Solar burst real-time monitoring method and system based on CNN-LSTM
By using a CNN-LSTM-based solar radio flux prediction model and a multi-judgment mechanism, the real-time and accuracy issues of solar radio burst detection are solved, and efficient monitoring of solar radio bursts is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANDONG UNIV
- Filing Date
- 2023-03-23
- Publication Date
- 2026-07-31
AI Technical Summary
Existing solar radio burst detection systems suffer from poor real-time performance, limited detection methods, frequent misjudgments, time-consuming large-scale data processing, and low detection accuracy.
A solar radio current prediction model based on CNN-LSTM is adopted, combined with a multiple judgment mechanism. The solar radio current prediction model is constructed by using convolutional neural network (CNN) and long short-term memory network (LSTM). Real-time observation data from solar radiation flowmeter is used to perform multiple judgments to improve detection accuracy.
It enables real-time detection of solar radio bursts, reduces outlier interference, improves detection accuracy and intelligence, and lowers equipment costs.
Smart Images
Figure CN116484720B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of solar radio detection, and particularly relates to a method and system for real-time monitoring of solar flares based on CNN-LSTM. Background Technology
[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.
[0003] Solar radio is radio radiation from the Sun. Based on the intensity of solar activity, it is divided into three types: quiet solar radio, slowly varying solar radio, and solar radio bursts. Solar radio bursts carry various information about the surrounding physical environment and radiation mechanisms to Earth. Among them, the fine burst structure carried by solar radio bursts can reflect the complex magnetic field structure and important physical processes such as the movement of high-energy particles during the burst.
[0004] With the continuous development of solar radio observation instruments, radio telescopes with high time and frequency resolution have gradually emerged, which has led to a significant increase in the amount of data stored for solar radio observation. The instruments generate a huge amount of solar radio data every day, but the number of bursts in the data is very small. It is extremely difficult for researchers to detect solar radio bursts in real time and efficiently from the huge amount of data using manual methods.
[0005] Currently, common solar radio burst monitoring systems typically fall into two categories: threshold judgment and image judgment. Threshold judgment can quickly identify solar bursts, but due to environmental and electronic noise, false positives can occur. Image judgment, on the other hand, preprocesses images, improving the accuracy of solar burst warnings. However, the processed images are drawn from a data set and cannot be processed upon receiving the data, resulting in poor real-time performance for this type of solar radio burst monitoring system.
[0006] Therefore, existing solar eruption detection methods suffer from the following drawbacks: 1. Poor real-time performance: Most current solar radio eruption detection methods analyze data collected on a given day to identify eruptions. 2. Limited detection methods: Most methods rely on analyzing solar radio eruption spectrum images, leading to potential misjudgments. 3. Lack of interference removal: Solar radiation flux assessments often rely on threshold settings. While single outliers can be removed through secondary checks, continuous outliers interfere with the system, resulting in low accuracy. 4. Massive amounts of eruption data: Summarizing eruption timing data requires significant time and effort. Summary of the Invention
[0007] To overcome the shortcomings of the prior art, this invention provides a real-time monitoring method and system for solar flares based on CNN-LSTM. Based on the CNN-LSTM network, a solar radio current prediction model is constructed to predict the radio current value at the next moment in real time. Multiple judgments are used to detect the flares, effectively solving the problem of real-time detection of solar radio flares and improving the accuracy of solar flare early warning.
[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:
[0009] The first aspect of this invention provides a method for real-time monitoring of solar flares based on CNN-LSTM;
[0010] Real-time monitoring methods for solar flares based on CNN-LSTM include:
[0011] A solar radio flux prediction model was constructed based on CNN-LSTM, and the model was trained using historical solar radio observation data.
[0012] Real-time observation of solar radio current values forms time-domain data, which is then input into a trained solar radio current prediction model to generate and output the predicted radio current value for the next moment.
[0013] Based on the predicted radio current values, multiple judgments are made to obtain the solar eruption monitoring results.
