A homotopy gated recurrent unit for predicting high dimensional hyperchaos

Due to recurrent neural network's (RNN) powerful ability to process sequential data, RNN attracts much attention in various fields. However, the long-term dependency has always been a major challenge for RNN. In this paper, a homotopy gated recurrent unit (H-GRU) model is proposed for predicting hyperchaos, which tries to improve the longterm dependence of RNN. We develop the model in three tasks: Mackey-Glass (tau = 30), hyperchaotic Rossler, and 4D hyperchaos of Chen et al. (2018). Compared with the four pre-existing prediction models, the prediction accuracy of H-GRU is the highest among baseline models, which demonstrates the merit of the proposed model. Furthermore, the proposed model has significant improvement in replicating hyperchaotic attractors. Especially in 4D hyperchaos tasks, not only is the prediction accuracy of H-GRU three orders of magnitude higher than the baseline model, but also H-GRU can replicate hyperchaotic attractors more accurately. (c) 2022 Elsevier B.V. All rights reserved.

Automated summary

This paper introduces a homotopy gated recurrent unit (H-GRU) model to improve the long-term dependency of recurrent neural networks (RNN) and validates it in hyperchaos prediction tasks. The results show that the proposed model outperforms baseline models in prediction accuracy and replicating hyperchaotic attractors.