CCS fast simulation method based on multi-feature fusion neural network

Abstract

The application relates to the technical field of carbon capture and storage (CCS) simulation, and particularly provides a CCS fast simulation method based on a multi-feature fusion neural network; the method constructs a U-FNO network of a fusion type U neural network (U-Net) and a Fourier neural operator (FNO), improves the prediction accuracy of pressure field and saturation field in a CO2 geological storage process on a long time sequence; meanwhile, a yield sequence prediction network based on transformer encoders (TE) is used, the output end of the U-FNO network is directly connected to the input end of the TE network, direct prediction from geological parameters and well control curves to time sequence distribution of pressure and saturation field and corresponding yield curves is realized; the method effectively overcomes the problem that existing CCS simulation technologies are difficult to simultaneously process multi-feature input and output, improves the prediction efficiency of the influence of the CO2 storage rate on the reservoir safety, and can provide an efficient and reliable intelligent simulation tool for CCS engineering optimization and history matching.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of carbon capture and storage (CCS) simulation, and particularly provides a CCS fast simulation method based on a multi-feature fusion neural network. BACKGROUND

[0002] CO2 capture and storage (CCS) has become one of the key strategies for global climate change mitigation and significant reduction of greenhouse gas emissions. As a CO2 geological storage reservoir, the saline aquifer has great storage potential. Developing efficient CO2 storage technology in the saline aquifer faces three major challenges: evaluating the storage capacity of the saline aquifer to maximize the injection of CO2, simulating the injection rate (injection performance) under geological uncertainty parameters, and evaluating the geological risks during CO2 injection. In order to apply carbon storage technology on a large scale in saline aquifers and depleted oil and gas reservoirs, the CO2 injection method must be technically feasible, safe, economically reasonable, and repeatedly simulated for different storage strategies.

[0003] During the CO2 injection process, excessive pressure accumulation can cause geological risks. In order to prevent the geological risk of rock rupture due to continuous pressure rise, deploying production wells can effectively alleviate pressure accumulation, thereby increasing the amount of injectable CO2. However, premature start-up of production wells can cause CO2 breakthrough in the production well, resulting in low efficiency of the CO2 injection cycle. Therefore, it is crucial to quickly simulate and predict the injection regime of the geological prediction well and the opening time of the salt water well, especially the impact of different CO2 injection rates on the occurrence of related geological risks.

[0004] With the advancement of artificial intelligence technology, machine learning and deep learning have promoted the development of proxy models in CCS and environmental science research, but existing CCS simulation methods are difficult to effectively process multi-feature input and output, thereby failing to efficiently predict the impact of CO2 storage rate on reservoir safety. SUMMARY

[0005] The present application aims to provide a CCS fast simulation method based on a multi-feature fusion neural network, which aims to solve the technical problem that the CCS simulation method in the prior art is difficult to effectively process multi-feature input and output, thereby failing to efficiently predict the impact of CO2 storage rate on reservoir safety.

[0006] To achieve the above-mentioned purpose, the present application adopts a CCS fast simulation method based on a multi-feature fusion neural network,

[0007] comprising the following steps:

[0008] establishing a U-FNO architecture suitable for CCS fast simulation and directly outputting pressure and saturation states in the time dimension;

[0009] Based on the U-FNO architecture, a Transformer module with self-attention mechanism is introduced to obtain a fast simulation surrogate model for building CCS;

[0010] The permeability and well control sequence are selected as the model input. 10 permeability samples are randomly generated by using SGeMS for sequential Gaussian simulation, and 500 well control sequences are randomly generated for model training;

[0011] Finally, the overall performance analysis method is proposed, and the error measurement mechanism is introduced to complete the simulation.

[0012] Among them, when building a surrogate model:

[0013] In the prediction of CO2 reservoir storage process under geological uncertainty, the well rate of injection and production wells is adjusted in each reservoir model at each control time step; The mapping relationship required to realize the related surrogate model is represented as:

[0014] X=[P,S,q i ,q p ]=g(m,w i ,w p )

[0015] Among them, the geological model is represented by m, which is an element of , where nb is defined as nx×ny×nz, representing the total number of grid blocks in the geological model, nx, ny and nz correspond to the number of blocks on each coordinate axis; N m represents the number of samples of the geological model. w i is an element of , representing the well control sequence of the injection well, ni represents the number of injection wells, N c represents the number of control time steps; N w represents the number of well control sequences; w p is an element of , representing the well control sequence of the production well, np represents the number of production wells, u t represents the opening time step of the production well.

