CN113344032A - Geomagnetic Kp index forecasting method and system based on similarity algorithm - Google Patents
Geomagnetic Kp index forecasting method and system based on similarity algorithm Download PDFInfo
- Publication number
- CN113344032A CN113344032A CN202110521691.XA CN202110521691A CN113344032A CN 113344032 A CN113344032 A CN 113344032A CN 202110521691 A CN202110521691 A CN 202110521691A CN 113344032 A CN113344032 A CN 113344032A
- Authority
- CN
- China
- Prior art keywords
- solar wind
- similarity
- index
- geomagnetic
- historical
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Optimization (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Pure & Applied Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Medical Informatics (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Computing Systems (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to interdisciplines of space physics and artificial intelligence, in particular to a geomagnetic Kp index forecasting method and a geomagnetic Kp index forecasting system based on a similarity algorithm. The method comprises the following steps: the method comprises the steps of receiving a current solar wind case, reading a pre-established historical solar wind case library, calculating the similarity between the current solar wind case and the historical solar wind case through a similarity calculation method, obtaining a plurality of historical solar wind cases with the highest similarity according to similarity ranking, and obtaining geomagnetic Kp index forecast corresponding to the current solar wind case according to Kp indexes corresponding to the historical solar wind cases. The invention applies the recommendation idea to geomagnetic index prediction, which is a brand new attempt; by adopting the method, the forecast result is vivid and friendly, and the artificial forecast experience can be added; the method of the invention has the advantages of simple input parameter prediction and advanced event prediction lead, and the prediction result is more accurate along with the continuous accumulation of historical data.
Description
Technical Field
The invention relates to interdisciplines of space physics and artificial intelligence, in particular to a geomagnetic Kp index forecasting method and a geomagnetic Kp index forecasting system based on a similarity algorithm, and is applied to a real business of space weather forecasting.
Background
The solar wind is supersonic plasma flow formed by the outward flow of the solar high-rise atmosphere. When the solar wind generated by corona substance ejection events or cavities reaches the earth, more energy and particles are injected into the magnetic layer, which may cause the magnetic layer to perturb, causing geomagnetic storms. Geomagnetic storms can seriously affect the performance and safety of aircraft, such as satellites, operating therein, and thus affect the technical systems throughout the space and the ground. Therefore, the method has important practical value for timely forecasting the geomagnetic disturbance caused by the solar wind.
At present, the technology of machine learning is broken through continuously, and the self-adaptive and self-learning strong parallel information processing capability of the machine learning obtains breakthrough progress in many aspects. Meanwhile, the space environment monitoring has long-term accumulation, and the geomagnetic Kp index which is one of important indexes for measuring the near-earth space global magnetic disturbance intensity has uninterrupted data for 80 years; the solar wind aspect also accumulates decades of satellite monitoring data. By combining the advantages of the two aspects, the geomagnetic disturbance prediction can be more effectively carried out.
Similarity calculation is a core thought of machine learning unsupervised clustering problems and recommendation algorithms, and at present, similarity calculation methods are various in variety, and have corresponding and relatively suitable algorithms for various problems. In the practical application of similar recommendations, many scenes achieve ideal results.
The conventional geomagnetic Kp index forecasting model is mainly based on the statistical relationship between the Kp index and solar wind parameters, the statistical relationship between the Kp index and the interior of a magnetic layer, the relationship between the Kp index and an energy coupling function and an artificial neural network, and a precedent for forecasting the Kp index based on a recommended idea is not provided. In the existing model, due to the complexity of the physical principle and the physical process, the types of commonly required parameters are many, the accuracy is uneven, and the time lead of the forecast is short.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a geomagnetic Kp index forecasting method and a geomagnetic Kp index forecasting system based on a similarity calculation method.
In order to achieve the above object, embodiment 1 of the present invention proposes a geomagnetic Kp index prediction method based on a similarity algorithm, the method including:
the method comprises the steps of receiving a current solar wind case, reading a pre-established historical solar wind case library, calculating the similarity between the current solar wind case and the historical solar wind case through a similarity calculation method, sequencing according to the similarity to obtain a plurality of historical solar wind cases with the highest similarity, and obtaining geomagnetic Kp index forecast corresponding to the current solar wind case according to Kp indexes corresponding to the historical solar wind cases.
