Solar radio burst prediction method and system based on whole solar magnetic map
Through the improved ResNet and LSTM networks, combined with custom residual layer and gating mechanism, and dynamic attention mechanism, the processing problems of small-scale features and time information in the entire solar magnetogram are solved, and the accuracy of solar radio burst prediction is improved. Model expression ability.
Patent Information
- Application Number
- CN202411176125.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-26
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-08-26
AI Technical Summary
When using a full-disc solar magnetic map to predict solar radio bursts, deep learning models have limited ability to capture small-scale features, complex spatial structure and large data volume, resulting in reduced prediction accuracy, and existing models fail to fully integrate space. and time information.
Using improved ResNet network and LSTM network, small-scale features of the full-disk solar magnetogram are extracted through custom residual layers, and gated mechanisms and dynamic attention mechanisms are introduced to process image feature vectors to improve prediction accuracy.
It enhances the model's ability to extract spatial information, effectively captures complex time dependencies and information of different time spans, and improves the accuracy of solar radio burst prediction and the generalization ability of the model.
Smart Images

Figure CN119048830A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of solar radio technology, and in particular to a solar radio burst prediction method and system based on a full-disk solar magnetogram. Background Art
[0002] The statements in this section merely mention background art related to the present invention and do not necessarily constitute prior art.
[0003] Solar radio bursts can be seen as a sign of predicting space weather disasters. Solar radio data analysis is an important means of predicting solar radio bursts based on the full-disk solar magnetic map. Existing research has applied deep learning and machine learning techniques to analyze solar radio data. For example, a solar radio database was developed using a deep belief network to extract features from solar radio spectrum maps; a recurrent neural network specifically designed for time series analysis was designed to classify solar radio spectrum data; the pre-trained visual geometry group network VGG-16 model was used to classify solar radio spectra, and the true positive rate (TPR) was improved by 12.2% compared to existing methods; a univariate time series prediction deep residual neural structure N-BEATS was established for predicting solar radio flux, and the model achieved better performance than the multi-flux neural network method in prediction; a radio burst detection system was designed using AlexNet, which is designed to adapt to different solar spectrum maps to provide high-accuracy predictions of solar radio bursts based on the full-disk solar magnetic map.
[0004] To sum up, in previous studies, most of the focus was on the solar radio spectrum. There is no doubt that solar radio bursts have very obvious characteristic information on the solar radio spectrum, but at the same time, the observation of the solar radio spectrum has extremely high requirements for instruments, and is therefore easily affected by sudden situations.
[0005] The full-disk solar magnetogram is a "snapshot" of the sun at a certain moment, which retains the solar activity information at that moment, such as the relative positions of the solar active regions. However, solar activity is continuous, that is, the full-disk solar magnetogram can be regarded as sequence data. However, in previous solar radio studies, this feature has been relatively rarely used, and the features existing in the sequence data cannot be well utilized. The existing algorithms are still applied to the processing of the full-disk solar magnetogram, and the following problems still exist:
[0006] (1) The full-disk solar magnetic map contains features of different scales, but existing deep learning models (such as VGG-16 and AlexNet) are not good at capturing small-scale features, resulting in limited ability to process fine structures.
[0007] (2) The full-disk solar magnetic map has a complex spatial structure and a huge amount of data. Existing models may lack generalization capabilities when dealing with magnetic maps at different time points and in different regions, resulting in reduced prediction accuracy on new data.
[0008] (3) Although some models (such as recurrent neural networks) are used to process time series data, they do not fully consider the combination of spatial and temporal information and cannot fully capture the dynamic changing characteristics of solar activity. Summary of the invention
[0009] In order to address the shortcomings of the prior art, the present invention provides a solar radio burst prediction method, system, electronic device, computer-readable storage medium and computer program product based on the full-disk solar magnetic map. The full-disk solar magnetic map is used as research data, and the improved ResNet network and the improved LSTM network are used to analyze the full-disk solar magnetic map, and the features hidden in the sequence are extracted to predict solar radio bursts.
