Comparison of the Single-Step and Multistep Approaches to Forecast of the Solar Activity Index

Abstract

By the example of a problem of forecasting daily values of solar activity index, three different approaches to forecasting a time series of the solar activity index were compared: iterations of a single-step model, an independent single-step forecast for each subsequent month, and a single multistep forecast for the entire period. As a model, each approach uses a machine learning model based on a neural network, as well as an auxiliary theoretical series of solutions, which is obtained from a physical model of solar dynamo.