Method and device for predicting the effectiveness of a coronal mass ejection on earth

Abstract

The application discloses a method and device for predicting the effectiveness of a coronal mass ejection (CME) on the earth, and the method comprises the following steps: S1, acquiring a CME data set and an ICME data set; S2, processing the CME data set and the ICME data set to obtain a CME effectiveness on the earth data set; and S3, processing the CME effectiveness on the earth data set by using an effectiveness prediction model to obtain a CME effectiveness on the earth prediction result data. The CME data is processed by using the effectiveness prediction model to obtain the CME effectiveness on the earth prediction result data, so that the automatic classification prediction of the effectiveness on the earth is realized, and the missing report rate is reduced.

Description

Technical Field

[0001] This invention relates to the field of automatic detection of coronal mass ejections (CMEs), and more particularly to a method and apparatus for predicting the effectiveness of CMEs on Earth. Background Technology

[0002] Coronal mass ejections (CMEs) are explosive releases of large-scale magnetized plasma in the solar atmosphere. As a major driving force of space weather, their physical mechanisms deserve in-depth investigation.

[0003] These phenomena typically originate from magnetic reconnection events in solar active regions. When magnetic field energy is rapidly released through magnetic reconnection, it drives plasma to be ejected outward at speeds of 100-3000 km / s, with masses reaching enormous heights. - kg.

[0004] In terms of propagation path, the CME originates in the corona region approximately 2-30 solar radii from the Sun's surface, and then propagates outwards. In interplanetary space, its speed changes due to the influence of the solar wind, reaching its vicinity near Earth (approximately 1.5 × 10⁻⁶ solar radii). (km) It usually takes 1-5 days.

[0005] When the CME propagates toward Earth and eventually reaches it, it will compress the Earth's magnetosphere, triggering geomagnetic storms. The geomagnetic disturbance index Dst can drop below -200 nT, which will have a serious impact on critical infrastructure such as satellite communications, navigation systems, and power grids.

[0006] For satellite systems, geomagnetic storms accelerate orbital decay, high-energy particle radiation can damage satellite electronics, and ionospheric disturbances can affect satellite communication and navigation signals. In the power system sector, geomagnetic induced currents can damage transformers, causing economic losses of up to billions of dollars. Aviation safety is also affected; high-latitude routes are more susceptible to high-energy particle radiation, necessitating route adjustments or flight delays. Therefore, the importance of CME Earth effectiveness forecasting for critical infrastructure in modern society is self-evident. However, traditional CME Earth effectiveness forecasting methods have many limitations. Some methods rely on expert experience and rules, making simple judgments based on CME velocity, angular width, and position angle, such as assuming a position angle in the 270-90 degree range, a velocity exceeding 400 km / s, and an angular width greater than [missing information]. CME is more likely to be effective on the ground, but this method has limited accuracy and a high false negative rate; there is also the physical propagation model based on magnetohydrodynamics, which, although theoretically sound, has high computational cost, poor real-time performance, and is sensitive to initial conditions, resulting in high prediction uncertainty; statistical methods are mainly based on historical data analysis, which makes it difficult to capture complex nonlinear relationships and has limited generalization ability.

[0007] With the development of machine learning technology, data-driven prediction methods have brought new hope to CME land effectiveness forecasting. This method can automatically learn patterns from large amounts of historical data, capture complex feature relationships, and predict quickly after training, while continuously optimizing with data accumulation. This invention utilizes a machine learning-based land effectiveness prediction model to achieve automatic classification and prediction of land effectiveness by analyzing CME physical parameter characteristics and CME image features, thereby reducing the false negative rate. Summary of the Invention

[0008] This invention addresses the need for automatic detection of coronal mass ejections (CMEs) by proposing a method and apparatus for predicting the Earth-based effectiveness of CMEs. It utilizes an Earth-based effectiveness prediction model composed of a CME physical parameter feature model, a TimeSformer image sequence model, and a multimodal feature fusion model to process CME data. This organically integrates two heterogeneous information sources—CME physical parameter features and CME image features—to obtain CME Earth-based effectiveness prediction results. This achieves automatic classification and prediction of Earth-based effectiveness, improving prediction accuracy and robustness while reducing the false negative rate.

[0009] To address the aforementioned technical problems, a first aspect of this invention discloses a method for predicting the Earth-based effectiveness of coronal mass ejections, the method comprising:

[0010] S1, obtain the CME dataset and ICME dataset;

[0011] S2, process the CME dataset and the ICME dataset to obtain the CME ground-based effective dataset;

[0012] S3. Using the land effectiveness prediction model, the CME land effectiveness dataset is processed to obtain CME land effectiveness prediction result data.

[0013] As an optional implementation, in a first aspect of the present invention, processing the CME dataset and the ICME dataset to obtain a CME-based ground-effective dataset includes:

[0014] S21, The CME dataset is populated to obtain a preprocessed CME dataset;

[0015] S22, perform time matching processing on the preprocessed CME dataset and the ICME dataset to obtain the CME ground-based effective dataset.

[0016] As an optional implementation, in the first aspect of the present invention, the step of filling the CME dataset to obtain a preprocessed CME dataset includes:

[0017] S211, Traverse the CME dataset to obtain all CME data;

[0018] S212, parse and process any of the CME data to obtain the target data segment to be filled;

[0019] S213, Using the data filling model, the target data segment is filled to obtain preprocessed CME data;

[0020] The data filling model expression is:

[0021] ,

[0022] in, Indicates missing values ​​to be filled; Indicates the number of valid data points; Indicates the first [number] after ascending sort. One data point; This represents the outlier filtering coefficient; This represents the floor function;

[0023] S214, all the preprocessed CME data are combined in sequence to obtain a preprocessed CME dataset.

[0024] As an optional implementation, in a first aspect of the present invention, the step of performing time matching processing on the preprocessed CME dataset and the ICME dataset to obtain a CME-based ground-effective dataset includes:

[0025] S221, Traverse the ICME dataset to obtain all ICME data;

[0026] S222, parse and process any of the ICME data to obtain the LASCO-CME time;

[0027] S223, Traverse the preprocessed CME dataset to obtain all preprocessed CME data;

[0028] S224, Extract the file time information from the file name corresponding to any of the preprocessed CME data;

[0029] S225, based on the LASCO-CME time, match and process the file time information to obtain the CME tag identifier value corresponding to the preprocessed CME data;

[0030] S226, Based on the CME tag identification value, the preprocessed CME dataset is extracted to obtain the CME ground-based effective dataset.

