Spatial Weather Observation and Simulation Fusion Method and Device Based on 3D Adversarial Network

Abstract

The present application discloses a method and device for fusing space weather observation and simulation based on a 3D adversarial network, relating to the technical field. The method includes: interacting with a space weather simulation database, and obtaining simulation data based on a preset selection constraint; performing adaptive grid division on the simulation data based on adaptive grid technology to obtain divided adaptive grids, and storing the divided adaptive grids in an associated manner with the simulation data; obtaining observation data, and performing grid matching of the observation data based on the divided adaptive grids and the simulation data; inputting the observation data and the simulation data after grid matching into a 3D GAN generator to obtain generated data, wherein the generated data is marked with gradient information; activating a 3D GAN discriminator, performing data discrimination based on the generated data, and outputting fused data based on the data discrimination result. Thus, the technical effects of improving the data fusion accuracy and enhancing the fusion adaptability are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field, and particularly to a method and device for fusing space weather observation and simulation based on a 3D adversarial network. Background Art

[0002] The observation and simulation of space weather are crucial for understanding and predicting the dynamic processes in the Earth's magnetosphere and ionosphere. There are complex interactions between these layers and differences between observation and simulation data. The existing fusion of observation data and simulation data usually adopts methods of data assimilation or machine learning. The data assimilation method is limited by model parameterization and the uncertainty of observation data, and its applicability in complex space weather systems is not good. The machine learning method requires a large amount of training data and has high requirements for the generalization ability and robustness of the model. The existing technologies have technical problems of low data fusion accuracy and poor fusion adaptability. Summary of the Invention

[0003] The purpose of this application is to provide a method and device for fusing space weather observation and simulation based on a 3D adversarial network, so as to solve the technical problems of low data fusion accuracy and poor fusion adaptability in the existing technology.

[0004] In view of the above technical problems, this application provides a method and device for fusing space weather observation and simulation based on a 3D adversarial network.

[0005] In the first aspect, this application provides a method for fusing space weather observation and simulation based on a 3D adversarial network. The method includes:

[0006] Interact with the space weather simulation database, and obtain simulation data based on a preset selection constraint;

[0007] Based on the adaptive grid technology, perform adaptive grid division on the simulation data to obtain divided adaptive grids, and store the divided adaptive grids in an associated manner with the simulation data;

[0008] Obtain observation data, and perform grid matching of the observation data based on the divided adaptive grids and the simulation data;

[0009] Input the observation data and the simulation data after grid matching into a 3D GAN generator to obtain generated data, where the generated data is marked with gradient information;

[0010] Activate the 3D GAN discriminator, perform data discrimination based on the generated data, and output fused data based on the data discrimination result.

[0011] In the second aspect, this application also provides a device for fusing space weather observation and simulation based on a 3D adversarial network. The device includes:

[0012] A data acquisition module, which is used to interact with a space weather simulation database and obtain simulation data based on preset selection constraints;

[0013] An adaptive mesh generation module, which is used to perform adaptive mesh generation on the simulation data based on adaptive mesh technology to obtain generated adaptive meshes, and the generated adaptive meshes are stored in an associated manner with the simulation data;

[0014] A mesh matching module, which is used to obtain observation data and perform mesh matching of the observation data based on the generated adaptive meshes and the simulation data;

[0015] A generation module, which is used to input the observation data and the simulation data after mesh matching into a 3D GAN generator to obtain generated data, wherein the generated data is marked with gradient information;

[0016] A fusion output module, which is used to activate a 3D GAN discriminator, perform data discrimination based on the generated data, and output fusion data based on the data discrimination result.

[0017] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0018] By interacting with a space weather simulation database, simulation data is obtained and screened according to preset constraints. Adaptive mesh technology is used to perform mesh generation on the simulation data, and the generated meshes are stored in an associated manner with the simulation data. Observation data is obtained and matched with the generated meshes and the simulation data. The observation data and the simulation data after matching are input into a 3D GAN generator to generate data with gradient information. The 3D GAN discriminator is activated to perform discrimination on the generated data, and fusion data is output according to the discrimination result. Thus, the technical effects of improving data fusion accuracy and improving fusion adaptability are achieved.

