Recently the data-driven learning of dynamic systems has become a promising approach because no physical knowledge is needed. Pure machine learning approaches such as Gaussian process regression (GPR) learns a dynamic model from data, with all physical knowledge about the system discarded. This goes from one extreme, namely methods based on optimizing parametric physical models derived from physical laws, to the other. GPR has high flexibility and is able to model any dynamics as long as they are locally smooth, but can not generalize well to unexplored areas with little or no training data. The analytic physical model derived under assumptions is an abstract approximation of the true system, but has global generalization ability. Hence the optimal learning strategy is to combine GPR with the analytic physical model. This paper proposes a method to learn dynamic systems using GPR with analytic ordinary differential equations (ODEs) as prior information. The one-time-step integration of analytic ODEs is used as the mean function of the Gaussian process prior. The total parameters to be trained include physical parameters of analytic ODEs and parameters of GPR. A novel method is proposed to simultaneously learn all parameters, which is realized by the fully Bayesian GPR and more promising to learn an optimal model. The standard Gaussian process regression, the ODE method and the existing method in the literature are chosen as baselines to verify the benefit of the proposed method. The predictive performance is evaluated by both one-time-step prediction and long-term prediction. By simulation of the cart-pole system, it is demonstrated that the proposed method has better predictive performances.
Usually learning dynamical systems by data-driven methods requires large amount of training data, which may be time consuming and expensive. Active learning, which aims at choosing the most informative samples to make learning more efficient is a promising way to solve this issue. However, actively learning dynamical systems is difficult since it is not possible to arbitrarily sample the state-action space under the constraint of system dynamics. The state-of-the-art methods for actively learning dynamical systems iteratively search for an informative state-action pair by maximizing the differential entropy of the predictive distribution, or iteratively search for a long informative trajectory by maximizing the sum of predictive variances along the trajectory. These methods suffer from low efficiency or high computational complexity and memory demand. To solve these problems, this paper proposes novel and more sampleefficient methods which combine global and local explorations. As the global exploration, the agent searches for a relatively short informative trajectory in the whole state-action space of the dynamical system. Then, as the local exploration, an action sequence is optimized to drive the system's state towards the initial state of the local informative trajectory found by the global exploration and the agent explores this local informative trajectory. Compared to the state-of-the-art methods, the proposed methods are capable of exploring the state-action space more efficiently, and have much lower computational complexity and memory demand. With the state-of-the-art methods as baselines, the advantages of the proposed methods are verified via various numerical examples.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.