[0031] As an optional implementation, in the first aspect of the present invention, the step of matching and processing the file time information based on the LASCO-CME time to obtain the CME tag identification value includes:

[0032] S2251, Calculate the difference between the time information and the LASCO-CME time to obtain the time difference;

[0033] S2252, determine whether the time difference is less than a set time difference threshold, and obtain a time difference judgment result;

[0034] S2253, when the time difference judgment result is yes, update the CME tag of the preprocessed CME data to 1, and obtain the CME tag identification value corresponding to the preprocessed CME data;

[0035] When the time difference determination result is negative, the CME label of the preprocessed CME data is updated to 0, and the CME label identification value corresponding to the preprocessed CME data is obtained.

[0036] As an optional implementation, in the first aspect of the present invention, the step of processing the CME land-effective dataset using a land-effective prediction model to obtain CME land-effective prediction result data includes:

[0037] S31, Construct an effective Earth prediction model;

[0038] S32, Based on the aforementioned effective land prediction model, the CME effective land dataset is processed to obtain CME effective land prediction result data.

[0039] As an optional implementation, in the first aspect of the present invention, the step of processing the CME land effectiveness dataset based on the land effectiveness prediction model to obtain CME land effectiveness prediction result data includes:

[0040] S321, preprocess the CME ground-based effective dataset to obtain a preprocessed CME ground-based effective dataset;

[0041] S322, according to a set ratio, the preprocessed CME ground-based effective dataset is divided into a training dataset, a test dataset, and a validation dataset;

[0042] S323, Using the training dataset and the test dataset, the effective ground prediction model is trained to obtain an optimized effective ground prediction model;

[0043] S324, Using the aforementioned land effectiveness prediction optimization model, the validation dataset is processed to obtain CME land effectiveness prediction result data.

[0044] The second aspect of this invention discloses a device for predicting the effectiveness of coronal mass ejections (CMEs) on Earth, the device comprising: a historical data acquisition module, a data extraction module, and an effectiveness processing module;

[0045] The historical data acquisition module is used to acquire CME datasets and ICME datasets;

[0046] The data extraction module is used to process the CME dataset and the ICME dataset to obtain the CME ground-based effective dataset;

[0047] The land effectiveness processing module is used to process the CME land effectiveness dataset using the land effectiveness prediction model to obtain CME land effectiveness prediction result data;

[0048] The historical data acquisition module, the data extraction module, and the ground effectiveness processing module are sequentially connected.

[0049] A third aspect of the present invention discloses another device for predicting the effectiveness of coronal mass ejections on Earth, the device comprising:

[0050] Memory containing executable program code;

[0051] A processor coupled to the memory;

[0052] The processor calls the executable program code stored in the memory to execute the method for predicting the effectiveness of coronal mass ejections to Earth disclosed in the first aspect of the present invention.

[0053] The fourth aspect of the present invention discloses a computer-readable storage medium storing computer instructions, which, when invoked, are used to execute the method for predicting the effectiveness of coronal mass ejections on Earth disclosed in the first aspect of the present invention.

[0054] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0055] This invention discloses a method and apparatus for predicting the Earth effectiveness of coronal mass ejections (CMEs). It utilizes an Earth effectiveness prediction model composed of a CME physical parameter feature model, a TimeSformer image sequence model, and a multimodal feature fusion model to process CME data. This organically integrates two heterogeneous information sources—CME physical parameter features and CME image features—to obtain CME Earth effectiveness prediction results. This achieves automatic classification and prediction of Earth effectiveness, improving prediction accuracy and robustness while reducing the false negative rate. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0057] Figure 1 This is a schematic diagram illustrating an application scenario of the method for predicting the effectiveness of coronal mass ejections on Earth disclosed in Embodiment 1 of the present invention.

[0058] Figure 2 This is a flowchart illustrating the method for predicting the effectiveness of coronal mass ejections on Earth as disclosed in Embodiment 1 of the present invention.

[0059] Figure 3 This is a schematic diagram of the confusion matrix of the method for predicting the effectiveness of coronal mass ejections on Earth disclosed in Embodiment 1 of the present invention;

[0060] Figure 4 This is a false alarm rate curve of the coronal mass ejection Earth-to-ground effectiveness prediction method disclosed in Embodiment 1 of the present invention;

[0061] Figure 5 This invention discloses a method for predicting the effectiveness of coronal mass ejections (CMEs) on Earth using CME image sequence data.

[0062] Figure 6 This is a schematic diagram of the structure of a coronal mass ejection Earth-based effectiveness prediction device disclosed in Embodiment 2 of the present invention;

[0063] Figure 7 This is a schematic diagram of another coronal mass ejection Earth-based effectiveness prediction device disclosed in Embodiment 3 of the present invention. Detailed Implementation

[0064] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0066] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0067] In this application, the term "exemplary" is used to mean "used as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0068] It should be noted that since the method in this application embodiment is executed in a computer device, the processing objects of each computer device exist in the form of data or information, such as time, which is essentially time information. It is understood that if size, quantity, position, etc. are mentioned in subsequent embodiments, they are all corresponding data that exist so that the computer device can process them. Specific details will not be elaborated here.

[0069] This application provides a method, apparatus, computer device, and computer-readable storage medium for predicting the effectiveness of coronal mass ejections on Earth, which will be described in detail below.

[0070] Please see Figure 1 , Figure 1This is a schematic diagram illustrating the application of the coronal mass ejection (CME) Earth-based effectiveness prediction device provided in this application within a space weather forecasting system. The space weather forecasting system may include a computer device 100, which integrates the CME Earth-based effectiveness prediction device. Figure 1 Computer equipment in the country.

[0071] In this embodiment, the computer device 100 can be a standalone server, a server network, or a server cluster. For example, the computer device 100 described in this embodiment includes, but is not limited to, a computer, a network host, a single network server, a set of multiple network servers, or a cloud server composed of multiple servers. The cloud server is composed of a large number of computers or network servers based on cloud computing.

[0072] It is understood that the computer device 100 used in the embodiments of this application can be a device that includes both receiving and transmitting hardware, that is, a device having receiving and transmitting hardware capable of performing bidirectional communication on a bidirectional communication link. Such a device may include: cellular or other communication devices having a single-line display, a multi-line display, or a cellular or other communication device without a multi-line display. Specifically, the computer device 100 may be a desktop terminal or a mobile terminal, and may also be one of a mobile phone, tablet computer, laptop computer, etc.

