Time series prediction for evolutions of complex systems: A deep learning approach

Evolutions of complex systems are expressions of the dynamic and uncertain behaviors and ever-changing causalities of the potential multivariate processes in a system, which exhibit the characteristics of uncertainties, nonlinear dynamics, and aperiodic fluctuation. These characteristics pose a major challenge for predicting the states of evolutions of complex systems accurately. Supervised learning methods in artificial intelligence, such as the artificial neural networks (ANNs) and the support vector machine (SVM), are suitable for nonlinear time series prediction. However, due to suffering from the issue of feature extraction relying on prior knowledge, and the weakness of shallow learning, traditional ANNs and SVM are hard to guarantee the prediction accuracy and stability. In this study, we propose a deep learning approach, which hybridizes a deep belief networks (DBNs) and a nonlinear kernel-based parallel evolutionary SVM (ESVM), to predict evolution states of complex systems in a classification manner. The proposed approach is tested by two case studies regarding the mobile robot navigation and the early-warning-of-risk of harmful algal blooms. Computational results are compared with those of SVM and DBNs approaches, which demonstrate that the proposed approach outperforms them, and can offer an intelligent decision support tool for predicting evolutions of complex systems.