Natural Data Analysis Method Based on Wavelet Filtering and NARX Neural Networks

Abstract

A method for analyzing natural data and detecting anomalies is proposed. The method is based on combining wavelet filtering operations with the NARX neural network. The analysis of natural data and the detection of anomalies are of particular relevance in the problems of geophysical monitoring. An important requirement of these methods is their adaptability, accuracy and efficiency. Efficiency makes it possible to detect anomalies timely in order to prevent catastrophic natural phenomena. Wavelet filtering operations include the application of a multi-scale analysis construction and threshold functions. The article proposes a wavelet filtering algorithm and a method for estimating thresholds based on a stochastic approach. The operations of the method implementation are described. It is shown that the use of wavelet filtering allows one to suppress noise, simplifies the data structure and, as a result, allows one to obtain a more accurate NARX neural network model. The effectiveness of the method for detecting ionospheric anomalies during periods of magnetic storms is shown using the data of the critical frequency of the ionosphere as an example.

Keywords

• data analysis
• wavelets
• neural networks
• ionosphere