[0016] Among them, the specific way to establish a U-FNO architecture suitable for CCS fast simulation and directly output pressure and saturation state in time dimension is as follows:

[0017] The U-shaped network is connected in parallel with the Fourier neural operator network to obtain the U-FNO architecture, which is as follows: an input observation a(x) is lifted to a higher-dimensional feature space through a fully connected neural network, represented as Then, a series of iterative FNO and U-FNO layers are used, which can capture complex spatial dependencies in data. These layers are applied to a function sequence and Transformations are applied, operating on the channel dimension c. Finally, the architecture uses another fully connected neural network to project the high-dimensional output back to the original space, denoted as

[0018] In each U-FNO layer, an integral kernel transformation K with learnable parameters, a U-Net CNN operator U, and a linear operator W are applied. The transformation is followed by a nonlinear activation function σ, and the result is a highly expressive model that can approximate multiphase flow with less data requirements.

[0019] where the Transformer module with self-attention mechanism is used to refine the output pressure and saturation states and generate production sequences over time by capturing temporal dependencies.

[0020] where the U-FNO architecture with Transformer module with self-attention mechanism works as follows:

[0021] After obtaining the pressure and saturation state images, convolutional layers are used for feature extraction, converting the state images into low-dimensional latent variables; subsequently, these latent variables are processed by a Transformer encoder to reconstruct the production rate sequence; the Transformer encoder component has a multi-layer architecture consisting of N f Transformer encoder layers, each layer is good at processing and refining the hidden sequence features of the input data;

[0022] The Transformer encoder layer mechanism is based on residual connections, supplemented by normalization layers, followed by multi-head self-attention layers, and finally multi-layer perceptrons; multi-head self-attention layers are particularly effective in capturing complex dependencies within sequences by focusing on different positions in the input sequence in parallel; for a given sequence feature Z = (z1, z2,..., z n ), the model uses linear layers to derive the corresponding transformation matrices Q, K, and V.

[0023] where the overall performance analysis method is proposed, and the error measurement mechanism is introduced, and the specific way to complete the simulation is as follows:

[0024] The overall performance analysis method is proposed, and the error measurement mechanism is introduced, and the training loss and the corresponding validation set root mean square error of the surrogate model are analyzed, which can complete the simulation.

[0025] The application discloses a CCS fast simulation method based on a multi-feature fusion neural network. BRIEF DESCRIPTION OF DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0027] Figure 1 is the overall architecture of the proxy modeling of the present application.

[0028] Figure 2 is a three-dimensional configuration and well site layout diagram of a reservoir of an embodiment of the present application.

[0029] Figure 3 is a six-permeability sample diagram of a saline aquifer reservoir of an embodiment of the present application.

[0030] Figure 4 is six well control sequence samples of an embodiment of the present application.

[0031] Figure 5 is the training loss (MSE) and the corresponding RMSE on the validation set of an embodiment of the present application.

[0032] Figure 6 is the average percentage absolute error (MPAE) and the structural similarity index (SSIM) evaluation of all test samples of an embodiment of the present application.

[0033] Figure 7 is the predicted reservoir (third layer), pressure and gas saturation diagram of the 6th and 16th control time step of an embodiment of the present application.

[0034] Figure 8Pressure and gas saturation map predictions of the reservoir (sixth layer) at the 3rd, 6th and 16th control time steps of an embodiment of the present invention.

[0035] Figure 9 Comparison of the simulation results of cumulative CO2 injection and cumulative CO2 production with the predictions of the alternative model of an embodiment of the present invention.

[0036] Figure 10 Crossplot of the predicted cumulative CO2 production versus the true value in all test samples of an embodiment of the present invention.

[0037] Figure 11 R 2 box plot of the cumulative CO2 injection and cumulative CO2 production of an embodiment of the present invention.

[0038] Figure 12 Flow chart of the steps of the CCS fast simulation method based on multi-feature fusion neural network of the present invention.

[0039] Figure 13 Schematic diagram of the basic reservoir parameters of an embodiment of the present invention. DETAILED DESCRIPTION

[0040] Embodiments of the present invention are described in detail below with reference to the attached drawing figures, wherein the embodiments described below are exemplary and do not limit the present invention, as described above and below.