[0010] In a first aspect, the present invention provides a method for predicting solar radio bursts based on a full-disk solar magnetogram;
[0011] A solar radio burst prediction method based on a full-disk solar magnetogram, comprising:
[0012] Obtaining a full-disk solar magnetic map, and inputting the full-disk solar magnetic map into a pre-trained improved ResNet model to obtain an image feature vector;
[0013] Dividing the image feature vector into pseudo time series and inputting the pseudo time series into an improved LSTM network to obtain a solar radio burst prediction result;
[0014] Among them, the improved ResNet model introduces a custom residual layer to extract the small-scale features of the full-disk solar magnetic map, and the improved LSTM network introduces a gating mechanism to process the image feature vectors of different time scales.
[0015] In some implementations, the custom residual layer includes a plurality of custom residual modules connected in sequence, and the custom residual module introduces a multi-layer perceptron to perform nonlinear transformation on the features of each channel.
[0016] In some embodiments, the custom residual module is connected to a first convolution unit, a second convolution unit, a global average pooling layer, a multi-layer perceptron and a ReLU activation function.
[0017] In some embodiments, the improved ResNet model includes a convolutional layer, a batch normalization layer, a ReLU activation function, a maximum pooling layer, a plurality of custom residual layers, a global pooling layer, and a fully connected layer connected in sequence.
[0018] In some embodiments, the improved LSTM network introduces a dynamic attention mechanism to dynamically adjust the attention level of the improved LSTM network to different time scales, and outputs solar radio burst prediction results based on the full-disk solar magnetic map.
[0019] In some embodiments, the gating mechanism is represented by:
[0020] G t =σ(ω G [h t-1 ,x t ]+b G );
[0021] Where σ is the activation function of the LSTM network, h t-1 is the hidden state at the previous moment t-1, x t is the input at the current time t, w G is the recurrent weight of the additional gating state, b G Bias for additional gating states.
[0022] In a second aspect, the present invention provides a solar radio burst prediction system based on a full-disk solar magnetogram;
[0023] A solar radio burst prediction system based on a full-disk solar magnetogram, comprising:
[0024] The image feature extraction module is configured to: obtain a full-disk solar magnetic map, input the full-disk solar magnetic map into a pre-trained improved ResNet model, and obtain an image feature vector;
[0025] A solar radio burst prediction module is configured to: divide the image feature vector into pseudo time series and input the pseudo time series into an improved LSTM network to obtain a solar radio burst prediction result;
[0026] Among them, the improved ResNet model introduces a custom residual module to extract the small-scale features of the full-disk solar magnetic map, and the improved LSTM network introduces a gating mechanism to process the image feature vectors of different time scales.
[0027] In a third aspect, the present invention provides an electronic device;
[0028] An electronic device comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-mentioned solar radio burst prediction method based on the full-disk solar magnetic map.
[0029] In a fourth aspect, the present invention provides a computer-readable storage medium;
[0030] A computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the steps of the above-mentioned solar radio burst prediction method based on the full-disk solar magnetic map.
[0031] In a fifth aspect, the present invention provides a computer program product;
[0032] A computer program product includes a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned solar radio burst prediction method based on the full-disk solar magnetic map.
[0033] Compared with the prior art, the present invention has the following beneficial effects:
[0034] 1. The technical solution provided by the present invention utilizes the improved ResNet to extract spatial features of the full-disk solar magnetic map, and introduces a multi-layer perceptron (MLP) through a custom residual layer to further process the features of each channel, thereby enhancing the expressive power of the model, enabling the network to better capture small-scale features and achieve full extraction of spatial information.
[0035] 2. The technical solution provided by the present invention introduces an additional gating mechanism to improve the LSTM network, so that the model has more degrees of freedom to control which information should be retained, updated or discarded, which can help the model better capture complex time dependencies and provide the model with additional linear transformation paths, so that it can better analyze complex sequence patterns and effectively capture information of different time spans in the full-disk solar magnetic map at a certain moment.
[0036] 3. The technical solution provided by the present invention introduces a dynamic attention mechanism, so that the LSTM network can dynamically adjust the degree of attention to different time steps through the context, so that the model can more centrally process those time steps that have a greater impact on the final decision. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.