As an improvement of the above method, the method further comprises the step of establishing a historical solar wind case library; the method specifically comprises the following steps:
acquiring historical solar wind parameter data and a geomagnetic Kp index;
carrying out data cleaning and normalization processing on the solar wind parameter data, and carrying out proximity value conversion on the geomagnetic Kp index;
selecting solar wind characteristic parameters by adopting an XGboost algorithm, and screening out two characteristic parameters which have the greatest influence on the geomagnetic Kp index;
calculating the weight of the influence of the two characteristic parameters on the geomagnetic Kp index by adopting a single-layer perceptron;
dividing historical solar wind parameter data into a plurality of solar wind cases by a sliding window with fixed step length, and defining a Kp index immediately after the historical solar wind parameter data as the Kp index corresponding to the solar wind case.
As an improvement of the method, the XGboost algorithm is adopted to select the solar wind characteristic parameters, and two characteristic parameters which have the greatest influence on the geomagnetic Kp index are screened out; the method specifically comprises the following steps:
the method comprises the steps of taking the average value of the sun wind speed V, the sun wind proton density N, the sun wind temperature T, the interplanetary magnetic field intensity B, the interplanetary magnetic field By and the interplanetary magnetic field Bz component in 1 hour as input, taking the corresponding Kp index as a label, iteratively establishing a decision tree through an XGboost model, scoring six input parameters, and taking two characteristic parameters with the highest scores.
As an improvement of the above method, the single-layer perceptron is adopted to calculate the weight of the influence of the two characteristic parameters on the geomagnetic Kp index; the method specifically comprises the following steps:
the two characteristic parameters are used as input layer parameters of the single-layer perceptron, the time resolution is 1 hour, the output layer is Kp indexes corresponding to each hour, the mean square error function is used as a loss function, the Adam algorithm is used as an optimizer, and the weight of the two characteristic parameters on the influence of the geomagnetic Kp indexes is obtained.
As a refinement of the above method, the size of the sliding window is divided in time resolution by Kp index, the sliding window step size being in the range of 0.5 to 1 hour.
As an improvement of the above method, the similarity between the current solar wind case and the historical solar wind case is calculated by a similarity calculation method; the method specifically comprises the following steps: and calculating the similarity of the current solar wind case and the historical solar wind case by using a dynamic time warping algorithm or a piecewise linear representation algorithm.
As an improvement of the above method, the calculating the similarity between the current solar wind case and the historical solar wind case by using the dynamic time warping algorithm specifically includes:
the time sequences of the current solar wind case and each historical solar wind case are subjected to local nonlinear scaling, so that the Euclidean distance accumulated value after the two sequence forms are aligned is the minimum distance of the two time sequences, and the similarity of the current solar wind case and each historical solar wind case is obtained.
As an improvement of the above method, the calculating the similarity between the current solar wind case and the historical solar wind case by using a piecewise linear representation algorithm specifically includes:
the method comprises the steps of adopting a piecewise linear expression algorithm for time sequences of a current solar wind case and each historical solar wind case to obtain a plurality of modes respectively, then conducting equal patterning to enable the lengths of the mode sequences of the current solar wind case and the historical solar wind cases to be equal, then obtaining a distance value of each mode by using an absolute value distance or Euclidean distance calculation method, and conducting weighting calculation on the mode distance values to obtain the similarity between the current solar wind case and each historical solar wind case.
The embodiment 2 of the present invention provides a geomagnetic Kp index prediction system based on a similarity algorithm, where the system includes: the system comprises a historical solar wind case library, a solar wind case receiving module, a similarity calculation module and a geomagnetic Kp index prediction output module; wherein the content of the first and second substances,
the solar wind case receiving module is used for receiving the current solar wind case;
the similarity calculation module is used for reading a pre-established historical solar wind case library and calculating the similarity between the current solar wind case and the historical solar wind case through a similarity calculation method;
the geomagnetic Kp index prediction output module is used for obtaining a plurality of historical solar wind cases with the highest similarity according to the similarity sequence, and obtaining geomagnetic Kp index prediction corresponding to the current solar wind case according to Kp indexes corresponding to the historical solar wind cases.
