The 10.7‐cm radio flux multistep forecasting based on empirical mode decomposition and back propagation neural network

Abstract

The 10.7‐cm solar radio flux (F10.7) is among the most widely used indices of solar activity, whose prediction plays a vital role in the field of meteorology and aerospace. In this paper, a novel approach for a specific amalgamation of back propagation neural network (BPNN) and empirical mode decomposition (EMD) is proposed, which is called EMD‐BP, aimed at forecasting the daily F10.7 values 1–27 days ahead. The daily F10.7 values, which are highly nonlinear and unstable time series, are transformed into a series of subsignals with different frequency components by EMD and trained by the BPNN. For estimating the performance of the description EMD‐BP, the support vector regression (SVR), Long Short‐Term Memory neural network and BPNN as the reference models are analyzed in different conditions. The results indicate that the proposed model improves the F10.7 forecasting within 27 days, compared with the common machine learning algorithms and transport model. Furthermore, a tremendous decline of forecast error is determined in high‐level solar activity, in contrast with SVR and BPNN. © 2020 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

Forecasting the 10.7-cm Solar Radio Flux Using Deep CNN-LSTM Neural Networks

“…In the experiment, multiple evaluation indexes are used to evaluate the prediction effect of comparative algorithms, as shown in Formulas( 12)- (15). Mean Absolute Percentage Error (MAPE) is calculated by dividing the difference between the actual value and the estimated value by the actual value.…”

“…Figure 3 showed the annual average MAPE results of comparative algorithms from 2003-2012 on 3-day forecasts. The comparison models include the BP model [14], a three-layer feed-forward network, the EMD-BP model [15] and a stacked model of EMD and BP. In addition, there is the traditional three-layer LSTM model.The MAPE is positive correlation with the level of annual average F10.7 for all algorithms.…”

“…In order to further improve the prediction effect, some scholars tried to extract the signal characteristics of the original data by the signal processing method and then predict them by machine learning method. Common ones include empirical mode decomposition (EMD) [15,16], Fourier transform [17] and wavelet analysis [18,19]. Relevant literature above have exemplified the effectiveness of machine learning methods, signal processing technique and a combination of the two, but still faces the following limitations: (1) Traditional machine learning methods are usually shallow neural networks.…”

Deep Temporal Convolutional Networks for F10.7 Radiation Flux Short-Term Forecasting

“…Xiao et al (2017) used back propagation neural network(BP) to predict the solar activity daily mean index F10.7 for short-term forecasting.The results showed that using BP neural networks to predict the solar activity daily index F10.7 was superior to the results of Wang et al(2009). Luo et al (2020) proposed a multi-step prediction method for the 10.7 cm radio flux. The method is a combination of the Empirical Mode Decomposition (EMD) and Back-Propagation (BP) network to construct an EMD-BP model for predicting F10.7 values.…”