A method and system for predicting coronal mass ejection events

Abstract

The present application provides a method and system for predicting a coronal mass ejection event, the method comprising: collecting relevant data of CME within a set time period, obtaining characteristic parameter values ​​of 10 coronal mass ejection events after preprocessing; inputting the characteristic parameter values ​​of the 10 coronal mass ejection events into a pre-established and trained CME judgment model to obtain a prediction of whether the CME will reach the earth; based on the parameter weights obtained by the CME judgment model, the characteristic parameter values ​​of the 10 coronal mass ejection events are matched with the historical events in the pre-established CME historical database to find the closest historical event; for the CME predicted to reach the earth, the arrival time of the earth is calculated. The advantage of the present application is that in terms of ensuring the completeness of the input features, it largely solves the accuracy and refinement requirements of the prediction model, and has special reference significance for the current space weather business forecast.

Description

Technical Field

[0001] The present application belongs to the interdisciplinary field of space physics and artificial intelligence, and the field of space environment forecasting, and specifically relates to a method and system for predicting coronal mass ejection events. Background Art

[0002] Coronal mass ejections (CMEs) are instantaneous solar explosions that can cause strong disturbances in the space environment, leading to geomagnetic storms, high-energy electron storms, hot plasma injections, ionospheric storms, and increased density in the middle and upper atmosphere. This series of changes will affect communications, navigation systems, aviation activities, and power grids. Among them, the aspect that is directly related to people's lives is the aurora phenomenon caused by geomagnetic storms. Therefore, accurate predictions of CMEs play an indispensable role in improving the ability to predict geomagnetic storms and auroras. Therefore, building an accurate coronal mass ejection prediction model can not only avoid potential damage and asset losses, but also has important significance for the actual forecasting business and people's daily lives.

[0003] At present, there are few studies on models based on recommending similar events of coronal mass ejections at home and abroad. Most of the research models for coronal mass ejection prediction still rely on manual experience prediction, and the manual experience prediction needs to fully consider the effects of similar historical events, and then combine the uniqueness of the current event to give the final coronal mass ejection prediction. At present, most models predict the arrival time of coronal mass ejections, without recommendation function, and the error of the predicted time is large, the result is unstable, and there is a great deal of randomness. In addition, there is no model system for predicting coronal mass ejection events at home and abroad. Summary of the invention

[0004] The purpose of this application is to overcome the defects of existing coronal mass ejection event prediction methods, such as large errors in prediction time, unstable results, and great randomness.

[0005] In order to achieve the above objectives, the present application proposes a method for predicting a coronal mass ejection event, the method comprising:

[0006] The central position angle of CME, the position angle of the fastest moving CME front edge, angular width, linear velocity, F10.7 index, average solar wind speed, temperature, proton density, southward component of interplanetary magnetic field and total magnetic field are collected within a set time period, and the characteristic parameter values ​​of 10 coronal mass ejection events are obtained after preprocessing;

[0007] Input the characteristic parameter values ​​of 10 coronal mass ejection events into the pre-established and trained CME judgment model to obtain a prediction of whether the CME will reach the Earth;

[0008] Based on the parameter weights obtained from the CME judgment model, the characteristic parameter values ​​of 10 types of coronal mass ejection events are distance-matched with historical events in the pre-established CME historical database to find the closest historical event; for CMEs predicted to reach the Earth, the time of arrival at the Earth is calculated.

[0009] As an improvement of the above method, the CME judgment model is implemented based on a logistic regression algorithm;

[0010] The CME judgment model is trained for multiple iterations to obtain the best parameter weights; a suitable threshold is selected according to the skill scoring standard, and a predicted probability greater than the threshold indicates that the CME will reach the earth.

[0011] As an improvement of the above method, the skill scoring standard is: F1 score>recall rate>false alarm rate>success rate>accuracy rate.