[0019] The above description is only an overview of the technical solutions of this application. In order to be able to more clearly illustrate the technical means of this application, it can be implemented in accordance with the content of the description. And in order to make the above and other purposes, features and advantages of this application more obvious and understandable, the following specifically gives the specific implementation manners of this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Embodiments of the present invention and the following brief description are described in conjunction with the drawings. The drawings are described as follows:

[0021] Figure 1 It is a schematic flowchart of a space weather observation and simulation fusion method based on a 3D adversarial network of this application;

[0022] Figure 2 This is a schematic structural diagram of the space weather observation and simulation fusion device based on a 3D adversarial network for this application.

[0023] Explanation of reference numerals: data acquisition module 11, adaptive grid division module 12, grid matching module 13, generation module 14, fusion output module 15. Specific implementation manners

[0024] This application provides a space weather observation and simulation fusion method and device based on a 3D adversarial network, solving the technical problems of low data fusion accuracy and poor fusion adaptability faced by the prior art.

[0025] The overall idea adopted by the solution in this technical embodiment to solve the above problems is as follows:

[0026] First, interact with the space weather simulation database, and obtain simulation data according to the preset selection constraints. Then, use the adaptive grid technology to perform adaptive grid division on these simulation data, obtain the divided adaptive grid, and store the divided adaptive grid and the simulation data in an associated manner. Next, obtain the observation data, and perform grid matching of the observation data according to the divided adaptive grid and the simulation data. Then, input the observation data and simulation data after grid matching into the 3D GAN generator to generate data with gradient information. Finally, activate the 3DGAN discriminator, perform data discrimination on the generated data, and output the fusion data according to the discrimination result.

[0027] To better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings of the specification and specific implementation manners. It should be noted that the described embodiments are only a part of the embodiments of this application, rather than all the embodiments of this application. It should be understood that this application is not limited by the example embodiments described here. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention. Additionally, it should be noted that for the sake of description, only the parts related to the present invention are shown in the drawings rather than all of them.

[0028] Embodiment 1

[0029] As Figure 1 shown, this application provides a space weather observation and simulation fusion method based on a 3D adversarial network, and the method includes:

[0030] Interact with the space weather simulation database, and obtain simulation data based on the preset selection constraints;

[0031] Optionally, interact with the database based on the user interface or application, and obtain the required data through the database language and the corresponding database interface. Among them, the space weather simulation database refers to a structured data set storing space weather simulation data. The preset selection constraint refers to the data selection criteria set according to specific requirements or conditions. That is to say, the selection constraint stipulates the range of the selected simulation data.

[0032] In some embodiments, interacting with the space weather simulation database and obtaining simulation data based on the preset selection constraint includes:

[0033] Obtain historical space weather simulation records and construct a space weather simulation database;

[0034] Interact with the target fusion scenario to obtain the target scenario requirements. Among them, the target scenario requirements include memory requirements, quality requirements, and richness requirements;

[0035] Preset the selection constraint according to the memory requirements of the target scenario;

[0036] Using the selection constraint as the call index, traverse the space weather simulation database to obtain simulation data.

[0037] Specifically, collect historical space weather simulation records. These records can come from previously conducted weather simulation experiments, meteorological observation data, climate model outputs, etc. Organize and process the collected historical simulation records to ensure the accuracy and integrity of the data. Then, standardize and format the data to ensure that the data can be correctly read and processed by the database, including the definition of data fields, unit unification, etc. Next, establish a space weather simulation database based on the organized and formatted data. Optionally, the database is stored and managed using a relational database management system (such as MySQL, PostgreSQL, etc.) or a NoSQL database (such as MongoDB, Cassandra, etc.).

[0038] Through the above steps, a database containing historical space weather simulation records can be constructed, providing basic support for subsequent simulation data acquisition and application.

[0039] Optionally, obtain the requirements of the target fusion scenario. These requirements include memory requirements, quality requirements, and richness requirements. Among them, the memory requirement refers to the time window requirement that the simulation data needs to meet. The quality requirement means that the simulation data needs to have a certain degree of accuracy and credibility. The richness requirement means that the simulation data needs to contain diversity and details to meet the application requirements in different scenarios.