[0073] Those skilled in the art will understand that Figure 1 The application environment shown is merely one application scenario of the solution in this application and does not constitute a limitation on the application scenario of the solution in this application. Other application environments may include those that are more specific to this application. Figure 1 The number of computer devices shown is more or less, for example Figure 1 Only one computer device is shown in the diagram. It is understood that the system may also include one or more other services, which are not limited here.

[0074] In addition, such as Figure 1 As shown, the space weather forecasting system may also include a memory 200 for storing sensor-acquired data, CME historical data, and processed result data.

[0075] It should be noted that, Figure 1The schematic diagram of the application scenario of the coronal mass ejection (CME) Earth effectiveness prediction device shown is merely an example. The CME Earth effectiveness prediction device and scenario described in the embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided in the embodiments of this application. As those skilled in the art will know, with the evolution of space weather forecasting systems and the emergence of new business scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0076] This invention discloses a method and apparatus for predicting the Earth-based effectiveness of coronal mass ejections (CMEs). It utilizes an Earth-based effectiveness prediction model composed of a CME physical parameter feature model, a TimeSformer image sequence model, and a multimodal feature fusion model to process CME data. This organically integrates two heterogeneous information sources—CME physical parameter features and CME image features—to obtain CME Earth-based effectiveness prediction results. This achieves automatic classification and prediction of Earth-based effectiveness, improving prediction accuracy and robustness while reducing the false negative rate. Detailed explanations follow.

[0077] Example 1

[0078] Please refer to Figure 2 In order to better understand the technical solution of the present invention. Figure 2 This is a flowchart illustrating a method for predicting the Earth-based effectiveness of coronal mass ejections (CMEs) according to an embodiment of the present invention. Figure 2 The described method for predicting the effectiveness of coronal mass ejections (CMEs) on Earth is applied to space weather forecasting systems, such as local servers or cloud servers used in space weather forecasting systems. This invention does not limit the application of this method. Figure 2 As shown, the method for predicting the Earth-based effectiveness of coronal mass ejections may include the following operations:

[0079] S1, obtain the CME dataset and ICME dataset;

[0080] It should be noted that in this embodiment, the CME dataset originates from the SOHO / LASCO-CME-Catalog. The observation instrument is the Coronagraph LASCO aboard the SOHO solar observation satellite launched in 1995, which includes two cameras, C2 and C3. C2 has an observation range of 2-6 solar radii and high resolution; C3 has an observation range of 3.7-30 solar radii and a wide field of view. The CME dataset spans from 2001 to 2023, covering two complete solar activity cycles, and contains 19,093 CME events. Each event is stored in a text file containing detailed observation parameters, including position parameters, velocity parameters, and dynamic parameters. The position parameters include the central position angle (PA), 0- This indicates the direction of the CME burst, approximately 270° from Earth's perspective. The range; angular width reflects the lateral spread angle of the CME; there is also the magnetic pressure anomaly (MPA), which indicates the most likely propagation direction of the CME; velocity parameters include linear velocity (average propagation velocity), initial velocity (velocity when first observed), final velocity (velocity when leaving the observation field of view), and velocity at a radius of 20 solar radii, among which the velocity at a radius of 20 solar radii is closer to the velocity in interplanetary space and is more valuable for prediction; dynamic parameters include acceleration (rate of change of velocity), estimated mass and kinetic energy, etc. Other parameters include event mass rating and altitude range, etc.

[0081] It should be noted that the ICME dataset is an observational dataset of interplanetary coronal mass ejections (ICMEs), recording various key parameters of ICMEs in near-Earth space since January 1996. This dataset contains 606 ICME events, spanning from 1996 to 2023, and includes information such as the ICME arrival time at Earth, the corresponding LASCO-CME time (if known), and the geomagnetic disturbance index (DST). Approximately 70% of the ICME records contain corresponding LASCO-CME time information, and we correlated CMEs with ICMEs through time matching (with a tolerance of 30 minutes).

[0082] S2, process the CME dataset and the ICME dataset to obtain the CME ground-based effective dataset;

[0083] S3. Using the land effectiveness prediction model, the CME land effectiveness dataset is processed to obtain CME land effectiveness prediction result data.

[0084] Optionally, the processing of the CME dataset and the ICME dataset to obtain the CME-based effective dataset includes:

[0085] S21, The CME dataset is populated to obtain a preprocessed CME dataset;

[0086] S22, perform time matching processing on the preprocessed CME dataset and the ICME dataset to obtain the CME ground-based effective dataset.

[0087] Optionally, the step of filling the CME dataset to obtain a preprocessed CME dataset includes:

[0088] S211, Traverse the CME dataset to obtain all CME data;

[0089] S212, parse and process any of the CME data to obtain the target data segment to be filled;

[0090] It should be noted that approximately 32% of the CME dataset contains partially missing values, primarily in the PA and Mass fields.

[0091] It should be noted that, in this embodiment of the invention, the PA and Mass fields are extracted based on the CME data structure to obtain the target data segment for filling;

[0092] S213, Using the data filling model, the target data segment is filled to obtain preprocessed CME data;

[0093] The data filling model expression is:

[0094] ,

[0095] in, This indicates missing values ​​to be filled; n represents the number of valid data points. This represents the k-th data element after ascending sorting. This represents the outlier filtering coefficient; This represents the floor function;

[0096] It should be noted that the outlier filtering coefficient is... The value range is (0,1], with a default of 0.9; the floor function... Ensure that k is an integer position to avoid non-integer indices;

[0097] S214, all the preprocessed CME data are combined in sequence to obtain a preprocessed CME dataset.

[0098] Optionally, the step of performing time-matching processing on the preprocessed CME dataset and the ICME dataset to obtain a ground-based effective CME dataset includes:

[0099] S221, Traverse the ICME dataset to obtain all ICME data;

[0100] S222, parse and process any of the ICME data to obtain the LASCO-CME time;

[0101] S223, Traverse the preprocessed CME dataset to obtain all preprocessed CME data;

[0102] S224, Extract the file time information from the file name corresponding to any of the preprocessed CME data;

[0103] S225, based on the LASCO-CME time, match and process the file time information to obtain the CME tag identifier value corresponding to the preprocessed CME data;

[0104] S226, Based on the CME tag identification value, the preprocessed CME dataset is extracted to obtain the CME ground-based effective dataset;

[0105] It should be noted that the CME ground-based effective dataset includes several CME ground-based effective datasets;

[0106] It should be noted that the CME ground-based effective data is parsed and processed to obtain CME ground-based effective parameter data and CME ground-based effective image sequence data;

[0107] It should be noted that the effective ground parameters data of the CME include LASCO-CME time data, CME velocity data, CME acceleration data, CME angular width data, and CME orientation angle data;

[0108] It should be noted that, in this embodiment of the invention, the determination of Earth-based validity is mainly achieved through time matching of CME and ICME events. From a physical perspective, after a CME erupts from the solar surface, it travels through interplanetary space for 1-5 days to reach Earth. If it propagates towards Earth, it will form ICME data upon arrival. The ICME data records the corresponding LASCO-CME time, i.e., the time of the CME eruption. Therefore, if a CME event eventually reaches Earth and forms an ICME, its time should match the corresponding LASCO-CME time in the ICME record.

