Optimizing the process of construction of NARX neural network model for time series of complicated structure based on threshold wavelet filtering

A method for optimization of the process of construction of a NARX model based on threshold wavelet filtering was proposed. NARX realizes the paradigm of the autoregressive moving average (ARIMA) model class. Some difficulties and restrictions are known when modeling noisy and nonstationary time series with seasonal patterns of long time delay by NARX. On the example of time series of ionospheric parameters, the paper shows that the developed procedure of wavelet filtering allows us to obtain an adequate NARX model. An algorithm for wavelet filtering was constructed and a method for obtaining stochastic thresholds was suggested. The efficiency of the method was shown for the problems of detection of ionospheric inhomogeneities based on the data of ionospheric layer F2 critical frequency. The work was carried out as part of the implementation of the state task AAAA-A21-121011290003-0. The work was carried out by the means of the Common Use Center "North-Eastern Heliogeophysical Center" CKP_558279, USU 351757.