second moment estimate is: ;
[0077] Deviation correction
[0078] ;
[0079] ;
[0080] The parameter update formula is:
[0081] ;
[0082] in, For the first Network parameters for the next iteration For the current gradient, and These are the first-order moment estimate and the second-order moment estimate of the gradient, respectively. and These are the exponential decay coefficients of the first and second moments, respectively. For learning rate, To prevent positive numbers with a denominator of zero, during training, each batch of training samples is input into the network, and forward computation, loss calculation, backpropagation, and parameter update are performed sequentially until the preset number of training rounds is reached.
[0083] Example 2
[0084] This embodiment provides a joint prediction method for geological carbon dioxide sequestration multi-response (QMLP) based on a quadratic mapping multilayer perceptron. This method trains a quadratic mapping multilayer perceptron, using grid spatial location, time, reservoir physical parameters, and carbon dioxide injection conditions as inputs. It utilizes a feedforward neural network containing quadratic mapping neurons to simultaneously predict gas saturation, pressure, and water production rate. 1) The input from the same layer is processed through three sets of primary neurons. The results of the first two sets are multiplied element-wise and then added to the result of the third set, with the result output after activation. The input consists of seven dimensions: grid coordinates, time, permeability, porosity, and injection rate. The same network jointly outputs gas saturation, pressure, and water production rate. The simulated operating conditions are expanded into spatial grid-time samples, and the state field and water production rate sequence are recovered after prediction based on grid and time indices.
[0085] In this embodiment, the specific training method for the secondary mapping multilayer perceptron includes the following process:
[0086] S1. Obtain and organize training data
[0087] Reservoir data were acquired under different geological models and carbon dioxide injection sites. The data included the spatial coordinates of the reservoir grid, permeability and porosity, the carbon dioxide injection rate for each injection site, time information corresponding to different output times, and the corresponding gas saturation, pressure, and water production rate. Specifically, the grid spatial coordinates represent the sample's location within the reservoir; permeability and porosity describe reservoir properties; the injection rate describes the carbon dioxide injection conditions; and time represents the moment of the reservoir's dynamic response. Gas saturation, pressure, and water production rate are the three types of response variables that the network needs to learn and predict.
[0088] S2, Constructing training samples
[0089] Training samples were constructed using data from each grid within each work area at different time points. Each sample contained seven input variables: the coordinates of the three grids, time, permeability, porosity, and carbon dioxide injection rate; and three output variables: gas saturation, pressure, and water production rate.
[0090] S3. Normalize the input and output variables.
[0091] Because the numerical ranges of grid coordinates, time, permeability, porosity, injection rate, gas saturation, pressure, and water production rate differ, each input and output variable is normalized to ensure its value falls within the 0-1 range. Training and testing data are normalized uniformly. After normalization, the variables have similar numerical ranges, thus reducing the impact of differences in the dimensions and numerical scales of different variables on network training.
[0092] S4. Establish a quadratic mapping multilayer perceptron.
[0093] A feedforward neural network consisting of an input layer, multiple hidden layers, and an output layer is constructed, and the regular neurons in the network are replaced with quadratic mapping neurons.
[0094] For any quadratic mapping layer, let the input of the layer be h, and calculate using three independent sets of trainable weights and biases: u1 = W1h + b 1, u2=W2h+b 2, u3 = W3h + b3; where W1, W2, and W3 are three sets of trainable weight matrices, and b1, b2, and b3 are the corresponding trainable bias vectors.
[0095] Multiply the first two sets of calculation results by corresponding elements, and then add them to the third set of calculation results to obtain the pre-activation value of this layer: h'=u1⊙u2+u3; where ⊙ represents the multiplication of corresponding elements.
[0096] Subsequently, the pre-activated value is input into the activation function to obtain the output of the layer: h"=σ(h').