[0041] Reference is made to Figures 1-13 wherein Figure 1 the general architecture of the proxy modeling of the present invention, Figure 2 the three-dimensional configuration and well placement map of the reservoir of an embodiment of the present invention, Figure 3 the six permeability sample maps of the saline aquifer reservoir of an embodiment of the present invention, Figure 4 the six well control sequence samples of an embodiment of the present invention, Figure 5 the training loss (MSE) and the corresponding RMSE on the validation set of an embodiment of the present invention, Figure 6 the MPAE and SSIM evaluation of all test samples of an embodiment of the present invention, Figure 7 Pressure and gas saturation map predictions of the reservoir (sixth layer) at the 3rd, 6th and 16th control time steps of an embodiment of the present invention. Figure 8 Pressure and gas saturation map predictions of the reservoir (sixth layer) at the 3rd, 6th and 16th control time steps of an embodiment of the present invention. Figure 9 Comparison of the simulation results of cumulative CO2 injection and cumulative CO2 production with the predictions of the alternative model of an embodiment of the present invention. Figure 10is a cross plot of the predicted cumulative CO2 production and the actual value in all test samples of the embodiment of the present invention, Figure 11 is the cumulative CO2 injection amount and the cumulative CO2 production amount of the embodiment of the present invention. 2 box plot, Figure 12 This is a flowchart of the CCS fast simulation method based on multi-feature fusion neural network of the present invention. Figure 13 It is a basic reservoir parameter diagram of an embodiment of the present invention.

[0042] The present invention provides a CCS fast simulation method based on a multi-feature fusion neural network, comprising the following steps:

[0043] S100. Based on the determined simulation process, a U-FNO architecture suitable for CCS fast simulation is established, which can directly output pressure and saturation status in the time dimension;

[0044] For this specific implementation, the specific method of determining the simulation process is:

[0045] In the prediction of CO2 reservoir storage process under geological uncertainty, the injection and production well rates are adjusted in each reservoir model at each control time step; the simulation process can be expressed as:

[0046] X=[P,S,q i ,q p ]=g(m,w i ,w p )

[0047] Among them, the geological model is represented by m, which is An element of N, where nb is defined as nx×ny×nz, which represents the total number of grid blocks in the geological model, and nx, ny, and nz correspond to the number of blocks on each coordinate axis respectively; N m Indicates the number of samples in the geological model. i Yes An element, representing the well control sequence of the injection well, ni represents the number of injection wells, N c Indicates the number of control time steps; N w Indicates the number of well control sequences; w p yes An element of , which represents the well control sequence of the production well, np represents the number of production wells, u t Indicates the start-up time step of the production well.

[0048] The U-FNO architecture is as follows Figure 1 As shown in ①. The U-FNO approximation of the mapping includes the following steps:

[0049] The input observations a(x) are lifted to a higher-dimensional feature space by a fully connected neural network, denoted as Next, a series of iterative FNO and U-FNO layers are used, which are capable of capturing complex spatial dependencies in the data. These layers transform the function sequence and in the channel dimension c. Finally, the architecture uses another fully connected neural network to project the high-dimensional output back to the original space, denoted as In each U-FNO layer, a learnable parameterized integral kernel transformation K, a U-Net CNN operator U, and a linear operator W are applied. The transformation is followed by a nonlinear activation function σ, resulting in a highly expressive model that can approximate multiphase flow with fewer data requirements.

[0050] S200, based on the U-FNO architecture, a Transformer module with self-attention mechanism is introduced;

[0051] For this specific implementation, a Transformer module with self-attention mechanism is added, which refines these states and generates a production sequence over time by capturing temporal dependencies;

[0052] In this implementation, a Transformer module with self-attention mechanism is added, whose architecture can be seen in Figure 1 ②, where the specific process is:

[0053] After obtaining the pressure and saturation state images, convolutional layers are used for feature extraction, converting the state images into low-dimensional latent variables. Subsequently, these latent variables are processed by a Transformer encoder to reconstruct the production rate sequence. The Transformer encoder component has a multi-layer architecture consisting of N f Transformer encoder layers, each specialized in handling and refining the hidden sequence features of the input data.

