EVALUATION OF THE EFFECTIVENESS OF USING GRU AND LSTM MODELS FOR PREDICTING THE BEHAVIOR OF THE LORENTZ SYSTEM WHEN GENERATING SEQUENCES OF NUMERICAL VALUES

Abstract

This article presents a comparative analysis of the applicability of neural networks, namely recurrent models Gated Recurrent Units (GRU) and Long Short-Term Memory (LSTM), in terms of their ability to predict the behavior of chaotic systems when generating numerical values. The main goal is to use the pattern recognition capabilities inherent in neural networks to make accurate predictions of the behavior of the chaotic Lorenz system when building telecommunications systems based on it. In the field of telecommunications systems, predicting the output values of a chaotic system helps to optimize the network, prevent failures, increase bandwidth, and build communication systems with extended spectrum. GRU and LSTM are two popular architectures of recurrent neural networks for sequential data processing. GRU has a simpler structure and learns much faster, but LSTM has a more complex structure, which gives it the ability to remember information for a longer period. In this work, the GRU and LSTM models were implemented using the Python programming language and trained to predict the possible output values of the multidimensional chaotic Lorenz system and to evaluate the possibility of using neural networks to improve the recovery of information flows on the receiving side of the communication system with spectrum expansion. These models were evaluated using the mean square error (MSE), mean absolute error (MAE), and coefficient of determination metrics, and the test results are presented in the corresponding graphs. The results of this study demonstrate the high efficiency of the models in predicting the behavior of a chaotic system.