A method of predicting long-term solar activity levels

Abstract

The application discloses a method for predicting long-term solar activity level, comprising the following steps: S1, establishing a solar activity level prediction model; S2, establishing a relative sunspot number monthly average database; S3, obtaining a target prediction month; S4, using an RBF neural network to predict the relative sunspot number yearly moving monthly average of the target prediction month and the following n months, and filling in the relative sunspot number monthly average database; S5, traversing the relative sunspot number monthly average database to obtain the relative sunspot number monthly average data of the target prediction month and the previous several months; S6, substituting the relative sunspot number monthly average data of the target prediction month and the previous several months into the solar activity level prediction model; and S7, outputting the relative sunspot number yearly moving monthly average and the prediction result thereof. The application can accurately predict the relative sunspot number yearly moving monthly average of the current month and the following months, and can be used as a reference basis for the solar activity level and an input parameter for the prediction of other related parameters.

Description

Technical Field

[0001] This invention relates to the field of solar activity levels in astrophysics, and in particular to a method for predicting long-term solar activity levels. Background Technology

[0002] Currently, the main parameter used internationally to describe long-term solar activity levels is the smoothed monthly average of the relative sunspot number, which is often used by those skilled in the art as input for forecasting models related to long-term solar activity levels.

[0003] The internationally accepted method for calculating the smoothed monthly average of the relative sunspot number is to use the relative sunspot number over a continuous 13 months. The smoothed monthly average of the relative sunspot number (hereinafter referred to as SSN) is calculated using the following formula: .

[0004] The index actually calculates the value of the median month out of every 13 months. For example, if you substitute data from a certain 13-month period, the SSN calculation result will be the value of the 7th month. The first drawback of this calculation method is that there is a time lag of several months in obtaining the latest value. Regardless of how many consecutive months of data are used, this calculation mode will objectively have some degree of delay. The second drawback of SSN calculation is that if you need to obtain the SSN of the current month in real time for subsequent research, the value of that month actually uses the measured values ​​of the next 6 months. If it is used as an input factor in the forecast, it is equivalent to incorporating future measured information. In other words, the forecast for the next 6 months made using this value includes information from the forecast month itself and even some information after the forecast month. It contains the factor of predicting itself, and it can even be said that the future is predicting the past, which is logically unreasonable. It cannot reflect real-time information, and therefore, it is impossible to use SSN for further research in academia.

[0005] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method for predicting long-term solar activity levels, which can solve the problems existing in the background technology.

[0007] To achieve the above objectives, the present invention provides a method for predicting long-term solar activity levels, characterized by comprising: S1: establishing a solar activity level prediction model; S2: establishing a database of monthly average relative sunspot numbers; S3: obtaining the target prediction month; S4: using an RBF neural network to predict the annual sliding monthly average of the relative sunspot numbers for the n months following the target prediction month, and filling it into the database of monthly average relative sunspot numbers; S5: traversing the database of monthly average relative sunspot numbers to obtain the monthly average data of the relative sunspot numbers for the target prediction month and the preceding several months; S6: substituting the monthly average data of the relative sunspot numbers for the target prediction month and the preceding several months into the solar activity level prediction model; S7: outputting the annual sliding monthly average of the relative sunspot numbers and its prediction result.

[0008] In one or more embodiments, step S1 includes: S11: inputting the calculation formula for the annual average of the relative number of sunspots; S12: determining the input and output quantities; S13: generating input ports and output ports based on the quantity of the input and output quantities.

[0009] In one or more embodiments, the formula for calculating the annual monthly average of the relative number of sunspots in step S11 is as follows: Where i month refers to the target predicted month, SSNyi refers to the annual monthly average of the relative number of sunspots, SNi is the monthly average of the relative number of sunspots in the i-th month, and SNi-n refers to the monthly average of the relative number of sunspots in the n months preceding the i-th month.

[0010] In one or more embodiments, step S2 includes: S21: obtaining monthly average data of the relative number of sunspots; S22: storing the monthly average data of the relative number of sunspots into the corresponding storage space in month order.

[0011] In one or more embodiments, step S4 includes: S41: setting the prediction algorithm of the RBF network and formulating a loss optimization strategy; S42: predicting the annual slip-month average of sunspot relative numbers and filling the predicted value into the corresponding month of the annual slip-month average of sunspot relative numbers database; S43: after the prediction process, adopting a loss optimization strategy to reduce the error of the predicted value and adjusting the prediction algorithm.