[0014] Furthermore, the solar radio flux prediction model consists of a CNN layer, a Dropout layer, an LSTM layer, an average pooling layer, a max pooling layer, and a fully connected layer, and is trained on an industrial control computer.
[0015] Furthermore, the solar radio flux prediction model uses the CBAM mechanism, which, based on the channel attention mechanism, selects the spatial dimension and uses average pooling to obtain the corresponding weights, and then distributes attention serially in two dimensions: space and channel.
[0016] Furthermore, the real-time observation of solar radio current values is performed using a solar radiation flow meter;
[0017] The solar radiation flow meter includes an antenna feed system, an analog receiver, and a digital receiver.
[0018] Furthermore, the turntable of the antenna system tracks the sun, the antenna points to the sun, receives solar radiation signals, converts them into solar electromagnetic wave signals, and sends them to the analog receiver;
[0019] The analog receiver receives solar electromagnetic wave signals, which are then amplified, power divided, filtered, and detected, and converted into analog voltage signals before being sent to the digital receiver.
[0020] The digital receiver consists of an ADC acquisition card and an FPGA connected in sequence. The ADC acquisition card acquires the analog voltage signal into a digital signal. The FPGA performs smooth integration based on the continuously input data to obtain the solar radio current value, which is then transmitted to the industrial control computer via a serial port.
[0021] Furthermore, the multiple judgments specifically refer to:
[0022] First judgment: Compare the predicted value of the radio current with the burst threshold. If the predicted value is not less than the burst threshold, proceed to the second judgment.
[0023] The second judgment: According to the Wright criterion, compare the predicted value of the radio current with the variance at the current time. If the predicted value is less than three times the variance at the current time, it is determined that the predicted value is not an outlier, and then proceed to the third judgment.
[0024] The third judgment: compare the actual observed value at the current moment with the burst threshold. If the observed value is greater than the burst threshold, a solar burst occurs; otherwise, calculate the average of the actual observed values at Z consecutive moments, compare the average value with the burst threshold, and if the average value is greater than the burst threshold, a solar burst occurs. Here, Z is the first preset number.
[0025] Furthermore, the solar eruption monitoring results include whether a solar eruption occurred and the eruption level;
[0026] The determination of the eruption level is specifically as follows: calculate the average value of the actual observed values of N consecutive solar eruption moments, and divide the solar eruption into three levels through two preset level thresholds, where N is the second preset number.
[0027] A second aspect of the present invention provides a real-time monitoring system for solar flares based on CNN-LSTM.
[0028] A real-time solar eruption monitoring system based on CNN-LSTM includes a model building module, a real-time prediction module, and a result judgment module.
[0029] The model building module is configured to: build a solar radio flux prediction model based on CNN-LSTM, and train the model using historical solar radio observation data;
[0030] The real-time prediction module is configured to: observe solar radio current values in real time, form time-domain data, input it into a trained solar radio current prediction model, and generate and output the predicted radio current value for the next moment.
[0031] The result judgment module is configured to perform multiple judgments based on the predicted radio current value to obtain the solar eruption monitoring result.
[0032] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of the CNN-LSTM-based real-time monitoring method for solar eruptions as described in the first aspect of the present invention.
[0033] The fourth aspect of the present invention provides an electronic device including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the CNN-LSTM-based real-time monitoring method for solar flares as described in the first aspect of the present invention.
[0034] The above one or more technical solutions have the following beneficial effects:
[0035] This invention employs multiple judgments to reduce outlier interference and improve the accuracy of solar radio burst detection. Compared to common solar radio observation instruments, the solar radiation flowmeter solution reduces the number of devices required for burst detection, thus lowering costs. Fully automated burst monitoring reduces manual operation and significantly improves the level of intelligence. Real-time data is used for prediction and burst detection, improving the real-time performance of the detection.
[0036] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0037] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0038] Figure 1 This is a flowchart of the method in the first embodiment.
[0039] Figure 2 This is a structural diagram of the solar radio flux prediction model for the first embodiment.
[0040] Figure 3 This is a structural diagram of the solar radio current detection device according to the first embodiment.