[0109] Optionally, the step of matching and processing the file time information based on the LASCO-CME time to obtain the CME tag identifier value includes:

[0110] S2251, Calculate the difference between the time information and the LASCO-CME time to obtain the time difference;

[0111] S2252, determine whether the time difference is less than a set time difference threshold, and obtain a time difference judgment result;

[0112] S2253, when the time difference judgment result is yes, update the CME tag of the preprocessed CME data to 1, and obtain the CME tag identification value corresponding to the preprocessed CME data;

[0113] When the time difference determination result is negative, the CME label of the preprocessed CME data is updated to 0, and the CME label identification value corresponding to the preprocessed CME data is obtained.

[0114] It should be noted that the ICME data contains corresponding LASCO-CME time information. By matching the time (with a tolerance of 30 minutes), the CME and ICME are associated to obtain the effective CME dataset.

[0115] It should be noted that the time is extracted from the filename of the CME data, and the difference between the time and the LASCO-CME time is calculated. If the time difference is ≤30 minutes, it is considered a match. In the case of one ICME corresponding to multiple CME candidates, the one with the closest time is selected and the CME label is updated to 1 (valid to ground). All unmatched CMEs are marked as 0 by default (invalid to ground).

[0116] It should be noted that, taking into account factors such as the time accuracy of LASCO observations, the duration of CME outbreaks, data recording errors, and balancing accuracy and recall, ±30 minutes was chosen as the time tolerance. This avoids missing true matches if the tolerance is too small, while potentially introducing false matches if it is too large.

[0117] It should be noted that, in this embodiment, among the 19,093 CME data, 141 (0.74%) were successfully matched and valid CME data were grounded, while 18,952 (99.26%) were not matched and invalid CME data were grounded. The imbalance ratio is approximately 1:134. Among the 141 matched CMEs, the average time difference is 8.5 minutes, the maximum is 28 minutes, and 95% of the matching time differences are less than 20 minutes.

[0118] It should be noted that, in this embodiment, the reasons for non-matching may include the LASCO-CME field in the ICME data being empty (approximately 30% of ICME data), the CME not propagating towards the Earth, the CME dissipating or changing direction during propagation, and inaccurate time recording exceeding the 30-minute tolerance.

[0119] Optionally, the step of processing the CME land efficiency dataset using the land efficiency prediction model to obtain CME land efficiency prediction result data includes:

[0120] S31, Construct an effective Earth prediction model;

[0121] S32, Based on the aforementioned effective land prediction model, the CME effective land dataset is processed to obtain CME effective land prediction result data.

[0122] Optionally, the effective prediction model for Earth includes a CME physical parameter feature model, a TimeSformer image sequence model, and a multimodal feature fusion model;

[0123] The CME physical parameter feature model is used to process the CME ground-based effective parameter data to obtain the first classification result data;

[0124] The TimeSformer image sequence model is used to process CME ground-based effective image sequence data to obtain second classification result data;

[0125] The multimodal feature fusion model is used to process the first classification result data and the second classification result data to obtain CME land effectiveness prediction result data;

[0126] It should be noted that the processing refers to multiplying the predicted probability given by the CME physical parameter feature model and the predicted probability given by the TimeSformer image sequence model, that is, multiplying the first classification result data and the second classification result data to obtain the class with the highest comprehensive probability as the class finally output by the multimodal feature fusion model, that is, the CME land effectiveness prediction result data.

[0127] The output of the CME physical parameter feature model and the output of the TimeSformer image sequence model are connected to the input of the data fusion model.

[0128] It should be noted that the resulting data fusion model organically integrates the results from two heterogeneous information sources: the CME physical parameter feature model and the TimeSformer image sequence model, thereby making a more accurate and robust prediction of Earth effectiveness than any single modality. We use a model probability weighted average method to obtain the multimodal fusion output result. The weighted average result of the CME physical parameter feature model and the TimeSformer image sequence model is the confidence level of the Earth effectiveness prediction model in predicting the Earth effectiveness of the CME event.

[0129] Optionally, the CME physical parameter feature model is the first effective prediction model for Earth.

[0130] It should be noted that the first effective prediction model for Earth is a logistic regression model;

[0131] The first effective prediction model for Earth includes: a first data input layer, a linear prediction layer, a mapping compression layer, a learning iteration layer, and a first data output layer;

[0132] The first data input layer is used to receive the CME ground-based effective parameter data and perform structured processing on the CME ground-based effective parameter data to obtain structured CME ground-based effective parameter data;

[0133] The linear prediction layer is used to perform linear superposition processing of the structured CME on the effective parameter data and model weights to obtain a linear score for the category.

[0134] The linear superposition processing expression is:

[0135] ,

[0136] in, This represents the linear score of the category for the i-th sample; Represents the intercept term; Indicates the first Each feature weight; Indicates the first One feature; Indicates the number of features;

[0137] The mapping compression layer is used to compress the linear score to obtain the category prediction probability;

[0138] The compression expression is:

[0139] ,

[0140] in, This represents the predicted class probability of the i-th sample; Represents the Sigmoid function;

[0141] The learning iteration layer is used to iteratively process the loss values ​​of the predicted class probability and the true label using a loss function and an optimizer to obtain the optimal weight values ​​of the model.

[0142] The loss function expression is as follows:

[0143] ,

[0144] in, Represents the global loss value; Indicates the number of samples; This represents the true label of the i-th sample;

[0145] The iterative processing expression is:

[0146] ,

[0147] in, Indicates the first Weights for each iteration; Indicates the learning rate; This indicates that the loss function is in gradient at;

[0148] It should be noted that by adding regularization This makes the weights tend to be smaller, avoiding excessive influence of a single feature on the model and preventing overfitting;

[0149] It should be noted that the optimizer mentioned is the ADM optimizer;

[0150] The first data output layer is used to transform the predicted probability of the category and output the category label;

[0151] The conversion processing expression is:

[0152] ,

[0153] in, Indicates the first i The final classification result of each sample, i.e., the category label; Indicates the decision threshold;

[0154] It should be noted that, in this embodiment, the decision threshold is set to 0.5;

[0155] The first data input layer, the linear prediction layer, the mapping compression layer, the learning iteration layer, and the first data output layer are sequentially connected in data.