[0054] The Transformer encoder layer mechanism is based on residual connections, supplemented by normalization layers, followed by a multi-head self-attention (MHSA) layer, and finally a multi-layer perceptron (MLP). The MHSA layer is particularly effective in capturing complex dependencies within the sequence by simultaneously focusing on different positions in the input sequence. For a given sequence feature Z = (z1, z2,..., z n ), the model uses linear layers to derive the corresponding transformation matrices Q, K, and V. This approach is crucial for efficiently processing sequence features and significantly improves the overall performance of the model.

[0055] S300, select the permeability and well control sequence as the model input, use SGeMS to perform sequential Gaussian simulation to randomly generate 10 permeability samples, and randomly generate 500 well control sequences for model training;

[0056] For this specific embodiment, in the present application, 10 permeability realizations are randomly generated using SGeMS (Stanford Geostatistical Modeling Software) for sequential Gaussian simulation. And 500 well control sequences are randomly generated.

[0057] S400, finally propose a general performance analysis method, introduce an error measurement mechanism, and complete the simulation.

[0058] For this specific embodiment, a general performance analysis method is proposed, and an error measurement mechanism is introduced to analyze the training loss and the corresponding validation set root mean square error (RMSE) of the surrogate model.

[0059] As an example, a three-dimensional (3D) actual saline reservoir is simulated and analyzed for CO2 storage:

[0060] The model used in this example represents a three-dimensional (3D) actual saline reservoir, providing a realistic case for simulation and analysis of CO2 storage. The reservoir domain is discretized into a Cartesian grid, divided into 101x135x6 grid cells along the x, y, and z axes, respectively, with each grid cell having a size of 40m x 40m in the horizontal plane. The top two layers of the model are caprock, providing containment for the CO2 storage process, while the lower four layers are designated as saline reservoirs.

[0061] The reservoir model contains a total of 81,810 grid cells, and is provided with four injection wells and one production well, each of which penetrates the lower four layers vertically. The three-dimensional configuration of the reservoir and the well placement are shown in Figure 2 , and the basic reservoir parameters are detailed in Table 1.

[0062] Considering permeability and well control sequence as model input, permeability samples are prepared for the surrogate model. In this example, 10 permeability realizations are randomly generated using SGeMS (Stanford Geostatistical Modeling Software) for sequential Gaussian simulation. Figure 3 reflecting the uncertainty of the geological model.

[0063] In addition, 500 well control sequences are randomly generated in this example. Specifically, the well control sampling range of the injection wells is set to 200,000m 3 / day to 1,000,000m 3between 0 and 16 control time steps, and well controls are sampled between 40 MPa and 45 MPa. Subsequently, for each model implementation, 500 simulations are performed using the generated 500 sets of well control sequences, resulting in a total of 5,000 samples. These constitute the training dataset. Figure 4 Six representative well control sequence samples are shown. Pressure and saturation maps, as well as cumulative CO2 injection and CO2 production from the production well, are calculated using the reservoir numerical simulator CMG-GEM. Simulation results for all samples are generated and saved at each time step. The simulation time step is set to 180 days, and the total simulation time is 180 days multiplied by 16. Well controls are adjusted every 180 days, and there are a total of 16 control steps throughout the simulation period. These datasets are then used to build and train proxy models for each layer of the CO2 saline reservoir, resulting in 14 proxy models. The 14 models are trained in parallel, with each model undergoing 50 epochs, a learning rate of 0.001, and a batch size of 8. Training is performed on eight NVIDIA Tesla A800 GPUs, each with 80 GB of memory. The entire training process takes a total of 625.5 seconds.

[0064] Figure 5 The training loss and root mean square error (RMSE) of the surrogate models for each layer are shown. The first and second layers are part of the cap rock, while the third to sixth layers constitute the reservoir. Notably, the second layer, which is part of the cap rock and is connected to the third layer of the reservoir, exhibits higher training loss and RMSE due to significant differences in geological parameters between the two layers, indicating poorer training performance and generalization ability compared to other layers. This suggests that sudden changes in geological properties at the interface between the cap rock and the reservoir can pose challenges to the effective learning of the model.