[0041] Figure 4 This is a system structure diagram of the second embodiment. Detailed Implementation
[0042] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0043] Example 1
[0044] The specific solution for this embodiment can be implemented by referring to the following content:
[0045] Currently, the identification of solar radio bursts mostly involves analyzing and processing data observed on the same day, which has the characteristic of poor real-time performance. In addition, due to the need to eliminate interference events, it requires professional personnel to handle the process. Therefore, a real-time monitoring method for solar bursts based on CNN-LSTM is proposed.
[0046] This embodiment processes solar radio current data observed by a solar radiation flowmeter to obtain solar radio time-domain data, which is then input into a pre-trained CNN-LSTM network in an industrial control computer to predict the solar radio current value at the next moment in real time. Based on the multi-judgment mechanism proposed by this invention, it determines whether an eruption will occur; when an eruption is detected, the data is recorded and the intensity is judged. Simultaneously, real-time early warning of solar eruptions is provided through the industrial control computer software interface and the cloud platform. This embodiment effectively solves the problem of real-time detection of solar radio eruptions by using a high-performance industrial control computer for network training, solar radio current prediction, and eruption detection; furthermore, the eruption judgment mechanism can effectively improve the accuracy of solar eruption early warning.
[0047] like Figure 1 As shown, the real-time monitoring method for solar flares based on CNN-LSTM specifically includes:
[0048] Step S1: Construct a solar radio flux prediction model based on CNN-LSTM and train the model using historical solar radio observation data.
[0049] This embodiment constructs a solar radio current prediction model based on CNN-LSTM and attention mechanism for real-time prediction of solar radio current values at the next moment.
[0050] Convolutional Neural Networks (CNNs) leverage the advantages of convolution operations to represent raw data at a higher level and with greater abstraction. Time-series data often exhibits local correlations, making CNNs even better suited for processing local features. The specific formula for convolution operations is as follows:
[0051]
[0052] Among them, y k-1 This is the input to the k-th convolutional layer, and * represents the convolution operation. Let be the weights of the i-th convolutional kernel in the k-th convolutional layer. This is the bias of the k-th convolutional layer.
[0053] The LSTM network, by adding forget gates, input gates, and output gates, makes the model more suitable for predicting long-term data. LSTM learns the temporal features in the input data, and by adding an attention mechanism, it can learn long-term data of solar radio bursts, reducing the gradient vanishing problem.
[0054] The conventional attention mechanism calculates channel weights using global average pooling, which leads to the loss of data spatial features. Therefore, this embodiment chooses to use the CBAM mechanism, which, based on the channel attention mechanism, uses average pooling to obtain the corresponding weights for the spatial dimension, thereby realizing the function of two-dimensional serial allocation of attention in both space and channel.
[0055] Figure 2 It is a structural diagram of the model, such as Figure 2 As shown, it mainly consists of CNN layers, Dropout layers, LSTM layers, average pooling layers, max pooling layers, and fully connected layers.
[0056] The raw input data is filtered and processed before being used as input to the model. Local spatial features are obtained through CNN layers. Temporal features are extracted using LSTM layers. Gradient vanishing is reduced by using Sigmoid and Tanh activation functions and Dropout layers. The temporal and spatial features extracted by CNN-LSTM are integrated through an attention mechanism and the solar radio flux prediction results are output through two fully connected layers.
[0057] Using solar radio observation data from Shandong University's Chashan Observatory from 2020 to the present as the raw data, and based on data from other observatories in similar frequency bands around the world, data with bursts were selected from the raw data, and incomplete data were removed. This data was used as the model training dataset. The dataset was trained using a time-domain data stream approach, with time-series solar radio observation data as input and the radio current value at the next moment as the output, to train the model's radio current prediction capability.