[0040] Optionally, when setting the preset selection constraints, determine the selection time range, spatial range, data granularity, etc. according to the memory requirements of the target scenario. For example, if the simulated data is used for comparison or analysis with historical data, the selected data should cover the relevant information in the historical records. At the same time, quality requirements and richness requirements also need to be considered to ensure that the simulated data has high quality and richness. The preset selection constraints will be used to filter and retrieve data in the space weather simulation database.

[0041] Furthermore, based on the preset selection constraints, obtain the corresponding simulated data from the database. By this method, data that meets the preset selection constraints can be retrieved from the space weather simulation database according to the specific needs of users to meet the requirements of space weather simulation. This method based on interaction and preset selection constraints can improve the accuracy and efficiency of space weather simulation and contribute to optimizing the application of space weather simulation.

[0042] Based on the adaptive grid technology, perform adaptive grid partitioning on the simulated data to obtain partitioned adaptive grids, and the partitioned adaptive grids are stored in an associated manner with the simulated data;

[0043] Optionally, through the adaptive grid technology, dynamically adjust the grid size and density according to data characteristics and requirements to better represent the characteristics and structure of the data, thereby improving both the data representation efficiency and the calculation efficiency. Among them, the grid partitioning can be adjusted based on factors such as the spatial distribution and feature changes of the data.

[0044] In addition, store the partitioned adaptive grids and the simulated data in an associated manner to ensure that each grid is associated with the corresponding data information. This can quickly access and process data when needed, while maintaining the structural integrity and accuracy of the data.

[0045] In some embodiments, before performing adaptive grid partitioning on the simulated data based on the adaptive grid technology to obtain partitioned adaptive grids, it includes

[0046] Interact with the target fusion scenario to obtain fusion comprehensive performance information, where the fusion comprehensive performance information includes computing power information, memory bandwidth information, and memory latency information;

[0047] Based on the fusion comprehensive performance information, perform a fusion computing power evaluation to obtain a fusion computing power index;

[0048] According to the fusion computing power index, combined with the timeliness requirements of the target fusion scenario, initialize the adaptive grid partitioning constraints.

[0049] Optionally, before performing adaptive mesh division on the simulation data and obtaining the divided adaptive mesh, first interact with the target fusion scenario to obtain the integrated fusion performance information. This integrated fusion performance information includes computing power information, memory access bandwidth information, and memory access latency information, which reflect the system's ability to process simulation data. Specifically, the computing power information is used to reflect the computing ability of the target fusion scenario, including the speed of the CPU, the capabilities of the GPU, the ability of parallel processing, etc.; the memory access bandwidth information refers to the speed of the system memory in the target fusion scenario, which can reflect the performance of the system memory; the memory access latency information can reflect the transmission time of data from the memory to the processor. The computing power information, memory access bandwidth information, and memory access latency information of the target fusion scenario jointly affect the ability and efficiency of the target fusion scenario to process the mesh division task.

[0050] Specifically, the integrated computing power evaluation is a quantitative evaluation of the performance of the fusion system or algorithm. During the evaluation process, various performance indicators need to be comprehensively considered to obtain the integrated computing power index to reflect the comprehensive performance level of the system or algorithm. Exemplarily, using the weighted average method, an empirical formula based on historical operation data, or other comprehensive evaluation methods, various indicators are integrated into a comprehensive performance evaluation value to obtain the integrated computing power index.

[0051] Among them, the timeliness requirement of the target fusion scenario refers to the time requirement for processing simulation data in the target scenario, that is, how fast the simulation data needs to be processed. The adaptive mesh division constraint is the division rule of the adaptive mesh division set according to the integrated computing power index and the timeliness requirement. Exemplarily, this adaptive mesh division constraint includes the mesh level and the corresponding mesh size. Preferably, it is represented as a mesh grading matrix, and each element of the matrix represents the mesh size of the corresponding mesh level. In actual applications, according to the current computing requirements and algorithm status, by querying the mesh grading matrix, determine which mesh level and the corresponding mesh size should be adopted, so as to perform adaptive mesh division.

[0052] In some embodiments, based on the adaptive mesh technology, performing adaptive mesh division on the simulation data to obtain the divided adaptive mesh includes

[0053] Based on the geodetic coordinate system, combined with the target space, construct a mesh coordinate system;

[0054] According to the mesh coordinate system, perform coordinate compression respectively to obtain a multi-dimensional compressed coordinate system;

[0055] Based on the multi-dimensional compressed coordinate system, perform parallel adaptive mesh division to obtain a multi-dimensional divided mesh;

[0056] Interact with the target fusion scenario to obtain the dimension weights, and fuse the multi-dimensional divided mesh according to the dimension weights to obtain the divided adaptive mesh.