[0065] The generalization performance of the models on the test set is also evaluated. Figure 6 The mean percentage absolute error (MAPE) and structural similarity index (SSIM) evaluation results for the entire test set are shown, with the gray area representing all samples. The second, fourth, and sixth layers of the test cases are selected for display. Additionally, the R Figure 11 2 In the box plots, the SSIM values for pressure and saturation are greater than 0.95 and 0.987, respectively, and the MAPE values are less than 0.04 and 0.025, respectively, for all samples. These evaluation results indicate that the errors for most samples remain within 0.06, demonstrating that the reconstruction of image features is suitable for actual CO2 storage prediction applications.

[0066] ​Meanwhile, the present application also expands the investigation range, including the prediction ability of pressure and saturation distribution in 3D saline aquifer reservoirs after CO2 storage. In order to evaluate the performance of the model, the simulation results are compared with the prediction results of the model under various random well control sequences and random permeability realizations. Figure 7 The pressure and saturation field of the caprock reservoir (the third layer) after CO2 storage is shown, which indicates that the model can accurately depict the migration plume state from the CO2 reservoir. Meanwhile, Figure 8 The CO2 plume state in the reservoir (the sixth layer) is shown, and the colored bars represent the absolute error of the pressure and saturation prediction of the entire reservoir, which is controlled within 0.05 MPa and 0.035, respectively.

[0067] Near the CO2 displacement front, the prediction error of saturation is greater than that in the internal area of the CO2 injection area, and the error distribution near the front is also affected by heterogeneity. The results clearly show that the prediction error of CO2 saturation of the model is mainly concentrated in the CO2 displacement front. This error distribution is a direct result of the training process using the MSE loss function. The MSE loss function amplifies the error penalty in areas with high saturation, and preferentially reduces the error in areas with high saturation in the CO2 injection area.

[0068] Figure 9 The CO2 injection and production volumes calculated by the surrogate model at 16 control time steps are shown. The dashed line in each subplot represents the calculation output of the full physical simulation, while the dotted line represents the prediction results of the surrogate model. It can be seen that the CO2 injection and production volumes of the wells simulated by the surrogate model are in good agreement with the results of the full physical simulation. However, it needs to be noted that compared with the smooth curve generated by the full physical process, the prediction results of the surrogate model show certain irregularities. The prediction difference between the surrogate model and the full physical simulation is due to the inherent characteristics of the deep learning network, which introduces different degrees of error in the prediction of pressure and saturation, resulting in uneven results in the prediction of CO2 injection and production volumes by the surrogate model.

[0069] The present application shows the cross plot of cumulative CO2 injection and production volumes of the surrogate model and the full physical simulation. Figure 10 The comparison results of all test cases at the 3rd, 6th, and 16th control time steps are shown. The surrogate model is basically consistent with the calculation output of the full physical model in terms of cumulative CO2 injection and production rate in the CO2 production well. However, as the cumulative CO2 injection rate increases, the difference between the surrogate model and the full physical model becomes more obvious. As CO2 injection proceeds, the increase in the number of active production wells in the test sample leads to significant changes in the reservoir pressure and saturation response. This increase in complexity makes it more difficult for the surrogate model to learn accurately, ultimately leading to an expanding prediction difference in cumulative CO2 injection and production volumes over time.

[0070] Figure 11 The determination coefficients (R 2 ) of the complete test samples are shown. In each box plot, the middle line represents the median of the determination coefficients, the bottom and top of the box correspond to the 25th and 75th percentiles, and the whiskers represent the minimum and maximum values observed. The median R 2 of the cumulative CO2 injection and CO2 production volumes are 0.972 and 0.98, respectively. Overall, the U-FNO + Transformer network can efficiently and accurately predict the pressure and saturation states, as well as accurately predict the cumulative CO2 injection and production volumes throughout the life cycle of the entire saline aquifer reservoir.

[0071] Experiments show that this method can efficiently predict production dynamics in the study of CCS problems.

[0072] When the method is used, the method constructs a U-FNO network that fuses a U neural network (U-Net) and a Fourier neural operator (FNO) to improve the prediction accuracy of pressure fields and saturation fields in the long time series of the CO2 geological storage process. At the same time, a production sequence prediction network based on transformer encoders (TE) is used, the output end of the U-FNO network is directly connected to the input end of the TE network, and direct prediction from geological parameters and well control curves to time series distribution of pressure and saturation fields and corresponding production curves is realized. The method effectively overcomes the problem that existing CCS simulation technologies are difficult to simultaneously process multiple feature inputs and outputs, improves the prediction efficiency of the influence of the CO2 storage rate on the safety of the reservoir, and can provide an efficient and reliable intelligent simulation tool for CCS engineering optimization and history matching.