[0156] Optionally, the CME physical parameter feature model is a second effective prediction model for Earth.

[0157] It should be noted that the second effective prediction model for Earth is a random forest model;

[0158] The second effective prediction model for Earth includes a second data input layer, a diverse generation layer, a basic prediction layer, a fusion processing layer, and a second data output layer.

[0159] The second data input layer is used to receive the CME's ground-based effective parameter data;

[0160] The diverse generation layer is used to perform differential processing on the CME ground-based effective parameter data to obtain a differential sample dataset;

[0161] The differential processing expression is:

[0162] ,

[0163] in, This represents the m-th differentiated sample data; Indicates the first hyperparameter; Indicates the first The local density of the preprocessed CME ground-based effective data; Indicates the first The label consistency value of the preprocessed CME ground-based valid data; This indicates the number of valid ground-to-ground data points of the preprocessed CME;

[0164] It should be noted that the first hyperparameter represents the sample weight balancing coefficient; the value range of the first hyperparameter is... ;

[0165] It should be noted that the local density refers to the first... The preprocessed CME ground-based effective data is extracted into a subset of the training samples; the first hyperparameter takes values ​​ranging from 1 to 10. ;

[0166] It should be noted that the diverse generation layer increases the sampling probability of "high-density, high-label consistency" samples and reduces the interference of noise on the tree, thereby improving the quality of the basic prediction unit; the basic prediction layer is used to build a model based on the decision tree, process the differential sample dataset, and obtain a differential decision tree set and a performance weight set.

[0167] The differentiated decision tree set includes several different decision trees;

[0168] The expression for the decision tree construction model is:

[0169] ,

[0170] ,

[0171] in, Indicates feature index; Indicates the first Gain improvement value for each feature; Indicates the second hyperparameter; Indicates the first Maximum redundancy of each feature; Represents a sample set; Represents a subset of samples; Indicates the first Probability of tree prediction; The category labels represent the decision tree; The weights represent the performance weights of the decision tree; This represents the prediction of the t-th decision tree. Category The probability of; Indicates category; Indicates the number of decision trees; Indicates the number of categories;

[0172] It should be noted that the second hyperparameter represents the feature redundancy penalty coefficient; the value range of the second hyperparameter is... ;

[0173] It should be noted that by using the decision tree to build the model, the feature combination of each tree is made more diverse through the maximum feature redundancy; at the same time, the high-performance tree is given higher weight, making the "integration and fusion module" more inclined to high-quality prediction results, thus improving the reliability of the final classification.

[0174] The fusion processing layer is used to aggregate the differential decision tree set and the performance weight set to obtain the first classification result data;

[0175] It should be noted that the aggregation process refers to the statistically weighted category probabilities, selecting the category with the highest probability as the final result to obtain the first classification result data;

[0176] The second data output module is used to output the first classification result data;

[0177] The second data input layer, the diverse generation layer, the basic prediction layer, the fusion processing layer, and the second data output layer are sequentially connected;

[0178] Optionally, the TimeSformer image sequence model includes a third data input layer, an image segmentation layer, an image encoding layer, a feature extraction layer, and a third data output layer; the TimeSformer image sequence model is used to process the CME ground-based effective image sequence data to obtain second classification result data;

[0179] The third data input layer is used to input the CME ground-based effective image sequence data of a specified size;

[0180] It should be noted that the specified size includes batch size, number of RGB channels, number of time frames, image height, and image width;

[0181] It should be noted that in this embodiment, the batch size is 64, the number of RGB channels is 3, the number of time frames is 8, the image height is 512, and the image width is 512.

[0182] The image segmentation layer is used to process the CME ground-based effective image sequence data of a specified size to obtain an image segmentation vector set and an image segmentation block set;

[0183] It should be noted that the specified size is 512×512;

[0184] It should be noted that the aforementioned processing means dividing each 512×512 image frame into 1024 16×16 image segmentation blocks, then flattening each image segmentation block and projecting it onto a 768-dimensional embedding vector through a linear layer to obtain an image segmentation vector set.

[0185] The image coding layer is used to perform spatial and temporal location coding on the image segmentation vector set to obtain a coded image segmentation vector set.

[0186] It should be noted that performing spatial and temporal location encoding on the image segmentation block set means that, in order to preserve the position of the image segmentation block and the temporal information of the frame, learnable spatial and temporal location encodings are added to all the image segmentation blocks to obtain the encoded image segmentation block set.

[0187] It should be noted that the spatial location encoding means that different values ​​are added according to different spatial locations, and the location information is injected into a preset data structure to obtain spatial location encoded data;

[0188] The time-location encoding means arranging the spatial location encoding data at the same time in spatial order to obtain a first spatial sequence; arranging the first spatial sequence in spatial order to obtain a time-space sequence; and encoding the time-space sequence to obtain time-location encoding data.

[0189] The feature extraction layer uses a 12-layer Transformer-Encoder to process the encoded image segmentation vector set to obtain image classification result data;

[0190] It should be noted that the feature extraction layer is a data processing layer within the 12 layers that uses the divided_space_time attention mechanism;

[0191] The third data output layer is used to map the image classification result data to obtain the second classification result data;

[0192] It should be noted that the mapping process of the image classification result data means extracting the 768-dimensional output of the image classification result data and mapping it to 2-dimensional second classification result data through a fully connected layer;

[0193] The third data input layer, the image segmentation layer, the image encoding layer, the feature extraction layer, and the third data output layer are sequentially connected.

[0194] Optionally, the step of processing the CME land effectiveness dataset based on the land effectiveness prediction model to obtain CME land effectiveness prediction result data includes:

[0195] S321, preprocess the CME ground-based effective dataset to obtain a preprocessed CME ground-based effective dataset;

[0196] S322, according to a set ratio, the preprocessed CME ground-based effective dataset is divided into a training dataset, a test dataset, and a validation dataset;

[0197] It should be noted that the set ratio is 7:1:2; that is, training dataset: test dataset: validation dataset = 7:1:2.