[0057] Among them, the geodetic coordinate system refers to a coordinate system for positioning on the Earth's surface through longitude, latitude, and altitude coordinates. The target space refers to a specific spatial range set by the user, that is, the area where grid division needs to be performed. Furthermore, it means combining the geodetic coordinate system with the target space to construct a grid coordinate system suitable for dividing the target space. Exemplarily, based on the geodetic coordinate system, a certain point in the target space is defined as the base point, and the relative coordinates of the spatial points in the target space with respect to the base point are used as coordinate values to construct the grid coordinate system.

[0058] Optionally, coordinate compression is performed on the grid coordinate system to simplify the calculation amount. The compression method includes mapping the coordinates in the grid coordinate system to a two-dimensional coordinate plane along the coordinate axes respectively, so as to achieve dimensionality reduction compression mapping and compact representation, and obtain a multi-dimensional compressed coordinate system. Among them, the multi-dimensional compressed coordinate system is characterized by the dimensionality reduction data under multiple coordinate planes to compactly represent the original three-dimensional space.

[0059] Furthermore, based on the multi-dimensional compressed coordinate system, an adaptive grid division with parallel processing is performed. In other words, through parallel processing technology, based on the multi-dimensional compressed coordinate system, according to the above-mentioned adaptive grid division constraints, adaptive grid division is respectively performed based on the dimensionality reduction data under multiple coordinate planes to generate a multi-dimensional divided grid. Specifically, it includes dividing the grid into multiple sub-grids according to the coordinates of the points in the grid, and performing adaptive division on each sub-grid.

[0060] Specifically, the multi-dimensional divided grid reflects the grid division results from multiple analysis perspectives. Exemplarily, it includes the grid division results from the front view, side view, and top view angles.

[0061] Optionally, it refers to the weight assigned to each dimension of the multi-dimensional divided grid according to the requirements of the target fusion scenario. The dimension weight is determined based on the space weather observation and simulation tasks corresponding to the target fusion scenario. If the corresponding space weather observation and simulation tasks focus on the analysis of the altitude dimension, the dimension weight of the corresponding vertical dimension is relatively large. According to the dimension weight, the multi-dimensional divided grid is fused. It includes weighting each point in the grid according to the weight to obtain a divided adaptive grid.

[0062] Through this grid fusion method based on dimension weight, the processing efficiency and accuracy of the simulation data can be improved.

[0063] In some embodiments, based on the adaptive grid technology, performing adaptive grid division on the simulation data to obtain a divided adaptive grid further includes:

[0064] Analyzing the simulation data based on the principal component analysis method to obtain a deflection coordinate system;

[0065] Perform adaptive mesh generation according to the deflection coordinate system to obtain the deflected meshes.

[0066] Map the deflected meshes to the geodetic coordinate system to obtain the adaptive meshes.

[0067] Optionally, analyze the simulation data based on the principal component analysis method to obtain the main directions of data variation, i.e., the principal components. These principal components form a new coordinate system, namely the deflection coordinate system.

[0068] Furthermore, perform adaptive mesh generation according to the deflection coordinate system. This adaptive mesh generation is based on the same method principle as that for obtaining the multi-dimensional meshes above and is carried out based on the adaptive mesh generation constraints. For the sake of simplicity of the specification, it should be understood that for the simplicity of the specification, no further elaboration is made here.

[0069] Optionally, map the deflected meshes back to the geodetic coordinate system through the principal component vectors obtained by the principal component analysis to obtain the adaptive meshes. By this method, according to the characteristics and requirements of the simulation data, using the principal component analysis method and the adaptive mesh technology for mesh generation can improve the processing efficiency and accuracy of the simulation data.

[0070] Obtain the observed data and perform grid matching of the observed data based on the adaptive meshes and the simulation data;

[0071] Specifically, first, obtain real-time or historical observed data from sensors, monitoring devices, or other data sources; then, based on the divided adaptive meshes, perform grid matching on the observed data, align the geographical locations or spatial coordinates of the observed data with the adaptive meshes, so that the observed data and the simulation data have the same reference frame in space. Exemplarily, perform grid matching based on the RANSAC registration algorithm, the four-point congruent set registration algorithm (4-Point Congruent Set, 4PCS), the Super4PCS algorithm, etc.