[0097] The first two sets of weighted calculation results, when multiplied, form the squared term and cross term of the input variable within the same neuron; the third set of calculation results retains the linear term of the input variable. Thus, the quadratic mapping neuron has a quadratic polynomial form before the activation function, while conventional neurons primarily use a linear weighted summation form before the activation function.
[0098] Multiple quadratic mapping layers are connected sequentially, with the output of the previous layer serving as the input of the next layer, until the output layer obtains three predicted values: gas saturation, pressure, and water production rate.
[0099] S5, Jointly trained network parameters
[0100] Training samples are input into the quadratic mapping multilayer perceptron, and predicted values of gas saturation, pressure, and water production rate are obtained through layer-by-layer forward computation. A joint loss function is constructed based on the errors between the three predicted values and the corresponding response data in the training samples. The gradient of the loss function with respect to the weights and biases of each layer is calculated through backpropagation, and the network parameters are updated using an optimization algorithm. The process of sample input, forward computation, loss calculation, backpropagation, and parameter update is repeated until a preset number of training rounds is reached. After training, the network weights and bias parameters are saved to form a quadratic mapping multilayer perceptron model that can be used for prediction.
[0101] In this embodiment, the trained quadratic mapping multilayer perceptron model is used to obtain the grid coordinates, time, permeability, porosity, and carbon dioxide injection rate of the area to be predicted, and the data is processed according to the same normalization rules as the training data. The processed input data is then fed into the trained quadratic mapping multilayer perceptron, and after feedforward calculations by each quadratic mapping layer, normalized predicted values of gas saturation, pressure, and water production rate are obtained. During the testing phase, the mean square error, absolute error, and coefficient of determination can be calculated based on the predicted values and corresponding target values to evaluate the model's predictive performance for the responses of the three types of reservoirs.
[0102] In this embodiment, three heterogeneous geological models were used, each with nine injection rates, resulting in a total of 27 numerical simulation scenarios. Each scenario used a 25×25×3 grid, simulated for 2161 days, and output 72 time points per month. The seven inputs included three grid coordinates, time, permeability, porosity, and injection rate, while the three outputs were gas saturation, pressure, and water production rate. All variables were normalized to the range of 0-1.
[0103] The network structure is 7-512-256-128-64-32-16-8-3, with each layer using quadratic mapping neurons; ReLU is the activation function, mean squared error is the loss function, the Adam learning rate is 1×10⁻⁴, the batch size is 1875, and the training run is 500 epochs. Conventional MLPs use the same number of layers and nodes, employing only conventional neurons; the first two layers of the LSTM have 512 and 256 LSTM units respectively, followed by 128, 64, 32, 16, and 8 fully connected neurons.
[0104] Appendix Figure 3 The gas saturation results under the same test conditions are given, including the true saturation value, the prediction results and absolute errors of this method, the prediction results and absolute errors of LSTM, and the prediction results and absolute errors of conventional MLP. Combined with... Figure 4 The spatial distribution of gas saturation obtained by this method is closer to the numerical simulation results, and the absolute error is smaller than that of LSTM and conventional MLP.
[0105] Example 3
[0106] This embodiment 3 provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they implement the geological carbon sequestration multi-response joint prediction method based on a secondary neuron multilayer perceptron as described above. The method includes:
[0107] Acquire reservoir data under the geological carbon-sealed reservoir area to be predicted. The reservoir data includes grid coordinates, time, permeability, porosity, and carbon dioxide injection rate. Among them, grid spatial coordinates are used to represent the location of the sample in the reservoir; time is used to represent the moment of the reservoir's dynamic response.
[0108] The reservoir data is processed using a pre-trained quadratic mapping multilayer perceptron. After feedforward calculations through the quadratic mapping layers, normalized predicted values of gas saturation, pressure, and water production rate are obtained. The quadratic mapping multilayer perceptron includes a feedforward neural network consisting of an input layer, multiple hidden layers, and an output layer. The neurons in the feedforward neural network are quadratic mapping neurons, which connect multiple quadratic mapping layers sequentially. The output of the previous layer serves as the input of the next layer until the output layer obtains the three predicted values: gas saturation, pressure, and water production rate.