[0058] The model is trained on a high-performance industrial control computer. After training, in order to ensure the real-time operation of the data stream, i.e. real-time observation, real-time prediction, and real-time judgment, a solar radio flux detection device is built. The following steps S2 and S3 are implemented based on this device. Figure 3 This is a structural diagram of a solar radio current detection device, such as... Figure 3 As shown, the system includes a solar radiation flow meter, an industrial control computer, and a disk array connected in sequence. The solar radiation flow meter is used for real-time observation, the industrial control computer is used for real-time prediction and judgment, and the disk array is used for data storage.
[0059] Step S2: Real-time observation of solar radio current values, after a series of processing steps, to form time-domain data, which is then input into a trained solar radio current prediction model to generate and output the predicted radio current value for the next time step. Specifically, this includes:
[0060] Step S201: The solar radiation flowmeter is used to observe the solar radiation current value in real time and upload it to the industrial control computer for further processing, such as... Figure 3As shown, the solar radiation flow meter includes an antenna feed system, an analog receiver, and a digital receiver. The specific workflow is as follows:
[0061] (1) The turntable of the antenna feeder system tracks the sun, the antenna points to the sun, receives the solar radiation signal, converts it into a solar electromagnetic wave signal, and sends it to the analog receiver.
[0062] (2) The analog receiver receives the solar electromagnetic wave signal, which is then converted into an analog voltage signal through amplification, power division, filtering, and detection operations, and sent to the digital receiver.
[0063] (3) Digital receiver, consisting of an ADC acquisition card and an FPGA connected in sequence. The ADC acquisition card acquires the analog voltage signal into a digital signal. The FPGA performs smooth integration based on the continuously input data to obtain the solar radiation current value, which is then transmitted to the industrial control computer via a serial port.
[0064] Step S202: After receiving the solar radio current value, the industrial control computer processes it into time-domain data and inputs it into the trained solar radio current prediction model to obtain the predicted radio current value for the next time step. Specifically:
[0065] The received solar radio current value is S[i][j], where i represents the current time and j represents the channel. The data before time i, together with the data at the current time, are used to construct time-domain data [S[i-n+1][j], ..., S[i-2][j], S[i-1][j], S[i][j]], which consists of observation data at n times. This data is then input into the solar radio current prediction model, and the predicted radio current value Y[i+1][j] at time i+1 is output.
[0066] Since solar radiation flux varies with solar activity, and the sun is divided into periods of maximum and minimum activity, the range of radiation flux differs across different activity cycles. To improve prediction accuracy, before activating the solar radio flux detection device and making predictions each day, the pre-trained solar radio flux prediction model is updated and warmed up using N points of pre-stored solar radio time-domain data from the previous day. This ensures that the predicted values output by the CNN-LSTM prediction model better reflect the current state of solar activity.
[0067] After preheating, the industrial control computer first finds the corresponding serial port and connects to the digital receiver. Then, it opens the serial port. If the serial port opens successfully, the connection is successful. Parameters can be set as needed. If the serial port fails to open, it continues to send commands until the connection is completed.
[0068] Once successfully connected, it can receive data from a digital receiver and perform predictions based on real-time solar radio time-domain data.
[0069] Based on the above steps, the solar radio flux observed by the solar radiation flowmeter is processed to obtain solar radio time-domain data, which includes two parts:
[0070] 1) Smoothing and filtering processing is performed inside the FPGA: When the analog receiver input signal is stored in the FPGA's FIFO, when the data storage time in the FIFO meets the integration time τ, a smoothing and filtering operation is performed, and then the real-time hexadecimal data is output to the industrial control computer.
[0071] FIFO is an IP core in an FPGA (which can be understood as a built-in function of the software), while Buffer is a data buffer in an FPGA used to temporarily store data.
[0072] 2) Output hexadecimal data and convert it to decimal in the industrial control computer.
[0073] Step S3: Based on the predicted radio flux value, perform multiple judgments to obtain the solar eruption monitoring results.
[0074] After obtaining the predicted value of radio current Y[i+1][j] at time i+1, it is necessary to determine whether a solar eruption has actually occurred. This embodiment proposes a real-time solar radio eruption judgment algorithm, which uses a multi-judgment mechanism to improve the accuracy of solar eruption monitoring.