[0072] Optionally, associate the observed data with the simulation data so that each grid cell has both observed data and simulation data.

[0073] The above process ensures the spatial consistency between the observed data and the simulation data, which is a key step for subsequent analysis, such as difference analysis or trend analysis. In this way, the relationship between the simulation data and the observed data can be compared and understood more accurately, thereby improving the accuracy and reliability of the analysis results.

[0074] Input the grid-matched observed data and the simulation data into a 3D GAN generator to obtain generated data, where the generated data is marked with gradient information;

[0075] In some embodiments, inputting the observed data and the simulated data after grid matching into a 3D GAN generator to obtain generated data includes:

[0076] Based on the preset simulation evaluation window, obtain historical simulation data;

[0077] Evaluating the historical simulation data to obtain historical simulation performance;

[0078] Calculating the transfer coefficient of the simulation data according to the historical simulation performance, and fusing the observed data with the simulation data using the transfer coefficient to obtain fused data;

[0079] The 3D GAN generator takes the fused data as input and outputs the generated data.

[0080] Optionally, before activating the 3D GAN generator to obtain generated data, first, data fusion of the observed data and the simulated data is performed to obtain fused data. The fused data includes the Hard-target part represented by the observed data and the Soft-target part represented by the simulated data. This provides a richer generation basis for the 3D GAN generator. Among them, the 3D GAN generator is a three-dimensional generative adversarial network (Generative Adversarial Network) used to generate high-quality three-dimensional data.

[0081] Specifically, the simulation evaluation window refers to the time window used to evaluate simulation data, such as the past week or the past 100 simulations. Historical simulation data refers to simulation data within the simulation evaluation window over a period of time in the past.

[0082] Optionally, the historical simulation data is evaluated to obtain the historical simulation performance. This performance includes various measurement dimensions, such as accuracy, precision, deviation, etc. Then, the transfer coefficient of the simulation data is calculated based on the historical simulation performance. Specifically, the better the historical simulation performance of the simulation data, the higher the transfer coefficient of the simulation data.

[0083] Optionally, the observed data and the simulated data are fused using a transfer coefficient to obtain fused data. The transfer coefficient is used to adjust the relationship between the two, and the fused data contains the characteristics of both the observed data and the simulated data. Exemplarily, the fusion method includes weighted evaluation fusion, that is, using the transfer coefficient to perform weighted averaging on the observed data and the simulated data to obtain fused data; probability replacement fusion, that is, according to the transfer coefficient, the features of the observed data or the simulated data are selected as part of the fused data with a certain probability. Through the above method, the characteristics of the observed data and the simulated data can be retained to varying degrees, so as to better fuse the two.

[0084] Optionally, the fused data is input into a 3D GAN generator to output generated data. The generated data is three-dimensional data similar to the actual observed data generated by the 3D GAN generator by taking the fused data as the input of random noise and learning the distribution of the data. Exemplarily, the 3D GAN generator consists of a multi-layer neural network, including a series of convolutional layers, fully connected layers or other types of layers (such as batch normalization layers, activation layers, etc.). Each layer has its own weights and biases, and these parameters are optimized during training to minimize the loss between the generator and the discriminator. Preferably, each layer in the multi-layer neural network is configured with batch normalization (BN) + rectified linear unit (ReLU) to improve the training speed and effect of the network.

[0085] In some implementations, the generated data is obtained, and the generated data is marked with gradient information, including:

[0086] Establish a multi-dimensional gradient space;

[0087] Define the gradient control granularity, calculate the data gradient of the generated data, and obtain a multi-dimensional gradient data set;

[0088] Perform normalization processing on the multi-dimensional gradient data set, and map the normalization processing result to the multi-dimensional gradient space to obtain a gradient grayscale image, and the gradient grayscale image is stored in an associated manner with the generated data.

[0089] Optionally, first, create a multi-dimensional gradient space, which is used to represent the respective feature dimensions of the data. That is, each dimension in the multi-dimensional gradient space represents an attribute or feature of the data.