[0012] In one or more embodiments, step S42 includes: S421: loading the annual sliding monthly average of the relative number of sunspots for the required month, the format of which includes timestamps and numerical fields; S422: setting the lag step size to construct a supervised learning format; S423: dividing the training set / test set; S424: initializing and setting the network parameters of the input layer, hidden layer, and output layer; S425: executing the core code of the prediction algorithm; S426: obtaining the predicted value and filling it into the annual sliding monthly average database of the relative number of sunspots.

[0013] In one or more embodiments, step S43 includes: S431: resetting the target prediction month, wherein the reset target prediction month is a number of months preceding the current month and months earlier; S432: traversing the monthly mean of the relative sunspot number database, reading all monthly mean of the relative sunspot number and corresponding months used to calculate the annual slip-month mean of the relative sunspot number; S433: hiding the annual slip-month mean of the relative sunspot number for the second half of the months, simulating and predicting the annual slip-month mean of the relative sunspot number for the second half of the months, and obtaining a predicted value; S434: comparing the predicted value with the annual slip-month mean of the relative sunspot number for the second half of the months, obtaining the loss gap, and optimizing the prediction algorithm through backpropagation.

[0014] In a second aspect, the present invention provides a system for predicting long-term solar activity levels, based on the same concept as a method for predicting long-term solar activity levels, comprising: a first establishment module for establishing a solar activity level prediction model; a second establishment module for establishing a database of monthly average relative sunspot numbers; a first acquisition module for acquiring a target prediction month; a first prediction module, using an RBF neural network, for predicting the annual slip-monthly average relative sunspot number for the n months following the target prediction month, and filling it into the database of monthly average relative sunspot numbers; the first acquisition module for acquiring the monthly average relative sunspot number data for the target prediction month and the preceding several months; a first substitution module for substituting the monthly average relative sunspot number data for the target prediction month and the preceding several months into the solar activity level prediction model; and a first output module for outputting the annual slip-monthly average relative sunspot number and its prediction result.

[0015] In a third aspect, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements a method for predicting long-term solar activity levels.

[0016] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements a method for predicting long-term solar activity levels.

[0017] Compared with the prior art, the various technical solutions and embodiments provided by the present invention include at least the following technical effects or advantages:

[0018] A prediction algorithm based on the RBF network is established to accurately predict the monthly relative slip mean of sunspots for the current month and the future. A loss optimization strategy is formulated to train the prediction algorithm so that its error accuracy can be comparable to the acceptable error range of the monthly relative slip mean of sunspots observed in actual meteorological observations. Therefore, the actual predicted monthly relative slip mean of sunspots for the current month can achieve the ultra-high accuracy recognized by the industry and can be used as a reference for the level of solar activity. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and are intended to explain the invention. They are intended to be seen as illustrative of various combinations of preferred embodiments and do not constitute an undue limitation of the invention. In the drawings:

[0020] Figure 1 A schematic diagram of the overall process for a method to predict long-term solar activity levels provided by the present invention;

[0021] Figure 2 The present invention provides a method for predicting long-term solar activity levels, including predictions, measurements, and errors of the SSNy index for the first month of the future from January 1950 to February 2025;

[0022] Figure 3 The present invention provides a method for predicting long-term solar activity levels, including predictions, measurements, and errors of the SSNy index for the second month of the future from January 1950 to February 2025;

[0023] For necessary explanation and distinction, the purple curve represents the measured value, the red curve represents the predicted value, the dashed line represents the range of measured value fluctuation within 10%, and the red curve being obscured by the purple curve indicates a high degree of agreement between the predicted and actual values; the blue curves, from top to bottom, represent absolute error and relative error respectively. Detailed Implementation

[0024] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises", "made for", etc., shall be understood to include the stated elements or components, without excluding other elements or other components.

[0025] The purpose of this invention is to provide a method for predicting long-term solar activity levels. Existing methods for calculating the smoothed monthly average relative sunspot number (SSN) preferably use the monthly average of the relative sunspot number over 13 months as a reference value for the median month's solar activity level. For example, the monthly average of the relative sunspot number from January 2024 to January 2025 (13 months) is used to calculate the smoothed monthly average SSN for July 2024 (the median month). The use of 13 months of data is based on proven and generally accepted data that satisfies the optimal number of months for the required sample size. Since the future monthly average of the relative sunspot number is unpredictable, existing methods for calculating the smoothed monthly average SSN have a 6-month delay. This invention, by calculating the annual smoothed monthly average of the relative sunspot number, eliminates this forecast delay and provides a logically rigorous method for accurately predicting solar activity levels.