[0075] Specifically, the real-time solar radio burst detection algorithm takes 6 sets of data as input and outputs 2 sets of data. The 6 sets of data are:
[0076] (1) Time-domain data of the current time i [S[i-n+1][j], ..., S[i-2][j], S[i-1][j], S[i][j]]
[0077] (2) The predicted value Y[i+1][j] for the next time step i+1.
[0078] (3) The variance δ[i] at the current time i
[0079] (4) Burst threshold K
[0080] (5) X-level outbreak threshold K1
[0081] (6) M-level burst threshold K2
[0082] The two sets of data are:
[0083] (1) Whether a solar flare occurred
[0084] (2) Outbreak Level
[0085] Based on the input and output of the above judgment algorithm, the specific judgment process for multiple judgments is as follows:
[0086] (1) First - level judgment: Compare the predicted radio - flux value with the burst threshold. If the condition that the predicted value is not less than the burst threshold is met, then enter the second - level judgment.
[0087] Compare the predicted value Y[i + 1][j] with the burst threshold K. If Y[i + 1][j]<K, then no burst occurs at this time, and continuously judge the magnitude relationship between the predicted value Y[i + 1][j] and the threshold K; if the condition is met, enter the first - level judgment.
[0088] (2) Second - level judgment: According to the Least - trimmed squares (LTS) criterion, compare the predicted radio - flux value with the variance at the current moment. If the condition that the predicted value is less than three times the variance at the current moment is met, it is determined that the predicted value is not an outlier, and then enter the third - level judgment.
[0089] Since the solar radiation flowmeter used for observation is set in the wild and is easily affected by external electromagnetic interference, resulting in spike - pulse situations. Such spike - pulses are usually short - time high - energy signals. To exclude the influence of such outliers, in this embodiment, according to the LTS criterion, when the predicted value is less than three times the variance, that is, Y[i + 1][j]<3δ[i], it is determined that the time - domain data at this moment is not an outlier. Then, the time - domain data from the i - th moment is stored in the buffer Buffer[j] of the corresponding channel j, and then enter the third - level judgment.
[0090] If the second - level judgment is not satisfied, it means that there is no solar burst or there is an outlier phenomenon, then return to the first - level judgment.
[0091] (3) Third - level judgment: Compare the actual observed value at the current moment with the burst threshold. If the condition that the observed value is greater than the burst threshold is met, then a solar burst occurs; otherwise, calculate the average value of the actual observed values at Z consecutive moments, and compare the average value with the burst threshold. If the condition that the average value is greater than the burst threshold is met, then a solar burst occurs, where Z is the first preset number.
[0092] After the first and second - level judgments are satisfied, perform the third - level judgment to further determine whether a real solar burst occurs; at this time, judge the magnitude relationship between the actual observed value S[i][j] at the current moment and the burst threshold:
[0093] When S[i][j] is greater than the set burst threshold K, that is, S[i][j]>K, it means that a solar burst occurs. The industrial control computer software gives a solar - burst prompt and simultaneously starts the cloud - service port to send a burst - prompt message.
[0094] If the conditions are not met, start the counter Count_T, Count_T = 0; when new data is stored in Buffer[j], increment it by one, i.e., Count_T = Count_T + 1; when the counter Count_T = Z, calculate the average value P[j] of the data stored in the buffer[j] of channel j. If the average value P[j] is greater than the burst threshold K, it proves that a burst has occurred. If it is less than the burst threshold K, there is no solar radio burst. Discard the pre-stored data in Buffer[j] and re-enter the first judgment.
[0095] Once the first burst condition is met, the start register CSR_A is activated and set to 1, i.e., CSR_A = 1, indicating that a burst has occurred at this moment; then, the burst level judgment and burst end judgment are performed in parallel loop.
