Search

This article considers the synchronization problem of delayed reaction–diffusion neural networks via quantized sampled-data (SD) control under spatially point measurements (SPMs), where distributed and discrete delays are considered. The synchronization scheme, which takes into account the communication limitations of quantization and variable sampling, is based on SPMs and only available in a finite number of fixed spatial points. By utilizing inequality techniques and Lyapunov–Krasovskii functional, some synchronization criteria via a quantized SD controller under SPMs are established and presented by linear matrix inequalities, which can ensure the exponential stability of the synchronization error system containing the drive and response dynamics. Finally, two numerical examples are offered to support the proposed quantized SD synchronization method. [ABSTRACT FROM AUTHOR]

We propose, in this paper, a framework for time series and nonlinear system modeling, called the basis function matrix-based flexible coefficient autoregressive (BFM-FCAR) model. It has very flexible nonlinear structure. We show that many famous nonlinear time series models can be derived under this framework by choosing the proper basis function matrices. Some probabilistic properties (the conditions of geometrical ergodicity) of the BFM-FCAR model are investigated. Taking advantage of the model structure, we present an efficient parameter estimation algorithm for the proposed framework by using the variable projection method. Finally, we show how new models are generated from the proposed framework. [ABSTRACT FROM AUTHOR]

The temperature distribution of a lithium-ion battery (LIB) belongs to a nonlinearly distributed parameter system (DPS), which is infinite dimensional in the space direction. Modeling of such systems often leads to the following challenges: time/space-coupled dynamics; time-varying dynamics; and spatial nonlinearity. For the first two challenges, Karhunen–Loève (KL) decomposition based data-driven spatiotemporal models have been widely researched and successfully applied to industrial thermal processes. However, this method is a global linear model reduction method that ignores the nonlinear properties of measurements. This will prohibit the model performance of a strong nonlinear DPS, which has strong spatial nonlinearity refers to the aforementioned third challenge. To address this problem, a local-properties-embedding-based modeling method is developed for the nonlinear DPS. First, the finite-dimensional nonlinear spatial basis functions that can be the representative of the nonlinear feature of the original space are learned by the local-properties-embedding-based method. Then, the low-dimensional representative can be derived using the time/space separation and the unknown temporal coefficients can be identified through the traditional neural learning algorithm. Finally, the spatiotemporal temperature distribution can be reconstructed by the time/space synthesis. Since the nonlinear structure feature of the spatiotemporal datasets has been considered, the proposed modeling method can be more effective than the traditional KL-based method for modeling the nonlinear DPS. Numerical simulations on the LIB showed that the proposed methods are more robust than the KL-based method in both the time direction and the space direction. [ABSTRACT FROM AUTHOR]

This paper presents a real-time estimation method for the temperature distribution of cylindrical batteries under boundary air cooling. A space-/time-separation-based analytical model is developed using Karhunen–Loève decomposition and Galerkin's method. The model parameters can be identified and optimized using data-based approaches. The developed analytical model demonstrates the robustness to variation of thermal parameters. However, the change of boundary cooling will significantly degrade the performance of the developed analytical model. For the known boundary cooling, the compensation model for cooling effects can be derived to improve the modeling performance. For the unknown boundary cooling in real practice, a dual-extended Kalman filter can be used to simultaneously estimate coupled parameters and convection coefficient in the compensation model. The proposed method can achieve satisfactory performance in the battery duty-cycle experiments. [ABSTRACT FROM PUBLISHER]

Traditional wavelet system is a two-domain (time and frequency domains) wavelet system (2DWS), which works only in time and frequency domains. The 2DWS is not able to treat time-frequency information and fuzziness simultaneously. For this reason, a three-domain (fuzzy, time, and frequency domains) fuzzy wavelet system (3DFWS) is proposed, where the three-domain mechanism provides a solution to handle fuzzy uncertainties and time-frequency information together. The major advantage of 3DFWS is able to use the prior knowledge via the novel fuzzy domain to analyze uncertain data and signals, which will enhance the potentials of 2DWS. Experimental and simulation studies show that the performance of the proposed 3DFWS is superior to the traditional one for simultaneous processing of time-frequency and fuzziness. [ABSTRACT FROM AUTHOR]

Most self-tuning control algorithms for nonlinear systems become invalid when the controlled systems have nonminimum phase property. In this article, a direct neural network-based self-tuning control strategy is developed to deal with this problem under the certainty equivalence principle. Based on an equivalent linearized model from the local linearization, the controller structure is designed using a modified Clarke index with the guaranteed closed-loop stability and without the traditional requirement of the globally boundedness. For the system with unknown parameters, the controller is self-tuned by an on line RBF neural network identifier. Satisfactory simulations illustrate the effectiveness and adaptability of the proposed strategy even under system parameter variations. [ABSTRACT FROM AUTHOR]

A novel neural approximate inverse control is proposed for general unknown single-input-single-output (SISO) and multi-input-multi-output (MIMO) nonlinear discrete dynamical systems. Based on an innovative input/output (I/O) approximation of neural network nonlinear models, the neural inverse control law can be derived directly and its implementation for an unknown process is straightforward. Only a general identification technique is involved in both model development and control design without extra training (online or offline) for the neural nonlinear inverse controller. With less approximation made on controller development, the control will be more robust to large variations in the operating region. The robustness of the stability and the performance of a closed-loop system can be rigorously established even if the nonlinear plant is of not well defined relative degree. Extensive simulations demonstrate the performance of the proposed neural inverse control. [ABSTRACT FROM AUTHOR]

Glaucoma is a serious eye disease that can cause permanent blindness and is difficult to diagnose early. Optic disc (OD) and optic cup (OC) play a pivotal role in the screening of glaucoma. Therefore, accurate segmentation of OD and OC from fundus images is a key task in the automatic screening of glaucoma. In this paper, we designed a U-shaped convolutional neural network with multi-scale input and multi-kernel modules (MSMKU) for OD and OC segmentation. Such a design gives MSMKU a rich receptive field and is able to effectively represent multi-scale features. In addition, we designed a mixed maximum loss minimization learning strategy (MMLM) for training the proposed MSMKU. This training strategy can adaptively sort the samples by the loss function and re-weight the samples through data enhancement, thereby synchronously improving the prediction performance of all samples. Experiments show that the proposed method has obtained a state-of-the-art breakthrough result for OD and OC segmentation on the RIM-ONE-V3 and DRISHTI-GS datasets. At the same time, the proposed method achieved satisfactory glaucoma screening performance on the RIM-ONE-V3 and DRISHTI-GS datasets. On datasets with an imbalanced distribution between typical and rare sample images, the proposed method obtained a higher accuracy than existing deep learning methods. [ABSTRACT FROM AUTHOR]