[0012] As an improvement of the above method, finding the closest historical event includes: calculating the cosine distance and Euclidean distance of the characteristic parameter values ​​of 10 weighted coronal mass ejection events and the characteristic parameter values ​​of historical events in a pre-established CME historical database, respectively, to obtain two closest historical events.

[0013] As an improvement of the above method, the calculation of the time of arrival on Earth includes: averaging the time of arrival on Earth of two historical events that are closest to each other as the time of arrival on Earth.

[0014] As an improvement of the above method, the CME history database includes a positive sample database and a negative sample database;

[0015] The positive sample database includes CME events that have reached the Earth in history;

[0016] The negative sample database includes CME events that have never reached the Earth in history;

[0017] When looking for the closest historical event, when the CME judgment model result is that it will reach the earth, the historical database used is the positive sample database; when the CME judgment model result is that it will not reach the earth, the historical database used is the negative sample database.

[0018] As an improvement to the above method, when preprocessing the collected data, the data is not normalized.

[0019] The present application also provides a coronal mass ejection event prediction system, which is implemented based on the above method, and the system includes:

[0020] The data acquisition module is used to collect the central position angle of CME, the position angle of the fastest moving CME leading edge, angular width, linear velocity, F10.7 index, average solar wind speed, temperature, proton density, southward component of interplanetary magnetic field and total magnetic field within a set time period, and obtain the characteristic parameter values ​​of 10 coronal mass ejection events after preprocessing;

[0021] The Earth-reaching prediction module is used to input the characteristic parameter values ​​of 10 CME events into the pre-established and trained CME judgment model to obtain a prediction of whether the CME will reach the Earth; and

[0022] The event matching and calculation of arrival time on Earth module is used to match the characteristic parameter values ​​of 10 types of coronal mass ejection events with historical events in the pre-established CME historical database based on the parameter weights obtained by the CME judgment model, and find the closest historical event; for CMEs predicted to reach the Earth, the arrival time on Earth is calculated.

[0023] Compared with the prior art, the advantages of this application are:

[0024] 1. For the first time, logistic regression and recommendation algorithms are combined to build a CME recommendation model, which not only enriches the prediction function of the model, but also greatly shortens the training time of the model and simplifies the model construction process;

[0025] 2. Applying the weight of logistic regression to the recommendation process greatly reduces the training time and computing resource consumption in the recommendation stage;

[0026] 3. Compared with the existing recommendation system for the arrival time of coronal mass ejections, the parameters used in this application are simplified, which is convenient and quick to use, and can better meet the timeliness of actual forecasting business;

[0027] 4. Compared with the existing recommendation system for the arrival time of coronal mass ejections, the steps for normalizing parameters are simplified. Since the extreme value normalization adopted by the existing system is only applicable to situations with obvious boundaries, the extreme value problem needs to be fully considered. When new data is added, the extreme value will change. This requires the model to be adjusted at any time according to the size of the actual parameters during subsequent use to ensure the normal operation of the model. The purpose of normalization is to speed up the speed of gradient descent to find the optimal solution, that is, to speed up the convergence of the training network. Through control experiments, it was found that the training time consumed is not much different whether the parameters are normalized or not, and the results are similar. Therefore, it was finally decided that this application does not need to normalize the parameters, which not only simplifies the entire model training process, but also saves the subsequent model maintenance work;

[0028] 5. Compared with the existing recommendation system for the arrival time of coronal mass ejections, the prediction function of this application is more comprehensive, and it solves the two most concerned issues in the research of coronal mass ejection prediction: Can it arrive? When will it arrive? In addition, it can not only make up for the fact that most current models only give a single numerical prediction result, but also provide real historical similar events as a reference, which has a stronger reference value for business forecasting;

[0029] 6. The prediction accuracy of this application can reach 86%, far exceeding the average level of other systems;

[0030] Therefore, this model has largely solved the accuracy and refinement requirements of the model in terms of ensuring the completeness of input features. More importantly, the recommendation of similar events has special reference significance for the current space weather operational forecast. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Shown is a schematic diagram of the principle of the method for predicting coronal mass ejection events;

[0032] Figure 2 Shown is a flow chart of the coronal mass ejection event prediction method. DETAILED DESCRIPTION

[0033] The technical solution of the present application is described in detail below with reference to the accompanying drawings.