Compared with the prior art, the invention has the advantages that:
1. the invention applies the recommendation idea to geomagnetic index prediction, which is a brand new attempt;
2. the method of the invention can lead the forecast result to be vivid: instead of a single geomagnetic index as an output, several similar historical solar wind cases are taken as outputs (including time-series change images of various characteristics of the historical solar wind and the geomagnetic influence (subsequent geomagnetic Kp index)) caused by the time-series change images);
3. the method of the invention ensures that the forecast result is friendly: for a new forecaster, even if the previous typical solar wind events are not known, a recommended solar wind result can be given like a 'hundred degree search';
4. by adopting the method of the invention, the forecast result can be added into the artificial forecast experience: the former model is limited by a black box process, only a single result is returned, but the current model does not provide a predicted value, and can also provide the time sequence change condition of historical solar wind characteristic parameters, and compared with the time sequence change of the current input solar wind characteristic parameters, a forecaster can add own forecasting experience into an image to select or adjust the predicted value;
5. the method of the invention can simplify the forecast input parameters: the traditional model has a plurality of types of input parameters, and the number of types of input characteristics can be reduced according to the XGboost characteristic selection result;
6. the method leads the advance of the forecast event to be advanced: historical cases have occurred, and we can see its subsequent long enough influence, so long as we can find the most similar case, its historical subsequent influence can be used as the reference for the current forecast;
7. by adopting the method, the forecasting result can be more accurate along with the continuous accumulation of the historical data.
Drawings
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2(a) is a matrix grid constructed of two time sequences;
FIG. 2(b) is a dynamic time warping algorithm principle;
fig. 3 is an equal-mode numeration of two time sequences in a piecewise linear representation.
Detailed Description
The method carries out prediction on the geomagnetic Kp index for the first time from the recommended angle. By recommending the obtained forecast result, not only the predicted value of the Kp index (the geomagnetic Kp index corresponding to the historical similar solar wind case is used as) but also the evolution process of the historical similar solar wind case can be output. The prediction result is more vivid, and a forecaster can add own prediction experience to geomagnetic index prediction according to the comparison between the historical solar wind condition and the existing solar wind condition.
The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and examples.
Example 1
As shown in fig. 1, embodiment 1 of the present invention provides a geomagnetic Kp index prediction method based on a similarity algorithm. The method comprises the following steps:
the method comprises the steps of receiving a current solar wind case, reading a pre-established historical solar wind case library, calculating the similarity between the current solar wind case and the historical solar wind case through a similarity calculation method, sequencing according to the similarity to obtain a plurality of historical solar wind cases with the highest similarity, and obtaining geomagnetic Kp index forecast corresponding to the current solar wind case according to Kp indexes corresponding to the historical solar wind cases.
1 data
The solar wind and interplanetary data originate from the OMNI website. The OMNI website integrates solar Wind data from IMP 8, Geotail, Wind and Ace satellites. The time range of the solar wind data selected by the model is from 1998 to 2019, the solar wind data is the monitoring data of the ACE satellite, and the time frequency of the data is 1 minute. Due to the fact that the solar wind data are missing, data cleaning is conducted on the missing data (the default value is-9999.9), and the data corresponding to the previous 1 minute are taken as individual default data values, so that the completeness of the data is guaranteed. The solar wind parameters provided By the OMNI website comprise solar wind speed V, solar wind proton density N, values of interstellar magnetic field By and Bz components under the coordinates of solar wind temperature T, GSM, and interstellar magnetic field strength B. In order to facilitate training and learning, the solar wind parameter data is subjected to normalization processing, and the influence of different dimensions on calculation is eliminated.
The geomagnetic Kp index is derived from results of the OMNI website integrated from the german center for geosciences research. For convenience of calculation and analysis, the Kp index is converted into a nearby numerical value, namely Kp index 30Conversion to 3, 3-conversion to 2.7, 3+Conversion to 3.3 and so on. According to the current general specification of space weather forecast service, the Kp index is between 5 and 6+(including 5-and 6+) Defined as small magnetic storm, with Kp index between 7-and 90(including 7-and 90) Defined as a large magnetic storm.
The model takes the solar wind data within 3 hours after the world time integral point as a solar wind case, and defines the Kp index immediately after the solar wind case as the Kp index corresponding to the solar wind case. The solar wind case precedes its corresponding Kp index by 1-3 hours in time.
