Prof.DENG Linhua from Yunnan Observatories of Chinese Academy of Sciences with collaborators (Prof.ZHENG Sheng, and LIU Xiaohuan from China Three Gorges University) used the long short-term memory (LSTM) deep learning technique to predict the amplitude and peak time of solar cycles 25 and 26. The monthly database of the relative sunspot number taken from the National Astronomical Observatory of Japan is used. This work was published in Publications of the Astronomical Society of Japan.

For solar cycle prediction, there are three general techniques: the precursory method, the dynamo model, and the extrapolation method. The main idea of the extrapolation method is to analyze and predict future data based on time series data of the manifestations of solar magnetic activity such as sunspot numbers or areas.

Deep learning aims at establishing and simulating the neural network of the human brain, and imitating its mechanism for interpreting. As a new field of machine learning, deep learning has been applied in many aspects, such as speech recognition, biology, foreign exchange markets, etc. The LSTM neural network is an artificial network used for time series data analysis. LSTM neural networks are being applied to astronomy by more and more scholars because of their excellent learning ability.

The researchers used an LSTM network to analyze the time series observations of sunspots from 1929 January to 2022 April, then predict the maximum value and its occurrence time in the coming solar cycle 25 and 26. The dataset is divided into eight schemes of two to nine slices for training, showing that the five-slice LSTM model with root mean square error of 11.38 is the optimal model.

According to the prediction, solar cycle 25 will be about 21% stronger than solar cycle 24, with a peak of 135.2 occurring in 2024 April. Solar cycle 26 will be similar to solar cycle 25 and reach its peak of 135.0 in 2035 January. Therefore, the LSTM neural network shows good performance in studies of long-term time series observations, which could be useful in solar activity and space weather prediction.

Contact:
DENG Linhua
Yunnan Observatories, CAS
Email: lhdeng@ynao.ac.cn