[0026] Example 1:

[0027] This embodiment provides a method for predicting long-term solar activity levels, including:

[0028] S1: Establish a solar activity level prediction model; specifically, the solar activity level prediction model includes prediction formulas and prediction steps. After inputting the original monthly average of the relative number of sunspots into the solar activity level prediction model, accurate solar activity level prediction data can be obtained.

[0029] In a preferred embodiment of this example, step S1 includes:

[0030] S11: Input the calculation formula for the annual monthly average of the relative number of sunspots;

[0031] S12: Determine the input and output quantities;

[0032] S13: Generate input ports and output ports based on the number of input and output quantities.

[0033] Specifically, in order to provide a solar activity level prediction model, prediction formulas and steps are set. The calculation formula for the annual slip-monthly average of the relative sunspot number is obtained, and input and output interfaces are defined. Preferably, the monthly average of the 12-month relative sunspot number is used as the input, and the annual slip-monthly average of the relative sunspot number is used as the output.

[0034] In a preferred embodiment of this invention, the formula for calculating the annual monthly average of the relative number of sunspots in step S11 is as follows:

[0035] Where i month refers to the target predicted month, SSNyi refers to the annual monthly average of the relative number of sunspots, SNi is the monthly average of the relative number of sunspots in the i-th month, and SNi-n refers to the monthly average of the relative number of sunspots in the n months preceding the i-th month.

[0036] In a preferred embodiment, the formula for calculating the annual average of the relative sunspot number is to take the weighted average of the monthly averages of the relative sunspot number for different months and find the correlation with the parameters predicted by the model, thereby improving the prediction accuracy of the model.

[0037] S2: Establish a monthly average database of relative sunspot numbers;

[0038] In a preferred embodiment of this example, step S2 includes:

[0039] S21: Obtain monthly average data on the relative number of sunspots;

[0040] S22: Store the monthly average data of the relative number of sunspots in the corresponding storage space in order of month.

[0041] Specifically, in order to store and retrieve solar activity level data at any time, this implementation establishes a monthly average database of sunspot relative numbers, allocates storage space according to the order of months, and can retrieve the corresponding information by month when data needs to be retrieved.

[0042] S3: Obtain the target forecast month;

[0043] S4: Use the RBF neural network to predict the annual slip-monthly average of the relative number of sunspots for n months after the target prediction month, and fill it into the annual slip-monthly average of the relative number of sunspots database;

[0044] Specifically, existing solar activity level indicators cannot be obtained in real time, which fails to meet the timeliness requirements for operational forecasting of parameters related to solar activity levels, leading to delays in the progress of operational forecasting of these parameters. However, this implementation method can calculate the solar activity level using a formula after the end of a calendar month, thereby improving the efficiency and progress of operational forecasting of these parameters.

[0045] In a preferred embodiment of this example, step S4 includes:

[0046] S41: Set the prediction algorithm for the RBF network and formulate loss optimization strategies, such as adjusting parameters like training step size;

[0047] S42: Predict the annual sliding monthly average of the relative number of sunspots and fill the predicted value into the corresponding month in the database of annual sliding monthly average of the relative number of sunspots;

[0048] S43: After the prediction process, a loss optimization strategy is adopted to reduce the error of the predicted value and adjust the prediction algorithm.

[0049] Specifically, in order to reflect the level of solar activity in real time, this embodiment found that the biggest technical problem in obtaining the level of solar activity is the lack of future monthly average data on the relative number of sunspots. This embodiment proposes a method to reflect the level of solar activity by predicting the annual slip-monthly average of the relative number of sunspots. The prediction method includes a prediction algorithm and a loss optimization strategy. The prediction algorithm is used to obtain the known monthly average of the relative number of sunspots from the monthly average database of the relative number of sunspots, calculate the annual slip-monthly average of the relative number of sunspots, and predict the future annual slip-monthly average of the relative number of sunspots on a monthly basis. The prediction algorithm has a drawback, namely, the accuracy of the prediction is uncertain. Therefore, the loss optimization strategy is set to train the prediction algorithm and make its prediction results more accurate.

[0050] In a preferred embodiment of this example, step S42 includes:

[0051] S421: Loads the annual slip month average of the relative number of sunspots for the required month, in a format that includes timestamps and numerical fields;

[0052] S422: Set the lag size to construct the supervised learning format;

[0053] S423: Divide the dataset into training and test sets;

[0054] S424: Initialize and set the network parameters for the input layer, hidden layer, and output layer;

[0055] S425: The core code that executes the prediction algorithm;

[0056] S426: Obtain the predicted value and fill it into the annual monthly average value database of the relative number of sunspots.

