Research on sunspot prediction based on CNN-LSTM

Sunspots are closely linked to the solar cycle and the behavior of the solar magnetic field. Therefore, it is significant for predicting solar activity. With the exponential growth in historical data, accurating sunspot prediction has become an urgent requirement. this study proposes a CNN-LSTM prediction model that utilizes the peak number of sunspots as a basis for forecasting the start and duration of the current and next solar cycles. Then, relevant sunspot data is collected and subjected to meticulous data preparation and pre-processing. This uses sample interpolation methods addressing missing values of historical sunspot data, also utilizes performing data exclusion and outlier treatment using the Grubbs’ test ($3 \sigma \sigma$ rule). In study, elucidate the architecture and training process of the CNN-LSTM model and demonstrate its efficacy in predicting the start and duration of the solar cycle. Additionally, the model is employed to forecast the time and intensity of the solar peak period. By quantifying data based on individual solar cycles, the total number and total area of sunspots are determined. Subsequently, establish both CNN-LSTM model and BP neural network model, with the number and area of sunspots as output variables, and the solar cycle duration, solar peak period circumference, and duration as input variables. Through comparative analysis of the two models, this research finds that the CNN-LSTM model outperforms the BP neural network model, and the CNN-LSTM model shows superior performance is demonstrated in predicting the number and area of sunspots.