[0090] Among them, the gradient control granularity refers to setting the granularity size of data gradient analysis, that is, the data range or precision considered when analyzing the data gradient. In other words, the gradient control granularity can be regarded as the step size or precision used when calculating the data gradient.

[0091] Optionally, calculate the data gradient of the generated data. The methods for gradient calculation include the bilateral difference method, the finite difference method, and the automatic differentiation method, etc. Exemplarily, for discrete data, the finite difference method is used to approximately calculate the gradient. For continuous data, an analytical method (if possible) or a numerical method is used to calculate the gradient.

[0092] Furthermore, perform normalization processing on the multi-dimensional gradient data set. Make the range of the data within a standard interval (such as 0 to 255), and then map the result of the normalization processing to the multi-dimensional gradient space to obtain a gradient grayscale image. Specifically, the data point coordinates of the gradient grayscale image correspond to the respective feature dimensions of the data, and the value of the data point characterizes the data gradient at that place.

[0093] The gradient grayscale image can provide a visual representation of the data gradient, which helps to understand the characteristics and patterns of the data. In addition, the gradient grayscale image is stored in association with the generated data for subsequent analysis and comparison.

[0094] Activate the 3D GAN discriminator, perform data discrimination based on the generated data, and output the fused data based on the data discrimination result.

[0095] Among them, the 3D GAN discriminator is used to distinguish the generated data output by the 3D GAN generator from the real data. Specifically, the 3D GAN discriminator performs data discrimination on the generated data based on gradient information, that is, the gradient grayscale image. The 3D GAN discriminator is obtained through supervised learning of the sample gradient information with discrimination marks.

[0096] Exemplarily, the gradient grayscale image is passed as input to the 3D GAN discriminator, the input gradient grayscale image is analyzed, and a value is output, which represents the probability that the 3D GAN discriminator believes the input data is real. Specifically, if the output value is close to 1, then the 3D GAN discriminator believes that the input data is very likely to be real; conversely, if the output value is close to 0, it means that the 3D GAN discriminator believes that the input data is very likely to be generated by the 3D GAN generator.

[0097] Optionally, according to the result of data discrimination, it is decided whether to accept the data generated by the generator. Exemplarily, based on a preset selection probability threshold, the discriminant output of the fused data is performed.

[0098] In summary, the space weather observation and simulation fusion method based on the 3D adversarial network provided by the present invention has the following technical effects:

[0099] Under the preset selection constraints, extract the simulation data from the interactive space weather simulation database. Then, by applying the adaptive grid technology, perform adaptive grid division on these simulation data, and jointly store the divided adaptive grid and the simulation data. Subsequently, obtain the observation data, and perform grid matching of the observation data based on the divided adaptive grid and the simulation data. Next, input the grid-matched observation data and simulation data into the generator of the three-dimensional generative adversarial network (3D GAN) to obtain the generated data marked with gradient information. Finally, activate the discriminator of the three-dimensional generative adversarial network, perform data discrimination according to the generated data, and output the fused data according to the discrimination result. Thus, the technical effects of improving the data fusion accuracy and enhancing the fusion adaptability are achieved.

[0100] Embodiment 2

[0101] Based on the same concept as the method for fusing space weather observation and simulation based on a 3D adversarial network in the above embodiments, as Figure 2 shown, the present application also provides a device for fusing space weather observation and simulation based on a 3D adversarial network, the device comprising:

[0102] A data acquisition module 11, configured to interact with a space weather simulation database and acquire simulation data based on a preset selection constraint;

[0103] An adaptive grid division module 12, configured to perform adaptive grid division on the simulation data based on adaptive grid technology to obtain divided adaptive grids, and the divided adaptive grids are stored in an associated manner with the simulation data;

[0104] A grid matching module 13, configured to acquire observation data and perform grid matching of the observation data based on the divided adaptive grids and the simulation data;

[0105] A generation module 14, configured to input the observation data and the simulation data after grid matching into a 3D GAN generator to obtain generated data, wherein the generated data is marked with gradient information;

[0106] A fusion output module 15, configured to activate a 3D GAN discriminator, perform data discrimination based on the generated data, and output fusion data based on the data discrimination result.