[0198] It should be noted that in this embodiment, we use a chronological division method to avoid data leakage. The training set accounts for 70% (13,365 records), covering the period from 2001 to 2015, of which 13,277 records are invalid and 88 are valid. The validation set accounts for 10% (1,909 records), covering the period from 2015 to 2017, of which 1,894 records are invalid and 15 are valid. The test set accounts for 20% (3,819 records), covering the period from 2017 to 2023, of which 3,781 records are invalid and 38 are valid.

[0199] S323, Using the training dataset and the test dataset, the effective ground prediction model is trained to obtain an optimized effective ground prediction model;

[0200] It should be noted that the effective prediction model for Earth includes the CME physical parameter feature model, the TimeSformer image sequence model, and the result data fusion model;

[0201] It should be noted that the CME physical parameter characteristic model includes a first effective prediction model for Earth and a second effective prediction model for Earth.

[0202] S324, Using the aforementioned land effectiveness prediction optimization model, the validation dataset is processed to obtain CME land effectiveness prediction result data;

[0203] It should be noted that in this embodiment, the first effective ground prediction model and the second effective ground prediction model are compared.

[0204] The first effective prediction model first randomly initializes the weights and biases, then calculates the prediction probability of each sample, then calculates the loss using the weighted cross-entropy loss function, then obtains the gradient through backpropagation, then updates the parameters, and repeats these steps until convergence or the maximum number of iterations is reached.

[0205] The first effective prediction model is simple, has few parameters, and is easy to understand and implement; it is highly interpretable, as the magnitude and sign of the weights directly reflect the influence of features on the results; it is also fast to train, can output probabilities, making threshold optimization convenient, and has no strict requirements on feature standardization. However, it also has drawbacks: it can only handle linear relationships and cannot capture complex nonlinear patterns; it relies heavily on feature engineering and is sensitive to outliers, but its performance may not be good in small sample sizes.

[0206] It should be noted that in this embodiment, the first pair of effective prediction models is used as a baseline model to compare the performance of other complex models; it can also analyze which features are more important through weight coefficients; it can also quickly verify the effect of feature engineering, and even serve as a meta-learner in Stacking integration in ensemble learning.

[0207] The second effective prediction model first sets the number of trees, then performs bootstrap sampling on each tree, then constructs a decision tree, and finally integrates the prediction results of all trees.

[0208] The second pair of effective prediction models can handle nonlinear relationships, provide interpretable feature importance, are robust to overfitting, do not require feature standardization, can handle partially missing features, and can be trained in parallel. However, it also has drawbacks: the model is relatively complex, has a long training time, and consumes a lot of memory. Although it can calculate feature importance, individual predictions are still difficult to interpret.

[0209] It should be noted that in this embodiment, the hyperparameter of the second effective prediction model is set to 15; the minimum number of split samples and the minimum number of leaf node samples are both to prevent overfitting and improve generalization ability.

[0210] It should be noted that in this embodiment, the second Earth-based effective prediction model is used to capture nonlinear patterns, since the relationship between the effectiveness and characteristics of coronal mass ejections on Earth may not be linear; it can also analyze which physical parameters are most important for prediction, and it has good robustness to outliers and noise, and can also be used as a representative of ensemble learning and compared with gradient boosting methods.

[0211] It should be noted that, using the first effective Earth prediction model, from Figure 3 The confusion matrix shows 239 false positives that were actually invalid but predicted as valid, and 16 false negatives that were actually valid but predicted as invalid. Figure 4As can be seen, the area under the ROC curve of the second effective Earth prediction model reaches 0.829, which shows good discrimination ability and successfully captured 22 effective Earth events, which is much better than random guessing and improves the accuracy of the prediction results. Therefore, in this embodiment, the second effective Earth prediction model is used as the CME physical parameter feature model.

[0212] Optionally, the preprocessing of the CME ground-based effective dataset to obtain a preprocessed CME ground-based effective dataset includes:

[0213] S3211, The CME effective ground dataset is parsed to obtain CME effective ground parameter data and CME effective ground image sequence data;

[0214] It should be noted that the CME's effective ground image sequence data is as follows: Figure 5 As shown;

[0215] The CME ground-based effective parameter data is normalized to obtain the first preprocessed CME ground-based effective dataset;

[0216] The first preprocessed CME ground-based effective dataset includes several first preprocessed CME ground-based effective datasets;

[0217] S3212, perform weight processing on the first preprocessed CME ground-based effective dataset to obtain the second preprocessed CME ground-based effective dataset;

[0218] The second preprocessed CME ground-based valid dataset includes several second preprocessed CME ground-based valid datasets;

[0219] It should be noted that the second preprocessed CME ground-based valid data represents the data credibility weight of the first preprocessed CME ground-based valid data;

[0220] It should be noted that the weight processing expression is as follows:

[0221] ，

[0222] in, Indicates the first The first preprocessed CME ground-based valid data;

[0223] S3213, Based on the CME ground-based effective dataset, construct CME feature data to obtain the third preprocessed CME ground-based effective dataset;

[0224] S3214, perform fusion processing on the first preprocessed CME effective ground dataset, the second preprocessed CME effective ground dataset, the third preprocessed CME effective ground dataset, and the CME effective ground image sequence data to obtain the preprocessed CME effective ground dataset.

[0225] Optionally, the step of constructing CME feature data based on the CME ground-based effective dataset to obtain a third preprocessed CME ground-based effective dataset includes:

[0226] S32131, Construct the CME feature set;

[0227] It should be noted that the CME feature set includes several CME features;

[0228] It should be noted that, in this embodiment, the CME feature set includes 30 features, of which 15 are original features and 15 are derived features;

[0229] The original features refer to the physical parameter features directly extracted from CME observation data.

[0230] These are physical parameter features extracted directly from CME observation data.

[0231] There are three location characteristics: the central position angle (PA) indicates the burst direction of the CME, 0- PA at 270- The range of CMEs is more likely to be ground-effective, with about 15% of the data missing, filled with the median; the angular width reflects the lateral extension angle of the CME, with a wider CME being more likely to be ground-effective and having complete data; the magnetic anomaly (MPA) is the most likely propagation direction of the CME, usually close to the PA, but magnetic field deflection is taken into account.

[0232] The velocity characteristics include four aspects: linear velocity is the average propagation velocity of CMEs. Fast CMEs are more likely to be effective on Earth. The average velocity of CMEs effective on Earth is about 650 km / s, which is significantly higher than the overall average. Initial velocity is the velocity when it is first observed, reflecting the initial burst energy. Final velocity is the velocity when it leaves the observation field of view, reflecting its evolution in the corona. The velocity at 20 solar radii is closer to the velocity in interplanetary space and has higher reference value.