[0109] Example 4
[0110] This embodiment 4 provides a computer device, including a memory and a processor, wherein the processor and the memory communicate with each other, and the memory stores program instructions that can be executed by the processor. The processor calls the program instructions to execute the geological carbon sequestration multi-response joint prediction method based on a secondary neuron multilayer perceptron as described above. The method includes:
[0111] Acquire reservoir data under the geological carbon-sealed reservoir area to be predicted. The reservoir data includes grid coordinates, time, permeability, porosity, and carbon dioxide injection rate. Among them, grid spatial coordinates are used to represent the location of the sample in the reservoir; time is used to represent the moment of the reservoir's dynamic response.
[0112] The reservoir data is processed using a pre-trained quadratic mapping multilayer perceptron. After feedforward calculations through the quadratic mapping layers, normalized predicted values of gas saturation, pressure, and water production rate are obtained. The quadratic mapping multilayer perceptron includes a feedforward neural network consisting of an input layer, multiple hidden layers, and an output layer. The neurons in the feedforward neural network are quadratic mapping neurons, which connect multiple quadratic mapping layers sequentially. The output of the previous layer serves as the input of the next layer until the output layer obtains the three predicted values: gas saturation, pressure, and water production rate.
[0113] Example 5
[0114] This embodiment 5 provides an electronic device, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the geological carbon sequestration multi-response joint prediction method based on a secondary neuron multilayer perceptron as described above. The method includes:
[0115] Acquire reservoir data under the geological carbon-sealed reservoir area to be predicted. The reservoir data includes grid coordinates, time, permeability, porosity, and carbon dioxide injection rate. Among them, grid spatial coordinates are used to represent the location of the sample in the reservoir; time is used to represent the moment of the reservoir's dynamic response.
[0116] The reservoir data is processed using a pre-trained quadratic mapping multilayer perceptron. After feedforward calculations through the quadratic mapping layers, normalized predicted values of gas saturation, pressure, and water production rate are obtained. The quadratic mapping multilayer perceptron includes a feedforward neural network consisting of an input layer, multiple hidden layers, and an output layer. The neurons in the feedforward neural network are quadratic mapping neurons, which connect multiple quadratic mapping layers sequentially. The output of the previous layer serves as the input of the next layer until the output layer obtains the three predicted values: gas saturation, pressure, and water production rate.
[0117] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can 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.) containing computer-usable program code.
[0118] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0119] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0120] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0121] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solutions disclosed in the present invention, various modifications or variations that can be made by those skilled in the art without creative effort should be included within the scope of protection of the present invention.
Claims
1. A multi-response joint prediction method for geological carbon sequestration based on a secondary neuron multilayer perceptron, characterized in that, include: Acquire reservoir data under the geological carbon-sealed reservoir area to be predicted. The reservoir data includes grid coordinates, time, permeability, porosity, and carbon dioxide injection rate. Among them, grid spatial coordinates are used to represent the location of the sample in the reservoir; time is used to represent the moment of the reservoir's dynamic response. The reservoir data is processed using a pre-trained quadratic mapping multilayer perceptron. After feedforward calculations through the quadratic mapping layers, normalized predicted values of gas saturation, pressure, and water production rate are obtained. The quadratic mapping multilayer perceptron includes a feedforward neural network consisting of an input layer, multiple hidden layers, and an output layer. The neurons in the feedforward neural network are quadratic mapping neurons, which connect multiple quadratic mapping layers sequentially. The output of the previous layer serves as the input of the next layer until the output layer obtains the three predicted values: gas saturation, pressure, and water production rate.