[0107] Among them, the data acquisition module 11 includes:

[0108] A historical data processing unit, configured to acquire historical space weather simulation records and construct a space weather simulation database;

[0109] A scenario requirement acquisition unit, configured to interact with a target fusion scenario and acquire target scenario requirements, wherein the target scenario requirements include memory requirements, quality requirements, and richness requirements;

[0110] A constraint preset unit, configured to preset a selection constraint based on the memory requirement of the target scenario;

[0111] An index call unit, configured to traverse the space weather simulation database with the selection constraint as a call index to obtain simulation data.

[0112] Further, the adaptive grid division module 12 further includes:

[0113] A performance information acquisition unit, configured to interact with a target fusion scenario and acquire integrated fusion performance information, wherein the integrated fusion performance information includes computing power information, memory access bandwidth information, and memory access latency information;

[0114] The fusion computing power evaluation unit is used to perform fusion computing power evaluation based on the fusion comprehensive performance information to obtain a fusion computing power index;

[0115] The constraint initialization unit is used to initialize the adaptive grid division constraint according to the fusion computing power index in combination with the timeliness requirements of the target fusion scenario.

[0116] Furthermore, the adaptive grid division module 12 further includes:

[0117] The coordinate system construction unit is used to construct a grid coordinate system based on the geodetic coordinate system in combination with the target space;

[0118] The coordinate compression unit is used to perform coordinate compression respectively according to the grid coordinate system to obtain a multi-dimensional compressed coordinate system;

[0119] The adaptive grid division unit is used to perform parallel adaptive grid division based on the multi-dimensional compressed coordinate system to obtain a multi-dimensional divided grid;

[0120] The grid fusion unit is used to interact with the target fusion scenario to obtain dimension weights, and fuse the multi-dimensional divided grids according to the dimension weights to obtain a divided adaptive grid.

[0121] Furthermore, the adaptive grid division module 12 further includes:

[0122] The dimensionality reduction analysis unit is used to analyze the simulation data based on the principal component analysis method to obtain a deflection coordinate system;

[0123] The deflection division unit is used to perform adaptive grid division according to the deflection coordinate system to obtain a deflection divided grid;

[0124] The grid mapping unit is used to map the deflection divided grid to the geodetic coordinate system to obtain a divided adaptive grid.

[0125] Furthermore, the generation module 14 further includes:

[0126] The historical simulation data acquisition unit is used to acquire historical simulation data based on a preset simulation evaluation window;

[0127] The simulation performance evaluation unit is used to evaluate the historical simulation data to obtain historical simulation performance;

[0128] The data fusion unit is used to calculate the transfer coefficient of the simulation data according to the historical simulation performance, and perform data fusion of the observation data and the simulation data with the transfer coefficient to obtain fused data;

[0129] The generated data output unit is used for the 3D GAN generator to take the fused data as input and output the generated data.

[0130] Furthermore, the generation module 14 further includes:

[0131] A gradient space construction unit for establishing a multi-dimensional gradient space;

[0132] A gradient calculation unit for defining a gradient control granularity, calculating the data gradient of the generated data, and obtaining a multi-dimensional gradient data set;

[0133] A normalization mapping unit for performing normalization processing on the multi-dimensional gradient data set and mapping the normalization processing result to the multi-dimensional gradient space to obtain a gradient grayscale image, and the gradient grayscale image is associated and stored with the generated data.

[0134] It should be understood that the key point of the embodiments mentioned in this specification lies in their differences from other embodiments. The specific embodiments in the foregoing Embodiment 1 are equally applicable to the space weather observation and simulation fusion device based on a 3D adversarial network described in Embodiment 2. For the sake of simplicity of the specification, no further elaboration will be made here.

[0135] It should be understood that the embodiments and the above descriptions disclosed in this application can enable those skilled in the art to implement this application using this application. At the same time, this application is not limited to this part of the embodiments mentioned above. Obvious modifications, combinations, and substitutions to the embodiments mentioned in this application also fall within the protection scope of this application.

