The continuously increasing amount of noisy data demands the development of accurate and efficient models for analysis, modeling, and control. In this article, we propose a novel data-driven moment matching method whi...
详细信息
ISBN:
(纸本)9798331540920;9783907144107
The continuously increasing amount of noisy data demands the development of accurate and efficient models for analysis, modeling, and control. In this article, we propose a novel data-driven moment matching method which employs Tikhonov regularization in the Reproducing kernel Hilbert Spaces (RKHSs). Specifically, considering a realistic scenario in which the system's plant is unknown and only noisy measured data are available, we provide an estimation of the moment of the unknown plant by solving a regularized optimization problem on RKHS. For, we first demonstrate that the estimation of the moment can be improved via tuning the regularization term, and further, we show under which condition the effect of the transient improves the performance of the estimation. Then, we construct a parameterized model characterized by a kernel-based output mapping. Finally, the proposed data-driven approach is validated and discussed by means of a DC-to-DC C ' uk converter driven by a Van der Pol oscillator.
Recommender systems are information filtering tools that seek to match customers with products or services of interest. Most of the prevalent collaborative filtering recommender systems, such as matrix factorization a...
详细信息
Recommender systems are information filtering tools that seek to match customers with products or services of interest. Most of the prevalent collaborative filtering recommender systems, such as matrix factorization and AutoRec, suffer from the "cold-start" problem, where they fail to provide meaningful recommendations for new users or new items due to informative-missing from the training data. To address this problem, we propose a weighted AutoEncoding model to leverage information from other users or items that share similar characteristics. The proposed method provides an effective strategy for borrowing strength from user or item-specific clustering structure as well as pairwise similarity in the training data, while achieving high computational efficiency and dimension reduction, and preserving nonlinear relationships between user preferences and item features. Simulation studies and applications to three real datasets show advantages in prediction accuracy of the proposed model compared to current state-of-the-art approaches.
With the wide spread use of energy storage systems, battery state of health (SOH) monitoring has become one of the most crucial challenges in power and energy research, as SOH significantly affects the performance and...
详细信息
With the wide spread use of energy storage systems, battery state of health (SOH) monitoring has become one of the most crucial challenges in power and energy research, as SOH significantly affects the performance and life cycle of batteries as well as the systems they are interacting with. Identifying the SOH and adapting of the battery energy/power management system accordingly are thus two important challenges for applications such as electric vehicles, smart buildings and hybrid power systems. This dissertation focuses on the identification of lithium ion battery capacity fading, and proposes an on-board implementable model parametrization and adaptation framework for SOH monitoring. Both parametric and non-parametric approaches that are based on kernel functions are explored for the modeling of battery charging data and aging signature extraction. A unified parametric open circuit voltage model is first developed to improve the accuracy of battery state estimation. Several analytical and numerical methods are then investigated for the non-parametric modeling of battery data, among which the support vector regression (SVR) algorithm is shown to be the most robust and consistent approach with respect to data sizes and ranges. For data collected on LiFePO 4 cells, it is shown that the model developed with the SVR approach is able to predict the battery capacity fading with less than 2% error. Moreover, motivated by the initial success of applying kernelbasedmodeling methods for battery SOH monitoring, this dissertation further exploits the parametric SVR representation for real-time battery characterization supported by test data. Through the study of the invariant properties of the support vectors, a kernelbased model parametrization and adaptation framework is developed. The high dimensional optimization problem in the learning algorithm could be reformulated as a parameter estimation problem, that can be solved by standard estimation algorithms such as the l
暂无评论