2. The multi-response joint prediction method for geological carbon sequestration based on a secondary neuron multilayer perceptron according to claim 1, characterized in that, For any quadratic mapping layer, let the input of the layer be h, and calculate using three sets of independent trainable weights and biases: u1=W1h+b1 u2=W2h+b2 u3=W3h+b3 Among them, W1, W2 and W3 are three sets of trainable weight matrices, and b1, b2 and b3 are the corresponding trainable bias vectors; Multiply the first two sets of calculation results by corresponding elements, and then add them to the third set of calculation results to obtain the pre-activation value of this layer: h'=u1⊙u2+u3; where ⊙ represents the multiplication of corresponding elements; The pre-activated value is input into the activation function to obtain the output of the layer.
3. The method for joint prediction of geological carbon sequestration based on a secondary neuron multilayer perceptron according to claim 2, characterized in that, After multiplying the first two sets of weighted calculation results, the square term and cross term of the input variable can be formed within the same neuron; the third set of calculation results retains the first-order term of the input variable. Thus, the quadratic mapping neuron has a quadratic polynomial form before the activation function.
4. The multi-response joint prediction method for geological carbon sequestration based on a secondary neuron multilayer perceptron according to claim 1, characterized in that, The training samples are input into the quadratic mapping multilayer perceptron, and the predicted values of gas saturation, pressure, and water production rate are obtained through layer-by-layer forward computation. A joint loss function is constructed based on the error between the three predicted values and the corresponding response data in the training samples. The gradient of the loss function with respect to the weights and biases of each layer is calculated through backpropagation, and the network parameters are updated using an optimization algorithm. The process of sample input, forward computation, loss calculation, backpropagation, and parameter update is repeated until the preset number of training rounds is reached.
5. The multi-response joint prediction method for geological carbon sequestration based on a secondary neuron multilayer perceptron according to claim 4, characterized in that, For a mini-batch of data consisting of B training samples, the joint loss function constructed using mean squared error is: Where L is the joint loss function, B is the number of training samples in each batch, and M is the number of output response variables. In this method, M=3. Indicates the first The first sample One predicted response value, Indicates the first The first sample There are three target response values, and the three response variables are gas saturation. ,pressure and water production rate .
6. The method for joint prediction of geological carbon sequestration based on a secondary neuron multilayer perceptron according to claim 4, characterized in that, The gradient of the loss function with respect to the weights and biases of each layer is calculated through backpropagation, and the network parameters are updated using the Adam optimization algorithm, including: assuming the input of the l-th layer is... Three sets of linear transformations are performed on the three predicted values to determine the gradients of the three sets of weights and biases in the l-th layer. At the same time, the error signal is propagated backward to the previous layer. After obtaining the gradients of each layer, the Adam optimization algorithm is used to update the network parameters.
7. A multi-response joint prediction system for geological carbon sequestration based on a secondary neuron multilayer perceptron, characterized in that, include: The acquisition module is used to acquire reservoir data under the geological carbon-sealed reservoir area to be predicted. The reservoir data includes grid coordinates, time, permeability, porosity, and carbon dioxide injection rate. Among them, the grid spatial coordinates are used to represent the position of the sample in the reservoir; the time is used to represent the moment of the reservoir dynamic response. The processing module is used to process the reservoir data using a pre-trained quadratic mapping multilayer perceptron. After feedforward calculation by the quadratic mapping layer, normalized predicted values of gas saturation, pressure, and water production rate are obtained. The quadratic mapping multilayer perceptron includes a feedforward neural network composed of an input layer, multiple hidden layers, and an output layer. The neurons in the feedforward neural network are quadratic mapping neurons, which connect multiple quadratic mapping layers sequentially. The output of the previous layer is used as the input of the next layer until the output layer obtains the three predicted values: gas saturation, pressure, and water production rate.
8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the geological carbon sequestration multi-response joint prediction method as described in any one of claims 1-6.
9. A computer device, characterized in that, The method includes a memory and a processor, the processor and the memory communicating with each other, the memory storing program instructions that can be executed by the processor, and the processor calling the program instructions to execute the geological carbon sequestration multi-response joint prediction method as described in any one of claims 1-6.
10. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the geological carbon sequestration multi-response joint prediction method as described in any one of claims 1-6.