[0096] Outbreak level determination process: Start the counter Count_B, Count_B = 0. When new data is stored in Buffer[j], increment it by one, i.e., Count_B = Count_B + 1. When Count_B = N, average the data in the buffer Buffer[j] of channel j to obtain the average value B[j] of N points. Take the maximum value B of the average values of all channels. max Compare B in turn max The outbreak level is determined by the magnitudes of K1 and K2, where K1 > K2, and N is a second preset number. The specific judgment logic is as follows:
[0097] When B max K1 represents an X-level burst;
[0098] When K1≥B max K2, on the other hand, is an M-level burst.
[0099] When K2≥B max That would be a C-level burst.
[0100] After the level determination is completed, Count_B is reset to 0. Count_B = 0, and the outbreak level is determined again in a loop. The industrial control computer updates the outbreak level in real time until the outbreak ends.
[0101] When both the predicted value and the actual observed value are less than the threshold K, the outbreak ends, the register CSR_A is set to 0, the outbreak detection and outbreak level judgment are terminated, and the data in Buffer[j] is uploaded to the disk array for storage.
[0102] The industrial control computer reads the collected solar radio burst data and determines whether a solar burst has occurred based on the aforementioned real-time solar radio burst judgment algorithm. When data meeting the solar burst judgment criteria is found, an alert is issued via pop-up window and SMS. The specific methods are as follows:
[0103] A pop-up window can be triggered within the industrial control computer to provide an alert to check if an outbreak has occurred.
[0104] The system sends out burst alerts via SMS, comparing the assessed burst level with the previous one. If the levels are the same, no update is sent to the SMS; the level is then determined again. If the levels are different, Visual Studio software reads the pre-set key and URL to connect to the cloud platform server for data transmission. The cloud platform then sends an SMS to a pre-set mobile phone number, notifying the user of a solar radio burst and displaying the latest burst level. The system then checks if the SMS was sent successfully. If it was not sent successfully, the system reconnects to the server until it is successfully sent. The first burst does not need to be compared with the previous one; the SMS alert is sent directly using the above method.
[0105] Adding CNN-LSTM time series prediction and Wright's criterion to the real-time detection of solar flares can effectively reduce the impact of spike pulses; introducing predicted values to form a triple-judgment burst detection mechanism can effectively improve the accuracy of solar flare determination.
[0106] Example 2
[0107] This embodiment discloses a real-time solar eruption monitoring system based on CNN-LSTM;
[0108] like Figure 4 As shown, the real-time solar eruption monitoring system based on CNN-LSTM includes a model building module, a real-time prediction module, and a result judgment module:
[0109] The model building module is configured to: build a solar radio flux prediction model based on CNN-LSTM, and train the model using historical solar radio observation data;
[0110] The real-time prediction module is configured to: observe solar radio current values in real time, form time-domain data, input it into a trained solar radio current prediction model, and generate and output the predicted radio current value for the next moment.
[0111] The result judgment module is configured to perform multiple judgments based on the predicted radio current value to obtain the solar eruption monitoring result.
[0112] Example 3
[0113] The purpose of this embodiment is to provide a computer-readable storage medium.
[0114] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the CNN-LSTM-based real-time solar eruption monitoring method as described in Embodiment 1 of this disclosure.
[0115] Example 4
[0116] The purpose of this embodiment is to provide an electronic device.
[0117] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the CNN-LSTM-based real-time solar eruption monitoring method as described in Embodiment 1 of this disclosure.
[0118] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for real-time monitoring of solar eruptions based on CNN-LSTM, characterized in that, include: A solar radio flux prediction model was constructed based on CNN-LSTM, and the model was trained using historical solar radio observation data. Real-time observation of solar radio current values forms time-domain data, which is then input into a trained solar radio current prediction model to generate and output the predicted radio current value for the next moment. Based on the predicted radio flux values, multiple judgments are made to obtain the solar eruption monitoring results; The multiple judgments are specifically as follows: First judgment: Compare the predicted value of the radio current with the burst threshold. If the predicted value is not less than the burst threshold, proceed to the second judgment. The second judgment: According to the Wright criterion, compare the predicted value of the radio current with the variance at the current time. If the predicted value is less than three times the variance at the current time, it is determined that the predicted value is not an outlier, and then proceed to the third judgment. The third judgment: compare the actual observed value at the current moment with the burst threshold. If the observed value is greater than the burst threshold, a solar burst occurs; otherwise, calculate the average of the actual observed values at Z consecutive moments, compare the average value with the burst threshold, and if the average value is greater than the burst threshold, a solar burst occurs. Here, Z is the first preset number.