[0233] There are three kinetic characteristics: acceleration is the rate of change of velocity, and an accelerated CME may have more energy; the mass is estimated through brightness and volume, with about 25% of the data missing; and the kinetic energy is calculated based on mass and velocity, with a CME with more energy being more likely to be effective on the ground.

[0234] Other features include five characteristics: event quality rating (0-4, data from high-quality observations are more reliable), starting altitude, ending altitude, and altitude range.

[0235] The derived features represent engineering features created based on prior physical knowledge and domain experience; these derived features can capture nonlinear relationships between features, handle scale issues, introduce domain knowledge, and improve the model's expressive power.

[0236] The derived features include three directional features. The directional features mainly handle the circularity of PA by converting the angle into cosine and sine values ​​to eliminate the circularity problem. A binary feature indicating whether the direction is towards the Earth is set to directly encode domain knowledge.

[0237] The derived features include velocity change features, which include binary features of velocity change amount, velocity ratio and whether acceleration occurs, reflecting the velocity change of CME in the field of view;

[0238] The derived features include momentum features and energy features; wherein the momentum features reflect the propagation capability of the CME; and the energy features represent energy density, which reflects the energy concentration of the CME.

[0239] The derived features also include classification features, which include whether it is a halo-shaped CME (angle width > 0.05). The two binary features are halo-shaped CME and whether it is a high-speed CME (linear speed > 400 km / s). Halo-shaped CME and high-speed CME are more likely to be effective for ground.

[0240] The derived features also include logarithmic transformation features, which are used to handle problems with large numerical ranges. Logarithmic transformation is performed on mass and kinetic energy to compress the numerical range and improve stability.

[0241] It should be noted that after model training, feature importance analysis can reveal which features are more critical. Based on physics, anticipated important features include whether the model faces Earth (potentially the most important), linear velocity, angular width, whether it forms a halo-like CME, directional features, and solar activity cycles. The design of these feature engineering techniques allows derived features to capture the nonlinear relationships between original features, logarithmic transformations to improve numerical stability, and binary features to directly encode domain knowledge, thereby improving learning efficiency.

[0242] S32132, Based on the CME feature set, match the CME ground-based effective dataset to obtain the corresponding CME feature data;

[0243] S32133, combine all CME feature data to obtain the third preprocessed CME ground-based effective dataset;

[0244] It should be noted that the combination refers to the combination according to the sample index order.

[0245] As can be seen, the method for predicting the Earth effectiveness of coronal mass ejections (CMEs) provided in this embodiment utilizes an Earth effectiveness prediction model composed of a CME physical parameter feature model, a TimeSformer image sequence model, and a result data fusion model to process CME data. This organically integrates two heterogeneous information sources—CME physical parameter features and CME image features—to obtain CME Earth effectiveness prediction results data. This achieves automatic classification and prediction of Earth effectiveness, improves the accuracy and robustness of prediction, and reduces the false negative rate.

[0246] Example 2

[0247] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of a coronal mass ejection (CME) Earth-based effectiveness prediction device disclosed in an embodiment of the present invention. Figure 6 The described coronal mass ejection (CME) Earth-based effectiveness prediction device is applied to space weather forecasting systems, such as local servers or cloud servers within these systems; however, this invention does not limit its application. Figure 6 As shown, the device may include:

[0248] Historical data acquisition module 101, data extraction module 102, and ground effectiveness processing module 103;

[0249] The historical data acquisition module 101 is used to acquire the CME dataset and the ICME dataset;

[0250] The data extraction module 102 is used to process the CME dataset and the ICME dataset to obtain the CME ground-based effective dataset;

[0251] The land effectiveness processing module 103 is used to process the CME land effectiveness dataset using the land effectiveness prediction model to obtain CME land effectiveness prediction result data;

[0252] The historical data acquisition module 101, the data extraction module 102, and the ground effectiveness processing module 103 are sequentially connected;

[0253] The device provided in this embodiment has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment for predicting the effectiveness of coronal mass ejections to Earth.

[0254] Example 3

[0255] Please see Figure 7 , Figure 7 This is a schematic diagram of another coronal mass ejection (CME) Earth-based effectiveness prediction device disclosed in an embodiment of the present invention. Figure 7 The described apparatus can be applied to space weather forecasting systems, such as local servers or cloud servers used in space weather forecasting systems, and the embodiments of the present invention are not limited thereto. Figure 7 As shown, the device may include:

[0256] Memory 201 storing executable program code;

[0257] Processor 202 coupled to memory 201;

[0258] The processor 202 calls the executable program code stored in the memory 201 to execute the steps in the method for predicting the effectiveness of coronal mass ejections on Earth as described in Embodiment 1.

[0259] Example 4

[0260] This invention discloses a computer-readable storage medium storing a computer program for electronic data exchange, wherein the computer program causes a computer to execute the steps in the coronal mass ejection Earth-to-ground effectiveness prediction method described in Embodiment 1.

[0261] Example 5

[0262] This invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps in the method for predicting the effectiveness of coronal mass ejections on Earth as described in Embodiment 1.

[0263] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0264] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), once programmable read-only memory (OTPROM), electronically erasable rewritable read-only memory (EEPROM), read-only optical disc (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0265] It should be noted that all calculation expressions or mathematical functions in the embodiments of the present invention have undergone dimensionless processing of the variables involved before calculation.

[0266] It should be noted that in all the calculation expressions or mathematical functions in the embodiments of the present invention, the values ​​of the input independent variables all meet the reasonable requirements of the input value range of the calculation expression or mathematical function, and can ensure that the calculation expression or mathematical function can be calculated smoothly without violating physical laws or mathematical rules.

