TimesNet-based real-time forecasting system for F10.7 index using DRAO and Chinese Langfang datasets

This repository contains the datasets and experimental code used in our research, as well as documentation for these files, named "TimesNet_F10.7." In this paper, we constructed the DRAO univariate dataset (http://www.celestrak.com/SpaceData/SW-All.txt), DRAO multivariate dataset, Langfang daily-averaged dataset, and Langfang hourly-resolution dataset. The prediction datasets for BGS, CLS, and SWPC are sourced from (https://swe.ssa.esa.int/en/forind-federated).

Folder "1" corresponds to Innovation Point 1, where we developed five deep learning models (TimesNet, iTransformer, PatchTST, N-beats, BiGRU) and a benchmark model ANN for F10.7 forecasting. On the DRAO univariate dataset, we conducted a comparative study of the forecasting performance of these six models.

Folder "2" corresponds to Innovation Point 2, where we studied the impact of different combinations of multivariate features on the performance of the recommended model, TimesNet, using the DRAO multivariate dataset.

Folder "3" corresponds to Innovation Point 3, where, based on univariate feature model (TimesNet-F) and multivariate feature model (TimesNet-FIAC), we developed a real-time forecasting system for F10.7.

Folder "4" corresponds to Innovation Point 4, where we compared the forecasting performance of our models with that of four foreign research institutions (BGS, SWPC, CLS, DRAO) during the same period.

Folder "5" corresponds to Innovation Point 5, where we applied TimesNet-F model to conduct daily-averaged and hourly-resolution forecasting studies for F10.7 index from the L&S telescope in Langfang, China.