[0038] Figure 1 A schematic diagram of a process flow provided by an embodiment of the present invention;
[0039] Figure 2 A schematic diagram of the network architecture of the improved ResNet model provided by an embodiment of the present invention;
[0040] Figure 3 A schematic diagram of a network architecture of a custom residual module provided in an embodiment of the present invention;
[0041] Figure 4 A schematic diagram of an image feature vector output by the improved ResNet model provided in an embodiment of the present invention;
[0042] Figure 5 A schematic diagram of the network architecture of the improved LSTM network provided in an embodiment of the present invention;
[0043] Figure 6 A schematic diagram of a flow chart of a dynamic attention mechanism provided by an embodiment of the present invention;
[0044] Figure 7 An example diagram of the visualization result after the image feature vector is divided according to an embodiment of the present invention;
[0045] Figure 8 An example diagram of a full-disk solar magnetic map provided by an embodiment of the present invention;
[0046] Fig. 9 This is an example diagram of the evaluation results provided by the embodiment of the present invention. DETAILED DESCRIPTION
[0047] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.
[0048] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "have" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0049] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.
[0050] Embodiment 1
[0051] Existing solar radio burst predictions rely on solar radio spectrum data, but the collection of solar spectrum data relies on high-precision instruments and is extremely susceptible to sudden situations, which makes the basic data less accurate and affects the prediction effect; therefore, the present invention provides a solar radio burst prediction method based on the full-disk solar magnetic map. Taking into account the continuity of solar activity, the full-disk solar magnetic map is regarded as sequence data, and the improved ResNet model and the improved LSTM network are used to perform spatial feature analysis and extraction, thereby predicting solar radio bursts.
[0052] Next, combine Figure 1-Figure 9 , a solar radio burst prediction method based on a full-disk solar magnetic map disclosed in this embodiment is described in detail. The solar radio burst prediction method based on a full-disk solar magnetic map comprises the following steps:
[0053] S1. Obtain the full-disk solar magnetic map.
[0054] The full-disk solar magnetic map directly displays the distribution of the magnetic field on the solar surface. The image is displayed in grayscale, and the black and white areas in the image represent opposite magnetic fields. The full-disk solar magnetic map is a single-frame image in time. Compared with the spectrum map, the full-disk solar magnetic map can clearly show the state of solar activity at a certain moment, and has a higher resolution, that is, the image of the full-disk solar magnetic map contains rich features and can directly display the magnetic field distribution on the solar surface. Compared with the spectrum map, the full-disk solar magnetic map can be likened to a "snapshot" of solar activity at a certain moment, that is, each image contains all the characteristics of the currently observed solar activity, which can well show the state of the solar activity zone.
[0055] In this embodiment, the full-disk solar magnetic map comes from the Michelson Doppler Imager (MDI) on the Solar and Heliospheric Observatory (SOHO) satellite. MDI conducts full-disk observations of the photospheric field every 96 minutes. These full-disk magnetic maps are recorded by its 1024×1024 CCD detector with a field of view of 34.
[0056] S2. Input the full-disk solar magnetic map into the pre-trained improved ResNet model to obtain the image feature vector.
[0057] The full-disk solar magnetogram is a complex image data containing a large amount of spatial information; it may contain features of different scales, such as local magnetic field activity and overall solar activity.
[0058] Compared with shallow convolutional neural networks, the ResNet model has stronger feature extraction capabilities and can effectively extract useful spatial features from the full-disk solar magnetic map. However, as the number of network layers increases, deep networks often face the problem of gradient vanishing or exploding, making the model difficult to train. In addition, although the ResNet network can capture features of different scales through its multi-layer convolution and pooling operations, in the deep network structure, small-scale features may be gradually ignored in the layer-by-layer convolution and pooling process. Therefore, in this embodiment, the ResNet model is improved and a custom residual layer is introduced to comprehensively extract small-scale features from the full-disk solar magnetic map.
[0059] As an implementation method, the improved ResNet model includes a convolutional layer, a batch normalization layer, a ReLU activation function, a maximum pooling layer, 4 custom residual layers, a global pooling layer and a fully connected layer connected in sequence, and the custom residual layer includes 2 custom residual modules connected in sequence.
[0060] Specifically, the full disk solar magnetic map is used as input, and after a convolution layer with a convolution kernel of 7x7 and a stride of 2, as well as a batch normalization stabilization training process and the introduction of nonlinearity by the ReLU activation function, the feature map is output; the feature map is then processed by the maximum pooling layer to reduce the size of the feature map, and then input into the custom residual layer. The first custom residual layer has an input channel of 64, an output channel of 64, and a stride of 1; the second custom residual layer has an input channel of 64, an output channel of 128, and a stride of 2; the third custom residual layer has an input channel of 128, an output channel of 256, and a stride of 2; the fourth custom residual layer has an input channel of 256, an output channel of 512, and a stride of 2.