Model 2 method
2.1 basic principle
The solar wind is a direct cause of geomagnetic storm, and therefore, the geomagnetic disturbance caused by solar wind with consistent characteristics should be similar. If the historical data is enough, the historical data can be always in the historical data
And searching for historical solar wind cases similar to the current solar wind. Geomagnetic variation caused by historical solar wind events can be used as a reference for forecasting the current Kp index.
The model mainly utilizes a similarity algorithm in machine learning to construct a recommendation for searching for similar solar wind. The specific steps are divided into 3 steps. Step 1, characteristic parameter screening. The solar wind characteristic parameters mainly comprise 6 parameters such as speed, density, temperature, interplanetary magnetic field and the like; in the similarity calculation, too many characteristic parameters may cause confusion of recommendation results, and parameters which mainly play a role in geomagnetic disturbance need to be selected as characteristic parameters of the similarity calculation. And step 2, calculating the weight of the parameter. In geomagnetic disturbance, various characteristic parameters of solar wind have different influences, and weight values of different characteristic parameters need to be calculated. And 3, constructing a recommendation model. And (3) taking the selected solar wind characteristic parameters as input, searching for similarity of solar wind characteristic time sequence changes based on a similarity algorithm in the robotics, and constructing a recommendation model for screening similar solar wind cases.
2.2 feature parameter selection
The XGboost algorithm is adopted by the model to select the characteristic parameters of the solar wind. The XGboost is a decision tree training model which scores each characteristic parameter according to the function and value of the input characteristic parameter in the iterative establishment of the decision tree.
The method comprises the steps of taking an average value of 1 hour of components of sun wind speed V, sun wind proton density N, sun wind temperature T, interplanetary magnetic field intensity B, interplanetary magnetic field By and Bz after normalization in 1998-2019 as input, taking a corresponding Kp index as a label, and iteratively establishing a decision tree through an XGboost model to score six input parameters. The grade is the importance degree of the XGboost model considering that different parameters influence the geomagnetic disturbance (Kp index). In order to avoid confusion of recommendation results caused by excessive characteristic parameters and increase of time cost of model operation, two parameters which are most important to influence on geomagnetic disturbance are selected as input parameters of the recommendation model.
2.3 parameter weight calculation
The model adopts a single-layer perceptron to evaluate the weight of the screened characteristic parameters on the influence of geomagnetic disturbance. The characteristic parameters are used as input layer parameters of the single-layer perceptron, the time resolution is 1h, the output layer is a corresponding Kp index per hour, the loss function is a mean square error function, and the optimizer is an adam (adaptive motion) algorithm
2.4 recommendation model construction
The model constructs a solar wind similarity recommendation model by a similarity algorithm in machine learning, and mainly adopts a dynamic time warping algorithm.
The euclidean distance, as in formula 1, refers to the real distance between two points in m-dimensional space, or the natural length of a vector, and the euclidean distance in two-dimensional and three-dimensional spaces is the actual distance between two points. When the input is two sets of characteristic parameters, the comprehensive difference between the two sets of characteristic parameters can be reflected.
Dynamic time warping (DTW algorithm for short) is one of popular algorithms for finding similarity of time sequences, and its principle is to construct a matrix grid of n × m, as shown in fig. 2(a), with euclidean distance D as a standard, and matrix elements (i, j) as distances D (Qi, Cj) between Qi and Cj, given two sequences Q ═ { Q1, Q2, …, Qi, …, Qn } and C ═ C1, C2, …, Cj, …, Cm }, whose lengths are n and m, respectively. And taking the matrix angle where the sequence starting segment is positioned as a boundary condition, and obtaining a path with the minimum distance accumulation value through dynamic programming while satisfying continuity and monotonicity constraints, namely the optimal path.
As shown in fig. 2(b), the dynamic time warping algorithm considers that the euclidean distance cumulative value after aligning two sequence morphologies as much as possible is the minimum distance between two time sequences by locally non-linearly scaling the time sequences.