2. The CNN-LSTM based solar burst real-time monitoring method of claim 1, wherein, The solar radio flux prediction model consists of a CNN layer, a Dropout layer, an LSTM layer, an average pooling layer, a max pooling layer, and a fully connected layer, and is trained on an industrial control computer.
3. The CNN-LSTM based solar burst real-time monitoring method of claim 1, wherein, The solar radio flux prediction model uses the CBAM mechanism. Based on the channel attention mechanism, it selects the spatial dimension and uses average pooling to obtain the corresponding weights, and then distributes attention serially in two dimensions: space and channel.
4. The CNN-LSTM based solar burst real-time monitoring method of claim 1, wherein, The real-time observation of solar radio current values is performed using a solar radiation flow meter. The solar radiation flow meter includes an antenna feed system, an analog receiver, and a digital receiver.
5. The real-time monitoring method for solar flares based on CNN-LSTM as described in claim 4, characterized in that, The turntable of the antenna system tracks the sun, the antenna points to the sun, receives solar radiation signals, converts them into solar electromagnetic wave signals, and sends them to the analog receiver. The analog receiver receives solar electromagnetic wave signals, which are then amplified, power divided, filtered, and detected, and converted into analog voltage signals before being sent to the digital receiver. The digital receiver consists of an ADC acquisition card and an FPGA connected in sequence. The ADC acquisition card acquires the analog voltage signal into a digital signal. The FPGA performs smooth integration based on the continuously input data to obtain the solar radio current value, which is then transmitted to the industrial control computer via a serial port.
6. The real-time monitoring method for solar flares based on CNN-LSTM as described in claim 1, characterized in that, The solar eruption monitoring results include whether a solar eruption occurred and the eruption level; The determination of the eruption level is specifically as follows: calculate the average value of the actual observed values of N consecutive solar eruption moments, and divide the solar eruption into three levels through two preset level thresholds, where N is the second preset number.
7. A real-time monitoring system for solar flares based on CNN-LSTM, characterized in that, It includes a model building module, a real-time prediction module, and a result evaluation module: The model building module is configured to: build a solar radio flux prediction model based on CNN-LSTM, and train the model using historical solar radio observation data; The real-time prediction module is configured to: observe the solar radio current value in real time, form time-domain data, input it into the trained solar radio current prediction model, and generate and output the predicted radio current value for the next moment. The result judgment module is configured to: perform multiple judgments based on the predicted radio current value to obtain the solar eruption monitoring result; The multiple judgments are specifically as follows: First judgment: Compare the predicted value of the radio current with the burst threshold. If the predicted value is not less than the burst threshold, proceed to the second judgment. The second judgment: According to the Wright criterion, compare the predicted value of the radio current with the variance at the current time. If the predicted value is less than three times the variance at the current time, it is determined that the predicted value is not an outlier, and then proceed to the third judgment. The third judgment: compare the actual observed value at the current moment with the burst threshold. If the observed value is greater than the burst threshold, a solar burst occurs; otherwise, calculate the average of the actual observed values at Z consecutive moments, compare the average value with the burst threshold, and if the average value is greater than the burst threshold, a solar burst occurs. Here, Z is the first preset number.
8. An electronic device, characterized in that it comprises: Memory is used to store computer-readable instructions in a non-transitory manner. as well as Processor, for executing the computer-readable instructions, When the computer-readable instructions are executed by the processor, they perform the method described in any one of claims 1-6.
9. A storage medium, characterized in that, The computer-readable instructions are stored non-temporarily, wherein when the computer-readable instructions are executed by a computer, the instructions of the method according to any one of claims 1-6 are executed.