[0034] In the actual space environment forecasting business, the prediction of the impact of coronal mass ejections on the earth has always been one of the difficulties. In daily forecasts, special attention is paid to forecast contents such as whether the coronal mass ejection will reach the earth and the arrival time. At present, most of the forecast work for coronal mass ejections in business forecasts relies on manual experience, and the forecast based on manual experience needs to fully refer to similar historical events, and then combine the characteristics of the current event to give the final forecast result. Therefore, constructing a coronal mass ejection prediction model based on logistic regression and recommendation algorithms can provide important reference value for forecasters in actual forecasting business. Compared with general prediction models, this application not only predicts whether CME will reach the earth and the arrival time, and realizes numerical forecasting, but the key point is that it can provide similar coronal mass ejection events in history, which is of great reference value for actual business forecasting.

[0035] The coronal mass ejection event prediction method and system provided in this application can not only predict whether a coronal mass ejection will reach the earth and the time of arrival, but also recommend coronal mass ejection events with similar history, providing a reference for the current space weather forecasting business.

[0036] Embodiment 1:

[0037] like Figure 1 and Figure 2As shown, Example 1 of the present application proposes a method for predicting a coronal mass ejection event, the method comprising:

[0038] The central position angle of CME, the position angle of the fastest moving CME front edge, angular width, linear velocity, F10.7 index, average solar wind speed, temperature, proton density, southward component of interplanetary magnetic field and total magnetic field are collected within a set time period, and the characteristic parameter values ​​of 10 coronal mass ejection events are obtained after preprocessing;

[0039] Input the characteristic parameter values ​​of 10 coronal mass ejection events into the pre-established and trained CME judgment model to obtain a prediction of whether the CME will reach the Earth;

[0040] Based on the parameter weights obtained from the CME judgment model, the characteristic parameter values ​​of 10 types of coronal mass ejection events are distance-matched with historical events in the pre-established CME historical database to find the closest historical event; for CMEs predicted to reach the Earth, the time of arrival at the Earth is calculated.

[0041] From the formation of data, the application of principles to the final system construction, each step is crucial and indispensable. The following will introduce the content of each part in detail.

[0042] 1. Data

[0043] The data currently used is derived from the SOHO LASCO CME list, which contains all events of the SOHO satellite from 1996 to August 2022. Among them, the CME events that have reached the earth in history are considered to be positive samples, and vice versa. In order to balance the number of positive and negative samples and reduce the redundancy of the data, the data was cleaned. According to the two conditions of missing data and CME angular width less than 90 degrees, 4866 events were eliminated. At the same time, the cleaned events are divided into positive sample historical event library and negative sample historical event library according to positive and negative samples for the construction of subsequent models.

[0044] Compared with the existing recommendation system for the arrival time of coronal mass ejections, this application simplifies the steps of normalizing the parameters. Since the extreme value normalization adopted by the existing system is only applicable to situations with obvious boundaries, the extreme value problem needs to be fully considered. When new data is added, the extreme value will change. This requires the model to be adjusted at any time according to the size of the actual parameters during subsequent use to ensure the normal operation of the model. The purpose of normalization is to speed up the speed of gradient descent to find the optimal solution, that is, to speed up the convergence of the training network. Through control experiments, it was found that the training time consumed was not much different whether the parameters were normalized or not, and the results were similar. Therefore, it was finally decided that this application does not need to normalize the parameters, which not only simplifies the entire model training process, but also saves the subsequent model maintenance work.