Applying the recommended idea to geomagnetic index prediction is a completely new attempt, and has the main advantages that: (1) the forecast result is vivid: instead of a single geomagnetic index as an output, several similar historical solar wind cases are taken as outputs (including time-series variation images of various characteristics of the historical solar wind and the geomagnetic influence (subsequent geomagnetic Kp index)) caused by the time-series variation images) (2) the prediction results are friendly: for new forecasters, even if not aware of the previous typical solar wind events, it is possible to like "Baidu search" to give recommended results (3) forecast results can add to the manual forecast experience: the former model is limited by a black box process, only a single result is returned, but the current model does not provide a predicted value, and can also provide the time sequence change condition of historical solar wind characteristic parameters, and compared with the time sequence change of the current input solar wind characteristic parameters, a forecaster can add own forecasting experience into an image to select or adjust the predicted value (4), and the forecasting input parameters are simple: the prior model has a plurality of types of input parameters, and the number of input feature types can be reduced according to the XGboost feature selection result (5), and the lead of the forecast event is advanced: the historical case has already occurred, and we can see the influence of the history long enough, so long as we can find the most similar case, the subsequent influence of the history can be used as the reference (6) of the current forecast when the historical data amount is accumulated continuously, and the historical case similar to the input case can be found certainly, so the forecast result will be more and more accurate in the future.
The alternative points are as follows: instead of using a dynamic time warping algorithm, a PLR (piecewise linear representation) algorithm may be used to find similarity in time series variations. A hybrid recommendation can also be made using both methods. PLR: (piece wise linear representation). As shown in fig. 3. A time series of three states: rising, falling and unchanged we denote this state correspondence as [1, -1,0]. Assuming that there is a certain sequence of length S, we equally divide it into K segments. For each segment a slope is calculated, a positive slope indicating a rise, a negative slope indicating a fall, and a 0 indicating no change. Then I amOne can denote this sequence as [1,1,0, -1 ].]Such a sequence, combining adjacent identical patterns, we get [1,0, -1 ].]The sequence of (a). This is the PLR algorithm. Since we merge adjacent identical patterns, we must have the pattern sequence 1, -1,0 spaced, each pattern possibly spanning different lengths of time, and after merging the two time series, the sequence S1 may have N patterns, and the sequence S2 may have M patterns. We now want to digitize them into equal patterns. It is common to say that they use a common segmentation point to obtain the final pattern sequence of equal length. Next we calculate the distance. Here we use absolute distance, although it is also possible to use euclidean distance. The modal distance formula is: d ═ m1i-m2iI, obviously D belongs to {0, 1, 2}, the closer the distance is to 0, the more similar the representation mode; closer to 2, indicating less similarity in the pattern. Adding all the pattern distances together, i.e. the pattern distances of the time series:
but each pattern may span a different length of time, and the longer a pattern is, the more information it contains for the entire sequence, so we need to weight it. Thus:
wherein t iswi＝ti/tN，tiThe length of time spanned by the ith mode, tNIs the total time length.
The solar wind is a direct cause of geomagnetic storm, and therefore, the geomagnetic disturbance caused by solar wind with consistent characteristics should be similar. If the historical data is enough, historical solar wind cases similar to the current solar wind can be found in the historical data. Geomagnetic variation caused by historical solar wind events can be used as a reference for forecasting the current Kp index.
And obtaining a plurality of solar winds which are most similar to the current input historically through calculation, and taking similar historical results as recommendation results. The forecasting result not only contains a geomagnetic index forecasting value, but also contains the evolution comparison between the recommended historical solar wind and the currently input solar wind, so that a forecaster can better combine the self forecasting experience.
The recommended idea avoids the complex nonlinear relation between the solar wind parameters and the geomagnetic indexes which are not completely clear at present, and directly starts from similarity and combines with the manual experience of a forecaster, so that the forecasting result is more friendly, intuitive and interpretable, more information can be seen in the forecasting result, and the forecasting time advance and the forecasting accuracy are also advanced.
The method mainly utilizes a similarity algorithm in machine learning to construct a recommendation model for searching for similar solar wind.