[0073] The above only discloses a preferred embodiment of the present application, of course, cannot limit the scope of the rights of the present application, and those skilled in the art can understand that all or part of the above-mentioned embodiments are implemented, and equivalent changes made according to the claims of the present application still belong to the scope covered by the present application.

Claims

1. A CCS fast simulation method based on multi-feature fusion neural network, characterized in that: The steps include: Establish a U-FNO architecture suitable for CCS fast simulation and directly output pressure and saturation status in the time dimension; Based on the U-FNO architecture, a Transformer module with self-attention mechanism is introduced to obtain a fast simulation alternative model for building CCS; Permeability and well control sequences were selected as model inputs. SGeMS was used to perform sequential Gaussian simulation to randomly generate 10 permeability samples and 500 sets of well control sequences for model training. Finally, an overall performance analysis method is proposed, an error measurement mechanism is introduced, and the simulation is completed; When building a surrogate model: In the prediction of CO2 reservoir storage process under geological uncertainty, the well control of injection and production wells is adjusted in each reservoir model at each control time step; the mapping relationship required to achieve the relevant alternative models is expressed as: X=[P,S,q i ,q p ]=g(m,w i ,w p ) Among them, the geological model is represented by m, which is An element of N, where nb is defined as nx×ny×nz, which represents the total number of grid blocks in the geological model, and nx, ny, and nz correspond to the number of blocks on each coordinate axis respectively; N m Indicates the number of samples in the geological model; w i Yes An element, representing the well control sequence of the injection well, ni represents the number of injection wells, N c Indicates the number of control time steps; N w Indicates the number of well control sequences; w p yes An element of , which represents the well control sequence of the production well, np represents the number of production wells, u t represents the start-up time step of the production well; The specific method to establish a U-FNO architecture suitable for CCS fast simulation and direct output of pressure and saturation status in the time dimension is as follows: The U-type network is connected in parallel with the Fourier neural operator network to obtain the U-FNO architecture, which is as follows: the input observation a(x) is promoted to a higher-dimensional feature space through a fully connected neural network, which is expressed as Next, a series of iterative FNO and U-FNO layers are used, which can capture the complex spatial dependencies in the data; these layers are used to train the function sequence and Transform, operating on the channel dimension c; finally, the architecture uses another fully connected neural network to project the high-dimensional output back to the original space, represented as In each U-FNO layer, an integral kernel transformation K with learnable parameters, a U-Net CNN operator U, and a linear operator W are applied; this transformation is followed by a nonlinear activation function σ, resulting in a highly expressive model capable of approximating multiphase flows with less data requirements.

2. The CCS fast simulation method based on multi-feature fusion neural network according to claim 1 is characterized in that: The Transformer module with self-attention mechanism is used to refine the output pressure and saturation states and generate time-varying production sequences by capturing temporal dependencies.

3. The CCS fast simulation method based on multi-feature fusion neural network according to claim 2 is characterized in that: The specific working process of the U-FNO architecture after introducing the Transformer module with self-attention mechanism is as follows: After acquiring the pressure and saturation state images, the convolutional layer is used for feature extraction to convert the state images into low-dimensional latent variables; these latent variables are then processed by the Transformer encoder to reconstruct the production rate sequence; the Transformer encoder component has a multi-layer architecture consisting of N f It consists of 100 Transformer encoder layers, each of which is good at processing and refining the hidden sequence features of the input data; The Transformer encoder layer mechanism is based on residual connections, supplemented by normalization layers, followed by multi-head self-attention layers, and finally multi-layer perceptrons; the multi-head self-attention layer is particularly effective in capturing complex dependencies within the sequence by focusing on different positions in the input sequence in parallel; for a given sequence feature Z = (z1, z2, ..., z n ), the model uses linear layers to derive the corresponding transformation matrices Q, K, and V.

4. The CCS fast simulation method based on multi-feature fusion neural network according to claim 3 is characterized in that: An overall performance analysis method is proposed, an error measurement mechanism is introduced, and the specific method for completing the simulation is as follows: The simulation can be completed by proposing an overall performance analysis method, introducing an error measurement mechanism, and analyzing the training loss of the alternative model and the corresponding validation set root mean square error.

Images