[0061] The output of the fourth defined residual layer is used as the input of the global pooling layer, and a 1x1 feature map of 512 channels is output; the 1x1 feature map is expanded and a 512-dimensional image feature vector is output through the fully connected layer.
[0062] Furthermore, each custom residual layer contains two custom residual modules, the custom residual module includes a first convolution unit, a second convolution unit, a global average pooling layer, a multilayer perceptron and a ReLU activation function connected in sequence, and the input of the first convolution unit and the output of the multilayer perceptron are residually connected; the first convolution unit and the second convolution unit have the same structure, including a convolution layer, a BN layer and a ReLU activation function connected in sequence; the multilayer perceptron includes an input layer, a linear layer, a ReLU activation function, a linear layer, a ReLU activation function, a Reshape layer and an output layer connected in sequence.
[0063] The specific implementation of the custom residual module is: after processing through two sets of convolution units using 3x3 convolution kernel convolution, batch normalization BN, and ReLU activation function, the output feature image is globally average pooled, and the features of each channel are compressed into a single value to form a two-dimensional tensor. The pooled features are nonlinearly transformed through MLP, and the output size remains unchanged. The output of MLP is then adjusted to the same size as the original feature map, and a residual connection is performed to finally output the processed feature image.
[0064] In this embodiment, by introducing the MLP module, the features extracted by the residual module are nonlinearly transformed, and the input image process is residually connected, which ensures that small-scale features are not ignored during the convolution training process.
[0065] The inputs received by the custom residual module described in this embodiment are in_chinnels, out_chinnels, blocks and stride, which represent input channels, output channels, number of modules and stride respectively. In this embodiment, the number of custom residual modules in the custom residual layer is fixed to 2.
[0066] S3. Divide the image feature vector into pseudo time series and input it into the improved LSTM network to obtain the solar radio burst prediction results based on the full-disk solar magnetic map.
[0067] Among them, the improved LSTM network introduces a gating mechanism to process image feature vectors of different time scales, introduces a dynamic attention mechanism, dynamically adjusts the attention level of the improved LSTM network to different time steps, and outputs the solar radio burst prediction results based on the full-disk solar magnetic map.
[0068] As an implementation method, the specific process of S3 is as follows:
[0069] S301 , dividing the input 512-dimensional image feature vector into pseudo time series.
[0070] Specifically, the 512-dimensional image feature vector is divided into four sub-vectors, each sub-vector corresponds to a time step, and each group of sub-vectors is input into the LSTM as sequence data.
[0071] Since the improved LSTM network only accepts time series data as input, the image feature vector is processed into a continuous pseudo time series. The time step is the number unit of the processed pseudo time series. That is, in the current case, the value corresponding to the time step is 4, which means that the data input into the improved LSTM network is a set of pseudo time series data containing four feature sub-vectors.
[0072] S302: Input the sub-vector into the improved LSTM network for processing, perform weighted calculation on the processing result through the dynamic attention mechanism, and output the solar radio burst prediction result.
[0073] In this embodiment, the improved LSTM network is used to process sub-vectors that are continuously distributed in the spatial dimension, so that the spatial information contained in the continuous spatial sequence data can be fully learned, and the information of different time spans in the full-disk solar magnetic map can be extracted.
[0074] LSTM (Long Short-Term Memory) is a special recurrent neural network (RNN) designed specifically for processing and predicting time series data. Traditional RNNs are prone to gradient vanishing or gradient exploding problems when processing long sequences, making it difficult for the model to capture long-term dependencies. LSTM effectively solves these problems by introducing memory cells and gate mechanisms (input gate, forget gate, output gate). In the LSTM network, memory cells are used to store the state in the network and can remember information for a long time. The input gate is used to control the extent to which information is added to the memory cell. The forget gate determines which information in the memory cell should be forgotten, and the output gate determines the extent to which the information in the memory cell is output. Through the combination of these gates, LSTM can selectively retain and discard information, thereby effectively learning and memorizing dependencies over long time spans.