The method comprises the following specific steps:
preparing:
1) cleaning historical data to ensure the completeness of the data;
2) normalizing the data to eliminate dimension influence of different parameters;
3) selecting characteristics from a plurality of solar wind parameters which possibly cause geomagnetic influence, and screening out two most important parameters;
4) dividing historical data of minute precision into a plurality of solar wind cases according to a sliding window (the size of the sliding window can be divided according to the time resolution of Kp index-three hours, and fixing the step length of 0.5 or 1 hour);
recommending:
1) the similarity of the input case and the historical solar wind case is calculated through a similarity calculation method (after the characteristic parameters are weighted), and the input case is sorted according to the similarity (the distributed calculation can be used, the efficiency is improved)
2) Recommendations (output)
Example 2
The embodiment 2 of the invention provides a geomagnetic Kp index forecasting system based on a similarity algorithm. The system comprises: the system comprises a historical solar wind case library, a solar wind case receiving module, a similarity calculation module and a geomagnetic Kp index prediction output module; wherein the content of the first and second substances,
the solar wind case receiving module is used for receiving the current solar wind case;
the similarity calculation module is used for reading a pre-established historical solar wind case library and calculating the similarity between the current solar wind case and the historical solar wind case through a similarity calculation method;
the geomagnetic Kp index prediction output module is used for obtaining a plurality of historical solar wind cases with the highest similarity according to the similarity sequence, and obtaining geomagnetic Kp index prediction corresponding to the current solar wind case according to Kp indexes corresponding to the historical solar wind cases.
Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention and are not limited. Although the present invention has been described in detail with reference to the embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (9)
1. A geomagnetism Kp index forecasting method based on a similarity algorithm, the method comprising:
the method comprises the steps of receiving a current solar wind case, reading a pre-established historical solar wind case library, calculating the similarity between the current solar wind case and the historical solar wind case through a similarity calculation method, sequencing according to the similarity to obtain a plurality of historical solar wind cases with the highest similarity, and obtaining geomagnetic Kp index forecast corresponding to the current solar wind case according to Kp indexes corresponding to the historical solar wind cases.
2. The geomagnetism Kp index forecasting method based on the similarity algorithm according to claim 1, characterized by further comprising the steps of establishing a historical solar wind case base; the method specifically comprises the following steps:
acquiring historical solar wind parameter data and a geomagnetic Kp index;
carrying out data cleaning and normalization processing on the solar wind parameter data, and carrying out proximity value conversion on the geomagnetic Kp index;
selecting solar wind characteristic parameters by adopting an XGboost algorithm, and screening out two characteristic parameters which have the greatest influence on the geomagnetic Kp index;
calculating the weight of the influence of the two characteristic parameters on the geomagnetic Kp index by adopting a single-layer perceptron;
dividing historical solar wind parameter data into a plurality of solar wind cases by a sliding window with fixed step length, and defining a Kp index immediately after the historical solar wind parameter data as the Kp index corresponding to the solar wind case.
3. The geomagnetic Kp index forecasting method based on the similarity algorithm according to claim 2, wherein the XGBoost algorithm is adopted to select the solar wind characteristic parameters, and two characteristic parameters which have the greatest influence on the geomagnetic Kp index are screened out; the method specifically comprises the following steps:
the method comprises the steps of taking the average value of the sun wind speed V, the sun wind proton density N, the sun wind temperature T, the interplanetary magnetic field intensity B, the interplanetary magnetic field By and the interplanetary magnetic field Bz component in 1 hour as input, taking the corresponding Kp index as a label, iteratively establishing a decision tree through an XGboost model, scoring six input parameters, and taking two characteristic parameters with the highest scores.
4. The method for predicting geomagnetic Kp index based on similarity algorithm according to claim 3, wherein the single-layer perceptron is adopted to calculate the weight of the influence of the two characteristic parameters on the geomagnetic Kp index; the method specifically comprises the following steps:
the two characteristic parameters are used as input layer parameters of the single-layer perceptron, the time resolution is 1 hour, the output layer is Kp indexes corresponding to each hour, the mean square error function is used as a loss function, the Adam algorithm is used as an optimizer, and the weight of the two characteristic parameters on the influence of the geomagnetic Kp indexes is obtained.
5. The geomagnetism Kp index forecasting method based on the similarity algorithm according to claim 3, characterized in that the size of the sliding window is divided by the time resolution of Kp index, and the step size of the sliding window is in the range of 0.5 to 1 hour.
