Selection and Prediction of the Trend of a Time Series Using a Recurrent Neural Network

For preprocessed time realizations, envelopes are constructed using interpolation methods and the signal trend is highlighted. The resulting trend is divided into two samples, according to which the recurrent neural network is trained. Various variants of the neural network architecture, including those using the LSTM module, are considered. The Adam optimization algorithm is used to train the model. The implementation is made in Python using the PyTorch framework. The dependence of the forecast quality on the size of the input data sequence is considered.