Deep Learning in Multi-step Prediction of Chaotic Dynamics: From Deterministic Models to Real-World Systems

The book is the first to systematically address the use of deep neural networks for forecasting chaotic time series, implementing a multi-step approach and comparing various neural network architectures. It also introduces an innovative training approach for recurrent structures and applies transfer-learning techniques for environmental time series forecasting.

Format: Paperback / softback
Length: 104 pages
Publication date: 15 February 2022
Publisher: Springer Nature Switzerland AG

The book represents a groundbreaking endeavor in the realm of forecasting chaotic time series using deep neural networks. Unlike most existing literature, it adopts a multi-step approach, enabling the prediction of an entire interval of future values. This is of immense importance in various applications, such as model predictive control, where the prediction of values across the entire receding horizon is crucial. The book progresses systematically, exploring deterministic models with varying degrees of complexity and chaoticity, moving towards noisy systems, and ultimately delving into real-world scenarios. It conducts a comprehensive comparison of various neural network architectures, including feed-forward and recurrent networks. Furthermore, it introduces an innovative and powerful approach for training recurrent structures tailored specifically for sequence-to-sequence tasks.

In addition to its technical contributions, the book also makes pioneering efforts in the field of environmental time series forecasting by applying transfer-learning techniques such as domain adaptation. By leveraging knowledge acquired from other domains, the book enhances the forecasting accuracy of chaotic time series, making it a valuable resource for researchers and practitioners in the field.

Overall, the book represents a significant milestone in the development of deep neural networks for forecasting chaotic time series. Its multi-step approach, comprehensive comparison of architectures, and innovative training techniques make it a valuable resource for anyone interested in this field.

Weight: 191g
Dimension: 235 x 155 (mm)
ISBN-13: 9783030944810
Edition number: 1st ed. 2021