6. The geomagnetism Kp index forecasting method based on the similarity algorithm according to claim 1, characterized in that the similarity between the current solar wind case and the historical solar wind case is calculated by the similarity algorithm; the method specifically comprises the following steps: and calculating the similarity of the current solar wind case and the historical solar wind case by using a dynamic time warping algorithm or a piecewise linear representation algorithm.
7. The geomagnetism Kp index forecasting method based on the similarity algorithm according to claim 6, wherein the calculating the similarity between the current solar wind case and the historical solar wind case by using the dynamic time warping algorithm specifically comprises:
the time sequences of the current solar wind case and each historical solar wind case are subjected to local nonlinear scaling, so that the Euclidean distance accumulated value after the two sequence forms are aligned is the minimum distance of the two time sequences, and the similarity of the current solar wind case and each historical solar wind case is obtained.
8. The geomagnetic Kp index forecasting method based on the similarity algorithm according to claim 6, wherein the calculating the similarity between the current solar wind case and the historical solar wind case by using a piecewise linear representation algorithm specifically comprises:
the method comprises the steps of adopting a piecewise linear expression algorithm for time sequences of a current solar wind case and each historical solar wind case to obtain a plurality of modes respectively, then conducting equal patterning to enable the lengths of the mode sequences of the current solar wind case and the historical solar wind cases to be equal, then obtaining a distance value of each mode by using an absolute value distance or Euclidean distance calculation method, and conducting weighting calculation on the mode distance values to obtain the similarity between the current solar wind case and each historical solar wind case.
9. A system for predicting a geomagnetic Kp index based on a similarity algorithm, the system comprising: the system comprises a historical solar wind case library, a solar wind case receiving module, a similarity calculation module and a geomagnetic Kp index prediction output module; wherein the content of the first and second substances,
the solar wind case receiving module is used for receiving the current solar wind case;
the similarity calculation module is used for reading a pre-established historical solar wind case library and calculating the similarity between the current solar wind case and the historical solar wind case through a similarity calculation method;
the geomagnetic Kp index prediction output module is used for obtaining a plurality of historical solar wind cases with the highest similarity according to the similarity sequence, and obtaining geomagnetic Kp index prediction corresponding to the current solar wind case according to Kp indexes corresponding to the historical solar wind cases.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110521691.XA CN113344032B (en) | 2021-05-13 | 2021-05-13 | Geomagnetic Kp index forecasting method and system based on similarity algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110521691.XA CN113344032B (en) | 2021-05-13 | 2021-05-13 | Geomagnetic Kp index forecasting method and system based on similarity algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113344032A true CN113344032A (en) | 2021-09-03 |
| CN113344032B CN113344032B (en) | 2021-12-28 |
Family
ID=77468436
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110521691.XA Active CN113344032B (en) | 2021-05-13 | 2021-05-13 | Geomagnetic Kp index forecasting method and system based on similarity algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113344032B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060229813A1 (en) * | 2005-03-30 | 2006-10-12 | Tobiska William K | Ionospheric forecast system (IFS) |
| CN104680167A (en) * | 2015-03-09 | 2015-06-03 | 西安电子科技大学 | Aurora oval position determining method based on deep learning |
| CN105118047A (en) * | 2015-07-15 | 2015-12-02 | 陕西师范大学 | Auroral oval boundary position prediction method based on interplanetary and geomagnetic parameters |
| CN107563574A (en) * | 2017-09-29 | 2018-01-09 | 北京理工大学 | A kind of earth's magnetic field Forecasting Methodology based on long memory models Recognition with Recurrent Neural Network in short-term |