[0075] The full disk solar magnetic map at a certain moment contains information of different time spans, and the existing LSTM model cannot capture these features well. Therefore, in this embodiment, an additional gating mechanism G is introduced to give the model more freedom to control which information should be retained, updated or discarded, which can help the model better capture complex time dependencies. The new gating mechanism is equivalent to providing the model with an additional sub-linear transformation path, enabling it to better analyze complex sequence patterns.
[0076] Furthermore, in the improved LSTM network, the forget gate f t It is expressed as:
[0077] f t =σ(ω f [h t-1 ,x t ]+b f );
[0078] Input Gate i t It is expressed as:
[0079] i t =σ(ω i [h t-1 ,x t ]+bi );
[0080] Output gate o i It is expressed as:
[0081] o t =σ(ω o [h t-1 ,x t ]+b o );
[0082] The calculation formula of the gating mechanism is as follows:
[0083] G t =σ(ω G [h t-1 ,x t ]+b G );
[0084] Based on this gating mechanism, the LSTM network state unit c at time t t It is expressed as:
[0085]
[0086] Candidate status information at time t It is expressed as:
[0087]
[0088] Candidate state information h at time t t It is expressed as:
[0089] h t =o t ·tanh(c t );
[0090] In the formula, x t is the current input, i t is the input gate value, ω f ,ω t ,ω o ,ω c are the cycle weights of each state, b f , b t , b o , b c are the biases of each state, h t-1 、c t-1 are the hidden state and candidate state of the previous moment, σ and tanh are the activation functions of the improved LSTM network, which are the sigmoid function and the hyperbolic tangent function respectively.
[0091] f t Control the state unit c of the previous moment t-1Whether the information in is transmitted to the current moment determines the retention ratio of the output information at the previous moment, i t Control the current input x t and the hidden state h t-1 So as to selectively obtain the current input information, o t Then the output at the current moment can be calculated based on the new input information and the output information at the previous moment, c t Used to store and transmit time series information.
[0092] G t This is equivalent to adding an extra control mechanism between the forgetting and input mechanisms of the classic LSTM, which is a supplementary adjustment to the memory unit and can be used to adjust the C t Additional addition or reduction is performed, compared with the traditional LSTM based on f t 、i t To update the memory state, G t It provides an additional adjustment channel to enhance the model's ability to adapt to the characteristics of different time scales of the input data.
[0093] When processing long sequences, the improved LSTM network can capture the temporal dependency of the sequence, but the focus on each time step is uniform, and there is no distinction between which time step's information is more important. Through the analysis of the full-disk solar magnetic map, each image contains solar active areas and non-solar active areas. It is obvious that the relevant characteristic information of the solar active areas should receive more attention because it is closely related to solar radio bursts. Therefore, in this embodiment, a dynamic attention mechanism is introduced so that the model can dynamically adjust the degree of attention to different time steps through the context, so that the model can more centrally process those time steps that have a greater impact on the final decision. The specific process is as follows:
[0094] (1) The dynamic attention mechanism receives the output of the improved LSTM as input data LSTM_out and passes it into the attention head. The default number of attention heads is 4 (because the number of previous time step divisions is 4).
[0095] (2) Each attention head independently calculates the attention weight and generates a feature context vector according to the following steps:
[0096] (201) Use linear transformation to process the input data LSTM_out and calculate the attention weight of the current attention head.
[0097] (202) The softmax function is used to normalize the attention weights to ensure that the sum of the attention weights is 1.
[0098] (203) The normalized attention weights are multiplied by the elements of LSTM_out one by one and summed along the time step dimension to generate the feature vector of the current attention head.
[0099] (204) The outputs of all attention heads are concatenated into an overall context feature vector, and the dimension is compressed through a linear layer to generate the final context feature vector.
[0100] (3) Feature output and classification:
[0101] Specifically, the generated context feature vector is passed to the fully connected layer, and then passes through the first fully connected layer fc1 for feature change, and then passes through the Droput layer to prevent overfitting. By randomly inactivating a part of neurons, the generalization ability of the model is enhanced. Then, it passes through the second fully connected layer fc2 to compress the feature dimension from 512 to 128, and the fully connected layer fc3 to compress the feature dimension from 128 to 2. The output is used as the prediction result of solar radio bursts.