[0057] Specifically, the Radial Basis Function (RBF) network is a type of feedforward network constructed based on function approximation theory. Learning this type of network is equivalent to finding the best-fitting plane for the training data in a multidimensional space. This network avoids the tedious and lengthy computations of backpropagation (BP) networks, possesses high computational speed and extrapolation capabilities, and also exhibits strong nonlinear mapping capabilities.

[0058] Radial basis function (RBF) networks achieve a nonlinear transformation from the input space RN to the output space RM through a linear combination of nonlinear basis functions. Sunspot numbers are a highly nonlinear time series. Predicting them, i.e., predicting the future M data from the previous N data, essentially involves finding the nonlinear mapping relationship from RN to RM. Therefore, it can be said that radial basis function networks are particularly suitable for predicting nonlinear time series such as sunspot numbers.

[0059] In a preferred embodiment of this example, step S43 includes:

[0060] S431: Reset the target forecast month, wherein the reset target forecast month is a number of months prior to the current month and earlier;

[0061] S432: Traverse the database of monthly average relative sunspot numbers and read all monthly average relative sunspot numbers and corresponding months used to calculate the annual average relative sunspot number;

[0062] S433: Hide the annual average relative number of sunspots for the second half of the month, simulate and predict the annual average relative number of sunspots for the second half of the month, and obtain the predicted value;

[0063] S434: Compare the predicted value with the annual average of sunspots in the latter half of the months to obtain the loss gap, and optimize the prediction algorithm through backpropagation.

[0064] Specifically, the radial basis function (RBF) network method is used to predict SSNy. A loss optimization strategy is adopted to modify the prediction algorithm. The problem encountered by the loss optimization strategy is how to find loss optimization data. Since the predicted value still needs to wait for n months to obtain the actual data for comparison, its optimization efficiency is not good. Therefore, to address this problem, this implementation uses historical monthly data for training. First, the target prediction month is reset, and the target is set from this month to the month when the monthly average data of the relative number of sunspots over several months before and after can be determined.

[0065] To avoid using data from a single month to predict its own performance, the system hides data from the following n months. The algorithm then uses data from the preceding n months to predict the data for the following n months, assuming that data for the following n months is unavailable. The predicted value is compared with the actual value, and the algorithm is optimized to make the predicted value closer to the actual value. Since the records of relative sunspot numbers have a long history, the monthly average database of relative sunspot numbers contains rich training data. Through extensive computation, the prediction algorithm can accurately predict the annual slip-monthly average of relative sunspot numbers. The loss optimization strategy in this implementation aims to reduce the difference between the predicted and actual values ​​to within the allowable error range of the internationally measured annual slip-monthly average of relative sunspot numbers. In other words, even the measured annual slip-monthly average of relative sunspot numbers has errors, and the predicted value in this implementation can be optimized to within this allowable error range. Therefore, the predicted value is generally considered by the industry to be equivalent to the measured value, and the annual slip-monthly average of relative sunspot numbers calculated based on this is also recognized by the industry as an accurate value.

[0066] For example, data from January 1750 to December 1949 is used to train a Radial Basis Function (RBF) network. The observations of the first three months, D(n-2), D(n-1), and D(n), are used as the network's input vectors, and the observations of the first and second months following, D(n+1) and D(n+2), are used as the output vectors. Then, data from January 1950 to February 2025 is used for a forecast experiment. The observations of the first three months, D(n-2), D(n-1), and D(n), are used as the network's input vectors. The output vectors are then the predicted values ​​D(n+1) and D(n+2) for the first and second months of the future. These predicted values ​​are compared with the actual values, and the prediction algorithm is optimized to ensure that the error between the predicted and actual values ​​is within the allowable error range of the annual sliding monthly average of the relative number of sunspots.

[0067] S5: Traverse the annual monthly average of the relative sunspot number database to obtain the monthly average of the relative sunspot number for the target prediction month and the previous few months;

[0068] S6: Substitute the annual average monthly sunspot data of the target prediction month and the previous few months into the solar activity level prediction model;

[0069] S7: Outputs the annual monthly average of the relative number of sunspots and its prediction results.

[0070] Example 2:

[0071] This embodiment provides a system for predicting long-term solar activity levels, based on the same concept as the method for predicting long-term solar activity levels described above, including:

[0072] The first module is used to establish a predictive model for solar activity levels.

[0073] The second module is used to establish a monthly average database of relative sunspot numbers.

[0074] The first acquisition module is used to acquire the target predicted month;

[0075] The first prediction module uses an RBF neural network to predict the annual monthly average of the relative number of sunspots for n months after the target prediction month, and fills it into the monthly average database of the relative number of sunspots.