[0045] From the perspective of practicality and taking into full account the physical laws involved in model construction, the model input features include CPA (CME central position angle, Central Position Angle), MPA (Measurement Position Angle of the fastest moving CME front edge), angular width, linear velocity, F10.7 index (the flux of optical radiation with a wavelength of 10.7cm emitted by the solar atmosphere), average solar wind speed, temperature, proton density, interplanetary magnetic field south component and total magnetic field, a total of 10 features. Compared with the existing coronal mass ejection space environment event similarity recommendation algorithm, in order to ensure the integrity of input parameters and the accuracy of prediction results, the characteristic parameters are reduced from 18 to 10. Reducing the types of input parameters has many advantages: first, it not only greatly reduces the redundancy of the model, but also reduces the size of the model, making the model run faster; second, some parameters involved in the existing model are difficult to obtain in time in actual forecasting business, which greatly reduces the timeliness and effectiveness of the forecast. In order to make the model more convenient and faster in practical applications and meet the timeliness requirements of actual business, 10 parameters that are easier to obtain from real-time data are finally selected as the input of the model; finally, experiments have proved that by selecting 10 feature parameters as input, the accuracy of the model still maintains the original level or even improves slightly. Therefore, it is very necessary to reduce the number of input parameters.

[0046] 2. Principle

[0047] The coronal mass ejection event prediction method and system provided in this application are based on logistic regression and recommendation algorithms, and include two types of algorithms:

[0048] One is the logistic regression algorithm. Logistic regression is often used for dichotomy. The binary logistic regression model used in this application is as follows:

[0049]

[0050] Among them, ω T x=ω0+ω1x1+ω2x2+ω3x3+…+ω n x n ;ω T x is the linear regression fitting function, and its range is [-∞,+∞]. i (i=1,2,…,n) are input parameters. ω0 is a constant, ω i (i=1,2,…,n) is the parameter x i Yes T The coefficient of the contribution of the x value, that is, the weight value. Sigmoid function g ω(x) finally maps the input parameters to the corresponding classification probabilities. T When the x value approaches positive infinity, the Sigmoid function is closer to 1, which means that the predicted CME event is a ground-reaching event.

[0051] The function is continuously optimized to obtain a set of suitable coefficients (ω) to minimize the value of the loss function. The loss function of logistic regression is:

[0052]

[0053] Among them, y (i) is the true value of the sample, x (i) is the eigenvalue. The loss function value used to average the m sample values.

[0054] Using the loss function, the gradient descent method can be used to solve the optimal coefficient (ω), and when the optimal coefficient is obtained, the optimal model is established. Gradient descent is to find the descent direction by taking the first-order derivative of J(ω) with respect to ω, and iteratively update the coefficient. The update method is:

[0055]

[0056] Where α is the learning rate, which is usually less than 1 and is used to control the size of the movement during gradient descent. k is the number of iterations, which is calculated by comparing || J(ω k+1 )-J(ω k )|| is less than a threshold or reaches a maximum number of iterations to stop the iteration. The learning rate, threshold and maximum number of iterations can be set manually.

[0057] The second is the recommendation algorithm. The recommendation algorithm measures the similarity between sample events by calculating the distance between them. The smaller the distance between two events, the more similar the two events are. The recommendation algorithm combines the cosine distance and the Euclidean distance to build a model. The following introduces the calculation principles of the two algorithms:

[0058] (1) Cosine distance

[0059] Cosine distance compares the similarity between two events by the cosine value of the angle between two vectors in n-dimensional space. The formula of cosine distance in n-dimensional space is as follows:

[0060]

[0061] A i =(ω1a1,ω2a2,…ω i a i ),B i=(ω1b1,ω2b2,…ω i b i )

[0062] In the above formula, A i and B i Respectively represent the spatial vectors of the two events to be compared, corresponding to different coronal mass ejection events. i (i=1,2,…,n) and b i (i=1,2,…,n) represents physical parameters, each parameter has a corresponding weight ω i , the subscript i represents the i-th dimension, that is, the sequence number of the feature. According to the characteristics of the cosine function, the distance between two events generally falls within the interval [-1, 1]. The larger the cosine value, the smaller the angle between the two vectors, which means the two events are more similar. Therefore, when building a system, the similarity between two events is measured by the size of the cosine value.