[0102] As an implementation method, before executing S2 and S3, the improved LSTM network and the improved ResNet model need to be trained. The specific process is as follows:
[0103] (1) Obtain the solar radio burst log file provided by an official website, which systematically records solar radio burst events from January 1967 to January 2011. In the log file, the seventh column describes the type of solar radio burst event that occurred within the specified time interval. This log file systematically compiles observation data from various observatories.
[0104] According to the information recorded in the log files of solar radio burst observations, the data are divided into two categories: "burst" and "no burst". For each set of recorded data, if any number of solar radio bursts occur in the corresponding daily data in the log file, it is classified as "burst"; otherwise, it is classified as "no burst". In the obtained full-disk solar magnetogram dataset, there are 2100 records with solar bursts and 936 records without solar bursts.
[0105] At the same time, in order to verify the accuracy of the implemented method in predicting solar radio bursts, the common ten-fold cross-validation method was used to divide the data set, that is, the data set was divided into ten sub-datasets of equal size, and the data ratio in each sub-dataset was the same as that of the original data set.
[0106] In addition, considering that there may be too many image features that may cause small pixel features to be ignored, the input data set was subjected to feature engineering such as random horizontal flipping, random vertical flipping, random rotation, and normalization to obtain the processed image data.
[0107] (2) The processed data is used as the training set to train the improved LSTM network and the improved ResNet model, and the prediction performance of the model is evaluated using evaluation indicators.
[0108] Exemplarily, solar radio burst prediction is considered as a binary classification task, in which burst samples are classified as positive samples, and non-burst samples are classified as negative samples. The evaluation parameters are: TP (True Positive), FP (False Positive), TN (True Negative) and FN (False Negative), and the detailed definitions are given in Table 1.
[0109] Table 1 Definitions of TP, FP, TN and FN
[0110]
[0111] Based on the evaluation parameters, four evaluation metrics are used to evaluate the performance of the prediction model: accuracy, precision, recall, and false alarm rate (FAR). Accuracy, precision, recall, and false alarm rate are all important metrics for evaluating predictions. Accuracy measures the proportion of correctly predicted samples, considering the prediction accuracy of the true positive class, recall evaluates the ability to successfully capture positive samples, and false alarm rate quantifies the proportion of incorrectly predicted positive samples in the negative class. The detailed definitions are given in Table 2.
[0112] Table 2 Definitions of accuracy, precision, recall rate, and false alarm rate
[0113]
[0114] Table 3 shows ten sets of confusion matrix results obtained based on the ten-fold cross validation method. It can be seen from the table that the number of samples predicted as true positives and true negatives is relatively high, which shows that the model has good performance in practical applications.
[0115] Table 3 Ten data sets and their evaluation parameter results obtained based on the ten-fold cross validation method
[0116]
[0117]
[0118] Table 4 gives the calculation results of the evaluation indicators for these ten sets of data. It can be seen from the table that the four types of evaluation indicators fluctuate within a very small range. It is also worth noting that the accuracy rate is always maintained above 0.87 and the precision is always maintained above 0.89, which shows that the model proposed in this embodiment has a good prediction effect.
[0119] Table 4 Ten data sets and their evaluation index results obtained based on the ten-fold cross validation method
[0120]
[0121] Combination Fig. 9 .Then the fluctuation range of the evaluation index is given more intuitively. It can be seen that the fluctuation range of the result is small.
[0122] Embodiment 2
[0123] This embodiment discloses a solar radio burst prediction system based on a full-disk solar magnetogram, comprising:
[0124] The image feature extraction module is configured to: obtain a full-disk solar magnetic map, input the full-disk solar magnetic map into a pre-trained improved ResNet model, and obtain an image feature vector;
[0125] A solar radio burst prediction module is configured to: divide the image feature vector into pseudo time series and input the pseudo time series into an improved LSTM network to obtain a solar radio burst prediction result based on a full-disk solar magnetic map;
[0126] Among them, the improved ResNet model introduces a custom residual module to extract the small-scale features of the full-disk solar magnetic map, and the improved LSTM network introduces a gating mechanism to process the image feature vectors of different time scales.