[0076] The first acquisition module is used to acquire the monthly average data of the relative number of sunspots in the target prediction month and the previous several months;

[0077] The first input module is used to input the monthly average data of the relative number of sunspots in the target prediction month and the previous several months into the solar activity level prediction model.

[0078] The first output module is used to output the annual sliding monthly average of the relative number of sunspots and its prediction results.

[0079] Example 3:

[0080] This embodiment provides an electronic device. Based on the same concept, this embodiment is used to implement a method for predicting long-term solar activity levels as described in Embodiment 1. The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method for predicting long-term solar activity levels.

[0081] Example 4:

[0082] This embodiment provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method for predicting long-term solar activity levels.

[0083] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A method for predicting long-term solar activity levels, characterized in that, include: S1: Establish a model for predicting solar activity levels; S2: Establish a monthly average database of relative sunspot numbers; S3: Obtain the target forecast month; S4: Use the RBF neural network to predict the annual monthly average of the relative number of sunspots for n months after the target prediction month, and fill it into the monthly average of the relative number of sunspots database; S5: Traverse the monthly average of the relative sunspot number database to obtain the monthly average of the relative sunspot number for the target prediction month and the previous few months; S6: Substitute the target prediction month and the monthly average of the relative number of sunspots in the previous few months into the solar activity level prediction model; S7: Outputs the annual monthly average of the relative number of sunspots and its prediction results; Step S1 includes: S11: Input the calculation formula for the annual monthly average of the relative number of sunspots; S12: Determine the input and output quantities; S13: Generate input ports and output ports based on the number of input and output quantities; In step S11, the formula for calculating the annual monthly average of the relative number of sunspots is as follows: Where i month refers to the target predicted month, SSNyi refers to the annual monthly average of the relative number of sunspots, SNi is the monthly average of the relative number of sunspots in the i-th month, and SNi-n refers to the monthly average of the relative number of sunspots in the n months preceding the i-th month. Step S4 includes: S41: Define the prediction algorithm for the RBF network and formulate a loss optimization strategy; S42: Predict the annual sliding monthly average of the relative number of sunspots and fill the predicted value into the corresponding month in the database of annual sliding monthly average of the relative number of sunspots; S43: After the prediction process, a loss optimization strategy is adopted to reduce the error of the predicted value and adjust the prediction algorithm; Step S42 includes: S421: Loads the annual slip month average of the relative number of sunspots for the required month, in a format that includes timestamps and numerical fields; S422: Set the lag size to construct the supervised learning format; S423: Divide the dataset into training and test sets; S424: Initialize and set the network parameters for the input layer, hidden layer, and output layer; S425: The core code that executes the prediction algorithm; S426: Obtain the predicted value and fill it into the annual monthly average value database of the relative number of sunspots; Step S43 includes: S431: Reset the target forecast month, wherein the reset target forecast month is a number of months prior to the current month and earlier; S432: Traverse the database of monthly average relative sunspot numbers and read all monthly average relative sunspot numbers and corresponding months used to calculate the annual average relative sunspot number; S433: Hide the annual average relative number of sunspots for the second half of the month, simulate and predict the annual average relative number of sunspots for the second half of the month, and obtain the predicted value; S434: Compare the predicted value with the annual average of sunspots in the latter half of the months to obtain the loss gap, and optimize the prediction algorithm through backpropagation.

2. The method for predicting long-term solar activity levels as described in claim 1, characterized in that, Step S2 includes: S21: Obtain monthly average data on the relative number of sunspots; S22: Store the monthly average data of the relative number of sunspots in the corresponding storage space in order of month.

3. A system for predicting long-term solar activity levels, based on the same concept as the method for predicting long-term solar activity levels described in any one of claims 1-2, characterized in that, include: The first module is used to establish a predictive model for solar activity levels. The second module is used to establish a monthly average database of relative sunspot numbers. The first acquisition module is used to acquire the target predicted month; The first prediction module uses an RBF neural network to predict the annual monthly average of the relative number of sunspots for n months after the target prediction month, and fills it into the monthly average database of the relative number of sunspots. The first acquisition module is used to acquire the monthly average data of the relative number of sunspots in the target prediction month and the previous several months; The first input module is used to input the monthly average data of the relative number of sunspots in the target prediction month and the previous several months into the solar activity level prediction model. The first output module is used to output the annual sliding monthly average of the relative number of sunspots and its prediction results.

4. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements a method for predicting long-term solar activity levels as described in any one of claims 1 to 2.

5. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements a method for predicting long-term solar activity levels as described in any one of claims 1 to 2.

Images