[0063] (2) Euclidean distance

[0064] Euclidean distance is also a commonly used distance formula in recommendation algorithms. Unlike cosine distance, it usually refers to the distance between two points in n-dimensional space, and is sometimes expressed as the length of a vector (i.e., the distance from a point in space to the origin of the coordinate system). In general, the Euclidean distance formula is as follows:

[0065]

[0066] In the above formula, A i , B i ,ω i , a i and b i Has the same meaning as in the cosine distance formula.

[0067] 3. System construction

[0068] The coronal mass ejection event prediction method and system provided in this application can be established using Python language. The prediction method is divided into two steps:

[0069] Step 1: Predict whether a CME will reach the Earth based on the logistic regression algorithm;

[0070] Input 10 features of CME events, and the output is "yes" or "no". "Yes" means the predicted event will reach the Earth, and "no" means it will not reach the Earth. The framework used in this module is the logit function of statsmodels models in Python, hereinafter referred to as "sm.logit". Through multiple iterations of training, the best parameter weights are obtained. According to the skill score (F1 score > recall rate > false alarm rate > success rate > accuracy), the appropriate threshold is selected. If the predicted probability is greater than the threshold, it means the event will reach the Earth, and vice versa.

[0071] Step 2: Recommend similar events based on the weight of logistic regression;

[0072] Based on the two types of results predicted in step 1, similar events are recommended respectively: historically similar events are recommended for events that are predicted not to reach the earth, and historically similar events are recommended for events that are predicted to reach the earth, and the arrival time of the events is predicted, which serves as a reference for the actual forecasting business.

[0073] The main difference between the two types of event recommendations lies in the different historical databases used for matching. For events that are predicted not to arrive on Earth, similar historical events are recommended. The event database used for matching is a negative sample historical event database, which outputs two similar historical events; for events that are predicted to arrive on Earth, similar historical events are recommended. The event database used for matching is a positive sample historical event database, which outputs two similar historical events and predicts the arrival time. However, the algorithms for the two types of recommendations are the same, both using cosine distance and Euclidean distance. The specific process is as follows:

[0074] The binary classification result output from step 1 - "yes" or "no" is the starting command for step 2, pointing to different event recommendation processes respectively.

[0075] The input of step 2 is the same as the input of step 1, which is also the 10 parameters of the coronal mass ejection. After the parameters are input, the weights obtained by logistic regression training are applied to the corresponding parameters to make event recommendations based on the recommendation algorithm. Among them, based on the cosine distance, the most similar historical event can be obtained as the output of "historical similar event 1". Similarly, based on the Euclidean distance, the most similar historical event can also be obtained as the output of "historical similar event 2". In addition, the predicted arrival time is the average of the true values ​​of these two similar historical events as the final prediction result.

[0076] Embodiment 2:

[0077] The present application also provides a coronal mass ejection event prediction system, which is implemented based on the above method and includes:

[0078] The data acquisition module is used to collect the central position angle of CME, the position angle of the fastest moving CME leading edge, angular width, linear velocity, F10.7 index, average solar wind speed, temperature, proton density, southward component of interplanetary magnetic field and total magnetic field within a set time period, and obtain the characteristic parameter values ​​of 10 coronal mass ejection events after preprocessing;

[0079] The Earth arrival prediction module is used to input the characteristic parameter values ​​of 10 CME events into the pre-established and trained CME judgment model to obtain a prediction of whether the CME will reach the Earth; and

[0080] The event matching and calculation of arrival time on Earth module is used to match the characteristic parameter values ​​of 10 types of coronal mass ejection events with historical events in the pre-established CME historical database based on the parameter weights obtained by the CME judgment model, and find the closest historical event; for CMEs predicted to reach the Earth, the arrival time on Earth is calculated.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present application and are not intended to limit it. Although the present application is described in detail with reference to the embodiments, a person skilled in the art should understand that any modification or equivalent replacement of the technical solution of the present application does not depart from the spirit and scope of the technical solution of the present application and should be included in the scope of the claims of the present application.