Claims

1. A space weather observation and simulation fusion method based on 3D adversarial network, characterized in that: The method comprises: Interactive space weather simulation database, based on preset selection constraints, to obtain simulation data; Based on the adaptive grid technology, the simulation data is adaptively gridded to obtain the partitioned adaptive grid, and the partitioned adaptive grid is associated with the simulation data and stored; Acquire observation data, and perform grid matching of the observation data based on the divided adaptive grid and the simulation data; Inputting the observed data and the simulated data after grid matching into a 3D GAN generator to obtain generated data, wherein the generated data is marked with gradient information; activating a 3D GAN discriminator, performing data discrimination based on the generated data, and outputting fused data based on the data discrimination result; Wherein, based on the adaptive grid technology, the simulation data is adaptively gridded to obtain the adaptive grid, before that, including Interact with the target fusion scenario to obtain fusion comprehensive performance information, wherein the fusion comprehensive performance information includes computing power information, memory access bandwidth information, and memory access delay information; Based on the fusion comprehensive performance information, a fusion computing power evaluation is performed to obtain a fusion computing power index; Initialize adaptive grid division constraints according to the fusion computing energy index and the timeliness requirements of the target fusion scenario; Based on the adaptive grid technology, the simulation data is adaptively gridded to obtain the adaptive grid, including Based on the geodetic coordinate system and combined with the target space, a grid coordinate system is constructed; According to the grid coordinate system, coordinate compression is performed respectively to obtain a multi-dimensional compressed coordinate system; Performing parallel adaptive grid division based on the multi-dimensional compressed coordinate system to obtain a multi-dimensional divided grid; The interactive target fusion scene obtains dimension weights, and fuses the multi-dimensional partitioned grids according to the dimension weights to obtain a partitioned adaptive grid; The method further comprises: performing adaptive grid division on the simulation data based on the adaptive grid technology to obtain the divided adaptive grid; Analyze the simulation data based on principal component analysis to obtain a deflection coordinate system; According to the deflection coordinate system, adaptive grid division is performed to obtain a deflection division grid; Mapping the deflected grid to a geodetic coordinate system to obtain an adaptive grid; Among them, the interactive space weather simulation database obtains simulation data based on preset selection constraints, including: Obtain historical space weather simulation records and build a space weather simulation database; Interactive target fusion scenario, obtain target scenario requirements, wherein the target scenario requirements include memory requirements, quality requirements and richness requirements; wherein the memory requirement refers to the time window requirement that the simulation data needs to meet, the quality requirement refers to the accuracy and credibility of the simulation data, and the richness requirement refers to the diversity and details of the simulation data to meet the application requirements in different scenarios; Preset selection constraints based on the target scene memory requirements; The space weather simulation database is traversed using the selection constraint as a call index to obtain simulation data.

2. The method according to claim 1, characterized in that Inputting the observed data and the simulated data after grid matching into the 3D GAN generator to obtain generated data includes: Based on the preset simulation evaluation window, obtain historical simulation data; Evaluating the historical simulation data to obtain historical simulation performance; Calculating the transfer coefficient of the simulation data according to the historical simulation performance, and fusing the observed data with the simulation data using the transfer coefficient to obtain fused data; The 3D GAN generator takes the fused data as input and outputs the generated data.

3. The method according to claim 2, characterized in that Obtaining generated data, where the generated data is marked with gradient information, including: Establish a multi-dimensional gradient space; Defining a gradient control granularity, calculating a data gradient of the generated data, and obtaining a multi-dimensional gradient data set; The multidimensional gradient data set is normalized, and the normalization result is mapped to the multidimensional gradient space to obtain a gradient grayscale image, and the gradient grayscale image is stored in association with the generated data.

4. A space weather observation and simulation fusion device based on a 3D adversarial network, characterized in that: For executing the method according to any one of claims 1 to 3, the device comprises: A data acquisition module, the data acquisition module is used to interact with the space weather simulation database and acquire simulation data based on preset selection constraints; An adaptive grid division module, the adaptive grid division module is used to perform adaptive grid division on the simulation data based on the adaptive grid technology, obtain the divided adaptive grid, and the divided adaptive grid is associated with the simulation data and stored; A grid matching module, the grid matching module is used to obtain observation data and perform grid matching of the observation data based on the divided adaptive grid and the simulation data; A generation module, the generation module is used to input the observed data and the simulated data after grid matching into the 3DGAN generator to obtain generated data, wherein the generated data is marked with gradient information; A fusion output module is used to activate the 3D GAN discriminator, perform data discrimination based on the generated data, and output fusion data based on the data discrimination result.

Images