[0267] Finally, it should be noted that the method and apparatus for predicting the effectiveness of coronal mass ejections to Earth disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the Earth-based effectiveness of coronal mass ejections, characterized in that, The method includes: S1, obtain the CME dataset and ICME dataset; S2, process the CME dataset and the ICME dataset to obtain the CME ground-based effective dataset; S3, using the land effectiveness prediction model, process the CME land effectiveness dataset to obtain CME land effectiveness prediction result data; The process of processing the CME dataset and the ICME dataset to obtain the CME-based effective dataset includes: S21, The CME dataset is populated to obtain a preprocessed CME dataset; S22, perform time matching processing on the preprocessed CME dataset and the ICME dataset to obtain the CME ground-based effective dataset; The process of using the land effectiveness prediction model to process the CME land effectiveness dataset to obtain CME land effectiveness prediction result data includes: S31, Construct an effective Earth prediction model; S32, Based on the land effectiveness prediction model, the CME land effectiveness dataset is processed to obtain CME land effectiveness prediction result data; The effective Earth prediction model includes a CME physical parameter feature model, a TimeSformer image sequence model, and a multimodal feature fusion model. The CME physical parameter feature model is used to process the CME ground-based effective parameter data to obtain the first classification result data; The CME physical parameter feature model includes a second data input layer, a diverse generation layer, a basic prediction layer, a fusion processing layer, and a second data output layer. The second data input layer is used to receive the CME's ground-based effective parameter data; The diverse generation layer is used to perform differential processing on the CME ground-based effective parameter data to obtain a differential sample dataset; the differential processing expression is: ， in, This represents the m-th differentiated sample data; Indicates the first hyperparameter; Indicates the first The local density of the preprocessed CME ground-based effective data; Indicates the first The label consistency value of the preprocessed CME ground-based valid data; This indicates the number of valid ground-to-ground data points of the preprocessed CME; The basic prediction layer is used to build a model based on decision trees, process the differentiated sample dataset, and obtain a differentiated decision tree set and a performance weight set; the expression for the decision tree model is: ， ， in, Indicates feature index; Indicates the first Gain improvement value for each feature; Indicates the second hyperparameter; Indicates the first Maximum redundancy of each feature; Represents a sample set; Represents a subset of samples; Indicates the first Probability of tree prediction; The category labels represent the decision tree; The weights represent the performance weights of the decision tree; This represents the prediction of the t-th decision tree. Category The probability of; Indicates category; Indicates the number of decision trees; Indicates the number of categories; The fusion processing layer is used to aggregate the differential decision tree set and the performance weight set to obtain the first classification result data; The second data output layer is used to output the first classification result data; The second data input layer, the diverse generation layer, the basic prediction layer, the fusion processing layer, and the second data output layer are sequentially connected; The TimeSformer image sequence model is used to process CME ground-based effective image sequence data to obtain second classification result data; The TimeSformer image sequence model includes a third data input layer, an image segmentation layer, an image encoding layer, a feature extraction layer, and a third data output layer. The TimeSformer image sequence model is used to process the CME ground-based effective image sequence data to obtain the second classification result data; The third data input layer is used to input the CME ground-based effective image sequence data of a specified size; The image segmentation layer is used to process the CME ground-based effective image sequence data of a specified size to obtain an image segmentation vector set and an image segmentation block set; The image coding layer is used to perform spatial and temporal location coding on the image segmentation vector set to obtain a coded image segmentation vector set. The feature extraction layer uses a 12-layer Transformer-Encoder to process the encoded image segmentation vector set to obtain image classification result data; The third data output layer is used to map the image classification result data to obtain the second classification result data; The third data input layer, the image segmentation layer, the image encoding layer, the feature extraction layer, and the third data output layer are sequentially connected in data. The multimodal feature fusion model is used to process the first classification result data and the second classification result data to obtain CME land effectiveness prediction result data; The output of the CME physical parameter feature model, the output of the TimeSformer image sequence model, and the input of the data fusion model are connected.

2. The method for predicting the effectiveness of coronal mass ejections on Earth according to claim 1, characterized in that, The process of filling the CME dataset to obtain a preprocessed CME dataset includes: S211, Traverse the CME dataset to obtain all CME data; S212, parse and process any of the CME data to obtain the target data segment to be filled; S213, Using the data filling model, the target data segment is filled to obtain preprocessed CME data; The data filling model expression is: ， in, Indicates missing values ​​to be filled; Indicates the number of valid data points; Indicates the first [number] after ascending sort. One data point; This represents the outlier filtering coefficient; This indicates the floor function, ensuring that k is an integer position (to avoid non-integer indices); S214, all the preprocessed CME data are combined in sequence to obtain a preprocessed CME dataset.

3. The method for predicting the effectiveness of coronal mass ejections on Earth according to claim 1, characterized in that, The step of performing time matching processing on the preprocessed CME dataset and the ICME dataset to obtain the CME-based ground-effective dataset includes: S221, Traverse the ICME dataset to obtain all ICME data; S222, parse and process any of the ICME data to obtain the LASCO-CME time; S223, Traverse the preprocessed CME dataset to obtain all preprocessed CME data; S224, Extract the file time information from the file name corresponding to any of the preprocessed CME data; S225, based on the LASCO-CME time, match and process the file time information to obtain the CME tag identifier value corresponding to the preprocessed CME data; S226, Based on the CME tag identification value, the preprocessed CME dataset is extracted to obtain the CME ground-based effective dataset.

4. The method for predicting the effectiveness of coronal mass ejections on Earth according to claim 3, characterized in that, The process of matching and processing the file time information based on the LASCO-CME time to obtain the CME tag identifier value includes: S2251, Calculate the difference between the time information and the LASCO-CME time to obtain the time difference; S2252, determine whether the time difference is less than a set time difference threshold, and obtain a time difference judgment result; S2253, when the time difference judgment result is yes, update the CME tag of the preprocessed CME data to 1, and obtain the CME tag identification value corresponding to the preprocessed CME data; When the time difference determination result is negative, the CME label of the preprocessed CME data is updated to 0, and the CME label identification value corresponding to the preprocessed CME data is obtained.

5. The method for predicting the effectiveness of coronal mass ejections on Earth according to claim 1, characterized in that, The process of processing the CME land effectiveness dataset based on the land effectiveness prediction model to obtain CME land effectiveness prediction result data includes: S321, preprocess the CME ground-based effective dataset to obtain a preprocessed CME ground-based effective dataset; S322, according to a set ratio, the preprocessed CME ground-based effective dataset is divided into a training dataset, a test dataset, and a validation dataset; S323, Using the training dataset and the test dataset, the effective ground prediction model is trained to obtain an optimized effective ground prediction model; S324, Using the aforementioned land effectiveness prediction optimization model, the validation dataset is processed to obtain CME land effectiveness prediction result data.

6. A device for predicting the effectiveness of coronal mass ejections (CMEs) on Earth, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the method for predicting the effectiveness of coronal mass ejections to Earth as described in any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, which, when invoked, are used to execute the method for predicting the effectiveness of coronal mass ejections on Earth as described in any one of claims 1 to 5.

Images