Claims

1. A method for predicting a coronal mass ejection event, the method comprising: The central position angle of CME, the position angle of the fastest moving CME front edge, angular width, linear velocity, F10.7 index, average solar wind speed, temperature, proton density, southward component of interplanetary magnetic field and total magnetic field are collected within a set time period, and the characteristic parameter values ​​of 10 coronal mass ejection events are obtained after preprocessing; When preprocessing the collected data, the data is not normalized; Input the characteristic parameter values ​​of 10 coronal mass ejection events into the pre-established and trained CME judgment model to obtain a prediction of whether the CME will reach the Earth; Based on the parameter weights obtained from the CME judgment model, the characteristic parameter values ​​of 10 types of coronal mass ejection events are distance-matched with historical events in the pre-established CME historical database to find the closest historical event; for CMEs predicted to reach the Earth, the time of arrival at the Earth is calculated.

2. The method for predicting a coronal mass ejection event according to claim 1, wherein: The CME judgment model is implemented based on a logistic regression algorithm; The CME judgment model is trained for multiple iterations to obtain the best parameter weights; a suitable threshold is selected according to the skill scoring standard, and a predicted probability greater than the threshold indicates that the CME will reach the earth.

3. The method for predicting a coronal mass ejection event according to claim 2, wherein: The skill scoring standard is: F1 score > recall rate > false alarm rate > success rate > accuracy rate.

4. The method for predicting a coronal mass ejection event according to claim 1, wherein: The method of finding the closest historical event includes: calculating the cosine distance and the Euclidean distance of the characteristic parameter values ​​of the 10 weighted coronal mass ejection events and the characteristic parameter values ​​of the historical events in the pre-established CME historical database, respectively, to obtain the two closest historical events.

5. The method for predicting a coronal mass ejection event according to claim 4, wherein: The calculation of the time of arrival on the earth includes: taking the average of the time of arrival on the earth of two historical events that are closest to each other as the time of arrival on the earth.

6. The method for predicting a coronal mass ejection event according to claim 4, wherein: The CME historical database includes a positive sample database and a negative sample database; The positive sample database includes CME events that have reached the Earth in history; The negative sample database includes CME events that have never reached the Earth in history; When looking for the closest historical event, when the CME judgment model result is that it will reach the earth, the historical database used is the positive sample database; when the CME judgment model result is that it will not reach the earth, the historical database used is the negative sample database.

7. A coronal mass ejection event prediction system, implemented based on the method according to any one of claims 1 to 6, characterized in that: The system comprises: The data acquisition module is used to collect the central position angle of CME, the position angle of the fastest moving CME leading edge, angular width, linear velocity, F10.7 index, average solar wind speed, temperature, proton density, southward component of interplanetary magnetic field and total magnetic field within a set time period, and obtain the characteristic parameter values ​​of 10 types of coronal mass ejection events after preprocessing; when preprocessing the collected data, the data is not normalized; The Earth-reaching prediction module is used to input the characteristic parameter values ​​of 10 CME events into the pre-established and trained CME judgment model to obtain a prediction of whether the CME will reach the Earth; and The event matching and calculation of arrival time on Earth module is used to match the characteristic parameter values ​​of 10 types of coronal mass ejection events with historical events in the pre-established CME historical database based on the parameter weights obtained by the CME judgment model, and find the closest historical event; for CMEs predicted to reach the Earth, the arrival time on Earth is calculated.

Images