[0127] It should be noted that the above-mentioned image feature extraction module and solar radio burst prediction module correspond to the steps in Embodiment 1, and the examples and application scenarios implemented by the above-mentioned modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Embodiment 1. It should be noted that the above-mentioned modules, as part of the system, can be executed in a computer system such as a set of computer executable instructions.
[0128] Embodiment 3
[0129] Embodiment 3 of the present invention provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of the above-mentioned method for predicting solar radio bursts based on the full-disk solar magnetic map are completed.
[0130] Embodiment 4
[0131] Embodiment 4 of the present invention provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps of the above-mentioned solar radio burst prediction method based on the full-disk solar magnetic map are completed.
[0132] Embodiment 5
[0133] Embodiment 5 of the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned solar radio burst prediction method based on the full-disk solar magnetic map.
[0134] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0135] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0136] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0137] The description of each embodiment in the above embodiments has different emphases. For parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.
[0138] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A solar radio burst prediction method based on a full-disk solar magnetic map, characterized in that: include: Obtaining a full-disk solar magnetic map, and inputting the full-disk solar magnetic map into a pre-trained improved ResNet model to obtain an image feature vector; Dividing the image feature vector into pseudo time series and inputting the pseudo time series into an improved LSTM network to obtain a solar radio burst prediction result; Among them, the improved ResNet model introduces a custom residual layer to extract the small-scale features of the full-disk solar magnetic map, and the improved LSTM network introduces a gating mechanism to process the image feature vectors of different time scales.
2. The solar radio burst prediction method based on the full-disk solar magnetic map according to claim 1, characterized in that: The custom residual layer includes a plurality of custom residual modules connected in sequence, and the custom residual module introduces a multi-layer perceptron to perform nonlinear transformation on the features of each channel.
3. The solar radio burst prediction method based on the full-disk solar magnetogram according to claim 2, characterized in that: The custom residual module is connected to a first convolution unit, a second convolution unit, a global average pooling layer, a multi-layer perceptron and a ReLU activation function.
4. The solar radio burst prediction method based on the full-disk solar magnetogram according to claim 1, characterized in that: The improved ResNet model includes sequentially connected convolutional layers, batch normalization layers, ReLU activation functions, maximum pooling layers, multiple custom residual layers, global pooling layers and fully connected layers.
5. The solar radio burst prediction method based on the full-disk solar magnetogram according to claim 1, characterized in that: The improved LSTM network introduces a dynamic attention mechanism, dynamically adjusts the attention level of the improved LSTM network to different time scales, and outputs solar radio burst prediction results.
6. The solar radio burst prediction method based on the full-disk solar magnetogram according to claim 1, characterized in that: The gating mechanism is expressed as: G t =σ(ω G [h t-1 ,x t ]+b G ); Where σ is the activation function of the LSTM network, h t-1 is the hidden state at the previous moment t-1, x t is the input at the current time t, w G is the recurrent weight of the additional gating state, b G Bias for additional gating states.
7. A solar radio burst prediction system based on the full-disk solar magnetic map, characterized in that: include: The image feature extraction module is configured to: obtain a full-disk solar magnetic map, input the full-disk solar magnetic map into a pre-trained improved ResNet model, and obtain an image feature vector; A solar radio burst prediction module is configured to: divide the image feature vector into pseudo time series and input the pseudo time series into an improved LSTM network to obtain a solar radio burst prediction result; Among them, the improved ResNet model introduces a custom residual module to extract the small-scale features of the full-disk solar magnetic map, and the improved LSTM network introduces a gating mechanism to process the image feature vectors of different time scales.
8. An electronic device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the solar radio burst prediction method based on the full-disk solar magnetic map as described in any one of claims 1-6.
9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the solar radio burst prediction method based on the full-disk solar magnetic map described in any one of claims 1 to 6 are implemented.
10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the solar radio burst prediction method based on the full-disk solar magnetic map described in any one of claims 1 to 6 are implemented.
Citation Information
Patent Citations
Solar radio burst forecasting method and system based on full-solar-surface photosphere magnetic map
CN114912674A
Power load prediction method and device based on bidirectional long short-term memory network
CN117728380A
Detection method using fusion network based on attention mechanism, and terminal device
US11222217B1
Neural networks with linear recurrence and feedforward units
WO2024156796A1
Cited By
Image analysis method based on deep learning and application thereof
CN119379693A