| US20180231686A1 (en) * | 2017-02-15 | 2018-08-16 | International Business Machines Corporation | Geomagnetic storm warning |
| CN110046697A (en) * | 2019-04-26 | 2019-07-23 | 北京理工大学 | A kind of Kp index based on K index shows reporting method |
| CN111507538A (en) * | 2020-04-26 | 2020-08-07 | 国网内蒙古东部电力有限公司检修分公司 | Geomagnetic induction current GIC prediction method for power plant station |
-
2021
- 2021-05-13 CN CN202110521691.XA patent/CN113344032B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060229813A1 (en) * | 2005-03-30 | 2006-10-12 | Tobiska William K | Ionospheric forecast system (IFS) |
| CN104680167A (en) * | 2015-03-09 | 2015-06-03 | 西安电子科技大学 | Aurora oval position determining method based on deep learning |
| CN105118047A (en) * | 2015-07-15 | 2015-12-02 | 陕西师范大学 | Auroral oval boundary position prediction method based on interplanetary and geomagnetic parameters |
| US20180231686A1 (en) * | 2017-02-15 | 2018-08-16 | International Business Machines Corporation | Geomagnetic storm warning |
| CN107563574A (en) * | 2017-09-29 | 2018-01-09 | 北京理工大学 | A kind of earth's magnetic field Forecasting Methodology based on long memory models Recognition with Recurrent Neural Network in short-term |
| CN110046697A (en) * | 2019-04-26 | 2019-07-23 | 北京理工大学 | A kind of Kp index based on K index shows reporting method |
| CN111507538A (en) * | 2020-04-26 | 2020-08-07 | 国网内蒙古东部电力有限公司检修分公司 | Geomagnetic induction current GIC prediction method for power plant station |
Non-Patent Citations (4)
| Title |
|---|
| TAN Y: "Geomagnetic index Kp forecasting with LSTM", 《SPACE WEATHER》 * |
| YURONG SHI等: "Predicting the CME arrival time based on the recommendation algorithm", 《RESEARCH IN ASTRONOMY AND ASTROPHYSICS》 * |
| 朱志鹏: "太阳风影响北半球冬季大气环流的特征及其数值模拟研究", 《中国优秀硕士学位论文全文数据库 (基础科学辑)》 * |
| 沈晓飞: "第23太阳活动周期太阳风参数及地磁指数的统计分析", 《地球物理学报》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113344032B (en) | 2021-12-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20220373984A1 (en) | Hybrid photovoltaic power prediction method and system based on multi-source data fusion | |
| US10041475B1 (en) | Reducing curtailment of wind power generation | |
| CN107944622B (en) | Wind power prediction method based on continuous time interval clustering | |
| US20180223812A1 (en) | Reducing curtailment of wind power generation | |
| CN110969290A (en) | Runoff probability prediction method and system based on deep learning | |
| CN116937579B (en) | Wind power interval prediction considering space-time correlation and interpretable method thereof | |
| CN107909212B (en) | Climbing event prediction method based on mesoscale wind power development process serialization | |
| CN115220133B (en) | Rainfall prediction method, device and equipment for multiple meteorological elements and storage medium | |
| Yu et al. | Improved convolutional neural network‐based quantile regression for regional photovoltaic generation probabilistic forecast | |
| CN111159900B (en) | Method and device for predicting wind speed of fan | |
| CN114781723A (en) | Short-term photovoltaic output prediction method based on multi-model fusion | |
| Ho et al. | Amic: An adaptive information theoretic method to identify multi-scale temporal correlations in big time series data | |
| CN113344032B (en) | Geomagnetic Kp index forecasting method and system based on similarity algorithm | |
| CN110826810B (en) | Regional rainfall prediction method combining spatial reasoning and machine learning | |
| CN116029440A (en) | Ultra-short-term power prediction method and device for photovoltaic power station | |
| CN114444763A (en) | Wind power prediction method based on AFSA-GNN | |
| Ma et al. | Anomaly Detection of Mountain Photovoltaic Power Plant Based on Spectral Clustering | |
| CN111008596A (en) | Abnormal video cleaning method based on characteristic expected subgraph correction classification | |
| Zhang et al. | Wind Speed Forecasting of High Plateau Airport Based on Informer | |
| CN117151303B (en) | Ultra-short-term solar irradiance prediction method and system based on hybrid model | |
| Wang | Predict Model on Air Quality and Characteristic of Shenyang | |
| CN117891005A (en) | Ground wire icing prediction method and system based on microclimate factor time sequence | |
| Anand et al. | Huber SVR-Based Hybrid Models for Significant Wave Height Forecasting Using Buoy Sensors | |
| Wan et al. | Modelling the Effects of Wind Characteristics on Wind Gust Statistics in a GEV Distribution | |
| Niu et al. | FsrGAN: A Satellite and Radar-Based Fusion Prediction Network for Precipitation Nowcasting |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |